doi
stringlengths 28
28
| title
stringlengths 19
311
| abstract
stringlengths 217
5.08k
| plain language summary
stringlengths 115
4.83k
| article
stringlengths 3.87k
161k
|
---|---|---|---|---|
10.1371/journal.pcbi.1007226 | Energy-efficient information transfer at thalamocortical synapses | We have previously shown that the physiological size of postsynaptic currents maximises energy efficiency rather than information transfer across the retinothalamic relay synapse. Here, we investigate information transmission and postsynaptic energy use at the next synapse along the visual pathway: from relay neurons in the thalamus to spiny stellate cells in layer 4 of the primary visual cortex (L4SS). Using both multicompartment Hodgkin-Huxley-type simulations and electrophysiological recordings in rodent brain slices, we find that increasing or decreasing the postsynaptic conductance of the set of thalamocortical inputs to one L4SS cell decreases the energy efficiency of information transmission from a single thalamocortical input. This result is obtained in the presence of random background input to the L4SS cell from excitatory and inhibitory corticocortical connections, which were simulated (both excitatory and inhibitory) or injected experimentally using dynamic-clamp (excitatory only). Thus, energy efficiency is not a unique property of strong relay synapses: even at the relatively weak thalamocortical synapse, each of which contributes minimally to the output firing of the L4SS cell, evolutionarily-selected postsynaptic properties appear to maximise the information transmitted per energy used.
| Compared to other organs, the brain consumes a vast amount of energy for its size. Most of this energy is used to power the electrical and chemical processes that support neural computation. As the energy supply to the brain is limited, it follows that this computation should be energetically efficient. Previously, we showed that this is indeed the case for transmission of information between cells at synapses. Synapses transferring information from the retina to the brain do not maximise information transmission—some information is lost and does not reach the visual cortex. Instead, these synapses maximise the information transmitted per energy used. Here, we demonstrate that this principle of energetic efficiency also holds at the next synapse in the visual pathway, the thalamocortical synapse. This synapse is weaker and competes with hundreds of other inputs to influence the output firing of the next cell. Using detailed simulations of cortical neurons, and electrophysiological recordings in rodent brain slices, we found that this relatively weak synapse also does not maximise information transmission. Instead, it maximises the amount of information transmitted per energy used. This suggests that energy efficiency at synapses could be a common design principle in the brain.
| Information transmission in the brain is energetically expensive [1–7], yet has to satisfy demands of speed and signal-to-noise reliability [8, 9]. In order to balance these competing demands, the brain may tend towards a design which prioritises energy efficiency at the expense of computational power. For instance, theoretical analysis has previously shown that the low mean firing rate of neurons [10] and the surprisingly low release probability of cortical synapses [4, 11] can be explained if axons and presynaptic terminals operate to maximise the information transmitted per energy used.
Experimentally, we have shown previously that the strong retinothalamic synapse, relaying information from the retina to the thalamus, also maximises energetic efficiency when transferring information [12]. At that synapse, the evolutionarily selected properties are not set to transmit the maximum amount of information possible—more information would be transmitted if larger excitatory postsynaptic currents (EPSCs) were evoked by presynaptic action potentials. However, EPSCs that are larger or smaller than physiological EPSCs decrease the information transmitted per energy used. The physiological EPSC size therefore maximises energy efficiency of information transfer rather than information transfer across the retinothalamic synapse. Crucially however, it is unclear whether energy efficiency at excitatory synapses is a special property of strong relay synapses, or a more general principle also governing synaptic inputs that contribute more weakly to determining the output of the postsynaptic cell.
To address this question, we investigated information transmission and EPSC energy cost at the next synapse along the visual pathway, from relay neurons in the thalamus to spiny stellate (SS) cells in layer 4 (L4) of the primary visual cortex (V1). This relatively weak thalamocortical synapse operates in the presence of many other synaptic inputs from the thalamus and from the cortex. Using a multicompartment Hodgkin-Huxley-type model of L4SS cells, we simulated random synaptic bombardment from thalamic and cortical synapses, while quantifying the energetic cost of information transmission across the synapses impinging on the cortical cell from a single thalamic axon. We assessed the energetic cost incurred by these postsynaptic cells in V1 by converting the ion flows across the membrane resulting from EPSCs or from action potentials into the corresponding amount of adenosine triphosphate (ATP) molecules necessary to pump these ions back out [1, 4, 12]. At the same time, we evaluated information transfer from an axon of interest to the output spike train of these cells using transfer entropy [13, 14]. Similar to what we observed at retinothalamic synapses, our simulations suggested that the energetic efficiency of information transmission was maximal at the physiologically observed level of thalamocortical synapse conductance.
Then, we tested this result experimentally on real L4SS cells patch-clamped in rat brain slices, evoking thalamocortical input while using dynamic-clamp to inject the random background synaptic conductance derived from our simulations. We found that increasing or decreasing the conductance at thalamocortical synapses decreased the energetic efficiency of information transmission from one, experimentally-stimulated, thalamic axon.
Thus, both simulations and experiments suggest that, like at the retinothalamic synapse [12], thalamocortical postsynaptic properties are evolutionarily set to be energy efficient.
P28 Sprague Dawley rats were killed by cervical dislocation following sedation with isoflurane. Animal procedures were carried out in accordance with the guidelines of the UK Animals (Scientific Procedures) Act 1986 and subsequent amendments.
The mathematical model of L4SS cells was adapted from an earlier model by Lavzin and colleagues [15] available at ModelDB (https://senselab.med.yale.edu/ModelDB/showmodel.cshtml?model=146565) [16]. The multicompartment simulations were conducted using the NEURON 7.3 simulation platform (http://neuron.yale.edu/) and will be available from our GitHub page (https://github.com/JolivetLab).
The cell was subdivided into 360 compartments, with a maximum length of 21 μm (14.6 μm on average). Following Lavzin et al. [15], the soma area was 757 μm2, the total dendritic area was 11,885 μm2, the resting membrane potential was -70 mV, the membrane resistance was 16,000 Ω·cm2, the axial resistivity was 100 Ω·cm and the membrane capacitance in all compartments was set to 1.5 μF/μm2 to account for the presence of spines.
The model included Hodgkin-Huxley-type voltage-gated channels. Fast sodium channels (reversal potential = 70 mV), and delayed rectifier and slow non-inactivating potassium channels (reversal potential = -77 mV), were distributed with a higher density at the soma (gNa = 300 mS/cm2, gKdr = 30 mS/cm2, gKs = 100 mS/cm2) than in the dendritic tree (gNa = 3 mS/cm2, gKdr = 1 mS/cm2, gKs = 1 mS/cm2). L-type voltage-gated calcium channels were distributed evenly across the cell (gCa = 0.03 pS/μm2). Calcium diffusion within and between compartments was modelled as described in Carnevale and Hines [17] with a diffusion coefficient of 0.6 μm2/ms, and an intracellular calcium buffer at 3 μM concentration with a dissociation constant of Kd = 1 μM. Calcium pumping across the surface membrane was modelled again as described in [17] with a plasma membrane calcium pump density of 10−13 mol/cm2. The pump is modelled with a two-step reaction. The first step describes binding of intracellular calcium to the pump, with a binding rate constant of k1 = 1 mM-1ms-1 and an unbinding rate constant of k2 = 0.005 ms-1 (giving an apparent dissociation constant of 5 μM). The second step describes loss of the calcium from the pump to the extracellular solution, with an unbinding rate constant of k3 = 1 ms-1 and a rebinding rate constant of k4 = 0.005 mM-1ms-1 (giving an apparent dissociation constant of 200 mM). See Chapter 9 of Carnevale and Hines [17] for further details.
To simulate cortical background input to the modelled cell, one lumped excitatory and one lumped inhibitory synapse (each representing a number of synapses) were placed on each compartment of the model (Fig 1A). The frequency of activation of each lumped synapse was adjusted by the size of the compartment area so as to model monosynaptic inputs onto 1870 excitatory and 460 inhibitory spines homogeneously distributed throughout the dendritic tree at densities of 0.167/μm2 and 0.041/μm2 respectively, which were randomly activated at 0.45 Hz and 0.09 Hz respectively (Fig 1B). Frequencies for the corticocortical background input were slightly adjusted from Waters and Helmchen [18], so that after adding thalamocortical synapses (see below), the modelled cell’s output frequency was ~4 Hz, the mean cortical firing rate in vivo [1]. Excitatory conductances were modelled with a time course of the form exp(-t/τfall)—exp(-t/τrise) with time constants τrise = 0.3 ms and τfall = 1.7 ms, and with the reversal potential at 0 mV. Inhibitory synapses were modelled with the same function with time constants τrise = 1 ms and τfall = 10 ms, and with the reversal potential at -75 mV. The conductance of these synapses was held constant at gcc = 0.0008 μS and ginh = 0.0005 μS. To test for the influence of inhibition, some simulations were run without inhibitory synapses.
To simulate thalamocortical background input to the modelled cell, one excitatory synapse was placed on each compartment and the frequency of activation of each one of those synapses was adjusted according to the distance of that compartment to the soma (Fig 1B). This is equivalent in the NEURON simulation environment to modelling 470 spines distributed as a Gaussian with respect to their position in the dendritic tree [19] and randomly activated at 4 Hz (the average output frequency we measured in LGN cells [12]), but is less computationally intensive. Thalamocortical synapses were modelled with a conductance time course of the form exp(-t/τfall)—exp(-t/τrise) with τrise = 0.3 ms and τfall = 1.7 ms, and with the reversal potential at 0 mV. Their conductance was initially set at gtc = 0.001 μS.
To evaluate the contribution of an individual thalamic cell projecting onto the modelled cell, three synapses were added, clustered onto the same dendritic branch, approximately 78 μm away from the soma [19] and typically spaced by 18 μm (see Fig 1A and 1B for details; located on compartments a3_122, a3_121 and a3_12 at positions 0.5, 0.2 and 0.9 respectively, see file l4sscell.hoc in https://senselab.med.yale.edu/ModelDB/showmodel.cshtml?model=146565 for details). These synapses were modelled using the same parameters as for other thalamocortical synapses. In particular, their conductance was initially set at gsyn = 0.001 μS. This is to approximate in the model three synaptic sites, which together produce an EPSC of a similar magnitude to that observed in experiments (3*(conductance of 10-9S)*(driving force of -70mV) ~ 210pA; see below). These synapses were synchronously activated using the same input spike train as was used for the experiments described below in the presence of both thalamocortical, and excitatory and inhibitory corticocortical background noise. To determine the optimal conductance value for all thalamocortical synapses, we modulated both gsyn and gtc by the same gain factor while maintaining gcc and ginh constant. None of the synapses exhibited plasticity, facilitation, depression or failures. Note however that these synapses typically express only mild depression (see below).
To test for the influence of the exact location of these three additional thalamocortical synapses on our results, we repeated all the simulations with a second set of such synapses (located on compartments a5_1111, a5_11111 and a5_11111 at positions 0.9, 0.15 and 0.1 respectively, see file l4sscell.hoc in https://senselab.med.yale.edu/ModelDB/showmodel.cshtml?model=146565 for details). These additional simulations revealed no qualitative differences from what had been obtained with the first set. The location of the first set of synapses is plotted in Fig 1A (syn).
Additionally, for experiments described below, we generated two synthetic “noise” conductance trains, one corresponding to the total excitatory corticocortical background input and one corresponding to the total thalamocortical background input. To generate these trains, simulations were run as described above with two exceptions. First, inhibitory input was ignored. Second, all synapses were relocated to the soma and their conductances were summed.
To calculate the metabolic cost incurred by the modelled cell (Figs 2 and 3), the total synaptic current generated by gsyn, which is the largest signalling cost [1, 4], was integrated over the three synaptic locations and converted to the corresponding ATP consumption per unit time [1, 12]. The energetic cost of other synapses and output action potentials was calculated in the same way. For the experiments described below, the ATP used to reverse the postsynaptic ion flux was calculated for voltage-clamp and dynamic-clamp recording modes as described in [12].
Using information theory in neuroscience has a long tradition [20]. In particular, it is common to use mutual information to quantify the flow of information from one neuron to another, or from a stimulus set to a recorded neuron [4, 12, 21, 22]. Numerous methods have been devised to measure mutual information in various contexts, and to correct for its inherent biases (see [23] for a review). However, mutual information is by construction a symmetric measure and thus does not strictly measure information being transferred from a sender to a receiver. Rather, when measuring mutual information between two random variables, one captures how much information can be inferred about one of those processes when observing the other one. Mutual information also suffers from significant estimation biases when the dataset is limited, a recurrent problem in experimental contexts like the one we will deal with here.
More recently, a measure analogous to mutual information called transfer entropy, which seeks to capture the unidirectional flow of information from one variable to another, was introduced [13], and has been increasingly applied to neuronal spike data (for comprehensive reviews of the application of transfer entropy in neuroscience, see [24–26]). Transfer entropy is non-symmetric. Let us define two processes I and J with joint probability pIJ(i,j). The transfer entropy from J to I is defined by:
TEJ→I=∑p(in+1,in(k),jn+1−u(l))logp(in+1|in(k),jn+1−u(l))p(in+1|in(k))
(1)
where in(k)=(in,…,in−k+1) is a shorthand notation for words of length k, and where n denotes the nth time bin and u an optional frame shift (see below) [13]. The sum runs over all possible combinations of in, in(k) and jn(l). For words of length 1, in simple cases where information flow is strictly unidirectional and when consecutive bins are both independent and conditionally independent given the source value, it is possible to show analytically that transfer entropy is equivalent to mutual information. Simulations also suggest that transfer entropy is largely superior to mutual information in that its estimate converges much faster when increasing the size of the dataset when analysing spike trains (Conrad and Jolivet, in preparation). Transfer entropy is used extensively in fields outside neuroscience and in systems neuroscience but is used relatively little in cellular neuroscience. While we decided to use here transfer entropy instead of mutual information because of the technical reasons highlighted above, essentially the same results were obtained when we used mutual information. Similarly, there is no reason to think that we would have found different results in ref. [12] had we used transfer entropy instead of mutual information.
Here, we binarized the 125 second input and output spike trains in 3 ms time bins (approximately the refractory period of a neuron) and measured the transfer entropy from the input axon to the output spike train using the package published by Ito and colleagues (https://code.google.com/archive/p/transfer-entropy-toolbox/) [14]. The results of this procedure were divided by the time bin (3 ms) to get information rates in bits/sec [27].
The package by Ito and colleagues allows consideration of a temporal frame shift (u) between the sender and the receiver. We systematically analysed the effect of this parameter on the results and observed that peak information transfer always occurs in the same time bin in our simulations, i.e. that transfer entropy is maximal without a frame shift between the input and output sequences when using a 3 ms temporal resolution (S2 Fig). Using shorter bins yielded similar results (see S3 Fig). Nevertheless, in all subsequent analysis, we allowed for temporal frame shifts of up to 10 time bins (30 ms). Changing this value had no significant impact on the results.
Changing the word length however does have an effect on the measure of transfer entropy. (We talk here about word length in the singular, as we systematically considered the same word length for both the input and output sequences, thus k = l. See refs. [26] and [28] for best practice in choosing word lengths). Specifically, increasing the value of k increases the measured transfer entropy, because of bias increase (this is well documented [26–28]). With a finite dataset, random coincidences can lead to mis-estimating probabilities, which will add up in the final calculation of the transfer entropy product (see Eq 6 in [14]).
In order to remediate that problem, one solution is to subtract from our raw estimates of TE, the value of TEnoise, calculated between a random permutation of the input sequence and the actual output sequence. We carried out a systematic analysis of the effect of varying k on TE—TEnoise and observed no significant changes in the main results of this paper when k ≤ 10. We thus carried out all of our analyses with k = l = 10 (30 ms). All values reported in the results section are of TE defined as:
TE=TEraw−TEnoise
(2)
with TEraw calculated following Eq (1) above and TEnoise calculated following Eq (1) above but with randomly permuting the input sequence (randomizing words instead of individual time bins produces almost identical results, see S1 Fig). In practice, for each condition, we generated multiple simulations with different seeds initialising the random number generator, averaging about 6 individual simulations for each condition. For each one of those, we calculated TEraw, 30 realisations of TEnoise over which we took an average, and TE. The values of TEnoise and TE reported in the Results section are the average over all these individual simulations.
For all conditions (simulations, and experiments with real stimulation and all dynamic-clamp gains described below), the information rate was divided by the rate of energy consumption on reversing the ion flux generating EPSCs and action potentials to get a measure of efficiency in bits/(ATP consumed). All data were analysed using custom-written MATLAB scripts (The Mathworks Inc.).
In the second part of the present study, we sought to replicate our simulation results in in vitro experiments using rat brain slices. The first part of each experiment was performed in whole-cell voltage-clamp mode, in order to calculate the energy cost per EPSC. L4SS cells were patch-clamped and thalamic axons were stimulated with a previously recorded thalamic relay neuron response train to retinal input (see below). The second part of the experiment was performed by injecting the same stimulation pattern in current-clamp in order to measure information transmission across the synapse. From the voltage-clamp recording, a conductance trace was then generated and injected into the cell with dynamic-clamp. The conductance trace, to which noise was added to simulate the physiological background of additional synapses, was scaled up or down to measure information transmission at conductances above or below the physiological value. Energetic efficiency for each scale factor (gain) was calculated by dividing the transfer entropy rate by the rate of energy consumption [4, 12]. This allowed us to assess whether, as in the LGN [12], energetic efficiency at the thalamocortical synapse was maximised at the physiological gain.
P28 Sprague Dawley rats were killed by cervical dislocation following sedation with isoflurane. The brain was rapidly removed and immersed in ice-cold, slicing solution containing (in mM): 87 NaCl, 25 NaHCO3, 7 MgCl2, 2.5 KCl, 1.25 NaH2PO4, 0.5 CaCl2, 25 glucose, 75 sucrose, 1 kynurenic acid, saturated with 95% O2/5% CO2 (modified from ref. [29]).
The hemispheres were separated along the midline and an off-coronal cut (20° from vertical, heading anteriorly while cutting towards the dorsal surface, between the cerebrum and cerebellum) was performed on each hemisphere to create an angled surface, which was glued onto the stage of the vibratome. Off-coronal 225 μm slices containing the thalamus and cortex including V1 were then made.
Slices were placed in a storage chamber containing continuously oxygenated slicing solution at 35°C, which was allowed to come to room temperature naturally. During the experiment, slices were continuously perfused with artificial cerebrospinal fluid (aCSF) containing (in mM): 124 NaCl, 26 NaHCO3, 10 glucose, 2.5 KCl, 2 CaCl2, 1 NaH2PO4, 1 MgCl2, 0.005 Gabazine (to block disynaptic inhibition during stimulation). The aCSF was heated to 35°C and constantly bubbled with 95% O2/5% CO2.
Whole-cell recordings from cortical L4SS cells were obtained using 2–3 MΩ borosilicate glass electrodes filled with internal solution containing (in mM): 130 K-gluconate, 10 EGTA, 10 HEPES, 4 NaCl, 4 MgATP, 1 CaCl2, 0.5 Na2GTP, 0.4 K2-Lucifer yellow. Spiny stellate cells in layer 4 of cortical area V1 were identified according to their location, morphology, and electrophysiological characteristics (Fig 4A and 4B). In contrast to the less-numerous pyramidal neurons also found in L4, the round or ellipsoidal spiny stellate cells do not have a prominent apical dendrite or dendrites reaching across several layers, but have a broader dendritic tree confined to L4 [30, 31] with a high density of spines, which could be seen once a cell was dye-filled (Fig 4A). L4SS cells were recorded at their resting potential of approximately -70mV, at which they respond to current injection with sustained, regular spiking (Fig 4B; [32]).
Online corrections were made for the junction potential of -14 mV for the gluconate-based internal solution used (e.g. neurons were held at an apparent potential of -56 mV to achieve a true potential of -70 mV). Recordings were made with an Axopatch 200B amplifier, filtered at 5 kHz and sampled at 20 kHz. Data were acquired using custom-made MATLAB software, kindly provided by Ho Ko and Tom Mrsic-Flogel (UCL).
The first part of the experiment was performed in voltage clamp. Upon seal formation, pipette capacitance was compensated. Once in whole-cell mode, the series resistance was compensated by up to 70% (after which the mean series resistance was 6.4 ± 1.0 MΩ). The second part of the experiment was performed in current clamp, using the I-CLAMP FAST mode (which was stable with the 2–3 MΩ pipettes used). In current clamp mode, series resistance compensation was set to 100%.
Thalamic axons in the subcortical white matter were stimulated extracellularly with a borosilicate glass electrode (gently broken to achieve a tip diameter of ~10–15 μm) containing aCSF. In voltage clamp, stimulation strength was gradually increased to achieve the smallest reliable EPSC (defined as an EPSC that, when it occurred, did not vary in size in response to a pulse delivered every 3 s). This stimulus intensity was then maintained throughout the experiment. The average minimal EPSC size was 356 ± 186 pA (Fig 4C) and paired thalamic stimulation elicited the mild EPSC depression characteristic of this synapse (PPR = 0.70 ± 0.13, Fig 4D and 4E; [31, 33, 34]).
We stimulated thalamic axons making synapses onto whole-cell patch-clamped L4SS cells of V1 in rat brain slices (Fig 4A). The spike train used for stimulation was recorded from a typical LGN relay neuron in response to retinal ganglion cell (RGC) input [12], which in turn was recorded from RGCs in response to natural movies [35]. The spike train was 25 seconds long, with an average frequency of 4.3 Hz. The stimulation procedure followed that of our previous experiments in the LGN [12]: After a 5 second input train to habituate the synapse (the data from this train were discarded), the 25 second spike train was repeatedly played five times, resulting in 125 seconds of stimulation in total. This procedure was followed once in voltage-clamp, once in current-clamp, and several times in dynamic-clamp with various conductance gains (see below).
A 125 second long conductance train (gsyn) was derived from the 125 second postsynaptic current recording obtained from each spiny stellate cell in voltage clamp. First, we removed the stimulation artefacts by setting the current value for the duration of the artefact to the current value immediately preceding the artefact. The resulting current trace (Isyn) was converted to a synaptic conductance trace (gsyn) via:
gsyn(t)=Isyn(t)/(Vm−Vrev)
(3)
where Vm is the membrane potential of the cell (the holding potential, -70 mV), and Vrev is the reversal potential of the synapse (0 mV, the reversal potential for glutamatergic ionotropic receptors).
In addition to the synaptic conductance trace, we mimicked the effect of this synapse operating in the presence of hundreds of other synaptic inputs. We generated two synthetic “noise” conductance trains, designed to have the characteristics of (1) thalamocortical (TC) inputs (4 Hz input to 470 spines: gtc) and (2) corticocortical (CC) inputs (0.45 Hz input to 1870 spines: gcc). These two trains were the result of simulations carried out before the experiment (see above the section ‘Mathematical model of spiny stellate cells’), and were the same for each cell. In subsequent dynamic-clamp experiments, the amplitude of the TC conductance noise was scaled with the conductance of the synapse being studied (gsyn), since we assume that all the TC synapses will have their conductances set to the same optimal value. In contrast, the corticocortical noise was not scaled.
The baseline amplitude of the noise conductance was set individually for each cell by combining both trains (gtc+gcc) and scaling them up or down (mean scaling factor 0.4 ± 0.1) until a firing frequency of approximately 4 Hz [1] was triggered upon injection into the L4SS cell (“pre-scaling”, Fig 4G). The pre-scaled thalamic noise train (gtc) and the thalamic single-input train recorded in that cell (gsyn) were then summed and scaled together (by a factor of 0.1, 0.3, 0.5, 0.75, 1, 1.5, and 3). In contrast to the LGN relay synapse [12], for which only the synaptic conductance was scaled up, in this experiment the combined synaptic and noise conductances could not be scaled above 3 times the physiological value without inducing oscillations or a depolarising block, resulting in a lack of action potential firing.
The pre-scaled corticocortical noise train (gcc) was then added to each of these scaled trains, to create 7 differently scaled composite conductance trains for each cell (Fig 4G). These trains were injected into the cell using dynamic-clamp (SM-1, Cambridge Conductance [36]), which injects a time-varying current Iinj(t), at time t, calculated from gsyn(t) and the instantaneous value of the cell membrane potential:
Iinj(t)=gsyn(t)·(Vm(t)–Vrev)
(4)
Because of the liquid junction potential, the Vm received by the SM-1 was 14 mV more positive than the real membrane potential. We therefore set Vrev on the SM-1 to 14 mV (rather than 0 mV), to account for this in the online calculation of Iinj. In this calculation, all of the synaptic current was assumed to scale linearly with membrane potential. The voltage response of the postsynaptic cell was simultaneously recorded.
After injecting gsyn+tc /(normal gsyn+tc)*1, the order of scaled conductances was randomized. This meant that not every cell experienced every scale factor, as it was not usually possible to maintain whole-cell recordings for long enough to carry out every conductance injection, after all initial steps were completed (i.e. minimal stimulation protocol, paired pulse ratio characterisation, voltage clamp response to real stimulation, current clamp response to real stimulation, noise pre-scaling to achieve 4 Hz firing rate, and finally various dynamic clamp conductance injections). As such, gsyn+tc /(normal gsyn+tc)*1 was applied to all 6 cells, *0.3 to 3 cells, *0.5 to 4 cells, *0.75 to 5 cells, *1.5 to 5 cells, and *3 to 4 cells. The mean and SEM were calculated separately across these n numbers for each condition.
Data were analysed using custom scripts written in MATLAB (The Mathworks Inc.). Postsynaptic current traces were used to calculate ATP consumption at the synapse as described above. Postsynaptic voltage traces were converted to binarised sequences of 1s (representing action potentials) and 0s (their absence) by identifying events whose amplitude exceeded a threshold defining action potential occurrences (set individually for each cell, between -15 mV and -30 mV). This output sequence could then be compared with the binary input spike train to look at simple transmission characteristics (Fig 4F), or used to calculate the amount of information that would be propagated to the visual cortex by the postsynaptic cell (Figs 5A and 6C).
Spike transmission characteristics were calculated as in [12]. The percentage occurrence of each possible input-output relationship (Fig 4F) was calculated by searching for an output (AP or EPSP) in the 18 ms following an input spike (top row) or for an input (AP) in the 18 ms preceding an output spike (bottom row), and summing each occurrence across all L4SS neurons studied.
Data are presented as mean ± standard error of the mean (s.e.m.), unless mentioned otherwise. Differences between means were assessed with Student’s t-tests (Fig 5). Note that we did not compare the means between different conductance scaling factors for the experimental results in Fig 6; we present ± s.e.m. for each point simply as a visual guide for the reader. This is because we want to be careful not to claim that the peak is precisely at 1. We do not have a large enough dataset nor fine enough sampling around gsyn = 1 to claim that this is the exact value of the peak.
We have previously shown that the strong retinothalamic synapse relaying information from the retina to the thalamus maximises energetic efficiency (information transmitted per ATP used) when transferring information [12]. It is however unclear whether this principle of energy efficiency also applies at synapses that contribute more weakly to determining the output of the postsynaptic cell. To address this question, we adapted a Hodgkin-Huxley-type multicompartment model of a layer 4 spiny stellate (L4SS) cell [15] (see Materials & Methods), the next step in the visual pathway after the thalamus. L4SS cells receive excitatory inputs from thalamocortical relay cells, and excitatory and inhibitory inputs from cortical neurons (Fig 1A). While corticocortical synapse surface density is assumed to be homogeneous throughout the dendritic tree, thalamocortical synapse density follows a Gaussian distribution with respect to the distance from the soma (Fig 1B) [19]. To create background ‘noise’, all thalamocortical synapses were randomly activated at 4 Hz [12], while corticocortical excitatory and inhibitory synapses were randomly activated at 0.45 Hz and 0.09 Hz respectively [18]. Synaptic conductances were initially set to the same value for every synapse in a given category so as to generate on average an ~0.8 mV depolarisation at the soma for thalamocortical synapses, a ~0.6 mV depolarisation for corticocortical synapses and a ~0.1 mV hyperpolarisation for inhibitory synapses (Fig 1C). This constant barrage of excitatory and inhibitory inputs generated random fluctuations of the membrane voltage at the soma and led to irregular spiking (Fig 1D, black trace).
We then modelled one extra thalamocortical axon. This axon of interest contacts the dendritic tree at three independent but relatively close synaptic locations (‘syn’ in Fig 1A). Activation of this single extra axon (at ~4 Hz, the firing frequency of the thalamocortical neurons for physiological input [12]) visibly affects the output of the neuron even in the presence of intense random background activity from all other synapses (Fig 1D, light violet trace; activation times for that specific simulation are labelled with vertical light violet bars, the background synaptic input was identical for the black and light violet traces). Thus, some information about this input spike train is carried across the synapses to be represented in the output spike train of the postsynaptic L4SS neuron. We then evaluated the amount of information transmitted across thalamocortical synapses.
Information transfer at synapses can be assessed in different ways. One commonly used method consists of computing the mutual information between the input and output spike trains across the synapse(s) of interest [20]. This method has been successfully applied in a number of studies (e.g. [21, 22]), including by ourselves [4, 12], and numerous studies have addressed the pitfalls and biases of using mutual information in that context (see [23] for a review).
We have recently used the so-called direct method of Strong et al. [21] to measure information transfer (mutual information) at the retinothalamic synapse [12]. That synapse differs significantly, however, from the one we study here, as it is a strong relay synapse, generating EPSCs sufficiently large that a single one of them is often sufficient to trigger an output action potential on its own. Indeed, the output of thalamic relay cells in the visual pathway is driven almost entirely by input from a single strong relay synapse impinging on their dendritic tree from the retina [37]. The case we study here is different as L4SS cells receive weak inputs from hundreds to thousands of thalamic and cortical synapses (Fig 1A and 1C), each one of them generating small EPSCs that individually contribute little to the output of the cell (Fig 1B–1D).
Preliminary simulations revealed that using the direct method would be inappropriate because of the relative weakness of the synapse and the presence of background noise. At the thalamocortical synapse, in the presence of a significant barrage of background inputs, most output spikes are not related to input spikes in the axon of interest (Fig 1D). In these circumstances, we could not feasibly run long enough simulations and experiments to estimate accurately the mutual information between input and output sequences, a recurrent problem in experiments using acute brain slices with a lifespan counted in hours [38]. The direct method we have used in ref. [12] computes the mutual information by subtracting the noise entropy (Hnoise) from the output’s total entropy (Htotal). While Htotal is estimated by looking at the distribution of binary ‘words’ over time, Hnoise is computed as the entropy of these words’ distributions across repetitions, and then averaged over time. It is thus crucial for correctly estimating Hnoise to collect a sufficient number of repetitions. Htotal and Hnoise impose competing constraints over the time available for data collection as the best estimate of Htotal will be obtained with the longest possible non-repeating sequence, while the best estimate of Hnoise will be obtained with the largest number of repetitions of a repeating sequence. In Harris et al. [12], in the case of thalamic relay cells, most output spikes were directly driven by an input spike and the noise was relatively small in comparison to Htotal. It was thus possible to reach a reasonable estimate of the mutual information with few repetitions. In the present case, only a few of the output spikes are related to an input spike from the thalamocortical axon of interest, and one expects that Hnoise becomes approximately equal to Htotal, necessitating a number of repetitions for proper estimation of the mutual information incompatible with the experiments presented here. Others have faced the same problem before. London and colleagues, for instance, addressed it by devising an alternative measure similar to mutual information called synaptic information efficacy [39]. However, another concern is that measures like the mutual information are symmetric, i.e. they measure how much information can be inferred about one process when observing the other one, but do not measure directional information flow.
To deal with these issues, we decided to employ the measure termed transfer entropy (see Eq 1 in Materials & Methods) [13]. Transfer entropy is designed to measure directional information transfer from a sender to a receiver. If the information flow is unidirectional (from the sender to the receiver), measuring transfer entropy from the receiver to the sender will return 0. Thus, we binarized the 125 second input and output spike trains in 3 ms time bins and measured transfer entropy from the input axon to the output spike train using the package published by Ito and colleagues (http://code.google.com/p/transfer-entropy-toolbox/) [14]. Due to the limited size of the dataset that can be realistically acquired, we additionally applied a correction for random coincidences that contribute to noise in the estimation of the transfer entropy (see Materials & Methods for a detailed discussion of these issues). In the following, we usually report TEraw, the raw value of transfer entropy calculated from the input to the output using Eq (1) (see Materials & Methods), TEnoise, the value of transfer entropy calculated using Eq (1) but using a scrambled input (scrambling the sequence of words instead of the sequence of individual time bins produces almost identical results, see S1 Fig), and TE = TEraw−TEnoise, the effective information transfer after correcting for the effect of random coincidences (Eq (2) in Materials & Methods).
In multicompartment simulations of the thalamocortical synapse, using TE for our axon of interest, we found that a small amount of information is indeed carried across the simulated synapse of interest, from input action potentials, even in the presence of strong background noise generated by other synapses (Fig 1A), yielding a transfer entropy of TE = 0.47 ± 0.06 bits/sec (mean ± s.e.m.) after subtraction of TEnoise (TEraw = 1.0 ± 0.1 bits/sec; mean ± standard deviation), which is roughly 40 times less than at thalamic relay cells [12].
We then investigated whether modulating the strength of thalamocortical connections modulates information transfer, and how this relates to energy consumption by the postsynaptic cell.
In order to assess how modulating the strength of thalamocortical connections affects the postsynaptic cell’s energy consumption and information transfer at our axon of interest, we scaled all thalamocortical synapses by a range of gain factors [0, 0.4, 0.6, 0.8, 1, 1.2, 1.4, 1.6, 2.4, 4.8, 7.2, 9.6] while injecting excitatory and inhibitory corticocortical background noise at a constant level.
Increasing the conductance of all the thalamocortical synapses increased both the EPSC size (because more current is injected from the axon of interest) and the energy consumption in the postsynaptic cell associated with the thalamic input (Fig 2A), because more sodium ions need to be actively extruded from the cell via the Na,K-ATPase pump due to extra Na+ entry at all the thalamocortical synapses (see [4] for a review of synaptic energetics). This increase in energy consumption is roughly proportional to the scaling factor for the thalamocortical synapse conductance (Fig 2A).
Altering the conductance of the set of thalamocortical inputs to the L4SS cell affects the output firing rate. The rate of information transmission across a single thalamocortical connection increases with the output firing rate as a sigmoid function, up to about 20 Hz, above which the information transmission plateaus (Fig 2B). This is true for the raw value of transfer entropy (TEraw) or after subtraction of TEnoise (TE). Increasing the conductance of the set of thalamocortical inputs by 20% increases the amount of information that is transmitted across a single thalamocortical connection by ~16%, showing that thalamocortical inputs to L4SS cells, as with retinothalamic inputs to relay neurons [12], do not maximise information transmission across single connections (Fig 2C). However, dividing the information transmitted by the energy used on this connection demonstrates that the physiological thalamocortical input characteristics are close to optimality for maximising the information transmitted per energy used for each individual connection (Fig 2D). This was true when placing the three synapses of interest at the location depicted in Fig 1A (syn) or at a different location on another dendrite. Specifically, to test for the influence on our results of the location of the three thalamocortical synapses from our single axon of interest (see Fig 1A), we repeated all the simulations with a second set of such synapses (see Materials & Methods for details). These additional simulations revealed no qualitative differences to what had been obtained with the first set. Results for this second set of synapses are plotted in dark violet in Fig 2A–2C, with the average over both datasets plotted in blue. The ratio of information transmitted to energy used is approximately maximised for both datasets when looking at TEraw (light and dark violet traces), averaging over both sets of locations (blue trace), or when looking at TE after noise subtraction and averaging over both datasets (red trace), either when considering the energetic cost associated with the synapses of interest only (Fig 2D), or when considering the total signalling energy budget of the cell (Fig 2E). While small differences are present between the results obtained with the two different synapse placements tested for the investigated thalamocortical axon, the average shows a sharp peak at gain = 1, suggesting that these synapses are, on average, tuned for energetic optimality of information transmission. These results were obtained with words of length 10 time bins (30 ms) and allowing for temporal frame shifts between the input and output sequences of up to 30 ms. Note however that transfer entropy is maximal without a frame shift between the input and output sequences when using a 3 ms temporal resolution and all results reported hereafter are for temporally aligned sequences with no frame shift. Systematic tests showed that changing the value of these parameters, while introducing small quantitative changes to the results, did not lead to significant changes in the conclusions reached above (see Materials & Methods for a detailed discussion of these parameters).
To test for the influence of inhibition at the thalamocortical synapse, we used the mathematical model of L4SS cells and ran a second set of simulations with the same synaptic placements as before but without inhibitory input (setting ginh to 0; see Materials & Methods). The results obtained without inhibition are plotted in Fig 3 and overall, match those from Fig 2 (with inhibition) very closely. As for when inhibition was present, the increase in energy consumption was roughly proportional to the scaling factor of thalamocortical synapse conductance (Fig 3A), the rate of information transmission across a single thalamocortical connection increased with the output firing rate as a sigmoid function (Fig 3B) and thalamocortical inputs to L4SS cells did not maximise information transmission across single connections (Fig 3C), although now increasing the conductance of the set of thalamocortical inputs could only increase the amount of information that is transmitted across a single thalamocortical connection by ~8%.
Interestingly, even though the changes between the results obtained with (Fig 2) and without inhibition (Fig 3) were small, they have an effect on the value of the energetically optimal conductance for information transmission. Specifically, removing inhibition slightly shifted the curve in Fig 3C to the left (towards lower gain factors) when compared to results obtained with inhibition (Fig 2C). As a result, dividing the information transmitted by the energy used by all thalamocortical connections now yielded an optimum synaptic conductance value for energetic efficiency that was slightly lower than obtained previously (Fig 2D). Energetic optimality of information transmission was obtained for TE for gain = 0.8 when considering the energetic cost associated with all thalamocortical synapses (Fig 3D) and again for gain = 0.8 when considering the total signalling energy budget of the cell (Fig 3E).
Overall, these results suggest that the finding that experimentally reported synaptic conductances are close to the energetic efficiency for information transmission is robust, even though the exact position of that optimum on the conductance scale can be slightly affected by the exact positioning and clustering of synapses on the dendritic tree, or by the fine balance between excitatory and inhibitory activity levels in the network.
We next explored whether these results could be replicated experimentally.
In order to experimentally test our simulation results, we stimulated thalamic axons making synapses onto whole-cell patch-clamped L4SS cells of V1 in rat brain slices (Fig 4A and 4B). As for the retinothalamic synapse [12], we used P28 animals, minimal stimulation to activate a single input, and gabazine to block GABAA receptors. The stimulus size in the experiments was adjusted to produce the smallest detectable EPSC (mean value 356±186 pA). Single EPSCs were smaller (Fig 4C) and showed less paired pulse depression (Fig 4D and 4E) than at the retinothalamic synapse (compare with S1 Fig in ref. [12]).
The L4SS cells were held at -70 mV, and the input synapse stimulus trains used were thalamic relay neuron responses to retinal input obtained by [12], which had a mean spike frequency of ~4.3 Hz. Upon presynaptic stimulation, fewer spikes were successfully transmitted and more spontaneous spikes occurred at the thalamocortical synapse (9% and 68%, respectively, Fig 4F) than at the retinothalamic synapse described previously (19% and 7% respectively, [12]), confirming that one lateral geniculate nucleus (LGN) relay neuron has a weaker effect on L4SS cell firing than does one retinal ganglion cell on LGN relay neuron firing.
In contrast to the strong one-to-one relationship between retinal ganglion cells (RGCs) and thalamic relay neurons, where RGC activation is sufficient to evoke a relay neuron spike 19% of the time [12], L4SS cells receive weaker input from ~100–600 spines targeted by thalamic neurons (cat V1: Peters and Payne [40]; rat barrel cortex: Bruno and Sakmann [41]), that individually evoke 0.5~2 mV EPSPs (cat V1: Stratford et al. [33]; rat barrel cortex: Bruno and Sakmann [41]). These EPSPs are sufficient to evoke a spike only 9% of the time (Fig 4F). The receptive field of a L4SS cell differs from those of its input relay neurons (cat V1: Hubel and Wiesel [42, 43]), and L4SS cell spiking is also generated by intracortical inputs (mouse V1: Lien and Scanziani [44]) with input strengths that tend to be slightly smaller (~1 mV in L4 and ~ 0.22 mV in L6), but more numerous, than the thalamocortical inputs (cat V1: Stratford et al. [33]; Tarczy-Hornoch et al. [45]). Is the ratio of information transmitted to energy used maximised at a single thalamocortical synapse, operating in the context of hundreds of similarly weighted synapses?
We recorded the sequence of L4SS cell EPSCs evoked by minimal stimulation with a physiological firing pattern (i.e. the thalamic relay neuron responses to retinal input described above), and converted this to a synaptic conductance train (gsyn; Fig 4G). To simulate the action of this synapse with background synaptic activity present, we used our computational model to synthetically generate two excitatory “noise” conductance trains (see above and Materials & Methods). One train represented other thalamocortical input synapses, which make up ~20% (5% in cat V1: Peters and Payne [40]; 30% in mouse V1: Lien and Scanziani [44]) of the excitatory synapses to L4SS cells (gtc), while the other represented corticocortical inputs, which make up the remaining 80% of excitatory synapses to L4SS cells (gcc) (Fig 4G). (L4SS neurons also receive inhibitory input in vivo, which was not injected experimentally, but was simulated in our Hodgkin-Huxley-type multicompartment model, see above.) The baseline levels of the two excitatory noise traces were scaled together for each cell (mean scaling factor 0.4 ± 0.1 in 6 cells) so that, when injected simultaneously using dynamic-clamp, the L4SS cell firing frequency was approximately 4 Hz, the mean cortical firing rate in vivo [1]. Then, gsyn was added to the noise trains and the resulting composite conductance time course was injected into the L4SS cell (using dynamic-clamp to generate the appropriate membrane current from the conductance time course), while the voltage response was recorded (Fig 4H). Adding gcc+gtc led to an increase in the proportion of spikes transmitted from the stimulated thalamocortical axon (from 9% to 16% of input spikes), presumably because depolarisation by the synaptic noise conductances (which would be occurring in vivo) enabled more synaptic depolarisations to reach threshold (Fig 4F).
The input action potential train was derived from the output spike train recorded in thalamic relay cells, which had a frequency of 4.3 Hz, and had an entropy rate of about 32 bits/sec [12]. The output spike train recorded in cortical L4SS neurons (without the simulated synaptic noise added) had a frequency of 1.2 ± 0.8 Hz (mean ± s.e.m.) and a transfer entropy rate of 1.9 bits/sec (1.9 ± 1.6 bits/sec; mean ± s.e.m.; Fig 5A) encoding ~1.6 bits/spike (1.9 bits/sec at 1.2 Hz).
Employing dynamic clamp to apply the recorded synaptic conductance at the cell soma, in addition to the thalamocortical and cortical noise conductances, gave a transfer entropy rate of 1.5 ± 1.1 bits/sec, which was not significantly different from that seen with real presynaptic stimulation (paired Student t-test, p = 0.41; Fig 5A). These rates were slightly higher than the rates observed in simulations (TE = 0.47 ± 0.06 bits/sec), probably due to the fact that unlike in simulations, in experiments, all input is injected at the soma.
We calculated the energy used on a single thalamocortical connection–either evoked by presynaptic stimulation in voltage-clamp mode or scaled and injected using dynamic clamp in current-clamp mode–by calculating the total number of Na+ ions that must be actively extruded at the expense of 1 ATP molecule for every three Na+ ions [1, 4]. When the postsynaptic conductance trace derived from presynaptic stimulation was not scaled (but noise from the corticocortical and thalamocortical inputs was injected), the postsynaptic energy use was very slightly reduced by ~4% with respect to that occurring with real presynaptic stimulation in voltage-clamp mode (p = 0.02; Fig 5B). This is because, in current-clamp mode, depolarisation of the cell by the injected conductances and any action potentials that are evoked reduce slightly the Na+ entry through postsynaptic channels compared to the stimulation condition in which the cell is voltage clamped at its resting potential.
We next used dynamic clamp to investigate experimentally the effects on information transmission and energy use of scaling the thalamic input conductances up or down, while keeping the corticocortical input conductance constant. Since we are seeking the optimal conductance value for all thalamocortical synapses, the values of gsyn and gtc were scaled together. This results in the thalamic noise level increasing when gsyn is increased.
As observed with the simulations (Figs 2A and 3A), the energy use calculated to be associated with gsyn increased approximately linearly with the effective postsynaptic conductance but appears to tail off as the gain increases above 2 (Fig 6A). The L4SS cell firing frequency increased with thalamocortical conductance (with gsyn and gtc scaled up or down together as described for Fig 4G), although, like for thalamic relay cells [12], this relationship begins to plateau at conductance gain values higher than 1 (Fig 6B).
Examining transfer entropy as a function of thalamocortical conductance showed that, on average, the physiological gain of the synapse is at, or very close to, maximizing information transmission from one LGN relay neuron to the cortex (Fig 6C), unlike in simulations where a modest increase in information transmission was observed when increasing the physiological level of the thalamocortical conductance (Figs 2C and 3C). This discrepancy is probably explained by the fact that, in experiments, the whole synaptic conductance had to be injected via the patch-pipette at the soma. We then assessed the relationship between information transmission and its associated energetic costs, as previously done for the thalamic relay synapse [12] and in simulations (Figs 2D, 2E, 3D and 3E).
We calculated the ratio of the information transmitted through one thalamocortical synapse to the energy used on its postsynaptic current, when dynamic clamp was employed to inject a thalamocortical conductance of different magnitudes (Fig 6D). Like for multicompartment simulations (Figs 2D, 2E, 3D and 3E), the resulting average peak energy efficiency for information transfer occurred at a conductance value close to the physiological value for the synapse. We cannot claim that the peak is precisely at the physiological value: this would require finer sampling of the conductance values around gsyn = 1, and a larger experimental dataset to justify statistical comparison between conductance gain factors. We simply wish to show that the shape of the experimental curve is comparable to the simulated results, recapitulating and supporting the predictions made by the simulations in Figs 2 and 3. Together, the simulation and experimental results suggest that, as for the much stronger retina-LGN synapse, the weak thalamocortical synapse displays an optimal value of postsynaptic conductance for maximizing the information transmitted per energy used on synaptic currents which is in the region of the physiologically observed value.
We have previously shown that information transmission is sacrificed in favour of energetic efficiency at the retinothalamic synapse between retinal ganglion cells and relay neurons in the dorsal LGN of the thalamus [12]. We wondered whether such energy efficiency is a special property of strong relay synapses like that in the LGN, where postsynaptic spiking is largely driven by a single input, or whether similar principles apply at synapses that contribute less dominantly to postsynaptic spiking. At the next synapse along the visual pathway, the thalamocortical synapse, the set of thalamic relay neurons makes up to 600 contacts with the spines of a single cortical L4 spiny stellate cell, the firing of which is also influenced by abundant cortical input. Strikingly, both simulations and experiments presented here show that, even in the presence of random background corticocortical and thalamocortical input, the ratio of information transmitted to energy used is maximised near the physiological conductance of the weak thalamocortical synapse.
Using an updated multicompartment Hodgkin-Huxley-type model of L4SS cells [15], we simulated a constant background level of excitatory and inhibitory cortical inputs, while varying the conductance strength of a set of thalamocortical inputs. We found that the information transmitted from one thalamic input was maximised when the conductance of the set of thalamic inputs was slightly larger than the physiological value (Figs 2C and 3C). However, when the information transmitted across a single thalamic input was divided by its associated postsynaptic energy use, we found a sharp peak in energy efficiency of information transmission, which was at its maximum when the set of thalamic inputs were close to their physiologically observed conductance value (Figs 2D, 2E, 3D and 3E). Thus, thalamocortical synaptic properties appear to be set to favour energy efficiency for information transmission.
To check these results experimentally, we recorded the postsynaptic current evoked in single L4SS cells in response to stimulation of a presynaptic thalamic axon. We converted this current to a conductance, which could then be scaled up or down (mimicking larger or smaller synapses) and injected into the soma of the cell using dynamic clamp. Extra thalamic and cortical conductances derived from the simulations were simultaneously injected into the L4SS cell. In the presence of this constant background level of cortical inputs, we found that the information transmitted from one thalamic neuron already is maximal in the range of the physiological gain of the synapse–increasing or decreasing the gain by the factors that we applied reduced the information transmitted (Fig 6C). Nevertheless, if information transmitted is divided by the energy used on postsynaptic ion pumping [4, 12], we found that this measure of synaptic energetic efficiency also decreased when larger or smaller experimental gain factors were applied (Fig 6D). The exact position of the peak cannot be revealed by our experimental data but, taken together with the predictions from our multicompartment simulations, it appears that–as for the retinothalamic synapse–weak thalamocortical synapses display maximal information transmitted per energy used at or near physiological conductance values.
To test for the influence of inhibition at the thalamocortical synapse, we used the mathematical model of L4SS cells and ran simulations with and without inhibitory input (see Materials & Methods). Results obtained in the Hodgkin-Huxley-type multicompartment model in the presence of inhibition recapitulate the experimental results of Fig 6C and 6D. Even though information transfer increases monotonically in the simulations (Figs 2C and 3C) while it eventually decreases when the gain is increased in experiments (Fig 6C), this did not affect the excitatory conductance magnitude which produced the peak metabolic efficiency. Removing the inhibitory input in simulations did however change slightly the synaptic conductance value for the peak metabolic efficiency, shifting it to lower values (Fig 3D and 3E), suggesting that the balance between excitation and inhibition in the network may play a role in fine-tuning synaptic conductance values for energetic optimality of information transfer. Specifically, we reason that, because increased inhibition would demand that more excitatory charge enter the postsynaptic cell to achieve the same voltage change, when inhibition does not have to be overcome, the peak position of energy efficiency shifts to a lower excitatory conductance value. The discrepancy observed between simulations (Fig 2C) and experiments (Fig 6C) in the relation between information transfer and gain, at gains higher than ~1.5, can probably be attributed to the slightly different stimulation scenarios, as in experiments we were constrained to inject all input at the cell soma.
In vivo, especially in sensory systems, it is likely that a postsynaptic neuron will receive inputs from cells that are transmitting correlated information. For instance, the receptive fields of “simple cells” in the visual cortex are thought be built from the receptive fields of spatially adjacent thalamic relay neurons [42, 43], and therefore the excitatory thalamic input to a cortical cell may be correlated not just in amplitude (as we have mimicked here), but also in time, as a single object passes through the visual field. Additionally, inhibitory input may be correlated with excitatory input. In assessing how synaptic conductances are set to regulate information transfer and energy use, it will be interesting to investigate whether the conductance of each input to a postsynaptic cell is set independently, or whether account is taken of the correlations in information passing through spatially adjacent cells.
Temporal correlations are also seen in the form of oscillatory activity throughout the thalamocortical loop in different states of arousal [46]. This activity may be important for putting the visual scene into context, for attention, or even for conscious “binding” of different attributes of a visual scene [47]. In the present paper, we have focused on the simple feed-forward connection from thalamus to cortex, but of course a feed-forward input could arrive at any stage of such temporal oscillation, thus impinging on a more or less hyperpolarized cortical neuron. It would be interesting to investigate how such network-level membrane potential rhythms affect the energy efficiency of input connections. However, we do not find it necessary to postulate that the same energetic principles should hold for all oscillation conditions. For instance, it is possible that the high amplitude, low frequency thalamocortical oscillations that are seen in deep sleep may play a role in synaptic renormalization, perhaps actively regulating synaptic strength with energy efficiency a prioritized parameter. Note that detailed investigation of these questions in the future might require the use of conditional transfer entropy.
How can the preference of single synapses for an optimal information-to-energy ratio be reconciled with the need for strongly varying synaptic strengths for learning? This is a question that we are very intrigued by, and our best explanation is that synapses are efficient on average. If each synapse were always set to its energetic optimum, then how could the relative weights of synapses store information? From what is known about experience-driven and homeostatic mechanisms of synaptic plasticity, it is clear that we cannot consider energy to be the only determinant of neuronal organisation. Nor is information the only alternative competing pressure. For instance, brain-wide activity as a whole must be protected from entering dangerous regimes of recurrent excitation, and such stable network dynamics have recently been shown to be prioritised above optimal information transmission [48]. Many organisational constraints need to be considered to gain a full understanding of the evolutionary pressures that have guided brain design. We do not claim to have considered all such constraints here, but simply argue that energetic efficiency appears to be one important pressure on the strength of any synapse.
Both our previous work (retinogeniculate synapse [12]) and the present study (thalamocortical synapse), focus on relatively low-level feed-forward synapses in the adult visual system, examined after the critical period of development. We would predict that the distribution of synaptic strengths would be broader in a highly plastic brain region, or during development, with fewer synapses sitting at the energetically optimal point at any one time.
We have found that postsynaptic properties maximise the information transmitted per energy used. While these properties are of course the result of evolutionary selection, how do we know that energy efficiency was the selective pressure for these properties? That the evolved synapse properties facilitate energetically efficient information transfer could be a happy coincidence. One way to investigate this question would be examine the energy efficiency of synaptic communication in phylogenetically related species, to see if efficiency improves on an evolutionary time scale.
Looking beyond synapses in the mammalian brain, there is evidence that many levels of neural organisation adhere to a principle of minimising energy use. At the most macroscopic level, the organisation of brain area localisation has been shaped by energy constraints by minimising the length of “wiring” required for signals to travel from one neuron to another [49]. The segregation of the brain into grey and white matter areas has also been suggested to promote energy efficiency [50]. At the single-cell level, surprising parameters such as the sparse firing [10] and low release probability [4] of many cortical neurons can also be explained when energy is viewed as a limiting factor. Even the dimensions of synaptic boutons and axons and the timescale of neuronal computation have apparently been shaped by the pressure to transmit information reliably given a limited energy supply [51, 52]. It therefore does not seem surprising that the postsynaptic site, where most energy in the brain is used [1, 2], would also show evidence of evolving within strict energetic constraints.
In conclusion, we have shown that weak, as well as strong, synapses in the visual system operate in the region of optimal information transmitted per energy used. This is consistent with energy use being a major constraint on the evolution of the CNS [53–55].
|
10.1371/journal.pntd.0005883 | Decoding the similarities and differences among mycobacterial species | Mycobacteriaceae comprises pathogenic species such as Mycobacterium tuberculosis, M. leprae and M. abscessus, as well as non-pathogenic species, for example, M. smegmatis and M. thermoresistibile. Genome comparison and annotation studies provide insights into genome evolutionary relatedness, identify unique and pathogenicity-related genes in each species, and explore new targets that could be used for developing new diagnostics and therapeutics. Here, we present a comparative analysis of ten-mycobacterial genomes with the objective of identifying similarities and differences between pathogenic and non-pathogenic species. We identified 1080 core orthologous clusters that were enriched in proteins involved in amino acid and purine/pyrimidine biosynthetic pathways, DNA-related processes (replication, transcription, recombination and repair), RNA-methylation and modification, and cell-wall polysaccharide biosynthetic pathways. For their pathogenicity and survival in the host cell, pathogenic species have gained specific sets of genes involved in repair and protection of their genomic DNA. M. leprae is of special interest owing to its smallest genome (1600 genes and ~1300 psuedogenes), yet poor genome annotation. More than 75% of the pseudogenes were found to have a functional ortholog in the other mycobacterial genomes and belong to protein families such as transferases, oxidoreductases and hydrolases.
| Members of the Mycobacteriaceae family, which are known to adapt to different environmental niches, comprise bacterial species with varied genome sizes. They are unique in their cell-wall composition, which is remarkably thick and lipid-rich as compared to other bacteria. We performed a comparative analysis at the proteome level for ten mycobacterial species that differ in their pathogenicity, genome size and environmental niches. A total of 1080 orthologous clusters with representation from all ten species were obtained, and these were further examined for their domain annotations, domain architecture similarities and enriched GO terms. These core orthologous clusters are enriched in various biosynthetic pathways. The proteins that are specific to each of the ten species were also investigated for their GO functions. The M. leprae genome has a large number of pseudogenes and we searched for their functional orthologs in other mycobacterial species in order to understand the functions that are lost from the M. leprae genome. The proteins present exclusively in M. leprae genome were studied in more detail, in order to predict putative drug targets and diagnostic markers. These findings, which have implications in understanding evolution of mycobacterial genomes, identify species-specific proteins that have potential for use in developing new diagnostic tools and therapeutics.
| Mycobacteriacea are known etiological agents for a variety of human infections and are broadly classified as Mycobacterium tuberculosis (M. tuberculosis) complex (MTBC) and Non-Tuberculous Mycobacteria (NTM). The MTBC includes several pathogenic species including M. tuberculosis that causes tuberculosis (TB) in ~10.4 million people across the globe each year. In the year 2015, ~1.4 million deaths were reported due to TB and additionally, 0.4 million deaths occurred as a result of TB infection in HIV patients [1]. Other obligate intracellular pathogenic species include M. leprae that causes leprosy in ~200,000 people annually and is mainly confined to endemic countries in the tropical zones [2]. NTMs on the other hand cause opportunistic infections and are a growing concern for a plethora of varied atypical systemic infections [3]. Currently there are more than 140 species of NTMs, some of which lead to pulmonary diseases, otitis media, osteomyelitis, lymphadenitis and skin and soft tissue infections (SSTIs) in humans [4]. One of the NTM that deserves a specific mention is a free-living rapidly growing species, M. abscessus, which is regarded as a new antibiotic nightmare that causes opportunistic infections in patients with cystic fibrosis or chronic pulmonary disease, and/or skin and soft-tissue infection [5].
The availability of the genome sequence data for several mycobacterial species, together with a variety of bioinformatics software and methods for genome analysis, makes it feasible for researchers to annotate genomes and collate information related to evolutionary traits, sequence homology, conserved regions, domain architecture, structural properties of gene products and gene ontology (GO) content. Comparative functional annotation of proteins from the genome sequencing data for pathogenic and non-pathogenic mycobacterial species can provide information related to phylogeny, frequency & distribution of orthologous protein clusters (clusters of gene families obtained from sequence comparisons of multiple species that usually reflect common functions), overlap between functional networks and species specific unique gene products [6]. This information is vital for identifying potential drug targets and unique regions/gene products that provide opportunities for developing effective diagnostic tools with considerable sensitivity and specificity.
The resulting mycobacterial genome annotations also provide an extremely useful resource for understanding strain variation and pathogenicity. The emergence of multidrug resistant and extremely drug-resistant strains underlines the need to understand orthologous genes and to identify potentially druggable targets. Earlier attempts to compare mycobacterial genomes provided information about pairwise whole-genome similarities and their predicted proteomes [7]. Since the determination of the complete genome sequence of M. tuberculosis, there have been efforts to develop inventories that record information on open reading frames (ORFs) annotations and gene expression [8–11], drug resistance mutations and drug targets [12–15], phylogenetic relationships [16,17], pathogenomics, and structure and function annotation of the mycobacterial genome [18,19].
In the current study, we have chosen ten different species—M. tuberculosis, M. abscessus, M. leprae, M. marinum, M. avium, M. kansasii, M. thermoresistible, M. smegmatis, M. ulcerans and M. vanbaalenii—for comparative analysis of genomes and protein functions. The set being investigated encompasses pathogens, opportunist pathogens and non-pathogenic species. Here we describe the mapping of orthologous clusters across the species in terms of their gene products to identify conserved regions and species-specific unique proteins from the predicted proteomes. Further, phylogenetic linkages are defined and GO annotations assessed to identify functional similarities and differences between protein targets from various species.
Of these mycobacterial species, M. leprae is under-represented in most of the known mycobacterial databases and comparative genome studies. It has the smallest genome (due to reductive evolution) among known mycobacterial genomes and a limited set of predicted proteins while half of its genome is occupied by pseudogenes [20]. Here we describe the search for functional orthologs of the pseudogenes in other mycobacterial species, to gain insight into the set of functions lost from the M. leprae genome. M. leprae has a genome size of 3,268,210 bp with only ~1600 genes, of which 22% are hypothetical proteins with unknown functions [20]. This poor annotation is also reflected by just a handful of solved protein structures (13 structures as of 23rd April 2017) in the Protein DataBank (PDB). Further, it poses clinical challenges as it has a very long generation time of 14 days, is an unculturable pathogen that lacks reliable and specific molecular markers for diagnosis of the disease. Here we report a study of the unique proteins present in the genome of M. leprae for their GO functions, subcellular and transmembrane localization, gene expression profiles from a GEO dataset, essentiality, virulence and the presence of human orthologs.
The ten mycobacterial genomes (Table 1) investigated in the present study were downloaded from UniProt [21]. These mycobacterial species cover different genome sizes and environmental niches. In the pathogenic group, we included the most virulent mycobacteria [22]: M. tuberculosis, M. leprae, M. marinum (infects broader variety of hosts and causes lesions characterized by granulomas) and M. ulcerans (causes third-most common mycobacterial disease after tuberculosis and leprosy). The opportunist pathogenic group includes NTMs that cause pulmonary and other peripheral infections in immunocompromised individuals. These definitions are adapted from an earlier comparative study of metabolic pathways of the mycobacterial species [22].
Orthologs were identified in the ten species using ProteinOrthov5 [23]. Sets of orthologs that are shared across all species and between a given pair of species were identified. The remaining sets of proteins from each species that failed to identify an ortholog in any other nine species were marked as species-specific proteins. The clusters of orthologs that have representation from all the ten species are called core orthologs (gene families present in all ten species).
For all the orthologous clusters identified, irrespective of the number of genes and species, we looked at the domain composition and architectural similarities in order to identify the functional similarities at the genome level.
For calculating the similarity in domain composition, for a given cluster, all pairwise orthologs were considered and were assigned Pfam domains using hmmscan from the HMMER3 package [24] and Pfam v30 database [25] at an E-value threshold of 10−3. For each of these pairs having representation in two species, we then calculated the fraction of shared domains, known as the domain composition similarity score (DCS, Eq 1), which can range from 0 to 1, where 1 indicates that the given pair of orthologs has exactly the same domain composition and a score close to zero reflects poor similarity in domain composition. If there are in-paralogs in the orthologous cluster, then the presence of a Pfam domain in at least one of these is sufficient to be included in the count as a shared domain.
DCS= sd12N
(1)
where sd12 is the number of shared domains between protein p1 and p2 and N is the total number of non-redundant domains in p1 and p2.
The second level of similarity is more stringent as it considers both the order and the content of domains and is called the domain architecture similarity score (DAS, Eq 2). DAS, adapted from Forslund et al. [26], is calculated for each pair in the orthologous cluster and considers the number of identical aligned Pfam domains compared to the total number of domains in the pair.
DAS= al12N
(2)
where al12 is the number of domains that are aligned with an identical domain in two given proteins p1 and p2 and N is the total number of domains in the two given proteins.
We mapped both core orthologs and species-specific sets of proteins to GO terms in all the three domains (biological process, molecular function and cellular component) by considering the following evidence codes as reliable: IMP: Inferred by Mutant Phenotype, IGI: Inferred by Genetic Interaction, IPI: Inferred by Physical Interaction, IDA: Inferred by Direct Assay, IEP: Inferred by Expression Pattern, ISS: Inferred by Structure/Sequence Similarity, TAS: Traceable Author Statement and IC: Inferred by Curation.
In order to reduce the number of GO terms and map them to broader categories, we used GOSlimViewer from AgBase [27] to map the set of proteins to GO Slim terms (broader versions of the GO ontologies that provide a summary of results of GO annotation).
To identify the GO terms enriched in a specific subset of interest, hypergeometric probabilities were calculated as:
P= CxMCn−xN−MCnN
(3)
where M is the total number of GO terms in the subset, N is total number of GO terms full set, n and x is the occurrence of a GO term of interest in the full set and the subset respectively. To identify significantly enriched GO terms in the core orthologous set, p-values were calculated using hypergeometric distribution. The GO terms were considered enriched if the p-values were less than 0.05.
We used the nucleotide sequence of 1320 pseudogenes present in M. leprae and performed BLASTX against the remaining nine-mycobacterial proteomes to determine whether there is a functional ortholog present. To gain insights into the functions of these lost genes, we mapped the functional orthologs of pseudogenes identified in the M. tuberculosis genome to the protein families
The species-specific proteins were further mapped to their GO functions to identify the enriched GO functions for all ten species using Eq 3. The M. leprae species-specific proteins were studied in detail in order to explore their potential to be used either as diagnostic markers or new drug targets. The linear B-cell epitopes for these specific proteins using BepiPred (at a threshold of 0.35) and selected the ones that are between 10–30 amino acids in length.
The genome sizes of the two obligate pathogenic species, M. tuberculosis and M. leprae, are smaller than those of the free living non-pathogenic and opportunist pathogenic genomes (Fig 1). This is in agreement with previous observations of genome reduction and loss of genes when free-living bacteria adapt to an obligate pathogenic lifestyle [28]. However, M. marinum, a pathogen that can cause tuberculosis-like infections in aquatic organisms (fishes and amphibians) and can also cause peripheral disease characterized by granulomas in humans, has retained a higher genome size and a larger number of genes. This can be explained by its ability to infect broader range of hosts and its capacity to survive outside the host. Also, its genome is reported to have large number of polyketide synthases and non-ribosomal peptide synthases, PE and PPE proteins, secretion system proteins [29].
We identified 6983 orthologous clusters that have representation from at least two of the ten species. Of these 6013 were single gene clusters (one-to-one orthologs), whereas remaining 970 clusters had in-paralogs. There were 1080 clusters that have representation of all the species and the proteins forming these clusters are labeled as core orthologs (Fig 2).
The orthologous clusters shared between any two-mycobacterial genomes were also recorded in order to identify closely related mycobacterial genomes (S1 Table). Among the pathogenic species, M. tuberculosis, M. leprae and M. ulcerans share the maximum number of orthologous pairs with another pathogenic species M. marinum, which has a remarkably larger genome (as discussed above). Consistent with its ability to live without the host, M. marinum shares maximum similarity with an opportunist pathogenic species M. kansasii. For the non-pathogenic species, the maximum similarity was shared within the other non-pathogenic species. The opportunist pathogenic species M. abscessus is observed to share maximum similarity with a free-living non-pathogenic species M. smegmatis. These observations correlate with the environmental niches of these species and that they have preserved the higher number of genes and have also acquired genes through horizontal gene transfer unlike pathogenic species, which have adopted an evolutionary route to minimalism (and genome reduction) to maintain their growth efficiency and competitiveness inside the host.
As M. leprae has the smallest genome with only 1600 protein coding genes, we excluded the genome of M. leprae and then repeated the ortholog identification step for the remaining nine species (S1 Fig). Although this was not observed to increase the number of orthologous clusters, the number of core orthologs (gene families present in nine mycobacterial species) increased by 40% (10,910 proteins from all ten species vs. 15,043 proteins from all nine species- excluding M. leprae).
In order to understand and explore the molecular and structural biology of the drug targets for pathogenic mycobacterial species such as M. tuberculosis and M. abscessus, non-pathogenic species (M. smegmatis and M.thermoresistibile) are usually used as surrogate systems and models in the lab [43,44]. This enables researchers to work with non-infectious strains on the bench and also M. thermoresistibile proteins can tolerate higher temperatures than the M. tuberculosis proteins and on average are more soluble [44].
To investigate the suitability of using non-pathogenic species as surrogate systems for pathogenic species, we checked the similarity of the orthologous pairs between non-pathogenic and pathogenic species (S4 Fig). The orthologs of pathogenic species (M. ulcerans, M. tuberculosis, M. leprae and M. marinum, S4A and S4B Fig in red) present in M. smegmatis and M. thermoresistibile genomes were observed to share more than 70% average percent identity. But for the opportunist pathogens (M. abscessus, M. avium and M. kansasii, S4A and S4B Fig in black), the distribution of percent identity is much wider and median is below 70% indicating that non-pathogenic species M. smegmatis and M. thermoresistibile are more suitable surrogate models for studying the proteins of pathogenic species such as M. tuberculosis.
M. leprae has adapted to become an obligate pathogen and its genome has undergone a huge reduction to only 1600 protein-coding genes and large number of pseudogenes (1320) [20,45]. We inspected the other nine-mycobacterial genomes for the presence of functional orthologs of these pseudogenes in order to gain insight into what functions have been lost from the M. leprae genome during the process of genome reduction.
More than 75% of the pseudogenes were found to have a functional ortholog in the genomes of pathogenic and opportunist pathogenic species (except for M. abscessus, which had an ortholog for 61% of the pseudogenes, Fig 5A). However, the fraction of pseudogenes having a functional ortholog in non-pathogenic species was around 70% (M. smegmatis- 70%, M. thermoresistible- 64% and M. vanbaalenii- 71%).
Upon mapping the functional orthologs of pseudogenes identified in the M. tuberculosis genome to the protein families (Fig 5B), we noted that these were mainly associated with the catabolic functions such as transferases (including acetyltransferase, acyltransferase, methytransferase, transaldolase, transketolase, transaminase), oxidoreductases (including dehydrogenase and peroxidase) and hydrolases (including lipase, amylase, protease, phosphatase) thereby limiting the availability of usable energy source for M. leprae to grow. This is consistent with other studies, where they have analyzed the genome reduction and loss of functions in M. leprae genome [20,46,47]. As mentioned earlier, the proteins for all major biosynthetic pathways are fairly conserved between all mycobacterial species (as they are found in core orthologous clusters) but the energy metabolism genes appear to be more tuned to different species needs as they have evolved to survive in specific environmental niches with different growth rates.
We also looked at the enriched GO functions in species-specific proteins for all ten species (S3–S5 Tables). In the pathogenic species, we noticed that the genomes of M. tuberculosis and M. marinum were enriched in functions that are involved in DNA metabolism such as DNA recombination, DNA repair, DNA integration and protection (S3 Table). This supports the observation that DNA repair mechanisms are active throughout the course of tuberculosis infection as observed in infection models and clinical samples [48]. Once these pathogens infect the host, they need to survive the hostile environments inside the host cells and hence DNA repair and recombination are required to preserve the integrity of their genomes. Apart from surviving in the host-cell environment during the various stages of the infection cycle, there is a need for DNA repair and recombination mechanisms to preserve the genome during dormant phases of infection [49,50].
In the set of opportunistic pathogens (M. abscessus, M. avium and M. kansasii) specific genes, functions and processes associated with membrane transport such as the ATP-binding cassette transporter complex, high-affinity iron permease complex and oxidoreductase activity are enriched (S4 Table). The fact that the genome of M abscessus is known to code for many drug-efflux proteins such as ATP-binding cassette transporters and MmPL proteins [51,52] is consistent with its observed multidrug resistance. Furthermore, as these are free-living bacteria, the presence of enriched and active transport systems helps their survival through uptake of nutrients and acquisition of genes via horizontal gene transfer.
Diagnosing and treating M. leprae infections remain huge challenges due to its slow growth rate, lack of specific and reliable clinical markers and emerging drug resistance. We have therefore studied this genome in detail to identify the gene(s) that are specific to its genome, in order to identify and propose genes that can be further tested and validated for use as diagnostic markers and/or drug targets [53].
While comparing the genes of the ten-mycobacterial species, 141 M. leprae proteins were identified that lack a homolog in any of the other nine species. We further screened these 141 proteins for homologs against other mycobacterial genomes (from NCBI), and identified 86 M. leprae proteins that lack an ortholog in any other mycobacterial genomes (S5 Fig). Firstly, we scanned these 86 M. leprae specific proteins for their GO functions, transmembrane regions, presence of a human ortholog, and for their predicted essentiality, using Flux-balance analysis, from PATRIC database [10,11]) and virulence (S6 Table). Interestingly, none of the 86 had a human ortholog nor was predicted to be essential or involved in virulence. As these essentiality predictions are based on only flux-balance analysis, it would be interesting to design experiments for testing their essentiality for M. leprae.
We studied these M. leprae-specific proteins for their potential to be used either as diagnostic markers or new drug targets. Therefore, synthetic peptides presenting these epitopes could be used to raise antibodies, which can be used to detect the specific protein in a diagnostic test. Predicting the linear antigenic determinants is usually an initial step to determine antigenicity followed by prediction of non-continuous or conformational epitopes that are generally linear epitopes that are in close structural proximity upon folding. We mapped the linear B-cell epitopes for these specific proteins using BepiPred [54] (S7 Table). This is an initial step towards a search for B-cell antigens in the genome of M. leprae, which would aid to identify the antigenic determinants in this pathogenic mycobacterial species. The linear B-cell epitopes are predicted using the sequence information only; hence these provide a feasible way to run predictions at the genome-wide level, however, their accuracy rates are only about ~60–70% and more experimental validations are required to test these predictions. We could find 127 antigenic determinants in 69 of these specific proteins, which might serve as a good starting point for further experimental validations and developing diagnosis tools for leprosy (S7 Table).
In the species-specific M. leprae proteins, we investigated in detail ML2177c, which encodes for a probable uridine nucleoside phosphorylase (an important enzyme in the salvage pathway for nucleotide synthesis). This enzyme is of interest due to the following observations: a. availability of a suitable structural homolog to model the structure; b. lack of a uridine phosphorylase in M. tuberculosis genome [55–57], hence it can be specific for leprosy infection; c. known immunogenicity in both animal models and infected humans [58], which might aid in diagnosis of leprosy infection; and d. it is already being explored as a drug target for other bacterial infections such as Salmonella typhimurium [59]. We have also performed a transcriptomics analysis, to check the expression levels of ML2177c in patients (n = 3) with M. leprae infection from endemic regions. We measured the fold change in the expression levels of ML2177c as compared to the basal level of expression (16S rRNA). For two of the samples, the change of expression was two-fold and for one of the samples four-fold, indicating ML2177c is significantly expressed during leprosy infection. We also observed that ML2177c is conserved in strains of M. leprae other than TN (namely Br4923, NHDP63 and Thai).
We inspected ML2177c for its druggability by predicting the hotspots in the protein structure model. We first modeled the protein structures using our in-house automated modeling pipeline (Vivace) [18], followed by prediction of the druggable sites using our software for fragment hotspot mapping (Fig 6A) [60], which provides insights into the ligand binding site for the target. Using the known oligomeric structure for uridine phosphorylase for Shewanella oneidensis (PDB ID: 4R2X, 30% identity with ML2177c) as a template, we modeled the hexameric complex for ML2177c. The inhibitor 2,2'-anhydrouridine was modeled into the hexameric structure using the S. typhimurium structure (PDBID: 3FWP). The fragment hotspot maps to the dimeric interface of the modeled structure and superposes with the inhibitor-binding site, hence suggesting the druggability of ML2177c (Fig 6B).
We believe that the comparative genomic studies provide insights into understanding the common mechanisms of mycobacterial pathogenesis, including pathways and functions conserved across different species. Also examining these different mycobacterial genomes for specific proteins should help distinguish the mycobacterial infection, as well as aid in designing new therapeutics and in testing some for their use in developing diagnostic kits for specific mycobacterial infections.
Here, we have highlighted functions both common and specific to different mycobacterial species. Interestingly, the drug targets predicted for M. tuberculosis were found to have orthologs in other mycobacterial genomes, suggesting their suitability as a drug target for treating other mycobacterial infections.
In our opinion, it is of value to explore the large number of pseudogenes that are retained in the M. leprae genome in more detail. Their orthologs in M. tuberculosis are reported to be non-essential but a significant proportion of these, ~43%, are observed to be expressed at different levels during different stages of disease progression [61,62]. However, the expressed pseudogenes are observed to have altered ORFs because of the large number of stop codons, the lack of start codons and their presence usually towards the end of the 3’ end of the operon. As essential and functional genes tend to be present towards the 5’ end, this appears to be an example of position-dependence of functionally significant genes [63]. The sequence comparisons of these pseudogenes in different strains of M. leprae reveal that some of these pseudogenes are strain specific, possibly implicating their role in generating genetic diversity, but more likely a reflection of selectively neutral evolution. Regarding their functions, it is proposed that they play important roles in regulation of gene expression at both transcriptional and post-transcriptional level, serve back-up functions and can be activated by phenomena such as gene conversion, regulate replication rates and rate of infection [20]. More detailed analysis on the pseudogenes in mycobacterial genomes will shed light into evolution and understanding their role during stages of infection.
|
10.1371/journal.pcbi.1002283 | Chemotaxis when Bacteria Remember: Drift versus Diffusion | Escherichia coli (E. coli) bacteria govern their trajectories by switching between running and tumbling modes as a function of the nutrient concentration they experienced in the past. At short time one observes a drift of the bacterial population, while at long time one observes accumulation in high-nutrient regions. Recent work has viewed chemotaxis as a compromise between drift toward favorable regions and accumulation in favorable regions. A number of earlier studies assume that a bacterium resets its memory at tumbles – a fact not borne out by experiment – and make use of approximate coarse-grained descriptions. Here, we revisit the problem of chemotaxis without resorting to any memory resets. We find that when bacteria respond to the environment in a non-adaptive manner, chemotaxis is generally dominated by diffusion, whereas when bacteria respond in an adaptive manner, chemotaxis is dominated by a bias in the motion. In the adaptive case, favorable drift occurs together with favorable accumulation. We derive our results from detailed simulations and a variety of analytical arguments. In particular, we introduce a new coarse-grained description of chemotaxis as biased diffusion, and we discuss the way it departs from older coarse-grained descriptions.
| The chemotaxis of Escherichia coli is a prototypical model of navigational strategy. The bacterium maneuvers by switching between near-straight motion, termed runs, and tumbles which reorient its direction. To reach regions of high nutrient concentration, the run-durations are modulated according to the nutrient concentration experienced in recent past. This navigational strategy is quite general, in that the mathematical description of these modulations also accounts for the active motility of C. elegans and for thermotaxis in Escherichia coli. Recent studies have pointed to a possible incompatibility between reaching regions of high nutrient concentration quickly and staying there at long times. We use numerical investigations and analytical arguments to reexamine navigational strategy in bacteria. We show that, by accounting properly for the full memory of the bacterium, this paradox is resolved. Our work clarifies the mechanism that underlies chemotaxis and indicates that chemotactic navigation in wild-type bacteria is controlled by drift while in some mutant bacteria it is controlled by a modulation of the diffusion. We also propose a new set of effective, large-scale equations which describe bacterial chemotactic navigation. Our description is significantly different from previous ones, as it results from a conceptually different coarse-graining procedure.
| The bacterium E. coli moves by switching between two types of motions, termed ‘run’ and ‘tumble’ [1]. Each results from a distinct movement of the flagella. During a run, flagella motors rotate counter-clockwise (when looking at the bacteria from the back), inducing an almost constant forward velocity of about , along a near-straight line. In an environment with uniform nutrient concentration, run durations are distributed exponentially with a mean value of about [2]. When motors turn clockwise, the bacterium undergoes a tumble, during which, to a good approximation, it does not translate but instead changes its direction randomly. In a uniform nutrient-concentration profile, the tumble duration is also distributed exponentially but with a much shorter mean value of about [3].
When the nutrient (or, more generally, chemoattractant) concentration varies in space, bacteria tend to accumulate in regions of high concentration (or, equivalently, the bacteria can also be repelled by chemorepellants and tend to accumulate in low chemical concentration) [4]. This is achieved through a modulation of the run durations. The biochemical pathway that controls flagella dynamics is well understood [1], [5]–[7] and the stochastic ‘algorithm’ which governs the behavior of a single motor is experimentally measured. The latter is routinely used as a model for the motion of a bacteria with many motors [1], [8]–[11]. This algorithm represents the motion of the bacterium as a non-Markovian random walker whose stochastic run durations are modulated via a memory kernel, shown in Fig. 1. Loosely speaking, the kernel compares the nutrient concentration experienced in the recent past with that experienced in the more distant past. If the difference is positive, the run duration is extended; if it is negative, the run duration is shortened.
In a complex medium bacterial navigation involves further complications; for example, interactions among the bacteria, and degradations or other dynamical variations in the chemical environment. These often give rise to interesting collective behavior such as pattern formation [12], [13]. However, in an attempt to understand collective behavior, it is imperative to first have at hand a clear picture of the behavior of a single bacterium in an inhomogeneous chemical environment. We are concerned with this narrower question in the present work.
Recent theoretical studies of single-bacterium behavior have shown that a simple connection between the stochastic algorithm of motion and the average chemotactic response is far from obvious [8]–[11]. In particular, it appeared that favorable chemotactic drift could not be reconciled with favorable accumulation at long times, and chemotaxis was viewed as resulting from a compromise between the two [11]. The optimal nature of this compromise in bacterial chemotaxis was examined in Ref. [10]. In various approximations, while the negative part of the response kernel was key to favorable accumulation in the steady state, it suppressed the drift velocity. Conversely, the positive part of the response kernel enhanced the drift velocity but reduced the magnitude of the chemotactic response in the steady state.
Here, we carry out a detailed study of the chemotactic behavior of a single bacterium in one dimension. We find that, for an ‘adaptive’ response kernel (i.e., when the positive and negative parts of the response kernel have equal weight such that the total area under the curve vanishes), there is no incompatibility between a strong steady-state chemotaxis and a large drift velocity. A strong steady-state chemotaxis occurs when the positive peak of the response kernel occurs at a time much smaller than and the negative peak at a time much larger than , in line with experimental observation. Moreover, we obtain that the drift velocity is also large in this case. For a general ‘non-adaptive’ response kernel (i.e., when the area under the response kernel curve is non-vanishing), however, we find that a large drift velocity indeed opposes chemotaxis. Our calculations show that, in this case, a position-dependent diffusivity is responsible for chemotactic accumulation.
In order to explain our numerical results, we propose a simple coarse-grained model which describes the bacterium as a biased random walker with a drift velocity and diffusivity, both of which are, in general, position-dependent. This simple model yields good agreement with results of detailed simulations. We emphasize that our model is distinct from existing coarse-grained descriptions of E. coli chemotaxis [13]–[16]. In these, coarse-graining was performed over left- and right-moving bacteria separately, after which the two resulting coarse-grained quantities were then added to obtain an equation for the total coarse-grained density. We point out why such approaches can fail and discuss the differences between earlier models and the present coarse-grained model.
Following earlier studies of chemotaxis [9], [17], we model the navigational behavior of a bacterium by a stochastic law of motion with Poissonian run durations. A switch from run to tumble occurs during the small time interval between and with a probability(1)Here, and is a functional of the chemical concentration, , experienced by the bacterium at times . In shallow nutrient gradients, the functional can be written as(2)The response kernel, , encodes the action of the biochemical machinery that processes input signals from the environment. Measurements of the change in the rotational bias of a flagellar motor in wild-type bacteria, in response to instantaneous chemoattractant pulses were reported in Refs. [17], [18]; experiments were carried out with a tethering assay. The response kernel obtained from these measurements has a bimodal shape, with a positive peak around and a negative peak around (see Fig. 1). The negative lobe is shallower than the positive one and extends up to , beyond which it vanishes. The total area under the response curve is close to zero. As in other studies of E. coli chemotaxis, we take this response kernel to describe the modulation of run duration of swimming bacteria [8]–[11]. Recent experiments suggest that tumble durations are not modulated by the chemical environment and that as long as tumbles last long enough to allow for the reorientation of the cell, bacteria can perform chemotaxis successfully [19], [20].
The model defined by Eqs. 1 and 2 is linear. Early experiments pointed to a non-linear, in effect a threshold-linear, behavior of a bacterium in response to chemotactic inputs [17], [18]. In these studies, a bacterium modulated its motion in response to a positive chemoattractant gradient, but not to a negative one. In the language of present model, such a threshold-linear response entails replacing the functional defined in Eq. 2 by zero whenever the integral is negative. More recent experiments suggest a different picture, in which a non-linear response is expected only for a strong input signal whereas the response to weak chemoattractant gradient is well described by a linear relation [21]. Here, we present an analysis of the linear model. For the sake of completeness, in Text S1, we present a discussion of models which include tumble modulations and a non-linear response kernel. Although recent experiments have ruled out the existence of both these effects in E.coli chemotaxis, in general such effects can be relevant to other systems with similar forms of the response function.
The shape of the response function hints to a simple mechanism for the bacterium to reach regions with high nutrient concentration. The bilobe kernel measures a temporal gradient of the nutrient concentration. According to Eq. 1, if the gradient is positive, runs are extended; if it is negative, runs are unmodulated. However, recent literature [8], [9], [11] has pointed out that the connection between this simple picture and a detailed quantitative analysis is tenuous. For example, de Gennes used Eqs. 1 to calculate the chemotactic drift velocity of bacteria [8]. He found that a singular kernel, , where is a Dirac function and a positive constant, lead to a mean velocity in the direction of increasing nutrient concentration even when bacteria are memoryless (). Moreover, any addition of a negative contribution to the response kernel, as seen in experiments (see Fig. 1), lowered the drift velocity. Other studies considered the steady-state density profile of bacteria in a container with closed walls, both in an approximation in which correlations between run durations and probability density were ignored [11] and in an approximation in which the memory of the bacterium was reset at run-to-tumble switches [9]. Both these studies found that, in the steady state, a negative contribution to the response function was mandatory for bacteria to accumulate in regions of high nutrient concentration. These results seem to imply that the joint requirement of favorable transient drift and steady-state accumulation is problematic. The paradox was further complicated by the observation [9] that the steady-state single-bacterium probability density was sensitive to the precise shape of the kernel: when the negative part of the kernel was located far beyond it had little influence on the steady-state distribution [11]. In fact, for kernels similar to the experimental one, model bacteria accumulated in regions with low nutrient concentration in the steady state [9].
In order to resolve these paradoxes and to better understand the mechanism that leads to favorable accumulation of bacteria, we perform careful numerical studies of bacterial motion in one dimension. In conformity with experimental observations [17], [18], we do not make any assumption of memory reset at run-to-tumble switches.
We model a bacterium as a one-dimensional non-Markovian random walker. The walker can move either to the left or to the right with a fixed speed, , or it can tumble at a given position before initiating a new run. In the main paper, we present results only for the case of instantaneous tumbling with , while results for non-vanishing are discussed in Text S1. There, we verify that for an adaptive response kernel does not have any effect on the steady-state density profile. For a non-adaptive response kernel, the correction in the steady-state slope due to finite is small and proportional to .
The run durations are Poissonian and the tumble probability is given by Eq. 1. The probability to change the run direction after a tumble is assumed to have a fixed value, , which we treat as a parameter. The specific choice of the value of does not affect our broad conclusions. We find that, as long as , only certain detailed quantitative aspects of our numerical results depend on . (See Text S1 for details on this point.) We assume that bacteria are in a box of size with reflecting walls and that they do not interact among each other. We focus on the steady-state behavior of a population. Reflecting boundary conditions are a simplification of the actual behavior [22], [23]; as long as the total ‘probability current’ (see discussion below) in the steady state vanishes, our results remain valid even if the walls are not reflecting.
As a way to probe chemotactic accumulation, we consider a linear concentration profile of nutrient: . We work in a weak gradient limit, i.e., the value of is chosen to be sufficiently small to allow for a linear response. Throughout, we use in our numerics. From the linearity of the problem, results for a different attractant gradient, , can be obtained from our results through a scaling factor . In the linear reigme, we obtain a spatially linear steady-state distribution of individual bacterium positions, or, equivalently, a linear density profile of a bacterial population. Its slope, which we denote by , is a measure of the strength of chemotaxis. A large slope indicates strong bacterial preference for regions with higher nutrient concentration. Conversely, a vanishing slope implies that bacteria are insensitive to the gradient of nutrient concentration and are equally likely to be anywhere along the line. We would like to understand the way in which the slope depends on the different time scales present in the system.
In order to gain insight into our numerical results, we developed a simple coarse-grained model of chemotaxis. For the sake of simplicity, we first present the model for a non-adaptive, singular response kernel, , and, subsequently, we generalize the model to adaptive response kernels by making use of linear superposition.
The memory trace embodied by the response kernel induces temporal correlations in the trajectory of the bacterium. However, if we consider the coarse-grained motion of the bacterium over a spatial scale that exceeds the typical run stretch and a temporal scale that exceeds the typical run duration, then we can assume that it behaves as a Markovian random walker with drift velocity and diffusivity . Since the steady-state probability distribution, , is flat for , for small we can write(4)(5)(6)Here, and . Since we are neglecting all higher order corrections in , our analysis is valid only when is sufficiently small. In particular, even when , we assume that the inequality is still satisfied. The chemotactic drift velocity, , vanishes if ; it is defined as the mean displacement per unit time of a bacterium starting a new run at a given location. Clearly, even in the steady state when the current , defined through , vanishes, may be non-vanishing (see Eq. 8 below). In general, the non-Markovian dynamics make dependent on the initial conditions. However, in the steady state this dependence is lost and can be calculated, for example, by performing a weighted average over the probability of histories of a bacterium. This is the quantity that is of interest to us. An earlier calculation by de Gennes showed that, if the memory preceding the last tumble is ignored, then for a linear profile of nutrient concentration the drift velocity is independent of position and takes the form [8]. While the calculation applies strictly in a regime with (because of memory erasure), in fact its result captures the behavior well over a wide range of parameters (see Fig. 4). To measure in our simulations, we compute the average displacement of the bacterium between two successive tumbles in the steady state, and we extract therefrom the drift velocity. (For details of the derivation, see Text S1.) We find that is negative for and that its magnitude falls off with increasing values of (Fig. 4). We also verify that indeed does not show any spatial dependence (data shown in Fig. of Text S1). We recall that, in our numerical analysis, we have used a small value of ; this results in a low value of . We show below that for an experimentally measured bilobe response kernel, obtained by superposition of singular response kernels, the magnitude of becomes larger and comparable with experimental values.
To obtain the diffusivity, , we first calculate the effective mean free path in the coarse-grained model. The tumbling frequency of a bacterium is and depends on the details of its past trajectory. In the coarse-grained model, we replace the quantity by an average over all the trajectories within the spatial resolution of the coarse-graining. Equivalently, in a population of non-interacting bacteria, the average is taken over all the bacteria contained inside a blob, and, hence, denotes the position of the center of mass of the blob at a time in the past. As mentioned above, the drift velocity is proportional to , so that . The average tumbling frequency then becomes and, consequently, the mean free path becomes . As a result, the diffusivity is expressed as . We checked this form against our numerical results (Fig. 5).
Having evaluated the drift velocity, , and the diffusivity, , we now proceed to write down the continuity equation (for a more rigorous but less intuitive approach, see [10]). For a biased random walker on a lattice, with position-dependent hopping rates and towards the right and the left, respectively, one has and , where is the lattice constant. In the continuum limit, the temporal evolution of the probability density is given by a probability current, as(7)where the current takes the form(8)For reflecting boundary condition, in the steady state. This constraint yields a steady-state slope(9)for small . We use our measured values for and (Figs. 4 and 5), and compute the slope using Eq. 9. (For details of the measurement of , see Text S1.) We compare our analytical and numerical results in Fig. 2, which exhibits close agreement.
According to Eq. 9, steady-state chemotaxis results from a competition between drift motion and diffusion. For , the drift motion is directed toward regions with a lower nutrient concentration and hence opposes chemotaxis. Diffusion is spatially dependent and becomes small for large nutrient concentrations (again for ), thus increasing the effective residence time of the bacteria in favorable regions. For large values of , the drift velocity vanishes and one has a strong chemotaxis as increases (Fig. 2). Finally, for , the calculation by de Gennes yields which exactly cancels the spatial gradient of (to linear order in ), and there is no accumulation [8], [11].
These conclusions are easily generalized to adaptive response functions. For , within the linear response regime, the effective drift velocity and diffusivity can be constructed by simple linear superposition: The drift velocity reads . Interestingly, the spatial dependence of cancels out and . The resulting slope then depends on the drift only and is calculated as(10)In this case, the coarse-grained model is a simple biased random walker with constant diffusivity. For and , the net velocity, proportional to , is positive and gives rise to a favorable chemotactic response, according to which bacteria accumulate in regions with high food concentration. Moreover, the slope increases as the separation between and grows. We emphasize that there is no incompatibility between strong steady-state chemotaxis and large drift velocity. In fact, in the case of an adaptive response function, strong chemotaxis occurs only when the drift velocity is large.
For a bilobe response kernel, approximated by a superposition of many delta functions (Fig. 1), the slope, , can be calculated similarly and in Fig. 3 we compare our calculation to the simulation results. We find close agreement in the case of a linear model with a bilobe response kernel and, in fact, also in the case of a non-linear model (see Text S1).
The experimental bilobe response kernel is a smooth function, rather than a finite sum of singular kernels over a set of discrete values (as in Fig. 1). Formally, we integrate singular kernels over a continuous range of to obtain a smooth response kernel. If we then integrate the expression for the drift velocity obtained by de Gennes, according to this procedure, we find an overall drift velocity , for the concentration gradient considered (). By scaling up the concentration gradient by a factor of , the value of can also be scaled up by and can easily account for the experimentally measured velocity range.
We carried out a detailed analysis of steady-state bacterial chemotaxis in one dimension. The chemotactic performance in the case of a linear concentration profile of the chemoattractant, , was measured as the slope of the bacterium probability density profile in the steady state. For a singular impulse response kernel, , the slope was a scaling function of , which vanished at the origin, increased monotonically, and saturated at large argument. To understand these results we proposed a simple coarse-grained model in which bacterial motion was described as a biased random walk with drift velocity, , and diffusivity, . We found that for small enough values of , was independent of and varied linearly with nutrient concentration. By contrast, was spatially uniform and its value decreased monotonically with and vanished for . We presented a simple formula for the steady-state slope in terms of and . The prediction of our coarse-grained model agreed closely with our numerical results. Our description is valid when is small enough, and all our results are derived to linear order in . We assume is always satisfied.
Our results for an impulse response kernel can be easily generalized to the case of response kernels with arbitrary shapes in the linear model. For an adaptive response kernel, the spatial dependence of the diffusivity, , cancels out but a positive drift velocity, , ensures bacterial accumulation in regions with high nutrient concentration, in the steady state. In this case, the slope is directly proportional to the drift velocity. As the delay between the positive and negative peaks of the response kernel grows, the velocity increases, with consequent stronger chemotaxis.
Earlier studies of chemotaxis [13]–[16] put forth a coarse-grained model different from ours. In the model first proposed by Schnitzer for a single chemotactic bacterium [14], he argued that, in order to obtain favorable bacterial accumulation, tumbling rate and ballistic speed of a bacterium must both depend on the direction of its motion. In his case, the continuity equation reads(11)where is the ballistic speed and is the tumbling frequency of a bacterium moving toward the left (right). For E. coli, as discussed above, , a constant independent of the location. In that case, Eq. 11 predicts that in order to have a chemotactic response in the steady state, one must have a non-vanishing drift velocity, i.e., . This contradicts our findings for non-adaptive response kernels, according to which a drift velocity only hinders the chemotactic response. The spatial variation of the diffusivity, instead, causes the chemotactic accumulation. This is not captured by Eq. 11. In the case of adaptive response kernels, the diffusivity becomes uniform while the drift velocity is positive, favoring chemotaxis. Comparing the expression of the flux, , obtained from Eqs. 7 and 8 with that from Eq. 11, and matching the respective coefficients of and , we find and . As we argued above in discussing the coarse-grained model for adaptive response kernels, both and are spatially independent. This puts strict restrictions on the spatial dependence of and . For example, as in E. coli chemotaxis , our coarse-grained description is recovered only if and are also independent of .
We comment on a possible origin of the discrepancy between our work and earlier treatments. In Ref. [14], a continuity equation was derived for the coarse-grained probability density of a bacterium, starting from a pair of approximate master equations for the probability density of a right-mover and a left-mover, respectively. As the original process is non-Markovian, one can expect a master equation approach to be valid only at scales that exceed the scale over which spatiotemporal correlations in the behavior of the bacterium are significant. In particular, a biased diffusion model can be viewed as legitimate only if the (coarse-grained) temporal resolution allows for multiple runs and tumbles. If so, at the resolution of the coarse-grained model, left- and right-movers become entangled, and it is not possible to perform a coarse-graining procedure on the two species separately. Thus one cannot define probability densities for a left- and a right-mover that evolves in a Markovian fashion. In our case, left- and right-movers are coarse-grained simultaneously, and the total probability density is Markovian. Thus, our diffusion model differs from that of Ref. [14] because it results from a different coarse-graining procedure. The model proposed in Ref. [14] has been used extensively to investigate collective behaviors of E. coli bacteria such as pattern formation [13], [15], [16]. It would be worth asking whether the new coarse-grained description can shed new light on bacterial collective behavior.
|
10.1371/journal.pcbi.1001039 | Widespread Compensatory Evolution Conserves DNA-Encoded Nucleosome Organization in Yeast | Evolution maintains organismal fitness by preserving genomic information. This is widely assumed to involve conservation of specific genomic loci among species. Many genomic encodings are now recognized to integrate small contributions from multiple genomic positions into quantitative dispersed codes, but the evolutionary dynamics of such codes are still poorly understood. Here we show that in yeast, sequences that quantitatively affect nucleosome occupancy evolve under compensatory dynamics that maintain heterogeneous levels of A+T content through spatially coupled A/T-losing and A/T-gaining substitutions. Evolutionary modeling combined with data on yeast polymorphisms supports the idea that these substitution dynamics are a consequence of weak selection. This shows that compensatory evolution, so far believed to affect specific groups of epistatically linked loci like paired RNA bases, is a widespread phenomenon in the yeast genome, affecting the majority of intergenic sequences in it. The model thus derived suggests that compensation is inevitable when evolution conserves quantitative and dispersed genomic functions.
| Purifying selection is a major force in conserving genomic features. It pushes deleterious mutations to extinction while conserving the specific DNA sequence. Here we show that a large proportion of the yeast genome evolves under compensatory dynamics that conserve genomic properties while modifying the genomic sequence. Such compensatory evolution conserves the local G+C content of the genome, which influences nucleosome organization. Since purifying selection is too weak to eliminate every weakly deleterious mutation in nucleosome bound or unbound sequences, the local G+C content is frequently stabilized by compensatory G+C gaining and G+C losing mutations in proximal loci. Theoretical analysis shows that compensatory evolution is inevitable when natural selection is weak and the genomic feature is distributed over many loci. These results imply that sequence conservation may not always be equated with overall selection. They demonstrate that cycles of weakly deleterious substitutions followed by positive selection for corrective mutations, which were so far studied mostly in RNA coding genes, are observed broadly and profoundly affect genome evolution.
| With the complete sequencing of a large number of genomes, and with the rapid progress in the development and application of methodologies for functional annotation of whole genomes [1], it is becoming evident that our basic concepts of genomic function must be updated. The view of genomes as “bags of genes” is challenged by multiple lines of evidence, such as the extensive transcription of short and long RNAs from a substantial fraction of the genome [2]–[4], and the identification of a dense grid of enhancers and transcription factor binding sites in regions that could not be previously associated with genes [5], [6]. Some of the properties of the newly emerging genomic encodings are clearly different from the prototypic example of the triplet genetic code. The direct mapping between genomic positions (codons) and function (peptides) which is a hallmark of the genetic code does not seem to hold for the majority of the genome. Instead, genomic encodings integrate small contributions from multiple positions to form complex and quantitative outcomes. These types of dispersed encodings may be involved in defining enhancer sequences, maintaining epigenomic switches, affecting widespread transcription, and contributing to chromosome structure and dynamics. The evolutionary implications of these new types of codes are still poorly understood. The classical models in molecular evolution assume fitness to be a function of a single evolving locus. Conservation of the function encoded by such a locus is quantitatively predicted to decrease its rate of evolution. What rates of evolution can be expected when each of the multiple positions have small contributions to some joint quantitative fitness?
Neutral compensatory substitutions were predicted by Kimura 25 years ago [7] to couple substitutions in pairs of interacting protein coding loci. Kimura's concept was that an evolving population trajectory may visit suboptimal fitness levels transiently, thereby invoking an adaptive corrective force that can bring the system back to optimality. Such a process will change the genomic sequence, fixating pairs of compensatory alleles. Kimura's compensatory dynamic may work in any group of loci that are associated with an epistatic (non linear fitness function) constraint and was quantified extensively in RNA coding loci where the epistatic coupling of paired loci has a clear structural interpretation [8]–[10]. Another important source of genomic information, transcription factor binding sites, poses evolution with a different type of epistatic constraint by forming a quantitative binding energy landscape that affects gene regulation [11], [12]. The evolution of binding sites was shown to drive compensatory effects at the single site level [13] and also at the level of binding site clusters (or enhancers) [13], [14]. Studies of enhancer evolution are continuously providing striking examples for plasticity and compensation [15]–[17], but due to their heterogeneity, it is currently difficult to develop a general understanding of their evolutionary dynamics.
A simple experimentally characterized example of a dispersed genomic encoding involves the effect of DNA sequence on nucleosome organization [18], [19]. In-vitro and in-vivo experiments in yeast [20], [21] and other species [22]–[24] showed that nucleosomal packaging is correlated with preferential binding of nucleosomes to specific dinucleotide periodicities, and is strongly anti-correlated with A+T content in general and with poly(A/T) sequences in particular [20], [23], [25], [26]. The correlation between nucleosome occupancy and the underlying DNA sequence is sufficiently powerful to allow sequence based nucleosome occupancy prediction, but this prediction is not based on a strict requirement for certain nucleotides to appear at precise positions. Rather, information from multiple sequence positions along the 147bp length of the nucleosome contributes to the affinity of nucleosomes to a given sequence and consequently, to the formation of stable or semi-stable nucleosome configurations [27]. The evolution of these sequence determinants thus serves as a test case for the dynamics of dispersed genomic encodings. Analysis of substitution rates in yeast suggested that genomic sequences that are unbound to nucleosomes are evolving slower than genomic sequences that are bound to nucleosomes [20], [28]–[30]. Whether this is an indication of classical purifying selection on nucleosome encoding sequences, increased abundance of transcription factor (TF) binding sites at low nucleosome occupancy loci, or nucleosome-associated mutability, is currently unclear [31].
Here we analyze patterns of divergence and polymorphisms in yeast intergenic sequences to substantiate an extended model of selection on a dispersed genomic encoding. The analysis shows that yeast low nucleosome occupancy sequences have maintained a high A+T content throughout the evolution of the Saccharomyces cerevisiae lineage. Contrary to standard evolutionary models, we show that this conservation was made possible not by pointwise sequence conservation, but by a compensatory coupling of decreased rates of A/T-losing substitutions and increased rates of corrective A/T-gaining substitutions. Theoretical analysis suggests that this type of evolutionary dynamics is largely unavoidable when the genome employs dispersed functional encodings. The evolutionary dynamics we reveal shuffle sequences continuously while preserving their encoded function, creating a dynamic yet balanced process that may be central to the evolution of gene regulation.
The global G+C content of the yeast intergenic genome is about 35% (Fig 1A) but there is a significant heterogeneity in the genome local nucleotide composition (Fig S1). Such heterogeneity must be the consequence of a variable evolutionary process working in G+C poor and G+C rich sequences. Recently it was shown that nucleosome occupancy patterns strongly correlate with local G+C content in yeast [32]. We define high nucleosome occupancy loci as those in the top 21% MNase-seq coverage percentiles in-vivo (total 540 kbp, Fig 1B), and low nucleosome occupancy loci as those in the bottom 14% MNase-seq coverage percentiles in-vivo (total 350 kbp). Overall, the intergenic G+C content at high occupancy sequences (∼40% G+C) is higher than the G+C content of low occupancy sequences (∼28% G+C). This heterogeneity is even more pronounced when studying the distribution of tri-nucleotides (Fig 1C, Fig S2), showing A/T tri-nucleotides to be more abundant in low occupancy sequences, and pointing towards additional nucleosome sequence preferences. It was shown before that in-vitro nucleosome occupancy can be robustly predicted from the distribution of 5-mers or even 3-mers in the sequence [21]. This suggests that the functionality and fitness contribution of DNA-encoded nucleosome organization, if such a contribution exists, is dispersed across multiple loci in a quantitative fashion and is not encoded by a strict requirement for precise sequence elements at one or a few positions. To prove or disprove the hypothesis that yeast intergenic G+C content heterogeneity is affected by nucleosome-related selection, we studied the evolutionary dynamics of yeast sequences bound and unbound to nucleosomes. We hypothesized that through characterization of these dynamics, we may reveal, in addition to the sequence constraints affecting yeast nucleosome organization, some general principles governing the evolution of dispersed genomic encodings.
To study the evolutionary dynamics that underlie G+C content heterogeneity and nucleosome occupancy in the yeast genome, we inferred substitution rates and ancestral sequences in the Saccharomyces sensu stricto clade. We performed evolutionary inference from alignments of five yeast genomes [33], [34] for sequences that were classified as high nucleosome occupancy loci in S. cerevisiae. We separately inferred the evolutionary trajectory at low nucleosome occupancy loci. The analysis omitted exonic sequences, since the evolutionary dynamics in these involve additional sources of selection relative to those affecting intergenic sequences. Differences in locus mutability are known to be associated with the flanking nucleotides [35], [36], and this effect may severely bias the comparison of evolutionary dynamics between regions with different nucleotide composition. For example, A+T rich regions, like low-occupancy sequences, may exhibit slower divergence of A/T nucleotides than G+C rich regions, simply because A/T mutability is reduced in the flanking context of A/T nucleotides. To account for this effect of flanking nucleotides on substitution dynamics, we independently estimated the rate of substitution at all 16 possible combinations of flanking nucleotides. Indeed, the substitution rates estimated by our model vary significantly among flanking contexts both in high and low occupancy loci and reflect context-dependency that is consistent among phylogenetic lineages (Fig 2A, Fig S3). For example, the C to T transition rate over the S. cerevisiae lineage in low occupancy regions varies between ∼0.14 in the context of tCc and ∼0.03 in the gCg context. The estimation of context-dependent substitution rates proved essential for the unbiased comparison of evolutionary dynamics between the low occupancy, G+C poor, and the high occupancy, G+C rich sequences. As we show next, it allowed us to robustly identify and validate major differences in the evolutionary regimes of these two classes of loci.
We first studied S. cerevisiae substitution rates inferred from intergenic sequences within 200 bp of annotated transcription start sites. It is known that this region in yeast promoters is enriched for transcription factor binding sites and exhibits a stereotyped nucleosome-depleted region of length ∼100–150 bp. As shown in Fig 2B–C (see also Fig S4 and Fig S5), the analysis reveals that the rates of A/T-losing transitions (A to G, T to C) and transversions (A to C, T to G) are ∼45% lower in low occupancy sequences than in high occupancy sequences. A decrease is observed for all 16 nucleotide contexts (within an estimation variance), and is slightly more pronounced in A/T contexts (AAA, AAT). Notably, the rates of A/T-gaining transitions (G to A, C to T) and transversions (G to T, C to A) are not decreased like the A/T-losing substitutions. In most sequence contexts, the rates of A/T-gaining substitutions are higher in low occupancy sequences or similar between the sequence classes. On the other hand, when flanked by G's or C's, the rates of A/T-gaining substitutions are four times slower in low occupancy compared to high occupancy sequences. Evolutionary theory could not predict these dynamics if the evolution of G+C content was neutral (unless an extremely unlikely mutational regime is separating high from low occupancy regions, as we disprove below using population genetics data). Moreover, a simple theory assuming average stronger evolutionary constraint on low occupancy sequences [20], [29] would predict a general decrease in the substitution rates in the region and would not explain the asymmetry between A/T-gaining and A/T-losing substitution rates.
An important assumption underlying our evolutionary analysis above is that the evolutionary regime operating in regions that are occupied (or unoccupied) by nucleosomes in the extant S. cerevisiae genome has been the same since the divergence of S. cerevisiae from S. paradoxus. Violations of this assumption can potentially affect our substitution rate estimations. For example, if nucleosome occupancy is determined by the genomic sequence, but is not under selection, nucleosomes may drift freely following substitutions spontaneously generating new A+T rich hotspots. Following that, we may enrich for substitutions that increase A+T content in extant low occupancy sequences by assuming nucleosome organization were conserved. To verify that such a scenario has not significantly affected our analysis of TSS-proximal substitution rates, we inferred the G+C content in the common ancestor of S. cerevisiae and S. paradoxus, for 10 ranges of S. cerevisiae nucleosome occupancy levels, and compared it to the extant G+C content (Fig 2D). We found that the G+C content at all levels of nucleosome occupancy did not change significantly during evolution in the S. cerevisiae lineage. Sequences proximal to TSSs therefore conserve their regional G+C content (at least on average). Consequently, the different rates of substitutions in high and low nucleosome occupancy loci do not represent net divergence in the sequence features that correlates with nucleosome occupancy. This is further confirmed by recent comparative analysis of nucleosome organization in S. cerevisiae and S. paradoxus, which revealed only limited divergence in nucleosome positioning for these species [37], [38]. The highly non symmetric substitution dynamics observed at different levels of nucleosome occupancy must therefore be explained by means of a stationary evolutionary process that conserves the underlying nucleosome-associated encoding.
One intriguing possibility that may explain the asymmetry between the rates of A/T-losing and A/T-gaining substitutions in low occupancy sequences is that while A/T-losing mutations are selected against, some can be sustained in the population. Consequently, positive selection is able to push to fixation corrective A/T-gaining mutations (possibly at different genomic positions). If this hypothesis is correct, we can predict that loci near sites of A/T-losing substitutions will be enriched with A/T-gaining substitutions and vice versa. Remarkably, the yeast divergence patterns confirm this prediction. The data reveal that rates of A/T-gaining substitution are accelerated next to sites of observed A/T loss (compared to rates near conserved loci, Fig 3A). Furthermore, as shown in Fig 3A, this effect does not represent general spatial coupling of substitutions, since the A/T gain rate is significantly higher near sites of A/T loss than it is near sites of A/T gain. Conversely, the rates of A/T losing substitutions are higher next to sites of observed A/T gain (Fig 3B). Unexpectedly, this coupling effect is observed robustly across the entire spectrum of nucleosome occupancy levels (p<1e-5 for high nucleosome occupancy, p<0.04 for low nucleosome occupancy). The coupling between contrasting substitutions on spatially linked loci suggests the involvement of a common selective constraint, without which the dynamics at these loci must be independent of each other. The data therefore suggest that compensating A/T-losing and A/T-gaining mutations work to conserve a heterogeneous G+C content (both high and low) in TSS-proximal sequences.
The trinucleotide distributions of low occupancy TSS-distal sequences (over 200 bp from an annotated TSS) are generally similar to those in TSS-proximal loci, but some important differences are notable (Fig 4A). First, for low occupancy sequences, G/C trinucleotides are rarer in TSS-distal than in TSS-proximal loci. Second, poly-A/T trinucleotides are enriched relative to other A/T rich nucleotides in TSS-proximal but not TSS-distal low occupancy loci. These differences may represent a lower fraction of TF binding sites in TSS-distal regions [19], [20] (Fig S6 for additional analysis). As shown in Fig 4B–C, TSS-distal A/T-losing substitution rates are decreased in low occupancy vs. high occupancy sequences, consistent with the observations in TSS-proximal loci. Furthermore, the rates of A/T-gaining substitution in many contexts are increased in low occupancy vs. high occupancy sequences, similar to their behavior in TSS-proximal regions (but with G/C-flanking contexts not highly conserved). Comparison of the ancestral and extant G+C content reveals conservation at high levels of nucleosome occupancy, but some average decrease in G+C content for low nucleosome occupancy loci (Fig 4D). Analysis of compensatory spatial correlation between A/T-gaining and A/T losing substitutions reveals significant coupling at high nucleosome occupancy levels (p<6e-4). Also shown is the tendency of A/T-gaining substitutions at low nucleosome occupancy to occur in clusters (Fig S7).
The data therefore support a compensatory substitution process that drives G+C content conservation in most TSS-distal loci, in a way analogous to the dynamics at TSS-proximal loci. This is demonstrated by the asymmetric rates of A/T gain and A/T loss, the conservation of G/C content and the compensatory substitution coupling at most ranges of nucleosome occupancy. An exception to this general trend is observed at some of the TSS-distal low occupancy loci. We hypothesize that during the evolution of the S. cerevisiae lineage, de-novo A/T-rich hotspots may have driven divergence of nucleosome organization in some TSS-distal loci (possibly since these were under weaker selection [37], [38]). This effect may explain the non-stationary G+C content and spatial clustering of A/T-gaining substitutions at extant TSS-distal low occupancy loci (Fig S7). Taken together, the data on TSS-distal sequences further support the idea that selection maintains heterogeneous G+C content across most yeast intergenic sequences (and in particular at TSS-proximal sequences), and that this selection drives changes in substitution rates that are difficult to explain using models of selection on a single locus.
To study the hypothesis that selection on dispersed nucleosome encodings drives asymmetric substitution patterns in yeasts, we devised a simple theoretical model (Fig 5). We assume that a population of 20 bp sequences (each representing a different “genome”) is evolving given a constant flux of mutations in some fitness landscape that depends only on the G+C content of the sequence. The mutations transform G/C nucleotides to A/T nucleotides faster than they transform A/Ts to G/Cs, driving the genomes' stationary G+C content to a neutral level of 30%. Working against this stationary G+C content, the fitness landscape defines a lower G+C content (20%) as optimal, with symmetrically decreasing fitness for suboptimal values. This landscape is designed to approximate the potential selective pressure on low nucleosome occupancy sequences. We studied the model behavior at various selection intensities both analytically and using computer simulations (Methods). For each intensity level, we determined the A/T gain and A/T loss substitution rates and stationary G/C content (Fig 5A–D). When selection is weak, the dynamics we observed are neutral, with the rates of substitutions being equal to the rates of mutations, and the G+C content converging to the neutral stationary G+C content (30%). In contrast, when selection is strong, the rates of both A/T gain and A/T loss decrease to zero and the G+C content is optimal (20%). These two regimes are compatible with the standard evolutionary theory of selection on a single locus. More notable are the substitution rates observed at intermediate levels of selection. When selection is not sufficiently strong to purify all A/T-losing mutations, A/T-losing substitution rates are only partially decreased. Interestingly, this decrease is matched by an increase in the rate of A/T-gaining substitutions to levels higher than the neutral rate. The new balance between A/T-losing and A/T-gaining rates is sufficient to stabilize the G+C content of the regime at near-optimal levels. Detailed analysis reveals that the increase in the rate of A/T-gaining substitutions is driven by cycles of A/T-loss mutation at one position, which are corrected by an A/T-gain mutation at another position. Similar but opposite dynamics are observed when the optimal G/C content is higher than the neutral one (modeling selection of high G+C content in high nucleosome occupancy sequences, Fig S8). Furthermore, the compensatory regime is observed over a much wider range of selection intensities when the fitness landscape is more tolerant as shown, for example, in Fig 5E–I. These theoretical predictions are consistent with the empirical behavior observed in yeast, showing that weak selection can be sufficiently powerful to increase specific substitution rates over the neutral level due to a compensatory regime.
Our evolutionary analysis above supports the idea that high and low nucleosome occupancy sequences in yeast evolve under a selective pressure to maintain their G+C content, or a refined nucleosome sequence potential that is approximated by the average G+C content. According to this scenario, in low occupancy sequences, which are generally A+T-rich, A/T-losing substitutions are weakly selected against, while A/T-gaining substitutions are frequently pushed to fixation by an adaptive force. According to our simulations and to the standard population genetics theory, such selection on A/T-gaining and A/T-losing mutations should affect the distribution of allele frequencies in the population. In low occupancy loci, A/T-losing single nucleotide polymorphisms (SNPs) are expected to have lower allele frequencies than A/T-neutral SNPs, while A/T-gaining SNPs should have higher allele frequencies. Analysis of polymorphic sites in a sample of 39 S. cerevisae strains [39] confirmed these predictions (Fig 6). We used data on 9185 SNPs in low occupancy loci and 16956 SNPs in high occupancy loci, approximating the minor allele frequency using majority voting and discarding sites with incomplete data or more than two alleles. In low occupancy loci, A/T-losing SNPs are more rare (<20%, alternative threshold generated similar results, Fig S9) than A/T-gaining SNPs in non G/C flanking context (p<2e–05). A reciprocal effect is observed at high occupancy loci, where A/T-gaining SNPs are more rare than A/T-losing SNPs in non G/C flanking context (p<3e–07). The reciprocality of the effect also confirms that our conclusions are not affected by general biases in the estimation of allele frequencies due to systematic sequencing errors. We note that as expected by the low divergence of A/T nucleotides in G/C flanking contexts of low occupancy sequences, the allele frequencies of A/T-gaining SNPs in such loci are reflective of stronger selection. This may be related to the enrichment of such flanking contexts at TF binding sites, as we discuss below.
We classified yeast intergenic regions according to their nucleosome occupancy, and used evolutionary analysis of context-dependent substitution rates to reveal remarkable variability in the evolutionary dynamics of sequences bound and unbound to nucleosomes. Our analysis shows that low occupancy sequences lose A/T nucleotides slowly compared to high occupancy sequences, but gain A/T nucleotides at similar rates. We also observe spatial coupling between substitutions that gain A/Ts and substitutions that lose them, which suggests that a compensatory process preserves G+C content at both high and low occupancy loci. These observations are compatible with a model in which the local G+C content in yeast is conserved through weak quantitative selection. Such weak selection allows occasional fixation of substitutions that disrupt the optimal G+C content of the region, but then respond by adaptive evolution of corrective mutations at the mutated locus or at any of the surrounding genomic positions. Data on allele frequencies of yeast SNPs independently confirm the predictions of such a model. This set of observations proves that the G+C heterogeneity of yeast intergenic sequences is not a consequence of a neutral process and suggests that nucleosome organization may play a major role in this lack of neutrality.
The role of DNA encoded nucleosome occupancy in regulating gene expression is difficult to isolate experimentally, mostly due to the challenge of separating cause and effect inside the complex system involving nucleosomes, remodeling factors and TFs. Previous analysis identified an anti-correlation between nucleosome occupancy and genomic conservation in yeast [20], [28]–[30] putting forward the hypothesis that low occupancy regions (nucleosome free regions, linkers) may be under selection, either due to their increased frequency of TF binding sites, or since they serve as anchors that organize the entire nucleosome landscape. According to our analysis nucleosome occupancy is tightly correlated with substitution patterns reminiscent of selection throughout the genome and not just at low occupancy regions. The data therefore strongly support the non-negligible contribution of DNA encoded nucleosome organization to fitness and therefore to genome regulation. This is further demonstrated by contrasting the G+C content related selection patterns at TSS-proximal sequences (Fig 2, 3), with the frequent cases of overall divergence of A/T rich hotspots and clustered A/T-gaining substitution in TSS-distal low occupancy sequences (Fig 4). The data suggest that when selection is not working, nucleosome occupancy drifts following changes in the encoding sequences [37], [38]. We note that according to our simulations and the empirical data, the selection on nucleosomal sequences must be weak, driven by the very small (but still specific) fitness contribution of any individual genomic position. We predict that such selection is sufficiently powerful to contribute significantly to the heterogeneity of the yeast intergenic sequences, but it is clearly much weaker (per base) than the selection working to conserve classical functional elements. These theoretical considerations underline the difficulty in proving the functionality of specific nucleosome positioning sequences using direct genetics experiments, which typically require large and easily quantifiable phenotypic effects for specific genetic manipulations.
One source of evolutionary constraint on yeast intergenic sequences is their interaction with transcription factors. TF binding sites are known to be conserved among yeast species [33], [34] and their increased concentration in TSS-proximal nucleosome free regions was previously proposed to impose overall conservation at these regions. According to our inferred evolutionary dynamics at TSS-proximal DNA, selection on TF binding sites indeed contributes to the evolution of low occupancy sequences. This is indicated, for example, by a very low A/T gain rates in G/C trinucleotides (Fig 2), which are part of some of the most abundant and conserved yeast binding sites (e.g., Ume6, PAC, Reb1, MBP1) [11], [12], [40]. Nevertheless, selection on binding sites, even those that are A/T rich (e.g. TATA boxes) is highly unlikely to explain the nucleosome occupancy-dependent substitution rates we observed throughout the yeast genome. Specifically, the compensatory coupling of A/T-losing and A/T-gaining substitutions is not compatible with any particular binding site model. We therefore hypothesize that a combination of purifying selection on TF binding sites (either strong [33], [34] or weak [11]) and composite selection on DNA encoded nucleosome organization together define a complex fitness landscape that shapes the evolution of yeast intergenic sequences.
We studied here a model of evolution as manipulating sequences in a complex fitness landscape that combines contributions from multiple coupled loci into a single dispersed encoding. As shown by theoretical and empirical analysis of the model, when selection on each individual locus is weak, purifying selection is incapable of completely purging mutations that are only slightly deleterious and these are continuously challenging the overall optimality of the sequence. This suboptimality is compensated effectively by adaptive evolution at multiple other loci that participate in the dispersed encoding. In contrast to other cases of compensatory evolution (proteins [41] or RNA molecules [8]-[10], [42]), the encodings we studied here provide ample direct ways to correct a slightly deleterious substitution, thereby increasing the rate of compensation. Our study builds on earlier work on codon bias [43], [44], but uses the global and experimentally characterized sequence classes at high and low nucleosomes occupancy loci to establish compensatory evolution as a major driving force in evolution under multi-site selection. This type of evolutionary dynamics may be generalized to other dispersed functional encodings [45], [46] including complex regulatory switches that typically involve a large number of TF binding sites of variable factors and specificities. The remarkably global nature of the compensatory effect we observed in yeast, which cause a measurable global increase in the substitution rate of specific mutations, supports the notion of an evolutionary process that conserves function without a strict requirement to conserve sequence. It is tempting to speculate that such a process may allow genomes to maintain diversity and continuously search the sequence space, without significantly compromising their existing regulatory circuits. Furthermore, this process may reduce, through compensation, the mutational load [47] resulting from the use of multiple loci to encode regulatory functions.
Multiple alignments of the Saccharomyces cerevisiae, Saccharomyces paradoxus, Saccharomyces mikatae, Saccharomyces kudriavzevii and Saccharomyces bayanus were downloaded from the UCSC database [48] (sacCer2 version). Alignments were based on the SGD June 2008 assembly. A genome wide in-vivo nucleosome occupancy profile for S. cerevisiae was used as previously described [21], indicating a nucleosome occupancy value for each genomic position. SNP data were downloaded from the SGRP website [39]. Gene Annotations and transcription start sites of S. cerevisiae were taken from the SGD known gene table which corresponds to sacCer2 [49]. Transcription factor binding sites were downloaded from the UCSC Genome Browser [48] and are based on the chip-chip experiments described before [50].
Our analysis focused on intergenic genome sequences which are defined based on the SGD gene annotations. Each intergenic locus was defined as TSS-proximal if it is not part of an exon, and has an annotated TSS within 200 bp of it. TSS-distal loci included the remaining non exonic loci. We defined low occupancy loci as positions with nucleosome occupancy value lower than −2.5 (relative to the genomic mean, detailed description in Kaplan et al. [21]) and high occupancy loci as positions with occupancy higher than 0.4. Alternatively, we classified all loci to equal sized bins of nucleosome occupancy (ten in analysis of ancestral G+C context and five in the analysis of spatial coupling). Alternative definition of low occupancy linker regions based on raw data of MNase restriction sites resulted in similar results (data not shown).
As described in the text, a refined context dependent substitution model is essential for the correct estimation of the different evolutionary dynamics in low G+C content, low occupancy loci and high G+C content, high occupancy loci. We therefore applied a flexible substitution model to perform ancestral inference and learn evolutionary parameters from alignment data (details available upon request). The model included parameters for the substitution rates at each of 16 possible contexts parameterized by the identities of the 3′ and 5′ flanking nucleotides. Independent substitution rates were assumed for each lineage in a phylogenetic tree which was fixed throughout the process. We note that the model does not assume parametric constraints on different substitution rates, and infers substitution rates on lineages, rather than a global substitution rate matrix and branch lengths. This approach has proved more robust given that a sufficient number of loci was available to learn robustly the parameters at each lineage, and given that the substitution process in the different lineages indicated gradual changes in dynamics that a model using a universal rate matrix could not have accounted for (for example, the extant G+C content in each of the species we used show some variability).
To perform ancestral inference, we used a customized loopy belief propagation algorithm on a factor graph approximation of the model [51]. Parameter estimation was then performed using a generalized EM algorithm. We validated some key results using parsimony analysis (Fig S10 and data not shown).
For analysis of the resulted model parameters, each context dependent substitution rate was averaged with its reverse complement. For example CAT->CCT is averaged with ATG->AGG. The averaged conditional probabilities are presented in Fig 2, 4, Fig S3 and Fig S4. A/T gaining is defined as any of the following substitutions in any flanking contexts: C->A, C->T, G->A, G->T. A/T loss in defined as any of the following substitutions in any flanking contexts: A->C, A->G, T->C, T->G. Analysis was generally focused on the S. cerevisiae lineage (data on the other lineages are shown in Fig S3, Fig S5).
In order to estimate the theoretical regional G+C content of S. cerevisiae intergenic sequence, we have simulated this sequence using a lineage specific evolutionary probabilistic model learned over the whole intergenic sequence (see above). Specifically, the common ancestor of the sensu stricto clade was simulated first based on the learned 2-order markov model. Following this, the sequences of the descendants were simulated based on the simulated ancestor sequence and the corresponding substitution model. Iteratively, the sequences of all species in the phylogeny were simulated, including the extant species. The regional G+C content of the simulated S. cerevisiae intergenic sequence is presented in Fig S1.
To estimate the coupling between A/T gaining and A/T losing substitutions in the yeast genome, we used our probabilistic model to infer at each genomic position j the posterior probability of each type substitution in the lineage leading to species i from its ancestor (pai):
When sji denotes the nucleotide at the j'th genomic position of the i'th species in the phylogeny, and sjpai denotes the sequence of the ancestor of this species at the same genomic position.
Given the posterior probabilities we computed for each genomic position j the expected numbers of A/T loss and A/T gain events in the sequence preceding it. This was done using a horizon parameter, which was set to 5 bp by default (for alternative horizon values see below):
Where the δgain, δloss functions were given by Table 1, and the net A/T divergence of the position was defined as:
We then identified all positions with A/T divergence <-0.9 (A/T losing contexts), with A/T divergence >0.9 (A/T gaining contexts) and with conserved A/T content (background). For each such set we computed the probability of A/T gain and A/T loss substitutions using the same inferred posterior probabilities. By using this approach (conditional probability given the events in the preceding 5 bp) we ensured each substitution is counted precisely once. By computing the probabilities for similar events (e.g. A/T gain) given different contexts (A/T losing, A/T gaining, or background), we could robustly asses compensation patterns while controlling for the different basal rates of A/T gain and A/T loss and the general clustering of substitution in the genome.
To statistically assess the coupling between A/T divergence context and A/T losing/gaining substitutions in the S. cerevisiae lineage we counted the numbers of A/T gains and A/T losses at A/T gaining and losing contexts:
= number of A/T gains in A/T gaining contexts
= number of A/T losses in A/T gaining contexts
= number of A/T gains in A/T losing contexts
= number of A/T losses in A/T losing contexts
In addition we counted the numbers of A/T and C/G occurrences in these contexts:
= number of A/T's in A/T gaining contexts
= number of A/T's in A/T losing contexts
= number of C/G's in A/T gaining contexts
= number of C/G's in A/T losing contexts
We wished to test whether the spatial compensation effect is significant even given the general clustering of substitutions. Our null hypothesis was therefore:
We test it using bootstrapping with 100,000 resamples. At each resample, a set of items are sampled without replacement out of the union of two sets of sizes and (denoted by A, B respectively). Similarly, we sample without replacement items out of the union of two sets of size , (denoted by C, D respectively). The number of sampled items belonging either to set A or C is collected across all resamples. We end up with 100,000 counts representing the background distribution for the statistic. P-value for the null hypothesis is calculated by counting the fraction of iterations in which the sampled counts are bigger than .
Analysis of the robustness of the observed compensation patterns for different values of the horizon parameter is shown in Fig S11, Fig S12, and Fig S13.
To study the hypothesis that selection on dispersed nucleosome encodings drives asymmetric substitution patterns in yeasts, we devised a simple theoretical model. For clarity we describe here the version of the model for low occupancy sequences. For nucleosome DNA the model is the same apart from the fitness function.
First we used a Wright-Fischer dynamics on a population of binary sequences of size L, : In each generation there is a probability of for each site containing 0 to be flipped to 1 and for sites containing 1 to flip to 0. The sequences are then sampled relative to their fitness , where and is equivalent to the GC content. We simulated this system for the following parameter set
We note that the population expected θ parameter may be estimated from the above parameters ( in haploid population, but given the two different mutation rates the empirical theta needs to be corrected). The parameters we used ensured θ<0.04.
The simulation was based on the following procedure:
Initialize: Create a population of identical sequences of length L. For simplicity sequences use a binary alphabet on A and G. We define the current reference genome sequence R using the same initial sequence. We introduce the following counters to accumulate sufficient statistics for computing the rates of A->G and G->A substitutions (NA, NG and NA->G,NG->A, such that the rate will be estimated as NA->G /NA, NG->A /NG).
Sample a new generation: to create a new generation, we sample times from the current population using weights that are proportional to the fitness of each individual. For each sampled individual, we introduce mutations with probability for G loci and for A loci. Starting after a minimal number of “burn-in” iterations (at least 4 coalescent times) we also incremented NA and NG for each sampled individual with the number of A's and G's in the respective sequence.
Updating the reference genome: given the new generation population, we tested the frequency of A and G at each of the L genomic loci. Whenever the frequency in the current population is larger than 0.95 and the major allele is different from the reference genome R, we incremented the counter NA->G or NG->A (after the burn-in period) and updated the sequence R.
We end up with counts of A's (NA), counts of G's (NG) (in units of generations X loci) and counts of the substitutions between them (NA->G, NG->A). Substitution rates are estimated by:
These rates are shown in Fig 5 and Fig S8 for the different fitness landscapes we defined next.
The goal landscape is defined symmetrically around an optimal number of G's denoted by nGC and the selection intensity η (e.g., X axis in Fig 5B, C, F, G):
The threshold landscape is defined using similar parameters to generate an asymmetric function:
Next, we studied the above model analytically in the regime of low mutation rates. In this regime, drift is the dominating mechanism and we can model the process by assuming the population is represented by a single genome (or GC content). Given the definitions above, the rate at which mutations that increase the GC content enter the population is
While the rate of mutations that decrease the GC content is
In such drift dominating regime, the fixation probability of a new mutation is:where is the marginal fitness of the mutation [52]. Therefore, the rate at which the GC-content increases or decreases is on average
Where .
Thus the set equations for the dynamics of is
Solving this for the steady state results in
Where and is set by normalization . From this distribution of the GC-content one can calculate the average GC-content and the substitution rates
As can be seen in Fig S8, the analytical result and Wright-Fischer simulation are in good agreement.
We used DNA sequences of 39 S. cerevisiae strains sequenced in the Saccharomyces genome resequencing project (SGRP). Here, only intergenic 2 allele SNP's with sequence data from more than 20 strains were considered informative. For each of those SNP's, major allele was defined as the most abundant allele in the population. Minor allele is defined as the least abundant allele. A/T gaining SNPs were defined when the nucleotide of the major allele was C or G and the minor allele is A or T. A/T losing SNPs were defined reciprocally. All other SNPs were defined as A/T conserving (see illustration in Fig 6). We further subdivided SNPs into two groups: SNP's in G/C flanking context and SNP's with at least one A or T in the flanking contexts, using the reference strain for determining the context. These subgroups are again subdivided to SNP's within low occupancy sequences and SNP's within high occupancy sequences (Fig 6). We analyzed the distributions of the frequency of minor alleles of these subgroups separately. In figure 6, shown are the fraction of rare alleles (minor allele frequency <0.20) among A/T gain, A/T loss and A/T conserved SNP's within low or high occupancy sequences. We used a chi-squared test to reject the null of hypothesis that the fraction of rare alleles is the same between A/T gain and A/T loss SNP's.
|
10.1371/journal.pcbi.1005984 | Self-crowding of AMPA receptors in the excitatory postsynaptic density can effectuate anomalous receptor sub-diffusion | AMPA receptors (AMPARs) and their associations with auxiliary transmembrane proteins are bulky structures with large steric-exclusion volumes. Hence, self-crowding of AMPARs, depending on the local density, may affect their lateral diffusion in the postsynaptic membrane as well as in the highly crowded postsynaptic density (PSD) at excitatory synapses. Earlier theoretical studies considered only the roles of transmembrane obstacles and the AMPAR-binding submembranous scaffold proteins in shaping receptor diffusion within PSD. Using lattice model of diffusion, the present study investigates the additional impacts of self-crowding on the anomalousity and effective diffusion coefficient (Deff) of AMPAR diffusion. A recursive algorithm for avoiding false self-blocking during diffusion simulation is also proposed. The findings suggest that high density of AMPARs in the obstacle-free membrane itself engenders strongly anomalous diffusion and severe decline in Deff. Adding transmembrane obstacles to the membrane accentuates the anomalousity arising from self-crowding due to the reduced free diffusion space. Contrarily, enhanced AMPAR-scaffold binding, either through increase in binding strength or scaffold density or both, ameliorates the anomalousity resulting from self-crowding. However, binding has differential impacts on Deff depending on the receptor density. Increase in binding causes consistent decrease in Deff for low and moderate receptor density. For high density, binding increases Deff as long as it reduces anomalousity associated with intense self-crowding. Given a sufficiently strong binding condition when diffusion acquires normal behavior, further increase in binding causes decrease in Deff. Supporting earlier experimental observations are mentioned and implications of present findings to the experimental observations on AMPAR diffusion are also drawn.
| The transmembrane AMPA receptors (AMPARs) prominently exhibit lateral diffusion in the postsynaptic membrane at excitatory synapses. Steric obstructions to AMPAR diffusion due to the crowd of other relatively static transmembrane proteins and binding of AMPARs to the submembranous scaffold proteins in the specialized region of postsynaptic density (PSD) are well known to retard receptor diffusion, which causes receptor trapping and accumulation within PSD. However, AMPARs are significantly bulky structures and may also obstruct their own diffusion paths in the presence of their high density. It is shown here that intense self-crowding of AMPARs may lead to highly obstructed and confined receptor diffusion even in the obstacle-free medium, and the presence of other obstacles further aggravates this effect. AMPAR-scaffold binding reduces confined diffusion arising from self-crowding and strong binding engenders normal diffusion even at high receptor density. However, it overall causes reduction in the effective diffusion coefficient of the receptor diffusion.
| Glutamate-binding transmembrane alpha-amino-3-hydroxy-5-methyl-4-isoxazole-propionic acid (AMPA)-type receptors (AMPARs) are the pivotal element of fast synaptic transmission at excitatory synapses in the central nervous system [1, 2]. At the site of synaptic contact, these receptors are present at high density within a specialized region on the postsynaptic membrane, termed as postsynaptic density (PSD), which is closely apposed to the presynaptic active zone of glutamate release [3–5]. The remaining extra-synaptic region of the postsynaptic membrane is distinguished with a relatively lower density of AMPARs [4]. This unique spatial arrangement of AMPARs is a natural adaptation to expose a sufficiently large number of the receptors to sufficiently high glutamate concentration in the synaptic cleft [6–8] and, thus, enhance the postsynaptic response, which would otherwise be comparatively weaker under a homogeneously distributed receptor condition [9–11].
Like other mobile transmembrane proteins, AMPARs also exhibit lateral diffusion in the postsynaptic membrane [12, 13]. AMPAR mobility is crucial for many essential processes associated with the efficiency of synaptic functioning. Lateral diffusion causes continuous exchange of the receptors between PSD and extra-synaptic region [14, 15]. This exchange brings about replacement of desensitized AMPARs in the PSD with active AMPARs of the extrasynaptic region after an event of glutamate release and, hence, assists in maintaining the strength of consecutive postsynaptic responses in the presence of high-frequency presynaptic spike train [16–18]. It also underlies the recruitment of new AMPARs into the PSD which are brought by exocytotic vesicles from the local intracellular reserve-pool and are initially unloaded on the extra-synaptic membrane [19, 20]. In a similar manner, the older receptors in the PSD diffuse out to the extra-synaptic region where they are endocytosed [21]. Accordingly, lateral diffusion of AMPARs assists in the appearance of long-term potentiation (LTP) [20] or long-term depression (LTD) [22, 23] at excitatory synapses and, hence, assists in the molecular basis of learning. Owing to such a crucial and indispensable role of AMPAR lateral diffusion in shaping the density and spatial localization of AMPARs in the PSD, it has always engaged attention of a wide scientific community.
A remarkable thing is the gathering of AMPARs in the PSD despite that PSD spans a much smaller area than the extra-synaptic region and the AMPARs are significantly mobile. It seems that the receptors get trapped in the PSD while diffusion because, in the absence of trapping, diffusion would lead to a homogeneous distribution of AMPARs on the entire postsynaptic membrane. In the backdrop of continuous receptor exchange between PSD and extra-synaptic region, the trapping can be viewed in terms of the considerably longer residence time [24] of an AMPAR in the PSD. Using techniques like fluorescence recovery after photobleaching (FRAP), electrophysiology with mutant variants of AMPAR, and different versions of single particle tracking such as sptPALM, uPAINT and quantum dot (QD)-tagging of receptors, various experimental studies [24–29] in hippocampal slices and live hippocampal neurons in dissociated cultures have provided a wealth of observations on the nature of AMPAR diffusion at excitatory synapses. In conjunction with the theoretical investigations [29–34], these studies have so far clearly shown that the molecular composition of the PSD is indeed responsible for the trapping of diffusing AMPARs. The PSD is rich in large number of transmembrane as well as submembrane proteins [3, 35–37], owing to which it possess a very high molecular weight [37, 38]. The crowding of inert transmembrane proteins strongly obstructs the AMPAR diffusion within the PSD region through steric repulsion [29, 31]. Further, the reversible binding of intracellular domain of AMPARs as well as their associations with transmembrane AMPAR regulatory proteins (TARPS) to the submembranous scaffold proteins, such as PSD-95, also substantially reduces the AMPARs mobility [29, 39, 40].
An AMPAR is a tetramer and is consist of any combination of the four kinds of subunits GluR1, GluR2, GluR3 and GluR4 [41, 42]. Typically, GluA1-GluA2 and GluA2-GluA3 heterotetramers are most abundant in the adult brain [43–45]. These receptors are very bulky [41] and carry along a large steric-exclusion volume. The bulkiness of AMPARs is further increased due to the various auxiliary proteins [46–50] associated with it. In fact, the size of the native complexes of AMPARs isolated through biochemical techniques have been found to be approximately double the original size of the tetramer [51]. Further, AMPARs reside at high density in the PSD and contributes to a substantial fraction of the local macromolecular crowding [3, 37]. Therefore, it is reasonable to envisage that these receptors may block the diffusion paths of each other and may lead to a situation of self-obstruction or self-crowding. Moreover, the distribution of AMPARs in the PSD is not strictly homogeneous. Rather, there are smaller subregions or nanodomains within the PSD which have more AMPARs cluttered [52, 53] and the self-crowding of these receptors would be more pronounced. Accordingly, besides inert transmembrane protein crowding and binding to scaffold proteins, self-crowding of AMPARs may appear an additional factor behind the reduced or hampered mobility and trapping of these receptors in the PSD. However, in the earlier experimental and theoretical studies, the possible role of self-crowding factor has remained completely unaddressed.
The above speculation regarding self-crowding of AMPARs is a seemingly interesting issue and, therefore, is the source of motivation for carrying out the present theoretical investigation. The effect of various crowding factors on the AMPAR diffusion can only be enquired through detailed numerical simulation of independent diffusing receptors. Therefore, the present study involves the Monte Carlo simulation of receptor diffusion using lattice model of diffusion, which has proven to be an effective approach in the earlier theoretical studies [31, 54, 55]. The main body of the present study is comprised of a purely abstract framework with a lattice used as a generalized spatially-discrete medium of diffusion, regardless of whether the lattice represents the entire PSD or a subregion within the PSD. Moreover, the AMPARs are represented by point diffusion tracers (DT) on the lattice [31, 54]. On the basis of the ensemble-averaged mean-squared displacement of tracer diffusion, the nature of diffusion is established in terms of two physical quantities viz. anomalousity and effective diffusion coefficient under the different pertinent conditions of crowding and binding events. Both the quantities serve as the suitable marker of resulting dwell time of the receptors within PSD and, hence, can be effectively used to comprehend receptor trapping [31]. It must be noted that the dynamics of the AMPAR accumulation in the excitatory PSD and exchange with extrasynaptic region is not the immediate interest of the present study. Rather, it focusses on capturing the emergent statistical behaviors of the receptor diffusion in the thermodynamic limit when self-crowding is considered in addition to the other obstacles and binding, which may be later used as the building block to comprehend the dynamics of accumulation.
The findings reveal that even in the absence of any steric crowding of other transmembrane and scaffold proteins in the postsynaptic membrane, very high density of AMPARs may itself lead to extraordinarily high anomalousity and reduced diffusion coefficient. Remarkably, anomalousity of receptor diffusion may also exhibit a switch-like behavior with respect to their self-crowding density, similar to the switch-like behavior with respect to increase in steric macromolecular crowding of other PSD proteins observed earlier [31] as well as here. Further, increase in the crowding by other PSD proteins may exacerbate the anomalousity and decline in diffusivity arising from self-crowding. Contrarily, binding appears to mark a reverse effect by decreasing the anomalousity of crowded receptor diffusion. The plausible mechanisms underlying these findings are discussed to details. Moreover, the relevance of the use of point tracers in capturing the picture of self-crowded diffusion of the non-zero lateral-sized AMPARs is also drawn. Eventually, the possible elements of self-crowding lying within the earlier experimental observations on the nature of AMPAR diffusion at excitatory synapses are pointed out and the physiological relevance of the present observations made through the abstract framework is established in regard of real biological scenario.
The lateral diffusion of AMPARs is realized here through the numerical simulation of the diffusion of DTs. For simplicity, the entire macromolecular crowding at the PSD can be broadly classified into two pools [31]: Completely-Reflecting Obstacles (CROs) and Partially-Reflecting-cum-Binding Obstacles (PROs) (Fig 1A). In general, the CROs represent various transmembrane proteins in the PSD [3, 35] which interact with a diffusing AMPAR only to reflect it away on collision via. steric repulsion. Since they never significantly bind the AMPARs, they behave as inert obstacles [31, 54]. On the other hand, the PROs are often the submembranous scaffold proteins which are intracellularly accumulated close to the PSD [3, 35]. These proteins generally offer only a partial obstruction to the diffusing AMPARs through steric repulsion of the intracellular C-terminal domains of the receptors [31]. Moreover, they preferentially bind the receptors almost at their location through reversible non-covalent interactions between specific domains of the receptors and the scaffold proteins. For instance, GluR1 subunit of AMPARs can directly bind to the scaffolding SAP-97 proteins [56] whereas GluR2 subunits can bind to submembranous PICK1 or GRIP [57]. However, AMPARs cannot directly interact with the one of the most abundant PSD-95/SAP-90 scaffold proteins at excitatory synapses due to their incompatible PDZ domains. Rather, the receptors require association with auxiliary transmembrane AMPA receptor regulatory proteins (TARPs), such as Stargazin, to bind with PSD-95 [39].
Accordingly, the probability of reflection (Preflect) for tracer collision with CROs is always 1 [31] whereas it may be variable for collision with PROs, as it depends on the size of the C-terminal domains of AMPAR subunits, the size of the specific scaffold proteins, presence of auxiliary proteins etc. For the present investigation, Preflect for collision with PROs is kept fixed at 0.5, signifying a 50% chance of reflecting the trajectory of a tracer on physical contact without binding [31]. Furthermore, the detailed multi-step kinetic scheme for the binding of AMPARs to scaffold proteins is still unknown. However, the intensity of binding through hydrogen bond interactions within PDZ domains have been estimated to be in the range of 2–13kBT [31, 58–60], which may serve here as a rough estimate for the binding energy of AMPAR-scaffold interactions.
Accordingly, three specific situations of homogeneous DT-PRO binding viz. weak, intermediate and strong bindings with binding energy 2, 6 and 10 kBT, respectively, are taken into account in the present investigation. Further, a conventional approach is to assume the entire crowding factors to be static at their spatial locations in the PSD while an AMPAR diffuses through the crowd [29, 31]. This approach is reasonably correct to a great extent as the mobility of the crowding factors is too low [61, 62] in comparison to that of AMPARs and their average life-time in the PSD (in hours) [63] is substantially high relative to the typical measurement time-duration of AMPAR diffusion (in seconds). Accordingly, the obstacles CROs/PROs are considered here immobile and their density preserved throughout the duration of tracer diffusion.
A two-dimensional square lattice (Fig 1B) is considered for performing the lateral diffusion of AMPARs in the postsynaptic membrane [31, 54, 55, 64]. The lateral diffusion coefficient of an AMPAR in the extra-synaptic membrane is known to be almost 0.2 × 10−3μm2.ms−1 [24, 31, 65]. Since the extrasynaptic membrane offers least macromolecular obstructions relative to the PSD, this estimate is considered here as the natural free diffusion coefficient of an AMPAR in an unobstructed lipid medium of the postsynaptic membrane and is here assigned to the effective diffusion coefficient of DT (Deff). The diffusion is performed at discrete time-steps Δt of fixed size 10−3ms [31]. The mean-squared displacement (MSD), 〈r2〉(t), of a DT undergoing free normal diffusion in a two-dimensional medium is given by,
〈 r 2 〉 ( t ) = 4 D e f f t (1)
Therefore, using Eq 1, the desired finite diffusion length Δl for the lattice diffusion can be computed to be 8.9 × 10−4μm [31]. This estimate is assigned to the edge length between any two lattice-sites in the square lattice. In this way, the size of the entire lattice in terms of the number of lattice-sites is kept 1119 × 1119 such that the lattice approximates to an area of 1μm2 [31]. It must be noted that the lattice employed here is a purely abstract framework to procure the salient features of the tracer diffusion under different crowding conditions. Therefore, depending on the requirement, it may be used to address the properties of AMPAR diffusion over the entire PSD as well as within a subregion of the PSD.
The two kinds of obstacles, CROs and PROs, are considered as point obstacles over the lattice (Fig 1B). The CROs and PROs are uniformly distributed over the lattice according to their desired area fractions aCRO and aPRO, respectively. The two classes of obstacles are dealt separately so that how these obstacles of different nature may affect tracer diffusion in their specific manners can be clearly examined.
Since the main objective of the present study is to investigate the effect of self-crowding of tracers on their lateral diffusion, a standard situation of “no-self-crowding” is also taken into account which serves as a benchmark for comparative analysis of the observations made under varying self-crowding situations. In this standard situation, DTs are uniformly placed on the square lattice where each tracer behaves as an independent diffusing entity. Therefore, while diffusion, two or more tracers can together occupy the same lattice site. No steric exclusion among the tracers is considered. In fact, this constitutes an ensemble of multiple copies of independently diffusing tracers but with different initial positions on the lattice under an identical distribution of CROs or PROs. While initially placing DTs on the lattice, it is taken care that a tracer should not lie at a lattice site already occupied by a CRO whereas it is allowed to lie on the lattice-site occupied by a PRO. Moreover, in the case of PROs, the tracers are allowed to diffuse for 2s after being initially placed to acquire thermal equilibrium. Only after this annealing period, the measurement of tracer diffusion is performed [55].
However, while dealing with the self-crowding conditions, the density of diffusing tracers placed on the square lattice would also matter (Fig 1B). Accordingly, the area-fractions of the lattice occupied by DTs, aDT, are taken in the increasing orders of the magnitudes such that conditions of six different aDT viz. 0.00001, 0.0001, 0.001, 0.005, 0.01 and 0.1, are investigated in the present study. Here too, the DTs are uniformly distributed at the desired aDT and considerations regarding their initial placement on the lattice depending on the CROs or PROs are taken care as described for the standard situation.
A periodic boundary condition is imposed on the boundary of the lattice mesh [31, 54]. Therefore, as a tracer leaves the mesh, it re-enters the mesh from the exact opposite side. Monte-Carlo simulation of tracer diffusion is performed. At each time-step, all the tracers present over the mesh are inspected for diffusion one by one. For each tracer, a random number is generated from the uniform random number distributed over the interval [0, 1] to decide the direction of its diffusion. If the random number is < 0.25, the tracer would move left. If the random number is ≥ 0.25 but < 0.5, the tracer would move right. If the random number is ≥ 0.5 but < 0.75, the tracer would move up. Finally, if the random number is ≥ 0.75, the tracer would move down. Based on this outcome, the tracer intends to hop to its nearest-neighbouring lattice site, referred to as the destination site. However, before accomplishing the hopping, the occupancy status of the destination site in regard of CRO or PRO is checked.
If the destination site is occupied by a CRO, no hopping is performed and the tracer stays at its original lattice-site. In the case of PRO occupying the destination site, a uniform random number is again generated over the interval [0, 1] to check for the partial reflection of the diffusing AMPAR with the Preflect = 0.5. If the random number is ≥ Preflect, the tracer is allowed to diffuse to the destination site. Once the tracer reaches the PRO-occupied lattice site, it is considered to be bound. It will unbind and diffuse at a further time-step only when another uniform random number generated in a similar manner is greater than or equal to the probability of escape, Pescape, which is defined from the binding energy as [55],
Pescape=e(−BindingEnergykBT)
(2)
Once the tracer unbinds from the PRO, it is allowed to diffuse in either of the directions isotropically. Therefore, it may be noted that rotational diffusion of DT is neglected in the present framework [31].
The above algorithm is identically shared by the standard condition as well as the conditions of self-crowding. However, the latter condition involves some additional restrains for hopping to the destination site. Since steric-exclusion of DTs among themselves is present in the case of self-crowding, the destination site already occupied by another tracer does not allow hopping of the subject tracer while diffusion. Under such a situation, a peculiar phenomenon of false self-blocking of tracers might appear during diffusion simulation. This condition and its implemented remedy are described in the following subsection. Further, as long as another DT is bound to PRO, the partial reflection of the PRO turns into a complete reflection and the destination site would behave as if it is occupied by a CRO.
While performing the lattice-diffusion of a population of DTs with steric exclusion for each other, there appears a computational problem regarding the sequence of performing the finite-step hopping of individual tracers at each time-step of the diffusion simulation. At every time-step, all the tracers are genuinely expected to diffuse simultaneously on the lattice in random directions and depending on the availability of unoccupied neighbouring sites. If the tracers do not have steric-exclusion property, more than one receptor can occupy a single lattice site. Under this assumption, the computational sequence of performing the hopping of receptors one by one during a single time-step of simulation does not matter. However, if the tracers sterically repel each other, the computational sequence of performing the hopping of tracers at each time-step may lead to different profiles of diffusion.
This issue becomes clearer when the lattice-diffusion of only two DTs, let’s say, DT1 and DT2 with steric repulsion is illustrated (Fig 1C). If the two tracers are sufficiently isolated from each other on the lattice, the sequence of performing the finite-step hopping of the individual tracers during a forward time-step of simulation does not matter, as either sequence, first DT1 and then DT2 or first DT2 and then DT1, leads to the same diffusion profile. Now consider that the two tracers are sitting at the neighbouring sites and the random number generation leads to the expected movements of DT1 towards DT2 and DT2 towards the upper unoccupied neighbouring site. The sequence where first DT1 is considered for hopping will lead to reflection of DT1 back to its position since the DT2 is presently occupying the lattice-site. Next, when DT2 is considered for hopping, it will easily move to the upper lattice site leaving behind an unoccupied lower lattice site. Here, only one receptor DT2 could practically diffuse. In another sequence where first DT2 is considered for hopping and then DT1, DT2 will move to the upper lattice site and DT1 will arrive at the earlier position of DT2. Here, both the tracers could diffuse. According to the theoretically-expected simultaneous diffusion of both the tracers, the diffusion profile deriving from the latter sequence is correct but the former sequence leads to an artefact owing to the computational sequence of performing the hopping of tracers.
To solve this problem, an algorithm with two recursive steps of performing a sequential hopping of tracers at each time-step of simulation is devised:
This scheme of hopping the tracers completely removes the possibility of false self-blocking of tracers while diffusion on the lattice.
The time-duration of the recording of spatial locations of the DTs is 2s [31]. Wherever necessary, the observation has been made for an extended duration of time. Using the DTs’ trajectories, the temporal profile of ensemble-averaged MSD is computed as,
〈 r 2 〉 ( t ) = 1 N ∑ i = 1 N ( x i ( t ) - x i ( 0 ) ) 2 + ( y i ( t ) - y i ( 0 ) ) 2 (3)
Where, N is the number of DTs on the two dimensional lattice. xi (t) and yi (t) are the x− and y− coordinates of the ith tracer at time t. xi (0) and yi (0) denotes the initial location of the ith tracer at the beginning of diffusion i.e. t = 0. The MSD profiles are further averaged over 250-700 ensembles of lattices for every crowding conditions. In theory, the MSD of two-dimensional diffusion is described in general as
〈 r 2 〉 ( t ) = 4 D t α (4)
Here, α is the anomalous exponent and D is the diffusion constant of the diffusing particle. If α = 1, the diffusion is normal. However, if 0 < α < 1, it characterizes anomalous sub-diffusion. In this regard, computation of the log (〈r2〉(t)/t) vs log (t) profile, referred in the following text as log-log profile, is very beneficial for procuring many important features of the tracer diffusion.
log( 〈 r 2 〉 ( t ) / t ) = log ( 4 D ) + ( α - 1 ) log ( t ) (5)
It may be noted that for normal diffusion with α = 1, the log-log profile would appear a flat horizontal line with slope zero. However, for anomalous diffusion, the log-log profile would have a negative slope of magnitude (1 − α). Higher will be the anomalousity of diffusion, sharper will be the decline in log-log profile. Therefore, the log-log profile can easily provide a clear demarcation for the diffusion to be called normal or anomalous and is useful for computing the anomalous exponent of the diffusion as well. As a matter of fact, one may also be interested in the spatiotemporal profile i.e. probability distribution function of tracer diffusion. However, the main interest of this paper is in properties directly pertinent to tracer mobility viz. anomalousity and effective diffusion coefficient. Since a Gaussian or non-Gaussian diffusion can be normal as well as anomalous [66], the mobility factors are ultimately described by the MSD.
Lattice diffusion with no-self-crowding condition of tracers would serve as a standard benchmark for this study and corresponds to the conventional approach [29, 31, 54, 64] adopted so far in the existing literature. Accordingly, each following subsection begins with mentioning the observations made under varying conditions of reflecting obstacles (CROs) or/and binding obstacles (PROs) but in the absence of self-crowding factor. Subsequently, features of diffusion in the presence of self-crowding factor would be discussed.
In the absence of self-crowding factor, the 〈r2〉 increases linearly with time for the lower aCRO = 0.00 − 0.25 (Fig 2A). On the other hand, the conditions of very high aCRO = 0.45 − 0.60 can be clearly recognized by the confined tracer diffusion as the associated 〈r2〉 rapidly reaches a plateau (Fig 2A, inset) where no further variation in it occurs with progression of time. This range of aCRO is close to or above the percolation threshold, θP, of a square lattice framework for tracer diffusion, which is known to be approximately 0.5 for a sufficiently large square lattice [54]. θP of a diffusion lattice signifies the area fraction of the lattice occupied by immobile completely-reflecting obstacles at and beyond which the possibility of an infinite percolation cluster to exist vanishes. In other words, there is no way left for a diffusing tracer to diffuse/percolate to extremely large distances over the lattice as the time progresses and, rather, gets trapped in small domains or confinements. Therefore, the trapping of tracers observed here at this range of aCRO is technically consistent with the concept of θP of a square lattice. However, for the intermediate range of aCRO = 0.30 − 0.40, the 〈r2〉 initially increases in a nonlinear manner but later adopts a linear profile (Fig 2A).
The effect of varying aCRO on the 〈r2〉 of tracer diffusion becomes more conspicuous by looking at the log-log profiles (Fig 3A). The log-log plots depict almost flat horizontal profile for the lower aCRO and indicates perfectly normal diffusion according to Eq 5. However, increase in aCRO is marked by a brief initial anomalous diffusion of tracers where the log-log profile bears a negative slope (see Eq 5) and a gradual transition to the later normal diffusion. Remarkably, the crossover length, i.e. the 〈r2〉 traversed by the tracer after which the anomalous diffusion turns into normal diffusion, and the associated crossover time of the transition are observed to increase with rise in aCRO (Fig 3A). Only at aCRO closer to or higher than θP, a long-term anomalous diffusion appears where the log-log profile steeply decreases in a linear fashion at longer time. This, in turn, depicts a power-law time-dependence of 〈r2〉 (see Eq 4). Here, the crossover length and time approach infinity.
Using the log-log profiles, the anomalousity of tracer diffusion for the different values of aCRO is computed in terms of the anomalous exponent, α, of 〈r2〉 using Eq 5. It can be easily noted that the flat horizontal log-log plots for lower aCRO have slope zero and, thus, α = 1. For very high aCRO characterized with long-term anomalous diffusion of tracers, the slope of the long-time tail of the log-log plots can be easily used to compute α, which turns out to be close to or equal to 0. However, for the intermediate values of aCRO observed with a transition from anomalous to normal diffusion, α is computed from the linear fitting to the initial segment of the log-log plot within crossover length associated with anomalous diffusion. Consequently, across the increasing aCRO, the anomalousity of tracer diffusion almost exhibits a sharp inverted sigmoidal profile (Fig 4A). There occurs a sudden decline in α (increase in anomalousity) close to aCRO = 0.4, which has been suggested earlier [31] as the switch-like behaviour leading to the trapping of AMPARs.
Further, using the log-log profiles, Deff of tracer diffusion is computed under different conditions of aCRO using Eq 5. The Deff shows a consistent decrease, unlike α, with increase in aCRO (Fig 4D). This indicates that although diffusion remains normal for lower aCRO, the receptor diffusivity indeed decreases with rise in the crowding conditions of completely reflecting obstacles. Under the extreme conditions of confined diffusion, the Deff becomes negligible. Altogether, it must be noted that these observations under the standard condition are identical to that reported earlier in a computational study by Santamaria et al. [31].
The temporal profiles of 〈r2〉 and the associated log-log plots for increasing aCRO under the additional consideration of the different self-crowding conditions of tracers are shown in Figs 2B–2G and 3B–3G, respectively. Particularly looking at the log-log profiles, it is clear that the self-crowding conditions with aDT = 0.00001 and 0.0001 (Fig 3B & 3C) exhibit almost an identical behaviour as well as identical to that noted in the no-self-crowding condition (Fig 3A). Even the anomalousity profile across increasing aCRO under these self-crowding conditions almost mimic that of the no-self-crowding condition (Fig 4A). Therefore, it appears that these self-crowding conditions are associated with sufficiently low tracer density such that they could not noticeably affect the features of tracer diffusion observed under no-self-crowding condition. However, aDT = 0.001 demonstrates a significant intensity of long-term anomalous diffusion across the entire range of aCRO (Fig 3D). Although this behaviour does not become clearly visible across the 2s measurement time duration, longer time duration of 4s makes it clearly visible (Fig 5), where the log-log plots for the selected values of aCRO exhibit a long-time sharp decay profile. Accordingly, the profile of α across increasing aCRO is significantly affected and shifted to lower levels in comparison to that for the lower aDT and no-self-crowding condition (Fig 4A). Further, extremely self-crowded conditions with aDT ≥ 0.005 lead to a very strong long-range anomalous diffusion across all values of aCRO (Fig 3E–3G) and the entire profile of α remains fairly close to zero (Fig 4A).
It can be seen that the range of aCRO = 0.0 − 0.4 associated with normal receptor diffusion (α = 1) is almost invariant (Fig 4A) for the lower aDT = 0.00001 and 0.0001 and the no-self-crowding condition. Moreover, for these aDT and the no-self-crowding condition, the window of aCRO associated with the transition of tracer diffusion from normal to strongly anomalous nature is very narrow. This signifies a sudden rise in anomalousity and a switch-like behaviour for tracer trapping due to high reflecting obstacles’ density. However, for the higher aDT = 0.001, the range of aCRO over which perfectly normal diffusion may occur completely vanishes (Fig 4A) and the window of transition from partially normal to strong anomalous diffusion is also very gradual. For aDT ≥ 0.005, the diffusion remains strongly anomalous irrespective of aCRO (Fig 4A).
To understand more about how the increase in self-crowding affects the tracer diffusion in the presence of CROs, the α is now plotted across the increasing values of aDT for a given value of aCRO (Fig 4B). It is interesting to note that intense self-crowding itself may bring strongly anomalous diffusion even in the absence of reflecting obstacles, as observed here for aDT ≥ 0.005. This contrasts a common fundamental assumption in the earlier theoretical studies [31, 64] that the AMPAR diffusion should be normal in the synaptic membrane in the absence of any non-binding completely-reflecting obstacles. Rather, the results suggest that it may also depend on the AMPAR density in the obstacle-free medium. At the same time, the present observations also support the above assumption to remain valid given the fact that the density of AMPARs in the extrasynaptic membrane is considerably low [4]. Another important thing to be noted is that the profile of α across increasing aDT exhibits a switch like behaviour when aCRO = 0, akin to that observed above in the case of variation in aCRO for lower aDT and no-self-crowding condition (Fig 4B). This switch-like behaviour appears to intensify with increase in aCRO as the transition becomes sharper. However, for aCRO ≥ 0.45, the profile is fairly close to or is identically zero across all values of aDT and the switch-like behaviour completely disappears.
Therefore, in regard of the anomalousity-driven trapping of AMPARs within PSD, self-crowding of AMPARs possibly appears as a new dimension to the causality, which was earlier thought to be driven only by the local macromolecular crowding other than the AMPARs. The cumulative effect of various densities of reflecting obstacles and diffusing tracers on the anomalousity of tracer diffusion observed here is summarized in the heat map shown in Fig 4C. It can be easily noted that, both high aCRO and/or high aDT can lead to strongly anomalous confined diffusion of the tracers. Further, the effect of self-crowding on tracer diffusion is distinguishable only at lower or moderate concentrations of reflecting obstacles and increase in aCRO catalyzes the anomalousity caused by higher aDT. However, for very high CRO concentration, tracer diffusion remains strongly anomalous for all self-crowding and no-self-crowding conditions, owing to the lack of percolation clusters on the 2D square lattice.
To observe the effect of self-crowding on tracer diffusion in terms of the effective diffusion coefficient, Deff is computed from the log-log profiles under the varying conditions of aDT and aCRO. As a matter of fact, for diffusion marked with α equal to or sufficiently close to 1, the computation of Deff is very straightforward (see Eq 5). On the other hand, for receptor diffusion marked with α equal to or sufficiently close to zero, the Deff will be certainly negligible as there occurs no apparent diffusion at a substantial timescale. However, for intermediate values of α, the diffusion is neither perfectly normal nor completely confined and it becomes difficult to conceive a term like a constant diffusion coefficient to describe the 〈r2〉 over the entire duration of time. Under such conditions, the diffusion coefficient becomes time-dependent and is generally described through the two kinds of time-dependent quantities viz. apparent diffusion coefficient and the instantaneous diffusion coefficient [67]. The apparent diffusion coefficient, Dapp, is a time-averaged quantity and signifies the Deff of normal diffusion which could efficiently lead to the identical 〈r2〉 at a given time which one gets through the anomalous diffusion. This is given as,
D a p p ( t ) = D t 1 - α (6)
Here, D is the original constant present in the Eq 4. On the other hand, the instantaneous diffusion coefficient, Dinst, represents the instantaneous rate of change of slope of the nonlinear increase in 〈r2〉 at a given time, which is given as,
D i n s t ( t ) = α D t 1 - α (7)
As evident, both the quantities decrease with progression of time in anomalous diffusion [67].
For the case here, use of Dapp is more suitable as it provides a sense of effective diffusion coefficient which could be used to describe diffusion conditions characterized with the intermediate values of α between zero and one. However, the choice of Dapp would necessarily depend on the time duration for which the process is observed. In the earlier studies [24, 29, 31], distribution of diffusion coefficient is also shown and the statistical parameters such as median diffusion coefficient is computed. Yet, there also the distribution is strictly dependent on the time at which the observation is made and the statistical parameters do temporally evolve. Therefore, the Dapp is computed for the time point of 2s, which is the time duration of diffusion measurement performed in the present study, and will be considered here as the Deff of tracer diffusion characterized with intermediate values of α.
For aDT = 0.00001 and 0.0001, the variation in Deff with increase in aCRO is completely overlapping with that for the no-self-crowding condition (Fig 4D) and, accordingly, the tracers mobility gradually decreases with increase in aCRO. However, the profile for aDT = 0.001 is shifted to slightly lower values depicting reduced mobility due to increased self-crowding of the receptors. Indeed, in this case too, the mobility appears to decrease with increase in aCRO. For the rest very high values of aDT ≥ 0.005, the entire profile of Deff is shifted to extraordinarily low levels (Fig 4D) depicting heavily hampered mobility of tracers owing to steric-exclusion and confinement among themselves as well as in the presence of CROs. Altogether, high density of completely reflecting obstacles and/or tracers engenders reduced mobility and confinement in terms of both the anomalousity as well as effective diffusion coefficient of the tracer diffusion.
In this regard, the earlier experimental studies involving monitoring of the properties of a diffusing entity in the presence of same entity acting as the crowders also appears to strongly corroborate the above observations resulting from the self-crowding. A recent study by Roosen-Runge et al. [68] on the diffusion of bovine serum albumin in the aqueous solution using neutron backscattering has revealed that increase in the volume fraction occupied by the protein (even upto 30%) causes strong decline in the translational diffusion coefficient and leads to shorter-time self-diffusion, implying anomalous nature in action. Similarly, another experimental study by Ramadurai et al. [69] involving fluorescence correlation spectroscopy of the lateral diffusion of a variety of integral transmembrane proteins of different sizes, such as monomeric LacY to trimeric glutamate transporters, at their different density on artificially reconstituted large lipid vesicles demonstrates that increase in the size and density of the subject protein leads to strong decline in the later diffusion coefficient. Further, it has been shown that there occurs a significant decrease in the anomalous exponent of the diffusion at sufficiently high density of the proteins and is evitable even for monomeric proteins, such as LacS. A very recent study by Houser et al. [70] on the lateral diffusion of a homogeneous population of transferrin membrane proteins using fluorescence correlation spectroscopy has also shown that increase in the membrane coverage by the protein leads to strong decline in the diffusivity and has emphasized on the steric-exclusion underlying the self-crowding of the protein. It must be noted that transferrin occupy much lesser membrane area (∼ 24nm2) in comparison of our subject protein, AMPAR.
Therefore, the self-crowding of bulky AMPARs implied here through the tracer diffusion indeed appears to be a significant factor at play in the anomalous diffusion and trapping of these receptors in the PSD, where these receptors are generally present at high density. The SI S1 Video demonstrates the temporal evolution of the position of a diffusing tracer under different self-crowding conditions, but in the absence of any other obstacles, as well as a control condition of free-diffusion.
Three levels of uniform binding energies representing weak (2kBT), intermediate (6kBT) and strong (10kBT) binding of tracers to the binding obstacles (PROs) are separately considered. Given a binding energy, four arbitrary densities of PROs, aPRO = 0.2, 0.4, 0.6 and 0.8, over the lattice are sampled to broadly capture the different situations of the accumulation of scaffold proteins, ranging from sparse to very dense, underneath the PSD. Subsequently, these combinations are examined for the different conditions of self-crowding of tracers. In this part of the study, reflecting obstacles are completely absent and only the role of binding obstacles in shaping the nature of tracer diffusion is examined.
The features of tracer diffusion under no-self-crowding condition is surely monotonous in the presence of binding obstacles. The tracer diffusion is always perfectly normal for all binding energies and values of aPRO, as the log-log profiles (Fig 6) remains fairly horizontal along the entire duration of diffusion monitoring and the α remains strictly close to one (Fig 7). However, for a given binding energy, the log-log profile shifts to lower values with increase in aPRO. Increase in binding energy further lowers the levels of these log-log profiles. This has implications in the decline of tracer mobility in terms of Deff. Accordingly, the Deff exponentially decreases with increase in aPRO for a given binding intensity (Fig 8). Moreover, increase in binding intensity shifts the Deff profile to lower orders of magnitude, depicting further decline in tracer mobility. These observations for no-self-crowding condition are equivalent to that observed in the computational study by Sanatamaria et al. [31]. Further, the absence of anomalousity in tracer diffusion in the presence of a wide range of PRO density and binding energy is also consistent with the previous study of anomalous diffusion in the presence of binding performed by Saxton [55], where it is implied that simple valley models of tracer binding always leads to normal diffusion under thermally-equilibrated initial condition.
The self-crowding conditions with aDT = 0.00001, 0.0001 and 0.001, are found to exhibit behaviors identical to that under the no-self-crowding condition. For a given binding energy and aPRO, the log-log plots across these self-crowding conditions are strongly overlapping with that of the no-self-crowding condition (Fig 6). Accordingly, diffusion is normal across all the values of aPRO and the levels of binding energies with α close to one (Fig 7). Further, the Deff for these self-crowding conditions demonstrate a consistent decrease in the tracer mobility with increase in binding energy and PRO density (Fig 7). Also, the Deff profiles are sufficiently overlapping for these conditions of self-crowding as well as no-self-crowding. Therefore, it appears that the increase in tracer density to 0.001 has no distinguishable effect on the tracer diffusion in the presence of binding obstacles. Rather the diffusion is being mainly governed by the PRO density and the binding energy.
On the other hand, the self-crowding conditions with aDT ≥ 0.005 exhibit a peculiar behaviour. For weak binding events, these self-crowding conditions clearly demonstrate a strong long-range anomalous diffusion for lower PRO density, aPRO = 0.2, (Fig 6A) and the values of α are close to zero (Fig 7A). However, as the PRO density is increased, the anomalousity of diffusion gradually reduces and the log-log profiles tend to approach normal diffusion behaviour. For the case of aDT = 0.005, diffusion becomes fairly normal at aPRO = 0.8 (Fig 6D) and α reaches 1 (Fig 7A). Tracer diffusion for aDT = 0.01 also tends to acquire normal behaviour with rising aPRO, though there remains slight anomalousity even at aPRO = 0.8. However, for = 0.1, the diffusion remains strongly anomalous even at = 0.8 (Figs 6D & 7A). An important thing to observe is that the log-log profile of normal diffusion that the anomalous tracer diffusion for aDT = 0.005 and 0.01 gradually approaches (Fig 6D)with increase in aPRO appears to overlap with that obtained for aDT ≤ 0.001 as well as for the no-self-crowding condition.
Remarkably, aDT = 0.005 and 0.01 consistently exhibit normal diffusion across all PRO densities for the intermediate and strong binding energies, as their log-log plots (Fig 6E–6L) remain horizontal with α = 1 (Fig 7B & 7C). Moreover, these log-log plots fairly overlap with that of the lower self-crowding conditions under the respective conditions of binding energies and aPRO. However, for the intermediate binding intensity, aDT = 0.1 exhibits significant anomalous diffusion for lower aPRO = 0.2 (Figs 6E & 7B). The anomalousity soon vanishes for aPRO ≥ 0.6 (Fig 6G & 6H) and α reaches 1 (Fig 7B). For strong binding intensity, aDT = 0.1 exhibits perfectly normal diffusion for all values of aPRO and identical to the lower self-crowding conditions (Figs 6I–6L & 7C).
At this point, if we remind the observations regarding tracer diffusion in obstacle-free medium, aDT ≥ 0.001 demonstrated a marked long-range anomalous diffusion (Fig 3D–3G) with significantly low α (Fig 5A–5C). Together with the observations made here in the presence of binding obstacles, it is strongly evident that increase in binding phenomenon, either through increase in PRO density or/and increase in binding energy, reduces the anomalousity in tracer diffusion arising from higher self-crowding. However, it is also observed that the intensity of amelioration of the anomalousity with increase in binding further depends on the intensity of self-crowding. Very intense self-crowding conditions would require a considerably large increase in binding energy and scaffold density to exhibit perfectly normal diffusion. As shown here, for strong tracer-PRO binding, receptor diffusion is completely governed by binding obstacles’ density, regardless of the self-crowding conditions.
Nonetheless, for weak binding intensity, tracers mobility in terms of Deff for aDT = 0.005 initially increases with increase in aPRO and later decreases along the profiles obtained for lower aDT (Fig 8A). However, for aDT = 0.01, Deff consistently increases with increase in aPRO. The increase in Deff under these conditions of aDT owes to the concomitant relaxation of anomalousity of tracer diffusion. For aDT = 0.1, Deff remains significantly close to zero for all values of aPRO (Fig 8A) due to strongly anomalous tracer diffusion. For intermediate binding intensity, the profile of variation in Deff for aDT = 0.005 and 0.01 is identical to that of the lower self-crowding as well as no-self-crowding conditions (Fig 8B). However, for aDT = 0.1, Deff sharply rises with increase in aPRO but soon gets along the decreasing profiles obtained for lower values of aDT. For the strong binding intensity, all conditions of aDT exhibit an identical profile of decrease in Deff with rise in aPRO (Fig 8C).
Therefore, increase in binding indeed ameliorates anomalousity of tracer diffusion arising from the self-crowding of the tracers and stronger binding favors normal diffusion even under high tracer density. However, in regard of Deff, increase in binding consistently reduces the mobility for low tracer density. But for high tracer density, increase in binding leads to higher mobility in the situation where concomitant reduction in the diffusion-associated anomalousity is observed. Otherwise, given a sufficiently strong binding condition, any further increase in binding leads to consistent reduction in the tracers mobility.
The present study deals with the aspect of how self-crowding of mobile bulky AMPARs may affect their lateral diffusion at different densities of the receptors in the postsynaptic membrane of the excitatory synapses and the way in which presence of obstacles and binding elements in the PSD may further influence the effect of self-crowding. In light of the above observations obtained through the Monte-Carlo simulation using the lattice model of diffusion of the representative point tracers, it would be reasonable to state that the density of AMPARs may significantly influence the nature of their diffusion and very high density may lead to strongly anomalous confined diffusion even in the absence of any other obstacles in the membrane. The presence of other transmembrane obstacles may further accentuate the appearance of anomalousity arising from the self-crowding of the receptors. Conversely, self-crowding may cooperate in the trapping of AMPARs effectuated by the intense crowding of transmembrane proteins in the PSD. However, partially-reflecting and binding scaffold proteins lying submembranously within PSD region may serve a contrary role where increase in binding, either through increase in the density of scaffold proteins or increase in the AMPAR-scaffold binding energy or both, reduces the anomalousity in receptor diffusion arising from their self-crowding. Therefore, in the context of anomalousity, the transmembrane obstacles and the binding submembranous scaffold may behave as the two opposing forces.
In the context of effective diffusion coefficient of the AMPARs, the receptor mobility may strongly decrease with rise in the self-crowding of the receptors, in concordance with the increase in anomalousity. And, the presence of transmembrane obstacles may further lead to the decrease in Deff. However, binding may have differential impacts on the receptor mobility and it depends on the intensity of self-crowding. For low and moderate intensity of self-crowding, increase in binding may consistently lead to decrease in the receptor mobility. However, for intense self-crowding conditions involving strong anomalousity of receptor diffusion, increase in binding may increase receptor mobility as long as it is associated with the reduction of anomalousity. Otherwise, once the diffusion acquires a normal behavior under sufficiently strong binding conditions, further increase in binding may lead to decline in the Deff. Yet, despite this differential behavior, binding can be considered to cause, in general, decrease in receptor mobility in terms of its Deff.
Noting these contrary implications of binding in anomalousity and effective mobility of the receptor, a genuine concern arises that which among these two features matters the most in regard of hampering the receptor diffusion. Indeed, it is the anomalousity which marks the most dominant contribution to declined receptor diffusion. If the diffusion is normal, a particle is at least able to diffuse (the MSD grows consistently with time) and escape a defined region of interest, no matter slowly due to a low diffusion coefficient. But in the case of anomalous diffusion, the Deff continues to decline to zero as the time progresses (see Eq 6) and the amount of decrease in Deff is directly proportional to the decrease in the anomalous exponent of the diffusion [66, 67]. Accordingly, it is possible that the particle may never escape a defined region under strongly anomalous condition. This is the reason that a diffusing particle exhibiting small anomalous exponent is widely referred to as confined within a region of certain confinement length, as the MSD comes to settle down at a plateau with time. Therefore, it must be recognized that, although binding reduces the receptor mobility, it certainly assists in bringing out the diffusing receptors from the more restricting situation of anomalous diffusion arising from their self-crowding and getting trapped.
In the case of obstacle-free diffusion of tracers, increase in the tracer density is responsible for the more frequent self-crowding collisions among the tracers and the resulting obstructions of their diffusion paths. Through the above observations, it is realized that, for the given specifications of the lattice dimension, a rise in tracer density (aDT) to the order of 0.001 commences the appearance of anomalousity in tracer diffusion and further increase in the tracer density leads to its more noticeable magnitude. The appearance of anomalousity can easily be better portrayed under an assumed condition of extreme self-crowding when aDT is sufficiently close to 1 and almost all lattice sites are occupied with the tracers. At every time-step of simulation, the hopping of a tracer to any random direction would be denied because the neighboring destination sites in almost all directions are occupied with the tracers. This will repeatedly occur across sampling of the entire population of tracers and, as a consequence, the tracers would remain stuck at their positions along a unit advancement in time. This condition would remain unchanged for every further time-steps and the MSD would not increase with time, depicting extraordinarily strict confinement of the tracers. It can now be extrapolated for the lower tracer densities that the confinement would be certainly reduced but the abundance of restricted diffusion would accordingly lead to anomalousity. Need not to say that, for the no-self-crowding condition, diffusion of single tracer on the obstacle-free lattice would always remain perfectly normal.
When reflecting obstacles (CROs) are added to the system, the unoccupied fraction of the lattice sites connected to each other through the diffusive edges decreases. In fact, this decrease in percolation paths is significant only when the CRO density reaches the percolation threshold of the square lattice. This is the reason that, for the conditions of single tracer diffusing in the lattice frame with no self-crowding at all or low tracer density with insignificant counts of self-obstructions during diffusion, what only shapes the nature of tracer diffusion is the extent to which CRO density is close to or beyond the percolation threshold. However, when tracer density is sufficient to effectuate a considerable amount of self-crowding against their free diffusion, slight decrease in percolation paths even at much lower CRO density can exacerbate the anomalousity of diffusion arising from the self-crowding. Furthering this description to the conditions of extreme self-crowding at very high tracer density, a situation appears where even in the absence of reflecting obstacles, the anomalousity of tracer diffusion is close to its possible maximum level and adding reflecting obstacles does not manifest into any significant change.
Unlike the above cases, stating the exact mechanism involved in the observed effects of increase in binding on the self-crowded diffusion of the tracers is not so straightforward. Therefore, the attempt here would be to carefully and systematically deduce the plausible mechanism, while keeping in mind the specific arrangements utilized in the above simulation experiments and the features associated with them in the background. If aPRO is the fraction of lattice sites occupied by binding obstacles (PROs), (1 − aPRO) is the fraction unoccupied by them. It essentially results into a partition of the lattice medium into two spatial subsets viz. non-binding and binding spatial subsets. The latter is capable of binding a diffusing tracer and freezing it at its location for a random size of waiting time. Notably, the mean waiting time is directly proportional to the intensity of binding such that higher is the binding energy, longer is the mean waiting time. Another important thing to note is the size of binding subset relative to the non-binding subset. Higher is the aPRO, larger is the binding subset.
At the beginning of the simulation, when tracers are uniformly distributed over the lattice with the desired area fraction aDT, aDT aPRO would be the fraction of aDT lying on the PROs and, thus, lying in the binding spatial subset whereas aDT(1 − aPRO) will be lying on completely empty nascent lattice sites and, thus, belongs to the non-binding spatial subset. As the lateral diffusion proceeds, there occurs diffusion of tracers within their own spatial subsets as well as diffusion-associated exchange of tracers across the subsets. Due to reduced tracer mobility in the binding subset, there would occur an initial drift of a certain fraction of the tracer population belonging to the non-binding subset towards the binding subset acting as a sink, until a thermal equilibrium is achieved. Higher is the binding energy and larger is the binding subset, the thermal equilibrium would be acquired with a larger fraction drifted. Once such an equilibrium distribution of the tracer population between the two spatial subsets is achieved, contribution of each population to the anomalousity of entire tracer diffusion can be easily compartmentalized and examined.
The process of equilibrium distribution engenders two consequences for the tracers belonging to the non-binding spatial subset. First, the resultant density of tracers within the non-binding subset is significantly reduced leading to a reduced self-crowding condition. Second, the binding subset-associated larger population of tracers appear as almost static reflecting obstacles (CROs) to the highly mobile tracers belonging to the non-binding subset. Here comes the role of longer mean waiting time under stronger binding condition which leads to larger decline in the hopping rate of the tracers belonging to the binding spatial subset. Therefore, under strong binding conditions, the entire diffusion system for the tracers associated with non-binding spatial subset turns into the diffusion of tracers at low density but in the presence of less or moderately dense CROs. And, according to the previous experiences with the reflecting obstacles, the contribution to anomalousity from the diffusion of unbound tracers is severely reduced. One can now envisage that decrease in binding will certainly violate this setup by bringing more self-crowding encounters amongst the tracers and their resulting anomalousity would be higher.
On the other hand, the diffusion of tracer population belonging to the binding subset within its own spatial subset appears, according to the results, less anomalous under stronger binding conditions. It seems that declined rate of hopping is beneficial in reducing anomalousity by frequently avoiding self-crowding encounters. This is even helpful for the case of encounters with highly mobile tracers belonging to the non-binding spatial subset, which are themselves in lesser density too. However, binding certainly reduces the mobility of the tracers belonging to binding spatial subset. Corollary, lesser and weaker binding would increase the tracers mobility but would concomitantly cause more frequent self-crowding encounters within the binding spatial subset as well as across the non-binding subset and lead to higher anomalousity. This entire description of the possible mechanism concludes at one interesting fact that self-crowding collisions are the main source of anomalousity. Although binding reduces the effective mobility of the tracers, it ameliorates anomalousity by avoiding such collisions. Therefore, the phenomenon of binding plays its role at a trade-off point between the effective mobility of the tracers and the anomalousity of their diffusion.
In the present study, the diffusing tracers are point particles diffusing on the lattice framework. Therefore, one may argue over how the self-crowded diffusion of point tracers may capture the crowded diffusion of AMPARs, which are bulky transmembrane structures with non-zero lateral span. The reply to this question is hidden in the description of area-fraction of the lattice occupied by the point tracers and the use of ensemble-averaged MSD. The extracellular domain of an AMPAR is the most bulky structure with lateral dimensions of length 16nm and width 8nm [35, 42]. Therefore, its two-dimensional projection on the lipid membrane would occupy a surface area of roughly 128nm2. For the purpose of realizing side-ways collisions during lateral diffusion, the complex details of an AMPAR structure can be essentially reduced to a transmembrane cylindrical structure [29] of radial width 6nm. This lateral radial span characterizes the exclusion area (128nm2) which avoids approach of another receptor closer than this radius and presumably reflects it away in an elastic manner. Certainly, association of the receptor with other auxiliary proteins [49–51] would further stretch the exclusion area, as it becomes more bulky along the lateral dimension. Given the density of the AMPARs and the areal span of the PSD or a subregion within the PSD, one may easily procure the resultant fraction of the PSD area occupied by the total exclusion area of the receptor population. This fraction amounts to the area-fraction of the self-obstructing point tracers, aDT, on the lattice referred here. Nevertheless, this approach gets complete only when ensemble-averaged MSD of the tracers is used to capture the bulk diffusion properties of AMPARs. Had it been time-averaged MSD observation of single tracers, the statistical approximation using aDT would not suffice to fully reproduce the time-averaged MSD behaviour of the non-zero size AMPARs [71].
For instance, the density of AMPARs in the extrasynaptic membrane has been experimentally measured to be 3–5μm−2 [4]. The estimated length scale of the region of extrasynaptic membrane on the spine head is approximately 1μm [72] and an effective surface area close to 1μm2. Therefore, the fraction of extrasynaptic region occupied by the total exclusion area of the receptor population would be 0.00038 − 0.00064. Given this fraction as aDT in the present study, it is shown that the tracer diffusion would be perfectly normal in the absence of any obstacle. The same is observed in the particle tracking experiments [24–29] and the receptor diffusion is normal in the extrasynaptic membrane, which contains least transmembrane obstacles [27]. However, within the PSD of typical radial size 100nm [72, 73], the AMPAR count may range from 20–100 [74] which is equivalent to receptor density 650–3000μm−2 [74, 75]. These estimates lead to an area fraction of 0.08 − 0.4 of the PSD to be occupied by the total exclusion area of the receptor population. Given this fraction as aDT, it is shown here that self-crowding of the tracers would immensely contribute to the anomalousity of tracer diffusion and their confinements.
Nonetheless, the convergence of lattice model of diffusion under the described conditions of reflecting and binding obstacles to continuous-space diffusion becomes important to be investigated. In regard of the earlier studies on AMPAR diffusion in the absence of self-crowding interactions amongst the receptors, a very recent study by Li et al. [29] has used the approach of continuous-space diffusion on the basis of Monte Carlo simulation of the Langevin dynamics. Using photoactivated localization microscopy (PALM) technique, the spatial distribution of the submembranous PSD-95 binding obstacles was determined and the other reflecting transmembrane obstacles in the simulation space were distributed accordingly. The findings of their study strongly asserts the observations made in an earlier lattice model-based work by Sanatamaria et al. [31] from the same research group. The present study additionally raises a significant factor of self-crowding in shaping the receptors diffusion. Hence, what appears important is to show here how convergent is the lattice-diffusion approach and the proposed recursive algorithm to the self-crowded continuous-space diffusion.
Accordingly, the Monte-Carlo simulation of the Langevin dynamics of receptor diffusion is performed with steric-exclusion under the vibrant conditions of self-crowding. In this approach, the AMPARs are modelled as flat circular disks of exclusion radius 6nm. The centre of the disc is moved in random directions over a Δt time-step of simulation as Δ x = 2 D Δ t ξ ( t ) and Δ y = 2 D Δ t ξ ( t ). Here, ξ(t) represents white Gaussian noise with mean zero and variance 1. The 2 D Δ t defines the standard deviation of the random-sized steps taken independently in x− and y− directions over single time-steps of the simulation. D is the free diffusion coefficient of AMPARs in the postsynaptic membrane. While diffusion, it is taken care that the centres of any two discs should not be at a relative distance shorter than the double of the radius of the discs to implement steric-exclusion. This minimum distance between the centres of two discs represent collision between the incompressible hard discs. Further, the collisions are considered elastic. The number of these circular disks in a square simulation space of area 1μm2 is computed from the desired area fraction aDT under investigation. The simulations begin with all the disks uniformly distributed across the simulation space with non-overlapping steric conditions. Three kinds of self-crowded conditions with sparse (aDT = 0.0001), fairly dense (aDT = 0.01 and 0.1) and extremely dense (aDT = 0.6) presence of AMPARs are enquired. These three kinds of self-crowded conditions are chosen in accordance with the order of AMPAR density typically observed in the extra-synaptic membranes and the PSD, as mentioned above. The presence of any other obstacles is not considered.
From the continuous-diffusion scheme, the log-log plots of the MSD are obtained by averaging over a sufficiently large size (700) of ensemble of the independent simulations. Based on this, the convergence of the newly-proposed lattice-based recursive algorithm for self-crowded diffusion to the continuous-space diffusion is checked and its efficacy over the other possibility without involving the recursion or repeated check of the labelled “DT-blocked tracers” in the same algorithm is also evaluated. Fig 9 demonstrates the overlaid normalized MSD plots obtained from the continuous-diffusion simulation, recursive lattice-based algorithm and the same algorithm without recursion under the different conditions of self-crowding. Interestingly, it is consistently observed that lattice-diffusion scheme with recursive algorithm is quantitatively sufficiently close to the MSD-profiles of the continuous-space diffusion in comparison to that without involving the recursive algorithm. For very low (aDT = 0.0001) as well as very high (aDT = 0.6) self-crowding density, it is quite apparent that the recursive algorithm and its absence are providing fairly identical convergence to the continuous-space model. Such convergence of the two lattice-diffusion schemes under very low density of tracers may be due to the lack of a substantial frequency of self-obstructing events. However, the same observed under very dense crowding of the tracers may arise from the fact that the fraction of false self-blocking events is negligible amidst the very frequent steric-collisions among the receptors, as most of the tracers are unable to diffuse under such strongly crowded condition. It is only for the intermediate densities (aDT = 0.01 and 0.1) of the tracers that the distinction between the MSD profiles obtained from the schemes with and without the recursive algorithm is more conspicuous. It describes that majority of the obstructions observed on first-attempt through the computational sequence of hopping appears to be false and, as a result, the observed log-log plot appears steeper (i.e. more anomalous and obstructed) than that obtained from continuous-space schemes and recursive algorithm. However, when such obstructions are checked back in a recursive manner, it leads to their diffusion. Here, the behaviour of receptor diffusion in the absence of recursive algorithm is substantially deviated from that obtained from the continuous-space diffusion. More specifically, the deviation is higher for aDT = 0.01 in comparison to aDT = 0.1.
Seeing these remarkable consequences, an important question appears: why has the factor of self-crowding of the bulky AMPARs remain unappreciated till now when we already have a sufficiently large body of experimental data on the nature of AMPAR diffusion at excitatory synapses? Possibly, the reason to this ignorance does not entirely or essentially owe to the experimental studies. In fact, the in vivo sophisticated microscopic tracking of endogeneously-expressed AMPARs or less bulky genetically-engineered transmembrane probes at excitatory synapses and the resulting observations regarding their anomalous diffusion within PSD indeed involve all the several factors which are simultaneously present there under the real physiological conditions of the experiments [24–29]. However, the mechanistic deduction of the effects of these pertinent factors to the finer details is beyond the scope of any existing experimental techniques. At this point, the theoretical studies [29–33] using detailed models of the receptor diffusion in the presence of PSD crowd comes forth as the only but efficient option to dig deeper into the mechanisms. Certainly, these studies have so far led us to realize the impacts of obstruction and binding by the local crowd of transmembrane and submembrane scaffold proteins on receptor diffusion in the PSD. On the basis of these factors, the previous experimental data on the tracking of receptor diffusion has also been explained to a great extent. Therefore, the existing ignorance towards the possibility of an additional role of the self-crowding of receptors owes merely to the lack of consideration of the self-crowding in the earlier theoretical approaches.
Yet, the possible contributions of self-crowding could be unknowingly by-passed in the earlier theoretical approaches by appropriate parameter estimation within the framework of the previous models and it is a strong possibility that self-crowding remained a hidden variable in the process. For instance, even in the detailed study by Li et al. [29], simulations used to describe the monitored diffusion of genetically-engineered single- or double-pass transmembrane probes using FRAP as well as sptPALM techniques considered a substantial fraction of AMPARs endogenously-expressed in the cultured hippocampal neurons as the part of static crowd only. It must be noted that transmembrane crowds like AMPARs are sufficiently mobile and may impact differently from the other relatively static crowd on the probe diffusion. Nonetheless, this immediately draws attention to the fact that the experimental data on the receptor diffusion should also contain the elements of self-crowding, besides the earlier recognized factors. To throw light on this aspect, some of the previous MSD data on the AMPAR diffusion at excitatory synapses are examined on the basis of the observations acquired in the present study and is discussed in the following subsection.
As stated above, the commonly observed density of AMPARs in a typically-sized PSD would lead to an occupied area fraction ranging 0.08 − 0.4. The present observations suggest that receptor diffusion would be strongly anomalous at this level of area fraction due to self-crowding, regardless of the other transmembrane obstacles. This leads to a confusing situation where importance of transmembrane obstacles becomes obsolete, whereas the earlier studies have shown that the steric repulsion by the obstacles is an indispensable and critically essential factor behind AMPAR trapping and accumulation within the PSD. This contradiction arises because of the difference between the configuration of the diffusion system employed here and that of the system under natural condition. The diffusion system used here has a periodic boundary condition at its edges, leading to a homogenous condition of obstruction or binding applied on a diffusing entity throughout an infinite two-dimensional space. On the other hand, receptors diffusing within the PSD at excitatory synapses can easily escape the local crowded condition by entering into the extrasynaptic space and, hence, the natural diffusion system is an open system.
Therefore, the configuration of the present system mainly captures the diffusive behaviour of a receptor as long as it is diffusing within the PSD region and the density of the receptors is in a perfect or quasi- steady state. This leads to the speculation that self-crowding of AMPARs cannot itself hold the accumulated density of the receptors if the steric repulsion by the other obstacles are completely removed. Rather, steric obstructions by the relatively static density of other transmembrane proteins may provide the initial as well as maintaining driving force by reducing the mobility of the receptors within the PSD and self-crowding may later come into action as the density rises to a certain required level. In fact, this might be possible as increase in reflecting obstacle density leads to consistent decrease in the Deff but a sudden increase in anomalousity occurs only beyond a certain very high CRO density. This speculation would have a remarkable impact on the required concentration of transmembrane obstacles predicted theoretically to effectuate anomalous confined diffusion of the receptors within PSD.
Fitting to the data on the diffusion of AMPARs in synaptic and extrasynaptic spaces acquired in the experimental study by Li and Blanpied [76] using single particle tracking and localization microscopy provides α = 0.22 and 0.99, respectively (Fig 10A). In the single-particle tracking experiment by Renner et al. [24] using quantum dots, two kinds of trajectories of the AMPARs diffusing in the PSD region were observed(Fig 10B). AMPARs, referred to as trapped, retained for longer durations within the PSD and exhibited strongly anomalous subdiffusion. The other population of AMPARs, referred to as passing, stayed for relatively shorter duration within the PSD but exhibited only a slightly lesser anomalous diffusion in comparison to the trapped receptors. Fitting to the MSD data of trapped and passing receptors provided α = 0.48 and 0.5, respectively. In a similar manner, through the fitting to the data on receptor diffusion within synaptic region obtained in the study by Renner et al. [77], the α comes out to be 0.42 (Fig 10C).
If the self-crowding factor is not considered, such high anomalousity of receptor diffusion could be possible only at an obstacle density (aCRO) between 0.4 and 0.44. Even, the theoretical study by Santamaria et al. [31] predicts a similar range of obstacle concentration (0.4 − 0.46) for achieving such low anomalous exponent of AMPAR diffusion. However, introduction of self-crowding can bring similar high levels of anomalousity even at lower levels of obstacle concentration (see Fig 4). Imagining that expression of transmembrane obstacles at a density lesser than the theoretically-predicted value would cause sharp loss in accumulated AMPAR density seems very strict and unrealistic for the natural scenario. In fact, self-crowding may provide a certain degree of flexibility to this aspect of synaptic homeostasis. This feature can be tested through an experiment where the nature of AMPAR diffusion and receptor accumulation within the PSD is examined under different densities of transmembrane obstacles. If a significantly anomalous receptor diffusion is observed at an obstacle density ammounting to occupied area fraction lesser than the abovementioned, it would be a strong evidence for the speculation drawn here for the self-crowding of the AMPARs.
Further, the average density of PSD-95 scaffold proteins in the PSD of an excitatory synapse is known to be 3000μm−2 [78]. The radial size of a PSD-95 protein is estimated to be almost 2.5nm [35], which results into a lateral span of 19.64 × 10−6μm2. Assuming a homogenous distribution of the PSD-95, the area fraction of the PSD occupied submembranously by the total PSD-95 proteins would amount to 0.059. Knowing that AMPARs are present at very high density within the PSD [74, 75], the present study suggests that binding in the presence of such low area fraction of PSD-95 would cause an insignificant effect on the anomalousity of receptor diffusion, unless the AMPAR-PSD-95 binding affinity is extremely high throughout the binding sites. Interestingly, this has also been noted in the earlier experimental study by Li and Blanpeid [76], stating that whole-synapse PSD-95 density would have inconsiderable impact on the diffusion of transmembrane proteins.
However, the experimental estimation of PSD-95 distribution demonstrates that, rather than homogeneously distributed, these proteins are enriched in smaller subregions or nanodomains within the PSD [52]. Accordingly, their local density and occupied area-fractions within these nanodomains may acquire considerably large magnitudes, such that even moderate binding affinity may appear effective in reducing the anomalousity of receptor diffusion at high receptor density within the nanodomains. Therefore, such PSD-95 distribution appears as a physiological strategy to enhance the effectiveness of binding on the AMPAR diffusion and exchange with the perisynaptic space.
Nonetheless, AMPARs have also been observed to accumulate at higher density within the nanodomains [52, 79]. The present observations suggest that, given PSD-95 density and AMPAR-binding affinity, the extent to which the anomalousity arising from the self-crowding would decrease further depends on the local receptor density within these domains. As the area fraction occupied by the higher density of AMPARs within the much smaller (∼ 80nm, [79]) PSD-95-rich domains would be very large, receptor diffusion within scaffold-rich domains would retain significant anomalousity, despite the ameliorating influence of binding on the anomalousity. In fact, experimental observations using super-resolution microscopy in the study by Hosy et al. [79] on the nature of AMPAR diffusion within these nanodomains and outside (Fig 10D) indicate that the AMPAR diffusion within these PSD-95-rich nanodomains indeed exhibit strongly anomalous diffusion. Fitting to their data on receptor diffusion within the nanodomains and in peripheral region provides α = 0.25 and 0.85, respectively. It must be noted that nanodomains do contain transmembrane obstacles [5], for instance the adhesion proteins LRRTM2 [80]. However, the AMPARs are present at an extraordinarily high density in these nanodomains and self-crowding appears to be a prominent factor underlying the intense anomalousity in receptor diffusion observed earlier [79].
In a recent study by Li et al. [29] using genetically-engineered single-pass transmembrane probes, it has been observed that rapalog-mediated cross-linked binding probes with intracellular PDZ-binding segments exhibit more anomalous and confined diffusing than the single binding probes (Fig 10E). Fitting to the data provides α = 0.34, 0.28 and 0.17 for the single probes, binding-non-binding cross-linked probes and binding-binding cross-linked probes, respectively. In a straightforward manner, it appears that enhanced binding with multiple PDZ-domains enhances anomalousity of receptor diffusion. However, it can be also be possible that increase in binding lowers the receptor mobility and causes accumulation of larger number of AMPARs in the local area. This may lead to increased self-crowding in a feedback manner and fuels stronger anomalousity to the receptor diffusion trapping more receptors.
Nonetheless, the major practical limitation towards drawing a definitive conclusion is that the earlier experimental studies haven’t taken into account the density of AMPARs at the PSD while tracking their diffusive properties and, in the absence of such mentions, it is difficult to establish the connection between the exact contribution of self-crowding of AMPARs to their overall nature of diffusion within the PSD. Besides this, the present study is also in its preliminary state and contains some important limitations when it comes to closely imitate the true biological condition. Given a population of AMPARs, the receptors may be in different states which may affect the strength of binding to the scaffold proteins, such as association with different kinds of auxiliary transmembrane proteins [39, 45] or no association at all [56, 57], glutamate-bound desensitized state of the receptor [81], differences in the cytoplasmic domains of receptor subtypes [56, 57] etc. The phosphorylation state of the auxiliary transmembrane protein, such as Stargazin [82] or that of the scaffold proteins, such as PSD-95 [83], also significantly affects binding. Together, these factors may lead to heterogeneous distribution of binding strength across the PSD. Further, the spatial distribution of the various scaffold proteins is inhomogeneous and smaller sub-regions within the PSD are found to have higher density of these proteins [52, 53]. In a similar manner, various transmembrane proteins such as N-cadherin, have specific spatial distributions [29, 35] rather than a perfectly homogeneous distribution in the PSD assumed here. Further, the presence of extracellular matrix in the synaptic cleft has also been observed to affect the lateral diffusion and accumulation of AMPARs in the PSD [84].
However, despite these limitations at the level of finer details, the present investigation indeed serves as an initial step towards gaining insight into the aspect of self-crowding of AMPARs, and similar other mobile bulky transmembrane proteins, and its effect on lateral diffusion in the postsynaptic membrane as well as in the specialized crowded PSD region. These features may serve as the conceptual nut-bolts for understanding the behaviour of more detailed models capturing the true irregular topology of synaptic PSD and the natural spatial distributions of the crowding and binding elements. Further investigations may involve the effects of self-crowding on the dynamics of AMPAR trapping and accumulation under the conditions of house-keeping maintenance of synaptic strength and evoked synaptic plasticity.
|
10.1371/journal.pcbi.1005151 | High-Accuracy HLA Type Inference from Whole-Genome Sequencing Data Using Population Reference Graphs | Genetic variation at the Human Leucocyte Antigen (HLA) genes is associated with many autoimmune and infectious disease phenotypes, is an important element of the immunological distinction between self and non-self, and shapes immune epitope repertoires. Determining the allelic state of the HLA genes (HLA typing) as a by-product of standard whole-genome sequencing data would therefore be highly desirable and enable the immunogenetic characterization of samples in currently ongoing population sequencing projects. Extensive hyperpolymorphism and sequence similarity between the HLA genes, however, pose problems for accurate read mapping and make HLA type inference from whole-genome sequencing data a challenging problem. We describe how to address these challenges in a Population Reference Graph (PRG) framework. First, we construct a PRG for 46 (mostly HLA) genes and pseudogenes, their genomic context and their characterized sequence variants, integrating a database of over 10,000 known allele sequences. Second, we present a sequence-to-PRG paired-end read mapping algorithm that enables accurate read mapping for the HLA genes. Third, we infer the most likely pair of underlying alleles at G group resolution from the IMGT/HLA database at each locus, employing a simple likelihood framework. We show that HLA*PRG, our algorithm, outperforms existing methods by a wide margin. We evaluate HLA*PRG on six classical class I and class II HLA genes (HLA-A, -B, -C, -DQA1, -DQB1, -DRB1) and on a set of 14 samples (3 samples with 2 x 100bp, 11 samples with 2 x 250bp Illumina HiSeq data). Of 158 alleles tested, we correctly infer 157 alleles (99.4%). We also identify and re-type two erroneous alleles in the original validation data. We conclude that HLA*PRG for the first time achieves accuracies comparable to gold-standard reference methods from standard whole-genome sequencing data, though high computational demands (currently ~30–250 CPU hours per sample) remain a significant challenge to practical application.
| Determining an individual’s HLA type (the sequence of the exons of the HLA genes) is important in many areas of biomedical research. For example, HLA types shape immune epitope repertoires, which are relevant in cancer immunotherapy, and influence autoimmune and infectious disease risk. Whole-genome sequencing data, currently being generated for hundreds of thousands of individuals, contains the information necessary for HLA typing–but inferring accurate HLA types from these is a challenging problem. First, the HLA genes are the most polymorphic genes in the human genome; second, these genes and their variant alleles exhibit high degrees of sequence similarity (due to a shared evolutionary origin). This makes it difficult to establish which specific HLA gene a given observed sequencing read derives from. We show that this problem can be addressed using a Population Reference Graph (PRG): for each gene, the PRG contains not only the reference sequence but also variant alleles, thus enabling, using a novel sequence-to-graph mapping algorithm, the accurate mapping of reads to HLA genes. We also show that HLA*PRG, the algorithm implementing our approach, achieves–based on standard whole-genome sequencing data–accuracies comparable to those of specialized gold-standard methods. HLA*PRG is open source and freely available.
| Genetic variation at HLA loci, both classical and non-classical, is associated with many medical phenotypes including risk of autoimmune [1–3] and infectious [4] disease, adverse drug reactions [5, 6], success of tissue and organ transplants [7], and, via epitope presentation preferences, the success of cancer immunotherapy [8]. The current gold standard for high resolution typing of HLA alleles, sequence-based typing (SBT), uses Sanger sequencing or targeted amplification of the HLA genes followed by next-generation sequencing. With the growth of high throughput genomic technologies, methods for inferring HLA genotype have been developed that use SNP genotyping [9–12] or next-generation sequencing [13–19]. These approaches, to date, are either limited to a subset of HLA loci, require targeted capture / amplification, or do not achieve the same degree of accuracy as SBT.
The main challenge in characterising the HLA genes from next-generation sequencing data is the correct alignment of sequencing reads. Multiple factors influence accuracy, including the sheer sequence and structural diversity of the region, the presence of multiple paralogous genes (including pseudogenes) and rare, but important, gene conversion events that generate mosaic allelic structures. The high degree of sequence similarity between alleles in certain groups of loci and its non-random spatial structure are illustrated in S1 Fig and S2 Fig.
To address these challenges, we have previously introduced structures to represent known genomic variation called population reference graphs (PRGs) and demonstrated their value in characterising variation across the major histocompatibility complex (MHC) and particularly within the HLA class II gene region [20]. Briefly, a PRG is a directed graph in which alternative alleles, insertions and deletions are represented as alternative paths through the graph, and in which orthologous and identical regions are collapsed locally to model potential recombination. Previously, we demonstrated that a prototype of this approach could identify the nucleotide-level variants at classical HLA alleles with high accuracy. However, we did not address the problem of inferring the alleles present at the gene level [20].
We set out to extend the PRG approach to provide accurate HLA typing at G group resolution (see below) using high coverage whole-genome sequencing data, such as is being generated by large-scale genomics projects. This study presents novel developments in 3 main areas:
First, we build a gene-only PRG that combines genomic haplotypes (spanning the complete MHC), gene haplotypes and exon sequences for 46 (mostly HLA) genes (S1 Table). In our previous work we utilized a whole-MHC PRG and didn’t attempt to integrate exon sequences. A gene-only PRG is smaller and therefore computationally advantageous, and integration of the exon sequences gives a more comprehensive model of genetic variation at the HLA loci.
Second, we present an algorithm to map short (e.g. 100 or 250bp) paired-end next-generation sequencing reads directly to the PRG. We had previously [20] described an approach for long non-paired reads. The short-read algorithm we present here follows a classical seed-and-extend paradigm; that is, it identifies areas of exact identity between the graph and the read to be mapped, and tries to extend these using dynamic programming, allowing for mismatches, insertions and deletions. Each alignment follows a possible walk through the PRG. Importantly, because the PRG encodes information on sequence variation and because the mapping algorithm utilizes this information, it enables accurate alignment in the presence of homologous variants and a more precise quantification of mapping ambiguity. The short read mapping algorithm is relatively slow and benefits from our decision to limit ourselves to a gene-only PRG. See Fig 1 for a schematic depiction of graph topology and read mapping.
Third, conditional on reads mapped to the PRG and a database of possible underlying haplotypes (i.e., the HLA allele sequences), we infer the most likely pair of underlying haplotypes and a quality score using a simple likelihood framework.
We implement our approach in an open source package called HLA*PRG and show in two validation experiments that the level of achieved accuracy is comparable to SBT. Allelic variants at HLA genes can be typed at different degrees of resolution; low resolution (“1-field”, formerly “2-digit”) types specify serological activity; intermediate resolution (“2-field”; formerly “4-digit”) HLA types specify the complete primary sequence of the HLA proteins and high-resolution (“3-field”; formerly “6-digit”) types determine the full exonic sequence including synonymous variants. Higher levels of resolution include non-coding variation. SBT is typically carried out at “G group” resolution, in which only the sequences of the exons encoding the peptide binding groove are considered: exons 2 and 3 for HLA class I genes and exon 2 for HLA class II genes. In most applications of typing, a set of 6–8 loci are typed (Class I: HLA-A, -B, -C, Class II: HLA-DQA1, -DQB1, -DRB1, -DPA1 and -DPB1), though there exist over 30 HLA genes and pseudogenes.
Like SBT, HLA*PRG reports HLA types at G group resolution. Lower-resolution types are only used when benchmarking against other HLA type inference algorithms that fall back to these in cases of ambiguity.
To assess the accuracy of HLA*PRG, we use two data sets with available high coverage sequencing data and independent SBT-based HLA type information for 6 classical class I and class II loci (Table 1).
First, we analyse NA12878, NA12891 and NA12892 from the Illumina Platinum Genomes Project, sequenced to 50 - 55x with a PCR-free 2 x 100bp protocol. We correctly infer all 36 HLA alleles.
Second, we analyse 11 samples from the 1000 Genomes Project, sequenced to 28 – 68x with a PCR-free 2 x 250bp protocol. In terms of diversity, the 11 samples represent 7 ethnicities (S2 Table); a wide range of HLA types, e.g. 9 different 1-field groups for HLA-DRB1 (S2 Table); and heterozygous as well as homozygous loci (S3 Table). Initial analysis identifies three discrepancies (S1 Text), though on re-typing these individuals two of three are the result of initial errors in the validation data. The remaining inconsistency, (HLA-DRB1*16:02:01 incorrectly typed as HLA-DRB1*16:23) is likely caused by HLA-DRB5 sequences incorrectly aligned to HLA-DRB1 (IMGT/HLA, the HLA sequence database, currently doesn’t contain genomic sequences for HLA-DRB5 and the representation of this gene in the PRG therefore remains incomplete).
We compare the performance of HLA*PRG with PHLAT [14] and HLAreporter [13], two state-of-the-art algorithms that support HLA class I and class II (Table 1). For the Platinum samples, we find that PHLAT also correctly infers all 36 alleles, whereas HLAreporter only reports 16 alleles (of which 14 are correct). For the 1000 Genomes Samples, we find that HLA*PRG outperforms both programs by a wide margin. Mean accuracy at 2-field resolution across all loci is 75% for PHLAT and 80% for HLAreporter, and HLAreporter achieves a call rate of only 38%. To confirm that the observed differences in performance are not merely driven by different approaches to encoding ambiguous alleles, we repeat the 1000 Genomes validation experiment at 1-field resolution, the lowest and most ambiguous level of HLA typing; at 100% accuracy, HLA*PRG remains ahead of PHLAT and HLAreporter, which achieve accuracies of 89% and 90%, respectively (S4 Table).
To evaluate sensitivity of HLA*PRG to whole-genome sequencing depth, we subsampled the NA12878 data from the Platinum and 1000 Genomes projects to average coverages of 40x, 30x and 20x in triplicates. We find that performance is stable (all alleles correctly predicted) down to 20x for the Platinum data and down to 30x for the 1000 Genomes data (S5 Table).
To assess to what extent HLA*PRG depends on the availability of whole-genome data, we carried out two additional experiments. First, we apply HLA*PRG to whole-exome sequencing data of a cohort of HapMap samples. Results are varied and accuracies consistently lower across all loci (ranging from 79% for HLA-C to 98% for HLA-DQB1, S6 Table). Second, we apply HLA*PRG to a cohort of 14 Ugandan samples that underwent targeted amplification and MiSeq-based high coverage sequencing of the HLA exons (this cohort also contains a novel HLA-B and a novel HLA-DQB1 allele, see below). Average accuracy is 95% at G group resolution (PHLAT: 74%; HLAreporter: 73%) and 96% at 2-field resolution (PHLAT: 97%; HLAreporter: 80%) (S7 Table). Of the 6 erroneous alleles at 2-field resolution, 2 are novel alleles; an additional 2 errors are associated with mis-inferred DRB1*14:141 alleles, which are exon 2-identical to DRB3 sequences (which we expect, due to linkage with DRB1*11/13 alleles, also to be present in the samples).
For each inferred allele, HLA*PRG reports three quality statistics: a parametric likelihood-based quality score (S1 Text) and the proportion of k-Mers associated with the allele present in the sample sequencing data, both ranging from 0 to 1; and the number of columns in the read-to-graph alignment that contain alleles with an allele frequency ≥0.2 that are not accounted for by the diploid HLA call for the sample (“unaccounted alleles”). For the first two metrics, samples with lower values are enriched for errors; there is, however, no clear cut-off between correct and incorrect alleles (S3 Fig). The two novel alleles cluster with the bulk of correct calls. For the “unaccounted alleles” statistic, we observe that the number of columns with high-frequency unaccounted alleles varies systematically between loci (S4 Fig), and that there is no systematic difference between correct and incorrect calls. Although the two novel alleles score comparatively highly on this metric, it doesn’t enable the clear distinction between novel and non-novel alleles (S4 Fig). Nevertheless, the combination of these quality metrics can help identify alleles with higher uncertainty.
To assess whether HLA*PRG could be applied to additional HLA loci beyond the set of the 6 classical genes validated here, we use it to genotype a set of 12 additional HLA genes and pseudogenes in the two trios that are part of our whole-genome cohorts (S8 Table). Across the 2 x 72 alleles inferred, we find one trio inconsistency in the CEU trio (pseudogene HLA-K, driven by an allele called with low confidence)); and two inconsistencies in the YRI trio (the HLA-DRB1 inconsistency described above and an additional inconsistency at HLA-K).
We have shown that HLA*PRG enables HLA typing from standard whole-genome next-generation sequencing data at accuracies comparable to those of the current gold-standard SBT technology (two errors in the original reference data compared to one from HLA*PRG at 2-field / G group resolution)–provided that high-quality whole-genome sequencing data are used as input (PCR-free protocol, read length of at least 100bp, coverage of at least 30x).
Importantly, our results apply to both the established 2 x 100bp and the more recent 2 x 250bp Illumina HiSeq protocols; they are therefore directly applicable to many of the large-scale sequencing projects currently ongoing. HLA*PRG will enable researchers to augment these cohorts with reliable HLA type information and can contribute to characterizing HLA signals in important medical phenotypes.
The current implementation of HLA*PRG is limited in three respects.
First, although the algorithm was designed explicitly for application to high-quality whole-genome sequencing data, it would be advantageous if comparable performance was achieved from other data sources. Our evaluation shows that this is not the case. The exome sequencing cohort exhibits the lowest accuracy of all cohorts examined; it is also the cohort with the lowest effective fragment length (2 x read length + insert size, S5 Fig). High effective fragment lengths help overcome local sequence homologies like those observed in the HLA region, and it is likely that this factor plays a role in explaining the poor performance of the exome cohort. Alignment issues also likely contribute to the slight reduction in accuracy observed when applying HLA*PRG to the MiSeq sequencing data (exon targeting by PCR) validation cohort. Although the effective fragment length of this cohort is higher, the vast majority of reads start and end within a few bases of the exon boundaries. In comparison to whole-genome data, where the majority of exon-spanning reads run into introns and the variation contained therein, this exacerbates the effects of exon sequence homologies between different genes (consistent with the observation that the problem of HLA*PRG mis-inferring a small number of alleles due to DRB1-DRB3 distinction issues arises only in the MiSeq, but not in the whole-genome, cohort). Of note, base quality doesn’t seem to strongly influence accuracy; when measured by the number of read bases agreeing with the graph location they are aligned to, average base quality is lowest for the 1000 Genomes cohort (87%) and highest for the exome cohort (99%). In summary, high (≥30x) uniform coverage across the whole length of the HLA genes and high fragment lengths seem to deliver best results; caution should be exercised when these conditions are not met (e.g. for targeted amplification, whole-genome amplification, targeted capture).
Second, HLA*PRG is optimised for accuracy rather than computational efficiency. Analysing the NA12878 data takes between 33 and 215 CPU hours (Platinum / 1000G data; AMD Opteron 6174 2.2GHz; 11–17 hours clock time). Analysing the same data with PHLAT and HLAreporter takes 466/626 and 53/50 CPU hours, respectively (Fig 2). Depending on the availability of high-performance compute infrastructure and the number of samples to analyse, computational demands might represent a significant barrier to adoption. We provide a detailed runtime (including CPU time) and memory analysis in S1 Text. Achieving improvements in computational efficiency is ongoing work, but it is worth noting that HLA*PRG can be run immediately after the raw sequence data has been mapped, in parallel with standard variant-calling. Future versions might make use of linear sequence alignments to seed graph alignment (similar to ALT-aware alignment in BWA-MEM[21]) and also incorporate population haplotype frequencies [22, 23].
Third, HLA*PRG doesn’t enable the discovery of new alleles and it is limited to G group resolution. It would be advantageous if it was possible to identify samples that harbour novel alleles from the quality metrics of the inferred alleles. This, however, was not the case for the two novel alleles in the Ugandan cohort and the 3 parametric and non-parametric statistics we analysed. It seems (S4 Fig) that there is a residual number of reads aligned to the wrong gene, and distinguishing between the signal generated by these and that of true novel alleles is non-trivial. Further research will be necessary to address this.
The good performance of HLA*PRG is consistent with the hypothesis that reference graph approaches are well-suited to enable accurate genome inference in highly complex, highly diverse regions of the human genome. At the example of the MHC as a whole, we had already shown that this was the case using small-scale (SNPs, k-Mers) and large-scale (long reads) metrics of genome inference accuracy [20]. By focusing on G group resolution HLA typing, this study complements the existing evidence with another important metric: exon-scale haplotype accuracy. Faster algorithms for sequence-to-graph alignment than the one used here are currently under development (https://github.com/vgteam/vg) and these will likely simplify the development of future variation graph-based approaches.
There are other regions in the human genome that could benefit from a tailored PRG-based inference approach. One example is the LRC/KIR region on chromosome 19, which is similar to the MHC in some aspects (hyperpolymorphism, availability of haplotype and allele databases[24]) and different in others (higher degrees of structural variation and paralogy [25–27]). One important implication of the results presented here is that a variation-aware read mapping algorithm that processes reads independently (in the sense that no attempt at local haplotype reconstruction is made during the read mapping process) is sufficiently accurate for HLA genotyping. Whether this also applies to the LRC/KIR region is an open question.
In this Section we present a high-level summary of PRG construction, read mapping, HLA type inference and validation. A full technical description of the algorithms is given in S1 Text.
We construct a gene-only PRG for 46 genes from 720 genomic sequences (from IMGT [28] / GRCh37), 10050 exonic sequences (from IMGT) and 8 MHC haplotypes (from GRCh37). For each gene independently, we combine (see next Section) exonic sequences (where available), genomic sequences and 0.5kb of surrounding non-genic “padding” sequence from the 8 MHC haplotypes (where appropriate; to enable the alignment of reads that span a gene boundary). We then construct a joint PRG [20], in which we separate genes with 2000 N characters (see Fig 1A for a schematic depiction of the topology of the PRG created).
We employ a multiple sequence alignment (MSA) merging algorithm (Fig 1C) to combine genomic, exonic and genomic “padding” sequences for PRG construction. Typically there are more exonic sequences than genomic sequences, and more genomic sequences than genomic “padding” sequences (Fig 1B).
We describe the base case of merging the MSA for one exon into the MSA of surrounding genomic sequences. The other cases follow immediately and are described in S1 Text.
The aim of the MSA merging algorithm is to find the “switch points” between the exon MSA and the genomic MSA; i.e. the coordinates at which the PRG switches from one MSA to the other and back (blue lines in Fig 1C). To compute the switch points we rely on shared alleles (i.e. alleles that are present in both MSAs)–for each shared allele, we should be able to identify the exon sequence as a substring of the genomic sequence alignment, defining the coordinates of the exon MSA in genomic MSA terms. If there is more than one shared allele, we compute consensus switch points.
Switch coordinate computation for the exon MSA:
The steps as described above define the area (from coordinates GL to GR) of the exon MSA to be utilized for PRG construction. For each individual genomic shared allele sequence, this leaves sequence to the left and to the right of the extracted exon sequence. We combine all such ‘left’ (genomic) sequences from all shared alleles and compute [29] a new MSA; we also combine all such ‘right’ (genomic) sequences from all shared alleles and compute a new MSA. Finally, the two MSAs so-created are used as the segments to the left and to the right of the extracted exon MSA block for PRG construction (Fig 1B).
A full description of the PRG construction algorithm is given in S1 Text and the output data are part of the HLA*PRG data package (available on the HLA*PRG website).
Let R be the set of paired-end read alignments that overlap with the peptide-binding site of a given HLA locus. We calculate the likelihood of the observed data R conditional on pairs of possible underlying HLA types at G group resolution. We note that each HLA type (i.e., each possible underlying allele) has a defined genotype (potentially including “gap” characters) at each position of the peptide binding site as represented in the PRG.
For an arbitrary pair (a1,a2) of underlying alleles, we define the likelihood functions
L(R|(a1,a2))≔∏r∈RL(r|(a1,a2))and
L(r|(a1,a2))≔12×L(r|a1)+12×L(r|a2).
L(r|a) is the likelihood of observed aligned read pair r, conditional on an assumed underlying HLA allele a.
By definition r overlaps with the peptide-binding site. At each overlapping position, we have a pair (gr,ga), where gr is the nucleotide (or gap) of the aligned read r (and its associated base quality, if applicable) at this position, and ga is the genotype of underlying HLA allele a at this position. We define the set Or as the set of pairs for all overlapping positions of r.
Finally, we define
L(r|a)≔∏(gr,ga)∈Orscore(gr,ga)
, with score being a base-quality-aware alignment scoring function for matches, mismatches, deletions and insertions.
We compute the likelihood function over all pairs of possible underlying alleles and normalize to obtain a probability distribution over possible underlying allele pairs. We call two “best guess” alleles and their associated qualities as described in [10]. Briefly,
We give a full technical description of the likelihood-based inference procedure in S1 Text.
Except for the Ugandan MiSeq cohort, HLA types for all validation samples are either publicly available [32, 33] or available from a previous study [10].
14 samples from Ugandan individuals were available through the Entebbe Mother and Baby Study courtesy of Alison Elliott[34]. DNA was extracted from EDTA-stored cell pellets using the QIAamp DNA Blood Mini Kit (QIAGEN, Germany) before undergoing SBT using two methods. The first method, a Sanger-based approach for validation, was undertaken at Addenbrooke’s Tissue Typing Laboratory, (Cambridge, UK) using the Fisher Scientific proprietary uTYPE (version 7) software (Fisher Scientific, MA, USA). Exon-targeted MiSeq sequencing was undertaken at Histogenetics (NY, USA) using proprietary protocols. After validating the MiSeq-based calls with the Sanger data, the MiSeq data were used for validation. All data will be made available to interested researchers upon request through the African Partnership for Chronic Disease Research Data Access Committee.
S1 Text contains a table of utilized samples and their HLA types.
HLA*PRG is implemented in C++/Perl and available under GPLv3 as part of the MHC*PRG repository https://github.com/AlexanderDilthey/MHC-PRG. A readme file (https://github.com/AlexanderDilthey/MHC-PRG/blob/master/HLA-PRG.md) describes how to install and run the software. A compiler with support for C++11 and openMP is required (e.g., GCC 4.7.2). We provide a separate HLA*PRG data package (download size ~41G), independent from the larger MHC*PRG data package.
|
10.1371/journal.pntd.0003500 | Species-Specific Antimonial Sensitivity in Leishmania Is Driven by Post-Transcriptional Regulation of AQP1 | Leishmania is a digenetic protozoan parasite causing leishmaniasis in humans. The different clinical forms of leishmaniasis are caused by more than twenty species of Leishmania that are transmitted by nearly thirty species of phlebotomine sand flies. Pentavalent antimonials (such as Pentostam or Glucantime) are the first line drugs for treating leishmaniasis. Recent studies suggest that pentavalent antimony (Sb(V)) acts as a pro-drug, which is converted to the more active trivalent form (Sb(III)). However, sensitivity to trivalent antimony varies among different Leishmania species. In general, Leishmania species causing cutaneous leishmaniasis (CL) are more sensitive to Sb(III) than the species responsible for visceral leishmaniasis (VL). Leishmania aquaglyceroporin (AQP1) facilitates the adventitious passage of antimonite down a concentration gradient. In this study, we show that Leishmania species causing CL accumulate more antimonite, and therefore exhibit higher sensitivity to antimonials, than the species responsible for VL. This species-specific differential sensitivity to antimonite is directly proportional to the expression levels of AQP1 mRNA. We show that the stability of AQP1 mRNA in different Leishmania species is regulated by their respective 3’-untranslated regions. The differential regulation of AQP1 mRNA explains the distinct antimonial sensitivity of each species.
| The degree of response to antimonial drugs varies widely between species and even among strains of the same species of the protozoan parasite Leishmania. However, the molecular mechanism(s) is unknown. In this study, we show that Leishmania aquaglyceroporin AQP1 drives this species-specific antimonial resistance. Aquaglyceroporins are channel proteins that facilitate the passage of small uncharged molecules, such as glycerol and water, across the biological membranes. AQP1 helps the parasite cope with the osmotic challenges it faces during its life cycle. Additionally, AQP1 is an adventitious facilitator of antimonite, the active form of pentavalent antimonial drugs. We show that AQP1 expression level is species-specific, and less AQP1 in visceral species compared to the cutaneous species results in increased resistance to antimonials. We also demonstrate that the 3’-untranslated regions (3’-UTR) of the AQP1 mRNA is a major determining factor of species-specific regulation of AQP1. Along with water homeostasis, aquaglyceroporins are also involved in directed cell migration. The variable levels of AQP1 in different Leishmania species may enable them to find their appropriate niches in vertebrate hosts and cope with the species-specific osmotic challenges during their life cycles.
| Leishmaniasis is a protozoan parasitic infection in humans and other mammals that is transmitted by the bites of sandflies. The infection is caused by more than 20 different Leishmania species. The clinical manifestations range from self-healing cutaneous leishmaniasis (CL) to a potentially life threatening mucocutaneous leishmaniasis (MCL) [1] to the lethal, if untreated, visceral lesihmaniasis (VL) [2]. The disease is endemic in parts of 88 countries in five continents—the majority of the affected countries are in the tropics and subtropics. Approximately two million new cases are estimated to occur annually, of which 1.5 million are categorized as CL and 500,000 as VL. The parasite exists in two distinct morphological forms. The promastigotes form resides in the insect gut and appears to have slipper-like bodies with long flagella. The vertebrate forms of the parasite, amastigotes, have spherical, oval-shaped, aflagelleted bodies that reside in the macrophages of mammalian hosts. The first line of treatment against all forms of leishmaniasis is the pentavalent antimony-containing drugs sodium stibogluconate (Pentostam) and meglumine antimonite (Glucantime). However, drug resistance is a major impediment to the treatment of leishmaniasis. For example, approximately 60% of the patients in India do not respond to antimonial treatment due to acquired resistance [3].
Mechanisms of antimonial resistance in Leishmania have been explored extensively for several decades and are considered to be multifactorial [4] [5]. We have shown that laboratory-raised arsenic resistant L. tarentolae, which are cross resistant to antimonials, overproduced trypanothione (T[SH]2) [6], the major reduced thiol in Kinetoplastida [7], and conferred resistance by providing excess Sb-[TS]2 conjugates for the efflux pump in the plasma membrane [8]. The Sb-[TS]2 conjugates were shown to be sequestered into small intracellular vesicles near the flagellar pocket [9]. These mechanisms also seemed to play an active role in the pathogenic Leishmania [4] and also in field isolates [10] [11]. Amastigote-specific pentavalent antimonial reducing capability has also been implicated in VL [12]. Variability in the frequency of incidence of clinical antimonial resistance among Leishmania species has been reported [13]. However, it is not known whether an intrinsic variation in antimonial sensitivity exists among different Leishmania species. We reported that L. major was 50–70 times more sensitive to antimonite when compared to L. infantum [14]. Sarkar et al (2012) reported that Leishmania strains causing self-healing CL exhibited greater susceptibility towards oxidative stress as a result of low thiol content [15]. However, a comprehensive species wide study of all antimonial resistance markers reported so far is absent, and hence, the mechanism(s) of species-specific antimonial sensitivity is unknown.
We discovered the first aquaglyceroporin from Leishmania (AQP1) and showed its direct relationship to antimonite [Sb(III)], the active component of Pentostam and Glucantime, uptake [14] and resistance [16]. Overexpression of AQP1 in Leishmania cells led to hypersensitivity to antimonite, and disruption of one of the two LmAQP1 alleles in L. major conferred a 10-fold increase in resistance to Sb(III) [14]. Later, these findings were corroborated in field isolates from India [17,18] and Nepal [19]. Besides the metalloids arsenite [As(III)] and Sb(III), the water conduction capacity of AQP1 is 65% of that of the classical water channel, human AQP1. Unlike the Trypanosoma and Plasmodium AQPs, AQP1 is a mercurial independent water channel. It also conducts glycerol, glyceraldehyde, dihydroxyacetone and sugar alcohols. We have also identified AQP1’s key role in osmoregulation and osmotaxis, which play crucial functions during parasite transmission [20]. Also, AQP1 is the first aquaglyceroporin to be exclusively localized in the flagellum of any organism. In intracellular amastigotes, it is localized in the flagellar pocket, rudimentary flagellum, and contractile vacuoles [20]. We have shown the involvement of flexible loop C of AQP1 in determining the substrate specificity of the channel [21,22]. Additionally, we showed that AQP1 was positively regulated at the post-translational level by a mitogen activated protein kinase 2 [17]. Therefore, AQP1 plays a major role in Leishmania cellular physiology and drug response.
During the course of our research with AQP1, we noticed that the muccocutaneous and cutaneous species were much more sensitive to Sb(III) when compared to the visceral species. Since AQP1 is the sole facilitator of Sb(III) in Leishmania, we asked whether AQP1 is driving this species-specific antimony sensitivity. In the absence of RNA polymerase II promoters, Leishmania genes are constitutively transcribed from large gene clusters as polycistronic pre-mRNAs. Steady-state levels of mature monocistronic mRNAs are regulated post-transcriptionally primarily by trans-splicing and polyadenylation [23,24]. Several examples in Leishmania species support the notion that post-transcriptional regulation of developmentally expressed transcripts involves sequences present mainly in the 3’-UTR [25,26,27,28], and more rarely in intergenic regions between tandemly repeated genes [29,30]. The 3’-UTR also regulates logarithmic-stationary phase gene regulation [31].
In this study, we mapped and cloned the 3’-UTRs of AQP1 mRNA from six different Leishmania species representative of different clinical pathologies and endemic regions. Each approximately 1.8-kb 3’-UTR is highly U-rich (30%) with only 49% GC (in a highly GC-rich [∼ 60%] genome), and contains several well-known instability elements described in higher eukaryotes. Although the AQP1 protein sequences among these six species are more than 80% homologous, 3’-UTRs of AQP1 mRNAs differ significantly. We show that the species-specific antimonial sensitivity in Leishmania is uniquely driven by AQP1, and that it is mediated by post-transcriptional regulation through the respective distinct 3’-UTR of each species-specific AQP1 mRNA.
Work here was carried out on six Leishmania species: three cutaneous (L. major, L. tropica and L. panamensis), one mucocutaneous (L. braziliensis) and two visceral (L. infantum and L. donovani). Our rationale for choosing these species was to represent every endemic continent (Asia [L. donovani, L. infantum, L. major and L. tropica], Africa [same as Asia], Europe [L. infantum, L. major and L. tropica] and Americas [L. infantum, L. braziliensis and L. panamensis]); clinical manifestation (VL- L. donovani, L. infantum; CL- rest of the species; MCL- L. braziliensis); and mode of transmission (anthroponotic- L. donovani and L. tropica; zoonotic- rest of the species). Intrinsic difference in Sb(III) sensitivity in the representative six selected species was determined by exposing the promastigotes at increasing concentrations of potassium antimonyl tartrate. L. infantum was the least sensitive species. The EC50 data showed that it was 1.4 times more resistant when compared to L. donovani and 46, 15, 20 and 7 times more resistant than the cutaneous species, namely, L. major, L. tropica, L. braziliensis and L. panamensis respectively (Table 1). However, the cutaneous species also differed in sensitivity to Sb(III) among themselves. L. tropica, L. braziliensis and L. panamensis were 3, 2.3 and 6 times more resistant to Sb(III), respectively, when compared to L. major (Table 1).
To study whether this species-specific antimony sensitivity was also operational in amastigotes, we determined the EC50 of potassium hexahydroxy antimonate [Sb(V)] using intra-macrophagial amastigote model. We did not use pentavalent organo-antimonials as contaminating Sb(III) levels can represent more than 30% of total Sb [32]. The EC50 data showed that L. infantum was the least sensitive species among the six we tested. It was 1.4, 25, 9.6, 20, and 8.6 times more resistant when compared to L. donovani, L. major, L. tropica, L. braziliensis, and L. panamensis, respectively (Table 1). Again there was a slight variation in sensitivity among the cutaneous species. L. tropica, L. braziliensis and L. panamensis were 2.6, 1.3 and 3 times more resistant to SbV when compared to L. major (Table 1). In general, the CL species were much more sensitive to antimony compared to the VL species.
To understand this differential intrinsic antimonial sensitivity among the species in greater detail, we first examined the time-dependent intracellular accumulation of Sb(III) in the promastigotes. L. braziliensis showed the fastest and highest accumulation of Sb(III) at any given time followed by L. major, L. panamensis and L. tropica (Fig. 1A). The lowest rate and total accumulation of Sb(III) were observed in L. donovani and L. infantum (Fig. 1A). Generally, the antimony sensitive CL species accumulated more total antimony when compared to the VL species. Did this signify lower uptake or increased efflux in the VL species and higher uptake or slower efflux in the CL species? To answer this question, we prepared everted membrane vesicles from promastigotes of each species, and Sb(TS)2 conjugate accumulation was measured in the presence of 10 mM ATP as an energy source. The rates of transport of Sb-[TS]2 conjugate in the everted membrane vesicles of the six species were not significantly different from each other (Fig. 1B). These results indicate that differential sensitivity among different species is not due to a change in the efflux rate of antimony.
Mechanisms of antimonial resistance in Leishmania have been proposed to be multifactorial. Four major components shown to be modulated in many laboratory-raised and clinical antimonial resistant Leishmania spp. are: (i) higher MRPA level for greater intracellular sequestration of Sb-[TS]2 conjugates; (ii) overproduction of total thiols, specifically T[SH]2; (iii) faster efflux of Sb-[TS]2 conjugates through an unknown efflux system in the plasma membrane; and (iv) downregulation of AQP1. Thus, we examined the MRPA mRNA levels in all six species first. The CL species showed much more MRPA mRNA when compared to the VL species. L. panamensis had the highest level of MRPA mRNA, which was about 7.4 fold more compared to L. donovani (Fig. 2A). L. donovani and L. infantum had similar levels of MRPA mRNA. The levels of MRPA mRNA in L. major, L. tropica, L. braziliensis and L. panamensis were approximately 2.3, 5.4, 4.5 and 7.4 fold respectively, compared to L. donovani (Fig. 2A). Therefore, having more MRPA mRNA does not justify the CL species being more sensitive to the antimonials. More mRNA does not necessarily result in more protein. However, the difference at the MRPA protein level between the CL and VL species does not explain the difference in antimony sensitivity between the species.
Next, we measured the total non-protein thiol levels among all the species. There was no correlation between total thiol levels and species-specific antimonial sensitivity (Fig. 2B). For example, three CL species, L. major, L. braziliensis, and L. panamensis, and one VL species, L. donovani, showed similar levels of non-protein total thiols (Fig. 2B). Another CL species, L. tropica, showed almost one-half of the thiol levels compared to other CL species, whereas L. infantum had approximately double that of L. donovani. Fig. 1B shows that the activity of the efflux pump was similar in all six species. Taken together, these data indicated that the levels of total Sb(III) accumulation can only be different if the rate of uptake differed among the species.
Next, we measured the mRNA levels of AQP1—the only Sb(III) facilitator in Leishmania. Relative (normalized against L. donovani) AQP1 mRNA levels were calculated by the 2-ΔΔCt method. The AQP1 mRNA levels corroborated with our EC50 data in promastigotes and intracellular amastigotes. L. infantum showed the lowest level of AQP1 mRNA among the species tested. L. major, L. tropica, L. braziliensis and L. panamensis produced 37, 13, 56 and 52 fold more AQP1 mRNA, respectively, when compared to L. donovani (Fig. 2C). Therefore, the CL species showed much more AQP1 mRNA accumulation than the VL species, which also corroborated our Sb(III) accumulation data (Fig. 1A).
We have shown that the physiological function of AQP1 in Leishmania is osmoregulation. The strategic flagellar localization of AQP1 helps the parasite sense the changing osmotic environments between the vector and the host [20]. We could not detect native AQP1 expression at the protein levels by our anti-peptide polyclonal anti-AQP1 antibody in wild type Leishmania promastigotes [20]. Thus, it was reasonable to determine the functionality of AQP1 in all six species by determining their osmoregulatory capacities under hypo-osmotic shock (50% reduction in extracellular osmolarity), as we previously showed that osmoregulatory capacity was directly proportional to the AQP1 protein levels [20] [17] in the membrane. We observed that the CL species osmoregulate more efficiently when compared to the VL species, suggesting that CL species express more functional AQP1 than the VL species (Fig. 3). This corroborated with their EC50 (Table 1), Sb(III) accumulation (Fig. 1A) and AQP1 mRNA levels (Fig. 2C), i.e., promastigotes with less AQP1 mRNA (VL species) accumulated less amount of Sb(III), which resulted in higher resistance to antimonials when compared to promastigotes with more AQP1 mRNA (CL species). L. donovani and L. infantum promastigotes swelled more rapidly (drops in absorbance) and recovered their volumes (rises in absorbance) slowly when compared to all CL species (Fig. 3). It is interesting to note that the CL species L. braziliensis and L. panamensis showed the highest levels of AQP1 mRNA (Fig. 2C), and they were the most efficient osmoregulators (Fig. 3). Among the VL species, L. infantum showed the highest antimonial resistance and was the poorest osmoregulator. It swelled more than L. donovani and recovered even slower from the hypo-osmotic shock (Fig. 3). There is a very strong correlation between the antimony sensitivity among the species, the osmoregulatory capacity and AQP1 mRNA levels of the promastigotes.
Since Leishmania does not have any transcriptional control, we investigated the stability of AQP1 mRNA in all six species to determine if any post-transcriptional regulation is active in lowering the mRNA levels in the VL species. To determine the turnover rate of AQP1 mRNA, the mid-log phase promastigotes of all species were treated with sinefungin (to stop pre-mRNA processing) [33] followed by actinomycin D (to inhibit transcription), [34] and cells were harvested up to 130 minutes. The decay of AQP1 mRNA was determined by qPCR and normalized against the 0 minute (time of addition of actinomycin D). The half-lives of AQP1 mRNA from VL species, i.e., L donovani and L. infantum, were determined to be 40 min and 26 min, respectively; on the other hand, the half-lives of the AQP1 mRNA from CL species, such as L. tropica and L. panamensis, were 57 and 117 min, respectively. The half-lives of the AQP1 mRNA in L. braziliensis and L. major were estimated to be > 130 min (Fig. 4). Therefore, AQP1 mRNA from the CL species was much more stable than from the VL species, which follows a similar trend that was observed in their respective steady-state levels of AQP1 mRNA: L. braziliensis ≥ L. panamensis > L.major > L. tropica > L. donovani > L. infantum (Fig. 2C).
Since mRNA stability in Leishmania is mostly controlled by the 3’-UTR [23], we cloned the individual 3’-UTRs of AQP1 mRNA from all six species. The length of the 3’-UTR of the AQP1 mRNA was approximately 1.8 kb in all species examined as mapped by 3’-RACE PCR. The protein sequence alignment of AQP1 from all six species showed that they were very close to each other, with an overall similarity of about 88% (S1 Fig.). Additionally, the overall identity between the open reading frames (ORF) of AQP1 in all six species was about 66% (S2 Fig.). However, 3’-UTRs from all six species were more divergent and showed an overall identity of about 24% (S3 Fig.). A greater similarity was seen between the pairs. L. infantum and L. donovani 3’-UTRs were 99% identical (S4 Fig.). The CL species 3’-UTRs can be divided into two groups. L. tropica and L. major were 78% identical (S5 Fig.), while L. panamensis and L. braziliensis 3’-UTRs were 94% identical (S6 Fig.). Thus, based on their 3’-UTR sequences, the six species we examined can be divided into three distinct groups: VL (L. donovani and L. infantum), CL1 (L. major and L. tropica) and CL2 (L. braziliensis and L. panamensis) groups.
Based on the data presented above, we concluded that the stability of AQP1 mRNA plays a significant role in dictating the species-specific antimonial sensitivity in Leishmania. Given that 3’-UTR sequences between VL and CL species are very divergent, we hypothesized that these sequences are responsible for the species-specific differential AQP1 mRNA stability. To determine the underlying mechanism, a series of chimeric constructs were made with the full length 3’-UTR (∼ 1.8 kb) of the AQP1 mRNA from each of the six species by cloning them at the 3’ end of the luciferase (LUC) reporter gene (Fig. 5A). To direct accurate 5’ and 3’ processing of the LUC chimeric transcripts, these cassettes were flanked by an upstream α-tubulin intergenic (IR) region and by a downstream IR region (∼ 200 bp) of each species-specific 3’-UTR of AQP1 mRNA. In trypanosomatids, polyadenylation is often directed by trans-splicing signals that are located 100–400 nucleotides downstream of the polyadenylation site [28,35,36]. These chimeras were expressed from an episomal plasmid pSPYNEOαLUC with neomycin phosphotransferase (NEO) as a marker. The vector alone, where LUC expression was only regulated by the intergenic regions of α-tubulin, is referred to in the present work as pLUC. Each of the species-specific LUC-AQP1-3’-UTR constructs were named pLUC-Ld (L. donovani), pLUC-Li (L. infantum), pLUC-Lm (L. major), pLUC-Lt (L. tropica), pLUC-Lb (L. braziliensis), and pLUC-Lp (L. panamensis) (Fig. 5). Six individual chimeric constructs and the vector alone were transfected into the six species generating 42 transfectants. We also evaluated the relative copy number of LUC–containing plasmids by qPCR using pteridine reductase 1 (PTR1) as the housekeeping control, the levels of which are similar in all transfectants (S1A Table). The proper processing of 3’ end of UTRs from all transfectants with chimeric plasmids was determined by 3’-RACE PCR and sequencing. They had identical processing when compared to the 3’-UTR ends of the native AQP1 mRNA. Therefore, it was reasonable to deduce that the LUC expression and activity in the transfectants would largely depend upon the steady-state levels of LUC mRNA dictated by their stability in each species. We thus tested the role of each of the six 3’-UTRs in regulating LUC mRNA steady-state levels, stability, LUC protein levels and LUC activity in a species-specific manner.
L. donovani transfected with pLUC-Li produced the lowest levels of LUC mRNA when compared to pLUC control, whereas pLUC-Ld produced slightly more (only 0.37 fold). However, accumulation of the LUC mRNA under the control of 3’-UTRs from the CL species was 3 to 8 fold higher than that of the pLUC control (Fig. 5B). pLUC-Lt, pLUC-Lm, pLUC-Lb, and pLUC-Lp transfectants produced 3.7, 5.7, 8.4 and 6.6 fold more LUC mRNA, respectively, compared to pLUC alone (Fig. 5B). These data corroborated with our LUC mRNA stability, LUC protein expression and activity, suggesting that regulation occurs at the level of AQP1 mRNA stability. Indeed, LUC mRNA was most unstable in pLUC-Li transfectants, with a half-life of 25 min (Table 2). pLUC-Ld extended that for 39 min, and for pLUC-Lt, the half-life was 83 min. pLUC-Lm, pLUC-Lb, pLUC-Lp transfectants gave rise to the most stable LUC mRNA, extending their half-lives to over 130 min (Table 2). Western blot with an anti-luciferase antibody and densitometric analysis using α-tubulin as a loading control revealed that luciferase protein expression was lowest in pLUC-Li transfectants, at only about 21%, which resulted in about 18% luciferase activity when compared to pLUC control (Fig. 5C and D). pLUC-Ld, pLUC-Lm, pLUC-Lt, pLUC-Lb and pLUC-Lp transectants expressed 1.5, 4.5, 2.6, 5.5, and 4.7 times more luciferase, respectively, when compared to pLUC-Li. This is also comparable with LUC activity, which was increased in the following order in the transfectants: pLUC-Li < pLUC-Ld < pLUC-Lt < pLUC-Lb < pLUC-Lm < pLUC-Lp (Fig. 5C). Similar results were obtained when transfecting the chimeric LUC constructs with the 3’-UTRs in L. infantum. The 3’-UTRs from the CL species generated 3 to 5 fold more LUC mRNA when compared to the VL species (S7A Fig.). The CL 3’-UTRs also made LUC mRNA more stable (higher half-life) in L. infantum (Table 2). These results also corroborated with LUC expression and activity (S7B Fig.).
When transfected in L. major, pLUC-Ld and pLUC-Li constructs produced similar basal levels of LUC mRNA when compared to pLUC alone. However, the 3’-UTRs from the CL species generated 4–5 fold more LUC mRNA (Fig. 6A). These data corroborated with our LUC mRNA stability, LUC protein expression and activity. LUC mRNA produced from all constructs in L. major were quite stable, and the half-life was determined to be >130 min (Table 2). Densitometric analysis revealed that LUC expression (Fig. 6C) is similar in all CL UTR constructs. pLUC-Li and pLUC-Ld transfectants express LUC at 77% and 93% of pLUC control (Fig. 6B), respectively. The highest LUC activity was observed in cells expressing LUC from pLUC-Lm, followed by pLUC-Lp, pLUC-Lt and pLUC-Lb. Similar results were obtained when transfecting the chimeric LUC constructs with the 3’-UTRs in L. tropica. LUC mRNA produced from all constructs in L. tropica were quite stable, and the half-life was determined to be >130 min, except in pLUC-Li and pLUC-Ld, which were 87 min and 89 min, respectively (S8A Fig., Table 2). Four CL 3’-UTRs regulated to produce 85–101% LUC protein compared to pLUC control, which resulted in 81–100% LUC activities in pLUC-Lm, pLUC-Lt, pLUC-Lb and pLUC-Lp transfectants (S8B Fig.).
Promastigotes of L. braziliensis transfected with pLUC-Ld construct produced the lowest levels LUC mRNA when compared to pLUC alone, whereas pLUC-Li produced a little more (only 0.42 fold). However, the UTRs from the CL species generated 4.5–6.0 fold more LUC mRNA compared to pLUC cells (Fig. 7A). pLUC-Lb transfectants produced 4.5 fold more LUC mRNA, whereas pLUC-Lm, pLUC-Lt, and pLUC-Lp transfectants generated about 6 fold more (Fig. 7A) compared to pLUC. This data corroborated with our LUC mRNA stability, LUC protein expression and activity. LUC mRNAs produced from pLUC-Lm, pLUC-Lt and pLUC-Lp constructs in L. braziliensis were quite stable, and the half-lives were determined to be >130 min. The half-lives of LUC mRNAs generated from pLUC-Ld, pLUC-Li and pLUC-Lb were 84 min, 80 min and 98 min, respectively (Table 2). Densitometric analysis of LUC expression showed 72–75% of pLUC control in pLUC-Ld and pLUC-Li transfectants, which resulted in 60–65% of pLUC control LUC activity in those cells (Fig. 7B). The CL 3’-UTRs behaved in a similar manner, producing 101–122% luciferase protein compared to pLUC control, which resulted in 100–110% of pLUC control LUC activity in pLUC-Lm, pLUC-Lp and pLUC-Lt transfectants (Fig. 7B and C). However, LUC activity from pLUC-Lb construct was 86% compared to pLUC alone. Similar results were obtained in L. panamensis transfectants. The CL 3’-UTRs generated more stable LUC mRNAs with longer half-lives compared to the VL species (S9A Fig., Table 2). Higher LUC protein expression and activity were also observed in the CL species (S9B Fig.).
Lastly, we argued that if 3’-UTR was driving the species-specific antimony sensitivity, swapping the 3’-UTRs between Ld and Lm-AQP1 should lead to contrasting antimonial sensitivity of a visceral species. Thus we cloned LdAQP1 and LmAQP1 ORFs into pSP72-YHYG-αtubIR with their native 3’-UTRs and also chimeric constructs where the 3’-UTRs were swapped. The four constructs, namely, LdAQP1-Ld3’-UTR, LdAQP1-Lm3’-UTR, LmAQP1-Ld3’-UTR, and LmAQP1-Lm3’-UTR, along with the vector alone control, were transfected into the L. donovani strain LdBOB. We evaluated the relative copy number of AQP1-3’-UTR-containing plasmids by qPCR using hygromycin phosphotransferase as the target and PTR1 as the housekeeping control, the levels of which were similar in all transfectants (S1B Table). Sb(III) sensitivity of the transfectants were measured in promastigotes (Fig. 8A) and intracellular amastigotes (Fig. 8B). As expected, overexpression of AQP1 made the VL strain hypersensitive to Sb(III) when compared to the vector alone control albeit to a different degree depending on which type of 3’-UTR the ORF had at its 3’ end in both promastigotes and amastigotes. The VL species was 6–24 times more sensitive to Sb(III) in both stages of the parasite whenever Lm-3’-UTR was present at the 3’ end of AQP1 when compared to the Ld-3’-UTR constructs (Fig. 8A and B). A similar trend was observed during intracellular accumulation of Sb(III). L. donovani promastigotes accumulated significantly more Sb(III) overexpressing AQP1 with Lm-3’-UTR constructs (Fig. 8C). It was interesting to note that the osmoregulatory capacity of the VL species improved considerably with Lm-3’-UTR at the 3’ end of AQP1, whereas promastigotes overexpressing AQP1 with Ld-3’UTR were still poor osmoregulators, although better than the vector alone controls (Fig. 8D). These data convincingly show that 3’-UTR of AQP1 plays a major role in determining the species-specific antimonial sensitivity of Leishmania and that the effect is not stage-specific.
Although antimonial drugs are still the first line of treatment against all types of leishmaniasis, treatment failure is often a major cause of concern. The issue of treatment in American CL (ACL) is even more complex because of the factors that often influence the efficacy of the drugs, including the intrinsic and acquired variation in the sensitivities of the different Leishmania species [37]. Pentavalent antimonial drugs are the most prescribed treatments for American CL and MCL. The WHO recommends treating ACL with pentavalent antimonials at a dose of 20 mg/kg daily for 28 days [38]. There is no single effective treatment for all species of Leishmania. The choice of treatment strategy is based on geographical location and the infecting species [39]. Because of regional and species variability in treatment, doses of antimonials cannot be standardized, and local physicians determine appropriate dosages based on experience [40]. Therefore, it is clinically recognized that species and even strain-specific (regional) antimonial sensitivities are prevalent, which creates major impediments to adopting a single dose strategy for all types of leishmaniasis. There is also molecular and phenotypic heterogeneities that emerged in a natural L. donovani population from Nepal under antimonial treatment pressure. It has been proposed that each genetically distinct population can develop an antimonial resistant phenotype with a different molecular basis [41]. However, no single mechanism was identified for L. donovani. On the other hand, Leishmania strains causing self-healing CL were proposed to have greater susceptibility towards oxidative stress because they produced less non-protein thiols when compared to the VL species from the Indian subcontinent [15]. Here, we address this controversial issue of species-specific antimonial sensitivity in Leishmania by examining four Old World Leishmania species such as L. donovani, L. infantum, L tropica and L. major, and two New World species, namely L. panamensis and L. braziliensis. Our goal was to examine all commonly used drug resistant markers and determine whether there are any direct correlations. It is interesting to note that when some L. infantum clinical isolates from the Middle East showed CL phenotype, they produced 3–4 fold more AQP1 mRNA [42]. Therefore, the results of our study can be utilized for a larger study with clinical isolates and strains, but outside the scope of the work presented here.
The most commonly used antimonial resistance markers are: (i) MRPA[9]; (ii) thiols [6,11,41]; (iii) an unknown efflux system [8,43]; and (iv) AQP1 [16,19,44,45,46]. We examined carefully each of these factors in all six species. First, there is a correlation between the amount of MRPA mRNA and species-specificity (Fig. 2A). However, this is in contrast to the experimental evidence that MRPA is generally overexpressed in drug resistant Leishmania isolates [47]. MRPA is known to transport drug-thiol conjugates in Leishmania [9] and higher eukaryotes, including mammals [48,49]. Therefore, if correlated to the species-specific antimonial sensitivity, the MRPA mRNA levels in the CL species should have been lower along with lower non-protein thiol levels compared to the VL species. However, there was no discrimination in non-protein thiol levels between the VL and CL species (Fig. 2B), the second commonly used marker in antimonial resistance. Thus, we conclude that MRPA and non-protein thiol levels are not correlated to the species-specific antimonial sensitivity in Leishmania. The third factor is the Sb-[TS]2 efflux pump, but the rate of efflux is similar in all six species (Fig. 1B). The fourth factor is downregulation of AQP1, and this seems to be driving the species-specific antimonial sensitivity. The two VL species consistently showed less AQP1 mRNA (Fig. 2C), Sb(III) accumulation (Fig. 1A) and osmoregulatory capacity (Fig. 3). The parameters we studied to determine AQP1 functionality at the protein level and 3’-UTR derived mechanism(s) were extremely difficult to achieve with intracellular amastigotes, the clinically relevant form of the parasite, which need to be generated by in vitro macrophage infections. Axenic amastigote was developed to overcome this limitation, but was not successful for virulent Leishmania and specifically the CL species [50]. Also, we showed a similar pattern of antimonial resistance in the intracellular amastigotes of the VL species when compared to the CL species (Table 1). Hence, we chose to work with promastigotes, the vector form of the parasite.
In this study, we established that the species-specific antimonial sensitivity in Leishmania is being driven by the regulation of AQP1 at the mRNA level. Additionally, we showed that AQP1 mRNA is highly unstable in the VL species compared to the CL species (Fig. 4). Since, Leishmania does not have any transcriptional control, we determined the role of the AQP1 3’-UTRs in the species-specific stabilization of AQP1 mRNA using the luciferase reporter assay. As expected, the VL species AQP1 3’-UTR renders LUC mRNA unstable in all six species except in L. major (Table 2), which was corroborated by their basal LUC mRNA levels (Figs. 5A, B, 7A and S7A–S9A) and LUC activity (Figs. 5C, 7B and S7B–S9B). Much less LUC mRNA accumulates under the control of the VL 3’-UTRs both in VL and CL Leishmania species (Figs. 5B–7A and S7A–S9A), suggesting that there is no species-specific factor responsible for the regulation at the level of AQP1 mRNA stability, but rather it is the differences in sequences of the 3’-UTRs that make the trans-acting factor(s) more conducive to binding or not and to stabilizing or destabilizing the AQP1 mRNA. The correlation between LUC mRNA levels and LUC protein activity is not linear in L. major (Fig. 6), L. tropica (S8 Fig.) and L. panamensis (S9 Fig.) for the VL 3’-UTRs, suggesting that less mRNA is not necessarily associated with reduced protein levels. This suggests the presence of other factors (or differential expression of these factors) in some CL species that could increase translation rates, despite the fact that amounts of mRNA are low for certain genes including LUC. However, levels of AQP1 native mRNA (Fig. 2C) seem to be correlated linearly with protein expression as it corroborated with functional properties of AQP1, such as osmoregulation (Fig. 3) and Sb(III) accumulation (1A). The CL 3’-UTRs are similarly stable in VL species as they are in CL species, suggesting that there is no effect from any species-specific trans-acting factor. It was interesting to note that all 3’-UTRs of AQP1 provided LUC mRNA with the highest level of stability in L. major (Fig. 7), which is the most antimonial sensitive species under investigation. The most compelling evidence that the 3’-UTRs play a major role in AQP1 species-specific functionality comes from the fact that Lm-3’-UTR made Ld-AQP1 function three times more efficiently compared to its native 3’UTR (Fig. 8) and vice-versa. Additionally, AQP1 3’-UTR of L. braziliensis stabilized LUC mRNA, resulting in more LUC activity and protein expression in the VL species than in its native environment (Figs. 5 and 8), leading us to propose that L. braziliensis might harbor unique factors that are specific for its own AQP1 3’-UTR. However, basal level and stability of AQP1 mRNA in L. braziliensis were comparable to other CL species, such as L. major (Figs. 2C and 4). This difference could be attributed to the presence of specific AQP1 ORF sequences upstream to the 3’-UTRs in the native mRNA. In this context, it is interesting to note that L. braziliensis and L. panamensis ORF sequences differ (S2 and S6 Figs.) significantly from the four other species. L. donovani, L. infantum, L. major and L. tropica ORF sequences are closer to each other, sharing 87% identity among them (S10 and S11 Figs.), whereas the overall identity among all six ORFs is 66% (S2 Fig.). Therefore, the role of AQP1 ORFs in determining the native mRNA stability warrants further research in this direction, which is in progress.
The fundamental question is why there is a species-specific regulation of AQP1 in Leishmania. AQP1 is an adventitious facilitator of Sb(III); therefore, this species-specific antimonial resistance driven by AQP1 is a bonus that the VL species enjoy during treatment. Although acquired antimonial resistance in Leishmania is multifactorial, it is tempting to speculate that more antimonial resistant cases are observed in VL [13] due to down regulation of AQP1, because it is easy to downregulate something in which the intrinsic trend of higher mRNA instability leading to less production is already in place because the VL species likely does not need efficient osmoregulation. On the other hand, the physiological function of AQP1 is osmoregulation, and we showed that the CL species are better osmoregulators (Fig. 3). Is it possible that the CL species face greater osmotic challenges during vector to host transmission, or vice-versa, and more AQP1 helps them to overcome that barrier? AQPs are also implicated in a number of unrelated physiologic processes and functions, such as lipid metabolism, cell migration, epidermal biology, cell adhesion, and neural signal transduction [51]. Thus, it is also tempting to speculate that species-specific AQP1 expression may help the respective species to find their appropriate niches, resulting in tissue tropism. In VL and most CL cases, a standard single dose of 20mg/kg/day for 28 days of antimonials has been mandated by WHO since 1990 [52]. However, our data emphasize that treating all types of leishmaniasis with the same systemic dosage of antimony may not be a good practice. A lower dosage of antimonials as treatment for the visceral infection (which may be the correct dose for the cutaneous species as they are more sensitive) may have been the reason for the emergence of a more drug-resistant phenotype in that species. These are ambitious and yet intriguing and fundamental questions in Leishmania biology. Thus, our novel finding of 3’-UTR driven species-specific regulation of AQP1 is going to drive new approaches in that direction.
Wild type Leishmania donovani strain LdBob (kind gift from Professor Stephen M. Beverley at the Washington University School of Medicine), L. infantum strain MHOM/MA/67/ITMAP-263, L. major strain LV39 (kind gift from Professor Marc Ouellette, Laval University, Quebec, Canada), L. braziliensis strain MHOM/BR/75/M2903 (from ATCC), L. tropica strain MHOM/IL/67/JERICHO II (from ATCC), and L. panamensis strain MHOM/PA/71/LS94 (from ATCC) were used in this study. Promastigotes were grown at 26°C as described before [17]. Promastigotes were also grown on blood-agar/ brain heart infusion (BHI) broth biphasic medium containing (a) solid phase of 1.7% agar, 3.7% BHI and defibrinated rabbit blood and (b) liquid phase of 3.7% BHI broth. Human leukemia monocyte cell line THP1 was purchased from ATCC and maintained according to supplier’s instruction.
Antimony sensitivity of the promastigotes was determined as described previously [17]. Briefly, log phase promastigote cultures were diluted to 2 X106 cells ml−1 in a culture medium containing various concentrations of Sb(III) in the form of potassium antimonyl tartrate (Sigma). Following 72 h incubation, cell growth was monitored from the absorbance at 600 nm using a microplate reader (Spectramax 340, Molecular Devices). Percentage survival was plotted against Sb(III) concentrations and EC50 was determined using SigmaPlot 11.0. Each assay was performed at least three times in triplicates. Error bars were calculated from the mean ± SE.
Antimony sensitivity of amastigotes inside macrophages was determined after infecting THP1 derived macrophages. Briefly, 5 X105 THP1 cells/well/200 μl of RPMI were seeded in 16 chamber LabTek tissue culture slides (Nunc) and treated with 5 ng/ml phorbol myristate acetate (PMA) for 48h to differentiate into macrophages. Macrophages were infected with stationary phase promastigotes harvested from blood-agar/BHI biphasic medium at a parasite-to-macrophage ratio of 20:1 for 6 hours at 37oC with 5% CO2. Non-internalized promastigotes were washed away, and infected macrophages were treated with increasing concentrations of Sb(V) in the form of potassium hexahydroxoantimonate (Sigma) for 7 days. Medium was replaced every alternate day, and fresh drug was added. After 7 days, cells were stained with the Giemsa using Quick III Statpak kit (Astral Diagnostics). Numbers of amastigotes per 100 macrophages were determined by light microscopy. EC50 was calculated as described for promastigotes. Each assay was performed at least two times in triplicates. Error bars were calculated from the mean ± SE.
Log phase Leishmania promastigotes were washed twice with phosphate-buffered saline (PBS), pH 7.4 (Invitrogen) and suspended in PBS at a density of 108 cells ml−1. Promastigotes were then incubated with 10 μM Sb(III), a 200- μl portion was filtered through a 0.22 μm nitrocellulose filter at different time points (1, 5, 10, 20 and 30 min), and the filter washed once with 5 ml of ice-cold PBS. The filters were digested with 0.4 ml of concentrated HNO3 (69–70%) (EM Science) for 1 h at 70oC, allowed to cool to room temperature, diluted with high pressure liquid chromatography grade water (Sigma) to produce a final concentration of HNO3 of approximately 3%, and then analyzed by a PerkinElmer SCIEX ELAN DRC-e inductively coupled plasma mass spectrometer. Standard solutions were prepared in the range of 0.5–10 p.p.b. in 3% HNO3 using antimony standards (Ultra Scientific). Each transport experiment was repeated at least three times with duplicate samples. Error bars were calculated from the mean ± SE.
Membrane vesicles were prepared from promastigotes of each species as described previously [6]. They were rapidly frozen in liquid nitrogen in small aliquots and stored at -80°C until use. The total protein content of the plasma membrane fractions was determined by a filter assay as described previously [53]. ATP dependent uptake of Sb(TS)2 was measured in the presence of 10 mM ATP as energy source, as described previously, with a few changes [6]. Briefly, vesicles were added at 0.5 mg of membrane protein/ml and incubated with 0.1 mM of Sb(TS)2 in a buffer containing 75 MM Hepes-KOH, pH 7.0/0.15 M KCl. Reaction was started by the addition of ATP at room temperature. At the indicated intervals, samples (0.1 ml) were removed and filtered on wet 0.22 μm nitrocellulose filter, and the filter washed once with 5 ml of ice-cold PBS. The membranes were digested with 70% HNO3, and total Sb content was measured by ICP-MS as described above. Each transport experiment was repeated at least two times with triplicate samples. Error bars were calculated from the mean ± SE
Relative changes in cell volume following the induction of hypo-osmotic shock were measured as described earlier [54]. Briefly, log phase promastigotes were washed twice in PBS and re-suspended at a density of 109 cells ml−1. One-hundred-microliter portions of the cell suspension were transferred to a microtiter plate. Hypo-osmotic shock was induced by dilution of the isotonic cell suspension with an equal volume of deionized water, and the absorbance at 550 nm was recorded every 15 sec for 3 min in a microplate reader (Spectramax 340, Molecular Devices). A decrease in absorbance corresponds to an increase in cell volume. Isosmotic control experiments consisted of dilution of cell suspensions with appropriate volumes of isosmotic buffer. All hypo-osmotic shock experiments were conducted at a final osmolarity of 150 mOsm (1:1 dilution of isosmotic buffer and water). Each experiment was repeated at least three times in triplicate. Error bars were calculated from the mean ± S.E.
Genomic DNA was isolated using DNAzol reagent (Life Technologies). Total RNA from Leishmania promastigotes was isolated using TRIZOL reagent (Life Technologies) according to the manufacturer’s protocol. DNA was removed from total RNA preparation using TURBO DNA-free Kit (Ambion) according to the manufacturer’s instructions. Integrity of total RNA preparations was confirmed by denaturing agarose gel electrophoresis.
The AQP1 3’-UTR fragments from all six Leishmania species were mapped using 3’-RACE kit (Invitrogen) with 2 μg of total RNA as template according to the manufacturer’s protocol. AQP1 gene specific primers (S2 Table) were designed according to genome sequences in TriTrypDB. The amplified 3’-UTR fragments for all the species were cloned into pGEMT-Easy vector (Promega) according to the manufacturer’s instructions and sequenced (Eton Biosciences). As the database sequence of AQP1 ORF of L. tropica is not available, we cloned its AQP1 ORF from L. tropica genomic DNA using primers (sense 5’- GAATTC ATGAACTCTCCTACAACCATGCC-3’ and antisense 5’- GCGGCCGC CTAACAGCTGGGCGGAATGAT-3’) designed against L. major AQP1 ORF. This L. tropica AQP1 ORF sequence was used to design primers (S2 Table) for subsequent mapping of L. tropica AQP1 3’-UTR by 3’-RACE as described above.
Previously described [28] luciferase (LUC) expression vector for Leishmania pSPYNEOαLUC is referred to as LUC-control or pLUC in this study. The full-length 3′UTR of AQP1 and 200 base pairs beyond the 3’ end of the poly(A) site was PCR amplified from genomic DNA of each species using AccuPrime Taq DNA Polymerase (Invitrogen) and primers (S2 Table) with BamHI or SalI restriction sites inserted at 5’ and 3’ ends. To clone a similar 200 base pair sequence from L. tropica, a 555 base pairs fragment downstream to the poly(A) site was cloned using L. tropica genomic DNA using the sense primer (5’- GATGAGTGCACACGGCGTACTTC-3’) designed against L. tropica AQP1 ORF and the antisense primer (5’- ATGGTCGTACCACGCAAAGTCACC-3’) designed against L. major dtatbase (TriTrypDB) sequence. The sequence of this fragment was used to design primers (S2 Table) for subsequent cloning of L. tropica full-length 3′UTR of AQP1 and 200 base pairs downstream to the 3’ end of the poly(A) site. PCR products were cloned (cloning primers are described in S2 Table) into the pGEMT-Easy vector (Promega) and sequenced (Eton Bioscience) as described above. Different LUC-chimeric constructs were generated by digesting the pGEMT-Easy clones with BamHI or SalI (New England Biolab) and subcloned into the BamHI or SalI sites of LUC gene (3’ end) in the vector pSPYNEOαLUC, respectively. Directions of the cloned DNA segments were confirmed by sequencing. All plasmid constructs with forward orientation in respect to LUC open reading frame (ORF) were purified using QIAprep Spin Miniprep Kit (Qiagen). Purified plasmid constructs were transfected into Leishmania by electroporation. Briefly, stationary phase promastigotes were washed and resuspended in ice cold electroporation buffer (21 mM HEPES, 150 μM NaCl, 5 mM MgCl2, 120 mM KCl, 0.7 mM NaH2PO4, 6 mM glucose) at a density of 108 cells/ml. Three hundred microliters of resuspended cells were transferred to 0.2 cm electroporation cuvette (Bio-rad) with 10 μg of plasmid DNA. Cells were electroporated in Gene Pulser (Bio-rad) at 0.45 kV and 500 μF. All transfectants were selected and maintained in presence of 60 μg/ml geneticin (G418) (Invitrogen). The relative copy numbers of different LUC constructs in different transfectants were determined by qPCR using total genomic DNA from each transfectant. The relative abundance of target amplicons between samples was estimated by the 2−ΔΔCT method [55] using pteridine reductase 1 (PTR 1) as loading control. The processing of 3’ end of each 3’-UTR expressing from episomal copy was reconfirmed by 3’-RACE PCR as described above with gene specific primers designed from LUC ORF (S2 Table).
The full-length ORF of AQP1 was PCR amplified from genomic DNA of L. donovani (LdAQP1) and L. major (LmAQP1) using AccuPrime Taq DNA Polymerase (Invitrogen) and primers (S2 Table) with BamHI and XbaI restriction sites inserted at 5’ and 3’ ends respectively. The full-length 3′-UTR of AQP1 and 200 base pairs beyond the 3’ end of the poly(A) site were similarly PCR amplified using genomic DNA of L. donovani (Ld3’-UTR) and L. major (Lm3’-UTR) and primers (S2 Table) with XbaI restriction site inserted at 5’ and 3’ ends. PCR products were cloned into the pGEMT-Easy vector and sequenced as described above. pGEMT-Easy clones containing LdAQP1 and LmAQP1 were digested with BamH1 and XbaI (partial digestion with BamH1 for LdAQP1) and the resultant AQP1 ORF fragments were subcloned into pSP72-YHYG-αtubIR to generate LdAQP1-pSP72-YHYG-αtubIR and LmAQP1-pSP72-YHYG-αtubIR constructs. pGEMT-Easy clones containing Ld3’-UTR and Lm3’-UTR were digested with Xba1 and the resultant 3’UTR fragments were subcloned into both LdAQP1-pSP72-YHYG-αtubIR and LmAQP1-pSP72-YHYG-αtubIR constructs linearized with XbaI to generate LdAQP1-Ld3’-UTR-pSP72-YHYG-αtubIR (pLd-Ld), LdAQP1-Lm3’-UTR-pSP72-YHYG-αtubIR (pLd-Lm), LmAQP1-Ld3’-UTR-pSP72-YHYG-αtubIR (pLm-Ld) and LmAQP1-Lm3’-UTR-pSP72-YHYG-αtubIR (pLm-Lm) constructs (AQP1-3’-UTR constructs). Directions and integrity of the cloned DNA segments in AQP1-3’-UTR constructs were confirmed by sequencing. pSP72-YHYG-αtubIR (vector) and all AQP1-3’-UTR constructs were purified and transfected into L. donovani and L. major promastigotes as described above. All transfectants were selected and maintained in the presence of 300 μg/ml hygromycin B (Invitrogen). The relative copy numbers of different constructs in different transfectants were determined against hygromycin phosphotransferase gene by qPCR as described earlier.
cDNA synthesis was carried out using 500 ng of total RNA and AccuScript High Fidelity 1st strand cDNA synthesis kit (Agilent) according to the manufacturer’s instructions. The first-strand cDNA reaction mix was treated with 0.25N NaOH at 650C for 30 minutes to degrade the template RNA molecules. The reaction mix was neutralized using equimolar hydrochloric acid and purified using Qiagen PCR purification kit according to the manufacturer’s instructions. For qPCR, 10 ng of genomic DNA and for qRT-PCR, 2 μl of diluted purified cDNA reaction corresponding to 6 ng of template RNA, were used in a 10 μl reaction containing forward and reverse primers for the target genes (S2 Table) and 1X iQSYBR Green supermix (Bio-rad). The reactions were run on an Eppendorf Realplex2 PCR machine in the following thermal cycling conditions: initial denaturation at 950C for 3 minutes followed by 40 cycles of 950C for 15 sec and 650C for 20 sec. A final melting curve analysis was performed for each reaction to confirm that the PCR generated a single amplification product. Multiple primer sets against each target were designed using PrimerQuest software (Intergrated DNA technologies; http://www.idtdna.com/Primerquest/Home/Index?Display=SequenceEntry) and tested for their efficiency using the afore-mentioned thermal cycling conditions; the set(s) of primers showing efficiency between 98% to 102% were included in the current study. The relative abundance of target amplicons between samples was estimated using glyceraldehyde 3-phosphate dehydrogenase (GAPDH) as loading control by the 2−ΔΔCT method [55]. Error bars were calculated from the mean ± SD of three independent experiments in triplicate. Similar expression levels of GAPDH in all six species were confirmed using β-tubulin as the loading control (S3 Table).
To determine the half-lives of AQP1 or LUC mRNA, mid-log phase promastigotes were treated with sinefungin (5 μM) (Sigma) for 15 min followed by incubation with 10 μg/ml of actinomycin D (Sigma) to arrest trans-splicing and transcription, respectively. Cells were harvested just before adding actinomycin D and considered as the zero min time point. Subsequently, cells were harvested at 15, 30, 45, 60 and 120 min; an additional eight min of processing time was added while presenting the data. Total RNA isolation and qRT-PCR analysis were performed as described above using AQP1 or LUC ORF specific primers (S2 Table). Error bars were calculated from the mean ± SD of three independent experiments in triplicate.
Mid-log phase promastigotes (5X106) were lysed with 50 μl of lysis buffer (62.5 mM Tris-phosphate pH 7.8, 5 mM DTT, 2.5% Triton X-100, 25% glycerol). A 10 μl of lysate was used to estimate the luciferase activity using Luc-Screen Extended-Glow Luciferase Reporter Gene Assay System (Life Technologies) according to the manufacturer’s instruction. Error bars were calculated from the mean ± SE of three independent experiments in triplicate.
Whole cell lysates were prepared by lysing 1x107 promastigotes from each transfectant in 100 μl of 1x Laemmli’s buffer [56]. 10 μl lysate was used to fractionate proteins on 12% SDS-PAGE. Fractionated proteins were electroblotted on nitrocellulose membranes (Whatman) and probed sequentially with polyclonal goat anti-luciferase (Promega) and monoclonal mouse anti-α-tubulin (Sigma). The labeling was visualized with horseradish peroxidase-conjugated mouse anti-goat (Pierce) and rabbit anti-mouse (Abcam) respectively using a Western Lightning Chemiluminescence Reagent Plus system (PerkinElmer). Amount of luciferase expression relative to cells transfected with pSPYNEOαLUC was estimated by densitometric analysis using ImageJ software followed by normalization against the amount of α-tubulin of the respective cells. Error bars were calculated from the mean ± SE of two independent experiments.
The level of total intracellular non-protein thiol was measured in deproteinized cell extracts as described previously [6]. Briefly, log phase promastigotes (6 x 108) were harvested, washed with PBS and suspended in 0.6 ml of 25% tricholoracetic acid. Cell debris and denatured protein were removed by centrifugation at 16,000g for 20 min at 4°C after 10 min incubation on ice. The thiol content of the supernatant solution was determined using 0.6 mM 5,5'-dithio-bis(2-nitrobenzoic acid) (DTNB) in 0.2 M sodium phosphate buffer (pH 8.0). The concentration of 2-nitro-5-thiobenzoate (TNB), derivatives of non-protein thiol-DTNB reaction, was estimated spectrophotometrically at 412 nm. The concentration of total thiols in the test supernatants was estimated against a standard curve of cysteine. Error bars were calculated from the mean ± SE of three independent experiments in triplicate.
|
10.1371/journal.pbio.1002369 | Chromosome Synapsis Alleviates Mek1-Dependent Suppression of Meiotic DNA Repair | Faithful meiotic chromosome segregation and fertility require meiotic recombination between homologous chromosomes rather than the equally available sister chromatid, a bias that in Saccharomyces cerevisiae depends on the meiotic kinase, Mek1. Mek1 is thought to mediate repair template bias by specifically suppressing sister-directed repair. Instead, we found that when Mek1 persists on closely paired (synapsed) homologues, DNA repair is severely delayed, suggesting that Mek1 suppresses any proximal repair template. Accordingly, Mek1 is excluded from synapsed homologues in wild-type cells. Exclusion requires the AAA+-ATPase Pch2 and is directly coupled to synaptonemal complex assembly. Stage-specific depletion experiments further demonstrate that DNA repair in the context of synapsed homologues requires Rad54, a repair factor inhibited by Mek1. These data indicate that the sister template is distinguished from the homologue primarily by its closer proximity to inhibitory Mek1 activity. We propose that once pairing or synapsis juxtaposes homologues, exclusion of Mek1 is necessary to avoid suppression of all templates and accelerate repair progression.
| Chromosome segregation errors during meiosis may cause infertility, fetal loss, or birth defects. To avoid meiotic chromosome segregation errors, recombination-mediated linkages are established between previously unattached homologous chromosomes. Such recombination events initiate with breaks in the DNA, but how these breaks are preferentially repaired using the distal homologous chromosome, rather than the physically more proximal sister chromatid of similar sequence, is not well understood. Meiotic repair-template bias in the budding yeast depends on the function of Mek1, a meiosis-specific protein kinase. Previous models suggested that Mek1 activity creates repair-template bias by suppressing repair with the sister chromatid. We found that Mek1 localizes on meiotic chromosomes until the homologues pair and closely align. Removal of Mek1 requires the assembly of a conserved zipper-like structure between meiotic chromosomes, known as the synaptonemal complex. DNA break repair is delayed in mutants in which Mek1 persists on closely aligned homologues. These findings suggest that persistent Mek1 activity can suppress repair from all templates, and that one function of the synaptonemal complex is to remove this activity from chromosomes. Our findings build on previous models to propose that Mek1 activity creates a local zone of repair suppression that is normally avoided by the spatially distant homologous chromosome to promote repair-template bias.
| Meiosis is a specialized cell division that produces haploid gametes from diploid progenitors and is essential for sexual reproduction. The reduction in ploidy is achieved by a unique chromosome division phase (meiosis I) that segregates homologous chromosomes (homologues). Errors in this process are a leading cause of infertility, miscarriages, and birth defects in humans [1]. Proper meiosis I chromosome segregation in most organisms requires that each homologue pair be linked by at least one crossover. Crossover formation occurs during the extended prophase preceding meiosis I and is promoted by the programmed induction of DNA double-strand breaks (DSBs). Resection of these breaks exposes single-stranded DNA tails that invade a donor template for repair. A subset of strand-invasion reactions subsequently matures to form double Holliday junctions, which are generally resolved as crossovers [2].
To promote linkages between homologues, meiotic DSB repair is strongly biased toward using the homologue rather than the physically more proximal sister chromatid [3,4]. Available evidence, stemming mostly from studies in the budding yeast Saccharomyces cerevisiae, suggests that homologue bias results primarily from suppression of repair from the sister template. In yeast, this barrier to sister repair is mediated to a large extent by the chromosomal kinase Mek1, a meiosis-specific orthologue of mammalian CHK2 kinase that is recruited to the axial element structures of meiotic chromosomes upon DSB formation [3,5]. Mek1 recruitment requires the phosphorylation of the chromosome axis protein Hop1 on threonine 318 (T318) by the checkpoint kinases Mec1 (ATR) and Tel1 (ATM) [6]. Chromosomal recruitment leads to the dimerization and activation of Mek1 [7]. Mek1, in turn, phosphorylates a variety of targets, including the repair factors Rad54 and Rdh54, as well as histone H3 [8,9]. Phosphorylation of Rad54 inhibits its interactions with the recombinase Rad51 and is thought to help suppress sister-targeted repair along with additional Mek1 targets that remain to be identified [8]. In addition, Rad51 is kept inactive by its meiosis-specific inhibitor Hed1, which further biases repair towards the homologue [10–12]. Current models suggest that a DNA “tentacle” formed by the assembly of Rad51 and the meiotic recombinase Dmc1 on one end of a DSB interprets the suppressive signal established by Mek1, leading to preferred repair engagement with the homologue [3,4,13–15].
For homologue bias to be established, the DSB repair machinery must be able to distinguish homologue and sister templates. A probable mechanism included implicitly or explicitly in many models of homologue bias is that the sister chromatid is identified either by spatial proximity and/or through the cohesive linkages resulting from DNA replication [3,4,13–15]. As a result, the sister is subject to Mek1-dependent repair suppression, whereas the generally spatially more distant or unlinked homologue is not. An extension of the spatial proximity model is that if the homologue were to be transported within the range of Mek1 activity, it would also become suppressed as a template. Indeed, experimental hyperactivation of Mek1 also delays interhomologue repair [16], suggesting that Mek1-dependent repair suppression does not inherently distinguish between sister chromatids and homologous chromosomes.
The close alignment of homologous chromosomes is an essential part of meiotic prophase that in many organisms culminates in the assembly of the synaptonemal complex (SC). The SC is a conserved tripartite structure that assembles along the entire length of paired homologues during meiotic prophase [17]. In S. cerevisiae, SC assembly (synapsis) initiates at sites of crossover designation and centromeres [18–20] and involves the progressive deposition of Zip1, an extended coiled-coil protein that aligns homologous chromosomes at a fixed distance [21]. The function of the SC remains obscure. The SC is thought to stabilize pairing interactions, but homologues often remain co-aligned, albeit at a greater distance, in the absence of Zip1 [22]. Recent experiments have hinted at a signaling role for the SC. In several organisms, SC assembly is associated with a loss of chromosomal proteins, most notably yeast Hop1 and the orthologous HORMAD proteins in mouse [23–26], as well as the yeast DSB regulators Rec114 and Mei4, which require Hop1 for recruitment [27–29]. As DSB levels are elevated in SC mutants, these observations have led to the model that the SC acts as a feedback signal to suppress DSB formation on chromosomes that have engaged in crossover repair [10,30], although some evidence suggests that this suppression is not absolute [31,32]. Given that Mek1 recruitment also depends on chromosomal Hop1, the synapsis-associated loss of Hop1 would also be expected to affect Mek1 binding. Unexpectedly, however, Mek1 was reported to persist on synapsed chromosomes [33].
Here, we reinvestigated the chromosomal dynamics of Mek1 and its role in regulating meiotic DSB repair. We demonstrate that Mek1 is in fact eliminated from synapsing chromosomes and that removal requires Zip1-mediated recruitment of the AAA+-ATPase Pch2. Moreover, we show that DSB repair on synapsed chromosomes requires the function of Rad54, a target of Mek1-dependent inhibition. Importantly, failure to remove Mek1 from synapsed chromosomes leads to delays in DSB repair, indicating that Mek1 must be inactivated on fully engaged chromosomes to ensure timely completion of meiotic DSB repair.
To investigate the dynamics of Mek1 binding to meiotic chromosomes, we analyzed chromosome spreads using a functional Mek1-GFP construct. Nuclei were staged based on the progressive deposition of Zip1, marking the assembly of the SC along meiotic chromosomes. An NDT80 deletion was used to prevent SC disassembly and Mek1 degradation due to exit from meiotic prophase [32,34]. Mek1 foci were abundant on chromosomes prior to the association of Zip1 (Fig 1A). However, in contrast to a previous report, which detected many apparent Mek1 foci on synapsed chromosomes [33], we observed a notable loss of chromosomal Mek1 from regions of extended Zip1 staining, such that Mek1-GFP foci were nearly undetectable when all chromosomes had assembled an SC (Fig 1A). The reason for this discrepancy is unclear but may be related to differences in strain background. The loss of chromosomal Mek1 signal was confirmed using a polyclonal antibody against Mek1 (S1A Fig), and was not due to a drop in Mek1 protein levels during meiotic prophase [34]. Examination of nuclei with partially assembled SC indicated that the disappearance of Mek1 foci was directly correlated with SC deposition even on individual chromosomes. As SC formation is relatively rapid in wild-type cells, we confirmed this observation in a zip3Δ mutant, in which chromosome synapsis is delayed and limited [18,20]. Like in the wild-type situation, Mek1-GFP signal was strongly reduced in synapsed regions but persisted on unsynapsed chromosomes in a zip3Δ mutant (Fig 1B). These data indicate that chromosome synapsis coincides with a loss of chromosomal Mek1.
The disappearance of Mek1 from chromosomes was mirrored by a loss of Mek1-dependent chromatin marks. Phosphorylation of histone H3 T11 requires Mek1 activity [9]. Immunostaining using an antibody specific for H3-pT11 revealed numerous foci on unsynapsed chromosomes but a near complete absence once chromosomes were synapsed (Fig 1C), implying that Mek1 is not active on synapsed chromosomes. To support this observation, we analyzed H3-pT11 by western blotting in a meiotic time course. Cells were blocked at the end of prophase using an ndt80Δ mutation to avoid secondary effects from Mek1 inactivation after prophase [38,39]. H3-pT11 first became detectable at 3 h after meiotic entry (Fig 1D), corresponding to the time of DSB induction. This timing correlated well with the phosphorylation of other meiotic checkpoint targets, including Zip1, Zip3 and Sae2. H2A-pS129 accumulated earlier presumably because of its role in premeiotic DNA replication [40]. Consistent with the analysis of chromosome spreads, H3-pT11 signal disappeared 5 h after meiotic induction, when most cells in the culture were completing SC formation (Fig 1E). The disappearance of H3-pT11 is in contrast to the other tested checkpoint targets, which remained phosphorylated during SC formation (Fig 1D and 1E).
Intriguingly, the SC-associated loss of Mek1 appeared to occur irrespective of persistent DNA repair intermediates. zip3Δ mutants are severely defective in DSB repair [18,41] and accumulate abundant repair foci marked by the Rad51 recombinase (S1B Fig). However, whereas Mek1 signal was largely restricted to unsynapsed regions in zip3Δ mutants, Rad51 foci were abundantly detectable on both unsynapsed and synapsed chromosomes. These data indicate that, at least in the absence of ZIP3, completed chromosomal DNA repair is not a prerequisite for the loss of Mek1 from synapsed chromosomes.
To test whether the SC is responsible for the loss of Mek1, we removed Zip1 from meiotic chromosomes. To circumvent potential pleiotropic effects of earlier roles of Zip1 in centromere pairing and DSB repair [42], we used the “anchor-away” technique [37] to conditionally deplete Zip1 from chromosomes that had already assembled SCs. In this technique, proteins tagged with the FRB domain of human mTOR are actively depleted from the nucleus after rapamycin addition due to interaction with a cytoplasmic anchor (a ribosomal protein fused to FKBP12; Fig 1F). Zip1 was quantitatively depleted from meiotic chromosomes within 2 h of rapamycin addition (S1C Fig). Nuclear depletion of Zip1-FRB throughout meiosis caused defects in sporulation and spore viability approximating the zip1Δ mutant, whereas untagged control strains treated with rapamycin retained wild-type spore viability (S1A and S1B Table). Strikingly, specific removal of Zip1 starting at the 6 h time point, when the vast majority of nuclei have fully synapsed chromosomes, caused rapid reaccumulation of Mek1 on chromosomes (17/20 nuclei; Fig 1G). We conclude that Zip1 assembly on chromosomes promotes the removal of Mek1 and is required to maintain Mek1 exclusion from synapsed chromosomes.
Chromosomal recruitment and activation of Mek1 requires the phosphorylation of Hop1-T318 [6], which may be subject to Zip1-dependent regulation. Consistent with this notion, the phosphorylation-dependent slower migrating forms of Hop1 disappear at the time of SC extension (Fig 2A) [40]. To more directly test the role of Hop1-T318, we raised a polyclonal antibody that specifically recognizes the phosphorylated form of this residue (Figs 2A and S2A). Immunofluorescence analysis revealed that, similar to Mek1, Hop1-pT318 foci were abundantly present in early prophase but disappeared coincident with SC assembly (Fig 2B and 2C). Moreover, we observed an increased accumulation of Hop1-pT318 signal in cell extracts and on meiotic chromosomes after Zip1 was depleted from the nucleus by anchor-away compared to the undepleted controls (Fig 2D and 2E). These data are consistent with a model whereby the disappearance of Mek1 from synapsed chromosomes is the result of a loss of Hop1-pT318 epitopes.
Loss of at least some Hop1-pT318 epitopes is likely a secondary consequence of the reduced binding of Hop1 to synapsed chromosomes (S2B Fig) [23,25]. Indeed, Hop1 re-accumulates on chromosomes after depletion of Zip1 (Fig 2F and 2G). Hop1 removal from synapsing chromosomes requires the SC-bound AAA+-ATPase Pch2 [24,43,44] and super-resolution microscopy of the SC in pch2Δ mutants revealed an over-enrichment of Hop1 in two parallel tracts along the length of the lateral elements (Fig 3A). This effect was specific for Hop1, as the staining patterns of other SC lateral and central element components with respect to Zip1 appeared similar in wild type and pch2Δ mutants (S3A–S3C Fig). Significantly, Hop1-pT318 and Mek1 foci were visible on fully synapsed chromosomes in pch2Δ mutants (Fig 3B and 3C), which are almost never seen in wild-type cells. Chromosomal accumulation of Mek1 occurred independently of the ndt80Δ arrest (Fig 3D). Furthermore, phosphorylated Hop1 was abundant in pch2Δ whole-cell extracts (Fig 3E). These data indicate that Pch2 is responsible for the loss of Hop1-pT318 and Mek1 from synapsing chromosomes.
The persistence of Hop1 on synapsed chromosomes in the pch2Δ mutant is associated with unusually distinct parallel tracts of DAPI-stained chromatin along the lengths of chromosomes, a conformation only occasionally observed in short stretches on synapsed wild-type chromosomes (Fig 3F). Previous analyses had shown that PCH2 is required for establishing separate domains of Zip1 and Hop1 along chromosomes, which fail to be formed in pch2Δ mutants [43,44]. We speculate that the distinctive parallel organization of chromosomes observed in pch2Δ mutants is another reflection of this altered chromosome structure, although we currently do not know whether the chromosomal persistence of Hop1 or Mek1 is responsible for this chromosome conformation in pch2Δ mutants.
Our data suggest that removal of Mek1 depends on a Pch2-associated function during synapsis. Further analysis identified a non-null allele of ZIP1 that assembles SC but fails to recruit Pch2 to synapsed chromosomes. Cells lacking a leucine-zipper in the coiled-coil region of Zip1 (zip1-4LA) [45] exhibited overall wild-type SC structure but lost all SC-associated Pch2 staining (Figs 3G and S3D–S3F). By contrast, the nucleolar pool of Pch2, which is independent of ZIP1 [43], persisted in these mutants. Consistent with the failure to recruit Pch2 to the SC, zip1-4LA mutants retained large amounts of Hop1 on synapsed chromosomes (Fig 3A) and accumulated high levels of phosphorylated Hop1 and Mek1 (Fig 3B, 3C and 3E). Furthermore, as seen upon loss of PCH2, nuclear spreads of zip1-4LA mutants exhibited distinctly parallel DAPI tracks (Fig 3F). We note that the Pch2-mediated checkpoint, which specifically involves the nucleolar pool of Pch2 [43], remains active in zip1-4LA mutants [45]. These observations suggest that the Zip1-mediated recruitment or stabilization of Pch2 couples SC assembly to the removal of Mek1.
We investigated whether the SC-associated loss of Hop1-pT318 is mediated by dephosphorylation in addition to Hop1 removal. PP4 protein phosphatase, comprising the catalytic subunit Pph3 and the cofactor Psy2, negatively regulates Hop1 phosphorylation [35,46]. To specifically interrogate the role of PP4 in Hop1 dephosphorylation during chromosome synapsis, we conditionally depleted the PP4-cofactor Psy2-FRB by anchor-away at the time of full synapsis. Nuclear depletion of Psy2 caused a modest accumulation of Hop1-pT318 signal in cell extracts (Fig 4A) and an increase in Hop1-pT318 focus number (Fig 4B), indicating that PP4 contributes to the removal of Hop1-pT318. Unlike in pch2Δ mutants, the increases in Hop1-pT318 signal were not associated with an increase in chromosomal Hop1 levels (S4A and S4B Fig). We do note, however, that the Hop1-pT318 signals often appeared at sites of discontinuity in Zip1 staining (Fig 4B), perhaps reflecting sites where Hop1 persists on synapsed chromosomes. These findings suggest that PP4 acts in parallel to Pch2 in eliminating Hop1-pT318 (and thus Mek1).
The reappearance of Hop1-pT318 foci upon PP4 depletion also presented a puzzle, as it implied an increasing number of unrepaired DSBs on synapsed chromosomes. DSB formation, a prerequisite for Hop1 phosphorylation and Mek1 recruitment [6,47], is thought to be largely shut down upon homologue engagement and synapsis [10,28,30,48,49], although several groups have reported continued presence of DSBs in ndt80 mutants [30–32]. To test for the presence of unrepaired DSBs, we analyzed Rad51 focus number upon Psy2-FRB depletion. Nuclei with fully synapsed chromosomes displayed very few Rad51 foci when Psy2 was present (Fig 4C). By contrast, Psy2-FRB depletion led to a significant increase in Rad51 focus number on synapsed chromosomes that matched Hop1-pT318 accumulation (Fig 4C), suggesting an increased presence of DSBs. The accumulating Rad51 foci may reflect DSB repair intermediates that became destabilized upon PP4 depletion. Alternatively, they may represent continued DSB formation on synapsed chromosomes in the absence of PP4 activity. This latter possibility would imply that DSB formation continues on synapsed chromosomes.
To begin to distinguish between these possibilities, we first asked whether DSB formation can be restored upon removal of the SC, which would indicate that DSB suppression associated with synapsis is reversible. We depleted Zip1-FRB by anchor-away and used immunofluorescence analysis of Rad51 to monitor DSB levels (Fig 5A). Zip1 depletion led to a significant increase in steady-state focus number of Rad51 (Fig 5A and 5B). This effect is not observed in untagged control cells (S5A Fig) and is mirrored by an increase in steady-state focus numbers of the single-stranded DNA-binding protein RPA (S5B Fig). Importantly, co-depletion of Zip1 and an essential DSB factor, Mer2, did not lead to an increase in Rad51 foci (Fig 5B). This outcome was not due to non-specific disruption of Mer2 by the FRB tag, because Mer2-FRB strains accumulated near wild-type levels of Rad51 foci prior to synapsis and produced fully viable spores in the absence of rapamycin (S5C Fig and S1A and S1B Table). These data indicate that new DSBs form in a Mer2-dependent manner after SC depletion.
We used physical assays at several endogenous DSB hotspots to monitor the occurrence of new DSBs upon depletion of Zip1 [30,50]. Electrophoretic separation of restriction-digested genomic DNA followed by Southern analysis allows detection of the larger unbroken DNA (parental size) as well as the faster migrating DSB fragments. DSB fragments reappeared at the ERG1 hotspot in the ZIP1-FRB strain but not in the untagged control strain following rapamycin addition (Fig 5C and 5D). These DSBs were absent when Mer2 was co-depleted (Fig 5D and S6A Fig), indicating that they represent newly formed DSBs and are not the result of destabilized repair intermediates. DSB signal may be further increased due to the loss of Zip1 repair functions upon depletion [42]. A similar increase in DSBs was also observed at the YIL094c hotspot after Zip1 nuclear depletion (Fig 5E and S6B Fig). However, Zip1 nuclear depletion did not lead to significant DSB reappearance at the YGR279c or the YCR047c hotspot (Fig 5F and S6E Fig). Thus, although the increase in DSBs after Zip1 nuclear depletion is consistent with the notion that Zip1 prevents the formation of new DSBs on fully synapsed chromosomes [30], our data suggest that this suppression may occur in a locus-specific manner. We note that following Zip1 depletion, the DSB bands at several hotspots migrated at a higher molecular weight than DSB fragments observed in early prophase (Fig 5C and S6B–S6D Fig), suggesting that processing of DSB ends is altered in this situation.
Given that depletion of PP4 led to an increase in Rad51 foci even in the presence of Zip1 and that previous studies have reported continued presence of DSBs in ndt80 mutants [30–32], we asked whether some DSB formation is maintained when chromosomes appear fully synapsed in late prophase. To test this possibility, we depleted DSB repair factors from synapsed chromosomes to trap newly formed DSBs. We chose Rad54, which promotes Rad51-dependent DSB repair, and Rdh54, a Rad54-like protein that activates the meiosis-specific recombinase Dmc1. Dmc1 and Rdh54 are required for homologue-directed repair in meiosis [51,52], whereas Rad54 is inhibited by Mek1 to suppress intersister repair [8]. We reasoned since Mek1 is nearly absent on synapsed chromosomes, Rad54 may become active in this situation. We used anchor-away to deplete Rdh54-FRB and Rad54-FRB from synapsed chromosomes (Fig 5G and 5H). No increase in Rad51 focus number was observed upon removal of Rdh54 (Fig 5H), although nuclear depletion of Rdh54-FRB throughout meiosis caused an expected reduction in sporulation efficiency, indicating effective depletion (S1A Table). By contrast, Rad54 removal led to a strong increase in Rad51 focus number on synapsed chromosomes (Fig 5G and 5H). Although this finding may indicate that DSB formation continues on synapsed chromosomes, previous studies indicated that Rad51 also associates with undamaged DNA in the absence of Rad54 activity [53]. To address this possibility, we co-depleted a DSB-cofactor Mer2 or the DSB-inducing enzyme Spo11 with Rad54-FRB. Co-depletion of either factor significantly reduced Rad51 focus formation on synapsed chromosomes (Fig 5H). These data strongly suggest that DSB formation continues even when chromosomes appear fully synapsed, and that DSB turnover depends on Rad54.
Southern analysis indicated that DSB accumulation on synapsed chromosomes upon Rad54 depletion is locus-dependent. The accumulation of unrepaired DSBs was apparent at the YGR279c and YCR047c DSB hotspots (Fig 5F and S6D–S6F Fig), whereas the DSB signal at the YIL094c and ERG1 hotspots did not increase substantially above background (Figs 5D, 5E and S6B). Interestingly, these patterns of DSB accumulation are opposite to the patterns observed upon Zip1 depletion (Figs 5D–5F and S6D–S6F). Thus, these differences may reflect the varying propensities of different genomic regions to synapse or differential dependence on ZIP1 function for DSB repair. Alternatively, individual hotspots may differ in their dependence on Hop1/Mek1 for DSB formation and/or repair. In contrast to the slower-migrating DSB fragments after nuclear depletion of Zip1, the DSB fragments that appeared at the YGR279c and the YCR047c locus after nuclear depletion of Rad54 were faster migrating compared to DSBs in early prophase (S6D and S6F Fig; compare DSB pattern at T = 3 h to rapamycin-treated sample in Rad54-FRB). This migration pattern is consistent with hyperresection of DSBs ends and is typically observed when strand-invasion activity is blocked [54].
Despite the accumulation of Rad51 foci in Rad54-depleted nuclei, we observed no defect in SC structure (Fig 5G) and no increase in Hop1-pT318 focus number, overall Hop1 phosphorylation, or total chromosomal Hop1 signal upon Rad54 depletion (S7A–S7C Fig). This behavior is in stark contrast to the commensurate increase in Rad51 and Hop1-pT318 foci upon depletion of Zip1 (Figs 2D, 5A and 5B) or PP4 (Fig 4A–4C). These observations support the model that Zip1-dependent Hop1 removal and PP4 activity collaborate to prevent Hop1-T318 phosphorylation on synapsed chromosomes. We conclude that unrepaired DSBs do not lead to Mek1 recruitment when chromosomes appear fully synapsed.
The loss of Mek1 activity upon SC formation suggests that DSB repair on already synapsed chromosomes may not be constrained by homologue bias. To test this possibility, we investigated the formation of intersister (IS) and interhomologue (IH) double Holliday junction (dHJ) intermediates over time in ndt80Δ mutants at two DSB loci, HIS4-LEU2 and GAT1. Engineered restriction site polymorphisms surrounding these DSB sites permit the separation of IS and IH repair intermediates by two-dimensional gel electrophoresis [30,55] (Fig 6A). As ndt80Δ mutants accumulate unresolved dHJs, analysis of IS and IH dHJs at a given time point will provide the cumulative average of template bias up until that time point. Analysis of the HIS4-LEU2 hotspot revealed a strong IH bias that persisted over time (S8A Fig), consistent with previous results [56]. GAT1 reproducibly exhibited a weaker IH bias (IH:IS ~1.5:1; Fig 6B and S8B Fig) than other strong DSB hotspots (IH:IS ~4:1) [10,56], but still substantially higher than the IH:IS ~1:9 template bias observed for mitotic DSB repair [57]. Notably, the cumulative IH:IS ratio at GAT1 became progressively lower (Fig 6B and S8B Fig) consistent with decreased IH bias at later time points. These data support the notion that, at least at the GAT1 locus, meiotic repair constraints are relaxed after chromosomes are fully synapsed. Because technical difficulties precluded us from analyzing IH bias at additional loci, we do not know to what extent this effect extends to other DSB hotspots.
A major mechanism of establishing homologue bias is to make repair from the sister chromatid more difficult [13,58]. One conceptually simple way to achieve this goal is to establish a Mek1-dependent “zone” of repair suppression, such that spatially proximal sequences (i.e., the sister) cannot easily be used as repair templates [13]. If so, then removal of Mek1 may be necessary once chromosomes are aligned, as alignment would also place the homologue in this zone of repair suppression, thereby rendering repair from all templates equally difficult (see Fig 7). This model predicts that unrepaired DSBs should accumulate in cells that fail to remove Mek1 from synapsed chromosomes. Indeed, we observed an accumulation of Rad51 foci on fully synapsed chromosomes of pch2Δ mutants (Fig 6C). Rad51 accumulation occurred independently of the ndt80Δ-mediated prophase arrest (S8C Fig) and is consistent with previous observations showing a delay in DSB repair in these mutants [44,59]. To test whether the increased Rad51 foci on synapsed chromosomes are due to the persistence of Mek1 activity, we used an allele of Mek1 (mek1-as) that can be conditionally inactivated upon addition of a small molecule inhibitor (1-NA-PP1) [60]. Addition of the inhibitor after chromosomes were synapsed led to the disappearance of Rad51 foci in pch2Δ mek1-as mutants, whereas the foci persisted in untreated control cells (Fig 6D and 6E), suggesting rapid repair of DSBs once Mek1 was inactivated.
To confirm these results, we performed Southern analysis at the ERG1 and YCR047c DSB hotspots. Consistent with the persistence of Rad51 foci, pch2Δ mutants accumulated DSBs at both hotspots (Fig 6F and S8D Fig). The persistent DSBs differed in their processing from DSBs formed in early prophase, similar to what was observed upon Zip1 depletion (S8E Fig). Importantly, the DSB bands were lost at both hotspots upon inactivation of Mek1 (Fig 6F and S8D Fig). Together, these results indicate that one function of the SC is to prevent Mek1 association with synapsed chromosomes in order to allow rapid DSB repair following homologue engagement.
Here, we used stage-specific depletion experiments to investigate the function of chromosome axis proteins and the SC in controlling meiotic DNA-repair signaling by Mek1 kinase. Our data suggest that Mek1 activity, while being essential for establishing meiotic repair template bias, creates a problem for DSB repair when all templates are in close proximity. We show that the SC transverse filament protein, Zip1, promotes Pch2-mediated exclusion of Mek1 from fully paired chromosomes. We further show that the ectopic presence of Mek1 on synapsed chromosomes prevents DSBs from undergoing Rad54-mediated turnover. Thus, by promoting the removal of Mek1, the assembly of Zip1 into SC structure can directly modulate DSB repair pathways. The chromosome-autonomous nature of SC assembly provides an obvious means to differentially control this process between chromosomes.
For repair template bias to be established, cells must be able to distinguish sister chromatids from homologous chromosomes. Our data point to a simple mechanism, whereby the primary determinant distinguishing sister from homologue is the spatial distance of the respective template from DSB-associated Mek1 activity (Fig 7). This model is in line with current models of template choice [3,4,13–15], which propose that a Mek1-dependent inhibitory domain suppresses repair progression from the proximal sister template, while the generally more distant homologue escapes this suppression. It is further supported by the observation that hyperactivation of Mek1 also delays interhomologue repair [16]. We argue, however, that a consequence of this simple setup is that once homologous chromosomes pair and establish close juxtaposition, Mek1 must be inactivated, so as not to place the homologue in the inhibitory domain and thus render all possible repair templates unfavorable. The stochastic nature of chromosome pairing would require this inactivation to be coupled to the behavior of individual chromosomes.
One prediction emerging from this model is that Mek1 activity along chromosomes must be spatially and temporally restricted, a notion supported by our experiments. In addition, Mek1 recruitment must be dynamic, as the genomic distribution of DSBs varies from cell to cell. Accordingly, Hop1 distribution is highly stereotyped and DSB-independent [27], whereas Mek1 recruitment is coupled to DSB induction [6,47].
The model that Mek1-dependent suppression of DSB repair is not inherently selective for the sister can also explain why DSBs persist in pch2Δ mutants. Chromosome pairing is unaffected in pch2Δ mutants [43], implying functional interhomologue repair interactions. However, repair completion may be suppressed because Mek1 activity persists on these chromosomes. This model may also explain why both crossover and non-crossover formation is delayed in pch2Δ mutants while the formation of single-end invasion intermediates occurs with wild-type kinetics [44]. Mek1 has been suggested to promote homologue bias in part by sequestering one DSB end in a quiescent state [14,61]. Perhaps persistent Mek1 activity hinders use of the sequestered end for completion of repair, thereby equally affecting crossover and non-crossover repair. Alternatively, the SC structure may create a situation that renders interhomologue repair structurally difficult, while Mek1 activity hinders repair with the sister template in pch2Δ mutants. Intriguingly, despite the severe repair delay, pch2Δ mutants display a wild-type level of spore viability [43,50]. This result implies that Mek1 suppression of DSB repair can be overcome with time and supports the notion that meiotic template choice is not absolute but rather the consequence of a kinetic barrier to repair [13].
We speculate that the presence of Mek1 may also contribute to the accumulation of Rad51 foci when PP4 or Zip1 are depleted from synapsed chromosomes (Figs 3C and 4A). In both experiments, Mek1 was bound to chromosomes that were allowed to fully pair prior to experimental manipulation (Zip1-FRB, Psy2-FRB), creating a situation similar to what is observed in pch2Δ mutants. Thus, although PP4 and Zip1 clearly have additional roles in recombination [35,42], the presence of Mek1 may further impair DSB turnover in these situations.
The loss of Mek1 upon chromosome synapsis implies that meiotic repair constraints become progressively relaxed at late stages of meiotic prophase, such that repair perhaps transitions into a mitotic-like state. One likely consequence of this transition is that at least some of the DSB repair on synapsed chromosomes depends on the mitotic DSB repair factor Rad54, which presumably promotes Rad51-dependent repair. Moreover, as Rad54 activity mediates the disassembly of Rad51 filaments [62], it may also promote release of the second DSB end, which is thought to be held in a quiescent state by Rad51 [14]. The loss of Hop1 and Mek1 from synapsed chromosomes may also cause a down-regulation of Dmc1 activity, as Dmc1 no longer promotes meiotic DSB repair in the absence of either Hop1 or Mek1 [14]. Such loss of activity may explain why the DSBs that persist on synapsed chromosomes upon Rad54 depletion are not repaired and why depletion of the Dmc1-cofactor Rdh54 had little effect.
The successive implementation of repair constraints may be particularly important for the repair of DSBs in regions that lack an allelic sequence for repair, such as inversions or deletions, which, in fact, are repaired efficiently using the sister [13]. Chromosome synapsis is not dependent on sequence homology [63] and can thus spread into such regions from sites of SC nucleation. Conversely, constitutive binding of Zip1, as observed at yeast centromeres [64], may constitutively prevent the recruitment of Mek1 and thus activation of meiotic template bias. Indeed, deletion of Zip1 leads to an increase in interhomologue recombination specifically around centromeres in the absence of increased DSB formation [65], consistent with the model that centromeric DSBs are primarily repaired from the sister.
Our results complement a growing body of evidence that identifies the SC as a macromolecular signaling conduit. By extending out from sites of crossover designation, the SC may communicate successful engagement in crossover repair to the rest of the chromosome and trigger a profound switch in meiotic chromosome behavior, including the remodeling of meiotic chromosome structure and the dampening of further DSB activity [10,28,30,44,66]. Our work adds to this list the relaxation of meiotic repair constraints as a result of the SC-dependent removal of Mek1. Work in mice suggests that SC-dependent changes in chromosome structure and DSB activity are conserved [26,67]. It remains to be determined whether the same is true for the loss of repair constraints. Like in yeast, phosphorylation of HORMAD proteins is limited to unsynapsed chromosome axes in mice [68], and synapsis leads to strong TRIP13/Pch2-dependent depletion of chromosomal HORMAD proteins [26]. However, higher eukaryotes do not encode a clear Mek1 orthologue. Although CHK2 kinase could conceivably fulfill the role of Mek1 in these organisms, mouse CHK2 was recently shown to be required for checkpoint function without having a direct role in repair [69]. However, a role for the SC in regulating repair pathway choice is apparent in Caenorhabditis elegans, as partial depletion of the SC central region structure leads to increased interhomologue crossover events [70,71]. Although synapsis initiates independently of meiotic recombination in this organism [72], the change in repair parameters is associated with altered axial compaction [70], which may be functionally related to the altered DAPI patterns apparent in yeast pch2Δ mutants (Fig 3).
Ultimately, the transition in meiotic recombination mediated by the SC likely has at least two functions. First, it may preserve the pattern of crossover distribution by limiting the formation of additional crossovers [73]. Second, it minimizes the risk of aberrant repair events by restricting DSB numbers and by promoting the rapid repair of the DSBs that do form. Importantly, by executing this transition in cis, this feedback is robust to the inherently stochastic nature of chromosome pairing and meiotic crossover formation, and allows chromosomes to respond individually in a shared nuclear environment.
Antibody production was approved by the University Welfare Committee of New York University.
All yeast strains used are in the SK1 background except strains AM2981 and K303 (S3 Fig), which are in the BR1919-8B background. Genotypes are listed in S2 Table. Epitope tags and gene deletions were made by standard PCR-based transformations, except in the case of ZIP1-FRB. For construction of ZIP1-FRB, a previously published internally tagged ZIP1-GFP::URA3 plasmid [74] was used and GFP replaced with the FRB sequence before integration at the ZIP1 locus. URA3 along with the wild-type ZIP1 sequences was looped out on 5-FOA and a clone with a single copy of ZIP1-FRB was selected for further analysis.
Cells were grown in liquid YPD culture at 23°C for 24 h and diluted at A600 0.3 into presporulation media (BYTA; 50 mM sodium phthalate-buffered, 1% yeast extract, 2% tryptone and 1% acetate). The cells were grown in BYTA for 16 h at 30°C, washed twice in water and resuspended in sporulation media (0.3% potassium acetate) at A600 2.0 to induce meiosis at 30°C. FACS analysis was used for all experiments to assay duplication of the genome and confirm synchronous meiotic initiation. Experiments to measure sporulation efficiency and spore viability were set up as synchronous meiosis as above and kept at 30°C in liquid sporulation media for 24 h.
The anchor away technique was used to conditionally deplete proteins from the nucleus upon addition of rapamycin [37]. Rapamycin was added at a final concentration of 1 μM to the meiotic cultures at either meiotic induction (T = 0 h) or during pachynema (T = 6 h) except for depletion of Spo11-FRB or Mer2-FRB, where 2 μM rapamycin was added to the cells. mek1-as1 [60] was conditionally inactivated by addition of the ATP analog, 1-NA-PP1 (Cayman Chemicals), at a final concentration of 2 μM, to the meiotic cultures during pachynema (T = 6 h).
1-D gel analysis was performed as described in [75]; 2-D gel analysis of the dHJs was performed as described in [55]. Briefly, 15 mL samples were collected for the different time points and treated with 0.1% sodium azide. The cells were resuspended in 1 mg/mL Trioxsalen (Sigma) and the DNA was UV-crosslinked as described [35,76]. DNA was extracted and digested with the appropriate enzyme and then separated by two-dimensional gel electrophoresis. The DNA was transferred onto a ZetaProbe membrane (Biorad) by capillary transfer and detected by Southern hybridization. Probes for detection of dHJs at the HIS4-LEU2 DSB locus are described [55]. A probe to assay the YCR047c locus is described [50]. Probes for GAT1 and ERG1 loci were amplified from genomic DNA with primers- 5′-caataagcaggtggagttgctgcg-3′, 5′-aaagatccaaagcccaccagattg-3′ and 5′-ggcagcaacatatctcaaggcc-3′ and 5′-tcaatgtagcctgagattgtggcg-3′ respectively. Primer pairs 5′ -attgtgcctgtaaccgaactgc-3′ and 5′ -agtggacgtagaaagaggagc-3′, 5′ -ttcctcgttcgtgacactactc-3′ and 5′ -tagctgccaaacccattctgc-3′ were used to generate the probes for YIL094c and YGR279c DSB hotspots, respectively. 32P-dCTP was incorporated into the probe using a Prime-It random labeling kit (Agilent). The Southern hybridization blot was exposed on a Fuji imaging screen and detected using a Typhoon FLA 9000 (GE). Hybridization signal was quantified using ImageJ software (http://imagej.nih.gov/ij/).
Antibodies against phosphorylated Hop1 peptides (KLH-conjugated peptides: [H]- CKKLGNLLNS-pS-QASIQP -[NH2] and [H]- CKKQASIQP-pT-QFVSNNP -[NH2]) were raised in rabbits by Covance. The serum was affinity purified with the respective phospho-peptide, followed by adsorption against the unphosphorylated peptide using a SulfoLink Immobilization kit (Thermo Fisher Scientific). Affinity-purified pT318-Hop1 antibody was used at 1:100 for western analysis and 1:50 for immunofluorescence. The anti-Pch2 antibody was raised against the recombinant N-terminal 300 amino acids purified from Escherichia coli. An open-reading frame of the truncated Pch2 was PCR-amplified and inserted into the pET15b plasmid (Novagen), in which the N-terminus of the PCH2 gene was tagged with 6x-Histidine. His-Pch2 protein was affinity-purified using a nickel resin as described by the manufacturers and used for immunization (MBL Co. Ltd). Rabbit anti-Hop1 antibody (kindly provided by N. Hollingsworth) was used at 1:10,000 for western analysis or 1:500 for immunofluorescence, rabbit anti-phospho-H3T11 (Millipore) and anti-Rfa2 antibodies (kindly provided by S. Brill) were used at 1:100 for immunofluorescence. Goat anti-Zip1 (Santa Cruz, SC-48716) was used at 1:200, goat anti-Zip1 (Santa Cruz, SC-15632) was used at 1:500, rabbit anti-Rad51 (Santa Cruz, SC-33626) was first pre-absorbed to rad51Δ meiotic spheroplasts and then used at 1:200, and rat anti-HA (Roche-11867431001) was used at 1:200. Secondary fluorescent-conjugated antibodies were obtained from Jackson Laboratory and were used for immunofluorescence after pre-absorption to yeast spheroplasts. HRP-conjugated secondary antibodies from Pierce were used for western analysis.
Meiotic cells were collected at various time points, treated with 200 mM Tris pH7.5/20 mM DTT for 2 min at room temperature and then spheroplasted in 2% potassium acetate/ 1 M Sorbitol/ 0.13 μg/μL zymolyase T100 at 30°C. The spheroplasts were rinsed and resuspended in ice-cold 0.1 M MES pH6.4/ 1 mM EDTA/ 0.5 mM MgCl2/ 1 M Sorbitol. Two volumes of fixative (3% para-formaldehyde/ 3.4% sucrose) were added to the cells on a clean glass slide (soaked in ethanol and air-dried) followed by four volumes of 1% lipsol. The slide was tilted to mix the contents. Four additional volumes of the fixative were added to the slide and the samples were spread with a clean glass rod. After spreading was completed, slides were rinsed in 0.4% Photoflo (Kodak), dried overnight and stored at -80°C.
Images were collected on a Deltavision Elite imaging system (GE) equipped with an Olympus 100X lens/1.40 NA UPLSAPO PSF oil immersion lens and an InsightSSI Solid State Illumination module. Images were captured using an Evolve 512 EMCCD camera in the conventional mode and analyzed using softWoRx 5.0 software. Structured illumination microscopy was carried out on an OMX Blaze 3D-SIM super-resolution microscope equipped with a 6-line SSI Solid State Illumination module, 100X lens/1.40 NA UPLSAPO PSF oil immersion lens (Olympus) and three EVOLVE EMCCD cameras (housed at the Bio-imaging Resource Center, Rockefeller University). Super-resolution images for Fig 3 were collected on a Deltavision OMX V4 equipped with a 60X/1.42NA PLAPON oil immersion lens (Olympus). 100mW solid-state lasers were used along with three PCO sCMOS cameras for detection. Structured illumination reconstructions were carried out in softWoRx 6.1. Scatterplots were generated using the Graphpad program in Prism and statistical significance was assessed using a Mann-Whitney test.
|
10.1371/journal.pntd.0004320 | Costs of Three Wolbachia Infections on the Survival of Aedes aegypti Larvae under Starvation Conditions | The mosquito Aedes aegypti, the principal vector of dengue virus, has recently been infected experimentally with Wolbachia: intracellular bacteria that possess potential as dengue biological control agents. Wolbachia depend on their hosts for nutrients they are unable to synthesize themselves. Consequently, competition between Wolbachia and their host for resources could reduce host fitness under the competitive conditions commonly experienced by larvae of Ae. aegypti in the field, hampering the invasion of Wolbachia into natural mosquito populations. We assess the survival and development of Ae. aegypti larvae under starvation conditions when infected with each of three experimentally-generated Wolbachia strains: wMel, wMelPop and wAlbB, and compare their fitness to wild-type uninfected larvae. We find that all three Wolbachia infections reduce the survival of larvae relative to those that are uninfected, and the severity of the effect is concordant with previously characterized fitness costs to other life stages. We also investigate the ability of larvae to recover from extended food deprivation and find no effect of Wolbachia on this trait. Aedes aegypti larvae of all infection types were able to resume their development after one month of no food, pupate rapidly, emerge at a large size, and exhibit complete cytoplasmic incompatibility and maternal transmission. A lowered ability of Wolbachia-infected larvae to survive under starvation conditions will increase the threshold infection frequency required for Wolbachia to establish in highly competitive natural Ae. aegypti populations and will also reduce the speed of invasion. This study also provides insights into survival strategies of larvae when developing in stressful environments.
| Dengue is currently the most important arboviral disease in the world. With no effective treatment or commercial vaccine available, strategies to control dengue focus on its mosquito vectors, primarily Aedes aegypti. A recent effort to reduce the burden of dengue aims to replace native Ae. aegypti with those refractory to the virus. This is achieved by infecting mosquitoes with Wolbachia, bacteria which can invade insect populations by exploiting host reproduction. Some strains of Wolbachia have harmful effects on the mosquito host which can inhibit its ability to spread. While these costs have been characterized comprehensively in the laboratory, we must also consider any impacts when mosquitoes experience stresses that commonly occur in nature. For instance, Ae. aegypti larvae often develop in highly-occupied habitats where food is scarce. We investigated the effects of Wolbachia on mosquito larvae when they develop under extremely nutrient-limited conditions and found costs to survival for all strains. This will translate to a reduced ability of Wolbachia-infected mosquitoes to replace native populations in competitive habitats.
| Dengue fever is an increasing threat to global health. An estimated 50 to 390 million new cases of dengue occur annually, with 2.5 billion people living in areas at risk of infection [1,2]. At present, dengue lacks an effective treatment or vaccine that protects against all serotypes of the virus. Thus, strategies to reduce infection incidence must rely on the control of its mosquito vector, principally Aedes aegypti [3,4]. While permanent eradication is unlikely to be achieved, many emerging genetic and biological approaches aim to reduce mosquito vectorial capacity [5,6].
A promising new approach to dengue control utilizes the obligate intracellular bacterium, Wolbachia. Wolbachia are maternally inherited [7] and usually manipulate the reproduction of their hosts to enhance their own transmission [8]. The most common manipulation induced by Wolbachia is cytoplasmic incompatibility; a mechanism where embryonic lethality occurs when an infected male mates with a female that is not infected with Wolbachia, providing infected females with a relative reproductive advantage [9,10]. Many Wolbachia infections also provide protection to their host against pathogens, including RNA viruses [11–13]. These traits have enabled Wolbachia to be implemented in strategies to both suppress [14,15] and replace [16–18] insect populations.
While Ae. aegypti does not harbour a natural Wolbachia infection [19,20], three infections have been stably introduced into the vector: the wMelPop and wMel strains originating from Drosophila melanogaster [21,22] and wAlbB from the mosquito Aedes albopictus [23]. All three infections are transmitted vertically at high rates and exhibit complete cytoplasmic incompatibility [21–23], and these effects have remained stable after many years in the novel host [24–26]. Crucially, they also suppress the replication and transmission of dengue virus in Ae. aegypti [22,27,28], giving them potential to reduce dengue incidence in transformed populations. Establishment of Wolbachia in a field population is facilitated largely by maternal transmission and cytoplasmic incompatibility [16,29,30]. However, because Wolbachia-infected mosquitoes must survive and reproduce in competition with the native inhabitants, lower relative fitness of infected mosquitoes can hamper the invasibility of Wolbachia [31–33].
The experimental Wolbachia infections established in Ae. aegypti vary considerably in their effects on mosquito life-history traits. The wMel infection is relatively benign and has invaded both caged [22] and field [18] populations. wMel remains at a high frequency in mosquitoes collected from the release sites, three years after releases of wMel ceased in two suburbs of Cairns, Australia[24]. Conversely, the wMelPop infection tends to overreplicate in host cells, leading to rapid tissue degeneration and early death [34–36]. It exacts a high fitness cost on Ae. aegypti; wMelPop shortens adult lifespan [21,37], while fecundity [38], blood feeding success [39,40] and quiescent egg viability [37,38,41] deteriorate rapidly with age. wMelPop also modifies behaviour and metabolism [42], reduces the response of larvae to light stimulation [43], delays larval development, and decreases viability and adult size when reared under crowded conditions [44]. The wAlbB infection has intermediate fitness costs, likely due to its moderate density in host tissues that lies between that of wMel and wMelPop [26].
While each of these infections can invade caged populations of Ae. aegypti [22,23,26], the mosquitoes were not exposed to many of the selective pressures that exist in the field [6]. Suitable habitats for immature development in the field are limited; as a consequence, larvae are often subjected to competition for space and nutrition [45–48]. Though Wolbachia infection has no clear effect on Ae. aegypti larval development in the absence of stress [22,26,37,38], some costs emerge when larvae are crowded [44]. Many fitness costs of Wolbachia in Ae. aegypti also tend to become clearer with age in both adults and eggs [26,37,39]. As larval development times can reach several weeks, or even months in the field [49] and often experience periods of food limitation [47,48], deleterious effects of Wolbachia on larvae undetected in laboratory studies could emerge when development times are prolonged, impacting Wolbachia’s invasive potential. This could explain a lack of invasion success by wMelPop in natural populations despite multiple attempts to establish the infection in the field [50].
Aedes aegypti larvae are adapted to nutrient poor-habitats as food limitation is a major regulator of their population size [47,51]. Larvae decrease their rate of development in response to food scarcity, delaying metamorphosis until reaching a critical threshold of nutritional reserves [52–55], and larvae can resist starvation for several weeks at a time [51,56–58]. This is achieved largely by expending their accumulated reserves [59–61], though larvae also scavenge on dead conspecifics [62,63] and may even prey on younger larvae [64] to increase their chance of survival. Wolbachia depend on their hosts for a wide range of resources they cannot synthesize themselves [65–68]. Since Wolbachia increase the activity and metabolic rate of Ae. aegypti in adults, at least for the wMelPop infection [42], we hypothesize that Wolbachia may also increase the rate at which energy reserves are depleted in larvae without food. Aedes aegypti breeding containers typically have low productivity and high food intermittency because leaf litter, animal detritus and the microorganisms that break them down are the primary source of nutrition [62,69,70]. Thus, the ability to survive periods of limited food is a critical aspect of larval fitness [47,51]. In the field, competition between Wolbachia and Ae. aegypti for resources could substantially reduce the survival of larvae, limiting the potential for Wolbachia to invade and persist in natural populations.
In this study we investigate the effects of wMel, wAlbB and wMelPop infection on the ability of Ae. aegypti larvae to survive and develop under extreme nutrient stress. We compare the survival and development of Wolbachia-infected and uninfected larvae under starvation conditions when held in groups, when infected and uninfected larvae are together in the same container, or when isolated, and test their ability to recover when an influx of resources is provided. We also examine the ability of Wolbachia to express their reproductive effects when Ae. aegypti larvae are held under starvation conditions for extended periods. We then consider the likely impact of any fitness costs imposed by Wolbachia on the potential for these infections to invade highly competitive populations.
Aedes aegypti mosquitoes were sourced from Cairns, Queensland and maintained under laboratory conditions for at least two generations before use in experiments. Wolbachia-infected lines were generated by crossing male uninfected Cairns mosquitoes to laboratory-reared female mosquitoes infected with wMel [22], wAlbB [23] or wMelPop [21] to maintain a similar genetic background (>98%) between colonies. Mosquitoes were kept in the laboratory at 26°C ± 1°C and 80–90% relative humidity with a 12:12 light: dark photoperiod, and maintained according to methods described by Axford et al. [26]. Within one week of emerging, female adults were allowed to feed to repletion on the forearm of a single human volunteer. Blood feeding of female mosquitoes on human volunteers for this research has been approved by the University of Melbourne Human Ethics Committee (approval 0723847). All adult subjects provided informed written consent (no children were involved).
Larvae were reared under a common regime before initiating the food-deprivation period for all experiments. At the beginning of each experiment, wMel-infected, wMelPop-infected, wAlbB-infected and uninfected eggs were hatched synchronously in separate trays containing 3 L of RO (reverse osmosis) water, 2–3 grains of yeast and one crushed tablet of TetraMin tropical fish food (Tetra, Melle, Germany). Within three hours of hatching, cohorts of 200 1st instar larvae were transferred to plastic trays filled with 700 mL of RO water and fed TetraMin ad libitum for 72 hours. This rearing environment was chosen as development times do not differ significantly between Wolbachia-infected and uninfected larvae with abundant nutrition at this density. After the feeding period, larvae were pipetted into fresh trays of RO water. To remove any remaining food particles, larvae were rinsed by passing them through two additional trays of water before being added to experimental containers. All experiments used 72 hour old 3rd instar larvae of approximately the same size, and were conducted at 26°C ± 1°C and 80–90% relative humidity with a 12:12 light: dark photoperiod.
We tested the ability of Wolbachia-infected and uninfected larvae to survive starvation conditions in the absence of conspecific larvae, removing any effects of resource competition and also the ability to scavenge on dead larvae. Two independent experiments were conducted; in each, 96 larvae per infection type (see rearing regime) were added individually to wells of Costar 12-well cell culture plates (Corning, Corning, NY) filled with ~4 mL of RO water only. Plates were enclosed in stockings and held in a tray covered with a mesh lid to minimize external sources of food input, and RO water was topped up daily to counter evaporation. For both experiments, wells were monitored for mortality daily until all larvae had died. A larva was considered dead when no movement was observed after fifteen seconds of physical stimulation.
In the first experiment, plates were unmanipulated with the exception of maintaining a consistent volume of water in each well. In the second experiment, water was replaced completely twice per week to reduce the accumulation of microorganisms as a potential source of nutrition (e.g., bacteria, algae, protozoa, fungi) and waste in the water [73]. For this experiment, larvae were removed from wells and rinsed by pipetting through multiple trays of RO water, then returned to wells filled with a fresh change of water.
Two independent experiments tested the ability of Wolbachia-infected and uninfected larvae to survive starvation conditions when held in the presence of conspecific larvae. Larvae (see rearing regime) were added to circular plastic containers (9.5–11.5 cm radius, 7 cm height) with mesh lids and filled with 200 mL of RO water only (no TetraMin was provided). Mortality was scored every second day by temporarily pipetting larvae into a separate container of RO water. Numbers of dead and live larvae were counted before all larvae (including dead larvae) were returned to the original container. Water was refreshed every four days by transferring all larvae to a new container of RO water. In the first experiment, larvae were added to containers in groups of 50. Each container was replicated eight times for the uninfected, wMel, wAlbB and wMelPop strains. The experiment was terminated when all larvae had died or had reached adulthood.
During field releases, preferential mortality of Wolbachia-infected larvae in nutrient-deprived containers could release the remaining larvae from food stress, providing an advantage to uninfected larvae [71,72]. A second experiment was therefore conducted to determine whether there were differences in survival when Wolbachia-infected and uninfected larvae were held together in mixed proportions within the same container. Cohorts of larvae were added to plastic containers filled with 200 mL of RO water in the following proportions (Wolbachia-infected to uninfected): 12:36, 24:24 and 36:12. Additional cohorts of 48 Wolbachia-infected and 48 uninfected larvae were set up as controls. Treatments (mixed proportions) were replicated eight times each, while the controls (pure cohorts) were replicated four times, and the experiment was repeated for the wMel, wAlbB and wMelPop infections. Containers were monitored as per the previous experiment, with the exception that the five longest surviving larvae in each container were removed and screened for their Wolbachia infection status (see DNA extraction and Wolbachia detection). The proportion of individuals infected with Wolbachia in the longest surviving larvae was then compared with the initial proportion of larvae infected with Wolbachia in each container (see statistical analysis).
In both experiments, a few percent of larvae were able to reach the pupal and adult stages due to the availability of dead conspecific larvae as a food resource. All adults emerging throughout the two group experiments were stored in ethanol for wing length measurement and later tested for their Wolbachia infection status (see wing length measurements and DNA extraction and Wolbachia detection). Their development time and sex were also recorded.
An experiment was carried out to test the ability of Wolbachia-infected and uninfected larvae to recover from starvation conditions after providing an influx of resources. Larvae (see rearing regime) were added to RO water in groups of 50 (see survival and development of larvae held in groups under starvation conditions). Containers were then divided into two treatments; larvae were re-fed TetraMin ad libitum after either 15 or 25 days of surviving starvation conditions. These two time points were chosen based on when substantial starvation-induced mortality had occurred; approximately 25% and 10% of larvae were remaining on Days 15 and 25 respectively (S1 Fig). For each infection type and treatment, the following observations were recorded: the number of surviving larvae upon the resumption of feeding, rates of pupation and survival to the pupal stage after re-feeding, rates of adult emergence and survival to adulthood, and the body size (see wing length measurements) and sex ratio of emerging adults. Containers were replicated between six and eight times for each infection type and treatment.
We ran a series of experiments to determine if the reproductive effects caused by Wolbachia remain robust when larvae are held under starvation conditions for an extended period. To test the level of cytoplasmic incompatibility induced by Wolbachia-infected males, larvae (see rearing regime) were added to containers of RO water and their development was suspended for ~30 days by maintaining them in the absence of TetraMin. After this period larvae were again fed TetraMin ad libitum until pupation. Pupae were sexed (males are smaller than females), and male pupae pipetted into small cups of RO water and allowed to emerge in 1.5 L plastic containers with mesh sides and a stocking lid. Female pupae emerging from this treatment were set aside for an additional experiment on reproductive effects (see fecundity and maternal transmission). After confirming the sex of all males as adults, newly-emerged uninfected females that were reared under standard laboratory conditions (see colony maintenance and mosquito strains) were added to each cage and allowed to mate freely with Wolbachia-infected males. Seven Wolbachia-infected males and seven uninfected females were held in each experimental cage, and crosses were replicated eight times for the wMel, wAlbB and wMelPop infections. Cages of adults were provided access to 10% sucrose solution and water throughout the experiment. Crosses between standard laboratory-reared Wolbachia-infected males and uninfected females were set up as controls, as these crosses are known to produce no viable offspring [21–23]. Females were then blood fed and eggs were collected according to Axford et al. [26] for three gonotrophic cycles.
This experiment assessed the rate at which Wolbachia-infected females transmit the infection to their offspring when their development time is greatly extended. Food-deprived and re-fed larvae from the wMel, wAlbB and wMelPop lines (see cytoplasmic incompatibility) were sorted by sex, and 100 females per infection type were added to 12 L plastic cages and provided with 10% sucrose solution and a source of water. 100 uninfected males reared under standard laboratory conditions were then aspirated into each cage and allowed to mate freely. Females were then blood fed and isolated for oviposition according to Axford et al. [26], and their progeny reared to adulthood and stored in absolute ethanol.
Ten progeny each from 30 isolated females per infection type were tested for the presence of Wolbachia using PCR to determine maternal transmission efficiency (see DNA extraction and Wolbachia detection). A set of control crosses was also completed for each infection type where both Wolbachia-infected females and uninfected males were reared under standard laboratory conditions. Ten progeny from 15 Wolbachia-infected females were tested for each of the wMel, wAlbB and wMelPop infections. These crosses have expected maternal transmission rates of close to 100% [21–23]. All female parents from the treatments and controls were scored for their fecundity, with a sample also measured for wing length (see wing length measurements). Data from uninfected females reared under standard laboratory conditions from a concurrent experiment were included as a point of comparison.
Linear measurements of wings were taken to give an indication of body size [74,75]. The right wing was removed from each adult and fixed on a slide under a 10 mm circular coverslip (Menzel-Gläser, Braunschweig, Germany) using Hoyer’s solution (dH2O: gum arabic: chloral hydrate: glycerin in the ratio 5: 3: 20: 2) [76]. Wings were observed under a dissecting microscope fitted with a camera and measured using NIS-Elements BR (Nikon Instruments, Japan). Wing length was determined by calculating the distance from the alular notch to the intersection of the radius 3 vein and outer margin, excluding the wing fringe scales [77]. Measurements in pixels were converted to millimetres by calibration with a graticule before the start of each set of measurements. Each measurement was repeated independently so that length represented the average of two measurements. Damaged or folded wings were excluded from the analysis.
To test for the presence of Wolbachia in adult and immature mosquitoes, we carried out DNA extraction and Wolbachia detection according to methods described previously [24,26,78]. DNA from whole adults or larvae was extracted using 150 μL of 5% Chelex 100 resin (Bio-Rad Laboratories, Hercules, CA). The PCR assay was conducted using a LightCycler 480 system (Roche Applied Science, Indianapolis, IN); mosquitoes were considered positive for Wolbachia when the mRpS6 (Aedes universal) and aRpS6 (Ae. aegypti-specific) primer sets were successfully amplified in addition to the appropriate Wolbachia-specific primer set (wMel, wAlbB or wMelPop). Wolbachia-free mosquitoes tested positive for mRpS6 and aRpS6 and negative for all Wolbachia-specific primer sets.
All data were analysed using SPSS statistics version 21.0 for Windows (SPSS Inc, Chicago, IL). Survival data were investigated using Kaplan-Meier analysis; log-rank tests compared rates of mortality between lines and treatments. Wolbachia infection frequency was calculated as the proportion of individuals that tested positive for Wolbachia. For containers where both Wolbachia-infected and uninfected larvae were present, deviations from expected infection frequencies in larvae and adults were analysed using Chi-squared tests. Maternal transmission rates of Wolbachia were expressed as the proportion of infected offspring produced by infected mothers, for which 95% binomial confidence intervals were calculated. All other data were tested for normality using Shapiro-Wilk tests. Data that were not normally distributed were arcsine square-root transformed (proportional data) or square-root transformed and tested again. Normally distributed data were then analysed with one-way ANOVA and Tukey’s honest significant difference tests, while data that failed Shapiro-Wilk tests were analysed with non-parametric Kruskal-Wallis and Mann-Whitney U tests. Associations between wing length and development time were assessed with Pearson’s correlation if data were normally distributed or Spearman’s rank-order correlation where data could not be transformed for normality.
Kaplan-Meier (KM) analysis revealed a significant effect of Wolbachia infection type (KM: χ2 = 123.273, df = 3, P < 0.0001) and water-replacement regime (KM: χ2 = 678.532, df = 1, P < 0.0001) on the survival of larvae when isolated under starvation conditions. Whether water was refreshed in each well or left unmanipulated had a dramatic effect on survival, with the former (mean ± SE = 20.682 ± 0.221 days) reducing the mean survival time of larvae by half compared with unmanipulated experimental wells (40.286 ± 0.573 days, S2 Fig). An increased survival in the latter experiment is likely due to the build-up of microorganisms which are an important resource for mosquito larvae [70,73,79].
When water was not replaced, all three Wolbachia infections reduced survival; the wMel, wAlbB and wMelPop infections decreased mean survival 15.8, 28.8 and 28.7% compared with uninfected larvae (Fig 1A). All pairwise comparisons between the infection types were highly significant (KM: all χ2 > 24.087, df = 1, all P < 0.0001), with the exception that wMelPop and wAlbB did not differ significantly in their survival patterns under starvation conditions (χ2 = 0.717, df = 1, P = 0.397).
Although there was a significant effect of Wolbachia infection type in both experiments, survival differences between Wolbachia-infected and uninfected larvae were reduced markedly when water was replaced every four days (KM: χ2 = 17.939, df = 3, P = 0.0005) compared with wells that were unmanipulated (χ2 = 150.024, df = 3, P < 0.0001, Fig 1). When water was replaced, all pairwise comparisons between infection types were significant (KM: all χ2 > 4.262, df = 1, all P ≤ 0.039) except for between uninfected and wMel (KM: χ2 = 1.707, df = 1, P = 0.191), and wAlbB and wMelPop (KM: χ2 = 0.630, df = 1, P = 0.427) (Fig 1B). No pupae or adults emerged in either experiment where larvae were isolated.
Wolbachia infection type also had a substantial effect on survival when larvae were held under starvation conditions in groups of 50 (KM: χ2 = 225.821, df = 3, P < 0.0001). Uninfected larvae had the greatest mean time of survival (mean ± SE = 28.289 ± 0.532 days), with the wMel, wAlbB and wMelPop infections reducing survival times by 5.7, 15.7 and 29.5% respectively (Fig 2). All pairwise comparisons between lines were significant (KM: all χ2 > 7.411, df = 1, all P ≤ 0.006). Note that emerging adults were excluded from Kaplan-Meier analyses rather than censored because the rate and number of adults emerging differed between infection types.
Larvae from both Wolbachia-infected and uninfected lines readily consumed dead conspecifics throughout the experiment. We inferred scavenging based on observations that the number of dead larvae in each container fluctuated with mortality rather than increasing proportionally (S3 Fig). Distributions of necrophagy closely matched larval mortality, with the mean time for larval consumption occurring less than one day after the mean time of death for both Wolbachia-infected and uninfected lines (S4 Fig). Necrophagy likely contributed to increased survival time; larvae lived for longer in groups compared with larvae kept in isolation under otherwise similar conditions. While survival began to decline earlier in the group experiment, rates of mortality became considerably slower when the majority of larvae had died (S2 Fig).
Less than five percent of larvae reached pupation or adulthood during this experiment (Table 1). Wolbachia infection type had a significant effect on the total number of larvae that survived to both the pupal (one-way ANOVA: F3, 28 = 3.417, P = 0.031) and adult (F3, 28 = 5.647, P = 0.004) stages, and also affected the development times of those pupae (Kruskal-Wallis: χ2 = 31.499, df = 3, P < 0.0001) and adults (χ2 = 14.200, df = 3, P = 0.003). Despite uninfected larvae having greater survival times under starvation conditions (Fig 2), they developed more slowly and pupated less often than Wolbachia-infected larvae, with the wMelPop infection displaying the greatest proportion of larvae reaching adulthood and the most rapid development on average (Table 1, S5 Fig). This observation is likely due to an earlier availability and greater abundance of conspecific carcasses as a source of nutrition in containers with wMelPop-infected larvae.
A second experiment was conducted where Wolbachia-infected and uninfected larvae were held together in the same container under starvation conditions. Control containers, where 48 larvae from each infection type were held separately, had a shorter starved survival period than in the previous experiment despite nearly identical methods, though the relative performance of each infection type was similar (S2 Fig). In each treatment container, the five longest-lived larvae were screened for their infection status to test for differential survival between infected and uninfected larvae when held together at different frequencies. The wAlbB and wMelPop infections were significantly underrepresented in the surviving larvae for all treatments, while for wMel there were no significant deviations from any starting ratio (Table 2).
Less than two percent of larvae from this experiment emerged as adults. Expected ratios of Wolbachia-infected and uninfected adults emerging were based on the initial proportion of larvae in each container. We found no significant deviations from expected proportions of adults for all treatments (Chi-squared test: all χ2 < 3.267, df = 1, all P > 0.071), except for the wMelPop infection which was significantly underrepresented when larvae were held in the ratio 36:12 (wMelPop: uninfected) (Chi-squared test: χ2 = 24.2, df = 1, P < 0.0001).
All adults that emerged from larvae held in groups were measured for wing length to test for effects on body size. Due to low numbers of adults, data were pooled across both experiments as they did not differ significantly (Student’s t test: P = 0.795). Wing length was not associated with development time for either males (Spearman’s rank-order correlation: ρ = 0.071, P = 0.455, n = 56) or females (ρ = -0.009, P = 0.924, n = 58). As expected, there was a significant effect of sex on wing length (one-way ANOVA: F1,106 = 285.910, P < 0.0001), where males (mean ± SE = 1.659 ± 0.009 mm) were considerably smaller than females (1.973 ± 0.015 mm). However, we found no effect of Wolbachia infection type (one-way ANOVA: F3,106 = 0.360, P = 0.782); wings of mosquitoes with any infection type were approximately the same size (Table 3).
25.5% and 12.3% of larvae across all infection types survived after 15 and 25 days of exposure to starvation conditions respectively. Wolbachia infection type had a significant effect on the number of larvae surviving after both 15 (one-way ANOVA: F3,56 = 4.152, P = 0.010) and 25 days (F3,26 = 4.114, P = 0.016). The wMelPop infection had the lowest survival at both time points (S1 Fig), consistent with other experiments (Figs 1B and 2).
Recovery from food deprivation was assessed by scoring the proportion of surviving larvae that pupated and reached adulthood upon resuming feeding. The majority of surviving larvae were able to recover, though larval and pupal mortality occurred across both treatments for all infection types (Fig 3). We found a significant effect of treatment (day of re-feeding) (one-way ANOVA: F1,52 = 5.576, P = 0.022), but not Wolbachia infection type (F3,52 = 1.461, P = 0.236), on the proportion of surviving larvae that reached adulthood. Surviving larvae that were deprived of food for 25 days were less likely to reach adulthood than larvae deprived for 15 days, with the percentage surviving of larvae that died after re-feeding averaging 10.4% and 22.9% respectively. This is, in part, due to an increase in pupal mortality at the later time point (2.1% for Day 15, 9.0% for Day 25, Student’s t test: P = 0.042, Fig 3). The proportion of surviving larvae that reached adulthood was less for wMelPop than for other infection types, though this difference was not significant (Fig 3). Larvae that reached pupation before re-feeding (33.3% of wMelPop-infected larvae and 3.3% of wMel-infected larvae) were counted as survivors. However, these individuals were excluded from development time and wing length analyses (see below) as they pupated before food was provided again ad libitum, and were similar in size to adults emerging from larvae held in groups under starvation conditions (Table 3).
The number of days taken for larvae to reach pupation after re-feeding was significantly affected by infection type (one-way ANOVA: F3, 488 = 5.377, P = 0.001) but not treatment (day of re-feeding) (F1, 488 = 2.128, P = 0.145), though infection types within treatments did not differ significantly from each other (Table 4). Development times of both male and female adults were unaffected by infection type and treatment (one-way ANOVA: all P > 0.053). Female wing length was significantly affected by treatment (one-way ANOVA: F1, 251 = 6.696, P = 0.010) but not infection type (F3, 251 = 1.432, P = 0.234). Females re-fed after 25 days of food deprivation were smaller than those fed after 15 days for all infection types, though no pairwise comparisons were significant (Table 4). Conversely, male wing length was unaffected by both infection type (F3, 194 = 0.844, P = 0.471) and treatment (F1, 194 = 0.032, P = 0.859). We found no correlation between development time and wing length for both males and females for each treatment (Pearson correlation: all P > 0.175).
Males deprived of food for 30 days as larvae and then re-fed were tested for their ability to induce cytoplasmic incompatibility when crossed to uninfected females. All food-deprived and re-fed Wolbachia-infected males exhibited complete cytoplasmic incompatibility, with no viable offspring produced across three gonotrophic cycles (Table 5). Control crosses using standard laboratory-reared adults were also completely sterile, with the exception that a low proportion of eggs hatched in the wMelPop control cross due to contamination with uninfected males (Table 5).
We also tested maternal transmission rates of Wolbachia when infected females were held under starvation conditions for 30 days as larvae and then re-fed. The wMel, wAlbB and wMelPop infections were transmitted with perfect fidelity by both standard laboratory-reared females (All infection types: maternal transmission rate = 1, lower 95% binomial confidence interval = 0.976), and females that were food-deprived then re-fed (All infection types: maternal transmission rate = 1, lower 95% binomial confidence interval = 0.988).
Female parents were also measured for their fecundity and wing length. Both Wolbachia infection type (one-way ANOVA: F3, 227 = 33.011, P < 0.0001) and treatment (F1, 227 = 8.787, P = 0.003) had significant effects on fecundity. The food-deprivation treatment reduced the mean fecundity of wMel, wAlbB and wMelPop-infected females by approximately 5–6 eggs relative to the controls, though no pairwise comparisons were significant (Table 6). All Wolbachia-infected females had considerably reduced fecundity compared with uninfected standard laboratory-reared females, regardless of the rearing treatment (Table 6). Female wing length was also significantly affected by both Wolbachia infection type (one-way ANOVA: F3, 108 = 6.935, P = 0.0003) and treatment (F1, 108 = 8.852, P = 0.004). For all infection types, females held under starvation conditions and then re-fed were smaller than standard laboratory-reared females, though only the wAlbB comparison was significant (Table 6).
We have demonstrated that Wolbachia infection reduces the tolerance of Ae. aegypti larvae to starvation conditions. Because Ae. aegypti larvae survive nutrient-poor conditions primarily by expending their own accumulated energy reserves [59,60], we suspect that Wolbachia reduce survival by increasing the rate at which these reserves are depleted. Wolbachia do not appear to affect the rate at which larvae accumulate reserves because development times are unaffected by infection when larvae are well-fed [26,37,38]. However, when food is limited, Wolbachia may increase the drain on host reserves due to various nutritional requirements [65–68]. Indeed, Wolbachia increase the metabolism of Ae. aegypti adults, at least for the wMelPop infection [42], though this remains to be tested in larvae.
All three infections negatively affected the survival patterns of nutrient-deprived larvae but differed in their severity; wMelPop was highly costly to survival across all experiments, wMel either had a slightly deleterious or no significant effect relative to uninfected larvae, and wAlbB had an intermediate effect. These relative costs are consistent with their effects on mosquito adults and eggs; wMelPop drastically reduces adult lifespan and quiescent egg viability [21,37,38], wMel has relatively minor costs or no detectable effect [22,24], and wAlbB has an intermediate cost to these traits [26]. Here, we demonstrate that infections with higher virulence in these life stages also have greater costs to the survival of larvae under starvation conditions. The differences between Wolbachia infections in terms of their deleterious effects are likely to be attributed to their density in mosquito tissues [80]. High bacterial densities and broad tissue tropisms in host cells are often implicated in increasing fitness costs imposed by Wolbachia infection, both in Ae. aegypti [22,26] and other insects [81–84].
We found that as the survival period of larvae increased, the deleterious effects of Wolbachia became clearer. In adults and eggs of Ae. aegypti, the fitness costs of Wolbachia are also enhanced with age; wMelPop has relatively little cost to the reproductive success of young females, but fecundity [38] and rates of successful probing [39,40] decline severely with subsequent gonotrophic cycles. Additionally, the wAlbB and wMelPop infections impose increased costs on the viability of quiescent eggs over time [26,37]. If these age effects also occur in larvae as suggested by our results, virulent Wolbachia infections could have difficulty invading populations where resources are scarce and thus development times are lengthened.
Adults emerging from starvation conditions were small in size, even in comparison with those produced through extreme crowding or nutrient limitation (e.g. [75,85,86]). Adult sizes were at the lowest end of natural variation found in Australian field populations of Ae. aegypti, from where these mosquitoes were sourced [87,88]. Adult body size reflects the feeding history of larvae after reaching a critical weight [54]; therefore adults emerging from starvation conditions likely obtained only the minimum nutritional reserves required for pupation. In contrast, larvae that were deprived of food for extended durations and then fed ad libitum emerged nearly as large as mosquitoes fed ad libitum throughout development, suggesting that they were able to attain a close approximation of their maximum weight despite the long interruption to feeding [52,53,55,58].
We found that Ae. aegypti larvae, regardless of Wolbachia infection type, recover well from long periods of nutrient deprivation. While the ability of larvae to resume their development has been reported previously [54,56,59], we show that larvae exhibit low mortality, pupate rapidly and emerge at a large size when fed again after being deprived of food for as long as three weeks. In addition, infected males deprived of food as larvae for one month exhibited complete cytoplasmic incompatibility and females transmitted Wolbachia to their offspring with perfect fidelity despite a greatly extended development time. Maternal transmission rates of Wolbachia also remain high when eggs are held in a quiescent state for several weeks [41]. In insects, the maternal transmission efficiency of Wolbachia [35,89–91] and the strength of cytoplasmic incompatibility [36,92–95] are known to be affected by bacterial density. Because environmental factors such as temperature [96–98] and nutrition [68,90,99,100] modulate Wolbachia density, extreme stress in the field could lead to changes in host effects derived from Wolbachia. However, the wMel infection of Ae. aegypti established in Australian field populations has so far remained stable in terms of its reproductive effects, fitness costs and dengue blockage [24,101].
We acknowledge some limitations of our laboratory study that should be addressed in future experiments. We were somewhat limited in our ability to discern any effects of Wolbachia on larval development time and survival to adulthood when held under starvation conditions, due to low pupation rates. Future experiments testing these traits specifically should use larger cohorts with greater replication. Furthermore, we demonstrated the fitness costs of Wolbachia under rather arbitrary and specific scenarios. Nutrient input in the field is dynamic [102], but in this study larvae were fed for a single time period before either being deprived of food completely or re-fed at a later point. Breeding containers in the field are often populated by multiple cohorts [45,103,104], and Suh and Dobson [43] recently reported differential survival of Wolbachia-infected and uninfected 1st instar Ae. aegypti larvae in the presence of later instars. Because predatory behaviour is more likely to occur under nutrient-poor conditions [64], future experiments on survival under starvation conditions should also test interactions between larvae of mixed age classes. Our experiments also were conducted over multiple generations, and while all infection types were outcrossed to an uninfected colony, the number of generations spent in the laboratory varied between experiments. Laboratory adaptation can have substantial effects on fitness [6,105], which could explain why larvae in some experiments had reduced survival under similar conditions (see S2 Fig)
Nevertheless, our study demonstrates consistent deleterious effects of Wolbachia on the survival of Ae. aegypti larvae under starvation conditions. To predict the impact on the invasion dynamics of Wolbachia in highly resource-limited habitats, we estimate changes to the unstable equilibrium frequency, denoted p^, when this cost to larval viability is considered. For Wolbachia to reach fixation in a population its frequency must reach or exceed p^; larger p^ values thus decrease the likelihood and speed of invasion, and will additionally reduce the potential for spatial spread once established in a population [31,106,107].
Based on the mean survival time of larvae under starvation conditions (averaged across all experiments where larvae were held in groups), we estimate the relative fitness of the wMel, wAlbB and wMelPop infections to be 92.3, 81.3 and 68.5% that of uninfected respectively. We detected no significant costs for other traits, thus only the cost to survival patterns under starvation conditions is considered. Following equation 17b of Turelli [108], this produces a p^ of 0.08, 0.19 and 0.32 for wMel, wAlbB and wMelPop respectively in the absence of any other fitness costs, assuming complete cytoplasmic incompatibility and no maternal transmission leakage as indicated by our results. Previous laboratory studies have estimated the fitness costs of the wMel, wAlbB and wMelPop infections to be approximately ~24% [22], ~15% [23,26] and ~43% [37,108] respectively. Using these estimates, p^ increases to 0.30, 0.31 and 0.61 for wMel, wAlbB and wMelPop respectively when both the costs to larval viability under starvation conditions and deleterious effects on other life stages are considered.
In a more extreme scenario, where larvae are deprived of food for 25 days before being provided access to food ad libitum, the invasive potential of Wolbachia decreases further. Assuming Wolbachia-infected larvae are equally as capable of recovering from food deprivation as suggested by our results, the relative fitness of the wMel, wAlbB and wMelPop infections decrease to 90.6, 73.9 and 42.5% that of uninfected respectively. This corresponds to increases of p^ to 0.31, 0.37 and 0.75 when taking into account other fitness costs. The deleterious effects demonstrated here could in part explain why wMelPop was able to establish in semi-field cages [22,41] but has had great difficulty invading wild mosquito populations, both in Australia and Vietnam [50]. In semi-field cages, any costs of Wolbachia infection to larval viability under nutrient stress were likely to be masked by the fact that larvae were relatively well-fed. On the other hand, survival of larvae under starvation conditions was likely to be a critical fitness component in the field releases. The deleterious effects of Wolbachia demonstrated here will, therefore, have an impact on the potential for these infections to invade natural mosquito populations where competition for resources is the major limiting factor of population size, particularly for wMelPop.
|
10.1371/journal.pcbi.1002164 | Network-Based Prediction and Analysis of HIV Dependency Factors | HIV Dependency Factors (HDFs) are a class of human proteins that are essential for HIV replication, but are not lethal to the host cell when silenced. Three previous genome-wide RNAi experiments identified HDF sets with little overlap. We combine data from these three studies with a human protein interaction network to predict new HDFs, using an intuitive algorithm called SinkSource and four other algorithms published in the literature. Our algorithm achieves high precision and recall upon cross validation, as do the other methods. A number of HDFs that we predict are known to interact with HIV proteins. They belong to multiple protein complexes and biological processes that are known to be manipulated by HIV. We also demonstrate that many predicted HDF genes show significantly different programs of expression in early response to SIV infection in two non-human primate species that differ in AIDS progression. Our results suggest that many HDFs are yet to be discovered and that they have potential value as prognostic markers to determine pathological outcome and the likelihood of AIDS development. More generally, if multiple genome-wide gene-level studies have been performed at independent labs to study the same biological system or phenomenon, our methodology is applicable to interpret these studies simultaneously in the context of molecular interaction networks and to ask if they reinforce or contradict each other.
| Medicines to cure infectious diseases usually target proteins in the pathogens. Since pathogens have short life cycles, the targeted proteins can rapidly evolve and make the medicines ineffective, especially in viruses such as HIV. However, since viruses have very small genomes, they must exploit the cellular machinery of the host to propagate. Therefore, disrupting the activity of selected host proteins may impede viruses. Three recent experiments have discovered hundreds of such proteins in human cells that HIV depends upon. Surprisingly, these three sets have very little overlap. In this work, we demonstrate that this discrepancy can be explained by considering physical interactions between the human proteins in these studies. Moreover, we exploit these interactions to predict new dependency factors for HIV. Our predictions show very significant overlaps with human proteins that are known to interact with HIV proteins and with human cellular processes that are known to be subverted by the virus. Most importantly, we show that proteins predicted by us may play a prominent role in affecting HIV-related disease progression in lymph nodes. Therefore, our predictions constitute a powerful resource for experimentalists who desire to discover new human proteins that can control the spread of HIV.
| Conventional high-throughput antiviral discovery often targets the activities of specific viral enzymes. These approaches have been ineffective in stemming the emergence of drug-resistant variants, especially in the face of rapidly-mutating RNA viruses. One powerful yet under-explored avenue is the evolutionarily resilient nature of host proteins. Viral pathogens are parasitic in nature owing to their limited genomes. In principle, disruptions to host-pathogen interactions would impede the propagation of pathogens. The recent identification of HIV dependency factors (HDFs) or “host cellular factors” highlights this point [1], [2], [3]. HDFs represent a class of host proteins that are essential for HIV replication, but are not lethal to the host cell when silenced. By measuring levels of viral protein expression or production of infectious viral particles in human cells after knocking down individual genes using RNA interference (RNAi), these studies search for human genes that are required by HIV. Such studies have also been performed for other viruses and bacteria pathogenic to humans [4], [5], [6], [7], [8]. HDFs not only provide critical insights into HIV pathogenesis by helping to identify potential mechanisms for manipulation of host pathways, but may also have the potential to serve as therapeutic targets.
The studies conducted by Brass et al. [1], Konig et al. [2], and Zhou et al. [3] identified 275, 296, and 375 HDFs, respectively. The Brass and Konig sets had an overlap of 13 proteins, the Konig and Zhou sets had an overlap of 10 proteins, while the Brass and Zhou sets had 17 common proteins. One potential reason for the small overlap is that the experiments were performed in different cell lines; the Brass and Zhou studies used HeLa cells while the Konig study used HEK293T cells. The small overlaps could also arise from differences in the HIV strains used, the assay time post-infection, the procedures used to measure infection, and other approaches used to analyze experimental data [9], [10]. Although the three siRNA screens showed little overlap at the level of individual genes, Bushman et al. [10] found that similar Gene Ontology (GO) terms were enriched in the three gene sets. Interestingly, Konig et al. noted that 64 HDFs reported by Brass et al. directly interacted (via a physical interaction between proteins) with a confirmed HDF in their study. In support of this observation, Bushman et al. constructed a network of protein-protein interactions among HIV proteins and 2,410 host cell genes identified in the three siRNA screens and six other HIV-related studies. Dense clusters within this network contained multiple proteins identified in two or more siRNA screens and were enriched in processes and complexes such as the proteasome and the mediator complex, which are known to be associated with HIV replication. In a related study, Wuchty et al. [11] found that HDFs and human proteins that interact with HIV also appeared in dense clusters. The proposed that such protein groups may serve as “infection gateways” that enable the virus to control specific human cellular processes. They also noted that transcription factors and protein kinases mediated indirect interactions between HDFs and viral proteins. Macpherson et al. [12] performed a complementary analysis. Starting from known human-HIV protein-protein interactions (PPIs), they used biclustering to identify sets of human proteins that participated in the same types of interactions with HIV proteins. They evaluated the functional information in each bicluster and further grouped the human proteins in biclusters into higher-level subsystems. By overlapping these subsystems with HDFs, they characterized host systems that were perturbed by HIV-1 infection and identified patterns of human-HIV PPIs that correlated to these perturbations.
We took these analyses as our starting point, since they suggested that the three siRNA genomic screens may be incomplete and that there are potentially many HDFs yet to be discovered. In particular, we hypothesized that the proximity of experimentally-detected HDFs within the human protein-protein interaction (PPI) network can be fruitfully exploited by machine-learning algorithms to predict novel HDFs. We treated the computational problem of predicting HDFs as an instance of semi-supervised learning: we combined HDFs identified by Brass et al., Konig et al., or Zhou et al. (positive examples formed by the union of these three sets) with non-HDFs (negative examples, see “Data and Algorithms” for details) in the context of a human PPI network. The other proteins in this network constituted the unknown examples. We used an intuitive graph-theoretic approach that we call SinkSource and other algorithms published in the literature [13], [14], [15] to predict undiscovered HDFs. Our results, along with those of other studies [10], [11], [12], suggest that many HDFs are yet to be discovered and that they have potential value as prognostic markers to determine pathological outcome and the likelihood of AIDS development.
The SinkSource algorithm can be understood via the following physical analogy. We consider the PPI network to be a flow network. Here, each edge is a pipe and its weight denotes the amount of fluid that can flow through the pipe per unit time. Each node has a reservoir of fluid. We maintain the level of the reservoir at each HDF at 1 unit and at each non-HDF at 0 units. We let fluid flow through this network. At equilibrium (when the amount of fluid flowing into each node is equal to the amount flowing out), the reservoir height at each node denotes our confidence that the node is an HDF. Our approach is reminiscent of the FunctionalFlow algorithm [14] developed for predicting gene functions, with one crucial difference. The FunctionalFlow algorithm does not use negative examples, permitting the reservoir level at a node to increase without bound. Hence, the algorithm stops after a user-specified number of phases. In contrast, our algorithm will converge to a unique solution.
We applied seven prediction algorithms to the HDF data in the context of a human PPI network integrated from seven public databases [16], (see “Data and Algorithms”). The algorithms were the SinkSource algorithm; a variant called SinkSource+ that does not need negative examples; the commonly-used guilt-by-association approach, both with and without negative examples (called Local and Local+ in this work); a method based on Hopfield networks [13]; the FunctionalFlow algorithm [14]; and another flow-based approach called PRINCE [15]. Guilt-by-association, Hopfield networks, and FunctionalFlow have been proposed to address the problem of gene function prediction. PRINCE is an approach to prioritize disease-related genes; we selected PRINCE since it outperformed many other methods for predicting disease related genes, including cluster and neighborhood based algorithms. We applied the algorithms to four sets of positive examples: the HDFs in the Brass et al. study (B), the HDFs in the Konig et al. study (K), the HDFs in the Zhou et al. study (Z), and the union of these three sets (BKZ). We restricted these sets to those proteins that participated in at least one interaction in the human PPI network. We used an unweighted version of the network for all results below.
Figure 1 displays the results of two-fold cross validation for the six algorithms tested on four datasets. Two-fold cross validation involves splitting the positive and negative examples into two halves, and using each half to make predictions for the genes in the other half. We used two-fold cross validation since we felt it better mimics our state of knowledge of HDFs than the more commonly used five-fold or 10-fold cross validations. We averaged the results over 10 independent runs for each algorithm-dataset combination. For each algorithm, it is evident from Figure 1(a) that the area under the precision-recall curve (AUPRC) value for the BKZ dataset is larger than the values for the B, K, or Z datasets. It is also clear that these results are robust to the randomization inherent in cross validation: the largest standard deviation in the AUPRC values is 0.033 (as indicated by the error bars in Figure 1(a) and data in Table S1). Figure 1(b) displays the precision-recall curve for SinkSource on the four datasets and Figure 1(c) shows the results for SinkSource+. The results for SinkSource+ were obtained with an internal parameter λ set to a value of 1 (see “Other Algorithms” for the role played by this parameter in the SinkSource+ algorithm). In each figure, we observed that the curve for the BKZ dataset dominated the other three curves at most values of recall. This result is consistent with the expectation that the Brass, Konig, and Zhou studies did not discover all true HDFs, and that combining the three sets provides a better coverage of the true HDF universe. We also noted that the variation in precision (indicated by the error bars in Figure 1(b) and Figure 1(c)) decreases with increasing recall, suggesting that high confidence predictions are more subject to variation than low confidence predictions. Finally, Figure 1(d) compares the performance of all seven algorithms on the BKZ dataset. Three of the algorithms that do not use negative examples (Local+, SinkSource+, and Functional Flow with 1 and with 7 phases) achieved higher precision values than the other algorithms for values of recall less than 20%. However, SinkSource has the best performance for values of recall greater than 20%. PRINCE, the fourth algorithm that did not use negative examples, had uniformly lower precision than SinkSource+. Its precision was superior to that of SinkSource for values of recall less than 10%. To obtain the results for PRINCE, we used 0.8 for the value of an internal parameter α, since PRINCE achieved the highest precision values for this setting of α (see “Other Algorithms” for the role played by this parameter in the SinkSource+ algorithm). Furthermore, the precisions of the algorithms that do not use negative examples dropped considerably beyond a recall of 20% (beyond 10% in the case of PRINCE). We believe that this performance drop is caused by an undue influence of positive examples, resulting in many false positives. The performance of FunctionalFlow did not vary much with an increase in the number of phases (see Figure S1). The performance of SinkSource+ was independent of the parameter λ (see Figure S2), as was the performance of PRINCE with respect to the parameter α (see Figure S3). We also noted that the AUPRC values for the BKZ dataset were 0.67 for Local, Local+, and for FunctionalFlow with 7 phases, 0.65 for PRINCE, 0.69 for SinkSource+, 0.73 for SinkSource, and 0.74 for Hopfield. There is a difference of 11% between the AUPRCs of the worst performing algorithms (0.67) and the best performing algorithm (0.74). The results for weighted versions of the network did not substantially differ from those for the unweighted network (see Figure S4 and Table S2).
The SinkSource algorithm achieved a precision of 81% at 20% recall. The precision dropped only to 70% at a recall of 60%. The corresponding precisions for SinkSource+ were 85% and 60%. Although the Hopfield network algorithm achieved an AUPRC of 0.74, we observed that the smallest recall value attained by the algorithm was 60%, since the algorithm assigned a confidence of either 1 or −1 to a large number of predictions. We concluded that the Hopfield network algorithm was not a good choice for prioritizing predictions for further experimental analysis.
It is surprising that the very simple guilt-by-association algorithms (Local+ and FunctionalFlow with one phase) perform nearly as well as more sophisticated methods (FunctionalFlow with 7 phases, Hopfield, PRINCE, and SinkSource) that attempt to optimize predictions by taking into account constraints imposed by the entire protein interaction network. However, across 10 runs of cross validation, both Local+ and FunctionalFlow with one phase showed higher variation in precision and recall than the other algorithms (see Figure S5). Therefore, these two algorithms are likely to be more susceptible to missing or erroneous information.
Based on these results, we concluded that SinkSource+ and SinkSource were the two best algorithms for predicting HDFs. When high precision is required, SinkSource+ is superior to SinkSource. Thus, the predictions made by SinkSource+ might be the most suitable as the basis for detailed experimental studies of candidate HDFs. In the rest of the paper, we focus on the results obtained by the SinkSource+ and SinkSource algorithms.
We compared how many predictions SinkSource+ and SinkSource made at confidence values that correspond to approximately 80% precision after cross validation. SinkSource+ achieved a precision of 85% (and a recall of 20%) at a confidence of 0.5. The corresponding numbers for SinkSource were a confidence of 0.71 at a precision of 81% (and a recall of 20%). To further compare the two algorithms, we computed the overlaps in their predictions for different cutoffs on the confidence values. Specifically, we computed the k highest confidence genes predicted by SinkSource+ and the k highest-confidence genes predicted by SinkSource, and measured the Jaccard coefficient of the pair of gene sets, for different values of k in increments of 100. Figure S6 demonstrates that the overlap between the predictions of the two algorithms is at least 0.34 up to the first 2000 predictions, with peaks at around 300 and 1000 predictions. These results are consistent with the relatively low recall (20–40%) predicted for the two algorithms at this level of precision. The data suggest that approximately half of the predictions may be ranked differently by the two algorithms. Predictions made by SinkSource+ for different values of the parameter λ did not vary much in their ranking (see Figures S7 and S8).
On the basis of these comparisons, we identified a set of high confidence predictions composed of the 1000 top-ranked predictions from SinkSource+ and from SinkSource respectively. These two sets contained 606 predictions in common and comprised a total of 1394 proteins in addition to the 908 BKZ HDFs. At the confidence levels of the 1000 SinkSource and SinkSource+ predictions, the precisions with two-fold cross validation are 88% and 81% respectively, suggesting that these predictions are relatively reliable. The corresponding recalls with two-fold validation are roughly 17% and 15% respectively, suggesting that these predictions are quite conservative.
In the rest of the paper, we use the phrases “BKZ HDFs”, “SS+ predicted HDFs”, and “SS predicted HDFs” to distinguish between the HDFs identified by one or more of the three siRNA screens [1], [2], [3], the HDFs predicted by SinkSource+, and the HDFs predicted by SinkSource, respectively. We extensively evaluated the predicted HDFs by comparing them to each other and to BKZ HDFs in terms of their functional annotations, interactions with HIV proteins, clustering with the PPI network, and role in disease pathogenesis. We based these evaluations on additional datasets that we did not use for predicting HDFs. Specifically, the new datasets we used were (i) Gene Ontology (GO) annotations for human proteins, (ii) interactions between HIV and human proteins, and (iii) gene expression data from two non-human primate species following infection with SIV. Hence, the analyses described below constitute independent evaluation of the relevance of our predictions to HIV infection and disease progression.
We summarized the functional roles of predicted HDFs by asking which GO terms were enriched in the HDFs, and whether any terms were considerably enriched in predicted HDFs but not in BKZ HDFs. We used the FuncAssociate software [23] for this purpose, since it can take ordered lists of genes as input, in which case it finds and utilizes the set of top-ranked genes displaying the greatest enrichment. FuncAssociate adjusts for multiple hypotheses testing by computing an experiment-wise p-value. Note that FuncAssociate operates solely on the ranked list of genes and the GO annotations. It does not utilize a network. (See “Methods” for details.) We invoked FuncAssociate with three inputs: (a) the unordered set of BKZ HDFs, (b) the SS+ predicted HDFs, ordered by confidence, and (c) the SS predicted HDFs, also ordered by confidence. We used default values of all other parameters used by FuncAssociate. FuncAssociate reported 52 GO terms as being enriched in BKZ HDFs with an adjusted p-value of 0.05 or less and 199 GO terms as enriched in SS+ predicted HDFs. We identified three classes of terms (see Table S3). We note that FuncAssociate may report many related terms as enriched, due to the hierarchical nature of GO. Therefore, we also manually inspected the directed acyclic graph connecting the enriched terms in order to make the observations below.
The trends were similar for the HDFs predicted by SinkSource (data not shown). Therefore, we compared the FuncAssociate results for SS+ predicted HDFs and for SS predicted HDFs in a similar manner. We only considered GO terms enriched with an adjusted p-value of 0.05 or less. As shown in Table S4, 280 GO terms were enriched in both sets of predictions, 182 GO terms were enriched only in SinkSource+ predictions, and 25 GO terms were enriched only in SinkSource predictions. The 280 common terms were related to processes such as RNA splicing (GO:0008380), translation initiation (GO:0003743), and oxidative phosphorylation (GO:0003743) and complexes such as the proteasome (GO:0003743), the kinetochore (GO:0000776), and the nuclear pore (GO:0005643); we discuss their relevance to HIV when we discuss clusters in the PPI network below (See “PPI Clusters Spanned by BKZ HDFs and Predicted HDFs Are Exploited by HIV”). The 182 GO terms enriched only in SinkSource+ predictions included the Ndc80 complex and MIS12/MIND type complex (mentioned above), apoptosis (including its induction and regulation) (GO:0006915, GO:0006917, and GO:0042981), and specializations of terms enriched in both sets of predictions. Among the 25 GO terms enriched only in SinkSource predictions, there were 12 GO terms whose specializations or near neighbors (in the GO directed acyclic graph) were enriched in SinkSource+ predictions. Each of the remaining 13 GO terms enriched only in SinkSource predictions were closely related to the assembly of glycosylphosphatidylinositol (GPI) anchors (GO:0006506). Based on these results, we concluded that, for the most part, similar functions were enriched in HDFs predicted by SinkSource+ and by SinkSource.
Bushman et al. observed that each of the Brass, Konig, and Zhou HDF sets were statistically significantly enriched with human proteins that interact with HIV proteins (as reported in the NCBI HIV interaction database [25]). We hypothesized that predicted HDFs might be significantly enriched with HIV interactors. Accordingly, for each algorithm, we selected the k top ranking predictions made by that algorithm, for different values of k starting at 100 and in increments of 100, computed the overlap of each set of predictions with the human proteins that interact with HIV, estimated the statistical significance of the overlap using the one-sided version of Fisher's exact test, and adjusted the p-values to account for testing multiple hypotheses [26]. The overlap fraction for SS+ predicted HDFs peaked at 26% (79 of the top 300 predicted HDFs interact with HIV proteins, p-value 2.1×10−7), better than the BKZ HDFs of which 20% (109 proteins, p-value 9.11×10−6) interacted with HIV proteins. The trend for SS predicted HDFs was mixed: the overlap ratio was as high as 17.5% (70 of the top 400 predictions interact with HIV proteins), slightly less than the BKZ HDFs, but in no case was the enrichment statistically significant. These results suggest that SinkSource+ HDF predictions are dominated by proteins that lie close to BKZ and HIV proteins in the joint HIV-human PPI network, whereas the SinkSource predictions are dispersed further away. We discuss specific SS+ predicted HDFs that interact with HIV in the context of MCODE clusters below.
The cross validation analysis suggested that HDFs are not randomly located in the human PPI network. Rather, HDFs are closer to each other within the PPI network than to the negative examples. Therefore, in order to better understand how BKZ HDFs and SS+ predicted HDFs are related to each other, we computed the subnetwork of PPIs spanned by these two sets of genes. We applied a modified version of the well-known MCODE [27] graph clustering algorithm to this sub-network (see “Modifying MCODE to Compute PPI Clusters”). The network contained 1,562 proteins and 30,855 PPIs. MCODE identified 41 clusters of varying sizes containing a total of 829 proteins and 16,721 PPIs. Table 1 contains statistics on the 10 clusters with the largest number of PPIs computed by MCODE. Using the one-sided version of Fisher's exact test, we checked the overlap of each of the 42 clusters with BKZ HDFs. Only eight clusters had overlaps that were statistically significant, as shown in Table S5. Table S6 contains a list of BKZ HDFs and HDFs predicted by SinkSource+, annotated with MCODE cluster membership and information on interaction with HIV proteins. Table S7 lists the human PPIs in each MCODE cluster.
We computed GO terms enriched in all clusters. Table 2 contains statistics on highly enriched GO terms in the 10 most highly-connected clusters discovered by MCODE. Among the top 10 clusters, only clusters #1, #4, #7, #8, and #9 have statistically significant overlaps with BKZ HDFs (see Table S5). The fraction of BKZ HDFs is small in clusters #1, #4, and #9, so we reasoned that any functions enriched in these clusters would not be overly influenced by annotations of BKZ HDFs. In contrast, more than half the proteins in clusters #7 and #8 are BKZ HDFs; the functions enriched in these clusters are likely to annotate a number of BKZ HDFs. We now discuss the enriched functions in all clusters in Table 2. We focus our discussion on selected predicted HDFs contained within these clusters and present the support in the literature for the relevance of these HDFs to HIV pathogenesis.
Since HDFs play a critical role in HIV replication [1], [2], [3], we hypothesized that some of them may have value as prognostic markers of HIV pathogenesis and of AIDS development and progression. We anticipated that both experimentally-detected (BKZ) and predicted HDFs would satisfy this hypothesis. To explore this question, we combined BKZ HDFs and predicted HDFs with DNA microarray data from a study detailing the host response to simian immunodeficiency virus (SIV) infection in African green monkeys (AGMs) and pigtailed macaques (PTMs). AGMs are natural reservoirs of SIV that do not develop AIDS, while PTMs are non-natural hosts that develop AIDS when infected with SIV. The virus replicates to the same viral load in both of these hosts. Lederer et al. [56] performed a longitudinal transcriptomic analysis comparing AGMs to PTMs. They analyzed the host response in the setting of acute SIV infection with the same primary isolate (SIVagm.sab92018). They studied three different tissues: blood, colon, and lymph nodes. They collected samples at 10 days and 45 days post-viral inoculation and compared each sample to a sample from the same animal pre-inoculation. For each day-tissue combination, they performed an analysis of three AGMs and three PTMs using rhesus macaque (Macaca mulatta) oligonucleotide microarrays. The probes in this microarray were based on the human Reference Sequence (RefSeq) collection. Thus, there is a direct mapping from these probes to human gene identifiers.
For each tissue (blood, colon, lymph node) and day (10 and 45 post infection) combination, we performed a separate ANOVA analysis, using the host system as factor, to identify genes that are differentially expressed between AGMs and PTMs. Such differentially expressed genes could potentially serve as diagnostic markers of AIDS development and progression. We constructed six lists (three tissues×two time points) of genes that were differentially expressed between AGMs and PTMs to a statistically-significant extent (p≤0.05). We used the one-sided version of Fisher's exact test to determine if BKZ HDFs had a significant intersection with each of these six lists. We repeated this test with the top k predicted HDFs, for values of k starting at 100 and in increments of 100. We used the method of Benjamini and Hochberg [26] to correct for testing multiple hypotheses.
Figure 2 displays plots of the fraction of BKZ HDFs or of predicted HDFs that are also differentially-expressed to a significant extent in the AGM-PTM comparison; Figures S9 and S10 plot the corresponding p-values. Note that the plot for BKZ HDFs is a horizontal line since changing the score cutoff for predictions has no effect on BKZ HDFs. Three notable trends emerged from this analysis. First, for many tissue-day combinations, the overlap fraction for predicted HDFs was larger than the overlap fraction for BKZ HDFs. These trends were most noteworthy in day 10 lymph nodes, where the overlap ratio for predicted HDFs was larger than that for BKZ HDFs over the entire range of prediction confidence values. In particular, in day 10 lymph nodes, the overlap fraction of SS+ predicted HDFs peaked at 0.26 (53 of the top 203 predicted HDFs were also differentially-expressed in day 10 lymph nodes, p-value 0.01). The largest overlap for SS predicted HDFs was also 0.26 (26 of the top 100 predicted HDFs, an insignificant p-value of 0.07). In contrast, the overlap ratio for BKZ HDFs with genes differentially expressed in day 10 lymph nodes was 0.19 (p-value, 0.59). Second, none of the overlaps of BKZ HDFs with differentially-expressed genes were statistically significant, for any tissue-day combination. In contrast, p-values for HDFs predicted by each algorithm were statistically significant (red points in Figure 2 and Figures S9 and S10) in day 10 lymph nodes, across a wide range of prediction confidences. Third, no statistically significant overlaps appeared for predicted HDFs in blood or colon samples at any time point or in day 45 samples from lymph nodes.
We re-estimated the significance of these results after randomizing the gene expression data, by permuting each gene's p-values independently. This process retained the distribution of p-values for each gene, but randomized the associations between p-values and tissue-day combinations. We repeated the overlap analysis for predicted HDFs with each of 10,000 randomized gene expression data sets, for a total of 60,000 randomized tissue-day combinations. We observed only one randomized dataset for which any overlap ratio was at least as large as 0.26, the largest overlap ratio between HDFs predicted by SinkSource+ and genes differentially expressed in day 10 lymph nodes. Thus, the p-value of the observed overlap ratio was 1.7×10−5. For predictions made by SinkSource, we obtained a p-value of 8.3×10−5, for the largest observed overlap of 0.26.
Thus, we concluded that the predicted HDFs have a significant overlap with genes that are differentially expressed between AGMs and PTMs in day 10 lymph nodes, indicating that many predicted HDFs show considerably different programs of expression in the two species in response to SIV infection, especially in early time points. These data suggest that the algorithms have identified a highly responsive subset of potential HDFs, and provide strong experimental support for the prediction that these proteins are in fact HDFs. This result further suggests that viral manipulation of these host factors in lymph nodes soon after infection may have an effect on long-term pathological outcome. We used FuncAssociate to perform GO enrichment analysis on predicted HDFs that were also differentially expressed between AGMs and PTMs in day 10 lymph nodes. The terms we found were almost identical to those reported in the PPI clusters (data not shown). In summary, these results suggest that not only are HDFs critical for viral replication and infection, they may have potential value as prognostic markers to determine pathological outcome and the likelihood of AIDS development.
We have used network-based approaches to predict HIV dependency factors (HDFs). Upon two-fold cross-validation, we found that combining the three experimental data sets yielded much higher precision and recall than using each data set on its own. A number of the algorithms we compared achieved both high precision and recall on cross validation. Our results suggest that global optimization techniques such as SinkSource and SinkSource+ perform slightly better than the simple guilt-by-association rule [57]. Furthermore, SinkSource+ and SinkSource had the most consistent and reliable performance. Software implementing the function prediction algorithms is available at http://bioinformatics.cs.vt.edu/~murali/software/gain. We also observed that estimating the reliability of PPIs did not confer an advantage; in fact, the cross validation results worsened slightly with edge weights (Table S2). The decrease in performance is likely to be a combination of the close proximity of HDFs within the PPI network and the high reliability of PPIs that HDFs are involved in, since the corresponding biological processes are well studied.
We found that the HDFs predicted by SinkSource+ were significantly enriched in proteins that interact with HIV proteins. On the other hand, SinkSource predicted a set of HDFs that were not significantly enriched in HIV-interacting proteins. We computed clusters within the subgraph of the PPI network that encompassed the BKZ HDFs and HDFs predicted by SinkSource+. These clusters were enriched in host cellular complexes and pathways known to be that are known to be manipulated by HIV and perturbed during HIV infection such as the spliceosome, the microtubule network, the proteasome, the mitochondrion, and nuclear import and export.
Finally, we integrated BKZ HDFs and predicted HDFs with gene expression data from a non-human primate study detailing the host response to SIV infection in non-human primates that do not develop AIDS (African green monkeys) and those that do (pigtailed macaques) [56]. We found that up to 26% of predicted HDFs are differentially expressed, when we compared their gene expression profiles in macaques to their profiles in African green monkeys. This differential expression of HDFs was time- and tissue-specific, being strongest in lymph nodes 10 days post-inoculation. These HDFs are excellent candidates for studying transcriptional programs relevant to AIDS progression in humans.
Our results support three conclusions. First, existing genomic screens are incomplete and many HDFs are yet to be discovered. The HDFs predicted by SinkSource+ may include many proteins required for HIV replication that could not have been uncovered experimentally because the predictions were not constrained to non-essential human proteins. Second, HDFs are clustered in the human PPI network and belong to cellular pathways or protein complexes that play a critical role in HIV pathogenesis and AIDS progression. Third, many HDF genes show differential expression during AIDS development in non-human primates. Thus, HDFs may play an important role in the control of initial infection and eventual pathological outcome.
It will be valuable to integrate other HIV-relevant functional genomic data with PPI networks to improve the quality and robustness of HDF prediction. Modeling the impact on off-target effects of siRNAs on false positive HDFs is also important. To date, experiments that have detected HDFs have been performed in cell lines. Approaches such as ours may help to prioritize HDFs for further experimental study in more disease-relevant models such as non-human primates. Ultimately, we anticipate that future extensions of our work may provide multiple new targets and strategies for combating HIV in humans.
Our approach is general purpose and can be applied to interpret other genome wide gene-level studies. In particular, if independent labs have conducted multiple studies to study the same biological system or phenomenon, we provide a methodology to interpret them simultaneously within the context of molecular interaction networks. Our approach can be used to ask if the studies reinforce or contradict each other and to prioritize new genes for further experimental analysis.
We downloaded all the HDF and PPI data used in this study between August and December 2008. We downloaded functional annotation data in December 2010. We used Entrez Gene IDs in all analyses.
We modeled the human protein interaction network as an undirected graph G = (V, E), consisting of a set V of nodes (i.e., proteins) and a set E of edges (i.e., interactions). We used wuv to denote the weight of the edge , computed as described earlier. We partitioned V into three subsets and V− as follows: V+ was the set of HDFs (positive examples), V− was the set of human proteins orthologous to essential mouse proteins (negative examples), and V0 was the remaining set of nodes (unknown examples). For each node v∈V0, our goal was to assess whether v should be a member of V+ or V−. We did so by computing a function that is “smooth” over G. Specifically, we set r(v) = 1 for every node v∈V+, r(v) = 0 for every node v∈V−, and required that r minimize the functionMinimizing S(G, r) enforces the smoothness of r in the sense that the larger the weight of an edge (u, v), the closer in value r(u) and r(v) must be. The function S(G, r) is minimized when, for each node v∈V0,(1)where Nv is the set of neighbors of node v [64]. The right-hand side of this equation can be split into two parts: one corresponding to contributions to r(v) from neighbors in V0 and the second to a constant contribution from neighbors in V+ and V−. Let r0 denote the vector of values taken by the function r at the nodes in V0. Let M denote the square matrix, where , for every . We see that r0 satisfies the equations r0 = Mr0+c, where c is a vector denoting contributions from V+ and V−. We computed r0 by initializing it to 0 for each node and repeatedly applying the operation r0 = M r0+c. This process is known to converge [64], yielding a value of r0 = (I−M)−1c, where I is the identity matrix. The matrix M is sparse, being the adjacency matrix of a PPI network. Therefore, this iterative approach is efficient in practice.
We implemented six other algorithms for the purpose of comparison. The first two algorithms use both positive and negative examples. The other four algorithms do not use negative examples for making predictions, avoiding the uncertainties associated with choosing an accurate set of negative examples. We used both types of algorithms in order to assess the impact of our choice of negative examples on the cross validation results. Table 5 summarizes these algorithms.
Although SinkSource+, Local+, FunctionalFlow, and PRINCE do not use negative examples when making predictions, we used negative examples when computing the performance of these algorithms on cross validation in order to count the number of true negatives and false positives.
A number of approaches are available for computing GO terms enriched in lists of genes [23], [65], [66], [67]. Since BKZ HDFs are unordered while predicted HDFs can be ranked by confidence, we used the FuncAssociate software [23], which can take both unordered and ordered lists of genes as input. For an ordered list of genes, FuncAssociate analyses each one of the list's prefixes, and reports results for the prefix with the smallest p-value. It asks if the genes annotated by each GO term have surprisingly low ranks in the ranked list. The final p-value computed by FuncAssociate can be informally interpreted as the probability that a given overlap between a GO term and a ranked list of genes could be observed if the genes were ranked randomly. Note that FuncAssociate operates solely on the ranked list of genes and the GO annotations. It does not utilize a network. Details on how FuncAssociate operates are provided at http://llama.mshri.on.ca/FuncAssociate_Methods.html.
To determine enriched GO functions in each cluster computed by MCODE, we did not associate any weights with the proteins, since MCODE had already incorporated protein weights. We used an in-house implementation of the Ontologizer [68] to compute enriched GO terms. We chose the Ontologizer because it accounts for annotation dependencies that arise from GO's true path rule. We retained only those functions for which the p-value is at most 0.05, after accounting for multiple hypothesis testing using the method of Benjamini and Hochberg [26].
We modified MCODE to multiply internally-computed node weights with externally-defined node weights. For our application, we supplied the SinkSource+-derived confidence as the weight of a predicted HDF. For every BKZ HDF, we defined its weight as 1. By imposing these externally-defined weights, we aimed to bias MCODE towards finding dense subgraphs in the vicinity of BKZ and SS+ predicted HDFs. Therefore, we included all SS+ predictions together with their confidence levels in the network and used the ability of MCODE to utilize the confidence levels to identify high confidence clusters.
|
10.1371/journal.pmed.1002472 | Estimated clinical impact of the Xpert MTB/RIF Ultra cartridge for diagnosis of pulmonary tuberculosis: A modeling study | The Xpert MTB/RIF (Xpert) assay offers rapid and accurate diagnosis of tuberculosis (TB) but still suffers from imperfect sensitivity. The newer Xpert MTB/RIF Ultra cartridge has shown improved sensitivity in recent field trials, but at the expense of reduced specificity. The clinical implications of switching from the existing Xpert cartridge to the Xpert Ultra cartridge in different populations remain uncertain.
We developed a Markov microsimulation model of hypothetical cohorts of 100,000 individuals undergoing diagnostic sputum evaluation with Xpert for suspected pulmonary TB, in each of 3 emblematic settings: an HIV clinic in South Africa, a public TB center in India, and an adult primary care setting in China. In each setting, we used existing data to project likely diagnostic results, treatment decisions, and ultimate clinical outcomes, assuming use of the standard Xpert versus Xpert Ultra cartridge. Our primary outcomes were the projected number of additional unnecessary treatments generated, the projected number of TB deaths averted, and the projected number of unnecessary treatments generated per TB death averted, if standard Xpert were switched to Xpert Ultra. We also simulated alternative approaches to interpreting positive results of the Ultra cartridge’s semi-quantitative trace call. Extensive sensitivity and uncertainty analyses were performed to evaluate the drivers and generalizability of projected results. In the Indian TB center setting, replacing the standard Xpert cartridge with the Xpert Ultra cartridge was projected to avert 0.5 TB deaths (95% uncertainty range [UR]: 0, 1.3) and generate 18 unnecessary treatments (95% UR: 10, 29) per 1,000 individuals evaluated—resulting in a median ratio of 38 incremental unnecessary treatments added by Ultra per incremental death averted by Ultra compared to outcomes using standard Xpert (95% UR: 12, indefinite upper bound). In the South African HIV care setting—where TB mortality rates are higher and Ultra’s improved sensitivity has greater absolute benefit—this ratio improved to 7 unnecessary treatments per TB death averted (95% UR: 2, 43). By contrast, in the Chinese primary care setting, this ratio was much less favorable, at 372 unnecessary treatments per TB death averted (95% UR: 75, indefinite upper bound), although the projected number of unnecessary treatments using Xpert Ultra was lower (with a possibility of no increased overtreatment) when using specificity data only from lower-burden settings. Alternative interpretations of the trace call had little effect on these ratios. Limitations include uncertainty in key parameters (including the clinical implications of false-negative results), the exclusion of transmission effects, and restriction of this analysis to adult pulmonary TB.
Switching from the standard Xpert cartridge to the Xpert Ultra cartridge for diagnosis of adult pulmonary TB may have different consequences in different clinical settings. In settings with high TB and HIV prevalence, Xpert Ultra is likely to offer considerable mortality benefit, whereas in lower-prevalence settings, Xpert Ultra will likely result in considerable overtreatment unless the possibility of higher specificity of Ultra in lower-prevalence settings in confirmed. The ideal use of the Ultra cartridge may therefore involve a more nuanced, setting-specific approach to implementation, with priority given to populations in which the anticipated prevalence of TB (and HIV) is the highest.
| Xpert Ultra is a new version of a widely used molecular test for tuberculosis (TB) that has a better ability to detect TB (higher sensitivity) but also more frequently gives false-positive results (lower specificity).
These differences in sensitivity and specificity will have different clinical implications in settings with different characteristics such as higher or lower TB and HIV prevalence among the people being tested.
The relative advantages and disadvantages of adopting Xpert Ultra are therefore likely to differ across different clinical contexts.
We modeled and compared the likely clinical outcomes, including number of TB deaths and number of unnecessary treatments, when using Xpert Ultra versus the standard Xpert assay.
We performed this comparison for 3 different hypothetical patient populations in different medium- to high-TB-burden settings (a South African HIV clinic, an Indian TB center, and a Chinese primary care clinic).
We found that the estimated clinical impact of switching from standard Xpert to Xpert Ultra differed dramatically between settings: Ultra yielded fewer than 10 additional unnecessary TB treatments per TB death prevented in the HIV clinic setting in South Africa, in contrast to more than 300 additional unnecessary TB treatments per TB death prevented in the general primary care setting in China.
Xpert Ultra is likely to provide a large clinical benefit over standard Xpert in patient populations with high TB prevalence, high HIV prevalence, and high case fatality ratios for untreated TB.
In populations with low TB prevalence or small proportions of HIV-associated or smear-negative TB, Xpert Ultra will require more cautious implementation and interpretation to avoid costly and harmful overdiagnosis of TB.
| Introduced in 2010, Xpert MTB/RIF (Xpert)—a molecular assay for the detection of tuberculosis (TB) and resistance to rifampin—provides substantial improvements in sensitivity over sputum smear microscopy, previously the cornerstone of TB diagnosis [1]. The sensitivity of Xpert remains imperfect, however, particularly in patients with paucibacillary TB disease (often seen in the context of HIV) [2,3]. More recently, a novel cartridge, the Xpert MTB/RIF Ultra cartridge (“Xpert Ultra”), was developed for TB diagnosis using the same GeneXpert platform, but with technical enhancements (including larger specimen volume, probes for repeated elements in the mycobacterial genome, and optimized fluidics and polymerase chain reaction cycling) designed to further increase the sensitivity of Xpert for detection of TB [4]. The performance of the Xpert Ultra cartridge was subsequently evaluated in a large 10-site, 8-country study [5], which confirmed its increased sensitivity for diagnosis of active pulmonary TB relative to the existing Xpert (G4) cartridge (“standard Xpert”), using sputum culture as a reference standard. In particular, Ultra was estimated to add 5% to the sensitivity of standard Xpert among all culture-positive study participants and 13% (increasing the sensitivity for TB detection from 77% to 90%) among those infected with HIV. However, these study data also suggested a loss of specificity with Ultra, particularly among individuals with a history of previous TB treatment; false-positives increased more than 2-fold with Ultra compared to standard Xpert in those with no prior TB and more than 3-fold in those with a history of TB.
Based on its improved sensitivity, the World Health Organization has endorsed the new Ultra cartridge [6], and it has been made available to eligible countries at the same concessional pricing as the standard Xpert cartridge [7]. In deciding how best to implement use of the Ultra cartridge for diagnosis of adult pulmonary TB, it is important to consider how this trade-off between sensitivity and specificity would translate into clinical and/or public health outcomes. We therefore constructed a simulation model to explore the downstream clinical consequences of replacing the standard Xpert cartridge with the Ultra cartridge as the initial diagnostic test for presumptive pulmonary TB in 3 emblematic settings.
We developed a Markov microsimulation model of TB, using cohorts of adults (≥15 years old) undergoing diagnostic sputum evaluation for suspected pulmonary TB in a setting with Xpert capacity. Our primary comparison was of expected diagnostic and clinical outcomes using standard Xpert versus Xpert Ultra. We selected 3 emblematic settings to illustrate a range of different patient populations in which Xpert might be used: a TB diagnosis and treatment center in India’s public health sector, an ambulatory HIV care setting in South Africa, and a primary care setting in China. As detailed in Table 1, these settings differ in demographic makeup, underlying TB prevalence, HIV prevalence, prevalence of rifampin resistance, empiric treatment practices, and TB treatment outcomes. The breakdown of the resulting cohorts according to TB, HIV, and rifampin-resistance status is shown in S2 Table.
Within each setting-specific cohort, the values of cohort-defining parameters from Table 1 were used to randomly assign each of 100,000 individual simulated patients an age, sex, underlying TB (and rifampin resistance) status, HIV status, and history of previous TB treatment. Using the additional parameters in Tables 1 and 2, we then simulated individual-level diagnostic evaluation, resulting treatment decisions, and ultimate clinical outcomes for each person in each setting-specific cohort (as illustrated in Fig 1). To accomplish this, we defined a “diagnostic episode” as consisting of all clinical decision-making from the time that a patient is considered at sufficient risk of pulmonary TB to merit Xpert testing to the time that the patient is lost to follow-up, started on TB treatment, or no longer thought to have TB. A diagnostic episode may span multiple visits but, to be included in this analysis, must at some point include a diagnostic evaluation for adult pulmonary TB using either standard Xpert or Ultra. We model the diagnostic episode for each patient in the simulated cohort, assuming that those who test positive with Xpert are initiated on TB treatment (including second-line TB treatment if rifampin resistance is detected). For those who test negative, we assume a setting-specific probability of empiric treatment (i.e., initiating treatment for TB in the absence of a bacteriological result). In order to focus on the clinical impact of Xpert testing, we do not explicitly model other ancillary tests (e.g., chest X-ray, antibiotic trials) but rather assume for simplicity that the results of any such tests performed, coupled with clinical judgement, result in empiric TB treatment for a proportion of Xpert-negative patients. We then vary this empiric treatment proportion directly in sensitivity analysis. We assume that all such empiric treatments involve first-line therapy. Importantly, we also assume outcomes for rifampin-resistant TB that are better than those currently reported, in order not to bias findings against Ultra in light of pharmaceutical and other advances that are likely to improve those outcomes in the future.
Following the outcome of the diagnostic episode (treated for drug-susceptible or rifampin-resistant TB, or not treated), we then model both treatment outcomes and the ultimate probability of TB death. For those who are treated, treatment outcomes include cure/treatment success, death (due to TB or other causes), and failure/relapse (with the possibility of acquired rifampin resistance), with probabilities based on data reported to the World Health Organization from each country. For individuals with active TB who are not successfully treated (or not treated at all), we do not explicitly model all future clinical care (including possible subsequent diagnostic episodes and/or TB treatment) but rather assume a probability of ultimate TB death equal to the reported case fatality ratio of TB or multidrug-resistant TB in each country, stratified by HIV status. This probability is also varied directly in sensitivity analysis.
For each simulated cohort, we compared clinical outcomes under 2 alternative scenarios: one with the use of the standard Xpert cartridge and one with the use of the Ultra cartridge. We defined 3 a priori co-primary outcomes, each measured as the expected incremental value if standard Xpert were switched to Ultra: (a) incremental TB-attributable deaths averted, (b) incremental unnecessary TB treatments, and (c) the ratio of these 2 competing outcomes (incremental unnecessary treatments per incremental TB death averted). TB-attributable deaths include all TB-attributable deaths during treatment, after unsuccessful treatment, or after a missed diagnosis. Unnecessary TB treatments include treatments of people without underlying TB, due to false-positive Xpert result or incorrect empiric treatment (assuming that switching from the standard Xpert to the Ultra cartridge does not change the proportion of Xpert-negative patients to whom empiric treatment is prescribed). As a proxy for avertible transmission potential, we also considered as secondary outcomes the difference in the number of TB cases and the number of rifampin-resistant TB cases that remained untreated after the diagnostic attempt using either standard Xpert or Ultra.
To compare the standard Xpert cartridge against the Ultra cartridge, we assumed accuracy values as shown in Table 2, reflecting data from the recently performed diagnostic accuracy study among adults with symptoms of pulmonary TB at 10 sites in 8 countries, using mycobacterial culture as a reference standard [5]. Basing estimates on study data, the sensitivity and specificity for TB of Ultra were represented as beta distributions conditional on standard Xpert result (i.e., different for those with a positive versus negative standard Xpert result), with mean and standard deviation based on the confidence intervals reported in the trial. The sensitivity of Ultra for TB was also stratified by HIV status, and the specificity for TB was stratified by prior TB history. We also estimated the sensitivity and specificity of each assay for rifampin resistance as shown in Table 2.
The Ultra cartridge also has an additional semi-quantitative category on the lower end of the spectrum (“trace call”) indicating very low levels of mycobacterial DNA amplified. In our primary analysis, we included this trace call as a positive result, per the existing configuration of the test (for maximum sensitivity of Ultra). In a secondary analysis, we considered alternative approaches to interpretation of Ultra in which a trace call was assumed to represent a negative result—either for all individuals or only for those individuals with a prior history of TB treatment (“conditional trace call” scenario). We also considered an approach in which a trace call triggered a repeat Ultra test for adjudication (“positive trace repeated” scenario).
S1 Supplemental Methods provides details of the estimation of other parameters not directly related to the diagnostic assays.
For each of the 3 clinical scenarios (100,000 simulated adults each, which for the Indian TB center setting represents annual presumptive TB patients drawn from a general population of approximately 5 million people), we used Latin hypercube sampling to repeatedly draw random sets of all the parameters shown in Table 2 and in the last three rows of Table 1, assuming triangular distributions with the mode and upper/lower bounds provided in the tables (except for the beta distributions used for Ultra sensitivity and specificity as described above). We sampled 5,000 random parameter sets after verifying that this was sufficient to yield consistent results between sets of simulations (S3 Table). Each parameter set was then used to inform a stochastic simulation of diagnostic, treatment, and clinical outcomes, in which we first ran a simulation assuming the use of standard Xpert and then performed a counterfactual simulation differing from the initial simulation only by the replacement of standard Xpert with Ultra. Incremental outcomes were then evaluated by comparing results between the initial and counterfactual simulations; this process was repeated for each of the 5,000 parameter sets, in each clinical scenario. We report 95% uncertainty ranges (URs) as the 2.5th to 97.5th percentile of results from these 5,000 paired (initial and counterfactual) simulations. These URs thus reflect uncertainty in underlying parameter values (over the 5,000 random sets drawn), stochastic process uncertainty (as each of the 5,000 simulations represents a different stochastic realization), and the expected correlation between the results of standard Xpert and Ultra (by evaluating incremental outcomes from paired initial and counterfactual simulations).
We performed 1-way sensitivity analysis on all model parameters across the ranges specified in Tables 1 and 2, using partial rank correlation coefficients to control for potential variation in other model parameters. We then performed 3-way sensitivity analysis across 3 influential setting-specific parameters (TB prevalence, HIV prevalence, and TB-associated mortality rate).
Several additional sensitivity analyses considered alternative estimates for specific sets of parameters. To capture the possibility of lower specificity of Xpert and Ultra in settings of higher TB incidence (observed in a post hoc analysis of diagnostic accuracy study data after adjusting for participants’ personal history of TB, and possibly reflecting a greater probability of previously unrecognized, spontaneously resolved TB or inhalation of nonviable Mycobacterium tuberculosis), we ran additional simulations for each setting using specificity estimates based only on data from study sites with correspondingly high or low TB incidence. For the Chinese primary care setting scenario, we repeated simulations with the specificities of standard Xpert and Ultra reestimated after restricting the primary study data to the 4 countries with estimated national TB incidence < 100/100,000 person-years, while for the Indian TB center and South African HIV clinic settings, we repeated simulations with assay specificities reestimated using data only from study sites in countries with national TB incidence ≥ 100/100,000 person-years (S5 Table). This stratification of specificity by national TB incidence corresponds to a post hoc analysis performed on data from the diagnostic accuracy study of Ultra. In another sensitivity analysis performed at the request of a reviewer, we considered worse treatment outcomes for rifampin-resistant TB, consistent with outcomes reported by WHO for 2013 multidrug- or rifampin-resistant TB cohorts: a 14% (10%, 20%) probability of TB mortality during treatment and a 63% (54%, 72%) probability of cure among those who survive treatment [10]. Other sensitivity analyses also performed at the request of reviewers consider empiric treatment in the Chinese primary care setting and changes in clinician behavior (e.g., lower levels of empiric treatment, increased use of confirmatory testing) resulting from their knowledge of Ultra’s lower specificity. Details of the parameter values used in these analyses are provided in S1 Supplemental Methods. Finally, we considered the potential impact of the imperfect sensitivity of TB culture as a “gold standard,” such that some positive Ultra results originally classified as false-positives (i.e., culture-negative) were reclassified as true positives (details in S1 Supplemental Methods).
The model was implemented using R version 3.2.2 [21]. We have made the model code available at https://github.com/eakendall/xpert-ultra.
The primary outcomes, by clinical setting, are shown in Table 3. In the Indian TB center setting, for example, switching from standard Xpert to Ultra resulted in appropriate treatment for a median of 3% (95% UR: 0.4%, 5.5%) more TB cases (where “appropriate” is defined as second-line treatment for those cases with rifampin resistance and any TB treatment for other TB cases), increasing the median proportion appropriately treated from 88.5% (95% UR: 84.5%, 92.0%) to 91.5% (95% UR: 87.2%, 94.8%). However, switching to Ultra also increased the median proportion of people without TB being unnecessarily treated by 2.1% (95% UR: 1.1%, 3.2%). Since we assumed in this setting that nearly 8 individuals without TB would be evaluated for every case of true TB, Ultra resulted in a median of 5.2 (95% UR: 1.9, 19.1) additional unnecessary TB treatments for every additional TB case detected (S1 Fig). After modeling long-term effects on mortality, use of Ultra rather than standard Xpert was projected to avert 0.5 TB deaths (95% UR: 0, 1.3) and generate 18 unnecessary treatments (95% UR: 10, 29) per 1,000 individuals evaluated—for a median ratio of 38 unnecessary treatments per death averted (95% UR: 12, indefinite upper bound due to simulations in which Ultra averted no deaths). In the HIV care setting in South Africa, where the amount of additional sensitivity added by Ultra is larger and the mortality rates associated with untreated TB are higher, this ratio was more favorable: a median 7 incremental unnecessary treatments per incremental TB death averted (95% UR: 2, 43). By contrast, in the Chinese primary care setting, with lower TB prevalence (15 non-cases evaluated per TB case) and mortality, the median ratio rose to 372 (95% UR: 75, indefinite upper bound) unnecessary treatments per TB death averted.
We also estimated the increase in the number of patients started on appropriate treatment after evaluation with Ultra versus standard Xpert (S6 Table). Per 1,000 individuals evaluated with Xpert in the Indian public TB center, for example, Ultra led to an additional 3.4 (95% UR: 0.7, 6.2) prompt treatment initiations for drug-susceptible TB and an additional 0.04 (95% UR: −0.1, 0.2) prompt second-line treatment initiations for rifampin-resistant TB (S6 Table). The resulting reduction in transmission (before these cases would otherwise be diagnosed) was not estimated in this analysis, but is likely to be small given the paucibacillary nature of those cases detected by Ultra but not detected by standard Xpert.
Exclusion of the trace call reduced the incremental number of unnecessary treatments (with Ultra versus standard Xpert) by more than 50% in all settings, but also reduced the incremental number of deaths averted by similar proportions (Fig 2; S4 Table). In general, the choice of whether and how to include the result of the trace call resulted in little change in the ratio of additional unnecessary treatments per TB death averted. Differences in outcomes between clinical settings were substantially larger than the differences in outcomes comparing different approaches to the trace call (Fig 2).
The estimated number of TB deaths averted was most sensitive to variation in the case fatality ratio for drug-susceptible TB (i.e., the probability of subsequent death following a missed diagnosis of TB) and to the sensitivity of Ultra among individuals with the cohort’s predominant HIV status (S3 Fig). By contrast, the incremental number of unnecessary treatments was highly sensitive to the estimated specificity of Ultra (S3 Fig). Characteristics of the clinical cohorts—which were held fixed for each setting in our primary analysis—were also important. For example, a 2-fold increase in the prevalence of TB (among those with symptoms) or a 4-fold increase in the prevalence of HIV each had similar effects as a 2-fold increase in TB case fatality (S4 Fig).
When specificity parameter estimates for the Indian TB center and the South African HIV clinic were based only on data from study sites in higher-incidence countries (a post hoc analysis of data from the diagnostic accuracy study), the performance of Ultra became somewhat less favorable in those settings, with an increase in the expected number of unnecessary treatments per death prevented from 38 to 55 in the Indian TB center and from 7 to 10 in the South African HIV clinic (Table 4). Conversely, because specificity estimates based only on data from study sites in lower-incidence countries had largely overlapping confidence intervals for standard Xpert and Ultra (S5 Table), the expectation that Ultra would result in more unnecessary treatments became uncertain when using data on specificity only from lower-burden settings; the median number of unnecessary treatments per death averted in the Chinese primary care setting was reduced from 372 to 14 in this sensitivity analysis, but with very large uncertainty, ranging from no added unnecessarily treatments to a ratio of >1,000 (Table 4).
Assuming more pessimistic future treatment outcomes for rifampin-resistant TB increased TB deaths in both the standard Xpert scenario and the Ultra scenario but had little impact on the primary results of deaths averted by Ultra or the ratio of unnecessary treatments per death averted (S7 Table). Adding empiric TB treatment in the Chinese primary care setting also had little impact on the results (S8 Table). Following positive Ultra results with a separate confirmatory test could have a mortality benefit only if the confirmatory test were highly sensitive as well as specific; for existing diagnostics, such as chest X-ray, that could be considered as a confirmatory test after screening positive by Ultra, the loss of sensitivity associated with the confirmatory test would outweigh the benefit of the improved specificity (S9 Table). In an HIV care setting with a high rate of empiric treatment following negative results on standard Xpert, a greater confidence in negative Ultra results could substantially reduce unnecessary treatments with a relatively small impact on case detection and TB mortality (S9 Table). Considering the possibility that some culture-negative, Ultra-positive results represented false-negative cultures changed the ratio of unnecessary treatments per death averted by at most a factor of 2 (S10 Table).
Public health decision-making about replacing standard Xpert with Ultra will involve difficult trade-offs. The increased sensitivity of Ultra can lead to lower TB mortality, morbidity, and transmission, but the reduced specificity can result in individuals without TB being unnecessarily exposed to the toxicity and inconvenience of prolonged therapy. A quantitative understanding of these trade-offs—which are likely to be very different in different epidemiologic and clinical contexts—can guide adoption and implementation decisions. We have used a simulation approach to quantify the anticipated clinical consequences of replacing the standard Xpert cartridge with the Ultra cartridge for the diagnosis of adult pulmonary TB in different medium- to high-incidence clinical settings. Depending on the setting and diagnostic algorithm employed, we estimate that use of Ultra for this indication could result in anywhere from fewer than 10 to more than 300 additional unnecessary treatments for every TB death averted. This ratio, which we offer as a tool for understanding the variation in clinical consequences between settings, is most favorable where TB prevalence (especially HIV-associated TB prevalence) and TB mortality are high (e.g., HIV care in South Africa), and, unless preliminary evidence suggesting higher specificity in lower-burden settings is confirmed, it is least favorable where TB prevalence and mortality are lower (e.g., general primary care in China). These findings suggest that the same changes in sensitivity and specificity would have dramatically different consequences in different clinical settings.
We emphasize that the settings we have modeled are intended to represent specific clinical contexts and not all TB diagnostic attempts within a given country. There will be other clinical contexts within these heterogeneous countries (e.g., HIV care settings in India or China) where the relative benefits of standard Xpert versus Ultra will differ from the settings we have modeled in those countries. The practicalities of supply, procurement, and training may make it difficult, however, to offer Ultra alongside standard Xpert for use on a case-by-case basis at the clinic or hospital level or for different diagnostic tasks within the same health system. Rather, decisions to implement Ultra are likely to be made at the level of entire countries and to involve a wide range of clinical settings and potential indications (e.g., active case finding versus symptom-driven diagnosis). The level of overtreatment considered to be acceptable will vary in different social contexts; considerations will include potential strain on healthcare systems (by multiplying the number of people being treated for TB), patient faith in healthcare systems, and preferences regarding the relative harm of under- versus overdiagnosis [22,23]. It is also important to recognize that Ultra may have additional advantages that were not included in this modeling exercise. These include the potential for improved sensitivity (without the same specificity cost) in children [24] and patients with extrapulmonary TB [25]. Although settings with extremely high prevalence of rifampin resistance were not included in the current analysis due to limited data on the relative performance of the 2 Xpert assays for rifampin resistance detection in clinical contexts, small analytical studies demonstrate increased fidelity in rifampin resistance detection and improved specificity in differentiation of non-tuberculous mycobacteria [26], suggesting that Ultra may offer particular benefit in such settings. The decision to implement Ultra may therefore be different in countries with different TB epidemics, different healthcare systems, different societal values, and different relative weightings of the advantages and disadvantages of Ultra.
Importantly, policy decisions about whether and how to implement Ultra may also eventually influence clinical decision-making. For example, confidence among clinicians in the higher sensitivity of Ultra could reduce empiric treatment practices—and the consequences could be either positive (fewer overtreatments) or negative (missed treatment of Ultra false-negatives). In addition, if data suggesting a lower specificity of Ultra are borne out in clinical experience, then the decision to adopt Ultra could, over time, result in more selective use of Xpert tests. Again, as illustrated by our sensitivity analysis regarding such potential changes in practices, this could have positive effects (reducing excessive use of this diagnostic resource) or negative ones (reducing the testing of true cases).
Our analysis is helpful in identifying the key characteristics of settings in which Ultra is likely to be most preferred—namely settings with high prevalence of TB among adult patients likely to be evaluated with Xpert, as well as high prevalence of HIV and high risk of TB mortality if diagnoses are missed. Similarly, we identify characteristics of settings where there is greater risk that the disadvantages of Ultra may outweigh its benefits, and where standard Xpert might be preferred—settings with lower TB prevalence, low HIV prevalence, and low risk of TB mortality. In settings falling between these extremes (such as the illustrative Indian TB center in our model), the choice of cartridge is likely to depend on local priorities, for example, whether it is judged acceptable to subject dozens of people to unnecessary treatment in order to avert 1 death from pulmonary TB. Our analyses also suggest that the clinical setting is likely a much stronger determinant of the risk–benefit ratio in using Ultra than is the use or non-use of the trace call. Within a given setting, use of the trace call appears to increase unnecessary treatments in proportion to the TB deaths it averts. Therefore, in settings where the trade-off between these outcomes is judged to clearly favor Ultra, it is likely to also favor inclusion of trace call diagnoses, whereas disregarding or confirming trace call results may make sense in health systems that adopt Ultra for TB diagnosis but have less confidence that the associated sensitivity gains outweigh the specificity losses in adult pulmonary TB.
The uncertainty in our quantitative estimates remains reasonably high, reflecting in part the challenges of estimating the precise magnitude of sensitivity and specificity differences between Ultra and standard Xpert in multiple types of patients. However, the clinical data to inform these estimates come from a multicenter study of over 1,500 patients, and it is unlikely that additional data on diagnostic performance would greatly improve decision-making. This is because setting-specific parameters (e.g., TB prevalence, TB case fatality) are at least as important as assay-specific parameters. A possible exception in this regard concerns differences in specificity observed in post hoc analysis between study sites with different TB incidence, even among individuals with no history of TB. If, as experience with this assay accumulates, higher specificity in lower-incidence settings continues to be observed, then our results may be pessimistic with respect to use of Ultra in those settings. On the other hand, if specificity in higher-incidence settings is lower than modeled here, use of Xpert Ultra in those settings could be more problematic. To the extent that Ultra test characteristics are consistent between epidemiologic settings, our model’s ability to inform decision-making in any given setting will be primarily limited by our ability to describe the epidemiology of that setting, not by uncertainty regarding the diagnostic accuracy of Ultra. For example, our model suggests that the probability of TB death after a missed diagnosis is a critical parameter value—and this value is poorly understood in most settings [27]. Moreover, clinical and policy decisions are likely to be made on a semi-quantitative basis at best. For example, narrowing the confidence intervals of these quantitative estimates is less likely to influence decision-making than is a qualitative assessment of whether the loss of specificity associated with Ultra is acceptably small or unacceptably large, compared to the deaths averted and other potential benefits.
This analysis has a number of important limitations. We did not model the transmission of TB and thus may have underestimated the impact of Ultra after accounting for secondary transmission from cases diagnosed by Ultra but not standard Xpert. However, the amount of transmission from Xpert-negative TB cases is uncertain and likely to be low in settings where these cases would eventually come to clinical attention [28–30]. We also did not account for long-lasting sequelae of TB disease and delayed diagnosis [31], focusing instead on TB mortality (which represents the vast majority of disability-adjusted life years in other studies of TB disease [32–34]). We restricted our analysis to evaluation of adults presenting with symptoms of pulmonary TB in medium- to high-TB-burden settings. For other potential uses (e.g., diagnosis of extrapulmonary [35] or pediatric [36] TB, and use in low-prevalence settings), preliminary data suggest that sensitivity increases may be substantial and may come with less specificity cost [24,25]. For these indications, additional analyses in these specific populations would be warranted once confirmed data are available. Our primary analysis uses pooled data from all diagnostic accuracy study sites (for assay characteristics that were expected to be consistent across sites); if preliminary suggestions of higher specificity in lower-prevalence sites are confirmed, differences in outcomes across settings would be attenuated. Finally, our model does not include data on costs or the implications of false-positive or false-negative results on health utility. Future context-specific health technology assessments would therefore be useful to convert these results into estimates of cost-effectiveness and budget impact across different settings for use in national-level decision-making.
In summary, this individual-based cohort model in 3 illustrative clinical settings demonstrates the clinical implications of the sensitivity/specificity trade-off when replacing standard Xpert with Xpert Ultra for diagnosis of adult pulmonary TB. We demonstrate that this replacement will likely prevent a substantial number of TB deaths in settings characterized by high TB and HIV prevalence and mortality. While less certain, our findings also suggest that switching to Xpert Ultra may result in substantial overtreatment in settings with moderate prevalence of TB and lower mortality risk. To optimize the use of Xpert Ultra to improve TB diagnosis in moderate- and high-burden settings, we must carefully consider the diversity of contexts into which it might be introduced and the complexity of policy recommendations that might ensue.
|
10.1371/journal.pgen.1003079 | The CPEB Protein Orb2 Has Multiple Functions during Spermatogenesis in Drosophila melanogaster | Cytoplasmic Polyadenylation Element Binding (CPEB) proteins are translational regulators that can either activate or repress translation depending on the target mRNA and the specific biological context. There are two CPEB subfamilies and most animals have one or more genes from each. Drosophila has a single CPEB gene, orb and orb2, from each subfamily. orb expression is only detected at high levels in the germline and has critical functions in oogenesis but not spermatogenesis. By contrast, orb2 is broadly expressed in the soma; and previous studies have revealed important functions in asymmetric cell division, viability, motor function, learning, and memory. Here we show that orb2 is also expressed in the adult male germline and that it has essential functions in programming the progression of spermatogenesis from meiosis through differentiation. Like the translational regulators boule (bol) and off-schedule (ofs), orb2 is required for meiosis and orb2 mutant spermatocytes undergo a prolonged arrest during the meiotic G2-M transition. However, orb2 differs from boule and off-schedule in that this arrest occurs at a later step in meiotic progression after the synthesis of the meiotic regulator twine. orb2 is also required for the orderly differentiation of the spermatids after meiosis is complete. The differentiation defects in orb2 mutants include abnormal elongation of the spermatid flagellar axonemes, a failure in individualization and improper post-meiotic gene expression. Amongst the orb2 differentiation targets are orb and two other mRNAs, which are transcribed post-meiotically and localized to the tip of the flagellar axonemes. Additionally, analysis of a partial loss of function orb2 mutant suggests that the orb2 differentiation phenotypes are independent of the earlier arrest in meiosis.
| Cytoplasmic Polyadenylation Element Binding (CPEB) proteins bind and recognize CPE sequences in the 3′ UTRs of target mRNAs and can activate and/or repress their translation depending on the mRNA species and the biological context. Drosophila has two CPEB family genes, orb and orb2. orb is expressed in the germline of both sexes and has critical functions at multiple steps during oogenesis; however, it plays only a limited role in spermatogenesis. Here we show that the second CPEB family gene orb2 has the opposite sex specificity in germline development. While it appears to be dispensable for oogenesis, orb2 has essential functions during spermatogenesis. It is required for programming the orderly and sequential progression of spermatogenesis from meiosis through differentiation. orb2 mutants fail to execute the meiotic G2-M transition and exhibit a range of defects in the process of sperm differentiation.
| Proteins in the Cytoplasmic Polyadenylation Element Binding (CPEB) family were first identified in Drosophila ovaries and Xenopus oocytes [1]–[4]. In both organisms the CPEB proteins function in the localization and translational regulation of mRNAs encoding key developmental and polarity determinants as well as factors controlling the process of egg maturation. Since then CPEB family proteins have been implicated in many other biological contexts. These include translational regulation of embryonic cell division [5], [6], regulation of p53 expression [7], [8], synaptic plasticity in the rat hippocampus [9], long-term memory in Aplysia [10], [11] and spermatogenesis in the worm [12]. The CPEB proteins bind to CPE elements in the 3′ UTRs of target mRNAs and can both repress and activate translation. Translation activation typically involves the phosphorylation of the CPEB protein and the subsequent recruitment of a cytoplasmic poly A polymerase which extends the poly A tail [13].
Most animals have two or more CPEB genes. Completed genome sequences reveal that humans, mice, and C. elegans have four genes, while there are only two CPEBs, orb and orb2, in Drosophila. The homology between the CPEBs is largely restricted to the C-terminal region of the protein, where two RNA-Recognition Motif (RRM) domains are found, while the N-terminal domain is highly divergent even amongst closely related species. Phylogenetic trees indicate that the CPEB genes fall into two different subgroups. One subgroup includes Drosophila orb, mouse CPEB1 and the canonical Xenopus CPEB, while the other subgroup contains the second Drosophila CPEB gene, orb2, as well as mammalian CPEB 2, 3, and 4 [12], [14].
The Drosophila orb gene has been extensively characterized. Its expression appears to be restricted to the germline as neither mRNA nor protein can be detected in somatic tissues of the embryo, larvae and adult. While a male-specific Orb isoform is expressed in the male germline, its activity is not absolutely essential since the fertility of orb null males is reduced but not eliminated (Agunwamba and Xu, unpublished). In contrast, orb plays a central role in the process of oogenesis. orb expression is first activated during the mitotic divisions that ultimately generate an egg chamber containing 15 nurse cells and an oocyte. At this stage orb activity is required for the proper specification of the oocyte. Subsequently, orb is required for establishing the anterior-posterior and dorsal-ventral axes of the egg and embryo. Amongst the key orb mRNA regulatory targets are the polarity determinants oskar and gurken [15]–[18].
Unlike orb, the second Drosophila CPEB gene, orb2, is broadly expressed in both the soma and germline. The highest levels of Orb2 are in the embryonic, larval and adult CNS, and in the germ cells of the male testes [19]. There are two Orb2 isoforms, one of 75 kD and the other of 60 kD. The larger isoform is expressed in somatic tissues and the germline of both sexes, while the smaller isoform is found in testes but is not detected elsewhere. The isoforms share a 542 C-terminal amino acid sequence, but have unique N-termini of 162 and 9 amino acids respectively. Included in the common region is the conserved C-terminal CPEB signature RRM type RNA binding and zinc finger domains. The N-terminal half of both isoforms has short conserved sequences rich in serine or histidine interspersed with poorly conserved sequences containing poly-glutamine or poly-glycine repeats [19].
As might be expected from its broad expression pattern, orb2 has a number of somatic functions. During embryogenesis it is required for asymmetric cell division of neuroblast and muscle precursor stem cells and appears to function by promoting the localized accumulation of atypical Protein Kinase C (aPKC) [19]. In addition, orb2 mutants have substantially reduced viability, a shortened life span, and defects in behavior and long-term memory [14], [19], [20], [21]. Here we report that orb2 is essential for spermatogenesis, and that it functions in programming the orderly and sequential progression of spermatogenesis from meiosis through differentiation.
In situ hybridization and antibody staining were used to examine orb2 expression in the testes. While there is little if any orb2 mRNA (Figure 1B, 1B′) or protein (Figure 1C, 1C(I)) in stem cells, low levels are detected in the mitotic cysts. After mitosis is finished and the interconnected spermatocytes begin to grow, there is a substantial upregulation in both mRNA and protein. This period of growth corresponds to the stage when many gene products needed for subsequent steps in spermatogenesis are synthesized [22], [23]. Though Orb2 protein is found throughout the spermatocyte cytoplasm, higher levels of protein are concentrated in a ring around the nucleus (Figure 1C(II), 1D). Orb2 expression peaks as the mature spermatocytes go through meiosis and high levels of Orb2 are found in 32 and 64 cell spermatid cysts (Figure 1C(III)).
Orb2 persists after the spermatids in the 64 cell cysts start differentiation and begin flagellar axoneme elongation (Figure 1C(IV)). As the axonemes begins to elongate, the 64 spermatid nuclei bundle together and then begin to condense into needle-like structures (Figure 1F, inset, Figure 1A). Though Orb2 is distributed along the entire axoneme bundle, the highest concentrations are found in a prominent band (Figure 1E, arrowhead) close to the distal tip of the growing flagellar axonemes (Figure 1A). The leading edge of the axonemes is just in front of the Orb2 band and this region contains small clumps of Orb2 (Figure 1E, arrow). In the region behind the band, Orb2 is organized into a series of striated lines that extend towards the sperm nuclei at the proximal (basal) tip of the spermatid (Figure 1E, bracket) and presumably correspond to individual flagellar axonemes in the spermatid bundle.
While Orb2 is present in elongating spermatids that have not yet completed nuclear condensation (* in Figure 1F), it disappears once elongation and nuclear condensation are completed (o in Figure 1F). To confirm this, we compared the accumulation patterns of Orb2 and Don Juan-GFP (DJ-GFP). While DJ-GFP is highly expressed once the nuclei have condensed and individualization begins, it is not found in spermatids that are still undergoing elongation [24], [25]. As expected, we did not observe spermatids that simultaneously had Orb2 and DJ-GFP. Moreover, since some fully elongated spermatids with condensed nuclear bundles have neither Orb2 nor DJ-GFP (x in Figure 1F), there seems to be a delay between the disappearance of Orb2 and DJ-GFP expression. This suggestion is supported by a comparison of the Orb2 and Orb expression patterns. orb is transcribed post-meiotically and orb mRNAs localize in a band at the distal tip of elongating spermatids [3], [26]; however, the localized mRNAs does not appear to be translated until the end of the elongation phase after Orb2 begins to disappear. Figure 2A–2D show that high levels of Orb are found in the tips of elongated spermatids that have neither Orb2 nor DJ-GFP. On occasion we observed spermatids that have activated Orb translation but still retain some residual Orb2 (Figure 2B, arrowhead).
While meiosis and differentiation require different gene products for execution and have their own regulators, there is a class of genes that control both aspects of spermatogenesis. Included in this group are always early (aly), spermatocyte arrest (sa), meiosis I arrest (mia) and cannonball (can) which encode testes specific TAFs (TATA Box Protein associated factors) [22], [27]. Mutations in these testes specific TAFs cause spermatocytes to arrest at the G2-M transition of meiosis I and block the expression of factors needed for differentiation [28]. However, though these genes encode factors essential for Pol II activity, the effects of mutations are not limited to general transcription. For example, twine (twe) mRNA is expressed in tTAF mutants, but is not properly translated [28]. Figure S1 shows that mutations in these four genes have two effects on Orb2 protein expression in the testes. First, Orb2 levels were substantially reduced (Figure S1A). Second, there was a noticeable reduction in the electrophoretic mobility of the larger Orb2 isoform. As illustrated for the sa mutation, treatment of the testes extract with lambda phosphatase removes the Orb2 signal with slow electrophoretic mobility and indicates that phosphorylation is responsible for the reduced mobility of Orb2 in the mutant testes (Figure S1B). As might be expected, these tTAF mutations do not seem to affect Orb2 in somatic tissues such as the head (Figure S1A).
To better understand how orb2 functions in spermatogenesis, we examined the effects of mutations. Previously we characterized a collection of 5 transposon insertions in the orb2 locus [19]. As shown in Figure 3, two of the transposons, 1556 and 4965 have no effect on Orb2 expression. This is expected as 1556 is inserted upstream of the orb2-1 promoter, while 4965 is located downstream of the Orb2 protein coding sequences. Two of the transposons, 6090 and 1925, are inserted downstream of both the orb2-1 and orb2-2 promoters and interfere with expression of orb2 mRNAs encoding the 75 kD isoform in the testes and head (Figure 3, Figure S2 and [19]). In contrast, the 1793 insertion, which is located farther upstream in between the orb2-1 and orb2-2 promoters, affects 75 kD expression in the testes, but not in the head, suggesting that orb2-1 is more heavily used in the testes, while orb2-2 is more heavily used in the head (Figure S2). As expected from their insertion sites, none of the transposons affect the 60 kD isoform. On the other hand, the reduction in the 75 kD isoform in 6090, 1925 and 1793 is accompanied by a small but reproducible increase in the 60 kD isoform (Figure 3B). This raises the possibility that a negative feedback loop might regulate the levels of the two isoforms.
Consistent with an important role for Orb2 in spermatogenesis, we find that the fertility of homozygous 6090, 1925, and 1793 males is substantially impaired (not shown). When trans to deficiencies that uncover orb2, 1925 is completely sterile, while 6090 and1793 occasionally give fertile males (Figure 3C). In contrast, the two insertions, 1556 and 4965, that have no effect on the expression of the 75 kD isoform, are fully fertile. That sterility is due specifically to the loss of the Orb2 75 kD isoform is supported by the finding that excision of the transposon insertions restores the expression of this isoform and reverts the sterility phenotype (6090−1, Figure 3).
Since the mutants still expressed the 60 kD isoform, along with residual 75 kD isoform, they could retain some orb2 function. For this reason, we generated orb2 nulls using FLP recombination (Figure S2) [29], [30]. Two upstream piggyBac insertions (1556 and 1925) contain correctly oriented FRT sites for deleting the orb2 protein coding sequence when paired with the downstream 4965 insertion. The resulting deletions, orb27 (1925×4965) and orb236 (1556×4965), eliminate orb2 mRNA and protein expression (Figure 3). They have substantially reduced viability (data not shown), while the surviving males are completely sterile (Figure 3). To exclude possible background effects, we combined the two null alleles with three different third chromosome deficiencies that remove small parts of the third chromosome including orb2 (Df(3L)ED4421, Df(3L)ED4415, and Df(3L)ED4416). These trans combinations also have reduced viability and are completely male sterile (Figure 3C; not shown). Similar results were obtained for an independently generated null allele, orb2Δ [14]. Since all null alleles behave the same in our assays, we used orb236 in the experiments described below.
Overall testes morphology and the pre-meiotic stages of spermatogenesis appear normal in orb236 and other orb2 mutants. The spermatogonia undergo the sequential mitotic divisions generating 16 interconnected spermatocytes, and the spermatocytes mature as in wild type. However, subsequent stages of spermatogenesis are abnormal. In wild type, the products of meiosis, the spermatids in the 64 cell cysts, have two characteristic spherical structures when observed by phase contrast microscopy: a light nucleus and a dark mitochondrial Nebenkern (Figure 4A). While pseudo-spermatids are present in orb236, the cells and their nuclei are unusually large and they have a poorly contrasted Neberken, which is abnormally shaped and sometimes fragmented (Figure 4B). As the overall DNA content is also increased (Figure 4C, 4D), it seems likely that the orb2 spermatids have replicated their DNA as in wild type, but failed to complete meiotic divisions. Consistent with this possibility, we never observe products of the first and second meiotic divisions, the 32 and 64 cell cysts respectively, in orb236 testes. By contrast, 32 and 64 cell cysts are seen in wild type.
To further characterize the meiotic defects, we examined chromosome morphology. During the prolonged G2 before the spermatocytes enter meiosis I, the three large chromosomes segregate into 3 domains and start the process of condensation. As illustrated in Figure 4I, the spermatocyte chromosomes initially coalesce into irregular rod-like structures located at vertices of a triangle (Figure 4E). They subsequently condense into 3 sharp dots (Figure 4F) before congressing to the metaphase plate in preparation for the first meiotic division (Figure 4G) [31]. In orb2, the spermatocyte chromosomes segregate into three domains, and start the process of condensation. However, condensation is incomplete and the chromosomes don't congress to the metaphase plate (Figure 4H).
These findings suggest that orb2 spermatocytes arrest meiosis at a step prior to the first meiotic division. To analyze the meiotic arrest further we examined Cyclin A accumulation. In wild type testes, Cyclin A accumulates in the cytoplasm during G2. However, just prior to the meiosis I G2 to M transition, Cyclin A is targeted to the spermatocyte nucleus, and then quickly degraded as meiosis proceeds [31], [32]. Since nuclear localization is only transient, cysts with nuclear Cyclin A are rarely seen in wild type (Figure 5A). However, in orb236 and orb236/Df(3L)4416, most cysts in the middle of the testes have high levels of nuclear Cyclin A (Figure 5B).
These orb2 meiotic phenotypes are similar to the phenotypes reported for mutations in boule (bol) and off-schedule (ofs) [32]–[34]. bol encodes a homolog of mammalian DAZ fertility factor, while ofs encodes a testes eIF4G. Like orb2, bol and ofs mutant spermatocytes arrest meiosis prior to the first meiotic division and the cysts have high levels of nuclear Cyclin A. The fact that all three proteins are needed for meiosis suggested that they might function together. To explore this possibility, we first tested whether Orb2 and Bol associate with each other in testes extracts. As shown in Figure 5G, Orb2 and Bol are in an RNase resistant immunoprecipitable complex.
We also examined the pattern of Bol accumulation in orb236 testes. In wild type spermatocytes, Bol localizes in a perinucleolar dot during spermatocyte maturation; however, once meiosis begins, Bol is relocalized to the cytoplasm where it is thought to promote the translation of target mRNAs [35]. Figure S3A, S3A′, S3B, and S3B′ show that both phases of Bol localization are observed in orb2 mutant testes. Also as in wild type, Bol is present in “post-meiotic” (see below) orb236 spermatids even though they haven't undergone meiosis (Figure S3C, S3C′, S3D and S3D′). As for Ofs, we were unable to demonstrate an association with Orb2 in testes extracts (Xu: unpublished data).
One reason that bol and ofs mutants are blocked in meiosis at the G2/M transition is that both factors are required for translation of twine (twe) mRNA [33], [34], [36]. twe encodes Drosophila Cdc25 phosphatase. In order for meiosis to proceed twe must remove an inhibitory phosphorylation on tyrosine 15 of Cdc2 (Ck1) [37], [38]. In bol testes, twe mRNA is present but it is not translated. In the absence of Twe protein, phosphorylated Cdc2 on Tyr15 accumulates and meiosis arrests at the G2/M transition [38]. Since our results indicate that orb2 also arrests meiosis at the G2/M transition, we anticipated that orb2 activity would be required to translate twe mRNA. To test this hypothesis, we first determined whether twe mRNA levels are normal. The RT-PCR experiment in Figure 5I shows that twe mRNA levels in orb2 testes are similar to wild type. We next used a chimeric twe-lacZ translational reporter to ascertain whether twe mRNA is translated in orb2 mutants. The reporter has sequences encoding β-galactosidase inserted in frame into the twe gene and expresses a chimeric mRNA including the twe 3′ UTR [36]. While we anticipated that the translation of the chimeric twe-lacZ mRNA would be blocked in orb2 testes as in bol (and ofs), this is not the case. Instead, Twe-lacZ expression in orb2 exceeds even wild type.
Figure 5E and 5F show that the pattern of Twe-lacZ expression differs in several respects from wild type. First, compared to wild type (E) there are many more cysts in orb2 testes (F) that express Twe-lacZ. Second, the amount of lacZ is typically much higher than in wild type (compare purple arrows in E and F). Third, while residual Twe-lacZ is degraded in wild type once meiosis is complete and the spermatids begin differentiation, it persists in elongating orb36 spermatids (green arrow in Figure 5F). Finally, we sometimes observe that Twe-lacZ is precociously expressed in immature spermatocytes that normally would not have Twe protein (orange arrows in Figure 5F).
Meiosis arrests at the G2-M transition in bol mutants because CDC2 remains phosphorylated on Tyr15 in the absence of Twe [37], [38]. This should not be the case in orb2 because high levels of Twe-lacZ and presumably Twe accumulate. To confirm this prediction we compared CDC2 Tyr15-P in wild type, bol and orb2 testes. As expected the ratio of phosphorylated to unphosphorylated CDC2 is elevated in bol mutants compared to wild type, while it is reduced in orb2 (Figure 5H). This finding indicates that CDC2 is activated in orb2 mutants and that meiosis I must be blocked at a subsequent step in the G2-M transition.
To further pinpoint the meiosis block we examined the expression of Cyclin B (Cyclin B). In wild type testes Cyclin B is expressed in primary spermatocytes when chromosome condensation starts. It persists during metaphase and is abruptly degraded at the beginning of anaphase [28]. Like Cyclin A, Cyclin B's transient nuclear accumulation is seen only very infrequently. Previous studies have shown that the upregulation of Cyclin B expression during chromosome condensation doesn't occur in ofs mutants. But other than that, Cyclin B expression and degradation seem normal [33], [34]. In bol testes, Cyclin B is found in the cytoplasm (Xu, unpublished data). In orb2 mutant testes, Cyclin B initially accumulates in the cytoplasm as in wild type (Figure 5C, 5D, arrow). However, instead of transiently accumulating in the nuclei and then disappearing, we find many orb2 cysts with high levels of nuclear Cyclin B (Figure 5D, arrowhead). In older orb2 cysts we often observe many small Cyclin B speckles in the cytoplasm. Taken together with the effects on twe expression and CDC2 phosphorylation, these findings place the meiosis arrest in orb2 at a step later than in bol and ofs.
Even though orb2 spermatocytes fail to undergo meiosis, the spermatids in the older cysts eventually exit the meiotic cycle and begin the process of differentiation. One of the first steps in differentiation is the elongation of the flagellar axonemes. In wild type, the elongating bundle of flagellar axonemes extends towards the apical tip in a roughly straight and smooth line (Figure 6A). In contrast, the elongating flagellar axonemes in orb2 zigzag back and forth and are much shorter than wild type. The individual axonemes also often splay out from each other instead of remaining in a tight bundle (Figure 6B). In addition, rather than having a smooth, regular internal morphology, their internal morphology is rough and irregular. This phenotype likely arises from underlying defects in the assembly or localization of axonemal proteins. One protein that is not properly localized is the meiosis regulator Bol. In wild type, Bol co-localizes with the prominent Orb2 band near the tip of the elongating flagellar axoneme bundle. In the region distal to this band extending towards the spermatid nuclei, there is a lower level of Bol and Orb2 and both are distributed uniformly along the individual axonemes (Figure 6C1–6C3). In orb2 testes, the prominent Bol band at the tip of the axoneme is missing, while in the remainder of the axoneme bundle, Bol is dispersed in an irregular fashion, and unlike wild type, its association with individual axonemes is difficult to discern (Figure 6D1–6D3).
At the end of meiosis just as spermatid elongation commences, the 64 spermatid nuclei cluster together and begin the process of condensation, eventually forming a cap-like structure (Figure 7A, arrowhead) [39]. This doesn't happen in orb236, and instead of coalescing into a tight bundle, the spermatid nuclei usually end up spread out along the partially elongated flagella axonemes (Figure 7B). The process of individualization begins once elongation is complete. In wild type testes, individualization is accomplished by a special structure called the Individualization Complex (IC). The IC is comprised of 64 individual actin cones that assemble around each nucleus in the condensed spermatid nuclear bundle (Figure 7C, inset) and then travels down the bundled axonemes, ensheathing each in a plasma membrane and pushing the excess cytoplasm into a waste bag [40], [41]. The IC is never assembled in orb236 testes and individualization never takes place. However, we do observe scattered triangular shaped actin cones (Figure 7D, inset). Based on the observed defects, the steps involved in organizing actin filaments into individual actin cones might be comparatively normal, while the subsequent assembly of the cones into the larger IC ensemble is not. Consistent with this idea, Myosin VI, a component of the Actin cone [39], is present in the orb236 cones (Figure 7E1, 7E2, 7F1, 7F2). As the bundled and condensed spermatid nuclei are believed to provide the scaffolding for assembling the IC [41], the defects in orb236 could be due to the failure in spermatid nuclei bundling and condensation. Alternatively or in addition, orb2 may be regulating genes directly involved in assembling the IC. As might be expected from the failure in IC assembly, Don-Juan GFP is not expressed in orb236 testes (Figure 7C, 7D) and mature sperm are never observed (Figure 6A, 6B).
orb mRNAs are expressed after meiosis is complete and localize to the tip of the elongating axoneme close to the band of Orb2 protein [3], [26]; however, these localized mRNAs don't appear to be translated until Orb2 protein begins to disappear at the end of the elongation phase (Figure 2A–2D). These observations suggested that the localized orb mRNAs might be a target of Orb2 repression. To test this hypothesis, we first probed Western blots of wild type and orb2 testes extracts with Orb antibodies. Figure 8A shows that Orb levels are elevated in orb2 mutants. In addition to this increase in Orb protein, orb mRNA translation appears to be ‘prematurely’ activated in orb2 mutant spermatids. As shown in Figure 2E and 2F, Orb protein is expressed in incompletely elongated orb2 mutant spermatids. Finally, the expression of Orb protein is not properly restricted to the tip of the elongated flagellar axoneme as in wild type. Instead, Orb is found throughout the mutant spermatid axonemes.
The orb mRNAs in the two sexes differ at their 5′ and 3′ ends. The male transcripts begin at an internal promoter and encode a protein that has a different N-terminus from the female Orb. At the 3′ end, the male UTR is only about 200 bases in length, while the female UTR is over a thousand [3]. While the male 3′UTR lacks most of the critical sequences for orb mRNA localization and translational regulation in ovaries, there are two CPE elements. Thus, it seemed possible that orb2 might repress orb mRNA translation by a mechanism that involves an association between Orb2 protein and orb mRNA. To test this idea we reverse transcribed RNA isolated from Orb2 immunoprecipitates of wild type and orb236 testes extracts, and then used primers specific for the orb male 3′ UTR for PCR amplification. Figure 8B shows that orb mRNA is readily detected in the Orb2 immunoprecipitates from wild type but not orb236 testes. In control experiments (not shown), neither boule nor twine mRNA was found in Orb2 immunoprecipitates. Taken together, these findings are consistent with the idea that Orb2 represses orb mRNA translation directly, rather than by regulating some other intermediate.
More than twenty other mRNAs are transcribed post-meiotically and localize to the tip of the elongating spermatid flagellar axonemes [26]. In addition to having similar expression and localization patterns to orb several of these mRNAs have CPE-like elements in their 3′ UTRs and could be regulatory targets of orb2. Consistent with this possibility we found that two of the CPE containing mRNAs, scotti and f-cup, can be immunoprecipitated with Orb2 antibody from wild type but not orb2 mutant testes (Figure S4). While the function of f-cup is unknown, Barreau et al. [26] found that scotti mutant males are sterile. The primary defect appears to be at a late step in spermatogenesis and involves the assembly or maintenance of the IC structure.
Although the experiments above show that orb2 is required for spermatid differentiation, it could be argued that the differentiation defects are the indirect consequence of the failure to undergo meiosis rather than because orb2 has special functions in this stage of spermatogenesis. To address this problem, at least in part, we took advantage of the hypomorphic orb2ΔQ allele, which has a small deletion that removes an N-terminal poly-Q domain (Figure S2A, Figure 3) [14]. As shown in Figure 3B, both the large isoform and the smaller, testes specific isoform are abundantly expressed in orb2ΔQ testes; however, they migrate more rapidly than the corresponding wild type isoforms due to the loss of the poly-Q domain.
We examined spermatogenesis in orb2ΔQ homozygous flies. In contrast to the mutants that reduce or eliminate expression of the 75 kD isoform, there are no meiosis defects in orb2ΔQ. Instead, like wild type, 32- and 64-cell cysts are observed, and each of the spermatids in the 64-cell cysts has a normal looking nucleus (Figure 9C) and Nebenkern (not shown). Also unlike orb236, elongating orb2ΔQ flagellar axonemes have a seemingly normal morphology and as in wild type the mutant Orb2 protein and Bol accumulate together in a prominent band near the growing tip of the flagellar axoneme (Figure 6E1–6E3). Likewise the assembly of the spermatid nuclei into a bundle and their condensation appear to be normal (not shown).
On the other hand, the process of differentiation is not normal in orb2ΔQ. In wild type testes, elongation of the flagellar axoneme stops before the tail reaches spermatogonia region (Figure 9D, arrow points to end of elongation). This is not true in orb2ΔQ. About 70% of the mutant testes have over-elongated flagellar axonemes that extend into the spermatogonia region. Elongation doesn't seem to arrest even at this point. Overgrowth of spermatid axoneme results in the swelling of testes tip region and an over-sized testes tip is often observed in orb2ΔQ testes (Figure 9E, compare d in Figure 9E and d′ in Figure 9D). In some cases, the elongating flagellar axonemes push against the testes wall and cause the muscle layer encasing the apical tip of the testes to rupture (not shown). On other occasions, when the flagellar axoneme bundle reaches the spermatogonia region, it changes direction and begins elongating towards the side of the testes or even reverses direction and elongates towards the base of the testes (Figure 9F).
Another differentiation defect is in the assembly and functioning of the IC. While fully elongated cysts with scattered actin cones are occasionally observed in wild type testes (∼6%), 35% of the fully elongated cysts in orb2ΔQ testes have scattered actin cones (Figure 9A, 9G, 9H). The IC defects range from actin cones that are not fully coalesced into the IC structure (Figure 9G, arrow) to completely dispersed actin cones (Figure 9H, arrows). These IC phenotypes resemble the phenotypes reported for scotti [26]. In the testes that have IC defects, there is always a mixture of both wild type and defective ICs (Figure 9H arrow and arrowhead), which may explain why orb2ΔQ males are still fertile. Also by comparison, all ICs in orb236 testes are defective. There is also a reduction in the number of ICs in orb2ΔQ testes. In wild type flies, over 90% of the testes have more than 19 ICs. In contrast, orb2ΔQ testes, 44% testes have less than 19 IC (Figure 9B). The fact that meiosis is normal in orb2ΔQ, but there are clear defects in both spermatid elongation and individualization, would provide further support for the idea that orb2 activity is required not only for meiosis, but also for proper differentiation.
Although CPEB family proteins play critical roles in germline development in many species, their germline functions differ between proteins within an organism and also between proteins in different organisms. For example, in C. elegans, Fog-1 and the Orb2-like CPB-1 function in the male germline and are required for sex determination and meiosis respectively. A third, Orb-like CPEB, CPB-3 is required for meiosis in females [42]–[44]. Similar functional specializations are evident for orb and orb2. While orb is essential for oogenesis, it is not absolutely required for spermatogenesis as orb mutant males produce functional sperm and their fertility is reduced but not eliminated. The opposite sex specificity is exhibited by orb2. Though genetic interaction studies (suppression of orb haploinsufficiency in the gurken dorsal-ventral polarity pathway: see for example [17], [55]) suggest that orb2 may negatively regulate orb in the ovary, orb2 females are fertile and oogenesis appear to be comparatively normal (Nathaniel Hafer, PhD thesis). In contrast, orb2 plays an essential role in the male germline, and is required for programming the orderly progression of spermatogenesis from meiosis through differentiation.
How CPEB proteins regulate meiotic progression is best understood in Xenopus oocytes. During oocyte maturation, CPEB1 acts as a repressor, blocking translation of mRNAs containing CPE motifs. However, after progesterone stimulation, CPEB1 is converted into an activator by Aurora kinase phosphorylation, initiating translation by stimulating the Gld-2 dependent polyadenylation of target mRNAs. Amongst the targets are mRNAs encoding Mos and the Cyclins B2 and B5. These cyclins activate Maturation Promoting Factor (MPF) which mediates entry into metaphase I. Although CPEB1 is degraded during metaphase I, it induces expression of CPEB4, which is a member of the second CPEB family. CPEB4 subsequently controls the transition to metaphase II by regulating Cyclins B1 and B4 expression [45]–[49]. Interestingly, though mouse CPEB1 is also essential for meiosis in both sexes, it controls meiosis at an earlier step by regulating mRNAs encoding synaptonemal complex proteins [50].
Since there is no recombination in Drosophila males, the function(s) of orb2 in meiosis are necessarily different from those of mouse CPEB1 [48]. Additionally, its role is distinct from that of Xenopus CPEB1. While Xenopus CPEB1 promotes meiotic progression by activating translation of Cyclin B mRNAs, orb2 pre-meiotic cysts accumulate high levels of Cyclin B. orb2 also differs from the fly translation factors ofs and bol. The meiotic phenotypes of mutations in these two genes suggest that they regulate different targets and likely function at earlier steps in meiotic progression than orb2. Unlike orb2 mutants, Cyclin B levels aren't properly upregulated during G2 in ofs mutants. However, it is not clear whether ofs regulates Cyclin B mRNA translation directly, or whether the defects are an indirect consequence of incomplete spermatocyte maturation [33], [34]. bol seems to function at a step after ofs, controlling the onset of metaphase I by activating twe mRNA translation. In bol mutants Twe is not expressed and meiotic progression is blocked because CDC2 remains phosphorylated and inactive. orb2 mutations have a very different effect on Twe. First, Twe is precociously expressed in cysts containing spermatocytes that have not fully matured. Second, very high levels of Twe accumulate in mature cysts that are arrested prior to metaphase I. Moreover, as would be expected, a substantial fraction of Cdc2 in orb2 testes is dephosphorylated. Finally, Twe persists in differentiating spermatids. These phenotypes, together with the high levels of the A and B Cyclins, argue that orb2 regulates meiotic progression at a step that is likely later than either ofs or bol. Additionally, these findings indicate that meiotic progression in male flies does not depend upon a single critical step or “switch” such as turning on twe or cyclin mRNA translation. Rather, it would appear that multiple steps in meiotic progression are subject to translational regulation, and that these steps are controlled by different translation factors.
One simple model for Twe (Twe-LacZ) misexpression is that orb2 represses the translation of twe mRNA, perhaps by antagonizing Bol dependent activation. However, there are complications with this model. For example, the high levels of Twe-LacZ that accumulate in cysts arrested before metaphase I could be the consequence of a prolonged arrest at a point after Bol activation of twe translation rather than a failure to repress twe mRNA translation. While an indirect effect of this type would not explain why Twe-LacZ is precociously expressed in immature orb2 spermatocytes, we were unable to demonstrate an association between Orb2 and twe mRNA. Additionally, twe 3′ UTR doesn't contain any obvious CPE-like recognition sequences. With the caveat that these are negative results, an alternative possibility is that the effects on Twe-LacZ expression are indirect.
The onset of spermatid differentiation in wild type normally proceeds only after the completion of meiosis. However, as is seen for twe, ofs and bol, differentiation becomes uncoupled from meiotic progression and the mutant cysts ultimately exit the pre-metaphase I arrest and begin the process of spermatid differentiation [32]–[34]. In all of these mutants the differentiation process is abnormal, with some steps being initiated, but not properly executed, while other steps are not even initiated. One of the key events in spermatid differentiation is the elongation of the flagellar axoneme. Little or no elongation is evident for bol, while twe and ofs spermatids begin elongating but quickly abort [32]–[34]. While the spermatid flagellar axonemes elongate in orb2 mutants, the axonemes don't extend straight back towards the stem cells at the tip of the testes, but instead zigzag irregularly and prematurely halt elongation. They also have an abnormal internal morphology and though they express Bol, they lack the prominent Bol band, which in wild type testes co-localizes with the Orb2 band near the tip of the elongating axonemes. Since Bol is essential for elongation, the absence of the Bol band is likely to be relevant to the elongation defects in orb2. While we didn't detect any association between Orb2 and bol mRNA, RNA independent Orb2-Bol proteins complexes are found in testes extracts. Thus, a plausible idea is that localization of Bol to the axonemal band is mediated by interactions with Orb2.
Once elongation is completed in wild type, the spermatid nuclei condense and coalesce into a nuclear bundle and this structure provides a scaffold for assembling the IC. In orb2 the spermatid nuclei don't properly condense and never coalesce into a tight nuclear bundle. Though the process of IC assembly is initiated and actin cones are generated, a complete IC is never formed. The individualization marker Don Juan is also not expressed in orb2 testes. Interestingly, though spermatid differentiation appears to be much less complete in ofs than in orb2, Don Juan is expressed in ofs testes [33].
An important question is whether the defects in differentiation evident in orb2 testes reflect functions for orb2 during this stage of spermatogenesis or are the indirect and perhaps non-specific consequence of the earlier meiotic arrest. Arguing against the later possibility is the fact that ofs, bol, and orb2 mutants have quite distinct differentiation phenotypes, yet all three fail to undergo meiosis. In the case of orb2, other lines of evidence point to functions at specific steps in differentiation. First, orb2 appears to be required for repressing the post-meiotic expression of Orb until after spermatid elongation is complete. In wild type, the orb gene is transcribed post-meiotically, but orb mRNA is not translated until after spermatid elongation is nearly complete. Since the timing of orb mRNA translation is correlated with the disappearance of Orb2, a plausible idea is that Orb2 represses orb mRNA translation. Consistent with this hypothesis, the levels of Orb protein are elevated in orb2 mutant testes, and it is expressed prematurely in incompletely elongated spermatids. In addition, instead of being expressed only at the tip of the flagellar axonemes, Orb is distributed all along the axonemes. As orb mRNA contains two CPE elements and can be detected readily in Orb2 immunoprecipitates, it seems possible that Orb2 could directly repress orb mRNA translation. As noted above, a role in repressing orb mRNA translation is also suggested by genetic interaction studies in females (Nathaniel Hafer, PhD thesis). Second, orb mRNA does not seem to be the only post-meiotic orb2 regulatory target. We found that scotti and f-cup, which are also expressed after meiosis and thought to encode proteins involved in differentiation, are found in Orb2 immunoprecipitates of testes extracts. Moreover, there could be additional targets besides these three mRNAs. Several of the other post-meiotically expressed genes have CPE-like elements in their 3′ UTRs [26]. Similarly, the mRNA encoding gld2 poly(A) polymerase, which is thought to be an Orb co-factor, also has a CPE-like element in its 3′ UTR and resembles Orb in that Gld2 protein preferentially accumulates near the tip of elongated flagellar axonemes [51]. Third, the hypomorphic poly Q deletion mutant, orb2ΔQ, makes it possible to separate meiotic arrest from at least some steps in differentiation. Meiosis appears to be completely unaffected by the ΔQ mutation; however, as is seen for orb236 there are defects in both flagellar axoneme elongation and IC assembly. On the other hand, since the differentiation defects in orb2ΔQ are much less severe than those in the null, the possibility remains open that the failure in meiosis interferes with some process(es) critical for proper differentiation. For example, the defects in chromosome condensation and spermatid nuclear bundle formation could be due to the fact that the orb2 spermatid nuclei have a large excess of DNA. In turn it could be argued that the failure in IC assembly in orb2 is due to the absence of a coalesced spermatid nuclear bundle. However, the fact that IC assembly is also defective in orb2ΔQ would argue that orb2 must have IC specific functions that are independent of any IC assembly steps that require completion of meiosis. Consistent with this possibility, mRNAs encoding Scotti, which has also been implicated in IC function, are found in orb2 immunoprecipitates.
Finally, our studies provide some insights into the functional properties of the N-terminal region of the Orb2 protein. First, the very modest phenotypes observed not only in the soma [14], [19] but also in the male germline for orb2ΔQ suggest that the prion forming poly-Q domain, which is present in both the 75 kD and the 60 kD isoforms [10], [11], is dispensable for most orb2 functions. Second, even though the testes differ from the soma in that there are readily detectable levels of the 60 kD isoform, it is not clear what function if any this isoform has in spermatogenesis. In the insertion mutants that reduce expression of the 75 kD isoform there are even higher levels than normal of the 60 kD isoform, yet these mutants exhibit meiotic and differentiation defects that resemble those seen for the orb2 deletions. Though their phenotypes appear less severe than the deletion mutants, this could be attributed to the fact that all express some residual 75 kD protein. Third, the 162 N-terminal sequence that is unique to the 75 kD isoform is critical for orb2 function in programming the orderly development of the male germline from meiosis through the process of spermatid differentiation. Since there is little if any of the 60 kD isoform in somatic tissues, it isn't certain at this point whether the smaller isoform would be able substitute for the 75 kD in the soma. Additional tools will be required to determine whether the smaller isoform has any role in spermatogenesis and also to further dissect how orb2 functions at different points in meiosis and differentiation.
We obtained P-element/Piggybac insertion (f01556, c06090, e01925, d01793, f04965) from the Exelexis collection maintained at Harvard [29]. Deficiency stocks Df(3L)ED4421, Df(3L)ED4415, Df(3L)ED4416 and the dj-GFP stock were obtained from the Drosophila stock center (Bloomington). bol1 is a kind gift from Steven Wasserman [37]. orb2Δ and orb2ΔQ are provided by Barry Dickson [14]. All twine-lacZ flies, twine, aly, sa, can, and mia mutants are kindly provided by Minx Fuller (Stanford).
20 individual males were placed with two w1118 females each in food vials for 5 days, after which adults were removed. Presence of larvae, pupae and adults were examined after another 2 weeks. Those with presence of larvae are considered fertile.
piggyBack (pBac) transposon insertions with FRT sites near the orb2 gene used to generate orb2 null alleles are: FRT1 (f01556), FRT2 (d01925), FRT3 (f04965) (Figure S2). The FRT sites are used in combination with FLP to create targeted deletions of genomic DNA (method as described in [29], [30]). Deletions were confirmed using PCR primers specific for pBac sequences flanking the deletion site and within the gene region. We recovered and established several independent deletion stocks from each transposon pair and they behave similarly. Experiments described here use deletions from f01556–f04965, which we named orb236. There are also deficiencies in the region that uncover the orb2 locus and have been mapped molecularly (Df(3L)ED4421, Df(3L)ED4415, Df(3L)ED4416). They behave the same when combined with orb236. In the text, Df(3L)ED4416 is used and referred to as 4416.
Western blotting was essentially performed as in [19]. Antibodies used were as follows: mouse anti-Orb2 2D11 (1∶25), mouse anti-Orb2 4G8 (1∶25), mouse anti-Snf 4G3 (1∶2000), rabbit anti CDC2 (PSTAIR) (1∶2000, millipore), rabbit anti CDC2Tyr15 (IMG668) (1∶2000, IMGENEX), mouse anti-Orb 6H4 (1∶60), mouse anti-Orb 4H8 (1∶60) [3], [4], goat anti-mouse conjugated HRP (1∶1000- Jackson Immunoresearch). Blots were then washed 4×10 minutes in TBST and developed with ECL-plus (Amersham).
In situ hybridization was performed as described in [52]. Fluorescent antisense probes for orb2 were synthesized by Biosearch Technologies (www.biosearchtech.com). Forty non-overlapping 17 bp probes targeted at orb2 mRNA sequence from cctggacgatcagatgt to atatgttatttaatctcac were synthesized and labeled with Quasar 670 and used at 1∶100 dilution. Detection is done on an inverted Zeiss LSM510 confocal microscope.
Whole mount staining is performed as in [33]. Antibodies used were as follows: mouse anti-Orb2 2D11 and 4G8 IgG (undiluted), rabbit anti-Bol (1∶1000, a gift from Steven Wasserman), mouse anti-Myosin VI 3C7 1∶25 (a gift from Kathryn Miller), monoclonal anti-β-Tubulin E7 1∶50 (Developmental Studies Hybridoma Bank), monoclonal anti-Orb 6H4 and 4H8 1∶30. DNA was stained with Hoescht (1∶1000). Actin was stained with Alexa488-phalloidin, Alexa546-phalloidin (Invitrogen, Carlsbad, CA). Secondary antibodies used were goat anti-mouse IgG Alexa 488 or 546, goat anti-rabbit Alexa 488 or 546 (Molecular Probes, Inc.) Samples were mounted in Aqua-polymount on slides for an inverted Zeiss LSM510 confocal microscope. Testes live squash and phase contrast was performed as described in [41]. β-galactosidase activity assay was performed as described by [28].
Immunoprecipitation was performed essentially as described by [53], except the followings: crude monoclonal anti-Orb2 antibodies 2D11 and 4G8 were affinity purified with Orb2 coupled HiTrap NHS-activated HP column (GE healthcare) before used for immunoprecipitation; purified Orb2 antibodies were mixed with testis extract for 0.5 h–2 h at room temperature before protein-A/G beads (Calbiochem/Millipore) were added in; the mixture was then incubated at 4 C° for 2 h to overnight. Putative Orb2 target mRNAs with CPE binding sites were predicted using software described in [54]. RT-PCR was done according to [19].
Primers used were as follows:
orb2 common exon among RA,B,C,D:
CAACAGTGCCACCAGCAGTGC and GCGCAGACTAACTTCGTCGTT.
Cg5741: ATGAGCAAAGCTCCGTTGAAAGCC and TATCCGGATTAACCGTGTTCCGCA.
orb :CAAGCCCTTGACTCGCAACTCC and CTCCGCCATATTTCTACGTCGCCTAC
scotti: AAGAACCTCTCTTGGACCTCGGAA and AATGGGATGCATATCGGCTGGTTG
f-cup: AACCAGCTGAGCACTTTGCCCAAT and AGATGAACTGTGGCACATAGCCGA
Phosphorylation assay was done as in [55]. Testis were squashed in cold PBS and treated with λ protein phosphatase for 1 hour at 30°C followed by Western blots.
|
10.1371/journal.pgen.1004439 | Recombination Accelerates Adaptation on a Large-Scale Empirical Fitness Landscape in HIV-1 | Recombination has the potential to facilitate adaptation. In spite of the substantial body of theory on the impact of recombination on the evolutionary dynamics of adapting populations, empirical evidence to test these theories is still scarce. We examined the effect of recombination on adaptation on a large-scale empirical fitness landscape in HIV-1 based on in vitro fitness measurements. Our results indicate that recombination substantially increases the rate of adaptation under a wide range of parameter values for population size, mutation rate and recombination rate. The accelerating effect of recombination is stronger for intermediate mutation rates but increases in a monotonic way with the recombination rates and population sizes that we examined. We also found that both fitness effects of individual mutations and epistatic fitness interactions cause recombination to accelerate adaptation. The estimated epistasis in the adapting populations is significantly negative. Our results highlight the importance of recombination in the evolution of HIV-I.
| One of the most challenging issues in evolutionary biology concerns the question of why most organisms exchange genetic material with each other, e.g. during sexual reproduction. Gene shuffling can create genetic diversity that facilitates adaptation to new environments, but theory shows that this effect is highly dependent on how different genes interact in determining the fitness of an organism. Using a large data set of fitness values based on HIV-1, we provide evidence that shuffling of genetic material indeed raises the level of genetic diversity, and as a result accelerates adaptation. Our results also propose genetic shuffling as a mechanism utilized by HIV to accelerate the evolution of multi-drug-resistant strains.
| Recombination, here broadly defined as the shuffling of genetic material, is widespread in nature and occurs among a wide range of taxa, including most eukaryotes but also bacteria and viruses. It has long been believed that sex and recombination facilitate adaptation by increasing the genetic variance upon which natural selection can act [1]. However, recombination can also reduce variation if there is a preponderance of co-adapted allelic associations. This cost, referred to as recombination load, arises because recombination tends to unravel combinations of genes that are favored by selection, thus impeding adaptation [2]–[4]. Several hypotheses have been developed to account for the conditions under which recombination accelerates adaptation [4]–[8]. Epistasis-based hypotheses state that if beneficial mutations increase fitness less in combination than expected based on their individual effects (negative epistasis), recombination can accelerate adaptation by increasing genetic variance and thus enhancing the efficacy of selection [6], [7], [9]. In contrast, recombination is predicted to decelerate adaptation with positive as well as sign epistasis (i.e., when the direction of selection on an allele depends on the allelic status at other loci) [10]–[12]. According to another class of hypotheses, random genetic drift resulting from finite population size is believed to provide an advantage to recombination [4], [13]–[15]. This occurs because in finite populations, beneficial mutations are likely to occur on different backgrounds and compete with each other, thus reducing selection efficacy. Recombination alleviates this competition by bringing beneficial mutations together on the same background and consequently speeds up adaptation (the Fisher-Muller effect) [14]–[16].
Prior experimental studies have demonstrated that sex and recombination can facilitate adaptation [17]–[21]. Yet, our understanding of the costs and benefits of recombination during adaptation on empirical complex fitness landscapes is still limited. The structure of the underlying fitness landscape is a decisive factor for the effect of recombination on adaptation since it determines how natural selection creates non-random combinations of alleles. Although realistic fitness landscapes are believed to exhibit a complex structure characterized by intricate patterns of fitness interactions among genes, not much is known about the structure of large-scale fitness landscapes. This lack of knowledge hampered efforts to obtain insights into the effect of recombination on adaptation in realistic situations. To our knowledge, only one study explores the effect of recombination on an empirical fitness landscape [22], but this fitness landscape comprises only six loci, and thus, the results may not be generalizable to larger landscapes.
Recently, Hinkley et al. [23] have estimated a fitness landscape in HIV-1 for 1859 mutations, based on an in-vitro assay for viral replicative capacity. This empirical fitness landscape, by far the largest available empirical database characterizing epistatic interactions, allows us for the first time to scrutinize the impact of recombination on the evolutionary dynamics of an adapting population on a realistic fitness landscape. Through simulations on this fitness landscape, we examined the effect of recombination on adaptation under different conditions. We found benefits of recombination are sufficiently high to accelerate adaptation under a wide range of parameters. Our findings highlight the important evolutionary role of recombination in adaptation, and in particular, in HIV evolution.
Recombination was found to produce a substantial increase in the rate of adaptation (Figure 1A). This effect was robust with respect to the initial composition of the population, as it is observed not only when starting from the reference sequence but also when initializing the population with a random sequence (Figure S1 and Section S1 in Text S1). The acceleration of adaptation by recombination can be attributed to increased genetic variance in fitness, which in turn enhances the efficacy of natural selection, as proposed by the fundamental theorem of natural selection [14]. Indeed, we see a markedly stronger increase in genetic variance in fitness over time in the recombining, compared to the non-recombining population (Figure 1B). These results on population fitness are also in agreement with patterns of genetic diversity within the evolving populations: the recombining populations accumulate within-population diversity faster than the non-recombining populations (Figure S2A and Section S2 in Text S1), and they diverge faster from the initial sequence (Figure S2B and Section S2 in Text S1).
We also observed that with increasing recombination rate the fittest genotypes that emerged by the end of each simulation became more divergent from each other across replicate runs (Figure S3A and Section S3 in Text S1). However, it appears that this diversifying effect of recombination is primarily a consequence of the fact that recombination increases the rate at which the populations adapt and traverse the fitness landscape: when sequences are compared when the population reaches a certain threshold mean fitness, recombination has little effect on divergence (Figure S3B and Section S3 in Text S1). In other words, recombination accelerates adaptation but does not increase the number of evolutionary trajectories available to the populations in the course of adaptation.
We also considered how different drug treatments affect the impact of recombination on adaptation. To this end, we measured the effect of recombination using fitness landscapes obtained for 16 environments with different drug treatments. Our results indicate that the effect of recombination is markedly stronger in the presence of antiviral drugs compared to the drug-free environment (Figure S4 and Section S4 in Text S1), which appears to be due to the higher selection pressure in environments with drug treatments (results not shown).
We next explored how the population size and mutation rate affect the extent to which recombination accelerates adaptation. Figure 2A indicates that, for a given population size, adaptation is already accelerated with moderate recombination rates and this effect increases monotonically with increasing recombination rates. In contrast, the effect of recombination depends non-monotonically on mutation rate (Figure 2B): whereas modest mutation rates enhance the accelerating effect of recombination on adaptation, at very high mutation rates this effect is reduced (see Discussion). Finally, our results indicate that with increasing population size, the accelerating effect of recombination becomes stronger (Figure 2A and S5 and Section S5 in Text S1). This is because with increasing population size, more beneficial mutations co-segregate in the population on different backgrounds and as a result the Fisher-Muller effect is enhanced. We expect that in very large populations this acceleration would become weaker again (because then all combinations of beneficial alleles would be present in the population even without recombination), but in the range of population sizes that was computationally feasible this was not observed. We also found the effect of recombination for populations with the same population mutation rate (population size×mutation rate) to be dependent on the population size: the effect is maximized for intermediate values of the population mutation rate and this maximum occurs at higher values for larger populations (Figure S6 and Section S6 in Text S1). This is presumably because for larger populations a higher mutation rate is required for multiple beneficial mutations to occur on the same background and thereby mitigate the Fisher-Muller effect.
Our fitness model incorporates both main and pairwise epistatic effects, both of which are known to influence the effect of recombination on the rate of adaptation. To assess the relative contribution of these effects, we simulated adaptation on fitness landscapes where, starting from the original MHL fitness matrix, we decreased the epistatic and main effects by varying amounts. Figure 3 indicates that both main effects and epistatic effects contribute to the acceleration of adaptation and that in combination these two effects appear to operate additively. Similar results were obtained for smaller recombination rates (Figure S7). We also found that both main effects and epistatic interactions increase the rate of adaptation and the effect of recombination becomes stronger with increasing adaptation rate (results not shown). These findings suggest that both main and epistatic effects can enhance selection. This seems to result in stronger interferences between arisen beneficial mutations and therefore a higher advantage of recombination (see Discussion).
We next determined the predominant form of epistasis in the adapting populations. To infer epistasis, one can use the fitness values of a set of sequences that are chosen irrespective of the composition of the evolving population and that therefore may not represent the sequences formed during adaptation (‘physiological epistasis’). Alternatively, only fitness values of sequences that are present in the adapting population can be utilized to estimate epistasis. This form of epistasis, referred to as population epistasis, provides a real time estimate of the epistasis that is responsible for generating the standing linkage disequilibrium in the population, and is therefore more accurate than physiological epistasis (see Discussion). We calculated population epistasis by regressing log fitness against Hamming distance, i.e., the number of sites where the corresponding sites are different between two sequences. The regression was done for sequences that are present in the population at the end of simulation according to , where Hamming distance is measured relative to the reference sequence. The parameter , determining the curvature, is used as a measure of epistasis. Our results indicate that in the majority of simulations, population epistasis is significant (ANOVA test for comparison of a quadratic and a linear model, p≪0.001 for the simulations in Figure 4) and predominantly negative, indicating diminishing returns with each additional beneficial mutation in increasing fitness. Population epistasis in the recombining population becomes less negative on average than in the non-recombining population (Wilcoxon test for the significance of epistasis in Figure 4, p≪0.001 and for the significance of the difference between recombining and non-recombining simulations Figure 4B, p≪0.001). We obtained similar results for populations at other time points (results not shown).
Thus far, we considered the effect of recombination on adaptation by comparing evolving populations characterized by different recombination rates (including the absence of recombination). To examine whether recombination is selectively favored within an adapting population, we performed additional simulations in which we competed a resident non-recombining with an invading recombining genotype during adaptation. Figure S8A shows that the frequency of the recombining genotype gradually increases over time if the recombination rate is high enough . In accordance with our previous findings, the results of the invasion analysis demonstrate that the benefit of recombination is most pronounced for intermediate mutation rates (Figure S8B and S8C and Section S7 in Text S1).
Our results can be interpreted as support for the proposed accelerating role of recombination in the adaptive process through the Fisher-Muller effect. In our simulations, this effect seems to be sufficiently strong to outweigh potential costs of recombination. The Fisher-Muller effect is based on strong selective interference between beneficial mutations in an asexual (non-recombining) population. Previous mathematical models have provided important insights into how the strength of selection, mutation rate and population size affect selective interference and the Fisher-Muller effect, but these models have ignored epistatic interactions between mutations [24]–[30]. Our results demonstrate that both epistatic interactions and main fitness effects contribute to the accelerating effect of recombination. It is clear that this effect in the absence of epistatic interactions is due to the Fisher-Muller effect. Adding epistatic interactions to this model enhances the effect of recombination but the interpretation of why this occurs is challenging. On the one hand, this may be because the epistatic interactions increase the overall strength of selection and thereby produce stronger clonal interference, but on the other hand we cannot exclude explanations based purely on epistasis (see below).
One important determinant of the Fisher-Muller effect is the mutation rate. With a finite number of loci, an increasing mutation rate leads to a higher number of co-segregating beneficial mutations and this augments the Fisher-Muller effect. However, at very high mutation rates, it becomes increasingly likely that several beneficial mutations arise on the same genetic background so that recombination becomes less important (Figure 2B). A similar effect is expected for population size (with finite sites for beneficial mutations) but for the range of population sizes we examined here the effect of population size was monotonic.
In addition to the Fisher-Muller effect, recombination can also accelerate adaptation in the absence of random genetic drift when there is epistasis [9], [31]. Several studies have attempted to determine the prevailing form of epistasis in nature but have yielded inconsistent results: sometimes strong positive epistasis [32]–[36], negative epistasis [37], [38] or pervasive sign epistasis [39]–[42] was reported. The HIV-I fitness landscape is characterized by extensive epistatic interactions [23]. We demonstrated that population epistasis during adaptation is predominately negative. This result is in apparent contrast with the predominant positive epistasis in HIV-I sequences reported in Bonhoeffer et al. [33]. We think this discrepancy arises mainly because Bonhoeffer et al. estimated physiological epistasis, which is based only on the structure of fitness landscape itself and may be very different from population epistasis that we estimated here [43], [44]. The difference between the two estimates is that in determining population epistasis only the mutations that pass the sieve of natural selection are taken into account, whereas in measuring physiological epistasis all mutations are used indiscriminately. In addition to these different measures of epistasis, the two studies also differ in the way fitness was estimated. First, Bonhoeffer et al. [33] obtained fitness values from a much smaller data set than was used to estimate the fitness landscape in our study (9466 vs. 70,081 sequences). Second, in their study the main effects of a mutation and the epistatic effects for a given pair of mutations were calculated by averaging over the fitness effects of other mutations in the genetic background. By contrast, we obtained fitness using a predictive fitness model [23] that explicitly accounts for mutational effects in different genetic backgrounds during estimation of the fitness landscape, and therefore provides a more accurate estimate.
It is tempting to interpret the significant negative population epistasis that we observed as a support for the mechanism proposed by the mutational deterministic hypothesis, i.e. acceleration of adaptation through reduction of negative linkage disequilibria generated by negative epistasis. However, we would like to caution that it is very difficult to explain how population epistasis arises in our model and how it impacts the effect of recombination. For example, population epistasis on a complex fitness landscape can also be generated by variation in main fitness effects of mutations. The underlying causes of negative epistasis in our model and the extent to which it contributes to the accelerating effect of recombination (in isolation from the Fisher-Muller effect) is difficult to determine on a complex fitness landscape because deterministic models are not feasible. Therefore, we cannot exclude the possibility that stochastic benefits of recombination due to the Fisher-Muller effect may be sufficiently strong to override any direct effect of epistasis in our simulations, as reported by other studies [11], [45], [46].
Our study is based on an estimated fitness landscape and therefore the limitations of this approach should be taken into account while interpreting the results. First, in the fitness landscape that we used only main and pairwise epistatic effects were estimated but higher order fitness interactions (>2) were neglected. To accurately estimate higher orders epistasis, a much larger number of sequences with measured fitness values would be required. It is not clear what the strength of these higher-order interactions is and how they affect the impact of recombination. Second, although the accuracy of our fitness model (predicting 54.8% of the variance; see Methods) is acceptable as the only available fitness landscape, the predicted fitness landscape is yet to become more realistic by using a greater number of empirical fitness data. This is important as the estimated fitness values become increasingly unreliable for the regions in the sequence space far from the reference sequence due to the lack of data. To account for this problem, we focused on the population dynamics in the region of the fitness landscapes that is close to the reference sequence. Finally, the empirical fitness data used to predict the structure of the fitness landscape was obtained from an in vitro assay, and therefore might not completely correspond to in vivo fitness values.
It should also be noted that this study mainly examines the effect of recombination at the population level and does not address the evolution of recombination rate. One interesting extension of this work would be to incorporate variation in recombination rates in the model and study the spread of a recombination modifier gene in a non-recombining population. One problem with using the modifier approach in a realistic way with the current fitness model is that probably the best candidate for a recombination modifier gene in the HIV-1 genome is the reverse transcriptase gene [see 47], which is itself part of the fitness landscape and therefore changes during adaptation because of direct selection.
Our study relates to the debate over the advantage of recombination in retroviral, and in particular HIV, evolution. Recombination is believed by some to be beneficial because it generates genetic diversity to facilitate the development of multidrug resistance [48]–[50] or escape from host immune reaction [51]. Nonetheless, some studies have suggested that recombination has not evolved to facilitate adaptation but is a mere by-product of other mechanisms such as genomic organization [reviewed in 52], [ see also 45]. Unlike some prior studies [53]–[55], our model does not include any specific feature of HIV biology, such as viral dynamics during infection or specificities of recombination in HIV. Nonetheless, we believe that our findings are generic enough to highlight the potential role of recombination in accelerating HIV evolution.
This study utilizes data derived from a high-throughput fitness assay to address one of the long-standing questions in evolutionary biology. The advent of systems biology approaches made it possible to obtain a comprehensive picture of a large-scale fitness landscape. This serves as a framework for us to demonstrate that recombination has a substantial accelerating effect on adaptation on a realistic complex fitness landscape.
Our model is based on a recently estimated fitness landscape of HIV-1 [23]. Briefly, to obtain this fitness landscape, the in vitro replicative capacity of 70,081 samples from HIV-1 subtype B infected individuals were measured and the corresponding amino acid sequences of the protease and partial sequences of the reverse transcriptase were obtained for all of these samples. This enabled estimation of the fitness effects of 1,857 single mutations and 257,536 pairs of mutations in these samples by fitting a fitness model to the data. This fitness model, as detailed in Hinkley et al. [23], invokes a generalized kernel ridge regression (GKRR) method to estimate the fitness effect of individual amino acid variants and the epistatic effects between variants.
Based on these results, we used the following fitness model to obtain fitness values for a given sequence:
Here, the amino acid sequence x is a binary vector indicating the presence or absence of amino acid variants. M is a triangular matrix where an entry Mii on the diagonal determines the main effects of the amino acid variant i and the off-diagonal entries Mij (with ) determine pairwise epistatic effects between variant i and j.
Higher order epistatic interactions were not considered. Note that the original model in Hinkley et al. [23] also includes an intercept term that gives the log fitness of a reference sequence (NL4-3) and that is added to Equation (1). However, since natural selection only depends on relative fitness in our model, this term was not considered in our simulation setting.
We used two different types of matrices M determining fitness. The first matrix type, MRL, describing the ‘reference fitness landscape’, was obtained by Hinkley et al. [23] by estimating both main and epistatic effects simultaneously. This estimation was done in 16 different environments: one drug-free environment and 15 environments each characterized by the presence of a different antiretroviral drug. On average, this matrix predicts 54.8% of the variance in fitness across different environments. Unless stated otherwise, we use the fitness landscape in the drug-free environment as the reference fitness landscape in our simulations.
To obtain the second matrix type, MHL, describing a ‘hierarchical fitness landscape’, two fitting steps were performed [56]. In the first step, MHL was estimated by assuming that there are only main effects (all off-diagonal elements set to zero), and in the second step, epistatic effects were estimated by fitting the residuals under the assumption that main effects are absent. This fitness landscape was obtained only for the drug free environment. Since the main and epistatic effects are estimated separately for these fitness landscapes, this approach allowed us to generate fitness landscapes where we could scale the magnitudes of main and epistatic effects and thus evaluate their relative contribution with respect to the effects of recombination. Hierarchical fitness landscapes with different magnitudes of epistatic effects were shown to provide accurate predictions of the reference fitness landscape [56].
For details about the estimation procedures, we refer to Hinkley et al. [23] and Kouyos et al. [56].
We consider a discrete-time model based on the classic the Wright-Fisher model to simulate adaptive evolution under mutation, recombination and natural selection on the HIV-1 fitness landscape. The population consists of a constant number of amino acid sequences, each of which contains the protease, as well as a partial reverse transcriptase, gene. Initially, this population is monomorphic, consisting only of copies of a reference sequence (NL4-3).
In each generation, the new population is formed from the previous one through three steps. First, reproduction and selection are implemented through random sampling of sequences, weighted according to the relative fitness value of each sequence. Second, to implement mutation events, sequences are randomly chosen from the population with replacement ( denotes the per genome mutation rate), thus allowing for several mutations per sequence. For <1, this number is treated as a random number with mean . For each of these sequences, the allele at a randomly selected site for which there exists more than one possible variant is substituted with one of the other possible allelic variant. Amino acid variants at a given site that are not present in the data set used to estimate fitness are neglected. In the final step, selected sequences undergo homologous recombination. We denote the recombination rate by . Here, pairs of sequences are chosen randomly (without replacement) and for each of these pairs, a single crossover site at which the two parental sequences exchange genetic material is determined at random. Note that recombination may result in daughter sequences that are identical to the parental sequences if identical pieces are exchanged.
The simulations were run for 100 generations. This period is long enough for the population to adapt but the adapting population still remains in the proximity of the reference sequence, where due to the availability of empirical fitness data, the estimation of the fitness by our model is reliable. To examine the effect of recombination on adaptation, we computed the ratio of the logarithm (base 10) of the population mean fitness of a population evolving with recombination to that of a population evolving without recombination. In this case, finding a proper definition of error bars is not straightforward since the data in question are the logarithms of ratios. However, this logarithm can be written as a difference (), so that we can use the standard deviation of the difference between two log normal distributions, , as error bars. This is justified because we found the log fitness values of sequences at generation 100 across 100 simulations to be normally distributed (for instance, non-significant Shapiro-Wilk and Anderson-Darling test to reject normal distribution for results in Figure 1, with p>0.05).
|
10.1371/journal.ppat.1000460 | Mycobacterium tuberculosis Universal Stress Protein Rv2623 Regulates Bacillary Growth by ATP-Binding: Requirement for Establishing Chronic Persistent Infection | Tuberculous latency and reactivation play a significant role in the pathogenesis of tuberculosis, yet the mechanisms that regulate these processes remain unclear. The Mycobacterium tuberculosis universal stress protein (USP) homolog, rv2623, is among the most highly induced genes when the tubercle bacillus is subjected to hypoxia and nitrosative stress, conditions thought to promote latency. Induction of rv2623 also occurs when M. tuberculosis encounters conditions associated with growth arrest, such as the intracellular milieu of macrophages and in the lungs of mice with chronic tuberculosis. Therefore, we tested the hypothesis that Rv2623 regulates tuberculosis latency. We observed that an Rv2623-deficient mutant fails to establish chronic tuberculous infection in guinea pigs and mice, exhibiting a hypervirulence phenotype associated with increased bacterial burden and mortality. Consistent with this in vivo growth-regulatory role, constitutive overexpression of rv2623 attenuates mycobacterial growth in vitro. Biochemical analysis of purified Rv2623 suggested that this mycobacterial USP binds ATP, and the 2.9-Å-resolution crystal structure revealed that Rv2623 engages ATP in a novel nucleotide-binding pocket. Structure-guided mutagenesis yielded Rv2623 mutants with reduced ATP-binding capacity. Analysis of mycobacteria overexpressing these mutants revealed that the in vitro growth-inhibitory property of Rv2623 correlates with its ability to bind ATP. Together, the results indicate that i) M. tuberculosis Rv2623 regulates mycobacterial growth in vitro and in vivo, and ii) Rv2623 is required for the entry of the tubercle bacillus into the chronic phase of infection in the host; in addition, iii) Rv2623 binds ATP; and iv) the growth-regulatory attribute of this USP is dependent on its ATP-binding activity. We propose that Rv2623 may function as an ATP-dependent signaling intermediate in a pathway that promotes persistent infection.
| Mycobacterium tuberculosis poses serious threats to public health worldwide. The ability of this pathogen to establish in the host a clinically silent, persistent latent infection that can subsequently reactivate to cause diseases constitutes a major challenge in controlling tuberculosis. Our study showed that an M. tuberculosis mutant that is deficient in a universal stress protein (USP) designated Rv2623 fails to establish a chronic persistent infection in animal hosts. The mutant strain exhibits a hypervirulent phenotype as assessed by increased bacillary growth, pathology, and mortality in infected animals relative to the parental strain. Consistent with this in vivo growth-regulating attribute, we demonstrated that Rv2623, when expressed in mycobacteria at levels higher than that of the wild-type strain, retards bacterial growth in vitro. Using biochemical and biophysical analyses, including the Rv2623 crystal structure, we showed that this USP binds to ATP within a novel ATP-binding pocket. Through targeted mutagenesis studies, we further determined that the ability of Rv2623 to regulate bacillary growth is dependent on its ATP-binding capacity. Our data strongly suggest Rv2623 as a critical component that regulates the entry of M. tuberculosis into a chronic persistent growth phase, and therefore provide valuable insight into tuberculous dormancy and uncover new opportunities for the development of novel anti-tuberculous therapies.
| Mycobacterium tuberculosis, one of the most successful human pathogens, infects one-third of the world's population, causing nearly two million deaths per year [1]. Epidemiological data estimate that, in the immunocompetent host, only ∼10% of M. tuberculosis infection progress to active pulmonary disease. The remaining 90% of the infected individuals are asymptomatic, and are generally believed to harbor latent bacilli that can reactivate to cause tuberculous diseases, sometimes decades after the initial infection. Recrudescence of latent bacilli contributes significantly to the incidence of adult tuberculosis [2], yet the physiological state of latent bacilli and the signals that promote dormancy in the host remain incompletely defined. Understanding the dynamic interaction between host and pathogen during the establishment of persistent M. tuberculosis infection will guide the design of novel treatment for the latently infected population.
An intracellular pathogen, M. tuberculosis must possess a finely tuned signaling network to sense and transduce complex environmental signals, ensuring survival of the bacilli within host cells. Nitric oxide (NO) produced by infected macrophages and relative hypoxia are signals likely to be encountered within tuberculous lesions that are believed, based on in vitro studies, to promote latency by prompting the M. tuberculosis dormancy response. Exposure to these stimuli results in the induction of ∼50 M. tuberculosis genes, designated the dormancy regulon, via the two-component regulatory system DosR-DosS (see Table S1 for accession numbers) [3],[4],[5]. Among this set of genes is rv2623, one of eight M. tuberculosis genes annotated as containing the universal stress protein (USP) domain [6],[7]. Members of this ancient and conserved family of proteins are found in all forms of life and can be induced by a variety of environmental stresses [8],[9]. However, the roles of USP proteins in microbial pathogenesis are incompletely understood.
Interestingly, rv2623 is one of the most strongly induced transcripts of the dormancy regulon [3],[4],[5]. Increased expression of rv2623 was also observed following phagocytosis by macrophages [10] and in the lungs of chronically infected mice [11], supporting a functional role during persistent M. tuberculosis infection. The present study reveals that: i) deletion of rv2623 confers hypervirulence on the tubercle bacillus in animal models, suggesting that expression of Rv2623 may be conducive to the establishment of persistence in vivo; ii) overexpression of Rv2623 results in growth retardation of recipient strains in vitro, further supporting a growth-regulatory role; iii) Rv2623 binds ATP; and finally, through mutagenesis study guided by crystallographic analysis of Rv2623 (the first such study for a tandem-domain USP), we show that iv) the growth-regulating attribute of this M. tuberculosis USP is linked to its ATP-binding capacity.
An rv2623-deletion mutant of the virulent M. tuberculosis Erdman strain was generated by specialized transduction [12]. The rv2623-specific allelic exchange construct was delivered via recombinant mycobacteriophage phAE159 and transformants were analyzed by Southern blot, confirming replacement of rv2623 with the hyg gene, which confers hygromycin resistance (Figure 1A). Aliquots of a single knockout clone, designated as Δrv2623, were stored at −70°C. Deletion of rv2623 is not likely to affect transcription of neighboring genes, given the sequence-confirmed precise excision of the rv2623 coding region and the gene organization at the rv2623 locus (the downstream rv2624c is transcribed in the direction opposite to that of rv2623) (Figure 1B).
Deletion of specific USPs in E. coli results in growth defects in vitro [8],[13],[14]. For example, an E. coli strain deficient for UspA exhibits reduced survival in stationary phase culture [14]. However, the in vitro growth kinetics of Δrv2623 M. tuberculosis in OADC-supplemented Middlebrook 7H9 or minimal Sauton's medium is comparable to that of wildtype Erdman up to 14 days post-inoculation (Figure 2A). We reasoned that a potential growth-regulating attribute of Rv2623 might be masked by functional redundancy among the M. tuberculosis USP homologs. Indeed, partial functional overlap has been demonstrated among the E. coli USPs [9],[15]. We therefore examined the effect of overexpression of this USP in the rapidly growing M. smegmatis strain mc2155 [16]. As seen in Figure 2B, constitutive overexpression of M. tuberculosis rv2623 using the multi-copy plasmid pMV261 resulted in growth deficiency of the recipient strain both on solid medium (Middlebrook 7H10 agar) and in the liquid medium-based BD BACTEC 9000MB system. These results strongly suggest that M. tuberculosis Rv2623 regulates mycobacterial growth in vitro.
Although USP family proteins are expressed by many bacterial pathogens [7],[8], to date, there has only been one in vivo study, which showed that a Salmonella USP promotes virulence in mice [17]. The observation that Rv2623 modulates mycobacterial growth in vitro prompted us to examine the effect of this USP on the in vivo kinetics of M. tuberculosis infection. Low dose aerosol infection of outbred Hartley guinea pigs with ∼30 CFU revealed a clear growth advantage of the Δrv2623 mutant strain relative to wildtype. As early as 20 days post-infection, the number of M. tuberculosis bacilli present in the lungs of Δrv2623-infected guinea pigs was ∼10-fold higher (p<0.05) than those infected with wildtype Erdman, and continued to rise, attaining a 15-fold (p<0.001) difference by 60 days post-infection (Figure 3A). Guinea pigs are able to control the growth of Erdman bacilli following the onset of adaptive immunity at ∼3 weeks post-infection, as evident by the relatively stable pulmonary bacterial burden beyond the 3 week time point, yet levels of Δrv2623 bacilli continued to increase at a reduced but steady rate resulting in a rapidly progressing infection. Moreover, Δrv2623-infected guinea pigs were moribund at 60 days post-infection, while those challenged with wildtype Erdman remained relatively healthy, providing further evidence that the mutant strain is hypervirulent in this model. Finally, complementation with a single integrated copy of rv2623 expressed from a constitutive mycobacterial promoter (Δrv2623 attB::Phsp60Rv2623) abrogated the growth advantage of the deletion mutant (Figure 3A). Also consistent with the fulminate disease progression displayed by Δrv2623-infected guinea pigs are the more severe pathological changes observed as early as 20 days post-infection in the lungs of these animals, as assessed by histopathological studies, including the semi-quantitative Total Lung Score analysis (Figure 3B and Protocol S1). Overall, the progression of pulmonic lesions was accelerated in Δrv2623-infected animals compared to those infected with wildtype Erdman, accompanied by more extensive necrosis and widespread fibrosis. This increase in lung pathology was also largely reversed in animals infected with the complemented Δrv2623 attB::Phsp60Rv2623 strain (Figure 3B and C). Results of the complementation experiments were further validated using a complemented strain Δrv2623 attB::Prv2623Rv2623, whose expression of the wildtype universal stress protein is driven by the native rv2623 promoter [18] (Figure 3D and E).
In contrast to the result of the guinea pig study, we observed no difference in the kinetics of infection between C57BL/6 mice infected with wildtype M. tuberculosis, Δrv2623, or the attB::Phsp60 Rv2623 complemented strain in a low dose aerogenic model [19], as assessed by lung bacterial burden (Figure 4A). However, the mouse is a relatively resistant host to M. tuberculosis, particularly in strains such as C57BL/6 [20],[21]. In fact, evidence exists that M. tuberculosis triggers an immune response in mice that is in excess of that required for controlling the infection [22],[23]. Thus, the hypervirulence phenotype of Δrv2623 observed in the susceptible guinea pig model could have been masked in the C57BL/6 mice. Consequently, we examined the virulence of Δrv2623 in the relatively susceptible C3H/HeJ mouse strain [24]. Indeed, the Δrv2623 mutant was markedly more virulent relative to wildtype Erdman M. tuberculosis following aerogenic infection, as assessed by the mean survival time of C3H/HeJ mice infected with these strains (62 and 25.5 days post infection for Erdman- and Δrv2623-infected mice, respectively, p = 0.0014; Figure 4B). In agreement with the survival data, quantification of tissue bacterial burden revealed a growth advantage for the Rv2623-deficient mutant relative to wildtype M. tuberculosis Erdman (Figure 4C). Manifestation of this hypervirulence phenotype is apparent as early as 3 weeks post-infection, with the lung bacterial burden of mice infected with Δrv2623 M. tuberculosis ∼100 fold higher than that in the wildtype-infected animals. As in the guinea pig studies, results of complementation experiments involving the reintroduction of a single copy of wildtype rv2623 into Δrv2623 M.tuberculosis reverses the hypervirulence (Figure 4C) exhibited in the C3H/HeJ model, thus indicating that the observed growth phenotype of the tubercle bacillus deficient for the universal stress protein is rv2623-specific. Finally, survival of Δrv2623-infected mice was also significantly reduced in another susceptible mouse strain, C3HeB/FeJ (Figure S1). Together, the animal studies provide strong evidence that Rv2623 regulates the growth of M. tuberculosis in vivo: in the absence of Rv2623, the tubercle bacillus fails to establish a chronic persistent infection, exhibiting a hypervirulent phenotype.
Although the functions of universal stress proteins have yet to be completely defined, there is evidence that many USPs play differential roles in protecting microbes against various environmental stresses [9]. Therefore, the hypervirulence of Δrv2623 in guinea pigs and susceptible mice is intriguing; if Rv2623 provides M. tuberculosis protection against stress, it might be expected that the Rv2623-deficient mutant would be attenuated in vivo. The growth kinetics and survival of the Δrv2623 strain was examined under various stress conditions, including those likely to be present during M. tuberculosis infection. These included oxidative stress (superoxide anion, O2−), DNA damage (UV irradiation, mitomycin C), heat shock (53°C), and acidic culture (pH 4.0). The use of streptonigrin, an antibiotic whose toxicity correlates with levels of free iron, was based on the observation that the intracellular environment of macrophages can induce a iron-scavenging response in mycobacteria [25], perhaps as a means of maintaining adequate levels of this important growth factor, and that an E. coli USP was shown to regulate iron uptake [9]. The results showed that the mutant strain was no more susceptible to growth inhibition than was wild type Erdman under all of the stress conditions tested (Figure S2). These results support the notion that it is unlikely that M. tuberculosis Rv2623 is essential for resistance to stresses encountered in the host, which is consistent with the observed in vivo hypervirulence phenotype of Δrv2623.
We began a biochemical characterization of Rv2623 in order to gain insight into the relationship between the molecular structure/function of this USP and it's growth-regulatory properties. M. tuberculosis Rv2623 was expressed in E. coli and purified to homogeneity for biochemical studies. SDS-PAGE analysis of affinity-purified His6-Rv2623 revealed a single band that approximates the predicted molecular mass of ∼31.6 kDa, which was identified by immunoblotting as Rv2623 (Figure S3). Gel filtration analysis of native His6-Rv2623 revealed that the purified protein exists primarily as a single species with an apparent molecular mass of 61±1 kDa; suggesting that Rv2623 is a dimer under native conditions (Figure S3), an observation that was later confirmed using nano electrospray ionization (nano ESI) mass spectrometry (data not shown).
The nucleotide-binding capacity of a subset of USPs was discovered following the observation that MJ0577, a single-domain USP from Methanococcus jannaschii, co-purifies and co-crystallizes with ATP [26]. On the basis of structures of ATP-binding and non-ATP-binding USPs, a G-2X-G-9X-G(S/T) motif was suggested to be essential for the binding of ATP [27]. The presence of this motif in each of the two tandem USP domains of Rv2623 [7] raised the possibility that this protein possesses ATP binding activity. An HPLC-based examination of supernatants from boiled samples of His6-Rv2623 demonstrated that His6-Rv2623 co-purifies with both ATP and ADP (Figure 5). Analysis of E. coli-expressed Rv2623 using nano ESI mass spectrometry also demonstrated that an ATP-saturated form of dimeric Rv2623 (composed of 2 bound ATP molecules per monomer) constitutes at least half of the purified sample (data not shown). Measurement of the binding stoichiometry, which comprised HPLC-based quantification of adenine nucleotides from the boiled supernatant and spectral analysis of heat denatured Rv2623 following reconstitution in 6 M guanidine-HCl, yields 1.4±0.2 nucleotide equivalents/monomer with an overall content of 86±4% ATP (14±4% ADP). Thus, Rv2623 binds endogenous adenine nucleotides in E. coli, and the association is sufficiently tight that nearly 75% of the nucleotide binding sites are occupied upon purification. Indeed, nucleotide did not completely dissociate from the protein following an extensive, two-week dialysis with multiple changes against nucleotide-free buffer (approximately 0.3 nucleotide equivalents per monomer remain). It is conceivable that the presence of ADP is the consequence of an Rv2623-associated ATP activity and this putative ATPase function is currently under investigation.
To examine the biochemical mechanisms responsible for Rv2623 function, we determined the crystal structure of wild-type Rv2623 at a resolution of 2.9 Å. The structure reveals a compact, 2-fold symmetric dimer. Each monomer is composed of tandem USP domains [residues 6–154 (domain 1), 155–294 (domain2)] that share 26% sequence identity and significant structural homology (residues 6–154 and 155–294 comprise domains 1 and 2, respectively; interdomain rms = 2.04 Å for 140 equivalent Cα's). Individual domains, which consist of a twisted, five-stranded, parallel β sheet flanked by four α helices, unite through an antiparallel, cross-strand (β5–β10) interaction that produces a central dyad axis between β5/β10 and a continuous, ten-stranded, mixed β sheet in the complete monomer. Each domain possesses a pair of conserved βαβ motifs (domain 1: β1-L1-α1- β2, β4-L2-α4-β5; domain 2: β6-L3-α5-β7, β9-L4-α8-β10) that encompass four loops (designated L1–L4) responsible for ATP recognition (Figure 6A and C). A “U-shaped” ATP molecule that lies within a cleft near the monomer surface is stabilized by 1) a cluster of hydrophobic residues (I14, V41, H42, V116/132/261/277/281, L136, A175) that forge the adenine/ribose-binding scaffold, 2) a pair of conserved L1/L3 aspartates (D15-L1/D167-L3), and 3) small phosphoryl/ribosyl-binding residues within the G-2X-G-9X-G (S/T) motifs that comprise L2/L4 (G120/265/267/268 and S131/276) (Figures 6A,C and 7A). Dimerization of Rv2623 occurs along a 2-fold axis orthogonal to the intramonomer dyad and juxtaposes ATP binding pockets from opposing monomers (Figure 6B).
Phylogenomic analysis places Rv2623 in a Uniprot/TrEMBL family (Q5YVE7) of 370 tandem-domain USPs, and a 113-member subfamily (N631) that consists almost exclusively of actinobacterial representatives (Text S1). Structure-based sequence alignments of both Rv2623 domains with the N631 consensus suggest that domain 2, which exhibits significantly higher conservation than domain 1 across global and ATP-binding subfamily consensus sequences, represents the ancestral domain among ATP-binding USPs with tandem-type architectures. Interestingly, the domain fold and interdomain organization observed for Rv2623 is broadly conserved: these features are shared among single domain USP structures, both monomeric and dimeric, that are presently represented within the PDB. As this manuscript was under preparation, a second, lower resolution (3.2 Å) crystal form of Rv2623 (PDB ID 2JAX) was released for public access. This structure is nearly identical to the present model as demonstrated by superposition over the ATP ligands and the monomeric and dimeric forms (rmsds are 0.57 and 0.81 for 258 and 517 matched CA's, respectively). The differences localize primarily to flexible loop regions (residues 44–58, 150–159) that, while disordered in 2JAX, are partially stabilized in the present structure by local crystal contacts.
To gain insight into the ATP-binding mode(s) exhibited by Rv2623, the structural features of the ATP-binding pocket of domains 1/2 were compared to the monomer fold of the representative ATP-binding USP, MJ0577 (PDBID 1MJH) [26]. Overlay of these structures reveals very considerable similarity for the residues that form the binding pockets and the associated ATP molecules, for which the triphosphoryl moieties assume virtually indistinguishable conformations. Relatively subtle structural and phylogenetic differences that exist between the ATP-binding pockets might nevertheless confer divergent binding and/or regulatory properties to the tandem domains.
To explore the relationship between the putative ATP-dependent biochemical function of Rv2623 and the growth-regulating attribute of Rv2623, we engineered mutations within the L1 (D15E) and β4 (G117A) conserved residues that were predicted, on the basis of the crystal structure, to disrupt ATP recognition (Figure 7A). In silico replacement of the β4 G117 side chain hydrogen with a methyl group suggested that any residue larger than glycine at this position is likely to perturb both of the conserved loop regions in contact with the nucleotide. Similarly, extension of the D15 side chain to glutamate was also predicted to interfere with the ATP-binding conformation (Figure 7A). HPLC analysis of nucleotides extracted from Rv2623D15E and Rv2623G117A revealed that the mutant proteins are indeed deficient in ATP-binding, exhibiting ∼34% (p<0.001) and ∼29% (p = 0.0018) of the amount of ATP bound by wild-type Rv2623, respectively (Figure 7B). Likewise, following an overnight incubation with [α-33P] ATP at 4°C, the amount of protein-bound radioactivity, which represented a very small fraction of the total ATP binding sites, was significantly less for the mutant proteins than wild-type Rv2623 (data not shown). Importantly, thermal denaturation profiles of wild-type Rv2623, Rv2623D15E and Rv2623G117A demonstrated virtually identical Tm values, implying that the native Rv2623 fold was not destabilized by these mutations (Figure 7C). It is therefore likely that the D15E and G117A mutations produced local structural changes in the ATP binding loops that contributed directly to the reduced levels of bound ATP in comparing to wild-type Rv2623.
We next sought to probe the relationship between the nucleotide-binding capacity and growth regulation by this mycobacterial USP. Both the D15E and G117A mutant proteins were overexpressed in M. smegmatis mc2155 at levels equivalent to that of wild-type Rv2623 (Figure S4). Results of these studies demonstrated that while overexpression of wildtype Rv2623 retards the growth of the recipient strain relative to cells transformed with vector alone, growth of the strains overexpressing ATP-binding-deficient mutant Rv2623 are only minimally affected by overexpression as assessed by spotting serial dilutions of the cultures of the appropriate strains onto solid Middlebrook 7H10 agar (data not shown) as well as by monitoring the time to detection using the BD BACTEC 9000MB system (Figure 8A). The distinct effects exhibited by the wild type and the G117A and D15E mutants defective in ATP binding suggests a direct correlation between growth attenuation and ATP binding (Figure 7B). To examine whether the effects of overexpression of Rv2623 on M. smegmatis are operative in virulent M. tuberculosis, the growth kinetics of the Erdman strain overexpressing wildtype Rv2623, as well as the Rv2623G117A and the Rv2623D15E mutant proteins, were evaluated in vitro using the BACTEC 9000MB system (Figure 8B). As in the M. smegmatis studies, the results show that overexpression of Rv2623 in M. tuberculosis results in marked retardation of growth. Furthermore, this growth attenuation is not observed in M. tuberculosis strains overexpressing the G117A or the D15E mutant Rv2623 (Figure 8B). Taken together, these data strongly suggest that the ability of Rv2623 to regulate growth of M. smegmatis and M. tuberculosis is dependent on an ATP-dependent process.
Despite the significance of M. tuberculosis latency in pathogenesis, the mechanisms by which the tubercle bacillus establishes and maintains the latent state remain incompletely defined. Identification of M. tuberculosis genes that are induced by hypoxia and nitric oxide (NO) in vitro provides a framework for understanding the physiology of dormant bacilli [3],[4],[5]. These genes, referred to as the dormancy regulon, are transcriptionally regulated by the mycobacterial two-component system DosR-DosS under hypoxic conditions [4]. Indeed, it has been shown that both the cognate sensor histidine kinase DosS (a member of the dormancy regulon) as well as an “orphan” kinase, DosT, functioning as redox and hypoxia sensors, respectively; can regulate DosR activity, and that O2, NO, and CO can modulate the activity of these two kinases via interaction with a haem prosthetic group [28],[29],[30],[31],[32]. The biological significance of the dormancy regulon has been underscored by in vitro studies of dosR mutants of BCG and M. tuberculosis, which demonstrated the requirement of this transcription factor for survival under hypoxic conditions [3],[33]. Further, upregulation of the expression of certain dormancy regulon genes have been implicated in tuberculosis transmission as well as the virulence of the epidemiologically important W-Beijing lineage of M. tuberculosis [34],[35].
There are eight genes in the M. tuberculosis genome annotated to encode USP family proteins [7]. We studied the M. tuberculosis USP rv2623 because it is one of the most highly induced genes in the dormancy regulon when bacilli are subjected to hypoxia and nitrosative stress [3],[4],[5],[36],[37]. More important, rv2623 was also shown to be up-regulated when the tubercle bacillus is internalized by human and mouse macrophages [10],[38] as well as in the lungs of mice with persistent M. tuberculosis infection [11]. These latter observations suggest that the induction of rv2623 may have biological relevance. The precise mechanisms by which Rv2623 expression is regulated remain to be defined. Recent transcriptional analysis of Rv2623, while confirming the essentiality of the two 18 bp palindromic DosR-binding motifs that are present in the promoter region of this gene [38] for induction of Rv2623 under low oxygen conditions, also demonstrated the presence of additional regulatory elements within the rv2623 5′-untranslated region [18]. These results suggest that the regulation of Rv2623 is likely complex.
The M. tuberculosis dormancy response features a dramatic decrease in metabolic activity, resulting in a rapid decrease in bacterial replication [39]. Therefore, it is possible that deficiency in certain members of the dormancy regulon could result in inability of the tubercle bacillus to enter a latent state in the infected host, leading to unrestrained growth and thus, hypervirulence. Indeed, specific members of the M. tuberculosis dormancy regulon whose insufficiency results in a hypervirulence phenotype have been reported [40],[41]. In certain experimental tuberculosis animal models, DosR deficiency has been associated with a hypervirulence phenotype [41]. However, DosR deficiency has also been reported to have no effect on M. tuberculosis virulence or to lead to an attenuated phenotype [42],[43]. The discrepancies regarding M. tuberculosis virulence in these DosR studies are unclear, but could be due to differences in experimental systems employed. Insufficiency of the chaperone-like α-crystallin encoded by M. tuberculosis hspX (acr) has also been shown to be associated with hypervirulence in a BALB/c mouse model of tuberculosis [40].
In the present study, an rv2623 knockout mutant of virulent M. tuberculosis Erdman fails to establish a chronic persistent infection, displaying a hypervirulent phenotype in susceptible hosts, as assessed by lung bacterial burden, histopathology, and mortality. Results of the complementation studies indicate that the phenotype is Rv2623-specific. This growth-regulating phenomenon is echoed by the observation that ectopic overexpression of Rv2623 results in attenuation of mycobacterial growth. Together, these data strongly suggest that the M. tuberculosis USP Rv2623 has the ability to regulate growth in vitro and in vivo, and is required for the establishment of a persistent infection. Intriguingly, ectopic overexpression of HspX by the same means employed by our study also resulted in an attenuated growth phenotype compared with LacZ-overexpressing controls [44], suggesting that these two tightly co-regulated “stress” proteins might have similar growth-regulatory roles during dormancy.
Bioinformatic and experimental evidence suggest that nucleotide-binding capacity represents a discriminating biochemical feature that facilitates USP protein classification. Putative functional differences between USPs are implied by their assignment to two subclasses: one whose members do not bind ATP and another whose constituents bind and hydrolyze adenine nucleotide substrates [8],[26],[27],[45],[46]. A structural comparison between the prototypic members of the two subclasses, the non-ATP-binding UspA homolog (H. influenzae, PDB ID 1JMV) and the ATP-binding USP, MJ0577 (M. jannaschii, PDB ID 1MJH) revealed that while both proteins exhibit a similar fold, conserved glycine residues within the ATP-binding loop of the latter are substituted with bulky amino acids that preclude ATP recognition in the former [26],[27]. The unique nucleotide-binding pocket of this protein family is structurally distinct from those commonly encountered in ATP-binding proteins [26],[47],[48]. Specific roles for USP family proteins are just beginning to be characterized, and early functional classifications have been informed by ATP-binding capacity [9]. While the non-ATP-binding UspA homologs appear to play diverse roles in promoting survival under a variety of environmental insults [15],[49],[50],[51], the function(s) of ATP-binding type USPs remain unclear [9]. Based on in silico analyses, Florczyk et al. classified Rv2623 as belonging to a novel class of ATPases [52], although formal evidence for ATP binding by this protein has not been reported.
This study has provided substantial biochemical and structural evidence that M. tuberculosis Rv2623 is a bona fide nucleotide-binding USP: i) E. coli-expressed His6-Rv2623 co-purifies with tightly bound ATP and ADP; ii) analysis of the 2.9 Å -resolution Rv2623 crystal structure, the first molecular model of a tandem-type USP, reveals four ATP-bound nucleotide-binding pockets; and iii) point mutations (D15E, G117A) within the conserved L1 (D15E) and β4 (G117A) regions of the structure, which were predicted to disrupt nucleotide-binding, yielded mutant proteins with attenuated ATP-binding capacity. Furthermore, given that the attenuated growth phenotype caused by overexpression of Rv2623 could be abrogated by mutations that interfere with the binding of this protein to its nucleotide substrate, it is likely that the mycobacterial growth-regulatory faculty of Rv2623 is mediated by an ATP-dependent function.
In summary, the results of the present study have revealed that the M. tuberculosis USP Rv2623 has the ability to regulate mycobacterial growth, as evident by the in vivo hypervirulence phenotype of Δrv2623, which fails to establish a persistent infection in susceptible hosts, as well as the growth attenuation observed in mycobacteria overexpressing this USP. Thus, M. tuberculosis Rv2623 may serve the function of promoting mycobacterial transition into latency. The latent state allows persistence in infected individuals of tubercle bacilli that can reactivate to cause active disease and to disseminate when the immune status of the host is compromised. As a result, Rv2623 may contribute significantly to the propagation of the tubercle bacillus in the human host and the difficulties in eradicating tuberculosis. Mechanistically, results of the mutagenesis studies have shown that Rv2623 regulates growth through ATP-dependent function. Clearly, much remains to be learned regarding how the ATP-dependent function of Rv2623 contributes to growth regulation. It has been proposed that a nucleotide-binding USP from M. jannaschii, MJ0577, whose ability to hydrolyze ATP is dependent on interaction with factor(s) present in the cell extract of this hyperthermophile [26], functions as a molecular switch much like the Ras protein family, whose GTP hydrolysis ability is modulated by interaction with a number of regulatory proteins [53],[54],[55]. The fact that E. coli-expressed Rv2623 co-purifies with ADP as well as ATP suggests the possibility that this mycobacterial USP, like MJ0577, is capable of ATP hydrolysis. It is therefore conceivable that M. tuberculosis Rv2623, as a component of the yet-to-be defined dormancy signaling pathway(s), functions as a molecular switch by virtue of its ATP-binding and putative ATP-hydrolyzing properties, to mediate the establishment of tuberculous latency. Experiments designed to investigate the potential ATP-hydrolyzing activity of Rv2623 are currently underway. Recent identification of the DosR-dependent dormancy regulon [3],[4],[5]; the DosR-independent enduring hypoxic response, which involves over 200 mycobacterial genes, including those known to regulate bacteriostasis [42]; and the demonstration that M. tuberculosis redox and hypoxia sensors can interact with multiple ligands that differentially modulate the activity of these important kinases [28],[29],[30],[31],[32], predict a complex regulatory network for tuberculous latency. Elucidation of how ATP-binding and, potentially, the hydrolysis of ATP by Rv2623 regulate M. tuberculosis dormancy-signaling pathways will likely illuminate the mechanisms by which the tubercle bacillus establishes persistence.
Liquid cultures of M. tuberculosis and M. smegmatis strains were grown in Middlebrook 7H9 medium (Becton Dickinson, Sparks, MD) supplemented with 0.2% glycerol (Sigma, St. Louis, MO), 0.05% Tween 80 (Sigma, St. Louis, MO), and 10% oleic acid-albumin-dextrose-catalase (OADC) enrichment media (Becton Dickinson, Sparks, MD). For the determination of the number of colony forming units (CFU) and examination of growth on solid media, Middlebrook 7H10 agar medium (Becton Dickinson) supplemented with 0.5% glycerol and 10% OADC was used. The Δrv2623 mutant strain was maintained in media supplemented with 50 µg/ml hygromycin B (Roche) and cultures of complemented, Rv2623-overexpressing strains contained kanamycin (40 µg/ml). Growth was also examined in minimal Sauton's medium (4 g asparagine, 2 g sodium citrate, 0.5 g K2HPO4·3H2O, 0.5 g MgSO4·7H2O, 0.05 g ferric ammonium citrate, 60 g glycerol in 1 L of H2O supplemented with 0.05% Tween80 and antibiotics as required). For some experiments, growth was monitored by the BD BACTEC 9000 system (Becton Dicinson). Stationary phase M. tuberculosis or M. smegmatis cultures were inoculated in triplicates (105 or 104 CFU) into vials of liquid medium containing a sensor compound that fluoresces upon depletion of oxygen as a result of bacterial growth. The time to detection reflects the rate of bacterial growth.
Replacement of genomic rv2623 was performed by allelic exchange using a specialized transducing phage delivery system as previously described [12]. Transformants were analyzed by PCR and Southern blot to confirm replacement of rv2623 with a hygromycin cassette, yielding Δrv2623. A complemented strain was generated as described previously [56] by transformation of Δrv2623 with a plasmid vector that integrates at the attB site and bears the rv2623 coding sequence under transcriptional control of the constitutive hps60 promoter or the endogenous rv2623 promoter [18], yielding Δrv2623 attB::Phsp60 Rv2623 and Δrv2623 attB::Prv2623 Rv2623; respectively. The M.tuberculosis hsp60 promoter fusion was also used to overexpress Rv2623 via subcloning of this region into pMV261, a non-integrating variant of pMV361 to yield pMV261::rv2623, which is self-replicating at 3–5 copies/cell [57] (Text S1).
Log phase cultures (OD600 = 0.8–1.0) of Erdman and Δrv2623 were diluted 1∶10 into Sauton's media containing various stress-inducing chemical agents (phenazine methosulfate, streptonigrin and mitomycin C) at the indicated concentrations or for acid stress into 7H9+10% OADC+0.05% Tween 80 (pH = 4.0) and grown at 37°C for several days. Growth was monitored by OD600. Survival of these strains following heat shock was compared after a shift of log phase cultures from 37°C to 53°C and determining the number of CFU/ml at various time points thereafter. For irradiation with UV light, cells were plated onto solid 7H10 agar supplemented with 0.05% Tween80 and exposed to increasing amounts of UV energy (UV Stratalinker 1800, Stratagene). Surviving cells were enumerated and the data are expressed as percent survival as compared to unexposed controls. In addition, cells were treated with the indicated concentrations of mitomycin C for a period of 1 hour followed by determination of surviving CFU/ml. See Figure S2 for details.
Outbred Hartley guinea pigs (∼500 g body weight) (Charles River Laboratories, North Wilmington, MA) were given a low dose of M. tuberculosis using a Madison chamber aerosol generation device calibrated to deliver ∼30 CFU [58]. Guinea pigs were sacrificed (n = 5) at 20, 40, and 60 days post infection for histological analysis and determination of organ bacterial burden. Histological analysis of infected tissues was performed by scoring individual tissue sections based on criteria described in Protocol S1.
For the murine tuberculosis model, six-to-eight-week-old mice (Jackson Laboratories, Bar Harbor, Maine) were infected with M. tuberculosis via aerosol (In-Tox Products, Albuquerque, NM) as previously described [19] with ∼100 CFU (C57BL/6) or 750–1000 CFU (C3H/HeJ, C3HeB/FeJ). For CFU determination, mice were sacrificed (n = 3) at the times indicated and portions of the lung, liver and spleen were homogenized in PBS+0.05% Tween 80, diluted, and plated onto solid 7H10 media.
The coding sequence of rv2623 was PCR-amplified from M. tuberculosis Erdman genomic DNA and subcloned into the expression vector pQE80L (Qiagen, Inc.), which encodes an N-terminal His6-tag, producing the plasmid pQE-rv2623 (Text S1). Expression was carried out following isopropyl beta-D-thiogalactoside (IPTG) induction of BL21 E. coli transformed with pQE-rv2623. His6-Rv2623 was then affinity-purified to homogeneity from BL21 cell lysates using Ni-NTA agarose (Qiagen, Inc.) according to the manufacturer's instructions. For crystallization, purified Rv2623 was concentrated to 12 mg/ml using a 10 kDa Molecular Weight Cut Off (MWCO) centrifugal filter (Amicon), and frozen at −80°C.
A Superdex 200 10/300 GL column (GE Healthcare Life Sciences) was equilibrated with Rv2623 dialysis buffer (Text S1) and calibrated using the following molecular mass standards: aldolase (158 kDa), bovine serum albumin (67 kDa), ovalbumin (43 kDa), chymotrypsinogen A (25 kDa) as described in the Amersham Pharmacia technical notes (GE Healthcare Life Sciences). The flow rate was set to 0.15 ml/min and elution of the protein was monitored at 280 nm.
Nucleotides were extracted from purified Rv2623 by boiling. Samples were then loaded onto an analytical, anion-exchange HPLC column (AX300, Eprogen Inc. or Mono Q HR 5/5, GE Healthcare). Samples were eluted isocratically using NaH2PO4, pH 5.5 (AX300) or using a ammonium phosphate pH 7.0 (0.02–1.0 M) step gradient (Mono Q) (Text S1). Nucleotides were identified on the basis of retention time relative to nucleotide standards, and quantified by peak area.
Following Ni affinity chromatography, His6-Rv2623 samples for stoichiometry measurements (500 µl, 8–120 µM protein) were attained by subjecting the Ni-purified fractions to rapid desalting over a HiPrep Desalting 26/10 column (GE Healthcare) equilibrated with 50 mM NaCl, 2 mM MgCl2, 10% glycerol, 20 mM HEPES pH 8.0 and/or an additional purification using a MonoQ 10/100 GL column (GE Healthcare) equilibrated with the desalting buffer and eluted with a linear salt gradient ranging from 50 mM to 250 mM NaCl. Amicon centrifugal filtration concentrators (MWCO = 30 kDa) were employed for final concentration steps prior to analysis. Bound nucleotide was then released by boiling and quantified by HPLC (under the Mono Q HR 5/5 conditions) according to the methodology described above and Text S1. The heat-precipitated protein was subsequently reconstituted in 6 M guanidine HCl and its concentration was determined by spectrophotometric measurement of the protein peak at 280 nM and a molar extinction coefficient for Rv2623 at 280 nm (54,640/Mcm). This extinction coefficient was determined using a 6 M GuHCl-reconstituted (nucleotide-free) sample of Rv2623 that had been subject to quantitative amino acid analysis at the Yale Keck Facility. The nucleotide binding stoichiometry was calculated as the molar ratio of the released nucleotide to protein.
Single amino acid substitutions were incorporated into the rv2623 coding region contained in appropriate expression vectors by mismatched PCR priming (Text S1). Individual PCR reactions were performed using either pMV261::rv2623 or pQE-rv2623 plasmid templates for mycobacterial overexpression and protein purification, respectively. Then the pMV261:: rv2623 mutants expression vector DNA was used to transformed into M.smegamatis mc2155 and the DNA of pQE-rv2623 mutants was transformed into E.coli BL21(DE3). Thermal denaturation curves were determined for purified wild type and mutant Rv2623 using an IQ5 Real Time PCR Detection System (Bio-Rad) following incubation with SYPRO Orange protein gel stain (Invitrogen) (Text S1).
ATP and MgCl2 were added to final concentrations of 0.8 mM and 1.8 mM, respectively, in the protein sample prior to crystallization by sitting-drop vapor-diffusion at 4°C (Text S1). Diffraction data were collected at the National Synchrotron Light Source beamline X29A on an ADSC Q315 detector through the Macromolecular Crystallography Research Resource (PXRR) mail-in crystallography program. Data processing and scaling was performed with the HKL2000 suite.
The structure of M. tuberculosis Rv2623, which contains two USP domains in tandem, and whose first domain shares 25% sequence identity with the USP M. tuberculosis Rv1636, was solved using the molecular replacement method and a CHAINSAW-generated search model consisting of the Rv1636 dimer (PDB ID 1TQ8 chains A, B), using a 2.9 Å, C2221 dataset (a = 173, b = 241.5, c = 241.7). A starting polyalanine model (R/Rfree = .56/.57) of four dimers was subject to four refinement cycles, each consisting of multi-domain rigid-body refinement in Molrep, a single cycle of restrained MLF refinement in Refmac5 (to obtain input FOMs for DM), 20 cycles of phase extension in DM (as above), and manual rebuilding of the polyalanine backbone in Coot. As R-factors converged (R/Rfree = .40/.42), ∼80% of the side chains were positioned and the Rv2623 dimers were further rebuilt and refined (R/Rfree = .31/.33) in CNS using high NCS restraint weights (400 kcal/mol) with rigid-body, energy minimization, grouped isotropic B factor, and simulated annealing refinement protocols. ATP and Mg2+ were built within composite omit density (calculated in CNS) during the final rebuilding/refinement cycles conducted with relaxed NCS restraints in Arp-waters and Refmac5, yielding final R/Rfree = 24.5/26.5. SigmaA-weighted difference maps calculated with the refined model reveal weak, fragmented density for a pseudotranslated copy of the Rv2623 dimer whose corresponding NCS translational vector (uvw = .500, .012, .494) appears in the native patterson at 7.7% of the origin peak height. Data collection and refinement statistics are summarized in Protocol S1. The coordinates of M. tuberculosis Rv2623 have been submitted to the protein databank (PBDID 3CIS).
|
10.1371/journal.pntd.0004856 | Spatiotemporal Co-existence of Two Mycobacterium ulcerans Clonal Complexes in the Offin River Valley of Ghana | In recent years, comparative genome sequence analysis of African Mycobacterium ulcerans strains isolated from Buruli ulcer (BU) lesion specimen has revealed a very limited genetic diversity of closely related isolates and a striking association between genotype and geographical origin of the patients. Here, we compared whole genome sequences of five M. ulcerans strains isolated in 2004 or 2013 from BU lesions of four residents of the Offin river valley with 48 strains isolated between 2002 and 2005 from BU lesions of individuals residing in the Densu river valley of Ghana. While all M. ulcerans isolates from the Densu river valley belonged to the same clonal complex, members of two distinct clonal complexes were found in the Offin river valley over space and time. The Offin strains were closely related to genotypes from either the Densu region or from the Asante Akim North district of Ghana. These results point towards an occasional involvement of a mobile reservoir in the transmission of M. ulcerans, enabling the spread of bacteria across different regions.
| Infection with Mycobacterium ulcerans causes the debilitating skin disease Buruli ulcer. Until today, transmission pathways and reservoirs of this emerging pathogen are not well understood. Generally, it is assumed that infection occurs after contact with potential environmental sources of M. ulcerans through puncture wounds or lacerations or via invertebrate vectors, such as aquatic insects contaminated with the bacteria. Comparative genome analyses of M. ulcerans strains isolated from patients living in the same BU endemic areas have revealed a close relationship between the genotype detected and the geographical origin, indicating that the reservoir of the pathogen is relatively fixed in space. In the present study, we report the co-circulation of two distinct M. ulcerans clonal complexes in the same BU endemic area over space and time. Since members of these two clonal complexes were closely related to strains from either the Densu river valley or the Asante Akim North district of Ghana, we conclude that a mobile reservoir of M. ulcerans may be involved in the occasional spread of the bacteria across different regions.
| Mycobacterium ulcerans is an emerging pathogen with elusive reservoirs and transmission pathways. It causes the devastating skin disease Buruli ulcer (BU) that mainly affects rural populations in West Africa [1]. M. ulcerans is a descendant of the fish and occasionally human pathogen Mycobacterium marinum [2], from which the new species has evolved through the acquisition of a plasmid encoding the enzymatic machinery for the synthesis of the macrolide toxin mycolactone [3]. From this common ancestor at least three different lineages or ecovars have evolved through genome reduction [4]. Clinical isolates from Africa belong to the classical lineage and differ from each other only in a very limited number of single nucleotide polymorphisms (SNPs) [4, 5], indicative for a highly clonal recent expansion of the pathogen in Africa. BU is characterized by a focal distribution of cases within endemic countries. Previous studies have revealed a strong association between genotype and the geographic origin of strains [4, 6, 7], speaking for the development of local clonal M. ulcerans complexes following the introduction of this pathogen into a new area. The limited genomic diversity found within these local clonal complexes is however sufficient for studies on the distribution of variants at a micro-epidemiological level [8]. Since human-to-human transmission seems to be rare, findings point towards infection from a relatively localized environmental reservoir of the pathogen. In view of the association of BU outbreaks with stagnant and slow-flowing water bodies, a reservoir in the aquatic ecosystem is considered likely [9]. While in several African endemic areas a single unique clonal complex has been identified [4, 6–8], a recent comparative whole genome sequencing study of isolates from residents of the Asante Akim North district of Ghana showed for the first time the concurrent presence of two distinct clonal complexes within one BU endemic area [10].
Within the framework of a comprehensive genome analysis of clinical M. ulcerans isolates from Ghana, we analyzed genomes of a limited number of strains from the Offin river valley and equally observed a co-existence of two clonal complexes.
Ethical approval for the study was obtained from the institutional review board of the Noguchi Memorial Institute for Medical Research (Federal-wide Assurance number FWA00001824). Written informed consent was provided by all study participants.
In an exhaustive active BU case search conducted in 2013 and a subsequent continuous monitoring of cases over a 17-months period in 13 randomly selected communities located in the historically highly BU endemic Offin river valley, an unexpectedly low prevalence of BU was revealed with only 11 laboratory-confirmed cases identified [11]. Two M. ulcerans strains that could be isolated from two of the 11 patients, as well as three M. ulcerans strains isolated in 2004 from lesions of two BU patients residing in the valley were analyzed in this study (Table 1). In addition, we included 48 M. ulcerans isolates from patients residing in BU endemic areas located in the Densu river valley of Ghana, which were part of a previous SNP typing study [8]. All M. ulcerans strains were subjected to whole genome sequencing.
Genomic DNA was extracted from M. ulcerans cultures by phenol-chlorophorm extraction and ethanol precipitation as described previously [12]. Multiplexed genomic DNA libraries were prepared and sequenced on an Illumina HiSeq 2000 on 75-bp paired-end runs [13]. Illumina reads were aligned to the complete reference genome of M. ulcerans strain Agy99 (GenBank accession number CP000325.1) with an insert size between 50 and 400 bp using BWA version 0.7.10. SNPs were identified using SAMtools [14] as described [15] and were filtered for a minimum mapping quality of 30 and a quality cutoff of 75%. SNPs called in repetitive regions of the M. ulcerans reference genome (737,280 bp) were excluded from the analysis and only the SNPs mapped in the core genome (4,894,326 bp) were used to construct the phylogenetic trees.
Maximum-likelihood phylogenetic analysis was performed using RAxML [16] on the alignment of identified SNPs from across the Ghanaian genomes sequenced here, together with genomes sequenced in previous studies [4, 10, 17]. Additional strains isolated from BU patients from other regions of Ghana and from Benin and Australia were included in the analysis to provide a comprehensive genetic context for the analysis of genetic diversity among the Offin and Densu isolates.
Phylogenetic analysis demonstrated the expansion of a single clonal complex in the Densu river valley (Fig 1). This complex has diversified substantially, but still forms a separate cluster, distinct from other African local clonal complexes (Figs 2 and 3). When compared to the second branch of the classical lineage of M. ulcerans—isolates from Australia—it is evident, that all African isolates are genomically extremely closely related (Fig 2). In contrast to the observation of a single clonal complex in the Densu river valley, our analysis revealed for the Offin river valley the presence of members of two distinct clonal complexes (Fig 3). Two Offin isolates (NM031 and NM997) were closely related to isolates from the Densu river valley of Ghana (Figs 1 and 3). The other three Offin isolates (NM022B, NM022D and NM972)–separated from the two Densu-like Offin strains by 29 SNPs–clustered with strains (belonging to a clonal complex designated Agogo-1; [10]) from the Asante Akim North district in the Ashanti region of Ghana (Fig 3). Intra-genotype average diversity was low with 12 and 17 identified SNPs among Densu-like and Agogo-1-like Offin isolates, respectively. Not a single SNP difference was found between the genomes of two strains (NM022B and NM022D) isolated from two different lesions of the same patient. In a next step, we combined the phylogenetic analysis of the Offin isolates with information on the residence of the patients within the river valley and the year of strain isolation (Fig 4). In both 2004 and 2013 one member each of the two clonal complexes, isolated from BU patients resident in different communities was found. Due to the limited number of isolates and missing details on the travel history of the BU patients from which these strains have been isolated, no firm conclusions could be drawn concerning the apparent lack of geographical clustering. However, the data revealed a co-circulation of two distinct M. ulcerans clonal complexes in the Offin river valley over space and time.
In contrast to the remarkably strong link between genotype and geographical origin of clinical M. ulcerans isolates reported in previous genotyping studies conducted in African BU endemic foci [6–8], two distinct M. ulcerans clonal complexes were recently found to co-exist in the Asante Akim North district of Ghana among strains isolated within the short time frame of two years and an area of only 30km2 [10]. It was concluded that M. ulcerans genotypes might be spread across larger areas, suggesting the presence of a rather mobile reservoir of infection in addition to the postulated more focalized aquatic niche environment typically associated with the pathogen [9]. In this context, recent data indicated that M. ulcerans is able to persist for several months in underwater decaying organic matter [18], possibly as a commensal in protective aquatic host environments [19–21]. While the specific factors favoring the persistence of M. ulcerans in the environment and its transmission are yet to be explored, a complex interplay between environmental factors as well as biotic and abiotic drivers is assumed [22, 23]. Reductive genome evolution of M. ulcerans speaks for niche adaptation [17].
In the present study we revealed the co-existence of both Densu-like and Agogo-1-like M. ulcerans genotypes in communities along the Offin river at two time points separated by ten years. Our data thus show that co-existence of clonal complexes in one BU endemic area may prevail over longer time periods. A mobile mammalian host, allowing the bacteria to replicate and to be shed to the environment, that way forming a reservoir [18] from which humans may be infected by unknown mechanisms, could be the missing link explaining the spread of M. ulcerans from an established BU endemic region to a new area. However, as demonstrated here by the presence of only a single clonal complex in the Densu river valley, the exchange of genetic M. ulcerans variants between BU endemic areas appears to be an extremely rare event.
While in Australia possums have been identified as a host for M. ulcerans [24], another line of evidence points to the involvement of humans with chronic ulcerative BU lesions in the spread of the bacteria in African BU environments. Extensive whole-genome sequencing studies are required to further unravel the evolutionary history and population structure of M. ulcerans in Africa.
|
10.1371/journal.ppat.1006533 | Structural basis of Zn(II) induced metal detoxification and antibiotic resistance by histidine kinase CzcS in Pseudomonas aeruginosa | Pseudomonas aeruginosa (P. aeruginosa) is a major opportunistic human pathogen, causing serious nosocomial infections among immunocompromised patients by multi-determinant virulence and high antibiotic resistance. The CzcR-CzcS signal transduction system in P. aeruginosa is primarily involved in metal detoxification and antibiotic resistance through co-regulating cross-resistance between Zn(II) and carbapenem antibiotics. Although the intracellular regulatory pathway is well-established, the mechanism by which extracellular sensor domain of histidine kinase (HK) CzcS responds to Zn(II) stimulus to trigger downstream signal transduction remains unclear. Here we determined the crystal structure of the CzcS sensor domain (CzcS SD) in complex with Zn(II) at 1.7 Å resolution. This is the first three-dimensional structural view of Zn(II)-sensor domain of the two-component system (TCS). The CzcS SD is of α/β-fold in nature, and it senses the Zn(II) stimulus at micromole level in a tetrahedral geometry through its symmetry-related residues (His55 and Asp60) on the dimer interface. Though the CzcS SD resembles the PhoQ-DcuS-CitA (PDC) superfamily member, it interacts with the effector in a novel domain with the N-terminal α-helices rather than the conserved β-sheets pocket. The dimerization of the N-terminal H1 and H1’ α-helices is of primary importance for the activity of HK CzcS. This study provides preliminary insight into the molecular mechanism of Zn(II) sensing and signaling transduction by the HK CzcS, which will be beneficial to understand how the pathogen P. aeruginosa resists to high levels of heavy metals and antimicrobial agents.
| P. aeruginosa inhabits diverse environments and is one of the most prevalent opportunistic human pathogens of immunocompromised patients. The high antibiotic resistance is a major cause of therapeutic failure in the treatment of P. aeruginosa infections. The opportunistic pathogen P. aeruginosa co-regulates cross-resistance between Zn(II) and carbapenem antibiotics by the CzcR-CzcS signal transduction system. The extracellular Zn(II) stimulus is sensed by the HK CzcS and further triggers metal detoxification and antibiotic resistance through intracellular regulatory pathway. Here, we provide the three-dimensional structure of CzcS SD in complex with the Zn(II). Based on the structure, several key residues for Zn(II) sensing and regulation are identified, and the signal transduction is disclosed to be modulated by the dimerization of N-terminal α-helices in the sensor domain. Our research will provide potential guidance for the treatment of clinical issues caused by co-regulation between heavy metals and antibiotics in P. aeruginosa.
| Bacteria are extremely versatile that can regulate cellular processes in a sophisticated manner and thereby survive in changing environments. The two-component system (TCS) is the predominant strategy for coupling various extracellular stimuli to appropriate cellular responses in microorganisms [1–5]. In Pseudomonas aeruginosa (P. aeruginosa), approximately 130 genes have been identified that encode various types of TCSs [5, 6]. These regulatory systems enable this organism to ubiquitously exist in diverse environments and to express various virulence factors [7, 8]. Thus, P. aeruginosa is one of the most prevalent opportunistic pathogens and causes severe hospital-acquired infections among immunocompromised patients [7, 9, 10]. It is capable of causing both chronic and acute pulmonary infections in cystic fibrosis (CF) patients, ventilator-associated pneumonia, and sepsis in burn patients [8]. Moreover, this pathogenic bacterium possesses intrinsically high levels of resistance to multiple classes of antimicrobial agents, presenting tremendous obstacles for anti-infective therapies [7, 11].
The CzcR-CzcS TCS in P. aeruginosa is responsible for numerous cellular processes, including Zn(II) resistance, carbapenem antibiotic resistance, quorum sensing, and virulence regulation (Fig 1A) [12–14]. Under direct stimulation by Zn(II), the histidine kinase (HK) CzcS auto-phosphorylates on its conserved histidine residue. It subsequently transmits the phosphoryl group to the conserved aspartate residue in the receiver domain of the response regulator (RR) CzcR. The phosphorylated CzcR up-regulates the expression of a metal efflux pump, CzcCBA. It also represses the expression of OprD, a porin that regulates the entry of basic amino acids and carbapenem antibiotics [13, 14]. This co-regulation between metal detoxification and antibiotic resistance is unusual, and its mechanism will provide significant guidance for the treatments of environmental and clinical issues.
In the CzcR-CzcS TCS, the HK CzcS is predicted to be the transmembrane sensor-transmitter, and it contains three functional domains [15]. The highly diverse N-terminal periplasmic sensor domain is arranged between two membrane-spanning segments and is followed by a conserved C-terminal cytoplasmic kinase domain. The stimulus is detected by the periplasmic sensor domain and transmitted across the membrane to the cytoplasmic kinase domain [3, 16]. Given the pivotal role of the extracellular sensor domain in signal recognition and transduction, we determined the crystal structure of the CzcS sensor domain (CzcS SD) in the presence of Zn(II) (referred as CzcS-Zn hereafter). Together with the biochemical and in vivo studies, the CzcS SD is identified to bind Zn(II) between N-terminal H1 and H1’ α-helices, which is the key first step in Zn(II) detoxification and meropenem resistance by HK CzcS. The N-terminal H1 and H1’ α-helices are also shown to play important roles in signal transduction via a series of structure-guided mutagenesis studies. The study reveals the CzcS SD appears to utilize a new mode which is not previously observed for sensor HKs to protect P. aeruginosa from high levels of Zn(II) and in parallel meropenem.
The structure of CzcS-Zn complex was solved by single-wavelength anomalous diffraction (SAD) using the data collected at the zinc peak wavelength (1.2823 Å) and was refined using data collected at a wavelength of 1.0000 Å. The Rwork and Rfree are 0.210 and 0.253, respectively. The data collection and other refinement statistics are summarized in Table 1. The structure belongs to the C2 space group and it contains two molecules (CzcS SD, amino acids 40–166 in the CzcS protein) per asymmetric unit (S1A Fig). As supported by the clear electron density, residues 40–132 and 135–161 in both molecules are well defined, whereas the C-terminal segment (residues 162–166) and the loop (residues 133–135) that connects the S4 and S5 β-strands are disordered. The tertiary structures of the two molecules are similar (S1B Fig). The root-mean-square deviation (r.m.s.d.) between them is 0.8Å with 104 pairs of corresponding Cα atoms superimposed. The structural deviations are mainly caused by the tilting of the N-terminal domain (residues 40–81) with respect to the C-terminal domain (residues 83–161).
The structure of CzcS SD is a mixed α/β-fold in nature, which can be divided into two domains (Fig 1B). The N-terminal helix-loop-helix domain is composed of the H1 α-helix (residues 40–60), the connecting loop (residues 61–66), and the H2 α-helix (residues 67–81). It is connected to the C-terminal domain by one residue, Thr82. The C-terminal domain contains five β-strands (S1-S5) and one 03B1-helix (H3). The five β-strands form one anti-parallel β-sheet. S5 (residues 150–161) is located in the middle and is flanked by S1 (residues 83–90) and S2 (residues 97–102) on one side and by S4 (residues 136–147) and S3 (residues 126–132) on the other side. The H3 α-helix packs against the anti-parallel β-sheet and is nearly perpendicular to the first two α-helices. A short kink at residues 106 and 107 divides the H3 α-helix into H3a (residues 103–105) and H3b (residues 108–120). The N-terminal H1 α-helix and the C-terminal S5 β-strand are oriented in the same direction, which connect to the transmembrane helix (TM1 and TM2 helices) in the transmembrane domain of HK CzcS.
The typical PDC members include the divalent cation sensor PhoQ [18, 19], the citrate sensor CitA [20, 21], and the C4-dicarboxylate sensor DcuS [22]. Despite of the negligible sequences identities, it reveals that the CzcS SD possesses a similar structural arrangement to those of other members of the PDC superfamily by the structural similarity searches performed with the Dali server program [23] (S2 Fig). The relatively high Z-scores of 5.4, 4.0, and 7.0 are yielded in the structure-based alignments of the CzcS SD with PhoQ [18], CitA [20], and DcuS [22], respectively. The sensor domain of CzcS and PhoQ can be largely superimposed in the C-terminal domain with a r.m.s.d. of 2.6 Å over 66 corresponding Cα atoms. Similarly, the CitA superimposes onto CzcS over 58 corresponding Cα positions with a r.m.s.d. of 2.9 Å, and DcuS superimposes onto CzcS over 54 corresponding Cα positions with a r.m.s.d. of 2.6 Å. The distinct difference between the structure of CzcS SD and those of the other PDC superfamily members is the orientation of the N-terminal helix-loop-helix domain (Fig 1C). In the structure of CzcS-Zn, the N-terminal helix-loop-helix domain is tilted away from the central five anti-parallel β-sheet, which may be caused by the Zn(II) binding at the H1 and H1’ α-helices.
Two Zn(II) ions are captured in the structure of CzcS-Zn. One of the Zn(II) ions is coordinated with His72 and Asp76 of CzcS molecule A (S1A Fig, gray). It is also coordinated with Asp62 of symmetry-related molecule B and His72 of symmetry-related molecule C (S3 Fig). The P. aeruginosa functions normally in response to extracellular Zn(II) with the double mutation of His72 and Asp76 on HK CzcS (S4 Fig). The result shows that this Zn(II) binding pattern is physiologically irrelevant and may be caused by crystallographic packing.
The other Zn(II) is identified to be functional relevant and is associated with the second CzcS molecule (S1A Fig, green) shown in the asymmetric unit. Double mutation of coordinated residues (His55 and Asp60) of second Zn(II) on HK CzcS causes severely defects on the regulation of Zn(II) resistance in P. aeruginosa (S4 Fig). In the symmetry operation, this CzcS molecule can form a homodimer (Fig 1B), and the Zn(II) is exclusively buried between the central parallel H1 and H1’ α-helices which constitute the dimer interface (Fig 2A). The H1 and H1’ α-helices are surrounded by multiple solvent water molecules, which facilitate the Zn(II) access to the active site. A distorted tetrahedral geometry is adopted by Zn(II) to coordinate with the symmetric ligands (His55/Asp60 and His55’/Asp60) from the H1 and H1’ α-helices, respectively (Fig 2B). The His55 and His55’ residues interact with the Zn(II) through their Nε2 nitrogen atoms, and Asp60 and Asp60’ residues contact Zn(II) through the Oδ2 atoms of their carboxylate side-chain (Fig 2B). The bond distances of the coordination center are 1.92 Å-2.17 Å with the bond angles ranging from 102.0° to 127.6°. In each monomer, the Oδ2 atom of Asp59 residue makes a hydrogen-bond interaction with the Nδ1 nitrogen atom of the His55 residue. Additionally, the O atom of His55 forms hydrogen-bond with the N atoms of Asp59 and Asp60, respectively (Fig 2B). These second shell interactions, particularly the carboxylate side-chain with the histidine ligand, are thought to play an important role in the stability of the coordination structure [24].
In view of the co-regulation of cross-resistance between metal ions and carbapenem antibiotics by CzcR-CzcS TCS, the physiological importance of the His55 and Asp60 residues in P. aeruginosa is investigated by using the Zn(II) and meropenem (MEPM) antibiotic tolerance assay (Fig 2C). All the strains keep consistent growth state on the LB solid medium in the absence of Zn(II), and their growing status are not influenced until the concentration of Zn(II) reaches 0.5 mM. However, when a higher concentration of Zn(II) (2.5 mM) are supplied, the P. aeruginosa PAO1 strain demonstrates an obvious growth advantage over the czcS-deficient strain. The abolished metal resistance of the czcS-deficient strain is restored to wild type levels by complementation with a plasmid (pAK1900) carrying the czcS gene. When His55 and Asp60 are substituted by the amino acids which can’t coordinate with Zn(II), the mutant strains (H55A, D60A, and H55R) dramatically attenuate their abilities in the Zn(II) detoxification for the destruction of Zn(II) binding site. Remarkably, the losses in responsiveness to Zn(II) of H55A and H55R mutant strains are similar in degree to the non-responsiveness of the czcS-deficient strain. Additionally, the aforementioned mutant strains and the czcS-deficient strain lose their Zn(II)-inducible resistance to the MEPM antibiotic. The mutagenesis analyses corroborate the crucial role of the His55 and Asp60 in Zn(II) sensing. Intriguingly, markedly different phenotypes are observed when the His55 and Asp60 are replaced with the coordinated cysteine residues (Fig 2C). The H55C mutant partially preserves the ability in Zn(II) detoxification and in parallel meropenem resistance. By contrast, the D60C mutant shows equivalent responsiveness to Zn(II) and MEPM to that of the wild type PAO1.
The Co(II) has similar radius to Zn(II), and it can bind to Cys2His2 coordination site in a tetrahedral geometry as well [25]. The wild type P. aeruginosa is blind to Co(II) that none Co(II)-inducible resistance to the MEPM antibiotic can be observed (Fig 3). With the mutation of residues (Asp60 and Asp60’) on the N-terminal H1 and H1’ α-helices, the D60C mutant strain displays Co(II)-inducible resistance to the MEPM antibiotic (Fig 3). Although the increased antibiotic resistance induced by Co(II) is not as strong as that of Zn(II), this experiment indicates that the binding of Co(II) between the N-terminal H1 and H1’ α-helices can also regulate the downstream signaling transduction in CzcR-CzcS TCS.
The linker region connects the H1 and H1’ α-helices to the transmembrane helices. It plays an important role in the process of signaling transduction from extracellular sensor domain to the transmembrane region. The cysteine substitution scanning is performed in the linker region on the basis of the H55A mutant (Fig 4A). As demonstrated above, the H55A mutant loses its intrinsic resistance to Zn(II) and MEPM as the czcS-deficient strain due to the destruction of Zn(II) binding site. In conjunction with the mutation L38C, the L38C H55A mutant restores the responsiveness to Zn(II) stimulus. This mutant strain can survive on the LB medium with high concentrations of Zn(II) and shows Zn(II)-inducible resistance to MEPM (Fig 4B). The experiment indicates that the strain with cysteine substitution in the linker region instead of that at the Zn(II) binding site can also sense and transmit the Zn(II) signal as well as wild type P. aeruginosa.
The chromogenic indicator 4-(2-Pyridylazo)resorcinol (PAR) is reported to form both 1:1 and 2:1 complexes with Zn(II) with stepwise affinity constants of 7.7×106 and 5.0×105 M-1, respectively (at pH 7.4, 0.15 M KCl, 22°C) [26, 27]. It has been widely used to determine the dissociation equilibrium constants of protein-Zn(II) complex in the range of nanomolar to picomolar [28, 29]. With the addition of Zn(II) to the PAR solution, the formative PAR-Zn and PAR2-Zn complex will cause an intense absorbance at 500 nm [29]. The absorption bands of PAR and Zn(II) complex at 500 nm are reduced by the addition of wild type and mutant CzcS SD (CzcS SD H55C, CzcS SD D60C, and CzcS SD L38C H55A) (S5 Fig). It indicates that the wild type CzcS SD and aforementioned mutants have the ability to compete with PAR for binding Zn(II). The representative titration spectrums are displayed in S6 Fig for the PAR with Zn (II) under the competition of wild type and mutant CzcS SD. The titration data at 500 nm and fitting binding isotherms of which are inserted in corresponding titration spectrums (S6 Fig). The dissociation constants determined by Dynafit software [30] for Zn(II) with wild type CzcS SD, CzcS SD H55C, CzcS SD D60C, and CzcS SD L38C H55A are 1.7 (±0.2)×10−6 M, 8.5 (±0.4)×10−7 M, 5.7 (±0.3)×10−8 M, and 9.4 (±0.8)×10−9 M, respectively. By using one-site fitting model, the coefficient of variation (CV) for each equilibrium binding experiments is approximately 10% regardless of whether the wild type or mutant CzcS SD is monitored (S1 Table). It indicates that the model for data fitting provides a good description of the available data.
The Zn(II) induced dimerization of CzcS SD was analyzed by the chemical crosslinking experiments (S7A Fig) with bis[sulfosuccinimidyl] suberate (BS3) as the primary amine reactive crosslinker [31–33]. Under denatured electrophoresis conditions, the CzcS SD primarily migrates at the position with the molecular weight of monomer, and negligible proportion of dimer is observed when it is treated by excess BS3 crosslinker (S7A Fig). It means that the CzcS SD without Zn(II) is mainly existed as monomer in solution that leads the poor efficiency of intermolecular crosslinking reaction. The efficiency of intermolecular crosslinking is significantly increased when Zn(II) is loaded into CzcS SD (S7A Fig). With the Zn(II) binding at dimer interface (Fig 2A), the dimer produced by the intermolecular crosslinking reaction is analyzed to be as high as 33 (±5) % of the original sample by using ImageJ analysis [34].
Other divalent cations such as Mg(II), Co(II), and Mn(II) are performed in the chemical crosslinking experiments, too. None of these divalent cations can induce crosslinked dimerization of wild type CzcS SD, which shows that Zn(II) is specific for this crosslinking (S7A Fig). However, the Co(II) show its ability in inducing the crosslinked dimerization of mutant CzcS SD D60C (S7B Fig). The proportion of dimer induced by Co(II) is less than that of Zn(II) (S7C Fig). This may be caused by the different coordination geometry preferred by Co(II) and Zn(II). As the data from Protein Data Bank (PDB) and Cambridge Structural Database (CSD), the octahedral geometry is preferred by Co(II) coordination and tetrahedral geometry is more preferred by Zn(II) coordination.
It is well known that proline residue will distort the regular structure of helices by introducing a kink between the segments adjacent to it [35–37]. In this study, we want to characterize whether the distortion of H1 and H1’ α-helices has any influence on the function of HK CzcS by the introduction of proline residue. Besides the residues adjacent to the C-terminal Zn(II) binding site, other residues along the H1 and H1’ α-helices are chosen to do the proline substitutions. The residues Gln52, Leu50, Leu48, Asn45, Arg43, and Arg41 arranged with almost equal interval are replaced by proline residues (Fig 5A). All these single proline substitutional mutants grow well in the low concentration of Zn(II) (0.5 mM) and respond to MEPM equivalently to that of wild type P. aeruginosa. However, most of the mutants (L50P, N45P, R43P, and R41P) display varying degrees of impairments in their resistance to higher concentration of Zn(II) (2.5 mM) and the Zn(II)-inducible cross-resistance to MEPM (Fig 5B). Unlike the proline substitutions, other non-conserved mutations (S2 Table) of these residues (Arg41, Arg43, and Arg45) do not profoundly affect the signaling response. We also find that some proline substitutions still play full functions in Zn(II) induced metal detoxification and MEPM antibiotic resistance, such as Q52P and L48P (Fig 5B). Further, we also do some double proline substitutions along the H1 and H1’ α-helices (S8 Fig). All the double mutants (R41P L48P, R43P L48P, N45P L48P, R41P N45P, R43P N45P, and R41P R43P) totally lose the functions in Zn(II) induced metal detoxification and MEPM antibiotic resistance.
TCSs are frequently used by bacteria to adapt to the dynamic environments by coping with external stimuli [1–5]. Since their first discovery approximately 25 years ago, TCSs have been extensively identified in microorganisms. Although more and more crystal structures of HKs have been determined in the last few years [38, 39], the crystallographic analyses of Zn(II)-binding senor domains of HKs have not been reported to date. What’s more, the mechanisms by which extracellular stimuli are transduced from the sensor domain to the intracellular kinase domains are one of the least understood aspects of TCS response [38]. With growing concerns about co-regulation between heavy metals and antibiotic resistance [13], the CzcR-CzcS TCS of the pathological bacterium P. aeruginosa is an excellent candidate to study on. Here, we present a high-resolution structure of the CzcS SD in complex with its cognate ligand, Zn(II). The characterized Zn(II)-bound CzcS SD is demonstrated as a functional dimer with its central parallel bundle formed by the H1 and H1’ α-helices (Fig 2A). The subunit of the CzcS SD reveals characteristic PDC folds with N-terminal helix-loop-helix domain that leads into the central five anti-parallel β-sheet scaffold (Fig 1C). The N-terminal and C-terminal ends of the CzcS SD are orient in parallel, which allows communication of the connected transmembrane segments with the structural character of dimeric four-helical bundles [40].
The topologically similar PDC members, such as PhoQ, DcuS, CitA, DctB, and CzcS sensors, exhibit an enormous sequence variability (S2 Fig). Within the common structural arrangement, the sequence varieties enable them to detect diverse stimuli. Most of the characterized PDC sensor domains (CitA, DcuS, and DctB) detect their cognate ligands by the internal cavity that is formed by the central conserved β-sheets (Fig 1C) [20, 22]. Differently, the extracellular sensor domain of CzcS utilizes the residues from the symmetrical N-terminal α-helices to coordinate with Zn(II) in a tetrahedral coordination geometry. This constitutes a special class of Zn(II) binding sites that form at the dimer interface in the biochemical Zn(II) sites [24]. The recently reported metal-ion sensor CusS binds the effector at the dimer interface as well. However, the CusS SD interacts with the Ag(I) by using the N-terminal and C-terminal α-helices separately from different monomers. In the crystal structure of CzcS-Zn, the average distance is approximately 14.9 Å between the H1 and H1’ α-helices (Fig 2A). In the absence of outer-shell constraints (S9A Fig), the H1 and H1’ α-helices are flexible in rearranging the structural orientation. Their minor reorientation will initiate a large readjustment that affects the Zn(II) binding site (S9B Fig). These structural features enable the rapid regulation of the active site for Zn(II) binding or releasing. It is similar to that of the Zn(II) binding site confined between TM2 and TM5 in the Zn(II) transporter YiiP [41].
In the CzcS SD, the coordination environment is symmetrical with the His55, Asp60, His55’, and Asp60’ residues from the H1 and H1’ α-helices, respectively (Fig 2A). This class of Zn(II) ligands that comprise the His and Asp residues is rare for the reported biochemical Zn(II) sites [42]. It makes the CzcS SD bind to Zn(II) with an affinity of 1.7 (±0.2)×10−6 M. The RT-PCR analysis also indicates that the expression levels of czcS, czcR, czcC as well as oprD have obviously up-regulation or down-regulation when the P. aeruginosa is stimulated by Zn(II) at a micromole level (S10 Fig). When the Zn(II) ligands of HK CzcS are substituted by cysteine residues, the mutants H55C and D60C can respond to Zn(II) stimulus as well and bind Zn(II) with higher affinities than that of the wild type construct in vitro. For the D60C mutant, a classic Cys2His2 zinc finger configuration with a tetrahedral coordination geometry (S11A Fig) can be properly formed by the residues Cys60, Cys60’, His55, His55’ with Zn(II) [43]. The similar coordination geometry to that of the wild type HK CzcS causes the D60C mutant to display equivalent activities in sensing and regulating Zn(II) signal (Fig 2C). We speculate that a linear coordination geometry may be formed by Cys55 and Cys55’ with Zn(II) in the H55C mutant, which is similar to the configuration formed on the dimer interface of the colicin E3 immunity protein (S11B Fig) [42, 44]. The H55C mutant strain maintains the ability to respond Zn(II) stimulus as well (Fig 2C).
The biologically relevant dimer is observed in the crystal structure of CzcS-Zn complex with Zn(II) binding at the dimer interface (Fig 1B). We ever made great efforts but failed in crystallizing the CzcS SD in the absence of Zn(II). The difficulties in crystallization may be predominantly caused by the high flexibility of the CzcS SD especially the swing of N-terminal α-helices. In the absence of Zn(II), the CzcS SD is exited as monomer in solution. When it binds to Zn(II), the CzcS SD transforms from monomer to dimer and seems to be more conformational stable with the H1 and H1’ α-helices confined by Zn(II) coordination. Along with in vivo biological evidences, the Zn(II) induced dimerization of the CzcS SD is supposed to be physically important for signal regulation.
This speculation is also confirmed by the different effects of Co(II) on the regulation of antibiotic resistance between wild type and mutant D60C strain (Fig 3). The D60C strain turns to be a Co(II)-responsive regulator that shows Co(II)-inducible resistance to MEPM. In vitro, the crosslinked dimerization of sensor domain induced by Co(II) is also observed for the D60C mutant. These experiments again indicate that the association of H1 and H1’ α-helices is necessary for the activity of HK CzcS. What’s more, the association state of H1 and H1’ α-helices maintains till to the linker region when HK CzcS is in the activated form. In the case that the original Zn(II) binding site is destroyed (mutant H55A), a cysteine substitution in the linker region (mutant L38C H55A) can bind to Zn(II) with high affinity in vitro and strongly respond to the Zn(II) stimulus in vivo as well (Fig 4).
The rigid structural features of H1 and H1’ α-helices are the pivotal guarantee for the signal transduction, and this is verified by the proline substitutional experiments. Proline residues are known to distort the structure of helices [45, 46]. Visual inspection of some helices with a proline residue demonstrates a range of helix distortion (S12 Fig). Obvious kink angle is observed in some proline-containing α-helices (S12A Fig), which may happen to the H1 and H1’ α-helices in the mutant strains L50P, N45P, R43P, and R41P. This kind of distortion of H1 and H1’ α-helices makes the mutant strains (L50P, N45P, R43P, and R41P) seriously impair in the functions of Zn(II) induced metal detoxification and antibiotic resistance. While, there also exist some proline-containing α-helices that are approximately straight (S12B Fig). It may be the reason that why the growing status of mutants Q52P and L48P is not influenced by the introduction of proline residue on H1 and H1’ α-helices. In addition, the H1 and H1’ α-helices can’t resist the double proline substitutions, which lead serious impairments in the abilities of responding to Zn(II) signal. Thus, it’s important to keep the conformation of H1 and H1’ α-helices in the signal transduction.
As above described, the H1 an H1’ α-helices are the key factors in the activity of HK CzcS. They interact with the Zn(II) and keep the Zn(II)-induced association state till to the linker region, which are physiologically important for extracellular signal sensing and transduction across the transmembrane helices to the cytoplasmic kinase (Fig 6). By other group research, the rearranged helical interactions are discovered within the dimeric four-helical bundles in the transmembrane domain when HK CzcS is activated [47]. There is a transition from intramolecular- to intermolecular-crosslinking within the transmembrane helices (Fig 6) [47]. We speculate it’s the association of H1 an H1’ α-helices that leads the structural rearrangements in the sensor domain, which will drive the interactional displacements of the helix bundles in the transmembrane domain. The aforementioned quaternary structural changes within the homodimer ultimately lead the trans autophosphorylation in cytoplasmic kinase domain on the conserved histidine residues (Fig 6) [48–51]. The promising model (Fig 6) presented here provides preliminary insights into the molecular mechanism of Zn(II) signal sensing and transduction by HK CzcS. It gives an implication for understanding the Zn(II) induced metal detoxification and antibiotic resistance in CzcR-CzcS TCS of P. aeruginosa. However, besides the structural information of extracellular sensor domain provided in this study, the structural characterizations of transmembrane domain and cytoplasmic kinase domain have not been reported to date for HK CzcS. Thus, further investigations are still needed to precisely characterize how the signal in the sensor domains results in the interactional rearrangement of the transmembrane helices and modulates the autophosphorylation events in the cytoplasmic kinase domains.
The DNA fragment of the CzcS SD (amino acids 40–166 in CzcS protein) was amplified from P. aeruginosa genomic DNA and cloned into the NheI and HindIII sites of the pET-28a vector (named pCSET). E. coli BL21 (DE3) cells transformed with the construct were cultivated in the LB medium supplemented with 30 μg/ml of kanamycin. The cells were cultivated at 37°C with constantly shaking at 250 rpm following a 1:100 inoculation from an overnight culture. Expression was induced with 0.5 mM IPTG when the culture reached an optical density of OD 600 ≈ 0.6. The induced cells were grown for 4 h at 30°C, and subsequent steps were performed at 4°C. Cells expressing the CzcS SD with an N-terminal His6-Tag were harvested by centrifugation and lysed by sonication on ice in 15 ml of lysis buffer (10 mM Tris-HCl, pH 7.4, 100 mM NaCl, 0.1 mM PMSF, 10% glycerol, and 1 μl DNaseI). The supernatant was obtained by centrifugation at 12,000 rpm for 15 min and loaded onto a 5-ml Ni-NTA column that was pre-equilibrated with 2–3 column volumes of buffer A (10 mM Tris-HCl, pH 7.4, 100 mM NaCl, and 25 mM imidazole). The fusion protein was eluted in a linear gradient with the concentration of imidazole ranging from 75 mM to 500 mM in buffer A. The N-terminal His6-tag was removed by digesting the fusion protein with a protease overnight. The His6-tag-cleaved protein was treated with 0.5 mM EDTA and purified on an 8-ml MonoQ anion-exchange column which was equilibrated with buffer B (10 mM Tris-HCl, pH 7.4, and 50 mM NaCl). The protein was eluted from the MomoQ column with 100 mM NaCl in buffer B. The purified CzcS SD were identified by SDS-PAGE and used in the following experiments.
The CzcS SD mutants (CzcS SD H55C and CzcS SD D60C) were obtained by using site-directed mutagenesis technology performed on the pCSET plasmid. The CzcS SD L38C H55A mutant (amino acids 27–175 in CzcS protein) was constructed in the same method as pCSET plasmid followed by site-directed mutagenesis of sites Leu38 and His55. All the CzcS SD mutants were expressed and purified in the same procedures as wild type CzcS SD.
To crystallize the CzcS-Zn, the CzcS SD with the concentration of 3–4 mg/ml was mixed with an equimolar amount of ZnSO4 in the buffer containing 10 mM Tris-HCl, pH 8.5, and 100 mM NaCl. The complex crystals were grown at 16°C by the sitting-drop vapor-diffusion against the reservoir buffer containing 0.2 M (NH4)2SO4, 0.1 M Bis-Tris pH5.5, and 25% w/v polyethylene glycol 3350. The irregular cuboid crystals came out after two days and continued to grow until reaching a suitable size for X-ray diffraction studies. The crystals were briefly soaked in a cryoprotectant containing 0.2 M (NH4)2SO4, 0.1 M Bis-Tris pH5.5, 25% w/v polyethylene glycol 3350, and 8% glycerol prior to flash-frozen in liquid nitrogen.
The diffraction datasets at the zinc K-edge were collected from single crystals at BL17U beamlines at the Shanghai Synchrotron Radiation Facility [52]. The X-ray diffraction datasets were integrated and scaled with the HKL2000 package software. The initial phase for automated model building was solved by zinc single-wavelength anomalous dispersion using the Phenix software [53]. Iterative rounds of refinement were performed using the Phenix software, which were followed by manual alterations using the WinCoot software [54]. Refinement was conducted until no significant improvements were achieved. All structural models for the current study were generated with the PyMOL software [55]. The data collection and refinement statistics are listed in Table 1. The atomic coordinates and structural factors for CzcS-Zn have been deposited into the Protein Data Bank with accession code 5GPO.
The czcS-deficient strain of P. aeruginosa was constructed with a homologous recombination assay [56]. A 2025-bp PCR fragment corresponding to the first 8 bp of the czcS gene was amplified from P. aeruginosa genomic DNA with the primers I and II which contain an EcoRI and an XbaI restriction sites, respectively. Another 1892-bp PCR fragment that contains the 3’-end of the czcS gene was amplified from P. aeruginosa genomic DNA with primers III and IV which contain an XbaI and a HindIII restriction sites, respectively. The intervening gentamicin resistance cassette was amplified from the pPS858 plasmid with the XbaI restriction site both at 5’- and 3’-end. The aforementioned three DNA fragments were ligated into an EcoRI /HindIII-cleaved pEX18AP plasmid. The constructed plasmid was transformed into the P. aeruginosa competent cells by electroporation [56]. The successfully homologous recombinants were screened on the LB medium containing 30 ug/ml gentamycin. The czcS-deficient strain was further identified by PCR and DNA sequencing. The oligonucleotides used to construct the czcS-deficient strain are listed in S3 Table.
The czcS operon and its encoding gene were amplified from P. aeruginosa genomic DNA. They were ligated by overlap PCR and cloned into the HindIII/BamHI restriction sites of pAK1900. The complementary plasmid pCSAK was verified by DNA sequencing and used to construct the mutants. Site-directed mutagenesis of CzcS was performed on the yielding pCSAK plasmid using the Quik Change site-directed mutagenesis kit (Agilent Technologies). All the variants were verified by DNA sequencing. The oligonucleotides used for the site-directed mutagenesis plasmids are listed in S3 Table.
The pCSAK plasmid was transformed into the czcS-deficient strain by chemical transformation to supply as the complementary strain. The empty pAK1900 plasmid was transformed into the wild type P. aeruginosa and czcS-deficient strain to supply as the positive control and negative control, respectively. All the variants were separately transformed into the czcS-deficient strain. The aforementioned strains were cultivated in LB mediums supplemented with 150 μg/ml carbenicillin. They were grown overnight at 37°C with constantly shaking at 250 rpm/min. Fresh LB mediums containing 150 μg/ml carbenicillin were inoculated with the overnight cultures at the proportion 1:100 and grown at 37°C until the density of OD 600 arrived 1.0. The cultures then underwent ten-fold serial dilutions with five gradients and were further seeded onto the LB plates with varying concentrations of Zn(II), Co(II) or MEPM. The plates were incubated at 37°C for 16 h before observation.
The chromogenic chelating agent PAR was selected as the competitor in the spectrometric determinations of binding affinity of Zn(II) with wild type and mutant CzcS SD (CzcS SD H55C, CzcS SD D60C, and CzcS SD L38C H55A). All Zn(II) binding experiments were performed under photophobic condition at 22°C in the buffer containing 10 mM Tris-HCl, pH 7.4, 150 mM NaCl, and 0.5 mM TCEP (for the mutants CzcS SD H55C, CzcS SD D60C, and CzcS SD L38C H55A only). A known concentration of PAR solution (36 uM) was mixed with the purified protein (50 uM-200 uM). The mixtures were divided into equal volumes followed by loading consistent volumes of ZnCl2 with increasing concentrations (in the range of 2 uM-42 uM). The UV-visible spectrum were recorded in the range of 200 nm to 700 nm until the reaction systems achieved competitive equilibrium. The titration data at 500 nm were fit with Dynafit software [30] by using one-site model to obtain the apparent dissociation constants of Zn(II) with wild type and mutant CzcS SD.
The CzcS SD was purified in the same procedures as described above in the buffer containing 10 mM HEPES, pH7.4, 100 mM NaCl. The primary amine reactive crosslinker BS3 was stored in DMSO at 100 mM and diluted to 1 mM in 20mM HEPES (pH7.4) immediately before use. A 50-fold molar excess of BS3 crosslinker was loaded into the CzcS SD and CzcS-Zn samples with a final concentration of 1 mM. The reaction systems were incubated at room temperature for 30 minutes and quenched with 50 mM Tris-HCl, pH 7.4. The quenching reaction was incubated at room temperature for 15 minutes. The CzcS SD, CzcS-Zn, and the products of the crosslinking reactions were analyzed by 14% SDS-PAGE and quantified by ImageJ [34]. The crosslinking experiments of other divalent cations, such as Mg(II), Mn(II), or Co(II), were performed in the same procedure.
The quantitative real-time RT-PCR with the rpsl gene as the reference was performed to monitor the expression changes of czcS, czcR, czcC and oprD genes when P. aeruginosa is stimulated by the Zn(II) at a micromolar level. The total RNA was extracted by traditional phenol-chloroform method and reverse transcribed by iScript cDNA Synthesis Kit (Bio-Rad). The cDNA samples were diluted for different folds and used as the templates in the PCR experiments. The real-time RT-PCR was performed on the Bio-Rad CFX96 equipment using the Ssofast SYBR Green Supermix (Bio-Rad). The experiments were performed at least three independent times with average results shown. The primer sequences used for real-time RT-PCR are designed using the Primer3 program and listed in S3 Table.
|
10.1371/journal.ppat.1003185 | Nosema ceranae Escapes Fumagillin Control in Honey Bees | Fumagillin is the only antibiotic approved for control of nosema disease in honey bees and has been extensively used in United States apiculture for more than 50 years for control of Nosema apis. It is toxic to mammals and must be applied seasonally and with caution to avoid residues in honey. Fumagillin degrades or is diluted in hives over the foraging season, exposing bees and the microsporidia to declining concentrations of the drug. We showed that spore production by Nosema ceranae, an emerging microsporidian pathogen in honey bees, increased in response to declining fumagillin concentrations, up to 100% higher than that of infected bees that have not been exposed to fumagillin. N. apis spore production was also higher, although not significantly so. Fumagillin inhibits the enzyme methionine aminopeptidase2 (MetAP2) in eukaryotic cells and interferes with protein modifications necessary for normal cell function. We sequenced the MetAP2 gene for apid Nosema species and determined that, although susceptibility to fumagillin differs among species, there are no apparent differences in fumagillin binding sites. Protein assays of uninfected bees showed that fumagillin altered structural and metabolic proteins in honey bee midgut tissues at concentrations that do not suppress microsporidia reproduction. The microsporidia, particularly N. ceranae, are apparently released from the suppressive effects of fumagillin at concentrations that continue to impact honey bee physiology. The current application protocol for fumagillin may exacerbate N. ceranae infection rather than suppress it.
| Nosema ceranae, a microsporidian pathogen described from Asian honey bees, was discovered infecting European honey bees coincident with early reports of Colony Collapse Disorder (CCD) and has been suggested to be a factor in honey bee declines. Use of fumagillin, the only known reliable treatment for the naturally occurring Nosema apis, subsequently increased but controversial field results were reported for N. ceranae. Fumagillin suppresses reproduction of microsporidia but disease prevalence and hive performance in treated apiaries were similar to untreated apiaries 6 months after treatment. We compared responses of N. apis and N. ceranae to the antibiotic at concentrations that simulated fumagillin degradation in hives during the honey bee foraging season when use of the drug is discontinued. N. ceranae spore production recovered at a higher fumagillin concentration than N. apis. At lower fumagillin concentrations, significantly more infective N. ceranae spores were produced in treated bees than in untreated infected bees. Protein profiles of bees fed fumagillin confirmed our hypothesis that fumagillin affects bee physiology at concentrations that no longer suppress N. ceranae. Use of fumagillin may increase the prevalence of N. ceranae and is potentially a factor in replacement of N. apis by N. ceranae in US apiaries.
| Bicyclohexylammonium fumagillin, an antibiotic isolated from the fungus Aspergillus fumigatus, has been the only widely used treatment for nosemosis, or “nosema disease”, in western honey bees, Apis mellifera, [1], [2] for nearly 60 years [2]. The antibiotic (hereafter “fumagillin”), in the form of a 3% concentration for veterinary use, is considered to be the only effective treatment for Nosema apis infection and also suppresses the recently discovered microsporidian pathogen, Nosema ceranae, in honey bees [3].
N. ceranae, originally isolated from the Asiatic honey bee, Apis cerana [4] was discovered infecting A. mellifera in 2004 [5], increasing concerns about the impact of nosema disease on honey bee health. Microsporidia were correlated with declining populations of honey bees in the US [6], [7] and Spain [2]. Although fumagillin can control N. ceranae as well as N. apis at the manufacturer's recommended concentrations [3], several field studies have contradicted these results [8], [9] but no in-depth studies have been published. Since the discovery of N. ceranae, fumagillin sales have increased, and residues of the antibiotic were detected in harvested honey in the U.S. [10]. To reduce the residues, fumagillin treatment is prohibited during the foraging season [11], a time period that exceeds 6 months in most areas of the U.S. [12]. Hives are typically treated with the antibiotic once in the late fall and once in the early spring, usually prophylactically [13]. Fumagillin persists inside hives [2], and degrades over time [14].
The practice of periodic fumagillin treatment results in decreasing but nearly constant exposure of multiple generations of bees and pathogens to the drug. Although this practice appears to provide an environment conducive to selection of fumagillin-resistant Nosema strains, N. apis has evidently not developed resistance to the drug; however, studies have shown that N. ceranae can reestablish to pretreatment prevalence 6 months after treatments are terminated [2], [8]. Lower natural susceptibility to fumagillin or faster recovery from treatment could be a factor in the replacement of N. apis by N. ceranae, which apparently has occurred in North America and elsewhere [15], [16], [17].
Fumagillin inhibits the enzyme methionine aminopeptidase-2 (MetAP2) [18] and is known to block MetAP2 in Encephalitozoon cuniculi, a microsporidian pathogen of humans [19]. Most eukaryotes possess genes for two MetAP isoforms, MetAP1 and MetAP2, and apparently require either MetAP1 or MetAP2 to survive [18]. Microsporidia do not possess the MetAP1 gene [19], [20], making MetAP2 a logical target for suppression of microsporidian infection. The microsporidian MetAP2 gene, however, is homologous with MetAP2 genes in other eukaryotes, with approximately 60% similarity among all eukaryotic organisms [21]. Although fumagillin and analogous drugs are currently used for treatment of human microsporidiosis and certain cancers [22], fumagillin is known to be toxic to humans and other vertebrates by interacting with the MetAP2 enzyme, which is involved in protein maturation and post translation processes [23]. Honey bee queens and workers feeding on fumagillin have been shown to have significantly shorter lifespans [13], [24] but the potential toxicity of the antibiotic to honey bees when used to control nosemosis has received little study. Possible negative effects of fumagillin in insects, however, have been demonstrated in the greater wax moth, Galleria mellonella [25].
We fed low fumagillin concentrations to honey bees to evaluate effects of diminishing concentrations of the drug reported in bee hives [14] and documented increased production of N. ceranae spores and, to a lesser extent, N. apis spores in the treated bees. To determine if differences in susceptibility of Nosema spp. to fumagillin are reflected in MetAP2 sequences among apid species, and if honey bees are potentially susceptible to the drug, we compared MetAP2 sequences of the honey bee and the three described apid Nosema species, including Nosema bombi, a commonly observed pathogen of bumble bees, Bombus spp. [26], [27], [28]. N. bombi shares a close phylogenetic relationship with N. apis and N. ceranae but is not responsive to fumagillin treatment [29]. Based on MetAP2 sequence similarity and shorter lifespans of bees treated with fumagillin [13], [24], we hypothesized that fumagillin could also interact with the MetAP2 enzyme in honey bees. Computational comparison based on MetAP2 sequences of the pathogens and the honey bee is not yet optimal and there is no available in vivo enzyme dynamic comparison method; therefore, we performed 2D-gel electrophoresis (2DE) to evaluate the protein profiles in midgut tissues of uninfected honey bees fed concentrations of fumagillin corresponding to the bioassays of infected and treated bees. Our results suggest that declining levels of fumagillin in treated hives provide a window for hyperproliferation of microsporidia and that fumagillin continues to interfere with honey bee midgut physiology at levels that no longer suppress reproduction and maturation of N. ceranae and N. apis.
Nosema apis was provided by T. Webster at Kentucky State University and N. ceranae was isolated from honey bees from the University of Illinois at Urbana-Champaign apiary using methods identical with those used in previous studies [5]. N. bombi was isolated from Bombus pensylvanicus midgut tissues that were stored in liquid nitrogen as previously reported [28]. N. apis and N. ceranae were reproduced in caged bees, and mature spores were harvested from midgut tissues. Tissues were homogenized in glass tissue grinders, filtered through fine weave hardware mesh and centrifuged. Spore pellets were resuspended in sterile tap water and counted for immediate use in bioassays.
Brood frames from fumagillin-free colonies were held in growth chambers at 34.5°C, 65% relative humidity, 24 h dark. Newly emerged bees were transferred on a daily basis to cages consisting of 480 ml HDPE lidded plastic cups with tops cut out and screened with 3-mm hardware cloth [30]. The bees were fed with 50% sugar water (w/w), and pollen patties (15% pollen, Megabee) ad libitum. Five-day post-emergence adult bees were used for all bioassays; bees from four different hives that had not received fumagillin treatments for at least one year were used for trials conducted in 2011 and 2012. Bees were immobilized on ice, secured to a foam board with insect pins, and orally inoculated with 105 spores of either N. apis or N. ceranae in 2 µl sugar water using a micropipetter. This dosage was selected to exceed the IC100 level of approximately 2×104 spores (unpublished data for this N. ceranae isolate). Additional bees were randomly selected from the same brood frames and treated with sugar water without spores to verify that experimental bees had no background infection (negative control). Inoculated and negative control bees were transferred to new cages, 30 bees per cage per treatment, and held in growth chambers (30°C; 65% RH) after treatment. Beginning 24 h post inoculation until the experiment was terminated at 20 days post inoculation (dpi), inoculated bees were fed 50% sugar water ad libitum with selected concentrations of fumagillin. The tested concentrations included the manufacturer's recommended concentration of 25 mg/l or 1.0×, and 0.02, 0.01, 0.002, 0.001, 0.0002, 0.00006, 0.00001, and 0.0000033× the recommended concentration, and no fumagillin treatment as a positive control (Table 1). We focused on N. ceranae, currently the dominant microsporidian pathogen in US apiaries, but also conducted a limited number of tests of N. apis-infected bees (Table 1).
Beginning 10 dpi for infected bees treated with 0.01, 0.001× fumagillin concentrations and untreated positive controls, five bees were randomly removed from each cage for evaluation, then five bees were sampled every 2 days until 20 dpi. Bees fed all other fumagillin concentrations were evaluated at 14 dpi based on the midpoint of peak spore production in the midgut tissues. Midgut and hindgut tissues were excised and separated before counting spores. Because caged bees seldom defecated inside the cages, the spore count in the hindgut was considered to be the accumulation of spores released from midgut cells for entire period of infection. The spores were isolated for counting by homogenizing the tissues of five bees per sample, midgut and hindgut separately, in a glass tissue grinder in 0.5 and 2 ml sterile distilled water, respectively, and were counted using a Petroff Hausser counting chamber under phase-contrast microscopy. Developmental stages of the spores, including primary (internally infective) spores, germinated primary spores, immature environmentally resistant spores (environmental spores) and mature infective environmental spores, were distinguished by refringence and morphological characters [31], and counted. Counts of mature spores were analyzed using one-way ANOVA (SPSS statistic software, IBM).
DNA was extracted from spores of N. apis and N. bombi using Chelex [28]. Each spore sample was mixed with Chelex buffer (5% Chelex, 5% Tween20, and 1 ng/ml proteinase K) and incubated in a thermocycler, 2 hr 56°C, 30 min 95°C. The samples were centrifuged at 13,000×g for 10 min and the supernatant containing DNA solution was used for amplification. The sequence of the N. ceranae MetAP2 gene was obtained from GenBank (acc. no. XM002996491). The degenerate primer set (NMetAP2F: GRG CDG CVG ARG CWC AYA G; NMetAP2R: TCR TCR CCT YTT GTW AGR AYY TC) was designed based on the alignment of MetAP2 genes of N. ceranae and Encephalitozoon spp. (GenBank acc. nos. AF440270, XM00307371, AY224694) and was used to amplify the MetAP2 gene from N. apis and N. bombi. Platinum taq (Invitrogen) was used for PCR following the manufacturer's suggested protocol with 3 nM (final concentration) of degenerate primers at annealing temperature 49°C. The DNA fragment was cloned into pGemT easy vector (Promega) and transferred into DH5α competent cells, and the DNA insertion was sequenced using vector primers. The sequences were identified and compared using BlastX, then aligned using ClustalX. The phylogenetic tree was analyzed by maximum likelihood using PhyML 3.0 [32] with settings suggested by ModelGenerator [33] with 1,000 times bootstrap.
Sugar water with fumagillin concentrations of 0.0, 1.0, 0.01 and 0.001× the manufacturer's recommended concentration were fed ad libitum to uninfected honey bees, 20 bees per cage. After 10 days of feeding, midgut sections of the alimentary tract were excised and cut vertically to remove the gut contents and peritrophic membranes. The tissues were homogenized in sterile phosphate buffered saline (PBS) and protein samples were prepared using Genotech Focus Total Proteome Kit with 1× protease inhibitor cocktail. The protein samples were processed using Genotech Perfect Focus dissolved in IPG rehydration buffer (GE) and loaded on IPG focusing strips (13 cm, pH 3–10, GE) for first dimension processing, and on BioRad Criterion 12% polyacrylamide gels in 1× MOPS buffer for the second dimension. The gels were stained in Sypro Ruby and scanned using a Typhoon 9400 multi-laser scanner. Major protein spots were excised from the 2DE gels and subjected to in-gel trypsin digestion and protein identification using ESI-LC/MS. Proteins were identified in Mascot (Matrix Science) and used standard protein BLAST to search NCBI-NR database specific for Apis mellifera proteins.
Partial MetAP2 gene sequences of Nosema apis and N. bombi have been deposited in the GenBank database, accession nos. JQ927010 and JQ927011, respectively. The accession number of N. ceranae MetAP2 is XM002996491 [20] in GenBank. MetAP2 accession numbers for honey bee and human amino acid sequences are XP624161 and NP006829, respectively. The accession numbers of other MetAP2 sequences in analyses are AAC05144 (Drosophila melangonster), S45411 (Saccharomyces cerevisiae), AF440270 (Encephalitozoon cuniculi), AY224694 (E. hellem), AY224693 (E. intestinalis), AEI69245 (Encephalitozoon sp.).
Counts of N. ceranae spores in midgut tissues and hindgut contents varied significantly among fumagillin treatments and compared to untreated infected (positive control) bees. No spores were detected in bees inoculated with sugar water only (negative controls). Production of spores in midgut tissues reached a plateau phase with highest production capacity at 10 dpi and remained similar during the entire sampling period, 10–20 dpi, for both N. ceranae and N. apis infections (data not shown). The number of spores in the hindgut contents increased continually over the same period (Fig. 1A) but, after 16 dpi, spore counts were more irregular, perhaps the result of selecting surviving bees or those that defecated in the cages. We used 10–16 dpi data to generate a linear regression of spore accumulation in the hindgut (Fig. 1B). Slopes of regression lines for N. ceranae and N. apis, indications of daily spore production, were significantly different for all fumagillin concentrations. The difference between N. ceranae and N. apis is greater for honey bees treated with fumagillin, 165% and 129% higher for N. ceranae with 0.01× and 0.001× fumagillin, respectively. The slope was 30% higher for N. ceranae when bees were not treated with fumagillin. Unlike the 2011 cohorts, the 2012 test and negative control bees defecated in the cages and hindgut contents were lost, therefore spore counts were only made for midgut tissues in 2012.
The manufacturer's recommended concentration of fumagillin (1.0×) suppressed reproduction of both Nosema species, and the spore counts for midgut tissues and hindgut contents during the sampling period were below the resolution of the counting chamber. At a concentration of 0.04×, N. ceranae produced significantly fewer spores in the midgut tissues (P<0.001) than untreated controls; spore counts trended lower than the positive control at 0.02× (not significant; P = 0.086) but N. ceranae production began to recover at this concentration. At a concentration of 0.01×, N. ceranae produced a similar number of mature spores as positive control bees (P = 0.98 for midgut tissues; P = 1.00 for hindgut contents). In contrast, N. apis remained significantly suppressed at 0.01× (Table 1; P = 0.047 and 0.028 for midgut and hindgut, respectively). Both Nosema species were released from fumagillin suppression at 0.002× the recommended concentration. Significantly higher numbers of N. ceranae mature spores were produced in midgut tissues of infected bees treated with concentrations lower than 0.01× and higher than 0.0000033× the recommended fumagillin concentration than in positive control bees (P<0.05; Fig. 2). There was no significant difference in spore counts for N. apis infected bees treated with ≤0.002× fumagillin concentrations and positive controls. Although, the mean N. apis spore counts in hindgut contents trended higher (approximately 30% higher than in positive control bees), they were not statistically significant over three trials. With the exception of the recommended concentration, at all fumagillin concentrations tested, both Nosema species produced sufficient numbers of mature spores to infect numerous bees (Table 1). Hyperproliferation of N. ceranae was strongest at fumagillin concentrations between 0.001 and 0.0002×, then decreased with lower concentrations, although spore production was still significantly higher than positive controls at 0.00001 (1E−5)×(P<0.01). The lowest concentration tested, 0.0000033 (3.3E−6)×resulted in insignificant enhancement of spore production (P = 0.182).
Midgut spore counts corresponded with −log10 fumagillin concentrations (P<0.01) and the regression curve (Fig. 2A) predicts that the spore count of treated infected bees will equal the average counts of positive controls (no fumagillin) at 6.026 E−6×the recommended concentration. When data are normalized among trials to reflect the ratio between N. ceranae-infected positive control bees (no fumagillin) with treated infected bees, the predicted fumagillin concentration at which hyperproliferation no longer occurs is 2.239 E−6×(Fig. 2B).
We investigated differences in the MetAP2 gene among the Nosema species that infect apid bees, N. bombi, N. ceranae and N. apis, and compared them to the honey bee and human MetAP2 genes. Degenerate primers amplified a partial coding domain sequence of the MetAP2 gene, 868 bp for N. bombi (GenBank acc. no. JQ927011) and 926 bp for N. apis (GenBank acc. no. JQ927010). These sequences lacked base pairs for 65 amino acids in the C-terminal and two amino acids in the N-terminal end that are reported for N. ceranae (GenBank acc. no. XM002996491) but included all fumagillin binding sites and metal ion coordinate sites necessary to evaluate the interaction between fumagillin and MetAP2. Similarity was 83% between N. ceranae and N. bombi, 73% between N. ceranae and N. apis, and 82% between N. apis and N. bombi.
The MetAP2 genes from Apis mellifera (GenBank acc. no. XP624121), N. ceranae, N. apis, N. bombi and the mammalian microsporidium Encephalitozoon cuniculi (GenBank acc. no. AF440270), were translated using standard codec and aligned (Fig. 3). Binding site and coordinate site amino acid sequences were identical for honey bee and human MetAP2, and were identical among the microsporidia; Nosema spp. sequences differed from those of honey bees and humans at two fumagillin binding sites.
We conducted protein assays (Fig. 4) to identify alterations in midgut proteins of uninfected honey bees caused by fumagillin treatment, including 1.0×, 0.01× and 0.001× the recommended concentration of the drug and no fumagillin treatment (control). Alterations in protein presence, quantity and position on 2DE gels were identified at all three fumagillin concentrations (Fig. 4). The ESI-LC/MS and Mascot (Matrix Science) library reliably identified 45 altered proteins related to energy metabolism, mitochondria, cellular structure and transport in the midgut tissues of honey bees (Fig. 4; Table 2). Alpha-actin protein isoforms were located in different positions on 2DE gels that correlated with fumagillin concentration and incremental changes were noted (Fig. 4). Another fumagillin concentration dependent protein was an isoform of alpha-glucosidase II; three isoforms were found in all treatments but with different ratios among the isoforms. These isoforms originate from the same gene in honey bee genome (Table 2). H+ transporting ATP synthase beta subunit isoform 1, located in mitochondria, appeared to have a different molecular weight for each fumagillin treatment, and a voltage-dependent anion-selective channel protein was found on the high PH value side of the gel in all fumagillin treatments (Fig. 4D) but not in the positive control bees.
Our laboratory bioassays corroborated field observations [2], [3] that fumagillin suppresses both N. ceranae and N. apis at the manufacturer's recommended concentration, but the two microsporidian species responded differently to decreasing fumagillin levels. At a concentration of 0.01× (250 µg/L) the recommended concentration, N. ceranae produced a similar number of mature spores in treated and untreated bees while N. apis remained suppressed. At a concentration of 0.001×, N. ceranae produced significantly more spores in the host midgut tissues (approximately 40%) than N. apis at the same treatment regime, and 24% more than in untreated N. ceranae infected bees. In addition, the average number of N. ceranae spores in the hindgut was approximately 80% higher than N. apis spores at 0.001× fumagillin concentration, and 150% higher than in the hindgut contents of untreated N. ceranae-infected bees. Spore production of both Nosema species increased at lower levels of fumagillin residue, although not significantly for N. apis, and N. ceranae spore production was double that of N. apis. We demonstrated that very low levels of fumagillin residue, possibly below the detection limit [14], affect the interaction between the microsporidia and the host. The higher number of N. ceranae spores produced in treated honey bees could potentially increase the pathological effects and transmission of the microsporidium but mortality did not significantly differ among fumagillin treatments of infected bees in our trials.
N. apis and N. ceranae produced similar numbers of mature spores in midgut tissues of untreated infected honey bees, results that corroborate those of Forsgren and Fries [34] and Paxton et al., [35]; however, the spore counts from hindgut contents of untreated bees were significantly higher for N. ceranae infections (approximately 45%) than for N. apis (P = 0.037). When results of mature spore counts in the midgut and hindgut are combined, N. ceranae produced significantly more spores than N. apis (P = 0.031), corresponding to results of Martin-Hernandez et al. [36]. Our results perhaps resolve some disparity in spore counts noted among laboratories [37].
Disparity between the midgut and hindgut spore counts was noted for N. ceranae in the fumagillin trials. The slope of the growth curve representing accumulation of spores in the hindgut from 10–16 dpi (Fig. 1B) indicates that mature spores are produced in the midgut of bees treated with 0.001× fumagillin faster than in untreated infected bees. The difference increases if 18–20 dpi data are included, but these data are based on one trial. Unfortunately, the honey bees we used in 2012 voided the hindgut contents during the treatment period and we could not continue this investigation. Nevertheless, the results suggest that a “snapshot count” of mature spores in midgut may not fully indicate the speed of N. ceranae proliferation, a phenomenon that has been also been suggested from tissue observations [38].
Field tests of apiaries in Spain reported the degradation of fumagillin to approximately 0.001× of the applied concentration 3 months after treatment termination [14]. Although the US Federal Drug Administration disallows fumagillin usage during the foraging season, marketable honey in U.S. was found to contain 60 ng/g of fumagillin residue [10], approximately 0.0024× the recommended treatment concentration. We calculated that fumagillin degradation in the field should be approximately −log10/month based on residues in the U.S. and field results from Spain. Using these calculations, the period during which hyperproliferation of N. ceranae occurs is approximately 2 to 5.5 months after cessation of fumagillin treatment. A more comprehensive field study, however, may be necessary to confirm these estimates.
Our results suggest that N. ceranae can resume spore production significantly earlier than N. apis and hyperproliferation of N. ceranae results in more than twice the spore production of N. apis at the same conditions. Without fumagillin, N. ceranae produces only slightly more spores than N. apis (Table 1; Fig. 1B) and has no other known significant competition advantage [37]. Less susceptibility to fumagillin and hyperproliferation in the presence of low residues of the drug, however, suggest a mechanism by which N. ceranae may outcompete N. apis. Although use of fumagillin apparently has not selected a resistant strain of N. apis, it may provide an advantage to the pathogen that possesses more natural resistance.
We looked for differences in the MetAP2 gene among apid microsporidia that might provide information about differences in susceptibility to fumagillin. The MetAP2 gene and amino acid sequences are most similar between N. bombi and N. ceranae, 83 and 80% respectively, but N. apis and N. bombi share a closer relationship in phylogenetic analysis of MetAP2 sequences. This suggests that MetAP2 is not obviously different among these Nosema species, yet N. bombi is apparently not affected by fumagillin treatment [29]. The relationship between MetAP2 phylogeny and sensitivity to fumagillin indicated that MetAP2 may not be the only factor influencing response to fumagillin, and different responses have been reported for other microsporidia and hosts [39], [40], [41]. Nevertheless, the high similarity of the honey bee MetAP2 suggested the possibility that honey bees are affected by fumagillin at concentrations at which N. ceranae hyperproliferates.
MetAP2 is known to be involved in various post translation modifications of multiple proteins [23], but the full extent of its action is not known. Protein analysis of midgut tissues of uninfected bees fed different concentrations of fumagillin confirmed our hypothesis that fumagillin interacts with host MetAP2 at concentrations that no longer suppress Nosema spp. At 0.001× the recommended treatment concentration, changes in structural and metabolic proteins were less dramatic than at higher concentrations but were still observable (Fig. 4). Although we did not identify specific proteins that, when altered, would allow hyperproduction of microsporidian spores, several proteins existing in isoforms, e.g. alpha-actin and glucosidase, were altered in all fumagillin treated groups. Alterations in actin proteins were shown to be necessary for successful infection by intracellular apicomplexan parasites [42], [43]. Alpha glucosidase II isoforms are necessary in carbohydrate metabolism [44], and other altered proteins (Table 2) are located in mitochondria. Nosemosis in honey bees causes additional energy stress [45], therefore changes in proteins involved in energy metabolism could influence the progress of infection. It is possible that the alteration of isoforms results from the interference of MetAP2 function in protein modification; most of the altered proteins we identified have universal effects on cell function.
Possible synergic effects between pesticides and N. ceranae were previously reported [46], [47], [48], and we found that effects of fumagillin were surprisingly similar and possibly stronger than the pesticides tested when it degraded to low levels. In contrast to pesticide exposure during honey bee foraging, fumagillin is applied to the hive directly by bee keepers to treat nosemosis. We have not investigated whether fumagillin usage has consequences for infection by other pathogens, but it is clear that the unintended consequence of its use could be exacerbation of N. ceranae pathogenesis. Fumagillin treatment is known to reduce microsporidian reproduction and is probably useful for protecting weak colonies [2], but the antibiotic may have unintended effects on the honey bee host, ultimately contributing to increased prevalence and pathogenicity of N. ceranae. Many variables could affect fumagillin concentration in hives post-treatment, including hive size, nectar flow and other factors. Current fumagillin application involves a treatment gap of 6 months or more and almost guarantees that the antibiotic will degrade to concentrations that allow release of the microsporidia and result in fast recovery of N. ceranae [2], [8]. In addition, the time period of N. ceranae hyperproliferation may reverse the benefits gained at the beginning of fumagillin treatment, resulting in indistinguishable performance between fumagillin treated and untreated hives [9]. Although field studies are necessary to determine if fumagillin use has value in specific situations, it is clear that new treatments for nosema disease are needed. Identification and development of drugs that will target the microsporidia without serious impacts on host physiology are critical for control of nosemosis.
|
10.1371/journal.pntd.0004260 | Quantifying Poverty as a Driver of Ebola Transmission | Poverty has been implicated as a challenge in the control of the current Ebola outbreak in West Africa. Although disparities between affected countries have been appreciated, disparities within West African countries have not been investigated as drivers of Ebola transmission. To quantify the role that poverty plays in the transmission of Ebola, we analyzed heterogeneity of Ebola incidence and transmission factors among over 300 communities, categorized by socioeconomic status (SES), within Montserrado County, Liberia.
We evaluated 4,437 Ebola cases reported between February 28, 2014 and December 1, 2014 for Montserrado County to determine SES-stratified temporal trends and drivers of Ebola transmission. A dataset including dates of symptom onset, hospitalization, and death, and specified community of residence was used to stratify cases into high, middle and low SES. Additionally, information about 9,129 contacts was provided for a subset of 1,585 traced individuals. To evaluate transmission within and across socioeconomic subpopulations, as well as over the trajectory of the outbreak, we analyzed these data with a time-dependent stochastic model. Cases in the most impoverished communities reported three more contacts on average than cases in high SES communities (p<0.001). Our transmission model shows that infected individuals from middle and low SES communities were associated with 1.5 (95% CI: 1.4–1.6) and 3.5 (95% CI: 3.1–3.9) times as many secondary cases as those from high SES communities, respectively. Furthermore, most of the spread of Ebola across Montserrado County originated from areas of lower SES.
Individuals from areas of poverty were associated with high rates of transmission and spread of Ebola to other regions. Thus, Ebola could most effectively be prevented or contained if disease interventions were targeted to areas of extreme poverty and funding was dedicated to development projects that meet basic needs.
| Despite recognition that resource constraints contributed to the difficulty in controlling West Africa’s ongoing Ebola outbreak, no previous study has been conducted to quantify the impact of poverty on transmission. In particular, the extent to which within country heterogeneity in socioeconomic status (SES) could be driving geographic spread or sustaining Ebola hotspots has important implications for control policies. In Liberia’s capital city Monrovia, approximately 68% of the population occupies slum neighborhoods characterized by overcrowding and lacking sanitation—conditions conducive to outbreaks of infectious disease. The researchers find that infected individuals residing in these most impoverished settings are associated with three times more Ebola transmission, as well as greater dissemination of disease between communities of different socioeconomic levels. Cases from low and middle SES areas reported significantly more contacts during their symptomatic and therefore infectious periods than cases from high SES areas. Overcrowding and lack of education on how the disease is transmitted could explain the observed differences in number of secondary cases. The findings support the need for targeted response measures that account for differential risk arising from socioeconomic heterogeneity within West African countries.
| The 2014–2015 Ebola outbreak continues to have a global impact. Prevention and preparedness measures remain in place at airports and hospitals across the United States and Europe, while ongoing transmission in West Africa has led to a case count exceeding 28,450 [1]. Since October 2014, however, the epidemic has been diminishing. Despite a recent resurgence in Liberia, efforts have shifted from emergency response to endgame strategies. Fundamental to preventing or mitigating future outbreaks is the identification of factors that exacerbate risks of Ebola emergence, transmission and geographical dissemination. While it has been appreciated that insufficient healthcare infrastructure in Liberia, Guinea and Sierra Leone has been a primary obstacle in the treatment of Ebola, socioeconomic heterogeneities within these countries have not been evaluated as determinants of Ebola transmission. In Liberia, specifically, more than 1.4 million of the country’s extreme poor have incomes less than $0.50 per day [2], adult literacy rates are under 43%, and public spending allocated to health is only 4% of the GDP [3]. Approximately, 68% of its urban population resides in a network of slums [4] characterized by overcrowding, high crime, and lack of sanitation [5–7]. Even for the remainder of the population, there are stark differences regarding population density and improved sanitation between middle and higher SES.
To determine the role of poverty in the Ebola outbreak, we analyzed transmission chains using a time-dependent stochastic model that was adapted to evaluate the heterogeneity of Ebola incidence and transmission factors for over 300 communities within Liberia between February and December 2014. We found that cases from low and middle SES regions have significantly more contacts when infectious and lead to much greater transmission than cases from higher SES regions. Nonetheless, all SES regions responded to instrumental interventions aimed at rapid hospitalization, concomitantly reducing community transmission and improving case fatality over the course of the outbreak.
We analyzed data on Ebola incidence and case fatality for Montserrado County, Liberia provided by the Liberian Ministry of Health and Social Welfare (MoHSW) from two sources: 1) Case Classification Data (CCD) (S1 Data), and 2) Contact Tracing Data (CTD) (S2 Data). The CCD were collected using the Viral Hemorrhagic Fever form [8] and provide information about 4,373 individuals reported as suspected, probable, or confirmed Ebola cases in Montserrado County between February 28, 2014 and December 1, 2014. The CCD includes case classification status (i.e. suspected, probable, or confirmed), survivorship status, date of symptom onset, date of isolation when applicable, as well as general information regarding age, gender and community of residence. Upon removing duplicate entries and entries with no reported community, our dataset consisted of 3,532 individuals. The CTD provided additional information for a subset of 1,585 individuals who were traced between July 7, 2014 and October 28, 2014. The CTD specified the contacts encountered by each index case following symptom onset. Contacts were monitored for 21 days or until loss to follow-up or onset of symptoms. For contacts who became cases, tracing information was also recorded.
We classified 324 out of the 452 unique communities of residence that were identified from the CCD into three levels of SES (high, middle, and low) according to key indicators [9]. Communities were classified as high SES if residents tended to occupy modern/concrete structures; more than a third of households had access to improved sanitation; and population density was below the average for Monrovia (108,692 population per square kilometer [10]). Communities were classified as middle-to-low SES if most residents occupied tin roof homes; less than a third of households had access to improved sanitation; and population density was higher than the average for Monrovia. Out of the communities meeting these criteria, the subset consisting of slum neighborhoods, including West Point, New Kru Town, Clara Town/Struggle Community, Doe Community, People’s United Community, Logan Town, Jallah, Slipway, Peace Island, and Sinkor’s 12th Street, was classified as low SES. These low SES slums are characterized by high population density, elevated crime rates, ambiguous land ownership, unimproved water sources, and limited or no health care facilities, schools, and sanitation infrastructure [5–7]. Cases reporting areas of residence that were individual compounds or more general areas that consisted of multiple neighborhoods were not included in the classification. Population density was determined using data collected by the Community-Based Initiative, which mapped Monrovia’s neighborhoods as part of an active Ebola surveillance program.
The case classification dataset was analyzed to consider differences in key factors of transmission among cases categorized at low, middle or high SES levels. Descriptive statistics were calculated as means and standard deviations for continuous variables and frequencies and percentages for binary variables. Analyses of variance with ordered levels were used to assess differences in numbers of reported contacts and days between symptom onset and hospitalization. Chi-square tests were used to compare frequencies of care seeking and survivorship among the groups. P-values less than 0.05 were considered statistically significant.
The case classification dataset was analyzed for the 3,532 cases meeting our inclusion criteria and with dates of symptom onset between February 28, 2014 and December 1, 2014. On average, relative to probable and confirmed cases of Ebola in high SES areas, cases in communities of middle and low SES reported seeking care less frequently, although the difference is not statistically significant (Table 2). For those who presented at an Ebola treatment unit (ETU) or other health care facility, there were no significant differences in time from symptom onset to hospitalization. A decreasing trend was observed in time to hospitalization for all three SES groups between July and November 2014 (S6 Fig).
Cases in areas of low and middle SES were associated with a statistically significantly higher number of contacts. In particular, cases reporting residence in a low SES community had an average of nearly three more contacts, as compared to individuals residing in the high SES areas (10.31 versus 7.41 average contacts, p<0.001). Although the number of contacts reported by all SES groups increased over the course of the epidemic, cases from low SES communities were consistently associated with more (S6 Fig). No statistically significant differences in mortality rate were observed across the three SES levels (p = 0.240), but pairwise comparisons suggested that the case fatality for the low SES communities (46.50%) and middle SES communities (41.70%) tended to be higher than that (42.53%) for the high SES communities (p = 0.257 and p = 0.106, respectively).
Source cases reporting a low SES residence were associated with 3.5 times as many secondary cases as those sources reporting a high SES residence (95% CI: 3.1–3.9) (Table 3). In addition, the majority of index cases in the high SES areas transmitted to secondary cases in high SES areas, while only about a third of secondary cases originating from index cases in low SES areas were also in low SES areas. These observations provide evidence that cases tended to be exported from poverty areas and result in transmission chains across Montserrado County.
Our analyses show that despite widespread poverty throughout Liberia, less developed, more resource-constrained areas in Montserrado County tended to have more contacts after Ebola symptom onset and lead to more widespread transmission than higher SES communities. The significantly higher number of reported contacts by sources from low SES areas is consistent with overcrowding and lack of education on routes of disease transmission and prevention. Overall, cases from middle and low SES communities were associated with 1.5 and 3.5 times as many secondary cases, respectively, as sources from high SES communities.
We found that infected individuals who were residents of low SES communities were more likely to export Ebola to other SES communities than infected individuals within higher SES communities. Thus, not only were low SES communities disease hotspots, but they were also catalysts of spread throughout Monrovia. Employment opportunities are typically outside of the lowest SES communities, leading to substantial daily movement back and forth from slums throughout Monrovia.
Our findings suggest that targeting areas of extreme poverty would have the greatest impact in terms of preventing or containing outbreaks of infectious disease. However, the construction and staffing of hospitals or other public health infrastructure in the most disadvantaged areas of Liberia is a daunting even if essential undertaking, particularly in the context of worsened socioeconomic conditions at the country level due to the recent Ebola outbreak [21]. Therefore, optimal resource allocation to prevent widespread infectious disease will require improvements beyond hospital infrastructure [22].
Poverty has been correlated with increased rates of other infectious disease. In malaria endemic settings, for instance, living in traditional or unimproved homes with mud walls, thatched roofs, and earth floors has been associated with over twice the risk of infection than living in modern homes [23]. In Liberia, more specifically, cholera and other diarrheal diseases are highly prevalent in densely populated coastal regions of Monrovia [24, 25]. Poor sanitation conditions in these urban slums, where less than 25% of households have access to improved facilities, along with a lack of trash removal, result in regular contamination of the high water table [25]. Furthermore, overcrowding provides an environment conducive to the rapid spread and challenging containment of infectious disease. Focusing on sustainable development in urban slums and other communities of low and middle SES could significantly reduce the risk of future infectious disease outbreaks.
Despite the lack of infrastructure in Liberia’s urban slums, strong social networks exist. These networks afford opportunities for grassroots efforts to effectively engage community members in combatting infectious disease. The potential of community-driven efforts to contain Ebola spread became apparent in September when culturally sensitive messaging was used with leaders in the West Point slum to facilitate active case finding. The effectiveness of this approach is evidenced by our finding of increasingly prompt hospitalization, reducing transmission within the community [26], as well as by the organized response to and control of the recent outbreak in Nedowein, Liberia. Thus although poverty substantially exacerbates Ebola transmission, the obstacles imposed by poverty are not insurmountable by a targeted approach tailored to the specific needs and challenges faced by impoverished communities.
|
10.1371/journal.pgen.1005839 | Repeat-Associated Fission Yeast-Like Regional Centromeres in the Ascomycetous Budding Yeast Candida tropicalis | The centromere, on which kinetochore proteins assemble, ensures precise chromosome segregation. Centromeres are largely specified by the histone H3 variant CENP-A (also known as Cse4 in yeasts). Structurally, centromere DNA sequences are highly diverse in nature. However, the evolutionary consequence of these structural diversities on de novo CENP-A chromatin formation remains elusive. Here, we report the identification of centromeres, as the binding sites of four evolutionarily conserved kinetochore proteins, in the human pathogenic budding yeast Candida tropicalis. Each of the seven centromeres comprises a 2 to 5 kb non-repetitive mid core flanked by 2 to 5 kb inverted repeats. The repeat-associated centromeres of C. tropicalis all share a high degree of sequence conservation with each other and are strikingly diverged from the unique and mostly non-repetitive centromeres of related Candida species—Candida albicans, Candida dubliniensis, and Candida lusitaniae. Using a plasmid-based assay, we further demonstrate that pericentric inverted repeats and the underlying DNA sequence provide a structural determinant in CENP-A recruitment in C. tropicalis, as opposed to epigenetically regulated CENP-A loading at centromeres in C. albicans. Thus, the centromere structure and its influence on de novo CENP-A recruitment has been significantly rewired in closely related Candida species. Strikingly, the centromere structural properties along with role of pericentric repeats in de novo CENP-A loading in C. tropicalis are more reminiscent to those of the distantly related fission yeast Schizosaccharomyces pombe. Taken together, we demonstrate, for the first time, fission yeast-like repeat-associated centromeres in an ascomycetous budding yeast.
| Centromeres aid in high fidelity chromosome segregation. Paradoxically, centromere DNA sequences are rapidly evolving in fungi, plants, and animals. Centromere DNA sequences in fungi can be unique in each chromosome or share conserved features such as motifs for sequence specific protein binding, pericentric repeats, or transposon-rich elements. Ascomycetous fungi, in particular, show a wide range of diversity in centromere sequence elements. However, no ascomycetous budding yeast species is known to possess repeat-associated centromeres in all of its chromosomes. Here, we identified and mapped all seven centromeres in an ascomycete, a rapidly emerging human pathogenic yeast, Candida tropicalis. The repeat-associated centromeres of highly homogeneous DNA sequences in C. tropicalis are significantly diverged from the mostly non-repetitive unique centromeric DNA sequences of its closely related sequenced species, Candida albicans, Candida dubliniensis and Candida lusitaniae. Structurally, the centromeres of C. tropicalis more closely resemble those of the distantly related fission yeast Schizosaccharomyces pombe. Thus, we discover rapidly diverging repeat-associated centromeres in an ascomycetous budding yeast and provide evidence of emergence of repeat-associated centromeres via two independent evolutionary events in ascomycetous fungi.
| The high fidelity segregation of replicated chromosomes to daughter cells during cell division is essential in maintaining genome integrity. It is achieved by a dynamic and well-coordinated kinetochore-microtubule interaction on a specialized chromosomal element, known as the centromere. Strikingly, the centromere DNA shows rapid diversification in its sequence, length, and the organization of sequence elements across different species [1–3]. The centromere has been categorized into point and regional primarily based on its length. In addition, there are kinetochore protein complexes which are associated specifically to either the point or regional centromere [4]. Point centromeres, which are typically <400 bp long with conserved DNA elements (CDEs) but lacking DNA sequence repeats, appear to have evolved only once and are restricted to the Saccharomyces lineage [4]. However, the centromeres of most other organisms are regional in nature and span from as small as a few tens of kilobases (kb) as in fission yeast Schizosaccharomyces pombe to as large as multiple megabases (Mb) in length as observed in plants and animals. The large regional centromeres of most plants (reviewed in [5, 6]) and animals (reviewed in [7]) are composed of an array of either repetitive sequences or transposable elements. A classic example is the human centromeres that are organized as 171 bp monomeric repeats arranged into a higher ordered alpha satellite sequence (reviewed in [8]). The regional centromeres of two ascomycetous fungi, Neurospora crassa and S. pombe, and a basidiomycetous fungus Cryptococcus neoformans are much shorter (40 to 300 kb in length) and composed of either transposon-rich repetitive sequences as in N. crassa [9, 10] and in C. neoformans [11], or a heterogeneous central core sequence (cnt) flanked by two distinct inverted repeats (imr and otr) that are conserved across the centromeres in S. pombe [12–14]. It is noteworthy that the repeat-associated fungal centromeres lack tandem arrays of repeats as observed in the centromeres of higher metazoans. Interestingly, centromeres of chicken [18], potato [19] and unicellular red alga Cyanidioshyzon merolae [20] represent a distinct class where both repetitive and repeat-less centromeres exist in the same genome. On the other hand, shorter small regional centromeres of 3 to 5 kb non-repetitive, unique sequences have been identified in three Candida species–Candida albicans [15], Candida dubliniensis [16] and Candida lusitaniae [17]. Interestingly, the centromeres in these organisms lack any sequence conservation shared among different chromosomes in the same species. However, CEN1, CEN5, and CENR in C. albicans as well as in C. dubliniensis possess pericentric inverted repeats which are unique to each centromere [16]. The driving force enabling the evolution of centromeres with such remarkable diversity both in the DNA sequence as well as structure, rather than a common optimized centromere configuration, across eukaryotes remains an enigma [1].
The centromere DNA sequence and the organization of the sequence elements are rapidly evolving even in closely related species of three major forms of eukaryotic life—fungi, plants, and animals [1, 18]. In addition, a series of events including–(a) neocentromere formation [19–24] by centromere repositioning at ectopic sites with no obvious DNA sequence homology to the native centromere, (b) selective inactivation of a centromere in a dicentric chromosome [25–28], and (c) the presence of identical sequences elsewhere in the genome that do not serve as centromere/neocentromere sites in various organisms support the conclusion that centromere specification is largely epigenetically regulated (reviewed in [29, 30]).
The centromere specific histone H3 variant CENP-A (also known as Cse4 in yeasts) [31] is considered to be an epigenetic hallmark of active centromeres [32]. The unique structure of CENP-A chromatin provides the foundation to recruit other kinetochore proteins belonging to the Constitutive Centromere Associated Network (CCAN), Ndc80 complex and Dam1/ Ska complex [33], and nucleates kinetochore assembly in most organisms [34]. However, the mechanism(s) of CENP-A loading at a particular locus across species required for centromere specification and its propagation in subsequent generations remains unclear. As shown in S. pombe, CENP-A loading at the centromere is probably regulated via distinct processes leading to the establishment and propagation of a centromere in most organisms [35]. De novo CENP-A recruitment without any pre-existing mark is crucial to establish a centromere, whereas loading of CENP-A molecules during every cell cycle is important for the propagation of already established centromeres [36].
A common feature of the large regional centromeres in ascomycetous fungi is their inherent association with DNA repeats. Detailed studies on the centromeres of S. pombe revealed that centromere associated repeats provide structural determinants in de novo CENP-A recruitment [37]. In contrast, studies in the human pathogenic budding yeast C. albicans, which possesses small regional centromeres [3] reveal that the centromere DNA sequence (CEN7), that lacks pericentric repeats, fails to form functional centromere de novo on a naked plasmid harboring the CEN7 because CENP-A could not be recruited to the plasmid CEN7 [38]. This result implies that centromeres are epigenetically specified in absence of the pericentric repeats in C. albicans [38]. However, it remains to be tested whether centromeres with inverted repeats (such as CEN5) can recruit CENP-A de-novo in C. albicans.
Candida species, the most commonly encountered human fungal pathogens, cause a wide variety of mucosal infections and organ invasion in immunocompromised patients [39]. Although C. albicans has been long known to be the most abundant Candida species isolated from patients, recent global surveillance programs suggest that non-albicans Candida (NAC) species are rapidly emerging as a serious threat due to widespread use of antifungal drugs [40, 41]. In particular, infections caused by Candida tropicalis, a parasexual human pathogenic yeast, has been increased dramatically worldwide. Particularly in sub-tropical regions of Asia-Pacific, the number of patients with C. tropicalis infection is higher than that caused by C. albicans [40, 42]. Earlier, we reported centromere properties of C. albicans [15] and C. dubliniensis [16]. Here, we report the identification of the centromeres as binding sites of four evolutionarily conserved kinetochore proteins in C. tropicalis, which has a 30 Mb sequenced diploid genome arranged into 23 supercontigs [43]. A comparative analysis of centromeres suggests a rapid divergence not only in the centromere DNA sequence but also in the organization of the sequence elements in these closely related Candida species. Interestingly, pericentric repeats are shown to be important for de novo CENP-A recruitment on C. tropicalis centromeres. Based on the striking structural resemblance of centromeres and the necessity of pericentric repeats for de novo centromere formation both in C. tropicalis and S. pombe, we propose an independent evolution of repeat-associated centromeres in budding and fission yeasts.
We identified four putative kinetochore proteins in C. tropicalis- CtCENP-A (Cse4), CtCENP-C (Mif2), CtNuf2 and CtDad1 (Fig 1A). Each of these proteins shares a high degree of sequence conservation to those of the closely related species C. albicans (S1A Fig). Subcellular localization of these proteins in C. tropicalis revealed localization patterns typical of kinetochore proteins in related yeasts [44–47]: a single punctate structure representing clustered kinetochores in unbudded G1 cells that then segregated into two puncta in large-budded cells undergoing mitosis (Fig 1B). In addition, indirect immunofluorescence microscopy with anti-Cse4 antibodies [45], which are specific to CtCENP-A (S1B Fig), and anti-tubulin antibodies revealed CENP-A to be localized near the spindle pole bodies (S1C Fig). On the basis of the sequence similarities and localization patterns at two different stages of the cell cycle, we conclude that these genes encode conserved kinetochore proteins in C. tropicalis.
Kinetochore proteins are important for chromosome segregation in eukaryotes and their depletion results in chromosome segregation defects due to improper microtubule-kinetochore interactions, which may lead to cell cycle arrest due to activation of the spindle assembly checkpoint. For conditional expression of genes, we identified the GAL1 promoter sequence in C. tropicalis (See Methods). To test the function of these putative kinetochore proteins on chromosome segregation in this diploid organism, one copy of each gene was replaced by a marker gene and the remaining copy was placed under the control of the GAL1 promoter. The inability of the conditional mutant strains to grow under non-permissive conditions confirmed that each of these four kinetochore proteins is essential for viability in C. tropicalis (Fig 2A). Moreover, flow cytometry (FACS) analysis revealed an accumulation of large budded cells at the G2/M stage during growth in non-permissive conditions (Fig 2B and S2 Fig). A significant number of the arrested cells had either an unsegregated nuclear mass at the bud neck, or unequally segregated nuclei indicating an arrest due to mitotic checkpoint activation (Fig 2C and S2 Fig). Taken together, these results strongly suggest that each of these proteins is essential for proper chromosome segregation in C. tropicalis.
Having identified authentic kinetochore proteins, we next sought to map the centromeres in the C. tropicalis genome as the binding sites of CENP-A and CENP-C by chromatin immunoprecipitation followed by next generation sequencing (ChIP-seq) [48]. The sequenced C. tropicalis strain MYA-3404 (CSE4/ CSE4) and its derivatives CtKS201 (MIF2/ MIF2-TAP) were used for CENP-A (anti-Cse4 antibodies) and CENP-C (anti-Protein A antibodies) ChIP experiments respectively. Analysis of the ChIP-seq reads against the C. tropicalis genome [43] identified seven CENP-A- and CENP-C-bound overlapping but unique regions as centromeres in C. tropicalis (Fig 3A and S3A and S3B Fig). Primers designed from the seven unique enriched regions identified by ChIP-seq were used to validate the enrichment of CENP-A and CENP-C binding by analyzing ChIP DNA of each of the four kinetochore proteins namely, CENP-A, CENP-C, Nuf2, and Dad1 on seven supercontigs as compared to a non-centromeric locus (CtLEU2) using semi-quantitative PCR assays (Fig 3B). Moreover, each of these seven regions resides within a long ORF-free region (S1 Table), a common centromeric feature observed in most organisms.
To determine chromosomal identity of each of these centromeres, the chromosomes of C. tropicalis (MYA-3404) were first separated on CHEF gels (see methods). The probes were PCR amplified from a unique region adjacent to each centromere. The specific signals on the Southern blot of the CHEF gels revealed that at least five regions reside on different chromosomes (S4 Fig). Due to limited resolution of higher molecular weight chromosomes, the chromosomal identity of two regions (Scnt 3 and Scnt 4) could not be unambiguously verified by the CHEF analysis. These analyses, together with the previously reported 14 telomeric-linked scaffold ends [43], strongly suggest that there are seven pairs of chromosomes in C. tropicalis.
ChIP-seq analyses show a complete overlap in binding of CENP-A and CENP-C to a 2 to 3 kb region on each of the seven centromeres (Fig 4A and S1 Table). To validate the length of the CENP-A/CENP-C binding regions obtained by the ChIP-seq analysis, we scanned the enrichment of each of the four above mentioned kinetochore proteins on the ORF-free region of Scnt 8 by ChIP followed by quantitative PCR (ChIP-qPCR) with primers designed at approximately 1 kb intervals across the 10 kb region of Scnt 8. This analysis revealed that these kinetochore proteins were enriched over a 3 kb (632950–636200) region on Scnt 8 (Fig 4B) and confirms the results obtained from the ChIP-seq experiment. Binding of evolutionarily conserved kinetochore proteins on the same locus proves that the region on each of the seven chromosomes is an important part of a functional centromere in C. tropicalis.
The sequence analysis of centromeric DNA revealed that all seven centromeres in C. tropicalis have common structural elements comprising a non-repetitive mid core flanked by inverted repeats (IRs) (Fig 4A and S2 Table). The average length of the mid core region is 3.5 kb and is flanked by IRs of an average length of each repeat of 3.5 kb. This is a dramatic transition in the centromere organization in comparison with the centromeres of other closely related Candida species, C. albicans [15, 49], C. dubliniensis [16] and C. lusitaniae [17]. Incidentally, CEN1, CEN5 and CENR of C. albicans and C. dubliniensis also contain non-conserved short pericentric repeats. The binding of both CENP-A and CENP-C is restricted to 2 to 3 kb non-repetitive mid core region in all centromeres in C. tropicalis (Fig 4A). A similar length of CENP-A binding (3 to 5 kb) has been observed in C. albicans [15], C. dubliniensis [16], and C. lusitaniae [17] suggesting a striking conservation in the length of CENP-A chromatin that provides the platform for kinetochore formation [50–52]. On the other hand, the AT-content of the CENP-A-bound mid core regions is found to be 64% in C. tropicalis, which is marginally less than the overall AT-content of the genome (67%). A similar AT-content of CENP-A bound centromere DNA (65%) was observed in C. albicans [15]. Thus, in spite of the observed rapid change in the centromere DNA sequence and its organization, these closely related species employ a similar length and composition (in terms of the AT-content) of the centromere DNA for the recruitment of kinetochore proteins.
In silico analysis of centromere sequences in C. tropicalis revealed that the inverted repeats (IRs) and mid core regions share a high degree of sequence homology across the centromeres (Fig 4C and S5A–S5C Fig). Between different chromosomes, the mid core regions share 80% homology and 63% identity while the IRs show an average of 92% homology and 82% identity. However, the conservation is much higher between the left and right repeats (LR and RR) of the same centromere, with an average of 97% homology and 93% identity (Fig 4C and S5D Fig). In addition, we also observed that tandem direct repeats are present within each inverted repeat (S5E Fig and S3 Table). These groups of tandem repeats were prominent across all arms (except Scnt 9, which lacks the final group). However, the copy number varied significantly among arms (S3 Table). The observed high level of sequence conservation of the mid core and inverted repeats among the different centromeres in C. tropicalis suggests that these regions might have undergone homogenization via intra- and inter-chromosomal recombinatorial events. Such process may be facilitated by the close association of centromeres in the clustered kinetochores of C. tropicalis (Fig 1B).
Rapid divergence in the centromere sequence and structure is often associated with karyotypic changes [53, 54], a hallmark of speciation [55–58]. Previously, we demonstrated rapidly changing DNA sequence at the centromeres of orthologous chromosomes in C. albicans and C. dubliniensis without any significant changes in synteny across chromosomes [16]. Here we performed a synteny dot plot analysis between C. albicans and C. tropicalis genomes. This analysis revealed massive chromosomal rearrangements involving several syntenic breaks happened between these two species. Unusually, it appears that intra-chromosomal transpositions and inversions are far more common than inter-chromosome recombination (Fig 5). Strikingly, inter-chromosome recombination, though uncommon, tends to occur more often near the centromeres (Fig 5 and S6A Fig). For example, in CtScnt3, the large number of genes to the left of the centromere, and a few on the right, map to CaChr3. But immediately after the centromere, there are some segments on CtScnt3 that are in synteny with CaChrR and CaChr5, and then the remainders of CtScnt3 are largely from CaChr6 (Fig 5). Similar patterns of rearrangement can be seen in most other supercontigs also except CtScnt 7. This pattern of rearrangement was plausible probably due to recombination at highly identical sequences of inverted repeats. A similar phenomenon of centromeric repeat-mediated rearrangement and subsequent gain of a chromosome has been observed in two laboratory strains of S. pombe [59]. Incidentally, there is a change in chromosome number from eight pairs in C. albicans and C. dubliniensis to seven pairs in C. tropicalis indicating a possible structural rearrangement involving centromere which might have given rise to a centromere gain or loss (S6B Fig). However, CtScnt7 comes almost entirely from CaChr7, but has been heavily rearranged (Fig 5). The unusual preponderance of intra-chromosomal transposition and reversal compared to the smaller numbers of inter-chromosomal translocation may merit further study.
In addition, a putative retrotransposon present at the centromere in C. tropicalis is found to be conserved at CEN7 in C. dubliniensis [16] (S5B Fig). A similar retrotransposon was also found to be present within 50 kb region of CEN7 in C. albicans (S6A Fig). This putative transposon is a member of the Ty3/Gypsy family but does not present at putative centromeres in any other Candida species. These results, together with the conservation in the CENP-A chromatin length, indicate that the centromere position of these related species was shared by a common ancestor and may have undergone chromosomal rearrangement involving the centromeres of more than one chromosome during evolution.
The structural features of C. tropicalis centromeres strikingly resemble those of the distantly related fission yeast S. pombe. To understand the function of the underlying centromere DNA sequences in C. tropicalis, we engineered plasmids carrying either the full length centromere (pCEN8) or a part of it (pmid8) on a replicative plasmid pARS2 (Fig 6A). The replicative plasmid pARS2 harbors CaARS2 [60], which functions as an autonomously replicating sequence (ARS) on a circular plasmid in C. tropicalis (S7 Fig). While the pmid8 plasmid conferred 10 to 13-fold increased mitotic stability as compared to pARS2, inclusion of the full length centromere sequence harboring inverted repeats in the pARS2 plasmid (pCEN8) resulted in a 37 to 42-fold higher mitotic stability after 10 generations of nonselective growth (Fig 6B). A size-dependent stabilization of circular replicative plasmids has been reported previously in S. cerevisiae [61]. To rule out this possibility, we cloned a 10 kb of hererologous DNA sequence from bacteriophage λ (pARS2-λ) and measured the mitotic stability of the same. This plasmid, which is of similar length (15 kb) to that of pCEN8, did not show an increase in the mitotic stability as observed in pCEN8 (Fig 6B). In addition, pCEN8 is 3 to 4-fold more stable mitotically than pmid8 carrying only the mid core sequence (Fig 6B). These results suggest that the inverted repeats flanking the mid core can significantly improve the mitotic stability of an otherwise unstable replicative plasmid in C. tropicalis. Because CENP-A is known to bind to only functional centromeres, functionality of a centromere sequence cloned into the replicative plasmid was further assayed by the extent of CENP-A enrichment on these exogenously introduced plasmid DNA constructs. It should be noted that a unique SalI restriction site was introduced at the edge of the mid region of the plasmid-borne to differentiate it from the endogenous chromosomal ones (see S8 Fig). CENP-A ChIP-qPCR analysis with plasmid specific primer-pair revealed that CENP-A is enriched at the mid core region on only the full length centromere DNA in pCEN8 (Fig 6C) suggesting that the inverted repeats (LR and RR) flanking the mid core are important for de novo CENP-A deposition. Thus, we conclude that the CENP-A recruitment process has been significantly rewired in closely related Candida species. Centromere function was shown to be dependent on the presence of inverted pericentric repeats in S. pombe as well [62]. On the other hand, Candida species and S. pombe shared a common ancestor more than 330 mya [63]. Thus, we demonstrate an extraordinary example of evolution of inverted repeat containing ‘fission yeast-like’ centromeres that appeared independently at least once in the Candida clade.
In order to find out the role of pericentric inverted repeats and the DNA sequence associated with them for centromere function in C. tropicalis, we have constructed two different engineered plasmids namely pCEN801 and pCEN802 (Fig 6A). The pCEN801 plasmid harbors the left repeat of CEN8 (CtLR8) cloned in a direct orientation with respect to the right repeat of the same centromere (CtRR8). Thus, the only difference between pCEN8 (inverted orientation) and pCEN801 (direct orientation) is the orientation of pericentric repeats with respect to each other. However, pCEN801 is found to be significantly less stable mitotically as compared to the pCEN8 (Fig 6B). Moreover, CENP-A ChIP-qPCR analysis revealed that CENP-A does not bind to this engineered pCEN801 plasmid (Fig 6C). These suggest that the inverted orientation of pericentric repeats is an important structural feature for centromere function in C. tropicalis.
To understand the function of the underlying DNA sequence of the inverted repeats, we cloned the inverted repeats of CEN5 (CaIR5) from C. albicans into the pmid8 plasmid (pCEN802). However, the mitotic stability of pCEN802 is found to be 4 to 6-fold lower in C. tropicalis as compared to pCEN8, which harbors pericentric inverted repeats (IR8) of C. tropicalis (Fig 6B). This observation has been further verified by CENP-A ChIP-qPCR analysis (Fig 6C) and confirms that the sequence of the inverted repeats per se is also crucial for centromere function in this species. Thus, we conclude that the DNA sequence of the repeats as well the arrangement of the repeats in an inverted fashion is both important for centromere function in C. tropicalis.
To elucidate the route of centromere diversification, we reconstructed a phylogenetic tree of 13 species representing all major lineages of Ascomycota (Fig 7A). It demarcates three distinct monophyletic subphyla within the Ascomycota—Taphrinomycotina, Pezizomycotina and Saccharomycotina (Fig 7A). Moreover, this study also supports that Taphrinomycotina and Pezizomycotina are the early radiating branches in Ascomycetes. Thus, it is evident from both the phylogenetic relationship and the centromere structures of S. pombe and N. crassa that the invasion of transposons or symmetric repetitive elements shaped centromere structure in Taphrinomycotina and Pezizomycotina during an early era of ascomycete evolution (Fig 7A). In contrast, a dramatic reduction in centromere length with a concurrent absence or loss of centromeric transposons or repeats, is evidenced from the centromeres of Candida and Saccharomyces species, and, therefore, evolved in Saccharomycotina (Fig 7A). The identification of the centromere in C. tropicalis in this study is the first report that shows the evolution of repeat-associated centromeres in the clade of Saccharomycotina (Fig 7A).
In this study, we identified and analyzed the centromeres in C. tropicalis. We demonstrate that each centromere consists of a central non-repetitive mid core region, which is bound by evolutionarily conserved proteins from various layers of the kinetochore, and is flanked by inverted repeats. This is the first known saccharomycetous yeast in which all seven native centromeres are repeat-associated. Moreover, the inverted repeats of the same chromosome as well as across different chromosomes of C. tropicalis are highly similar in sequence. Taking together these centromere properties of C. tropicalis and those of other saccharomycetous yeasts, it is now evident that centromeres of all types—point centromeres with conserved motifs that are < 400 bp in length (as in Saccharomyces cerevisiae), shorter non-repetitive regional centromeres with unique CENP-A-rich regions of 3 to 5 kb long (as in C. albicans, C. dubliniensis and C. lusitaniae) as well as repeat-associated regional centromeres of 10 to 11 kb (as in C. tropicalis)–evolved in Saccharomycotina (Fig 7B). Although centromere structures are known in only a limited number of organisms, the discovery of all major types of centromeres in the saccharomycetes makes it a unique sub-phylum for tracing the path of evolution of monocentric chromosomes.
The CENP-A-bound DNA sequence is the most preferred site of kinetochore assembly in an entire chromosome. In spite of sharing conserved motifs among centromeres, the CENP-A- bound DNA sequences are often variable, even in the genetically defined point centromeres of S. cerevisiae. Intriguingly, a comparative analysis between S. cerevisiae and its closest relative Saccharomyces paradoxus, identified that the CENP-A-bound CDE-II elements are the fastest evolving region of the genomes [64]. Similarly, in S. pombe flanking repeat sequences are conserved among the different chromosomes but the CENP-A-rich central core sequences are heterogeneous [65]. The most extreme cases of rapid divergence have been observed in the centromeres of C. albicans [15], C. dubliniensis [16], and C. lusitaniae [17], where CENP-A-rich centromere DNA sequences are all unique and different in each species. In contrast, CENP-A is found to be enriched on highly homogenized arrays in most plants, mouse, and humans (reviewed in [7]). Thus, homogenization of CENP-A-bound mid core regions in C. tropicalis, as observed in this study, provides a unique feature of yeast centromeres that is more reminiscent of metazoan centromeres. It has been proposed that transposable elements are a major source of centromeric satellite repeats, which gradually homogenized over time by an unknown mechanism in a metazoan system (reviewed in [66]). We also observed a similar association of a retrotransposon in one centromere in C. tropicalis. More recently, it has been shown that the CENP-A-bound central core has a sequence feature enabling de novo recruitment of CENP-A molecules in S. pombe [67]. Thus, it will be intriguing to investigate a feature of CENP-A-enriched mid core regions in C. tropicalis that may facilitate CENP-A recruitment.
Centromeres are known to be species-specific as centromeres of one organism do not function even in a related species [68]. Inter-species crosses, mostly in plants, suggest that functional incompatibility of centromeres is a frequent cause of uniparental genome elimination [69–71]. Recently, it has been reported that perturbation of the length of the CENP-A binding domain to adopt a uniform size is a prerequisite for a successful inter-species hybridization between maize and oat [69]. Thus, the length of the CENP-A-rich region at the centromere may be a key factor for centromere incompatibility in close relatives. In addition, the length of the CENP-A binding domain is found to be uniform in an organism regardless of the chromosome size or the nature of the centromere. Indeed, we observed that the length of the CENP-A binding region (3 to 5 kb) is surprisingly conserved in related Candida species, in spite of the dramatic transition in the centromere organization. A uniform length of the CENP-A-bound regions in these related species may thus suggest a possible role in maintaining uniform kinetochore-microtubule interactions. This is further supported by the fact that the Dam1 complex is essential in C. tropicalis. Essentiality of the Dam1 complex has been previously correlated to a one microtubule-one kinetochore type of interaction as observed in S. cerevisiae and C. albicans [44, 46]. Recently, it was proposed that DNA sequence repeats might have evolved to provide a ‘safety buffer’ against drifts in kinetochore position [72]. Interestingly, we found that the binding of kinetochore proteins is restricted to a non-repetitive mid core region in all cases in C. tropicalis and does not spread to the surrounding inverted repeats. CENP-A chromatin is generally repressive (reviewed in [73]) and thus the safety buffer provided by the pericentric inverted repeats perhaps act as a barrier to prevent the drift of kinetochore position and maintain the size of CENP-A binding domain in this organism.
A series of growing lines of evidence suggest that fungal centromeres are rapidly evolving genomic loci (reviewed in [2]). It has been proposed that rapid evolution of centromere DNA may contribute to its functional incompatibility and perhaps aids in speciation [1, 18]. Speciation is, however, a poorly defined and less understood process in asexual organisms [74]. Some Candida species with known centromere structures (C. albicans, C. dubliniensis and C. tropicalis) are primarily parasexual and capable of mating but lack a recognized meiotic program. In spite of this, we observed in this study a high degree of divergence in the centromere DNA sequences as well as in the organization of centromere elements in these related Candida species. Why does the centromere structure diverge so rapidly in these related organisms? It has been proposed that the loss of centromere function followed by the birth of a centromere in a new position can be viewed as a life cycle of a centromere that operates during evolution (reviewed in [75]). For such an event, massive chromosomal rearrangements including the loss of an existing centromere would have to occur. Coincidentally, a comparative analysis among the relatives of both yeasts [76] and mammals [77] identified frequent breakpoints adjacent to centromeres. These results suggest that centromeres are among the most fragile sites in a genome. We also observed a gross chromosomal rearrangement between C. albicans and C. tropicalis specifically at the centromeres. It is also clear that the centromere loss or gain happened in these two organisms during their divergence from a common ancestor. Being both commensal and opportunistic pathogens, Candida species show considerable genome plasticity possibly as a means to survive in a hostile host environment. Genome rearrangements including karyotype changes, aneuploidy, and loss of heterozygosity have been frequently observed in clinical isolates of Candida species (reviewed in [78]). Thus, it is likely that the evolutionary life cycle of a centromere may have contributed to their rapid divergence in these related pathogenic yeast species.
Evolution is typically thought to proceed to generate diversity [79]. However, independent evolutionary origins of similar biological structures or functions in distantly related taxa challenge this common paradigm [80]. In this study, we observed that structural features of the C. tropicalis centromeres resemble a shorter version (10 to 11 kb) of the distantly related S. pombe centromeres (40 to 110 kb) [12, 65]. However, the pericentric inverted repeats observed in C. tropicalis have no sequence identity to either the pericentric repeats of S. pombe or the centromere associated inverted repeats of C. albicans or C. dubliniensis. A notable difference between the centromeres of S. pombe and C. tropicalis is the absence of outer repeats (otr in S. pombe) in C. tropicalis. The otr is the site of small RNA (siRNA) generation and subsequently otr recruits other heterochromatin proteins (such as Swi6 and Clr4 in S. pombe) to make the centromeric region heterochromatic in S. pombe (reviewed in [81]). Heterochromatin proteins and siRNAs play a vital role in centromere identity in this organism. Unlike S. pombe, C. tropicalis genome neither possesses the full RNAi machinery nor several key players required for heterochromatin formation such as an ortholog of Clr4 (H3K9 methyltransferase) [82]. Thus involvement of repeat elements in establishing RNAi-dependent H3K9me heterochromatin formation, as observed in S. pombe, is unlikely in C. tropicalis. In conclusion, we demonstrate for the first time the evolution of repeat-associated centromeres in an ascomycetous budding yeast (Fig 7B). The most reasonable explanation for the appearance of the repeat-associated centromere structure is the contribution of repeats to de novo CENP-A deposition. CENP-A is a universal marker of functional centromeres and does not localize at inactivated centromeres. Studies on artificial CENP-A recruitment, either by direct tethering of CENP-A or its chaperone HJURP (also known as Scm3 in yeasts) to an ectopic locus [83, 84], suggest that de novo CENP-A deposition is in general one of the most significant rate limiting steps to the acquisition of centromere function. The process of CENP-A recruitment is known to be regulated by both genetic and epigenetic means (reviewed in [2, 29]). However, neither the DNA elements nor epigenetic factors are conserved across the kingdom implying an astounding flexibility in centromere specification. In this study, we demonstrate that a dramatic transition in centromere organization has rewired the genetic and epigenetic regulation of CENP-A deposition in related species. Thus, the ways in which the genetic and the epigenetic factors are co-evolving to orchestrate de novo CENP-A recruitment on a DNA sequence to establish a functional centromere may determine the shape of the centromere structure in an organism.
C. tropicalis strains were grown either in YPDU (1% yeast extract/ 2% peptone/ 2% glucose/ 0.010% uracil), or in complete minimal (CM) media unless stated otherwise. C. tropicalis cells were transformed by the standard lithium acetate method as stated previously [45]. It is important to note that C. tropicalis requires uracil and not uridine in the medium to supplement the Ura auxotrophy.
The centromeric histone H3 CENP-A homolog in C. tropicalis [85], was identified in a BLAST analysis using C. albicans CENP-A (CaCse4) as the query sequence against the Candida tropicalis genome [43]. The BLAST analysis revealed that the proteins with high scores (score >213) were the putative CENP-A homologue, CtCse4 (CTRG_02639.3), and histone H3 proteins (CTRG_04732.3, CTRG_00676.3 and CTRG_05645.3). The CtCse4 (Scnt 3: 1334129–1334845) is a 238-aa-long protein that shows 90% homology with the C-terminal histone fold domain of CaCse4 (S1A Fig). Similarly, CENP-C (Mif2), Nuf2, and Dad1 homologs of C. tropicalis were identified in a BLAST analysis. The CtMif2 (CTRG_05763.3) is a 523-aa-long protein (Scnt 9: 474053–475624+) with a conserved CENP-C box, which is identical in sequence between the CaMif2 and CtMif2 (S1A Fig). CtNUF2 (CTRG_05381.3) and CtDAD1 (CTRG_03625.3) encode 492-aa- and 99-aa-long proteins respectively. Both of these proteins show a high degree of sequence conservation in comparison to those of C. albicans (S1A Fig).
The sequence upstream of the GAL1 gene in S. cerevisiae, harboring the upstream activation sequence (UAS), is used as the GAL1 promoter to regulate the expression of desired genes [86, 87]. However, no such regulatable promoter has been identified previously in C. tropicalis to control the expression level and study the essentiality of proteins. The C. tropicalis homolog of GAL1 was identified as the ORF (CTRG_04617) by BLAST using S. cerevisiae GAL1 as the query sequence. Further, on analyzing the genomic location of this gene, we found that the synteny of GAL1 and GAL10 genes was maintained as observed in S. cerevisiae.
The primer sequences and all C. tropicalis strains used in this study are listed in S4 and S5 Tables respectively. The detailed information about the strain construction is available in the S1 Text.
Cells of C. tropicalis strains expressing GFP tagged kinetochore proteins were grown overnight, harvested, and washed twice with sterile distilled water. Cells were then resuspended into sterile distilled water to obtain the desired density before taking the images with a Delta Vision Microscopy Imaging system. Indirect immunofluorescence was done as described before [45]. Asynchronously grown C. tropicalis cells were fixed with a 1/10th volume of formaldehyde (37%) for 1 h at room temperature. Antibodies used were diluted as follows: 1:500 for rabbit anti-Cse4 antibodies [45] and 1:30 for rat anti-tubulin antibodies (Abcam, Cat No. ab6161). The dilutions for secondary antibodies used were Alexa flour 568 goat anti-rabbit IgG (Invitrogen, Cat No. A11011) 1:500 and Alexa fluor 488 goat anti-rat IgG (Invitrogen, Cat No. A11006) 1:500. DAPI (4, 6-Diamino-2-phenylindole) (D9542 Sigma) was used to stain the nuclei of the cells. Cells were examined under 100 (multi) magnifications using a confocal laser scanning microscope (LSM 510 META, Carl Zeiss). The digital images were processed with Adobe Photoshop.
C. tropicalis cells were harvested at two different time points and processed as described before [45]. Prior to injection of the sample into the flow cytometer, the cell suspension was sonicated briefly (30% amplitude, 7s pulse). The sonicated samples were diluted to a desired cell density in 1X PBS and injected into the flow cytometer (BD FACSCalibur) for analysis. The output was analyzed using the BD CellQuestPro software.
The conditional mutant strains of C. tropicalis grown in both permissive and non-permissive media were harvested, washed, and resuspended in 300 μl of sterile distilled water. These cells were fixed by adding 700 μl absolute ethanol and incubated at room temperature for 1 h. After fixing, the cells were washed with 1ml of sterile distilled water twice and resuspended in sterile distilled water to obtain desired cell density prior to imaging. To 5 μl cell suspension, 3 μl DAPI (100 ng/ml) was added in the well, mixed gently by pipetting, and the cover slip was then placed. After 5 min of incubation, the cells were imaged using a fluorescence microscope (Olympus BX51) under 100 (multi) magnifications.
C. tropicalis strains were grown overnight in YPDU and cells were harvested. The harvested cells were washed with lysis buffer (0.2 M Tris, 1 mM EDTA, 0.39 M ammonium sulphate, 4.9 mM magnesium sulphate, 20% glycerol, 0.95% acetic acid, pH 7.8) and resuspended in 0.5 ml of the same buffer. The cells were disrupted using acid-washed glass beads (Sigma, Cat. No. G8772) by vortexing 5 min (1 min vortexing followed by 1 min cooling on ice) at 4°C. C. tropicalis cell lysates were electrophoresed on a 12% SDS-PAGE gel and blotted onto a nitrocellulose membrane in a semi-dry apparatus (Bio-Rad). The blotted membranes were blocked with 5% skim milk containing 1X PBS (pH 7.4) for 1 h at room temperature and were then incubated with following dilutions of primary antibodies: anti-Cse4 antibodies [45] 1:500; anti-H3 antibodies [Abcam, Cat No. ab1791] 1:2500; for 1 h at room temperature. Next, the membranes were washed three times with PBST (0.1% Tween-20 in 1X PBS) solution. Anti-rabbit HRP conjugated antibodies [Bangalore Genei, Cat No. 105499] in 1:1000 dilutions were added and incubated for 1 h at room temperature followed by three to four washes with the PBST solution. Signals were detected using the chemiluminescence method (SuperSignal West Pico Chemiluminescent substrate, Thermo scientific, Cat No. 34080).
The ChIP assays were done as described previously [15]. Briefly, each strain was grown until exponential phase (~2×107 cells/ml) and cells were cross-linked with formaldehyde (final concentration 1%). Chromatin was isolated and sonicated to yield an average fragment size of 300–500 bp. Then the DNA was immunoprecipitated with anti-Cse4 antibodies [45] (final concentration is 6 μg/ml) or anti-protein A antibodies (final concentration is 24 μg/ml) or anti-V5 antibodies (Life Technologies, Cat No. R960-25) (final concentration is 0.94μl/ml) and purified. The duration of cross-linking varies—15 min for CENP-A, 20 min for CENP-C, 1 h 45 min for Nuf2 and 3 h 15 min for Dad1. The total, immunoprecipitated (IP) DNA, and beads only material were used to determine the binding of kinetochore proteins in all seven putative centromeres by semi-quantitative PCR. PCR conditions for primers (as listed in S5 Table) were used as follows: 94°C for 2 min, Tm for 30 s (Tm varies with the primers), 72°C for 1 min, for 1 cycle; 94°C for 30 s, Tm for 30 s, 72°C for 1 min for 24 cycles in case of CENP-A and CENP-C; and 27 cycles for Nuf2 and Dad1; 72°C for 10 min.
ChIP-seq analyses were conducted as described previously [23]. The detailed procedure of ChIP-seq and analysis are provided in the S1 Text.
C. tropicalis strain MYA-3404 was grown until exponential phase (~2×107 cells/ml). Cells were washed with 50 mM EDTA and counted with a hemocytometer. Approximately 6×108 cells were used for the preparation of 1 ml genomic DNA plugs. The plugs were made according to the instruction manual protocol (BioRad, Cat No. 170–3593) with cleancut agarose (0.6%) and the lyticase enzyme provided by the kit. A 0.6% pulsed field certified agarose gel was prepared using 0.5X TBE buffer (0.1 M Tris, 0.09 M boric acid, 0.01 M EDTA, pH 8) and the PFGE was performed on a CHEF-DR II (Bio-Rad) for 72 h (24 h at 4.5 V/cm/106° with an initial and final switch times 200 s; 48 h at 3 V/cm/106° with an initial and final switch time 700 s). The gel was stained with ethidium bromide (EtBr) and analyzed by using the Quantity One software (Bio-Rad).
To determine the extent of binding of kinetochore proteins on the centromere of Scnt 8, real time PCR (qPCR) was performed. The template used was as follows: 1 μl of 1:100 dilutions for input and 1 μl of 1:5 dilutions for IP. The conditions used in qPCR were as follows: 94°C for 2 min; 94°C for 30 s, Tm for 30 s (Tm varies with the primers), 72°C for 45 s for 30 cycles. The results were plotted on a graph according to the percentage input method using the formula: 100*2^ (adjusted Ct input−adjusted Ct of IP). Here, the adjusted value is the dilution factor (log2 of dilution factor) subtracted from the Ct value of diluted input or IP [88]. Similar conditions were used to determine the enrichment of CENP-A proteins on the centromeric plasmids.
To determine conservation rates for inverted repeats (IRs) within and across centromeres, and mid regions across different centromeres, we used Sigma version 2 [89], a program that aims to minimize spurious alignments by using a stringent p-value for all local alignments, and uses a background model with correlations and an evolutionary model to link sequences. The background model and substitution matrix were drawn from S. cerevisiae and close relatives, and are not expected to vary significantly across Saccharomycetes. The branch lengths were determined dynamically. The conservation rates in Fig 4C were determined from these alignments using custom python scripts. The visual representation of the alignments as shown in S5A, S5C and S5D Fig was created with an in-house program. The dotplot in S5E Fig was created with dotmatcher, from EMBOSS 6.3.1 [90]. Additionally, the inverted repeats and mid regions were scanned for tandem repeats using the Tandem Repeats Finder version 4.04 [91]. The parameters used were “filename 2 5 5 80 10 2 2000” (maximum period size 2000). The results are summarized in S3 Table. For the synteny analysis as in Fig 5, orthology information was obtained from the Fungal Orthogroups Repository (http://www.broadinstitute.org/regev/orthogroups) [92]. Genes in each species within 100 kb of each centromere were examined, and orthologous genes were plotted using an in-house program.
Approximately 1 μg of DNA of both pARS2 and the control parental (pUC19-CaURA3) plasmids were used to transform CtKS04 strain using both the lithium acetate and the spheroplast transformation methods as stated before [93]. After transformation, the cells were plated on the complete media lacking uracil (CM-Ura) and incubated at 30°C for 3 to 5 days before taking photographs. The ARS activity of pARS2 was determined as the transformation efficiency (i.e., the number of transformants/ μg of DNA). Each transformation was performed at least 3 times.
The mitotic stability assay was performed to determine the loss rate of pARS2, pmid8, pCEN8, pARS2-λ, pCEN801 and pCEN802 in C. tropicalis. Briefly, the C. tropicalis strain, CtKS102 transformed with above mentioned plasmids were streaked on CM-Ura plates for single colonies. Single colonies thus obtained were subsequently inoculated in a nonselective media (YPDU) and incubated at 30°C for overnight for at least 10 generations. Next day, equal numbers of cells were simultaneously plated on YPDU and CM-Ura and incubated at 30°C for 2 days. Colonies grown on both plates were counted and the mitotic stability was calculated in percentage as follows: Mitotic stability = (S/NS), where S and NS denote the number of colonies grown on selective and nonselective media respectively.
A phylogenetic tree with estimated geological time was created via a multiple alignment of 573 gene orthologue sets in 13 sequenced species of Ascomycetous fungi (as shown in Fig 7A)–namely, C. tropicalis, S. cerevisiae, C. glabrata, K. lactis, A. gossypii, N. casetelli, C. dubliniensis, C. albicans, C. lusitaniae, D. hansenii, C. guilliermondii, Y. lipolytica, N. crassa, A. nidulans, S. japonicus, S. octosporus, S. pombe. The orthologous genes were identified using the Fungal Orthogroups Repository (http://www.broadinstitute.org/regev/orthogroups/) [92], except in the case of C. dubliniensis for which orthologues to C. albicans were used as annotated in the gene sequences from the Candida Genome Database (http://www.candidagenome.org/). Only genes for which there were unique reciprocal orthologues between S. cerevisiae and each of the 13 other species, and which lacked introns (or from which we could easily remove introns) were considered. To remove bias from outliers, the orthologous genes in all species were further sub-selected for genes that evolve uniformly. For this, the average rates of synonymous (ds) and non-synonymous (dn) substitution were calculated separately from codon-level alignments. Only genes whose ds and dn both fell within 1.5 standard deviations of the mean for the full set were considered. This yielded a list of 573 genes. These coding DNA sequences were aligned at the codon level with FSA [94] (command line option “—translated”), concatenated with gaps removed, and a tree was generated with codonphyml [95] (command line “-d codon -q”). Since the quantity of sequence was very large (nearly 10 Mbp, or over 0.5 Mbp per species) bootstrapping was not done.
|
10.1371/journal.pcbi.1004342 | Posterior Probability Matching and Human Perceptual Decision Making | Probability matching is a classic theory of decision making that was first developed in models of cognition. Posterior probability matching, a variant in which observers match their response probabilities to the posterior probability of each response being correct, is being used increasingly often in models of perception. However, little is known about whether posterior probability matching is consistent with the vast literature on vision and hearing that has developed within signal detection theory. Here we test posterior probability matching models using two tools from detection theory. First, we examine the models’ performance in a two-pass experiment, where each block of trials is presented twice, and we measure the proportion of times that the model gives the same response twice to repeated stimuli. We show that at low performance levels, posterior probability matching models give highly inconsistent responses across repeated presentations of identical trials. We find that practised human observers are more consistent across repeated trials than these models predict, and we find some evidence that less practised observers more consistent as well. Second, we compare the performance of posterior probability matching models on a discrimination task to the performance of a theoretical ideal observer that achieves the best possible performance. We find that posterior probability matching is very inefficient at low-to-moderate performance levels, and that human observers can be more efficient than is ever possible according to posterior probability matching models. These findings support classic signal detection models, and rule out a broad class of posterior probability matching models for expert performance on perceptual tasks that range in complexity from contrast discrimination to symmetry detection. However, our findings leave open the possibility that inexperienced observers may show posterior probability matching behaviour, and our methods provide new tools for testing for such a strategy.
| Decision making is partly random: a person can make different decisions at different times based on the same information. The theory of probability matching says that one reason for this randomness is that people usually choose the response that they think is most likely to be correct, but they sometimes intentionally choose the response that they think is less likely to be correct. Probability matching is a theory that was developed to describe how people try to predict the outcome of a partly random event, e.g., whether a patient has some medical condition, given the result of a medical test that does not provide perfectly accurate information. Recently, modified probability matching theories have been used to understand perceptual decision making, e.g., judging whether a sound and a visual flash were produced by the same event or by different events. We show that probability matching predicts that peoples’ perceptual decisions on difficult tasks are highly random and make poor use of the available information. We show experimentally that expert perceptual decisions are less random and more efficient than probability matching predicts. These findings help us understand how people perform a wide range of important real-world perceptual tasks, such as evaluating medical images and detecting targets in airport screening scans.
| Human decision making is partly random, in the sense that a person can make different decisions on different occasions based on the same information. Probability matching is a theory of decision making that aims to account for this randomness. Suppose a person believes that response A has a 70% probability of being correct, and response B has a 30% probability of being correct. A person who exhibits probability matching chooses response A with 70% probability and response B with 30% probability. This is a surprising decision strategy, because it means that the person sometimes chooses the response that is less likely to be correct according to the available evidence. Nevertheless, many studies support probability matching as a model of decision making in cognitive tasks such as probability learning [1–2].
Probability matching originated in models of cognition, and variants of probability matching have also been used in models of perception. For example, Mamassian and Landy [3] had observers judge the three-dimensional shapes depicted in line drawings. In their model of this task, subjects used the intrinsically ambiguous shape information from line drawings along with assumptions about the statistical distribution of shapes in the real world to estimate the probabilities that a line drawing depicted an elliptical shape or a saddle shape. Subjects then used probability matching to choose their response, either “elliptical” or “saddle”. Acerbi et al. [4] call such a strategy “posterior probability matching,” because the subject matches their response probabilities to the posterior probability of each response being correct, instead of the prior (i.e., baseline) probability as in classic probability matching models. Posterior probability matching and related approaches have become increasingly common in models of perceptual decision making [3–8].
Posterior probability matching has some appealing features for models of perception. It offers an explanation of why perceptual decisions have a random component at all. Furthermore, it has no free parameters, which one might hope could explain why subjects’ decisions show a limited range of randomness across many perceptual tasks [9, 10].
However, little is known about whether posterior probability matching is consistent with well-supported models of perception that have been developed within signal detection theory, which take a very different approach to modelling randomness in decision making. Signal detection models typically assume that observers’ decisions depend both on information received from stimuli, and on random fluctuations in perceptual mechanisms, i.e., internal noise. If the same stimuli are repeated on different trials, the subject may make different responses, because the internal noise contributions may be different. A large literature supports signal detection models of perceptual and cognitive decision making [11]. Furthermore, posterior probability matching models represent an unusual mix of optimal and suboptimal behaviours: these models state that observers calculate the posterior probabilities that would enable them to make optimal responses based on the available stimulus information (e.g., maximizing payoff according to some utility function), but then instead behave suboptimally by making stochastic responses that match their response probabilities to the posterior probabilities.
Here we compare posterior probability matching and signal detection models of perceptual decision making. We begin by defining two very general classes of these models, and we examine the models’ behaviour using two psychophysical methods. First, we use the two-pass response consistency method, which quantifies the randomness in an observer’s decisions by examining how consistent decisions are across repeated presentations of identical trials [9, 12]. In a two-pass response consistency task the observer views one of two possible signals shown in noise, and attempts to identify the signal. This continues for some number of trials. The observer then sees the identical sequence of trials a second time (i.e., the same signals in the same samples of noise), without knowing that they are being repeated, and again attempts to identify the signal on each trial. The experimenter measures the proportion of correct responses, PC, and also the proportion of consistent responses, PA, i.e., the proportion of repeated trials on which the observer gives the same response twice. (Here the subscript “A” stands for “agreement”.) Burgess and Colborne [12] show how to use PC and PA to calculate the relative amounts of internal and external noise, σI/σE, in the decision variable of an observer.
As a second test of posterior probability matching models, we use ideal observer analysis, which measures an observer’s efficiency at a task by comparing the observer’s performance to the best performance that is theoretically possible on the task [13, 14].
We examine posterior probability matching and signal detection models' predictions as to how response consistency and efficiency should vary as a function of proportion correct. To test these models we compare their predictions to human observers’ behaviour on a two-pass response consistency task, and to results of previous experiments on response consistency and efficiency. In Experiment 1 we compare human observers' response consistency on repeated trials in a two alternative forced choice (2AFC) discrimination task to the predictions of posterior probability matching and signal detection models. In Experiment 2 we test a larger number of inexperienced observers on a shorter version of the same task in order to sample a wider range of observers and examine the effect of practice.
We examined the performance of posterior probability matching and signal detection models in a task where there are two possible signals (A and B) and two possible responses. This includes two-alternative identification tasks, and it also includes 2AFC tasks because we can take “signal A” to mean the two stimulus intervals in one order, and “signal B” to mean the other order. To make the models as general as possible, we modelled observers at the level of the decision variable instead of choosing specific visual or auditory stimuli. The decision variable was D = E+I, where E is a normal random variable representing the contribution of the stimulus to the decision variable, and I is a normal random variable representing internal noise. E had mean zero on signal A trials, mean μE on signal B trials, and standard deviation σE on both types of trials. The internal noise I had mean zero and standard deviation σI, and was statistically independent of E. Signals A and B were equally likely.
We used the decision variable D to represent the information that the observer computes from a complete 2AFC trial. The most common model of 2AFC tasks, the difference model [11], assumes that observers compute one decision variable D1 from the first stimulus interval and another decision variable D2 from the second interval, and base their decisions on D = D1−D2. Our approach does not rely on the difference rule, and we do not need to consider the single-interval decision variables D1 and D2.
On each trial the model observers received a sample d from the decision variable D, and calculated the likelihoods that the value d would be generated on signal A and signal B trials:
P(d|A)=ϕ(d,0,(σE2+σI2)1/2)
(1)
P(d|B)=ϕ(d,μE,(σE2+σI2)1/2)
(2)
Here ϕ(x, μ, σ) is the normal probability density function. The observers used these likelihoods in Bayes’ theorem to find the posterior probability that the signal was A or B:
P(A|d)=P(d|A)P(A)P(d|A)P(A)+P(d|B)P(B)=P(d|A)P(d|A)+P(d|B)
(3)
P(B|d)=1−P(A|d)
(4)
Here we have used the fact that P(A) = P(B). The posterior probability matching observer chose response A with probability P(A|d) and response B with probability P(B|d). We call this 'veridical posterior probability matching' (VPPM), because the observer makes veridical estimates of the posterior probabilities given the value of the decision variable, and uses these probabilities in the probability matching rule. In the rest of this article, when we speak of 'posterior probability matching' we mean the VPPM model unless we specify otherwise. The signal detection observer used a maximum a posteriori (MAP) decision rule, and chose the response that had the greater posterior probability, P(A|d) or P(B|d). When the goal is to maximize the number of correct responses, the MAP rule is the statistically optimal strategy in this task.
We modelled the VPPM and MAP observers in a two-pass task as follows. On the first pass of each trial, the decision variable was D = E+I, with E and I drawn from their respective distributions. On the second pass the external component E of the decision variable was the same as on the first pass, and the internal component I was a new, independent sample. In S1 Text we show that the posterior probability matching observer's proportion correct in this task is
pC=∫−∞∞ϕ(u)2ϕ(u)+ϕ(u−d′D)du
(5)
Here ϕ(x) is the standard normal probability density function, and d′D is the signal-to-noise ratio of the decision variable D, d′D=μD/(σE2+σI2)1/2. We also show that with ρ = σI/σE, the probability of two correct responses on repeated trials is
pCC=(1+ρ−2)(1+ρ2)1/2∫−∞∞(∫−∞∞ϕ(u+v)ϕ(u+v)+ϕ(u+v−d′D)ϕ(v(1+ρ−2)1/2)dv)2ϕ(u(1+ρ2)1/2)du
(6)
and the probability of two incorrect responses is
pII=(1+ρ−2)(1+ρ2)1/2∫−∞∞(∫−∞∞ϕ(u+v−d′D)ϕ(u+v)+ϕ(u+v−d′D)ϕ(v(1+ρ−2)1/2)dv)2ϕ(u(1+ρ2)1/2)du
(7)
The probability of consistent responses across two repeated trials is PA = PCC+PII.
Green and Swets [11] show that the unbiased MAP observer's proportion correct is
pC=Φ(d′D/2)
(8)
Here Φ(x) is the standard normal cumulative distribution function. In S1 Text we show that the probability of the MAP observer making two correct responses on repeated trials is
pCC=ρ∫−∞∞Φ(d′D2(1+ρ−2)1/2−u)2ϕ(ρu)du
(9)
and the probability of two incorrect responses is
pII=ρ∫−∞∞(1−Φ(d′D2(1+ρ−2)1/2−u))2ϕ(ρu)du
(10)
Again, the probability of consistent responses is PA = PCC+PII.
We used Eqs (5) to (10) to find the proportion correct and proportion of consistent responses for the VPPM and MAP model observers at several signal-to-noise ratios (d′D) and internal-to-external noise ratios (σI/σE). We used eight values of d′D, evenly spaced from 0 to 2.6. We used σI/σE = 0, 1, and 2, which spans the range of internal-to-external noise ratios typically found with human observers [10]. Then, following Burgess and Colborne [12], we calculated the model observers’ apparent internal-to-external noise ratios from their proportion correct and proportion of consistent responses. Eqs (8), (9), and (10) give the proportion correct PC(d′D, ρ) and proportion of consistent responses PA(d′D, ρ) for a MAP observer as a function of d′D and ρ = σI/σE. We found the apparent internal-to-external noise ratio for the model observers by numerically minimizing the following sum-of-squares error:
(d′D^,ρ˜)=argmin(d′D,ρ) (pC^−pC(d′D,ρ))2+(pA^−pA(d′D,ρ))2
(11)
Burgess and Colborne’s method estimates the internal-to-external noise ratio of the decision variable of a MAP observer. Thus when we applied Burgess and Colborne’s method to the MAP model observer, we simply recovered the internal-to-external noise ratio that we had used to calculate PC and PA in the first place. When we applied the same method to the posterior probability matching observer, though, we did not recover the internal-to-external noise ratio that we had used to calculate PC and PA, because this observer has an additional source of internal variability, namely the probability matching rule. We emphasize this point: we applied Burgess and Colborne’s method to the posterior probability matching model observer, even though this observer does not satisfy that method’s assumption that additive internal noise is the only source of randomness in the observer’s decisions. We did this in order to discover what internal-to-external noise ratios Burgess and Colborne’s method would attribute to a posterior probability matching observer. We then used Burgess and Colborne’s method with human observers (see below) to see whether they showed the pattern of internal-to-external noise ratios that is predicted by the VPPM or MAP models. We will use σI/σE (or sometimes ρ for brevity) to denote the internal-to-external noise ratios that we used to calculate PC and PA for the model observers, and we will use ρ˜ to denote the apparent internal-to-external noise ratios that we calculated from PC and PA using Burgess and Colborne's method.
An observer’s efficiency at a task can be defined as the squared ratio of their d′ to the ideal observer’s d′: η=(d′/d′ideal)2 [15, 16]. The ideal observer is a theoretical model observer that achieves the best possible performance on the task. The VPPM observer’s proportion correct as a function of d′D is given by Eq (5), and the observer is unbiased, so its sensitivity is d′ = 2Φ−1(PC). (However, this method of calculating d′ is really only meaningful when the observer chooses responses by comparing the decision variable to a criterion, and the VPPM observer does not do this. Consequently, this is the apparent d′ that an experimenter who assumes signal detection theory would attribute to the VPPM observer, based on its proportion correct. This is fine for our purposes, since our goal is simply to find the VPPM observer’s efficiency.) To examine the best-case scenario, we assumed that the VPPM observer had no internal noise and that the decision variable captured all relevant information from the stimulus. In this case the ideal observer’s d′ is the signal-to-noise ratio of the decision variable, d′ideal = d′D. Thus the VPPM observer’s highest possible efficiency is η = (2Φ−1(PC)/d′D)2. We calculated the VPPM observer's efficiency with σI/σE = 0 and several values of d′D evenly spaced from 0.3 to 4.7.
Fig 1a–1c shows the model observers’ proportion correct and proportion of consistent responses at several signal levels and internal-to-external noise ratios σI/σE. At any given level of signal and noise, the posterior probability matching observer has lower proportion correct than the MAP observer. Furthermore, at any given proportion correct the posterior probability matching observer’s responses are less consistent than the MAP observer’s, i.e., they have lower PA.
Fig 1d shows the model observers’ apparent internal-to-external noise ratios ρ˜, calculated from the proportion correct and proportion of consistent responses shown in Fig 1a–1c. The posterior probability matching observer has very high apparent internal-to-external noise ratios at low proportion correct, and lower but still quite high ratios at higher proportion correct. The MAP observer has constant apparent internal-to-external noise ratios across all performance levels, as expected, since as explained earlier the apparent internal-to-external noise ratio ρ˜ simply recovers the internal-to-external noise ratio σI/σE we used to calculate the proportion correct PC and the proportion of consistent responses PA.
Fig 2 shows the posterior probability matching observer’s efficiency as function of proportion correct. Posterior probability matching is very inefficient at low performance levels, and even at a typical threshold performance level of 75% correct it reduces an otherwise ideal observer’s efficiency to around 50%. Efficiencies of 50% have been found in perceptual tasks at 75% threshold [13], and to be this efficient a posterior probability matching observer would have to make optimal use of the stimulus in all other respects, aside from probability matching. Furthermore, in a contrast increment detection task, Burgess, Wagner, Jennings, and Barlow [15] found efficiencies as high as 83% (standard error 15%) at 69% correct performance, and Fig 2 shows that at this proportion correct the posterior probability matching observer's maximum efficiency is 44%. Even in a task as complex as symmetry detection, Barlow [16] found efficiencies of 50% at 60% correct performance, whereas the posterior probability matching observer's maximum efficiency at this proportion correct is just 26%. Many authors have noted that probability matching is a suboptimal strategy, but perhaps it has not been realized previously how very inefficient it actually is.
In contrast, the MAP observer makes optimal use of the decision variable. To calculate the posterior probability matching observer's efficiency (Fig 2), we assumed that the internal-to-external noise ratio was zero, and that the decision variable captured all task-relevant information. Under these conditions the MAP observer is the ideal observer, so its efficiency is 100% at all values of proportion correct, and naturally no human observer can be more efficient than this.
Fig 3a shows the results of the two-pass response consistency experiments with human observers, and also a section of the lowest red curve in Fig 1d, which represents the posterior probability matching observer with no internal noise (solid red line). For reference, the dashed red lines show internal-to-external noise ratios at twice the value and half the value of the solid red line. The observers’ apparent internal-to-external noise ratios (coloured circles) are approximately constant across performance levels, and do not show the sharp increase at low proportion correct that is predicted by the posterior probability matching model. For the three observers, least-squares linear regressions of the apparent internal-to-external noise ratio against proportion correct have slopes and bootstrapped 95% confidence intervals of -0.37 (-2.85, 2.03) (red data points), 0.28 (-1.97, 2.21) (green data points), and -0.46 (-4.24, 2.36) (blue data points). Observers' performance approximately spans the range 58% to 75% correct, and at these values the noiseless posterior probability matching observer (solid red line) has apparent internal-to-external noise ratios of 1.89 and 0.93, respectively, corresponding to a slope of -5.6, which is well outside the 95% confidence intervals of the linear regression slopes for all three observers. Furthermore, at low proportion correct (<70%) the observers’ apparent internal-to-external noise ratios are lower than is ever possible according to the posterior probability matching model. These results show decisively that observers did not follow a posterior probability matching strategy in this task.
Observers clearly did not use a posterior probability matching strategy in Experiment 1. However, only three observers ran in the experiment, and they ran in almost 10,000 trials. Previous studies on classic probability matching (i.e., prior probability matching) have found large individual differences, with only some observers showing probability matching behaviour [1, 2]. Furthermore, previous studies have found that classic probability matching behaviour decreases with practice, e.g., Shanks et al. [1] found that probability matching declined over the course of 1800 trials, with less than half the participants showing probability matching by the end of the experiment. These findings raise concerns that in Experiment 1 we may simply have chosen three observers who did not happen to exhibit probability matching behaviour, or that any probability matching behaviour may have been eliminated over the course of the experiment.
To address these concerns, in Experiment 2 we ran a larger number of observers in a shorter version of the same task. Fig 3b shows observers’ apparent internal-to-external noise ratios as a function of proportion correct. Each observer is represented by a different coloured symbol, so each coloured symbol appears twice, once for the observer’s low performance trials (at the estimated 65% threshold) and once for the observer’s high performance trials (at the estimated 80% threshold). We clipped apparent internal-to-external noise ratios at a maximum of 4.0 in order to make them visible on the plot. The internal-to-external noise ratios are higher than in Experiment 1, probably because observers were less psychophysically experienced and did not run in the task as long. Confidence intervals are also much larger than in Experiment 1, for two reasons. First, there were fewer trials, which increased the standard error of the estimates of proportion correct and proportion of consistent responses. Second, internal-to-external noise ratios were higher, and Fig 1a–1c shows that iso-ρ lines are closer together at higher values of ρ, meaning that standard errors on estimates of proportion correct and proportion of consistent responses translate into larger confidence intervals on estimates of ρ.
Although the data is noisy, we can still test the VPPM model’s predictions. As explained earlier, the VPPM model predicts that apparent internal-to-external noise ratios increase sharply at low performance levels (Fig 1d). We will denote each observer’s apparent internal-to-external noise ratio at the low performance level by ρ˜low, and the apparent internal-to-external noise ratio at the high performance level by ρ˜high. Fig 4 shows the ratio ρ˜low/ρ˜high for each observer, versus the ratio ρ˜low/ρ˜high predicted by the VPPM model with no additive decision noise. We calculated the VPPM model’s predictions by looking up the predicted ρ˜low and ρ˜high for the noiseless VPPM model in Fig 1d (lowest red line), at each observer’s lower and higher proportion correct, and taking the ratio ρ˜low/ρ˜high. We clipped the ratios at a maximum of 4.0 to make them visible on the plot. Most data points in Fig 4 fall below the main diagonal, indicating that the ratio ρ˜low/ρ˜high is not as high as predicted by the noiseless VPPM model. A sign test shows that ρ˜low/ρ˜high is significantly lower than predicted by the noiseless VPPM model (17 of 21 data points below the diagonal, p<0.001).
However, we can also run less stringent tests of the VPPM model. The sign test reported in the previous paragraph tests the noiseless VPPM model, but observers may have sources of internal noise besides the posterior probability matching rule. If we re-run the sign test, comparing the actual ratio ρ˜low/ρ˜high for human observers to the ratio predicted by the VPPM model with an internal-to-external noise ratio of σI/σE = 1 (Fig 1d, middle red line), we find that the human observers’ ratio is lower than predicted in only 10 of 21 cases, which is not statistically significant according to a sign test (p = 0.50). Similarly, we can test the VPPM model simply by asking whether ρ˜low>ρ˜high for human observers, as is predicted by the VPPM model with any value of σI/σE. The prediction ρ˜low>ρ˜high is true for 12 of 21 observers, which is not a significant difference (p = 0.19). Thus our results with unpractised observers can rule out the noiseless VPPM model (which is what has been tested in previous studies of posterior probability matching in perceptual decision making), but they cannot rule out the VPPM model with additional sources of internal noise.
Our findings in Experiment 1 show that a broad class of posterior probability matching models is inconsistent with the results of two-pass response consistency experiments with practised human observers. Posterior probability matching predicts a sharp increase in the apparent internal-to-external noise ratio at low proportion correct that we do not find with practised observers, and at low proportion correct (<65%) it predicts apparent internal-to-external noise ratios that are higher than those we find with practised observers. Our results in Experiment 2 with less practised observers have the large measurement error associated with small numbers of trials and high internal-to-external noise ratios, but nevertheless we find evidence that internal-to-external noise ratios do not vary as much with performance level as the noiseless posterior probability matching model predicts. However, our results in Experiment 2 are consistent with VPPM models with additive Gaussian internal noise sources.
The contrast between our findings in Experiments 1 and 2 led us to examine previous two-pass studies. Fig 3c shows apparent internal-to-external noise ratios measured by Burgess and Colborne [12], Neri [10], and Murray, Bennett, and Sekuler [18]. All these studies used visual 2AFC tasks. The data points and confidence intervals representing Burgess and Colborne [12] and Murray et al. [18] show results for single observers, and those representing Neri [10] show averages across several observers. We recovered data from Burgess and Colborne using data capture software on their Figs 2 and 4, and the values from Murray et al. and Neri were given in their text (Neri’s Table 1, lines 4–8 and 14–20; Murray et al.’s results section for their Experiment 2). Fig 3c shows that most of these experiments do not span a wide range of performance levels, so they do not provide a strong test of the VPPM model’s prediction that internal-to-external noise ratios are higher at low performance levels. The most decisive evidence comes from Burgess and Colborne’s sine wave detection tasks (black circles and white circles). In these tasks the apparent internal-to-external noise ratios are only slightly higher at low performance levels than at high performance levels, and in fact at the lowest performance level (leftmost black and white circles) the internal-to-external noise ratios are much lower than is possible according to the VPPM model (solid red line).
Gold, Bennett, and Sekuler [19] found similar results in a two-pass response consistency experiment with a 1-of-10 identification task. We cannot show their data in Fig 3c, because a 1-of-10 task is not directly comparable to a 2AFC task, but Gold et al. found that apparent internal-to-external noise ratios were approximately constant over performance levels ranging from just above 10% correct to nearly 100% correct.
Our modelling results show that posterior probability matching causes a steep increase in apparent internal-to-external noise ratio at low performance levels. Burgess and Colborne [12] found that the internal-to-external noise ratio rose at low performance levels in a 2AFC sine wave detection task (black circles and white circles in Fig 3c), but not in a 2AFC disk detection task (black squares in Fig 3c). They noted that previous studies had found that intrinsic spatial uncertainty (i.e., observers being uncertain as to where a signal will appear, and so monitoring a range of possible locations) interfered much more with detection of low-contrast periodic signals such as sine waves, than with detection of low-contrast aperiodic signals such as disks. They suggested that the increase in internal-to-external noise ratios at low performance levels was due to intrinsic uncertainty. If this is correct, then we must be careful not to take an increase in apparent internal-to-external noise ratio as an unambiguous sign of posterior probability matching. In any case, this pattern in Burgess and Colborne’s data cannot be due to posterior probability matching, because as noted above, the internal-to-external noise ratios at low performance levels were much lower in their tasks than is possible according to the VPPM model.
It is worth noting that in most of the experiments we have examined, including those where we can reject the VPPM model statistically, apparent internal-to-external noise ratios are within a factor of two of the values predicted by the noiseless VPPM model, i.e., between the two dashed red lines in Fig 3a–3c. Some authors have argued that the randomness of a probability matching strategy can be adaptive, e.g., by allowing an organism or group of organisms to explore many possible courses of action, allocating effort according to the probability of success [20]. In signal detection theory, internal noise is usually thought of as an imperfection in perceptual processing. However, we speculate that under conditions where probability matching is adaptive, observers may also benefit from similar levels of randomness that are created by Gaussian internal noise and a MAP decision rule.
Would minor adjustments to the VPPM model produce better matches to human performance? A simple argument shows that many details of the model are unimportant, and that our findings are very general. In the VPPM model, the observer calculates veridical posterior probabilities P(A|d) and P(B|d) on each trial. On trials where PC = 0.5, the observer’s posterior probability estimates are P(A|d) = P(B|d) = 0.5, since any other values imply a higher proportion correct. On these trials the observer chooses responses A and B with probability 0.5, and so the proportion of consistent responses across repeated trials is also 0.5. However, PC = 0.5 and PA = 0.5 imply an apparent internal-to-external noise ratio of ρ˜=∞ (i.e., the lower end of the blue line labelled ∞ in Fig 1a–1c passes through the point (0.5,0.5)). This means that when the VPPM observer is uncertain (P(A|d)≃P(B|d)≃0.5), it must also be highly inconsistent across repeated trials. This is very different from a MAP observer, which may be uncertain and yet highly consistent if it has little or no internal noise. Consequently, if we use Burgess and Colborne's method to interpret randomness in terms of internal noise, we must attribute a near-infinite internal-to-external noise ratio to a posterior probability matching observer as its performance approaches chance, so long as the observer bases their responses on veridical estimates of posterior probabilities.
In contrast, a non-veridical posterior probability matching observer, i.e., one that makes inaccurate estimates of P(A|d) and P(B|d), could produce very different results from those in Fig 1. Suppose that a posterior probability matching observer always assigns zero or one to P(A|d) and P(B|d), even when performing at chance, and suppose that the observer has no additive internal noise (σI/σE = 0). This observer’s two-pass responses will be completely consistent, and the observer will have an apparent internal-to-external noise ratio of zero at all levels of proportion correct. It may also be possible to construct a non-veridical posterior probability matching observer that produces a more psychophysically plausible internal-to-external noise ratio. However, it is not clear what would be gained by constructing such a model, where probability matching is applied to inaccurate estimates of posterior probabilities that are, in effect, artificially chosen to avoid the high variability of posterior probability matching responses at low confidence levels. Nevertheless, this observation helps to define the range of models that our findings rule out.
The VPPM model should not be confused with an earlier probability matching model that was developed within signal detection theory [21–23]. In the earlier model the observer’s responses depend on whether the decision variable is greater than or less than a criterion. The observer chooses the criterion that makes their long-term response probabilities match the baseline signal probabilities, i.e., P(R = A) = P(A) and P(R = B) = P(B), instead of choosing the criterion that maximizes either the number of correct responses, or the expected payoff based on the signal probabilities and a payoff matrix. This is simply a standard signal detection model with a suboptimal criterion, and is quite different from the intrinsically random VPPM model that we have tested here and that has appeared in recent models of perceptual decision making. Furthermore, VPPM differs from prior probability matching models such as the one considered (and rejected) by Eckstein et al. [24], in which observers' response probabilities on each trial are simply matched to the baseline probabilities of the various signals, and the observer's response on a given trial does not even depend on the stimulus shown on that trial. A similar model has also been tested (and rejected) as an account of how observers distribute attention across possible signal locations [25].
It may be that people exhibit posterior probability matching in some types of perceptual tasks but not others. For example, there is at least one notable difference between our contrast discrimination task, where some observers did not exhibit posterior probability matching, and Gifford et al.’s [6] task, where model fitting suggests that they did. In our task, subjects discriminated between black-disk-first and white-disk-first stimuli, and the performance-limiting factors were the faintness of the signals and the power of the external noise. In Gifford et al.’s task, subjects categorized auditory signals of various frequencies as having been generated from one of two broad, overlapping frequency ranges. Subjects’ ability to perceive the signal frequency precisely was not the performance-limiting factor. Rather, the task was difficult because the two categories overlapped, and contained many of the same frequencies. Thus, even though this was a simple frequency categorization task, it was not limited by low-level sensory information, but instead by the fact that clearly perceived signals gave ambiguous information about the correct response. In this way it is similar to the cued probability learning tasks that have supported prior probability matching models in the past [26], i.e., models in which observers match their response probabilities to the baseline probabilities of the signals being viewed. We suggest that human observers may be more likely to show posterior probability matching in tasks where performance is limited by the ambiguity of clearly perceived signals, than in simple discrimination tasks that are limited by the perceptual signal-to-noise ratio.
In summary, posterior probability matching causes highly inconsistent responses, especially at low performance levels, and it is also inefficient. We can definitively rule out posterior probability matching models for practised observers in four typical perceptual tasks, where human observers are more consistent or more efficient than such models allow: the contrast polarity discrimination task reported here, sine wave detection [12], sine wave contrast discrimination [15], and symmetry detection [16]. We also find evidence that less practised observers are more consistent at low performance levels than noiseless posterior probability matching models predict, but these results are consistent with a posterior probability matching model with additional internal noise. We conclude that posterior probability matching is not a promising general-purpose model of expert perceptual decisions that are limited by low-level perceptual information. However, it may be viable as a theory of untrained perceptual decision making, and of decision making in tasks where performance is not limited by low-level perceptual information. The methods we have introduced also provide new tools for testing posterior probability matching models in more complex perceptual and cognitive tasks.
|
10.1371/journal.pcbi.1005337 | A Looping-Based Model for Quenching Repression | We model the regulatory role of proteins bound to looped DNA using a simulation in which dsDNA is represented as a self-avoiding chain, and proteins as spherical protrusions. We simulate long self-avoiding chains using a sequential importance sampling Monte-Carlo algorithm, and compute the probabilities for chain looping with and without a protrusion. We find that a protrusion near one of the chain’s termini reduces the probability of looping, even for chains much longer than the protrusion–chain-terminus distance. This effect increases with protrusion size, and decreases with protrusion-terminus distance. The reduced probability of looping can be explained via an eclipse-like model, which provides a novel inhibitory mechanism. We test the eclipse model on two possible transcription-factor occupancy states of the D. melanogaster eve 3/7 enhancer, and show that it provides a possible explanation for the experimentally-observed eve stripe 3 and 7 expression patterns.
| Biological regulation-at-a-distance, whereby a transcription factor (TF) is able to generate susbstantial regulatory effects on gene expression even though it may be bound a large distance away from its target (500 bp–1 Mbp), is only partially understood. Using a biophysical model and a computer simulation that take dsDNA and TF volumes into account, we identify a downregulatory mechanism which functions at large distances, whereby a TF bound within ∼ 150 bp from an activator decreases the probability of looping-based interaction between the activator and the distant core promoter. This “eclipse” mechanism provides insight into the question of how enhancer architecture dictates gene expression.
| Polymer looping is a phenomenon that is critical for the understanding of many chemical and biological processes. In particular, DNA looping has been implicated in transcriptional regulation across many organisms, and as a result plays a crucial role in how organisms develop and respond to their environments. While DNA looping has been studied extensively over the last several decades both experimentally [1–5] and theoretically [6–11], many aspects of looping-based transcriptional regulation remain poorly understood.
In vivo, the simplest looping-based regulatory architecture is comprised of a protein, or activator, that interacts with a distal site via DNA looping. Such simple architectures can be found in bacteria, where, for example, σ54 (σN) promoters are activated via such a mechanism [12–14]. In eukaryotes, DNA-looping-based regulation is associated with the interaction between the core promoter and distal regulatory regions called enhancers. These ∼500 bp enhancers typically contain clusters of transcription factor (TF) binding sites, and may be located from 1 kbp to several Mbps away from their regulated promoters. Detailed studies of enhancers from several organisms [15] have revealed that TFs can upregulate, inhibit, or both upregulate and inhibit gene expression via a variety of mechanisms. For example, repressors like D. melanogaster Giant, Knirps, Krüppel, and Snail inhibit expression either by partial overlap of their binding sites with that of an activator, or via a short-range repression mechanism termed “quenching”, whereby TFs positioned several tens of bps either upstream or downstream from the nearest activator inhibit gene expression [16–19].
In quenching, a bound protein inhibits gene expression, apparently without any direct interaction between the protein and either the promoter or the nearest-bound activator. In the prevailing model, quenching is assumed to be a result of histone deacetylation, which is facilitated by the formation of a ∼450 kDa DNA-bound chromatin remodeling complex made of a DNA-binding protein such as Knirps, C-terminus binding protein (CtBP) [20–22], and the histone deacetylases (HDACs) Rpd3 [23] and Sin3 [24]. However, this model falls short of providing a full description of quenching. In particular, short-range repressors such as Knirps have been shown to retain a regulatory function even without the CtBP binding domain [22, 25, 26], and tests for HDAC activity using either Rpd3 knock-outs [27] or the HDAC inhibitor trichostatin A failed to alleviate the observed repression effects [26]. Consequently, the mechanistic underpinnings of quenching are still poorly understood.
Previously, we studied looping-based regulatory mechanisms for synthetic bacterial enhancers, in which the activator was bound close to the promoter (within ∼1 Kuhn length, or ∼300 bp) [28]. We successfully modeled the experimental results using a modified worm-like chain model that takes excluded-volume considerations into account [29–31]. We found that the regulatory effects are maximal for TFs bound at the center of the looping segment (delimited by the activator and the promoter), and decrease as the looping segment length is increased.
In this report, we focus on regulation for looping segments much longer than the Kuhn length, which are more typical for eukaryotic enhancers. In this case, the elastic regulatory effects that we observed previously are negligible. Using the modified worm-like chain model, we show that the excluded volumes of DNA and a TF bound within ∼1 Kuhn length (∼300 bp) either upstream or downstream of one of the loop termini can block the “line-of-sight” of the other terminus, generating an eclipse-like effect, which reduces the probability of looping. Unlike the elastic effects that we reported previously, this eclipse-like effect is independent of looping-segment length for sufficiently long looping segments. Thus, our model offers a looping-based mechanistic model for quenching repression.
Linear polymers have been studied via simulations using a variety of methods [32, 33]. To model DNA fully it is necessary to take into account the experimentally observed non-entangled chromosomal DNA structure [34], and to include details regarding bound proteins. Recently, polymer rings have been studied as a model system for topologically constrained polymer melts, such as chromosomal DNA [35, 36]. In this work, we address looping of a linear polymer, neglecting the additional topological constraints, but including bound protrusions. We chose to simulate a linear DNA chain with a bound protein using a sequential importance sampling Monte-Carlo approach that we used previously to simulate the configurational space of bare DNA [29]. To adapt our algorithm to the case of protein-bound DNA, we take into account not only the growing chain but also the location of the protrusion (see Eq (4)). During chain generation, upon reaching link k, the simulation adds a hard-wall spherical protrusion with radius Ro at the location robject. If the protrusion overlaps any of the previously-generated chain links or protrusions, the chain is discarded. After generating the configurational ensemble, we identify the subset of “looped” chains. We provide the essential details of the simulation in this Section. Additional details can be found in S1 File.
To model the quenching effect of a bound repressor on DNA looping, we generated configurational ensembles for DNA with a spherical protrusion of size Ro = 9.2 nm or Ro = 18.4 nm located a distance of K = 95 bp or K = 135 bp from the chain origin along the chain, oriented either in the same direction as the looping volume δr or 180° from it (see Fig 1). We plot F(L) for the various configurations of Ro and K in Fig 2A. The data show that in the elastic regime (L ≤ b), protrusions bound in-phase with δr (solid lines, ↑) strongly reduce the looping probability relative to that of the bare DNA, while protrusions positioned out-of-phase to δr (dashed lines, ↓) increase the looping probability, as we showed previously [28]. However, in the entropic regime (L ≫ b), all chains converge to values of F → L ≫ b F ∞ ≤ 1. For protrusions that are bound in-phase (i.e. γk = 0°), F∞ is distinctly smaller than one, and strongly depends on both the distance to the nearest terminus and the size of the protrusion. Conversely, for protrusions bound out-of-phase (γk = 180°), F∞ is only slightly smaller than 1, with weak dependence on both protrusion size and position.
To further explore the extent of the quenching effect in the entropic or long-chain-length regime, we plot in Fig 2B the value of F∞ as a function of a wide-range of Ro and K, for γk = 0°. The heatmap shows both a non-linear decrease in F∞ as a function of protrusion size and a non-linear increase as a function of protrusion distance from the chain origin. The red line in the figure demarcates the closest possible location at which a protrusion can be bound without physically penetrating part of the looping volume δr. Thus, for volumes and protrusion positions that fall below this line, a second excluded volume effect contributes to the reduction in the probability of looping, leading to a sharp increase in the overall effect. Together, the panels in Fig 2 show that a sufficiently-large protrusion can substantially reduce the probability of looping, independent of loop length, provided that its binding site is within a small distance from either of the loop termini. For further details regarding the computation of F∞ and its error estimation, see S1 File.
To understand the long-range, length-independent effect shown in Fig 2, we examine the chains’ terminating segments of length T ≪ L. If T ∼ max r ∈ δ r | r − r o b j e c t | ≪ b, these segments resemble stiff rods, and the object obstructs the line-of-sight of one chain terminus from the other. This eclipse-like phenomenon is manifested by a reduction in the number of polymer chains that are able to reach δr. This, in turn, results in a smaller Plooped as compared with the case in which no protrusion is present. In the entropic regime and in the absence of protrusions, the generated “rods” approach the looping volume δr from all directions that are unobscured by the volume of the polymer in a homogeneous fashion [7]. Due to this isotropy in the distribution of the chain termini orientations within δr, the reduction in Plooped can be approximated by the solid angle that the eclipsing object subtends at δr. Consequently, F (Eq (6)) can be approximated for this “rod model” by:
F ∞ r ˜ ′ , R o = P looped object L , { r ˜ ′ , R o } P looped baseline L L ≫ b ≈ 4 π δ r - I chain - I object r ˜ ′ , R o + I chain ∩ object r ˜ ′ , R o 4 π δ r - I chain = 1 - I object r ˜ ′ , R o 4 π δ r - I chain + I chain ∩ object r ˜ ′ , R o 4 π δ r - I chain , (12)
where Ro is the radius of the spherical object, r ˜ ′ is the location of the object, which could be located statically at point r′ (in which case r ˜ ′ ≡ r ′), or located on the chain a distance K from the chain origin (in which case we use the terminology r ˜ ′ ≡ r ˜ ′ ( K ) figuratively to specify the progression of the protrusion along the chain). ℐ c h a i n ≡ ∫ δ r Ω c h a i n ( r ) d 3 r, where Ωchain(r) is the solid angle subtended at r by the polymer chain links. ℐ o b j e c t ( r ˜ ′ , R o ) ≡ ∫ δ r Ω o b j e c t ( r ˜ ′ , R o ) ( r ) d 3 r, where Ω o b j e c t ( r ˜ ′ , R o ) ( r ) is the solid angle subtended at r by the object, and ℐ c h a i n ∩ o b j e c t ( r ˜ ′ , R o ) corresponds to the solid angle contained in both ℐ c h a i n and ℐ o b j e c t.
In order to test the eclipsing hypothesis, we first computed F∞ for the case of an object statically positioned at an off-chain location r ′ = d u ^ 0, and without chain-chain interactions. In this simplified case, F∞ in Eq (12) can be approximated by the following eclipsing expression:
F ∞ r ′ , R o ≈ 1 - I object r ′ , R o 4 π δ r = 1 - ∫ δ r 2 π 1 - 1 - R o + w / 2 r - r ′ 2 d 3 r 4 π δ r ≡ F ∞ static , (13)
where we substituted
Ω object 2 π = ∫ 1 - R o + w / 2 2 | r - r ′ | 2 1 d cos θ = 1 - 1 - R o + w / 2 2 | r - r ′ | 2 . (14)
In Fig 3, we compare the value computed from Eq (13) (dashed cyan line) to F(L) computed by our sequential importance sampling algorithm for the same conditions (solid blue line). The data show that the eclipsing approximation F ∞ s t a t i c overestimates F∞. We reasoned that the main cause for this estimation error is that Eq (13) disregards the flexible polymer nature of the chain. We ran an additional Monte-Carlo simulation to quantify the correction resulting from polymer flexibility. Here, we generated pairs consisting of an end-terminus point in δr and a direction vector of the terminal link, both distributed uniformly. Short polymer chains of length T originating at the chosen points were grown with their first links oriented in the chosen directions. These chains can be thought of as the terminating segments of long chains that have a uniform distribution of their end-termini in δr. We found that the probability of a flexible-polymer chain to overlap the object increased relative to the probability within the “rod model”, resulting in a decrease in the probability of the chain to form a loop (magenta dashed line in Fig 3). Using this “terminating-segments” correction, the discrepancy between F∞ from the simulation and F ∞ s t a t i c from Eq 13 is partially accounted for. We attribute the additional reduction in the simulated F∞ to interactions between the object and the remaining L − T length of the chain.
In Fig 4 we plot F∞ as a function of an on-chain object of radius Ro, for several values of K. To compare the results of the numerical simulation to the full eclipsing model (Eq 12), we first note that when K is kept constant, ℐ c h a i n ∩ o b j e c t ( r ˜ ′ , R o ) | K = c o n s t ≈ c o n s t, as can be seen from the inset in Fig 4: the overlap between Ωchain (orange cones) and Ωobject (red cones) changes only slightly when the object grows by a factor of two. Furthermore, | r ˜ ′ ( K ) − r | K = c o n s t is approximately independent of Ro if ( R o + w / 2 ) ≪ | r ˜ ′ ( K ) − r | for all r ∈ δr. Thus, the dependence of F∞ on the radius Ro of an on-chain object can be derived from Eq (12):
F ∞ r ˜ ′ , R o | K = c o n s t ≈ ≈ 1 - I object r ˜ ′ , R o 4 π δ r - I chain + I chain ∩ object r ˜ ′ , R o 4 π δ r - I chain ≈ 1 - A I object r ˜ ′ , R o + B K ≈ 1 - A K R o + w / 2 2 + B K ≡ f K R o , (15)
where we approximated ℐ o b j e c t ( r ˜ ′ , R o ) ≈ ( R o + w / 2 ) 2 π ∫ d 3 r | r ˜ ′ − r | 2 using Eq (14) and ( R o + w / 2 ) ≪ | r ˜ ′ − r | r ∈ δ r. In Fig 4, we fit the numerical results for different values of K with functions of the form fK(Ro) (Eq (15)). The fits are in excellent agreement (R2∼ 0.99) with the numerical data.
A salient feature of enhancers is that binding sites for TFs are positioned both upstream and downstream of the activators. The eclipse model predicts that F ∞ ( r ˜ ′ ( K ) , R o ) = F ∞ ( r ˜ ′ ( − K ) , R o ), since for long chain lengths the correlation between the positions of both chain termini is completely abolished. Thus, from the perspective of one terminus, looping events can be initiated from any possible direction. To see if our simulation captures this symmetry, we explored a geometry in which the protein-like protrusion was positioned at negative K values, corresponding to a location outside of the looping segment delimited by the looping-volume (activator) and distal-terminus (promoter) locations. To do so, we generated an additional chain segment of length Q in the direction opposite to t ^ 1, starting from link 0, where Q ≫ K. We plot the results in Fig 5. The data show that in the elastic regime, the “outside” geometry (green line, K < 0) generates a significantly smaller effect on the probability of looping as compared with the “inside” architecture (blue line, K > 0). This lack of symmetry is probably due to the propensity of short loops to form a “tear-drop” shape, thereby reducing the quenching effect of any elements bound outside the looping segment [28]. However, for sufficiently large L/b, F(L) for both ±K enhancer geometries converge to the same value, as predicted by the eclipse model.
So far, we modeled an enhancer-promoter region by a chain of discrete semi-flexible links, with the activator and the promoter located at both ends of the chain. However, in vivo, the DNA chain extends far beyond the enhancer-promoter region in both directions, and is subject to confinement. We address the first issue in this section, and the issue of DNA confinement in the next section.
We computed the probability of looping for a generic loop of length L, located between two internal links of the chain. To do so, we extended the original length of the chain L by two flanking segments of 10 Kuhn lengths (10b) on both ends of the chain. We assigned the cumulative Rosenbluth weight of the extended chain to the central segment of length L containing the enhancer-promoter region. See S1 File, Section 1.3.
We ran a simulation with K = 60 bp and R0 = 11.9 nm (Fig 6, magenta line) and compared F(L) to the same case without flanking segments (Fig 6, blue line). The data show that there is a discrepancy in the long-range looping probability ratio between the two looping models. In particular, the down-regulatory effect of the protrusion is stronger when the enhancer-promoter region is modeled as part of a larger chain. In order to determine the individual contributions of the two flanking segments to the discrepancy, we ran additional simulations with different flanking-segment configurations. We found that only the trailing segment beyond the distal terminus of the looping segment contributes to this discrepancy. The addition of a single chain link of length w = 4.6 nm (0.092b) after the terminus of the looping segment (Fig 6, green line) diminishes F(L), accounting for approximately half of the discrepancy. The addition of only 4 links (0.368b) after the terminus of the looping segment (Fig 6, aqua line) accounts for the entire discrepancy, fully agreeing with the results for the case with 10b flanking segments on both ends of the looping segment. While an addition of a leading segment before the looping volume diminishes the looping probability (data not shown), it does not alter the looping probability ratio F(L). This can be seen from the comparison between F(L) for the case with a single link after the terminus of the looping segment (Fig 6, green line) and the case of a segment of length b before the looping volume and a single chain link after the terminus of the looping segment (Fig 6, red line).
We believe that these results depend strongly on the looping conditions. For the conditions used here (δω′ = 0.1 × 2π, see Fig 1) the leading flanking segment is inconsequential to the looping probability ratio. However, utilizing a uniform looping volume (δω′ = 4π) would lead to a diminished looping probability ratio as a result of leading segment addition. Similarly, the trailing flanking segment diminishes F(L) due to the fact that chain configurations that approached the looping volume from a direction colliding with the chain are no longer possible when the chain is extended by a trailing segment. If, however, the looping conditions were such that the chain terminus was required to approach the looping volume from a direction perpendicular to the looping volume cone axis, the trailing segment would become inconsequential to the looping probability ratio.
These results suggest that while there is a discrepancy between the two simulation modes (with flanking regions and without them), it arises from regions that are immediately adjacent to the enhancer-promoter region, and further downstream or upstream segments of the chain do not contribute to the down-regulatory effect.
To check the effects of polymer confinement on the looping probability, we generated chains confined by spheres of various radii (for full simulation details, see S1 File). In Fig 7, we present the normalized looping probability P l o o p i n g s p h e r e / P l o o p i n g f r e e and the looping probability ratio for the confined chains, for the range of confining-sphere radii 125–625 nm. Note that the looping probability increases relative to that of the unconstrained polymer when the gyration radius of the chain becomes comparable to the size of the confining sphere (Fig 7A). However, as can be seen in Fig 7B, the introduction of a bound protrusion does not influence the looping probability ratio F(L) up to a chain length of 10 kbp for confining spheres with radii ≥250 nm. For the smallest confining sphere (radius of 125 nm) we were only able to simulate chains of length ≤4.5 kbp (see S1 File Section 3.4), and here too the looping probability ratio remained unaffected.
Finally, we applied our model to a real enhancer-promoter system. We chose the experimentally well-characterized eve 3/7 enhancer of D. melanogaster, which is separated from its promoter by about 4 kbp. See Discussion for a detailed examination of the validity of this application.
We computed our model’s predictions for the active eve 3/7 enhancer, i.e. in the early developmental stages (before gastrulation—stage 14), and in the anterior region of the D. melanogaster embryo. To check whether the resultant probability of looping could provide a mechanistic explanation for the observed expression pattern, we simulated two states: first, a chain representing the enhancer that is fully occupied by spherical protrusions representing the activators dStat, Zld and the Knirps repressor–co-repressor complexes, and a second state where the chain is entirely devoid of bound protrusions. We tested three possible protrusion sizes, corresponding to reported repressor complex sizes: Knirps alone (46 kDa), Knirps bound to CtBP dimers (130 kDa), and a full putative 450 kDa complex, as reported by [23]. Since the exact looping conditions of the enhancer looping are unknown, we considered three choices of δω′ (see Materials and Methods) that represent different extreme cases.
The results, plotted in Fig 8, show that for all chosen looping conditions, there is a decrease in the probability of looping when the chain is fully bound by the protrusion representing Knirps. However, only the protrusion representing the full Knirps–co-repressor complex (containing dCtBP and the HDACs Rpd3 [23] and Sin3 [24]) was large enough to generate a significant reduction in looping.
We previously established [28] that a bound protein inside a loop can alter the probability of looping in a manner proportional to its size, when the chain length is of the order of the Kuhn length (i.e. < 300 bp). The simulations and theory presented here identify a separate regulatory effect that is relevant to much longer chain lengths. In particular, our model predicts a decrease in the probability of looping that is independent of chain length for long chains in the entropic regime (i.e. L ≫ 300 bp), provided that a sufficiently large protrusion oriented in-phase with the looping volume δr is positioned within one Kuhn length of one of the chain termini. We further showed that the reduction in looping probability resembles an eclipse-like phenomenon, where the protrusion blocks the line of sight of one chain terminus from the other.
Our model provides a biophysical mechanism for so-called short-range repression or quenching by enhancers, for sytems with sufficiently long separation between the enhancer and the core promoter (L ≫ b) [16–19]. The model successfully captures many of this phenomenon’s salient features. These include lack of dependence on chain length, symmetry with respect to binding-site positioning inside or outside of the looping segment, the dependence of the regulatory effect on both the size of the bound complex and the distance of the TF from the nearest terminus, and the typical distances of the TF from the terminus (≲ 150 bp) for which significant quenching can be generated.
We computed the looping probability ratio for the D. melanogaster eve 3/7 enhancer-promoter system in the segment of the embryo where the repressor Knirps is expressed. We found that using realistic repressor-co-repressor-HDAC complex sizes for the simulated protrusions, a significant reduction in looping probability can be obtained, showing that the looping-based mechanism can account for at least some of the quenching generated by Knirps in the context of early fly development. Since there is also a substantial body of evidence supporting HDAC catalysis as a major mechanism for repression [40], we conclude that it is likely that both HDAC activity and looping-related effects combine to generate the quenching observed in D. melanogaster and other organisms.
There have been previous attempts to model enhancer regulatory logic for the gap genes of early fly development [41–43]. Using a semi-empirical approach, which coupled a thermodynamic model to an empirically-based regulatory scoring function, these works demonstrated that TF occupancy of enhancers can determine the gene-expression outcome. By contrast, our model does not compute the actual expression pattern, but rather the probability of looping for a given occupancy state. For this computation, our model requires only knowledge of the enhancer structure (TF binding sites, protein-complex sizes, and over-all enhancer-promoter distance). This implies that no gene-expression experimental data is needed as an input to predict the down-regulatory effect of bound TFs.
While this work focused on bare dsDNA with a protrusion, our results are applicable to any linear polymer as long as its length is significantly larger than its Kuhn length. However, to apply our results to chromosomal DNA, other constraints must be taken into account, some of which we attempted to address in this paper. First, natural DNA loops occur between sites within the chromosome, and not between ends. We showed that looping of a linear segment with additional upstream and downstream flanking segments qualitatively resembles looping without flanking segments. Second, chromosomal DNA is subject to confinement. Third, DNA is typically a mixture of chromatinized and non-chromatinized DNA. We adress the second and third points below.
Due to confinement, chomosomal DNA is compacted into a globular state. The nature of the globular state varies between organisms (equilibrated globule in yeast vs. a possibly non-crosslinked globule in higher eukaryotes [34, 44–46]) and chromatin spans a wide range of volume fractions [36]. For lower volume fractions, DNA can be viewed as a semi-dilute polymer solution, in which coils are strongly overlapping. This semi-dilute solution may be pictured as a system of domains (blobs). Inside each blob, the chain behaves as an isolated macromolecule with excluded volume, and different blobs are statistically independent of one another [8, 47]. If the volume fraction is sufficiently low, chromosomal DNA within a blob can explore nearly the entire volume of the blob without interacting with other parts of the chromosome. Our model is applicable to non-chromatinized dsDNA within a blob, provided that the characteristic blob size of the organism is large enough to contain a sufficiently-long length of dsDNA.
For active eukaryotic promoters, there is evidence that the enhancer and the promoter regions are depleted of nucleosomes for up to 500–1000 bp both upstream and downstream from the center of each regulatory element [48]. This implies that the ends of the looping segment are non-chromatinized dsDNA, while the looping segment itself may be chromatinized. We argue that our model may be applied also for this case, provided that the looping segment is contained within a blob. This is because the eclipse effect is a result of the loss of correlation between the two chain termini. Thus, the effect should occur for any sufficiently-long chain (Lchain ≫ bchain), independent of the exact chain parameters (Lchain, bchain, wchain, etc.). The effect is sensitive to the TF size and location, and to the looping criteria. However, these are assumed to equal the conditions of non-chromatinized DNA at the chain termini (i.e. in close proximity to the enhancer and promoter).
We calculated typical blob sizes for a few model organisms (see S1 File Section 2). For bacteria, there are few examples of looping with lengths considered in this work, while the vast majority of the loops are short and are covered by the model described previously [28]. For D. melanogaster, a typical blob contains many kbp of chromatinized DNA. Thus our model may be applicable to enhancer-promoter systems in D. melanogaster and other organisms with low chromatin volume fractions, and to the eve 3/7 enhancer in particular.
For enhancer-promoter systems for which a non-interacting blob cannot be assumed (small blob size due to high chromatin volume fraction, or very long looping segment), we must consider the interaction of the looping segment with the rest of the chromosome. Recent results regarding chromatin structure [34] indicate that regions within eukaryotic chromosomes are unentangled but do interact (for example, by confining each other sterically). We attempted to simulate such confinement using a hard-wall sphere around the simulated chains. We found that the probability of looping is increased by confinement. This is perhaps not surprising, since regions of the order of 1 Mbp within chromosomes have been found to interact with a contact probability with power-law constant ∼-1 [34, 44], as compared to the ∼-1.5 prediction of the equilibirum globule model [8]. This behavior can be explained by the formation of self-organized non-entangled regions, which can form via a variety of mechanisms [45, 46, 49, 50]. Unlike the looping probability, the eclipse effect was found to be unaffected by confinement, up to the minimal confining sphere size that we were able to simulate. Note that the simulation dimensions are scalable: if we scale the bare dsDNA parameters by α = 10/4.8, the 104-link simulation confined by a sphere of radius r also describes a polymer with width of αw = 10 nm, length of αL ≈ 6800 nm, and persistence length of αlp ≈ 110 nm, which is comparable to ∼100 kbp of 10 nm chromatin fiber confined in a sphere of radius αr, if we assume a density of 15 bp/nm for the fiber [51]. The range of confining radii and chain lengths corresponds to polymer volume fractions of up to 0.2%. Despite being limited to low volume fractions, the result that the eclipse effect is independent on volume fraction suggests that our model might be applicable to chromatinized enhancer-promoter looping segments that extend beyond the blob size.
Our work indicates that we should consider three ranges for the physics of looping of chromatinized DNA. For short ranges which we studied previously [14, 28], looping is dominated by elastic energy. As shown in this work, there is an intermediate entropic looping range of up to ∼10 kbp of non-chromatinized DNA (or ∼100 kbp of partially-chromatinized DNA). Recent studies indicate that there may be an additional long-range regime in which interactions between neighboring regions of the globular DNA must be taken into account [34, 46].
|
10.1371/journal.pgen.0030030 | Phenotypic Plasticity in Drosophila Pigmentation Caused by Temperature Sensitivity of a Chromatin Regulator Network | Phenotypic plasticity is the ability of a genotype to produce contrasting phenotypes in different environments. Although many examples have been described, the responsible mechanisms are poorly understood. In particular, it is not clear how phenotypic plasticity is related to buffering, the maintenance of a constant phenotype against genetic or environmental variation. We investigate here the genetic basis of a particularly well described plastic phenotype: the abdominal pigmentation in female Drosophila melanogaster. Cold temperature induces a dark pigmentation, in particular in posterior segments, while higher temperature has the opposite effect. We show that the homeotic gene Abdominal-B (Abd-B) has a major role in the plasticity of pigmentation in the abdomen. Abd-B plays opposite roles on melanin production through the regulation of several pigmentation enzymes. This makes the control of pigmentation very unstable in the posterior abdomen, and we show that the relative spatio-temporal expression of limiting pigmentation enzymes in this region of the body is thermosensitive. Temperature acts on melanin production by modulating a chromatin regulator network, interacting genetically with the transcription factor bric-à-brac (bab), a target of Abd-B and Hsp83, encoding the chaperone Hsp90. Genetic disruption of this chromatin regulator network increases the effect of temperature and the instability of the pigmentation pattern in the posterior abdomen. Colocalizations on polytene chromosomes suggest that BAB and these chromatin regulators cooperate in the regulation of many targets, including several pigmentation enzymes. We show that they are also involved in sex comb development in males and that genetic destabilization of this network is also strongly modulated by temperature for this phenotype. Thus, we propose that phenotypic plasticity of pigmentation is a side effect reflecting a global impact of temperature on epigenetic mechanisms. Furthermore, the thermosensitivity of this network may be related to the high evolvability of several secondary sexual characters in the genus Drosophila.
| The phenotype of an individual is not fully controlled by its genes. Environmental conditions (food, light, temperature, pathogens, etc.) can also contribute to phenotypic variation. This phenomenon is called phenotypic plasticity. We investigate here the genetic basis of the phenotypic plasticity of pigmentation in the fruit fly Drosophila melanogaster. Drosophila pigmentation is strongly modulated by temperature, in particular in the posterior abdominal segments of females. The development of these segments is controlled by the homeotic gene Abdominal-B (Abd-B). Abd-B sensitizes pigmentation patterning in this region of the body by repressing several crucial pigmentation enzymes. It makes the regulation of their spatio-temporal expression in the posterior abdomen particularly sensitive to temperature variation. We show that temperature modulates the mechanisms regulating the dynamic structure of the chromosomes. Chromosomal domains can be compacted and transcriptionally silent, or opened and transcriptionally active. Temperature interacts with a network of chromatin regulators and affects not only the regulation of pigmentation enzymes but several traits under the control of this network. Thus, we conclude that the phenotypic plasticity of female abdominal pigmentation in Drosophila is a visible consequence for a particularly sensitive phenotype, of a general effect of temperature on the regulation of chromosome architecture.
| Phenotypic plasticity and buffering are concepts describing the phenotypic outcome of genotype-environment interactions. Phenotypic plasticity is the ability of a given genotype to produce different phenotypes in different environments [1]. It has been the subject of increasing interest as it is involved in adaptation and evolution [1–7]. Buffering, or canalization, is the ability of an organism to maintain a stable phenotype despite genetic variation or environmental fluctuations [8]. Buffering can be challenged by environmental stress, such as extreme temperatures. Thus, the question arises whether the plasticity of a particular phenotype is a specifically targeted reaction of the organism to changes in the environment or whether it is a side effect, reflecting a global process at the level of the transcriptome/proteome, but visible for weakly buffered phenotypes. To answer this question, we investigated the genetic basis of a particularly well described trait subject to phenotypic plasticity: the abdominal pigmentation of female Drosophila melanogaster, which strongly depends on the temperature conditions during development [9,10]. In the posterior abdomen, the differences of pigmentation between females grown at 20 °C and 29 °C are comparable to the phenotypic effect of mutations in major structural or developmental regulatory genes. The extreme plasticity of this phenotype makes it a particularly suitable model to dissect the responsible mechanisms. Within the last ten years, key studies have identified structural and developmental regulatory genes playing major roles in abdominal pigmentation patterning [11–16]. Because these studies focused on genetic factors, they were performed under standard temperature conditions [11,13–17]. Following a classical developmental genetics approach, we use mutations in key regulatory or structural genes to destabilize the underlying genetic networks and analyze how they interact with temperature.
Pigmentation is sexually dimorphic in D. melanogaster. In males, the abdominal tergites 5 and 6 are black and maintain this pigmentation at all temperatures. In contrast, the posterior abdominal pigmentation in females is highly polymorphic and plastic [18,19]. Figure 1A shows the abdominal pigmentation phenotypes of females from three different wild-type genotypes grown at different temperatures. At a given temperature, the extent of the dark region of the segments in females can differ dramatically between Drosophila lines, showing strong genetic basis [18,19]. NO1 and Samarkand are outliers and most lines have a pattern similar to that of BV1, comparable to the patterns described previously through pigmentation score [10]. Differences in plasticity are observed within each segment along the antero-posterior axis [10] and along the dorso-ventral axis (Figure 1A). This is extreme for A7, which can shift from completely black at 20 °C to completely yellowish at 29 °C. In addition, the transition border between the yellowish and the dark region of the tergites is not smooth but variegated (Figure 1B), implying that the control of pigmentation is not robust.
The spatial restriction of the phenotypic plasticity of pigmentation suggests the involvement of developmental regulatory genes. The morphology of abdominal tergites A5, A6, and A7 is specified by the posterior homeotic gene Abdominal-B (Abd-B) [20]. In tergites, Abd-B expression is low in A5, intermediate in A6, and high in A7 [21]. Thus, the increasing plasticity observed in the abdomen along the antero-posterior axis [10] perfectly correlates with the expression level of Abd-B. We used the Transabdominal mutation [17] to test whether ectopic expression of this gene in another body region is sufficient to generate a plastic pigmentation pattern. This mutation is a chromosomal rearrangement that fused the regulatory region of the stripe gene to the Abd-B locus [17,22]. It induces an ectopic expression of Abd-B on the thorax at the flight muscle attachment sites. This phenotype was previously described as sexually dimorphic, inducing melanin production in the whole sites of ectopic expression in males and in only restricted areas in females [17]. The pattern is indeed sexually dimorphic, but it is also extremely plastic (Figure 2). Remarkably, in females, the regions that are not brown at the sites of ectopic Abd-B expression show a very strong reduction in the production of yellowish pigments (Figure 2G–2L). This indicates that Abd-B plays opposite roles in melanin production. It either increases melanin production or represses the production of all pigments. Furthermore, these two roles are extremely thermosensitive. The increase of melanin production is much higher at low temperature, whereas the decrease in pigment production is much stronger at high temperature. These two roles of Abd-B are concomitantly observed within the same spot of ectopic expression, which suggests that they are influenced by other developmental pathways.
In order to quantify the effects of Abd-B and temperature on pigmentation, we tested how modifications of Abd-B expression level interact with temperature in the development of the pigmentation pattern. We varied the copy number of Abd-B using a deficiency of Abd-B (Df(3R)-RS-1–98) and a duplication of Abd-B (Dp-P5). Both mutations are carried in the same stock, which reduces background effects as much as possible. We found that high temperature decreases melanin production in all genotypes, but the effects of Abd-B level differed in A6 and A7 and along the dorso-ventral axis within A6 (Figure S1A–S1I). Thus, we quantified the melanin production along the antero-posterior axis in the lateral, median, and dorsal region of A6, in the lateral region of A7, and along the dorso-ventral axis in A7 (Figure S2). Temperature, Abd-B as well as the Abd-Bxtemperature interaction, strongly influenced pigmentation in each of these regions (p < 0.001 for all), except the dorsal midline (Table S1). These effects explained a large proportion of the total variation in abdomen pigmentation. The Abd-Bxtemperature interactions are particularly striking in the lateral regions of A6 and A7 and in the median region of A6 (Figure 3). In line with previous studies [13], Abd-B strongly increases melanin production in the lateral region of A6 under all temperature regimes (Figure 3A). The Abd-Bxtemperature interaction in the lateral region of A6 was mainly attributable to the pronounced reduction in pigmentation at high temperatures when the expression of Abd-B is low (Figure 3A). There was also a significant interaction between Abd-B and temperature in the lateral region of A7 (Figure 3C). However, in contrast to the lateral region of A6, Abd-B represses melanin production in the lateral region of A7. The opposite roles of Abd-B on melanin production are best illustrated by the very pronounced Abd-Bxtemperature interaction in the median region of A6. At low temperature, high Abd-B levels increase melanin production, whereas at high temperature they reduce melanin production (Figure 3B).
Based on these data, we conclude that Abd-B has two opposite roles on melanin production in females and can either increase or decrease melanin production. This makes the balance between melanin production and repression very unstable in the posterior abdomen, generating phenotypic plasticity in pigmentation. Indeed, this balance is very sensitive to temperature and is most pronounced in A7 showing the highest Abd-B level.
Abd-B is a developmental regulatory gene encoding a homeodomain transcription factor [23]. Its opposite roles on melanin production must be ultimately mediated by pigmentation enzymes. Indeed, pigment precursors move only a few cell diameters; thus, the spatial restriction of some of the enzymes synthesizing them is directly responsible for the pigmentation pattern observed in the adult [14,24,25]. A consensus model of pigment synthesis pathway is discussed in Text S1. Two classes of enzymes can be distinguished. Enzymes of the first class such as the tyrosine hydroxylase (TH) or the dopa decarboxylase (DDC) are required for the production of pigment precursors involved in the synthesis of all pigments. Enzymes of the second class, such as Ebony, Yellow, or Tan, are involved in the switch between the production of yellowish (NBAD sclerotin) or black-brown (dopa-melanin and dopamine-melanin) pigments [14,26].
The strong reduction in the production of all pigment observed in regions expressing Abd-B on the thorax of Tab/+ females suggests that Abd-B represses melanin production through the downregulation of one or several enzymes of the first class. The strong production of melanin observed at low temperature in the posterior abdomen and in some of the regions expressing Abd-B on the thorax of Tab/+ females suggests that Abd-B also regulates one or several enzymes of the second class. Mutations in genes encoding enzymes of the first class are homozygously lethal but loss-of-function mutations for enzymes of the second class are homozygously viable and can be used to identify the target(s) of Abd-B involved in plasticity. We postulated that if phenotypic plasticity of pigmentation is caused by temperature-dependent activity or regulation of a particular pigmentation enzyme, loss-of-function mutations in the gene encoding this enzyme should generate a nonplastic pigmentation phenotype. We used mutations in yellow (y1), tan (t1), and ebony (e1) (Figure S3). We observed that the plasticity of abdominal pigmentation is still visible in females mutant for y (Figure S3D–S3F), which lack black melanin but still have brown melanin. In t mutants, melanin is produced in males and females in the posterior abdomen at 20 °C, but is strongly reduced at higher temperatures in both sexes (Figure S3G–S3I and S3U). In contrast, e flies remain very dark at high temperature and show very limited plasticity of pigmentation (Figure S3J–S3L). Thus, a functional e gene is required for the plasticity of pigmentation. In flies mutant for t, which antagonizes e [15], the role of e is magnified. This suggests that the system responsible for plasticity in A5 or A6 also exists partly in males, but that it is normally hidden by the activity of Tan. The production of melanin in t mutants requires the repression or the strong downregulation of e, as even the gain-of-function of y cannot induce melanin production in the presence of Ebony [14]. Thus, the pigmentation pattern of t mutants at different temperatures implies that e is differentially regulated at different temperatures, relative to the production of melanin precursors, and that a major temperature-induced regulatory switch occurs between 20 °C and 25 °C.
We thus investigated the expression of the genes encoding Ebony and the two enzymes of the first class, TH and ddc. In order to visualize the expression of these enzymes, we stained pharate adults dissected out of their pupal case. We first investigated the expression pattern of e, ddc, and TH at 25 °C using e-LacZ, ddc-LacZ, and UAS-LacZ; TH-Gal4 flies, respectively. We observed that e, ddc, and TH are highly expressed in the pattern of the thoracic trident (Figure S4). The trident is a cryptic pattern fully visible in e mutants, y; e double-mutants (Figure S4B and S4C) [14], or when flies are grown at extreme temperatures [27] (Figure 2A). A similar pattern was described with an antibody against Ebony [14]. This suggests that the coexpression of e with TH and ddc assures that most of the locally produced pigment precursors are normally converted into yellowish NBAD sclerotin by Ebony. In the absence of Ebony, the excess of dopamine is converted into dopamine- melanin [14]. Thus, the melanin pattern in the absence of e is completely correlated to the spatial expression of TH and DDC in the epidermis.
We then looked at the expression of these enzymes in abdominal segments to see how their combined spatio-temporal expression could explain the pigmentation pattern. We observed with the e-LacZ transgene an expression similar to that previously reported using an antibody against Ebony [14]. It starts at the base of the bristles around 90 h after puparium formation (Figure 4A) and then becomes progressively uniform in the epidermal cells of the tergites (Figure 4B and 4C). We observed that the epidermal expression starts in the anterior region of the segment, as a weak antero-posterior gradient is first visible (Figure 4B). e-LacZ is stronger in the anterior region (Figure 4B, arrowhead) than in the posterior region of the segment (Figure 4B, arrow). We observed that the expression of ddc is also very dynamic in the posterior abdomen (unpublished data), but it is even more pronounced for TH, which encodes the first and limiting enzyme in the pigment synthesis pathway. Thus, we focused on TH. In the abdomen, TH expression started at the base of the large bristles on the posterior border of the segments before complete maturation of the bristles (Figure 4D). Expression at the base of more anterior bristles begins later (Figure 4E). Finally, TH is later expressed in epidermis of the whole tergites (Figure 4F). In the abdomen, it is expressed along an antero-posterior gradient as the expression starts much later in the more posterior segments (Figure 4D–4F). In particular, no strong expression is visible before hatching in A7 (Figure 4F). Thus, the expression of TH is lower in the posterior abdomen where the Abd-B level is the highest. We then looked at the expression of TH on the thorax of Tab/+ pharate females. We observed that it is also very dynamic. The expression is first visible at the base of the bristles located in the region of the trident (Figure 4G); then, a uniform expression is visible in the epidermal cells of the trident (Figure 4H). In the regions of ectopic Abd-B expression, expression starts at the base of the bristle located close to the teeth of the trident (Figure 4H, arrowhead). Later on, it is visible in the epidermal cells of these regions, but the regions that are devoid of pigments in Tab/+ flies show a much reduced staining (Figure 4I, arrows). Thus, the most plastic regions, i.e., the posterior abdomen and the regions of Abd-B ectopic expression in Tab/+ females, express TH very late. Furthermore, regions of ectopic expression of Abd-B devoid of pigments correspond to strongly reduced expression of TH. The delayed pigmentation in the posterior abdomen and the loss of TH expression in the regions of ectopic expression of Abd-B in Tab/+ females suggest that Abd-B represses TH, at least indirectly. There is an obvious difference between e and TH expression at 25 °C in the posterior abdomen. In particular, at 25 °C, e is already expressed in A7 before hatching, whereas TH is not yet expressed. We assume that TH is expressed in A7, which is pigmented, but this expression probably occurs after hatching. This is likely, as the activity of TH was reported to peak 50 min after hatching [28]. This means that when TH starts to be expressed, e being already expressed, DOPA and dopamine can be used to produce NBAD, the precursor of the yellowish pigment. Because the phenotype of tan mutants suggested that a major regulatory switch occurs between 20 °C and 25 °C, we analyzed the expression of e and TH in females grown at 20 °C. When flies are grown at 20 °C, the expression of e just before hatching is much weaker in the posterior abdomen than at 25 °C (Figure 4J). Furthermore, at 20 °C, TH expression can be observed very clearly before hatching in A7, but is mainly seen in association with bristles in the inner region of the tergite (Figure 4K).
Thus, Abd-B plays opposite roles in melanin production by repressing at least two genes encoding pigmentation enzymes with different roles in melanin production: TH required for the production of all pigments and Ebony required for the production of the yellowish pigment. It makes the expression of these enzymes particularly sensitive to temperature in the posterior abdomen. At low temperature, the stronger repression of e and the reduced repression of TH correlate with the increased melanin production observed in the posterior abdomen and on the thorax of Tab+/ females. In contrast, at higher temperature, the strong repression of TH and its delayed expression correlate with the reduced pigment production observed in the posterior abdomen and on the thorax of Tab/+ females. The effect on expression timing is visible on the pigmentation phenotype of the A7 tergite in limiting conditions, for example, in females with three doses of Abd-B grown at 25 °C: the melanin remaining is clearly associated with bristles, which mark the first sites of TH expression (Figure 4L).
How could temperature influence these opposite roles of Abd-B on pigmentation? Abd-B was previously shown to induce melanin production both via the repression of the transcription factor bric-à-brac (bab) and independently of bab [13]. bab, which was shown to repress melanin production, is strongly repressed in males by Abd-B in A5 and A6 [13]. This leads to the melanic pigmentation observed in the posterior abdomen [13]. bab is activated by the female-specific isoform of doublesex (dsxF), which compensates for the repression of bab by Abd-B, and thus reduces the amount of melanin produced in this part of the abdomen compared to males [13]. In female A7, bab is not repressed by Abd-B [13]. In order to analyze potential interactions between bab and environmental temperature, we used the babAR07 mutation that completely abolishes the expression of the two paralogs bab1 and bab2 and induces a well characterized haplo-insufficient melanic phenotype [13,18,29]. We observed that this phenotype is fully visible at 25 °C in A6, but is less obvious at other temperatures compared to wild-type (Figure S1J–S1L). Multivariate analysis of the effect of bab and temperature on melanin production (Table S2) revealed a very strong effect of bab and babxtemperature interaction in the lateral and median region of A6 and along the dorso-ventral axis of A7 (p < 0.001 for all, Figure 5A and 5B; Table S2). No significant effect was observed in the lateral region of A7 (Figure 5C). Thus, BAB level is less limiting in wild-type flies in A7, where bab is not repressed by Abd-B, than in A6. The role of bab on melanin production has been described previously [13,29], but these experiments did not reveal whether bab acts mainly by regulating pigmentation enzymes of the first or the second class. To identify the main targets of bab, we overexpressed bab1 in the dorsal domain using the pannier driver as previously described [13], but in an e or in a y background (Figure 6A and 6B). We observed that the production of both melanin and yellowish NBAD sclerotin is strongly repressed by the overexpression of bab1 (Figure S5A and S5B, arrows). It suggests that bab represses an enzyme of the first class. BAB has been reported to physically interact with products of the Broad-Complex [30], a direct regulator of ddc in pharate adults [31]. We used a ddc-lacZ transgene and observed that ddc-lacZ is downregulated in the dorsal domain of flies overexpressing bab1 (Figure S5C, compared to Figure S5D, arrows). Thus, the effect of bab on melanin production is mediated at least through the repression of ddc.
Interestingly, temperature was shown previously to modulate the effect of bab loss-of-function on another phenotype: the presence of an ectopic sex comb observed in males on the second tarsal segment of the first leg [32]. The sex comb is a structure made with modified bristles present on the first tarsal segment in Drosophila melanogaster males. Ectopic sex comb are extremely informative phenotypes frequently used to identify and quantify particular genetic interactions. The sex comb phenotype of bab mutants (distal sex comb) is observed also in some chromatin regulator mutants of the Polycomb group (PcG) and Enhancer of Trithorax and Polycomb group (ETP) [33–36]. The Polycomb group (PcG) and the antagonizing Trithorax-group (TrxG) proteins were identified through their role in the regulation of homeotic genes (Hox) [37], but it is now clear that they regulate hundreds of genes [38,39]. The PcG are involved in silencing of Hox genes, whereas the TrxG are involved in their activation. A third class of chromatin regulators has been described, the Enhancers of trithorax and Polycomb (ETP), required for both TrxG and PcG normal functions [40]. Most PcG mutants induce the formation of ectopic sex combs on the first tarsal segment of the second and third legs caused by ectopic expression of the homeotic gene Sex-comb reduced (Scr) [37,41]. However, ectopic distal sex combs are observed in mutants of only a few PcG or ETP genes [34–36]. This suggests that these different ectopic sex comb phenotypes correspond to the disruption of two distinct processes, and that bab and a subset of chromatin regulators are required for the repression of the distal sex comb. Other data suggest that these genetic interactions probably correspond to physical interactions between BAB and chromatin regulators. The bab locus encodes two closely related transcription factors with a BTB/POZ domain [29]. This interaction domain is present in many chromatin regulators [42] or transcription factors recruiting chromatin regulators [43]. In particular, BAB has been reported to bind to Batman/LolaL, which is part of PcG and TrxG complexes [42]. Interestingly, the activity of chromatin regulators such as members of the Trithorax/Polycomb system are known to be temperature-sensitive [38,44]. Silencing by PcG through characterized regulatory sequences known as Polycomb Response Elements was shown to be stronger at high temperature [38,44,45]. Thus, we hypothesized that the modulation of bab activity by temperature could take place via an effect of temperature on a network of chromatin regulators interacting with BAB.
We used the sex comb phenotype to test for genetic interactions between bab, genes encoding ETP or PcG, and temperature. We observed strong genetic interactions between bab and corto, cramped (crm), batman/lolal, and Trithorax-like (Trl) that encodes GAGA (Figure 6). Temperature strongly enhances the sex comb phenotype of crm7/Y; babAR07/+ males. At 29 °C, they die in their pupal case with sex comb teeth also visible on the third tarsal segment of the first leg (Figure 6B) in a large fraction of the individuals (6/18 observed legs). This phenotype is not observed at lower temperature or in single mutants. In addition, the second tarsal segment is inflated and shortened, which reduces the size of the ectopic sex comb and makes quantification impossible. Therefore, we quantified the crm-bab interaction only at 25 °C using the number of teeth in the ectopic sex comb on the second tarsal segment of the first leg (Figure 6C). Wild-type flies have no sex comb teeth on the second tarsal segment (0 ± 0, n = 16). The crm7/Y; babAR07/+ males have many more teeth (6.18 ± 0.25, n = 22) than crm7/Y hemizygotes (2.75 ± 0.18, n = 12) and babAR07 heterozygotes (0.18 ± 0.05, n = 36). This strong genetic interaction is shown in Figure 6C. We also quantified the effect of temperature on the interactions between bab and other chromatin regulators (Figure 6A and 6D, Tables S3 and S4). The single heterozygote mutants for corto, ban, or Trl do not show ectopic sex combs. We analyzed how these mutations modify the babAR07/+ phenotype. The genotype and the temperature accounted for 71% of the variance (Table S4). Temperature had a strong effect and increased sex comb teeth number across all genotypes. All heterozygote double mutants were significantly different from babAR07/+ (Tukey post hoc test, p < 0.001), which shows that the effect of bab mutation is strongly enhanced by mutations in corto, ban/lolal, or Trl. We tested three corto alleles. All alleles showed a similar trend, but the effects were stronger for corto420 and cortoL1. In addition, there was a significant genotype/temperature interaction (Table S4) visible in the curves corresponding to babAR07 single mutants and double heterozygotes (Figure 6D).
Given these strong effects, we analyzed the effect of mutations in chromatin regulators on the abdominal pigmentation of babAR07 heterozygote females. All the mutants we looked at had been induced in different backgrounds, so it was not possible to clearly differentiate the effect of the mutation itself from the background. Balancers are frequently used as control when the mutant stock is out-crossed to a wild-type line. However, most of the balancer chromosomes from the mutant stocks carry mutations in pigmentation genes, which is particularly inadequate in our case. Thus, we focused on genes for which we observed very strong phenotypes and interactions, and/or for which we could test different alleles. We observed very strong effects for the three corto alleles. They dominantly induce a reduced pigmentation in A7 at 25 °C. This is extreme for corto420 and cortoL1 (Figures S6 and S7) and weaker for corto07128 (Figure S7E). We tested how they would modify the haplo-insufficient pigmentation seen in babAR07 females. We observed a strong temperature-sensitive effect on pigmentation in babAR07/corto420 in A6 with a completely black phenotype at 20 °C, a strong variegation at 25 °C, and a completely white pigmentation at 29 °C (except for the dorsal midline) (Figure S7G–S7I). In A7, the pigmentation was very weak at 25 °C. A similar effect was observed with cortoL1 (Figure S8) and a visible but weaker effect with corto7128 (Figure S8). Quantification of melanin production revealed very strong effect for corto420 and strong interactions between corto and temperature and between bab, corto, and temperature (Figure 5A–5C, Table S1). In particular, whereas reducing bab level by half has no significant effect on melanin production in the lateral region of A7, it interacts very strongly with corto for this phenotype (Figure 5C, Table S1). In addition, in the median region of A6, babAR07/corto420 females produce less melanin than wild-type or single heterozygous females, whereas reducing bab level alone has the opposite effect (Figure 5B). This corresponds to the variegated phenotype observed at 25 °C (Figure S7H) and shows that bab and corto work together to increase melanin production in this region of A6. The females homozygous for the crm7 allele of the PcG gene crm show an absence of melanin in A7 and a very reduced and variegated pigmentation in A6 (Figure S8A). This is not observed in their heterozygote siblings crm7/FM7c (Figure S8B) or when out-crossed to a wild-type stock (Figure S8C). We also quantified how heterozygosity for crm7 would modify the pigmentation phenotype of babAR07/+ females. Except for the dorsal midline, we observed strong effect of crm and interaction between crm and bab, between crm and temperature, and between crm, bab, and temperature (Table S1).
The genetic interactions between bab and chromatin regulators for abdominal pigmentation and sex comb development, and previously reported physical interaction between BAB and the ETP Batman/Lolal [42], suggest that BAB and chromatin regulators cooperate in the regulation of particular targets. In order to test this hypothesis, we used antibodies against BAB, CRM, and Corto to localize their products on salivary gland polytene chromosomes. We observed many colocalizations on polytene chromosomes of BAB with Corto and BAB with CRM (unpublished data). In particular, we observed clear staining for BAB, Corto, and CRM in the cytological region corresponding to the locus of TH (65C) (Figure 7). BAB and CRM colocalized in the cytological region of the ddc (37C) (Figure 7G and 7H). We detected BAB alone in the cytological region of e (93C) (Figure 7E, 7F, 7K, and 7L).
Chaperones, in particular Hsp90, have been shown to buffer against the effect of cryptic genetic variation and environmental stress, in particular against high temperature [46,47]. Recent studies revealed a link between the chaperones and chromatin regulators [48–50], which suggested that the effect of temperature on chromatin regulators might be partly mediated by chaperones. We tested this hypothesis using two different alleles of Hsp83, the gene encoding Hsp90. In females, the allele Hsp83e6D dominantly induced a very low pigmentation in A7 at 25 °C (Figure S6Q), not observed with the allele Hsp83e6A (unpublished data), which had a weaker TrxG-like effect than Hsp83e6A [48].
We tested the effect on abdominal pigmentation of the two Hsp83 alleles in babAR07 heterozygous females. We observed that Hsp83e6D (Figure S6T and S6U), but not Hsp83e6A (unpublished data), strongly reduced the pigmentation phenotype of babAR07/+ at 25 and 29 °C. Quantification of melanin production revealed strong effect of Hsp83e6D and a strong interaction with temperature (Table S1). In contrast, interactions between Hsp83e6D and bab were only significant in the median region of A6. Significant interactions were observed between bab, Hsp83e6D, and temperature in the median region of A6 and along the dorso-ventral axis of A7 (Table S1). We also quantified the effect of Hsp83e6D on babAR07 heterozygote sex comb phenotype and found that it increased the number of teeth in the ectopic sex comb at 20 °C and 25 °C, but slightly decreases it at 29 °C (Figure 6D, Table S4). Because of the similarity of effects of Hsp83 and corto on the pigmentation phenotypes, we tested for potential genetic interactions between these two genes. We observed extragenic noncomplementation (lethality) between Hsp83e6D and cortoL1 (observed when crossed in both directions), whereas cortoL1 is viable with Hsp83e6A. The Hsp83e6D/corto420 and Hsp83e6D/corto07128 genotypes were viable, but strongly enhanced the loss of pigmentation observed in Hsp83e6D/+, corto420/+, or corto7128/+ females at 25 °C (Figures S6V–S6X and S7P–S7R). Quantification of melanin production revealed very strong interactions between Hsp83 and corto and between Hsp83, corto, and temperature (Figure 5D–5F; Table S1). In particular, the double heterozygote Hsp83e6D/corto420 females have an extremely reduced melanin production (Figure 5D–5F) and are the only genotype we analyzed where the pigmentation in A6 at the dorsal midline is affected at 25 and 29 °C. Furthermore, the Hsp83e6D allele (Figure S9) also dominantly induced pigmentation defects in males in the A5 segment observed at 25 °C and 29 °C, but not at 20 °C (Figure S9B and S9C). Most importantly, corto alleles strongly increased the pigmentation defects observed in Hsp83e6D/+ males at 25 °C and 29 °C (Figure S9D–S9I). Loss of pigmentation was visible in A6 with both alleles. In contrast, corto420/+ and corto07128/+ males had a normal pigmentation in A5 and A6 at all tested temperatures (unpublished data). We observed a similar phenotype in Hsp83e6D/babAR07 males (Figure S9J–S9L). This suggests that Hsp83, bab, and corto work tightly together to control abdominal pigmentation in both males and females and are much more required at 25 °C and 29 °C than at 20 °C.
Buffering or canalization describes the ability of individuals of a given species to show a constant phenotype despite genetic variations or environmental fluctuations. Phenotypic plasticity could therefore be interpreted as a weaker buffering of some phenotypes. Chaperones, and in particular Hsp90, have been identified as components of this buffering system and are thought to become limiting under stressful conditions such as high temperature [46,48]. The chaperone Hsp90 was proposed to act as a general evolutionary capacitor by releasing the effect of cryptic genetic variation under stressful environment [46]. However, more recent studies have revealed that the influence of Hsp90 on the variation of particular traits was very limited [51]. This suggests that the ability of chaperones, and Hsp90 in particular, to buffer phenotypic variation is not so general, but might rely on very specific interactions more tightly involved in particular phenotypes. Recent studies revealed a link between chaperones and chromatin regulators [48–50]. In particular, Hsp90 and several TrxG chromatin regulators were shown to buffer the same phenotype [48]. These studies on buffering in flies were based on the penetrance of deleterious phenotypes caused by cryptic genetic variation or known introduced mutations [46,48]. We found that mild modulation of a similar system by environmental temperature is involved in phenotypic plasticity of abdominal pigmentation. The chromatin regulator network we found to be sensitive to environmental temperature interacts genetically with the chaperone Hsp90 and the transcription factor BAB. It contains the PcG gene crm and the ETP gene corto. We observed very strong genetic interactions between corto and Hsp83, the gene encoding Hsp90. In particular, we observed extragenic noncomplementation between Hsp83e6D and cortoL1, whereas the viable trans-heterozygote combinations with the two other corto alleles induce strong reduction in melanin production in both sexes. This suggests that Hsp90 and Corto are involved in a common process. Hsp90 has been shown to physically interact with histones, to induce chromatin condensation, and to interact with topoisomerase II, which plays a crucial role in chromosome condensation [52,53]. Interestingly, corto is also required for the normal condensation of chromosomes during mitosis [33]. Therefore, the interactions between corto and hsp83 in gene regulation might be linked to a general role and common involvement of these genes in chromatin condensation.
The pigmentation in the posterior abdomen is particularly sensitive to environmental temperature because it is sensitized by the input of the homeotic gene Abd-B. Abd-B plays opposite roles in melanin production. The positive role of Abd-B in melanin production in females has already been described [13]. It is linked to the establishment of the sexually dimorphic pattern of pigmentation, and, for this role, Abd-B works antagonistically with bab by repressing it in A6 and A5 [13]. The role of Abd-B in repressing melanin production has not been described previously. It is very strong in A7 and is probably linked to the very peculiar morphology of this segment in females. In A7, bab is not repressed by Abd-B, and both genes work together not only to repress melanin production, but also to control some aspects of the particular development of this segment such as the absence of fusion of the tergites at the dorsal midline [29,54]. Abd-B plays these opposite roles in melanin production by repressing several pigmentation enzymes such as TH and Ebony. These enzymes start to be expressed at the end of pupal development [14,28]. Modulation of the regulation of these enzymes by temperature induces a local difference in their relative timing of expression in the abdominal epidermis. The effect is particularly visible in A7, which exhibits the highest Abd-B level. Studies in Drosophila wing have shown that pigment precursors can also be provided by the hemolymph [55]. Hence, it is possible that a change in the general level of pigment precursors in the hemolymph might contribute partly to the phenotypic plasticity of abdominal pigmentation. However, the diffusion of pigment precursors from the hemolymph does not seem to play an important role in the pigmentation of the body: recent studies in lepidopterans showed that the local production of DOPA and dopamine by TH and DDC in the epidermis are major components of the pigmentation pattern [56]. In Drosophila abdomen, epidermal clones of cells mutant for TH or DDC are albino [55], which shows that pigment precursors potentially available in the hemolymph cannot contribute significantly to pigment production in the body epidermis. In addition, in the thorax, the pigmentation patterns visible in e and y, e mutants perfectly correlate with the epidermal expression of TH and DDC, the enzymes providing pigment precursors (Figure S4). Furthermore, we clearly show that the spatial restriction of plasticity is strongly conditioned by Abdominal-B expression and the repression of pigment precursor production in the thoracic epidermis in Transabdominal mutants (Figure 2). Thus, we conclude that the modulation of the relative temporal expression of TH and ebony by temperature in the epidermis of the posterior abdomen is responsible for the phenotypic plasticity of female abdominal pigmentation. Mutations in corto, crm, hsp83, and bab enhance the effect of temperature on melanin production in the posterior abdomen. The colocalization of bab, corto, and crm at the locus containing TH in polytene chromosomes suggest that they might all cooperate in the direct regulation of this pigmentation enzyme, and that they might counteract the effect of temperature and Abd-B on TH expression. Their mutations enhance the repression of TH by Abd-B and high temperature, which explains why it has a particularly strong effect in A6 and A7, at 25 °C and 29 °C. We therefore propose the model presented in Figure 8 to explain some aspects of the pigmentation pattern plasticity in the posterior abdomen.
It does not exclude that temperature also modifies the expression of other genes. This is likely as the PcG/TrxG have hundreds of targets [38,39], and the thermosensitivity of the PcG/TrxG system is a general phenomenon observed with PRE from several different genes [38,44]. It is possible that other genes (developmental regulators or structural genes) are also modulated by temperature and contribute to the phenotypic plasticity of pigmentation. We observed, indeed, many colocalizations of Corto and BAB, and Corto and CRM, suggesting that this particular network of chromatin regulators regulate many targets. We demonstrate that this network is involved in at least two different phenotypes, abdominal pigmentation in females and sex comb development in males, both showing high temperature sensitivity. Thus, we propose that the plasticity of Drosophila pigmentation is a visible side effect, at particularly sensitive loci, of a process affecting the whole genome through alteration of epigenetic mechanisms.
Interestingly, abdominal pigmentation and morphology of the sex comb along the proximo-distal axis of the first leg evolve very rapidly in the Drosophila genus [57–59]. Remarkably, we found that these two morphological traits are under the control of a common thermosensitive network including the transcription factor bab, chromatin regulators, and the chaperone Hsp90. This suggests that the thermosensitivity of this particular regulatory network might be linked to the high evolvability of several secondary sexual characters in the genus Drosophila. Our results corroborate other studies, which have shown that the plasticity of specific traits is correlated to their evolvability [60].
Most of the fly stocks used in this study were provided by the Bloomington Drosophila Stock Center (http://flystocks.bio.indiana.edu). The following ones were kindly sent to us by various researchers: babAR07and UAS-bab1 (Jean-Louis Couderc), crm7 (Neel Randsholt), Df(3R)RS-1–98/Dp-P5 (Artyom Kopp), ebony-LacZ (Bernard Hovemann), and Pc3 Tab (Ian Duncan). The Tab mutation is associated with the Pc3 inversion in the stock we used, but the pigmentation phenotype visible on the thorax has been shown to be caused by the ectopic expression of Abd-B [17]. Flies were grown on standard agar-corn medium. Standard balancer chromosomes were used. For the effect of temperature, crosses were kept at 25 °C and tubes transferred after 2 d to the desired temperature. Oregon-R was used as a wild-type stock to outcross mutant stocks. All fly stocks are described in Flybase (http://flybase.bio.indiana.edu).
The interaction between bab, chromatin regulators, and chaperones was analyzed using sex comb phenotypes (except for the bab/crm interaction) and female abdominal pigmentation in the progeny of crosses between females Oregon-R or babAR07/+ and males carrying the mutation to be tested, or wild-type Oregon-R males. The interactions between crm (located on the X chromosome) and bab on sex comb was analyzed in the male progeny of crosses between crm7/FM7c females and males babAR07/TM6b, and compared to the effect of these mutations alone when crossed to Oregon-R. The effect and interaction with corto and bab of the hsp83e6D allele on male abdominal pigmentation was observed and analyzed in the male progeny of crosses made with females hsp83e6D/TM6b and wild-type males or carrying the mutation to test.
Flies were fixed in 75% ethanol 3 d after hatching to allow proper pigmentation of the cuticle to develop. Abdominal cuticles were cut on one side of the dorsal midline, cleaned, and mounted in Euparal (Roth). At least 15 flies were observed for each genotype/temperature condition, except the crm7 homozygote female escapers (Figure S3A) for which we had only five individuals. Thoraces were dissected in 75% ethanol, fixed 5 min in 100% ethanol, and mounted in Euparal.
Flies were dissected out of their pupal case, fixed, and stained without further dissection with X-Gal according to previously described protocols for pharate abdomen [61,62]. Because developmental time is conditioned by temperature [63], we staged pharate adults according to progression of eye color, bristle and wing melanization, meconium appearance, and ability of the fly to walk prematurely when dissected out of the pupal case [64]. Stainings were performed overnight at 37 °C for e-LacZ and ddc-LacZ, or 2 h for TH-GAL4; UAS-LacZ genotypes. Thoraces or abdomen were dissected after staining in PBS. This allowed us to make sure that absence of staining was not caused by tissue disruption during dissection prior to staining. Tissues were then dehydrated 5 min in 75% ethanol, 5 min in 100% ethanol, and mounted in Euparal.
Immunostaining of polytene chromosomes was performed as described by Cavalli (http://www.igh.cnrs.fr/equip/cavalli/link.labgoodies.html) on larvae of the w1118 genotype. The rabbit anti-CRM [35], rat anti-BAB2 [30], and rabbit anti-Corto [65] antibodies were used respectively, at 1:50, 1:200, and 1:25 dilutions.
All statistical analyses were performed using the software SPSS.10.0 or SPSS.13 [66]. We scored sex comb teeth number in the ectopic sex comb on the second tarsal segment. The number of teeth on the left and right legs of individuals was highly correlated for the bab-crm interaction (rs = 0.892, p < 0.001), as well as for bab and other chromatin regulator interactions across temperatures (rs = 0.667, p < 0.001). Therefore, we averaged the number of teeth of the left and right leg for all subsequent analysis. We analyzed the effect of mutations in bab and chromatin regulators on sex comb teeth number at three temperatures using a parametric two-way ANOVA. Log transformed data (ln + 1) were not exactly normally distributed; however, the residuals of the analysis did not deviate from normality (Kolmogorov-Smirnov test, p > 0.05).
We observed that the different genetic combinations affected melanin production differently in abdominal segments A6 and A7. Furthermore, within segments the effects differed along the dorso-ventral axis. Thus, we scored the proportion of melanin visible along the antero-posterio axis at the dorsal midline (A6D), on both sides in the lateral region of A6 (A6L1, A6L2), and in the median region of A6 (A6M1, A6M2) (Figure S2). In A7 we scored the proportion of melanin on both sides along the dorso-ventral axis (A7DV1, A7DV2) and the lateral region along the antero-posterior axis (A7L1, A7L2) (Figure S2). Ten individuals were scored for each genotype/temperature combination. Pigmentation scores varied between 0 (no melanin) and 4 (fully black) (Figure S2). Pigmentation scores between the left and right side were highly correlated within all regions (all: rs ≥ 0.925, p < 0.001). They were averaged in each individual (A6L, A6M, A7L, A7DV). We analyzed these pigmentation data using a multivariate analysis of variance. A6L, A6M, A6D, A7L, and A7DV were used as dependent variables; temperature and genotypes at Abd-B (one, two, or three doses), bab (one or two doses), corto (wild-type or corto420/+), Hsp83 (wild-type or Hsp83e6D/+), and crm (wild-type or crm7/+) as fixed factors. We included all main effects as well as possible interaction terms in the model (Table S1).
The model includes genes interacting with bab (encoding putative cofactors). Thus, in this model, although the effect of bab and the interaction between bab and temperature are highly significant (Table S1), they are also allocated to the interactions between bab and other genes, and between bab, other genes, and temperature. In order to test more generally the effect of bab and its interaction with temperature before dissecting the network (see Results), we also performed a multivariate analysis using the same dependant variables: temperature and genotype at bab as fixed factors in a reduced dataset with only wild-type and babAR07/+ females (Table S2). |
10.1371/journal.ppat.1004429 | Novel Cyclic di-GMP Effectors of the YajQ Protein Family Control Bacterial Virulence | Bis-(3′,5′) cyclic di-guanylate (cyclic di-GMP) is a key bacterial second messenger that is implicated in the regulation of many critical processes that include motility, biofilm formation and virulence. Cyclic di-GMP influences diverse functions through interaction with a range of effectors. Our knowledge of these effectors and their different regulatory actions is far from complete, however. Here we have used an affinity pull-down assay using cyclic di-GMP-coupled magnetic beads to identify cyclic di-GMP binding proteins in the plant pathogen Xanthomonas campestris pv. campestris (Xcc). This analysis identified XC_3703, a protein of the YajQ family, as a potential cyclic di-GMP receptor. Isothermal titration calorimetry showed that the purified XC_3703 protein bound cyclic di-GMP with a high affinity (Kd∼2 µM). Mutation of XC_3703 led to reduced virulence of Xcc to plants and alteration in biofilm formation. Yeast two-hybrid and far-western analyses showed that XC_3703 was able to interact with XC_2801, a transcription factor of the LysR family. Mutation of XC_2801 and XC_3703 had partially overlapping effects on the transcriptome of Xcc, and both affected virulence. Electromobility shift assays showed that XC_3703 positively affected the binding of XC_2801 to the promoters of target virulence genes, an effect that was reversed by cyclic di-GMP. Genetic and functional analysis of YajQ family members from the human pathogens Pseudomonas aeruginosa and Stenotrophomonas maltophilia showed that they also specifically bound cyclic di-GMP and contributed to virulence in model systems. The findings thus identify a new class of cyclic di-GMP effector that regulates bacterial virulence.
| Cyclic di-GMP is a bacterial second messenger that acts to regulate a wide range of functions including those that contribute to the virulence of pathogens. Our knowledge of the different actions and receptors for this nucleotide is far from complete. An understanding of the action of these elements may be key to interference with the processes they control. Here we have used an affinity pull-down assay using cyclic di-GMP-coupled magnetic beads to identify cyclic di-GMP binding proteins in the plant pathogen Xanthomonas campestris. This analysis identified XC_3703, a protein of the YajQ family that was able to bind cyclic di-GMP with high affinity. Mutation of XC_3703 led to reduced virulence of X. campestris to plants and alteration in biofilm formation. Genetic and functional analysis of YajQ family members from the human pathogens Pseudomonas aeruginosa and Stenotrophomonas maltophilia showed that they also specifically bound cyclic di-GMP and contributed to virulence in model systems. The findings thus identify a new class of cyclic di-GMP effector that regulates bacterial virulence and raise the possibility that other members of the YajQ family, which occur widely in bacteria, also act in cyclic di-GMP signalling pathways.
| Cyclic di-GMP (bis-(3′-5′) cyclic di-guanylate) is a second messenger in bacteria that acts to regulate a wide range of functions that include adhesion, biofilm formation, motility, synthesis of polysaccharides and synthesis of virulence factors in pathogens (recently reviewed by [1], [2], [3], [4]). The cellular level of cyclic di-GMP results from a balance between synthesis and degradation. Three protein domains are implicated in these processes: the GGDEF domain catalyzes synthesis of cyclic di-GMP from 2 molecules of GTP whereas EAL and HD-GYP domains catalyze hydrolysis of cyclic di-GMP, firstly to the linear nucleotide pGpG and then at different rates to GMP [1], [2], [3], [4]. All of these domains are named after conserved amino acid motifs. Most proteins with GGDEF/EAL/HD-GYP domains contain additional signal input domains, suggesting that their activities are responsive to signals or cues from the bacterial cell or its environment.
A number of cellular effectors or receptors for cyclic di-GMP have already been described in different bacteria [5], [6]. These include proteins with a PilZ domain, enzymatically inactive variants of GGDEF and EAL domains and a number of transcriptional regulators that do not possess a common domain organization [3], [5]. Furthermore, cyclic di-GMP is able to bind to untranslated regions of different mRNAs thereby affecting gene expression via riboswitches [3], [5]. Nevertheless, despite considerable progress, the mechanisms by which cyclic di-GMP exerts its action on diverse cellular processes remain incompletely understood, so that the discovery of further classes of cyclic di-GMP effectors is to be expected.
Here we have used an affinity pull down assay in order to identify potential cyclic di-GMP effectors in the phytopathogen Xanthomonas campestris pv. campestris (hereafter Xcc), the causal agent of black rot disease of cruciferous plants. As well as being a plant pathogen of global importance, Xcc is a model organism for molecular studies of plant-microbe interactions [7], [8]. Xcc has 37 proteins implicated in cyclic di-GMP synthesis and degradation and several of these are known to modulate synthesis of different virulence factors and virulence to plants [9], [10]. Thus far the only cyclic di-GMP effectors identified in Xanthomonas species are the transcription factor Clp, the enzymatically inactive GGDEF-EAL domain protein FimX and several PilZ domain proteins [5], [11], [12], [13], [14].
Our approach to reveal cyclic di-GMP binding proteins in Xcc described here identified XC_3703, a member of the YajQ family of proteins that is broadly distributed in bacteria. Mutational analysis showed that XC_3703 contributed to Xcc virulence to plants. Other members of the YajQ family from the human pathogens Stenotrophomonas maltophilia and Pseudomonas aeruginosa were also shown to preferentially bind cyclic di-GMP and to contribute to virulence in model systems. The findings thus identify a sub-group of the YajQ family of proteins as a new class of cyclic di-GMP effector.
To identify cyclic di-GMP receptor proteins, we performed an affinity pull-down assay using cyclic di-GMP–coupled magnetic beads and soluble protein extracts derived from the Xcc wild-type strain 8004. The selectively bound proteins were separated by SDS-polyacrylamide gel electrophoresis (Figure 1A) and were identified by peptide mass fingerprinting. Overall, 7 putative cyclic di-GMP binding proteins were identified from three cyclic di-GMP pull down experiments on Xcc 8004 lysates (Table S1). Three of these proteins were previously characterized cyclic di-GMP binding proteins from Xcc: the two PilZ domain-containing proteins XC_0965 and XC_3221 and the transcriptional regulator Clp, which is XC_0486. In addition, the analysis identified XC_1036, a protein containing a GGDEF domain with a predicted I-site (allosteric) cyclic di-GMP binding motif. Of particular interest to the work here was XC_3703 (MASCOT score 1156) a member of the highly conserved YajQ family of bacterial proteins. XC_3703 is related to YajQ of Escherichia coli (BLASTP E value is 10−20), a protein of unknown function that has motifs characteristic of nucleotide-binding proteins (Figure S1). YajQ family proteins are encoded by many bacterial genomes, to include both Gram-negative and Gram-positive bacteria. An amino acid sequence alignment of XC_3703 with sequences from a selected range of organisms is shown in Figure S1.
The ability of XC_3703 to bind different nucleotides including cyclic di-GMP was further examined using isothermal titration calorimetry (ITC) with the purified protein (see Materials and Methods). The full-length XC_3703 protein bound cyclic di-GMP with high affinity with a Kd of 2 µM (Figure 1B; Table S2). It appeared that binding was endothermic in nature, which is a characteristic of a binding mechanism that favours hydrogen bonding and hydrophobic interactions [15], [16]. The fitting of the binding isotherm data suggested a stoichiometry in which one molecule of cyclic di-GMP was bound by two molecules of the XC_3703 protein. The observed binding affinity is within the range of physiological cyclic di-GMP levels observed in Xcc and is similar to the binding affinity of a number of established cyclic di-GMP effector proteins. XC_3703 showed a much lower affinity for binding of ATP, GTP and cyclic di-AMP and no detectable binding of cyclic GMP or cyclic AMP (Figure 1B; Table S2). Taken together, these findings suggest that XC_3703 preferentially binds cyclic di-GMP.
Cyclic di-GMP is known to have a broad regulatory action in Xcc that includes influences on the synthesis of virulence factors and the formation of biofilms [7], [8]. The possible role of XC_3703 in these diverse regulatory actions was investigated by comparative phenotypic and transcriptomic analyses of the wild-type and XC_3703 deletion mutant. The XC_3703 mutant showed reduced virulence to Chinese Radish (Figure 2A) when tested by leaf clipping (see Materials and Methods) and reduced biofilm biomass when grown in complex medium (Figure 2B). Complementation restored these phenotypes towards wild-type.
Comparison of the transcriptome profiles of the wild-type and XC_3703 mutant by RNA-Seq showed that deletion of XC_3703 led to alteration in levels of transcript of a number of genes (Table S3, S4). These genes are associated with a range of biological functions that include bacterial motility and attachment, stress tolerance, virulence, regulation, transport, multidrug resistance, detoxification and signal transduction (Table S4).
The effect of XC_3703 mutation on the level of transcript was validated by quantitative reverse transcription polymerase chain reaction (qRT-PCR) (Figure 2C). The genes selected for these analyses represented those that have been previously implicated in virulence and biofilm regulation in Xcc but with a range of fold change of expression and of diverse functional classes. The relative expression levels of 12 genes measured using qRT-PCR reflected the differences in gene expression observed by transcriptome analysis (Figure 2C).
A number of cyclic di-GMP effector proteins exert their action through protein-protein interactions [6], [17]. To identify potential interacting proteins for XC_3703, we employed yeast two-hybrid (Y2H) analysis using the full-length XC_3703 protein as bait. Of the 36 preys isolated, the identities of 15 could be established by sequencing (Table S5). These represented partial or full-length proteins of diverse function including a putative octaprenyl-diphosphate synthase (XC_1377), a putative transcriptional regulator of the LysR family (XC_2801) and outer-membrane lipoprotein carrier protein (XC_2211). These particular protein interactions were confirmed when DNA fragments encoding the full length proteins were cloned and constructs were used individually as prey in Y2H analysis. The Y2H analysis was extended by Far-Western blotting experiments in which lysates of bacteria overexpressing the XC_3703 protein were separated by SDS polyacrylamide gel electrophoresis, transferred to nitrocellulose membranes and probed with the His6 tagged XC_1377, XC_2211 or XC_2801 which was then detected with an anti-His6 antiserum (Figure S2).
Since XC_2801 gave the strongest interaction signal we continued analysis with this transcriptional regulator. XC_2801 comprises a conserved N-terminal DNA binding domain and C-terminal LysR substrate binding domain (Figure 3A). Bacterial two-hybrid analysis using different truncated versions of XC_2801 showed that only full-length XC_2801 and the derivatives retaining the substrate binding domain interacted with XC_3703, while no interaction was seen with the isolated DNA binding domain (Figure 3A).
The findings outlined above indicate a potential interaction between XC_3703 and XC_2801. As a first step towards establishing a functional link between the two proteins, the phenotypic and transcriptional effects of mutation of XC_2801 were compared with mutation of XC_3703.
Mutation of XC_2801 significantly influenced the expression of 43 genes in Xcc; the regulated genes are associated with diverse functions that include extracellular enzyme production, attachment, biofilm formation and flagellar biosynthesis (Table S4). Quantitative RT-PCR methods were used to confirm alterations in expression of selected genes as revealed by RNA-Seq (Figure S3). Notably, comparison of the effects of mutation of XC_3703 and XC_2801 on the transcriptome revealed a significant overlap of regulatory influence (Figure 3B).
As outlined above, the XC_3703 mutant showed reduced virulence in Chinese Radish when plants are inoculated by leaf clipping. In contrast, the XC_2801 mutant showed no alteration in virulence when this method of inoculation was used. However, when plants were inoculated by spraying both XC_3703 and XC_2801 mutants showed reduced virulence, as indicated by a reduction in the percentage of inoculated leaves that develop black rot symptoms (Figure 3C). The effect of XC_2801 mutation on virulence was less pronounced than that of XC_3703 mutation however. Complementation of the XC_2801 mutant by introduction of XC_2801 cloned in pLAFR3 restored virulence to near wild-type level (Figure 3C).
Mutation of XC_2801 also had an effect on Xcc biofilm formation, although the effect was not as pronounced as that following mutation of XC_3703 (Figure 3D). Complementation with a clone expressing the full-length XC_2801 protein restored biofilm formation of the XC_2801 mutant to near wild-type levels. Taken together, these observations support the contention the effects of XC_3703 are exerted, at least in part, by interaction with XC_2801.
The transcriptional analyses outlined above showed that levels of transcripts of the flhB, aaeA, fliL and flgG genes were decreased approximately fivefold in both the XC_3703 and XC_2801 mutant compared to the wild-type. The regulation of these genes by XC_2801 was also examined by the use of promoter fusions to gusA (see Materials and Methods). Differences in the level of GusA between the XC_2801 mutant and wild-type were seen with all four fusions (Figure 4A), demonstrating that XC_2801 (directly or indirectly) regulates the expression of flhB, aaeA, fliL and flgG.
To further investigate the role of XC_2801 in regulation of transcription and any potential interplay between XC_2801 and XC_3703, electrophoretic mobility shift assays (EMSA) were used to examine protein binding to target promoters. His-tagged expression constructs of XC_2801 and XC_3703 were generated and both were subsequently purified by nickel affinity column chromatography. Purified His-tagged XC_2801 fusion protein caused no mobility shift of DNA probes spanning the upstream region of flhB, aaeB, fliL and flgG (Figure 4B). Interestingly, incubation of His-tagged XC_2801 together with XC_3703 led to a mobility shift of a DNA probe spanning the upstream regions of flhB and aaeB but not fliC or flgG (Figure 4B) whereas XC_3703 alone had no effect (Figure S4). These observations, taken together with the two-hybrid and far-western analyses presented above, suggest that under the conditions used, XC_2801 only binds to the upstream region of flhB and aaeB as a complex with XC_3703.
As LysR-type transcriptional regulators (LTTRs) typically bind a specific motif, T-N11-A, the upstream region of flhBA was examined for such a motif. The sequence TCCCGAATCCCGA was identified 80–67 bp upstream of the putative translational start site of flhB (Figure 4C). To investigate if this motif was essential for protein binding, EMSA was performed using an overlapping DNA fragment that lacked the full putative binding site. The presence of both XC_2801 and XC_3703 proteins did not cause a mobility shift of this DNA fragment demonstrating that XC_2801//XC_3703 binding requires the complete TCCCGAATCCCGA motif (Figure 4D).
As outlined above, XC_3703 preferentially binds cyclic di-GMP with a high affinity. This raised the question as to whether cyclic di-GMP could influence the binding of the XC_2801//XC_3703 complex to target promoters. To investigate this, the EMSA assay for the binding of XC_3703//XC_2801 to the flhBA promoter was repeated in the presence of added cyclic di-GMP at different concentrations. Addition of cyclic di-GMP at 1 µM had no apparent effect, but at 10 µM completely prevented DNA binding (Figure 5A). Other nucleotides did not have the same effect (Figure S4). XC_2801 alone did not bind to the flhBA promoter (as shown above) and addition of cyclic di-GMP did not alter this outcome (Figure 5B). Furthermore, XC_2801 appeared to have no affinity for the nucleotide as measured by ITC (Figure S5). These results suggest that cyclic di-GMP inhibits the interaction of the XC_3703//XC_2801 complex with DNA by binding to XC_3703, thus preventing the transcription of flhBA.
Bioinformatic analysis reveals that proteins of the YajQ family are widely conserved in bacteria (Figure S1). To examine whether cyclic di-GMP binding was a common feature of this family, recombinant YajQ-like proteins from various bacteria including Escherichia coli, Pseudomonas aeruginosa, Stenotrophomonas maltophilia, Bacillus cereus and Clostridium species were purified and their ability to bind different nucleotides was assessed. Interestingly, affinity for cyclic di-GMP binding was only seen in YajQ family proteins from P. aeruginosa (PA4395) and S. maltophilia (Smlt4090) (Table S1). YajQ family proteins from E. coli, Clostridium sp. and B. cereus by contrast exhibited a greater affinity for ATP and/or GTP than for any cyclic mono- or di-nucleotide.
The finding that XC_3703 can influence the virulence of Xcc prompted us to test whether PA4395 and Smlt4090 have a role in the virulence of P. aeruginosa and S. maltophilia respectively. Previous studies have established a correlation between the ability of strains of P. aeruginosa and S. maltophilia to attach and induce cytotoxicity in cultured mammalian cells and the virulence of these strains to mice [18], [19]. Mutation of either PA4395 or Smlt4090 led to reduced adhesion to monolayers of human bronchial epithelial cells compared to the wild-type (Figure 6 A, B). Importantly, the reduction in adhesion of these mutants could be restored to the level of the wild-type strain by complementation (Figure 6 A, B).
On the basis on these findings, we extended these studies to utilize the persistence mouse model of pulmonary infection. C57Bl/6 mice were intra-tracheally infected with wild-type P. aeruginosa, the PA4395 mutant, wild-type S. maltophilia or the Smlt4090 mutant and clearance of strains from the lung was examined at 1, 3 and 5 days post infection (see Materials and Methods). Both the PA4395 and Smlt4090 mutants had reduced persistence in C57BL/6 mice compared to their respective wild-type strains (Figure 6 C, D).
The findings from these infection models indicate that YajQ family proteins of P. aeruginosa and S. maltophilia significantly contribute to virulence of these opportunistic pathogens.
The YajQ family of proteins is broadly distributed in bacteria, with typically one member of this family in each species. YajQ family proteins have motifs characteristic of nucleotide or nucleic acid-binding proteins [16], [20]. Accordingly, YajQ from Escherichia coli has been shown to bind GTP and tRNA [16]. Here we show that some members of the YajQ family are novel cyclic di-GMP effectors that selectively bind this second messenger and act to regulate bacterial virulence. In particular we show that XC_3703 from Xanthomonas campestris influences the transcription of genes that contribute to virulence in plants and biofilm formation. We provide evidence that this action of XC_3703 is exerted, at least in part, through protein-protein interactions with the LysR family regulator XC_2801 in a manner that is negatively regulated by cyclic di-GMP. XC_3703 must exert a regulatory effect via additional pathways not involving XC_2801, but these currently remain unidentified.
We further identified the YajQ protein PA4395 of P. aeruginosa as a cyclic di-GMP binding protein. Previous studies to identify cyclic di-GMP-binding proteins from P. aeruginosa in a global fashion have used pull-downs with a cyclic di-GMP analog (2′-aminohexylcarbamoyl-cyclic di-GMP) covalently coupled to Sepharose beads [21] or a tri-functional capture compound incorporating cyclic di-GMP, biotin and a photo-activated reactive group [22]. Both of these studies identified a range of proteins, including six common proteins with an established function in cyclic di-GMP binding, but neither identified PA4395. Interestingly, of the novel proteins described in these two studies, only five were identified by both approaches, leading to the suggestion that the different compounds used for the experiments might be specific for a certain subset of proteins [22].
Knowledge of the cellular function of proteins of the YajQ family has been thus far restricted to the protein of Pseudomonas syringae that is implicated as a host factor controlling transcription in the bacteriophage Φ6 [23]. YajQ of P. syringae exerts an effect on phage RNA polymerase by protein-protein interactions involving the major protein (P1) of the phage core. Intriguingly this action is modulated in later phases of viral infection, even though YajQ is still present [23]. The underlying mechanism is unknown, although it has been suggested that a change in the chemical environment of the host cell between early and late phases of infection is responsible [23]. Our findings raise the possibility that one controlling factor could be the level of cyclic di-GMP in the bacterial cell.
In this study, XC_3703 was observed to bind cyclic di-GMP with an affinity of ∼2 µM. Measurements of cyclic di-GMP levels in wild-type Xcc strains under various conditions indicate that this value is within the physiological range of the concentration of the nucleotide [24], [25], [26]. The affinities of cyclic di-GMP effectors from a variety of organisms have been found to vary considerably, ranging from 10–15 µM down to ∼100 nM [27]. In Xcc, the transcription factor Clp binds cyclic di-GMP with an affinity of 3.5 µM, whereas the EAL domain of FimX has an affinity of 0.4 µM [11], [12]. The difference in binding affinities of these effectors has been suggested to represent a method whereby different responsive elements are selectively activated (or inactivated) as the cellular concentration of cyclic di-GMP changes [27].
Our experiments also show that not all members of the YajQ family preferentially bind cyclic di-GMP, suggesting that they have other cellular roles. This contention is supported by the observation that YajQ family members are found in bacteria that do not apparently have cyclic di-GMP signaling, such as Haemophilus influenzae [20], [28]. Proteins of the YajQ family show high level of amino acid sequence similarity but with some sequence divergence in the central part of the protein. Further work including structural studies will be required to identify residues that are critical for cyclic di-GMP binding, which may then allow bioinformatic discrimination of those family members that have this capability.
XC_3703 represents the first member of a new class of cyclic di-GMP binding protein and our findings suggest that the regulatory action of XC_3703 is exerted in part through protein-protein interaction with XC_2801, a protein of the LysR family that is the most abundant type of transcriptional regulator in bacteria [29]. Studies of members of the LysR family of transcriptional regulators indicate considerable structural flexibility, which may account for their variety of regulatory modes and versatility [30]. In general, the binding of small molecular weight molecules to the C-terminal co-inducer-binding domain modulates the activity of this family of regulators. XC_3703 also interacts with the C-terminal domain of XC_2801, suggesting a variation on the co-inducer mode of regulation (Figure S6). It appears that cyclic di-GMP specifically binds to XC_3703 with high affinity and that binding prevents the interaction of XC_3703//XC_2801 with a target promoter sequence (flhBA). In this manner, although XC_2801 alone does not bind cyclic di-GMP, transcriptional activation by XC_2801 is prevented when the concentration of cyclic di-GMP is high. Consistent with this view, the levels of expression of genes that we show here to be regulated by XC_2801 are reduced under conditions in which the cellular levels of cyclic di-GMP are high [30]. The absence of binding of XC_2801 or XC_2801//XC_3703 to the promoters of the fliL and flgG genes in EMSAs may indicate that the expression of these genes is indirectly regulated by XC_2801. We cannot however exclude that XC_2801 may also respond to a small-molecular weight co-inducer, which would be missing from these assays.
Notably, mutation of XC_2801 affects expression of only a subset of the genes that are regulated by XC_3703. Furthermore, loss of XC_3703 influences virulence of Xcc when the bacteria are introduced into the leaf vascular system by leaf clipping, but loss of XC_2801 has no effect. From these observations we surmised that XC_3703 must exert additional regulatory effects, independently of XC_2801, that contribute to bacterial virulence. The Y2H analysis of proteins that interact with XC_3703 may give clues to the identity of such additional pathways. Intriguingly, interaction is seen between XC_3703 and HpaR1 (XC_2736), a GntR family transcriptional regulator that has been previously shown to control the hypersensitive response and virulence in Xcc [25]. Further work is required to determine the importance of this interaction to the virulence of Xcc.
Finally, we determined that YajQ proteins from the opportunistic human pathogens P. aeruginosa and S. maltophilia are capable of binding cyclic di-GMP and showed by mutational analysis that these proteins significantly influence virulence, biofilm formation and persistence in models of human infection. S. maltophilia is a xanthomonad, related to Xcc whereas P. aeruginosa is more distantly related. One challenge is to establish if regulatory interactions between proteins of the YajQ and LysR families have a role in cyclic di-GMP signaling in these other bacteria. Interference with cyclic di-GMP signaling has emerged as a promising approach towards treatment of bacterial biofilm formation and the control of virulence and disease. Potential targets include the enzymes involved in cyclic di-GMP synthesis or degradation and the effectors by which cyclic di-GMP exerts its action. Our knowledge of the mechanisms of cyclic di-GMP signaling and associated effectors is however incomplete, and further understanding will be key to the progress of a promising approach for disease control into effective therapeutic strategies.
Xanthomonas campestris pv campestris (Xcc) strains and culture conditions have been described previously [31], [32], [33]. Most experiments were carried out in NYGB medium, which comprises 5 g liter−1 bacteriological peptone (Oxoid, Basingstoke, U.K.), 3 g liter−1 yeast extract (Difco), and 20 g liter−1 glycerol. For biofilm formation, Xcc was grown in L medium, which comprises 10 g liter−1 bactotryptone (Difco), 5 g liter−1 yeast extract, 5 g liter−1 sodium chloride, and 1 g liter−1 D-glucose. E. coli strains were grown in LB medium at 37°C. Other plasmids and strains used are shown in Table S6. Where required antibiotics were used at concentrations of 100 µg ml−1 for ampicillin, 50 µg ml−1 for rifampicin, 20 µg ml−1 kanamycin and 15 µg ml−1 tetracycline.
Common molecular biological methods such as isolation of plasmid and chromosomal DNA, polymerase chain reaction (PCR), plasmid transformation as well as restriction digestion were carried out using standard protocols [34]. PCR products were cleaned using the Qiaquick PCR purification kit (Qiagen) and DNA fragments were recovered from agarose gels using Qiaquick minielute gel purification kit (Qiagen). Oligonucleotide primers were purchased from Sigma-Genosys.
In-frame deletion of selected genes was carried out using pK18mobsac as described previously for XC_3703 and XC_2801 [31], [32]. Mutants were also created by the disruption of genes with the use of the plasmid pK18mob as described previously for Xcc [31].
The DNA fragments encoding the proteins of interest were synthesized by Gene Oracle in pGOv4 and sub-cloned into pET47b or pLAFR3 before transformation into E. coli BL21 (DE3). Genomic regions are described in Table S7. BL21 (DE3) cells were grown in LB media and induced with 0.25 mM IPTG; protein overexpression was carried out at 37°C for 1 h. Purification was achieved by Ni2+ affinity chromatography using the N-terminal His6 tag followed by tag cleavage using recombinant 3C protease.
Reporter plasmids pG2277 (flhB), pG3487 (aaeB), pG2239 (flgG) and pG2262 (fliL) were constructed by cloning the putative promoter region and ribosome binding site (∼500-bp region upstream of the start codon) of XC_2277, XC_3487, XC_2239 and XC_2262 respectively into the broad-host-range cloning vector pLAFRJ, which harbors the coding region (without promoter and RBS) of β-glucuronidase (gusA) gene in its MCS (multiple cloning site). The putative promoter region and RBS of XC_2277, XC_3487, XC_2239 and XC_2262 were amplified from the chromosomal DNA of Xcc8004 using the primer pairs detailed in Table S7. The amplified DNA fragment, confirmed by sequencing, was inserted 9 bp upstream of the promoterless gusA ATG start codon in the vector pLAFRJ to create the recombinant plasmid. The recombinant plasmid obtained was further confirmed by restriction analysis and PCR. All reporter plasmids were introduced into the Xcc strains of interest through conjugation as previously described [25].
A volume of 50 ml of an Xcc culture with an OD600 of 1 was harvested and suspended in 1 ml 10 mM Tris·HCl (pH 7.5), 50 mM NaCl buffer containing EDTA-free complete protease inhibitor (Roche). Cells were mixed with 0.1-mm glass beads and lysed in a Fast-Prep machine twice for 45 seconds. Samples were centrifuged for 5 min at 17,000× g and subsequently for 1 h at 100,000× g to obtain cytoplasmic protein extracts. A volume of 50 ml streptavidin dynabeads (Invitrogen) coupled with 2.4 µM biotinylated cyclic-di-GMP (BioLog) were incubated with 1.2 mg cytoplasmic proteins in 1.5 mL 10% (vol/vol) glycerol, 1 mM MgCl2, 5 mM Tris (pH 7.5), 230 mM NaCl, 0.5 mM DTT, and 4 mM EDTA containing 50 µg/ml BSA for 30 min at room temperature. Samples were washed four times with the same buffer lacking BSA and suspended in 50 µl protein sample buffer. Samples were boiled for 5 min, beads removed, and 18 µl run on 12% (wt/vol) SDS/PAGE gels. Protein identification was carried out by in-gel proteolysis and the resultant peptides were analysed by LC-MS/MS on the LTQ Orbitrap Classic (FingersPrints, UK).
The dissociation constant (Kd) between wild-type XC_3703 (or selected homolog) with cyclic di-GMP, cyclic di-AMP, ATP, cyclic GMP or cyclic AMP was measured by using an VP-ITC MicroCalorimeter. Titrations were carried out at 25°C in an assay buffer containing 20 mM Tris-Cl (pH 8.0), 80 mM NaCl. Samples of XC_3703 for ITC measurements were dialyzed extensively against the assay buffer overnight. The concentrations of protein in the cell ranged between 10 to 300 µM and that of the nucleotide in the syringe was in the range of 1–4 mM. A volume of 8 µl of nucleotide was injected into the cell with a time lag of 180 s between each injection for a total of 30 times. ITC data were analyzed by integrating heat exchange amount after subtracting background dilution heat from the apparent values. Data fitting was based on a one-site binding model using the using the MicroCal ORIGIN version 7.0 software.
Three independent cultures of each selected Xanthomonas strain were sub-cultured and grown to logarithmic phase (0.7–0.8 OD600) at 30°C in NYGB broth without selection. 800 µl of RNA protect (Qiagen) was added to 400 µl culture and incubated at room temperature for 5 min. Cell suspensions were centrifuged, the supernatant was discarded, and pellets were stored at −80°C. After thawing, 100 µl TE-lysozyme (400 µg/ml) was added and samples were incubated at room temperature. Total RNA was isolated using the RNeasy Mini Kit (Qiagen) whereby cells were homogenised utilising a 20-gauge needle and syringe. Samples were treated with DNase (Ambion) according to manufacturer's instructions and the removal of DNA contamination was confirmed by PCR.
RNA quality was assessed on a Bioanalyser PicoChip (Agilent) and RNA quantity was measured using the RNA assay on QuBit fluorometer (Life Technologies). Ribosomal RNA was depleted with Ribo-Zero rRNA Removal Kits for Gram-Negative Bacteria (Epicentre). The percentage of rRNA contamination was checked on a Bioanalyser PicoChip (Agilent). The rRNA-depleted sample was processed using the Illumina TruSeq RNA v2 sample preparation kit. In brief, the sample was chemically fragmented to ∼200 nt in length and the cleaved RNA fragments were primed with random hexamers into first strand cDNA using reverse transcriptase and random primers. The RNA template was removed and a replacement strand was synthesised to generate ds cDNA. The ds cDNA was end repaired to remove the overhangs from the fragmentation into blunt ends. A single ‘A’ nucleotide was added to the 3′ ends on the blunt fragments, which is complementary to a ‘T’ nucleotide on the 3′ end of the Illumina adapters. At this stage, adapters containing 6 nt barcodes were used for different samples to allow the pooling of multiple samples together. The resulted barcoded samples were enriched by 10 cycles of PCR to amplify the amount of DNA in the library. The final cDNA libraries were sequenced on an Illumina HiSeq2000 as per manufacturer's instructions. The RNA-Seq raw data files are accessible through XanthomonasGbrowse: http://browser.tgac.bbsrc.ac.uk/cgi-bin/gb2/gbrowse/Xanthomonas_8004/.
Cluster generation was performed using the Illumina cBot and the cDNA fragments were sequenced on the Illumina HiSeq2000 following a standard protocol. The fluorescent images were processed to sequences using the Real-Time Analysis (RTA) software v1.8 (Illumina). Raw sequence data were demultiplexed by assigning the sequenced reads of each sample with their corresponding indexed reads. The reads for each sample were then processed through the primary analysis pipeline to generate FASTQ files.
Quantitative RT-PCRs were used to validate RNA-Seq data. Reverse transcription PCR was achieved using a cDNA synthesis kit (Promega) according to the manufacturer's instructions. Specific RT-PCR primers were used to amplify central fragments of approximately 200 bp in length from different genes. For qRT-PCRs, quantification of gene expression and melting curve analysis were completed using a LightCycler (Roche) and Platinum SYBR Green qPCR Supermix-UGD (Invitrogen) was used according to manufacturer's instructions. The constitutively expressed housing keeping gene, 16S rRNA was used as a reference to standardize all samples and replicates.
The methods used for construction and analysis of the Xcc genomic DNA prey library in the plasmid vector pOAD were as described previously [17]. Xcc DNA sequence coding for the XC_3703 used as bait was amplified by PCR using Xcc genomic DNA as template and primers designed based on the Xcc genome sequence. These primers were designed to contain unique restriction sites to facilitate cloning into the pOBD vector downstream of the Gal4 DNA-binding domain. After transformation into E. coli DH5α cells, individual colonies were picked for plasmid isolation and confirmation by DNA sequencing. The bait construct was transformed into Saccharomyces cerevisiae, and this strain was used to screen the Xcc genomic DNA prey library. Approximately 2×108 yeast transformants (carrying both and prey vectors) were plated on selective media that lacked adenine, histidine, tryptophan, and leucine. Fifteen hundred transformants displaying prototrophy were transferred to plates with adenine and histidine but lacking tryptophan and leucine. Preys were identified by isolation of plasmids and sequencing.
For measurement of interactions of the segments of XC_2801 with XC_3703, doubly transformed yeast strains were grown in YPD medium (10 g liter−1 Bacto-yeast extract (Difco), 20 g liter−1 Bactopeptone (Difco), and 2 g liter−1 D-glucose) were performed as described [17].
In vitro binding assays were performed as described previously [34]. Briefly, ≈20 µg of purified recombinant proteins or total lysates of E. coli cultures expressing recombinant proteins were resolved by SDS/PAGE (12% acrylamide) and transferred to nitro- cellulose membranes. The membranes were blocked with TBS-TTmilk [140 mM NaCl, 20 mM TrisHCl (pH 7.4), 0.1% Triton X-100, 0.1% Tween-20, 5% powdered milk] for 1 h and then probed with anti-His6 antiserum as described previously [35].
The DNA probes used for EMSA were prepared by PCR amplification of the desired upstream region of flhB, using oligonucleotides as the primers (primers and amplified regions are described in Table S7). The purified PCR products were 3′-end-labelled with digoxigenin following the manufacturer's instruction (Roche). The EMSA was carried out using the DIG Gel Shift Kit 2nd Generation (Roche) as recommended by the manufacturer with some modifications. Ten fmoles of the DIG-labelled fragment and a range of 0–1000 nM of XC_2801 protein were added to the binding reaction. The mixture was allowed to proceed at room temperature for 45 min. The samples were separated by electrophoresis on 6% native polyacrylamide gels and transferred to Hybond-N blotting membrane (Amersham). Protein–DNA complexes were visualized by NBT/BCIP according to the manufacturer's instructions (Roche).
The virulence of Xcc to Chinese Radish was estimated after bacteria were introduced into the leaves by leaf clipping and spraying as previously detailed [10], [31], [36]. For leaf clipping assay, bacteria grown overnight in NYGB medium were washed and re-suspended in water to an OD at 600 nm of 0.001. For leaf clipping the last completely expanded leaf was cut with scissors dipped in the bacterial suspensions. Thirty leaves were inoculated for each strain tested. Lesion length was measured 14 days after inoculation. Each strain was tested in at least four separate experiments. For spraying assays, five-week old seedlings with fully expanded leaves were used. Bacteria were grown overnight in NYGB medium, centrifuged and re-suspended in water to a cell density OD of 0.01 at 600 nm for spraying. A volume of 50 ml of the bacterial suspension was inoculated onto the leaves of 25 plants (approximately 100 leaves) by spraying. Three replicates of each independent experiment were carried out. Ten days after inoculation the relative virulence was determined as the percentage of the total inoculated leaves that showed the typical black rot disease symptoms at the leaf margin. The experiment was repeated three times.
Biofilm was assessed by attachment to glass and was determined by crystal violet staining. Log-phase-grown bacteria were diluted to OD600 nm = 0.02 in L media broth and 5 ml was incubated at 30°C for 24 h in 14-ml glass tubes. After gently pouring off the media, bacterial pellicles were washed twice with water and were then stained with 0.1% crystal violet. Tubes were washed and rinsed with water until all unbound dye was removed [25], [31]. Bound crystal violet was eluted in ethanol and measured at OD595. Three independent assays were carried out for each strain.
Animal experiments were conducted in compliance with a protocol (# 16-051-11) approved by The Animal Experimentation Ethics Committee (AEEC) of University College Cork or Cork University Hospital. The Minister for Health accredits the University College Cork for conducting animal experiments. AEEC is guided by legislative requirements, in particular the Cruelty to Animals Act as amended and supplemented by the European Communities.
Mouse infection was performed using a variation on the mouse model described by [35], [37]. Briefly, P. aeruginosa or S. maltophilia strains were grown in Luria broth at 37°C overnight with shaking, after which bacteria were collected by centrifugation and resuspended in PBS. The exact number of bacteria was determined by plating serial dilutions of each inoculum on Luria broth agar plates. Female C57BL/6 mice (approximately 8-weeks old) were anesthetized and infected by the intratracheal route with 20 µl of culture of wild-type strains or mutants at a final inoculum of 5×106 colony forming unit (CFU) per mouse. Mice were killed at 1, 3 and 5 days post-infection by intraperitoneal injection of 0.3 ml of 30% pentobarbital. Lungs and spleens were harvested aseptically and homogenized in sterile PBS. A 10-fold serial dilution of lung homogenates was plated on Pseudomonas isolation agar (for P. aeruginosa strains) or NGBA media (for S. maltophilia strains). The results (means ± standard deviation) are expressed as CFU/organ.
P. aeruginosa or S. maltophilia strains were grown on bronchial cells using a co-culture model system as previously described [37], [38], [39]. Briefly, bronchial epithelial monolayers were co-cultured with 100 bacteria per cell for 2 hours. The bronchial epithelial cells were then washed and bacteria from cell lysates were plated on LB agar. Colonies were counted after 48 hours.
|
10.1371/journal.ppat.1005644 | Neutrophil Attack Triggers Extracellular Trap-Dependent Candida Cell Wall Remodeling and Altered Immune Recognition | Pathogens hide immunogenic epitopes from the host to evade immunity, persist and cause infection. The opportunistic human fungal pathogen Candida albicans, which can cause fatal disease in immunocompromised patient populations, offers a good example as it masks the inflammatory epitope β-glucan in its cell wall from host recognition. It has been demonstrated previously that β-glucan becomes exposed during infection in vivo but the mechanism behind this exposure was unknown. Here, we show that this unmasking involves neutrophil extracellular trap (NET) mediated attack, which triggers changes in fungal cell wall architecture that enhance immune recognition by the Dectin-1 β-glucan receptor in vitro. Furthermore, using a mouse model of disseminated candidiasis, we demonstrate the requirement for neutrophils in triggering these fungal cell wall changes in vivo. Importantly, we found that fungal epitope unmasking requires an active fungal response in addition to the stimulus provided by neutrophil attack. NET-mediated damage initiates fungal MAP kinase-driven responses, particularly by Hog1, that dynamically relocalize cell wall remodeling machinery including Chs3, Phr1 and Sur7. Neutrophil-initiated cell wall disruptions augment some macrophage cytokine responses to attacked fungi. This work provides insight into host-pathogen interactions during disseminated candidiasis, including valuable information about how the C. albicans cell wall responds to the biotic stress of immune attack. Our results highlight the important but underappreciated concept that pattern recognition during infection is dynamic and depends on the host-pathogen dialog.
| Opportunistic fungal infections, including those caused by C. albicans, have emerged as a significant global health burden and the disseminated form of these infections still have unacceptably high mortality rates despite modern antifungal treatments. The fungal cell wall controls its interaction with the host environment and immune recognition, although cell wall dynamics during infection are poorly understood. C. albicans organizes its cell wall to mask the inflammatory β-glucan as a form of immune evasion and it is known that during infection this β-glucan becomes exposed. Here, we investigated how β-glucan becomes exposed and discovered a dynamic interaction where host NETs provoke an active fungal response that disrupts cell wall architecture and unmasks β-glucan. We revealed an unexpected level of local fungal cell wall dynamics in response to immune mediated stress, suggesting this may represent a model that can be leveraged to identify novel drug targets. Our results highlight the understudied concept that the cell wall is a dynamic landscape during infection and can be influenced by the host.
| Innate immune recognition of pathogen-specific patterns plays a crucial role in initial infection control and activation of appropriate adaptive immune responses [1, 2]. Recognition through Toll-like, C-type lectin, Nod-like and Rig-I-like receptors elicits production of autocrine, paracrine and endocrine immunity. This includes activities as varied as deployment of neutrophil extracellular traps to directly attack pathogens and production of proinflammatory cytokines that recruit, activate and polarize additional innate and adaptive immune cells.
Pattern recognition receptors have evolved over millions of generations, and pathogens have concurrently developed creative ways to avoid these receptors by hiding specific epitopes. Epitope masking is practiced by many pathogens including bacteria, viruses, fungi, protozoans and helminths [3–9]. Work from a number of groups, including ours, has described how fungal cell wall architecture limits recognition of the β-glucan sugar by immune receptors that include Dectin-1, a C-type lectin crucial for resistance to fungal infections [5, 6, 10]. This epitope masking can be observed in Candida albicans, an opportunistic human pathogen which can cause both superficial mucosal and life threatening disseminated disease, particularly in immune compromised patients. However, C. albicans β-glucan epitope availability increases dramatically in vivo during a phase of neutrophilic influx in experimental murine candidemia [11, 12]. Although the dynamics of immune recognition during infection have implications for the trajectory of the immune response, the fungal and host mechanisms that lead to eventual β-glucan masking in vivo are unknown.
It is possible that the host, the fungus or both contribute to these changes in immune recognition during infection. On the fungal side, the cell wall integrity (CWI) pathway is critical in maintaining this compartment in response to abiotic stresses, but we still don’t understand how it functions in the context of immune attack in the challenging host environment [13]. We have previously described how a highly interconnected cell wall remodeling network creates and maintains the cell wall architecture that masks β-glucan from Dectin-1 under steady-state conditions, and this network may also act in vivo [7]. On the host side, cell-mediated immune attack by neutrophils can kill or incapacitate pathogens using reactive oxygen and nitrogen species, antimicrobial peptides, proteases, glycosidases, and extracellular traps (ETs) [14, 15]. Proteases and glycosidases could act on the outer mannan layer to directly expose underlying β-glucan, or phagocyte attack could indirectly trigger active fungal cell wall remodeling that unmasks underlying epitopes.
Changes in C. albicans cell wall β-glucan exposure due to early host-pathogen interaction during infection may sufficiently alter availability of cell wall epitopes to affect subsequent immune responses. However, the complexity of in vivo systems has limited our understanding of whether immune attack regulates subsequent immune cytokine elicitation. Here, we use a combination of in vitro and in vivo tools to show that neutrophils counter β-glucan masking by creating NETs that are required to trigger fungi to actively remodel local cell wall architecture. These disruptions of cell wall epitope masking alter recognition of the fungi and could enhance subsequent secondary immune responses.
Changes to the cell wall during infection alter C. albicans recognition by pattern recognition receptors, but the mechanisms driving these changes are unknown [12, 16]. Host defense against invasive candidiasis relies critically on neutrophils, evidenced by the increased susceptibility of neutropenic patients to candidemia [17]. We reasoned that they may disrupt the fungal cell wall and mediate β-glucan unmasking because neutrophils can damage the C. albicans cell wall and are present in high numbers during infection when β-glucan unmasking appears [11, 12, 18]. To determine the spatiotemporal dynamics of neutrophilic damage, we labeled biotinylated fungi with streptavidin-Alexa 647 and incubated with neutrophils. Time-lapse imaging shows that streptavidin fluorescence is lost rapidly at sites of neutrophil attack (Fig 1A–1D, S1 and S2 Movies). Controls demonstrate that there is also a loss of labeled protein (S1 Fig). Fluorescence of an Hwp1-GFP fusion protein, present in the hyphal cell wall, is also rapidly reduced upon neutrophil attack (Fig 1E–1I, S3 and S4 Movies). Overall, this suggests that neutrophils rapidly damage cell wall protein at sites of attack, in agreement with and extending previous work [18].
To determine if neutrophil attack disrupts or triggers disruption of other aspects of cell wall architecture to alter immune recognition, we stained attacked filaments with soluble Dectin-1-Fc (sDectin-1-Fc) to assess β-glucan availability and Calcofluor white (CFW) to stain chitin. Because CFW has been shown to be a stain in live yeast cells that is selective for new chitin fibers [19], we refer to sites in the lateral wall with increased CFW staining as “Sites of chitin deposition.” This staining of attacked filaments revealed areas of the lateral cell wall with β-glucan unmasking and increased chitin deposition at sites with cell wall protein loss (Fig 2). These overlapping sites of cell wall disruption occur uniquely in the neutrophil challenged samples but not in the absence of neutrophils, demonstrating that they are a direct or indirect result of immune activity. Taken together, these results show that neutrophil attack can result in and may be an important trigger for the disruption of C. albicans’ cell wall architecture and β-glucan unmasking in vitro.
We have previously shown β-glucan unmasking occurs during infection and our in vitro data suggests that neutrophils can mediate this exposure [12]. To test if neutrophils are required for these fungal cell wall changes in vivo, we examined C. albicans epitope exposure in neutropenic mice at day 5 post-infection, when there is normally β-glucan unmasking. To interrogate the native state of the C. albicans cell surface we used the ex vivo fluorescence method, which involves no fixation or permeabilization [12]. There is a significant reduction in β-glucan unmasking in neutropenic mice, demonstrating that neutrophils are critical for β-glucan unmasking in vivo (Fig 3A–3D). Similar results are seen in a second model of neutropenia, and as expected both methods of inducing neutropenia also led to increased susceptibility to infection (S2 Fig). Levels of β-glucan unmasking were similar when detected via either anti-β-glucan antibody or sDectin-1-Fc staining, demonstrating that this is not an artifact of a specific probe (S2 Fig). Further, chitin staining revealed that fungi from control mice have significantly stronger chitin deposition than neutropenic mice, suggesting that neutrophil attack is also important for increased chitin deposition in vivo (Fig 3A–3C and 3E). Fungi from neutropenic mice have slightly increased β-glucan unmasking and chitin levels as compared to in vitro RPMI-grown control cells, suggesting that growth in the host or possibly attack by other immune cells may also yield minor but significant cell wall changes even without neutrophil attack. Taken together, these results demonstrate that neutrophils are critical drivers of β-glucan unmasking and increased chitin deposition during disseminated infection in vivo.
It was not previously known that neutrophils alter innate pattern recognition of fungi in vivo, so we sought to characterize the mechanisms required to alter epitope unmasking. NET production, in which neutrophils create traps out of DNA and numerous antimicrobial factors, is a means of neutrophil attack against C. albicans and other fungi in vivo and in vitro [20–24]. Despite the poor NET production of mouse neutrophils relative to human neutrophils, we find strong evidence of NET formation in vitro. These NETs stain positive with the membrane impermeant Sytox green DNA dye, anti-citrullinated histone antibody, and anti-myeloperoxidase (MPO) antibody (Fig 4A and 4B). Furthermore, we observed that neutrophils could rapidly create ETs on C. albicans hyphae (S5 and S6 Movies, S3 Fig). Strikingly, treatment with DNase I to degrade extracellular DNA and prevent the establishment of NETs blocks both chitin deposition and β-glucan unmasking, functionally implicating NETs in driving this interaction (Fig 4C). In further support of NET-triggered changes during attack, inhibition of myeloperoxidase (MPO) with 4-aminobenzoic acid hydrazide (ABAH) prevents neutrophil attack from resulting in chitin deposition or β-glucan unmasking (Fig 4D). Taken together, these data provide strong evidence that NETs provide the initial stimulus that results in fungal cell wall changes including β-glucan exposure.
Neutrophil proteases are thought to be an important component of NET formation in some contexts and neutrophil elastase trafficking is regulated during NETosis against C. albicans [20, 25, 26]. However, neutrophils from mice deficient in the dipeptidyl peptidase (DPPI), which is required for the activation of the three major neutrophil proteases: elastase, cathepsin G and proteinase 3 [27], show no defect in their ability to cause β-glucan unmasking, chitin deposition or streptavidin loss (S4 Fig).Thus, these three proteases do not appear to play an important role in the downstream cell wall remodeling triggered by neutrophil attack of C. albicans in this system.
Phagocyte NADPH oxidase is important in defense against candidemia and plays an important role in NET formation under many conditions, including in response to fungi [20, 28, 29]. We find that disruption of NADPH oxidase function, either by using neutrophils from gp91phox-/- mice lacking a key component of the NADPH oxidase or using chemical inhibitors, decreases cell wall damage and prevents immune attack from resulting in β-glucan unmasking or chitin deposition in vitro (S5 Fig, Fig 4C, 4E and 4F). This was not due to a complete lack of neutrophil attack on the hyphae (S5 Fig). Similarly, fungi from the kidneys of gp91phox-/- mice had significantly less chitin staining and β-glucan unmasking, demonstrating the importance of phagocyte oxidase for causing cell wall remodeling in vivo (Fig 4G–4K). As expected, WT mice were able to control fungal growth while gp91phox-/- mice were unable to do so (S5 Fig). Importantly, immune cells including many neutrophils were found surrounding hyphae in gp91phox-/- mice (S5 Fig), suggesting that the loss of β-glucan unmasking was not due to lack of immune cell recruitment. Taken together, these data suggest that NETs also trigger C. albicans cell wall remodeling and enhanced Dectin-1 recognition in vivo and reveal a new way that immune cells counter fungal immune evasion.
Although NET attack could directly cause these cell wall changes, NET damage could also initiate conserved fungal stress signaling pathways that are known to both respond to cell wall insults and mask β-glucan in steady-state [7, 30]. We reasoned that if neutrophil-triggered changes are passive from the fungal perspective, they should occur rapidly and simultaneously, and should also occur in inactivated fungi. Surprisingly, although initial cell wall protein damage occurs within seconds (Fig 1), chitin deposition is not apparent until 30 minutes post-challenge, and enhanced Dectin-1 recognition lags even further (Fig 5A and 5B). A similiar succession of events also occurs in experiments conducted in imaging dishes (S6 Fig).
The nature of these sequential changes over hours suggests that unmasking results from an active fungal response rather than by direct immune mediated damage. In support of this hypothesis, UV-inactivated fungi lose streptavidin at attack sites but fail to develop sites of chitin deposition or β-glucan unmasking (Fig 5). UV inactivation is a minimally invasive means of killing fungi, so these results indicate that only initial cell wall damage is a direct result of immune attack (Fig 5C and 5D, S7 and S8 Movies). Thus, it appears that immune attack triggers β-glucan unmasking and chitin deposition only indirectly, by promoting active fungal signaling in response to immune mediated attack.
The fungal cell wall integrity (CWI) signaling pathway plays a key role in stress responses and in maintaining the normal cell wall architecture that masks β-glucan [7, 30]. However, it is not known how C. albicans responds to immune-mediated cell wall damage, so we sought to identify which signaling pathway(s) drives localized cell wall remodeling. Targeted screening of mutants deficient in individual CWI signaling components for defects in responding to neutrophil attack revealed that HOG1 is important for this process while CEK1 and MKC1 are not required (Fig 6A and 6B, S7 Fig). The HOG1-deficient strain has a significantly decreased ability to respond to neutrophil attack with both chitin deposition and β-glucan unmasking (Fig 6A and 6B, S7 Fig). This defect is not due to differences in fungal cell viability or attack rates between strains (S7 Fig, S8 Fig). Interestingly, C. albicans deficient in CAP1, which is involved in responding to some types of oxidative stress [31, 32], is not required for this chitin deposition response (S8 Fig). This primary dependence on Hog1p suggests that chitin deposition and enhanced Dectin-1 binding result from post-transcriptional activities, as Hog1p plays a limited role in regulating stress-mediated transcription [13]. It is important to note that, while deficient in its responses, the residual responses of the hog1 strain suggest other pathways are also involved.
Increased chitin levels can rescue C. albicans from stress, including antifungal treatment [33]. To implicate a specific synthase in enhanced chitin deposition at attack sites, we examined post-attack chitin deposition in mutants in either the major chitin synthase, CHS3, or both stress-activated synthases CHS2 and CHS8 [34]. Both WT and chs2∆/∆ chs8∆/∆ strains have dramatic increases in areas with localized chitin deposition and β-glucan unmasking following interaction with neutrophils when compared to their no neutrophil controls (Fig 6C and 6D). However, while the abnormal morphology of the chs3∆/∆ deletion mutant results in a high baseline number of areas with increased chitin deposition and to a lesser extent β-glucan exposure, there is no significant increase in cell wall changes after neutrophil attack. This defect was not due to differences in cell viability, number of attacked sites, or lack of cell wall damage (S8 Fig). Analysis of the intensity of chitin staining is consistent with a major role for Chs3p in driving neutrophil-triggered chitin deposition (S8 Fig). The chs3∆/∆ mutant was not completely deficient in responding to attack with chitin deposition, however, suggesting that other chitin synthases may play a very limited role in this process. In support of the idea that Chs3p is the major synthase in the response to neutrophil damage, timelapse of a Chs3-YFP fusion strain demonstrates recruitment of Chs3p-YFP to most sites of increased chitin deposition (Fig 6E, S9 Movie).
Cell wall remodeling is also crucial for response to stress, but we know little about the spatiotemporal dynamics of these responses, especially in the context of immune attack. We therefore characterized the post-attack movement of cell wall remodeling and biogenesis proteins, including Sur7p and Phr1p. Sur7p is deposited in new cell wall and marks eisosomes, and Phr1p is a β(1,3)-glucan remodeling enzyme crucial to cell wall integrity [35, 36]. Both Phr1p and Sur7p are recruited to sites of neutrophil attack. Sur7p was recruited early and coincident with chitin deposition while Phr1p accumulated at later times (Fig 6F–6I, S10 and S11 Movies). Overall, these results help elucidate important components of the fungal response to immune cell attack, suggesting Hog1p is important for the initial signaling response which leads to chitin deposition mainly through Chs3p localization and the later cell wall remodeling possibly involving Phr1p and Sur7p.
Neutrophil-triggered enhancement of Dectin-1 binding may result in an altered secondary immune response to C. albicans or have no impact due to redundant recognition modalities. To assay secondary immune responses, we challenged C. albicans hyphae with neutrophils, then lysed the neutrophils. We treated the samples with DNase1 to reduce activation of macrophages by neutrophil debris before UV-inactivating the fungi and adding them to murine macrophages. A mixture of unattacked fungi with neutrophil lysate served as a control for activation by remaining neutrophil debris in the context of fungal stimulation [37]. ELISA assays revealed that attacked fungi induced higher production of the proinflammatory cytokine IL-6 when compared to any of several controls (Fig 7). Interestingly, this increased cytokine production was not completely dampened by Dectin-1 inhibition suggesting other receptors may also be involved in this response. These results indicate that neutrophil attack and the resulting cell wall changes, including β-glucan unmasking, can lead to enhanced recognition and responses by other immune cells.
Recognition of pathogens based on conserved molecular patterns is a cornerstone of innate immunity but it is a dynamic battlefield between host and pathogen. The host’s task is complicated because pathogens conceal essential molecular patterns from detection, thereby denying the host the knowledge it needs to initiate a response. Here, we build on previous work to show that although C. albicans masks β-glucan during infection, host immune cells can damage the invader and trigger the disruption of cell wall architecture in a manner that could enhance innate immune recognition, including the unmasking of β-glucan. Our findings echo work in diverse animal and plant hosts that suggest pathogen recognition and responses are dynamic and can impact immunity during infection [16, 38–40]. Other types of “unmasking” might take place in a number of infections, as masking of epitopes has been demonstrated for bacteria [3], viruses [4], fungi [5–7], protozoans [8] and helminths [9]. While the microbial cell wall is an adaptable landscape that is capable of responding to numerous stimuli, we still have a limited understanding of how cell wall architecture changes in vivo and how host-pathogen interactions influence PAMP availability during infection.
We describe here how the host subverts a fungal evasion strategy, unmasking C. albicans to reveal fungal-specific epitopes like β-glucan. Surprisingly, this is a two-step process where NET-dependent neutrophil attack results in β-glucan unmasking via an active fungal process. The fungal response to localized cell wall stress includes a cascade of events, with chitin deposition and enhanced β-glucan exposure mediated by Hog1p signaling and the major chitin synthase Chs3p. Remodeling of cell wall architecture enhances recognition and could enhance responses by the host, but is also likely to protect the fungus by strengthening the cell wall, as is the case for plant cell walls that remodel upon fungal attack [41, 42]. These mechanisms of cell wall architecture control during fungal infection are likely relevant for other fungi that hide immunogenic β-glucan from the host and thereby limit immune responses [5, 6, 43]. Given the importance of Dectin-1 signaling in anti-fungal defense, and the fact that NETs are deployed against other fungi, it seems likely that immune mediated unmasking may take place during other fungal infections [21, 44, 45].
The ability of a host to recognize and respond to microbe-specific components is a key determinant to mounting an effective defense. Neutrophil unmasking of hyphal β-glucan, which is structurally distinct and elicits greater inflammatory cytokine responses than yeast β-glucan, could help the host discriminate between commensalism and opportunistic disease [46], especially since C. albicans hyphae are typically associated with invasion and greater recognition could enhance the “danger” response to invading hyphae [47]. Unmasked epitopes could also assist in “trained immunity” for other innate immune cells like monocytes, which has been shown to depend on fungal β-glucan and Dectin-1 signaling for protection against C. albicans [48]. Our results demonstrate that cell wall changes following neutrophil attack do increase recognition of β-glucan and also specifically elicit more IL-6, but not TNF-α, from macrophages when compared to controls. Elevated IL-6 production, in particular, may be important as it participates in the induction of Th17 responses which are important for antifungal immunity [49–51]. Interestingly, some of the increased macrophage response was not blocked by neutralizing anti-Dectin-1 antibodies, suggesting that either other cell wall changes beyond β-glucan unmasking may play a role or the antibody is less effective in our system. In addition to enhanced Dectin-1 recognition, it is likely that post-attack disruption of cell wall architecture results in alterations to other fungal cell wall epitopes. Our preliminary data using wheat germ agglutinin as a probe suggests there is also increased availability of chitin at attack sites. Because fungal chitin recognition can modulate inflammation, altered chitin recognition may contribute to secondary immune responses [52, 53].
The impact these epitope changes might have on the outcome of infection remains to be explored. While elevated IL-6 could contribute to the development of protective Th17 response after recognition of exposed fungal epitopes, hyperinflammation of the IL-17 axis and excessive neutrophilic influx are also risk factors, and recent work suggests that myeloid-derived suppressor cell-mediated immunomodulation is protective during this phase around 4–7 days post-infection [54–58], [59]. Whether for protection or pathogenesis, the potential of immune attack to alter subsequent immune response suggests that immune dynamics may play an important regulatory role.
Extracellular traps function in pathogen containment and killing, and we find that they can also influence pathogen epitope exposure. The specific requirements for neutrophils and phagocyte oxidase to damage C. albicans and initiate unmasking in vivo fits with previous in vitro findings that ETs made by macrophages don’t damage Candida [60] and our own preliminary observations that macrophage attack doesn’t elicit the same changes to the C. albicans cell wall. NADPH oxidase and MPO are known players in NET formation [61], but NET formation has also been shown to occur in a reactive oxygen species independent manner so it is still unknown if or how these components contribute to the fungal damage that induces cell wall remodeling. While some NETs were still produced upon NADPH oxidase inhibition, our data demonstrates the requirement for functional NADPH oxidase, MPO and extracellular DNA in NETs for inducing fungal cell wall changes following immune attack. This suggests that the role of the NADPH oxidase and MPO is not primarily in NET creation but instead may contribute to decorating NETs with damaging components which provoke fungal integrity responses (with the role of the NETs themselves to hold these components in close proximity to the fungal cell wall). It is possible that MPO activity alone could provoke fungal cell wall changes but the inhibitory nature of DNAse1 treatment combined with the requirement of NADPH oxidase all point to NETs playing a key role. Further experiments will be required to fully understand the exact role NETs and their components are playing. The MPO requirement suggests this is not strictly due to the neutrophil respiratory burst causing localized hypoxia, which is an environmental condition previously associated with fungal cell wall changes [62]. Surprisingly, our experiments suggest that neutrophil proteases are not required for NET-dependent unmasking in our system, although they have been previously implicated in human and mouse NET formation [20, 25, 26]. We tested this by using neutrophils deficient in DPPI, in which the three major neutrophil proteases could not be processed properly [27]. Interestingly, it has also been seen that DPPI deficiency limits PMA and ROS-induced NET production [26]. As more research emerges, the requirement for different components in NET production has been found to be highly context and stimulus dependent, even for elements like the phagocyte NAPDH oxidase which were previously thought to be absolutely critical [29, 61]. It is therefore possible that we have identified a situation in which these proteases do not play a critical role in NET formation or that the defects which result are not severe enough to compromise their function in the unmasking process. Further work will be required to determine if this lack of a requirement is because other proteases can fill in during this situation or if no protease activity is required at all.
The early initiation of NET-dependent C. albicans cell wall remodeling, by 30 minutes post-challenge, suggests a deployment of NETs within thirty minutes—much earlier than previous reports of NETosis attack of C. albicans [20, 24, 63]. Rapid NET formation has been previously described to occur in a subset of neutrophils in response to certain stimuli [25]. It is notable that the relatively early NET formation we describe here as triggering fungal cell wall remodeling is dependent on NADPH oxidase activity, in contrast to “live” NETosis that has been previously described as NADPH oxidase-independent [64]. The mechanism of NET release here is unknown and remains to be analyzed. A better understanding of neutrophil and NET function during infection could have clinical benefits, as defects in either result in increased susceptibility to many infections, including those caused by Candida [17, 65].
The mechanism whereby neutrophil attack reveals fungal epitopes is unexpected, as cell wall changes are not a direct result of immune attack but rather are initiated by signaling in the fungus. The importance of the Hog1p MAPK in response to neutrophil attack is consistent with its established roles in interactions with phagocytes and host immunity both in vitro and in vivo [66, 67]. While Hog1p clearly plays an important role, the hog1Δ/Δ strain is not completely deficient in responding to neutrophil attack, suggesting that other pathways can play a minor compensatory role. The lack of a requirement for CEK1 and MKC1 is intriguing, as they both play important roles in cell wall homeostasis and stress adaptation [30, 68]. This suggests that the importance of each CWI MAPK is context-dependent and specific for given set of stress factor(s). Considering previous reports, our data offer the possibility that hog1Δ/Δ hypersensitivity to neutrophil-mediated killing may be due to a failure to deposit chitin and reinforce its cell wall, a process that rescues C. albicans from other stresses [33]. This role for Hog1p adds to its previously described activities in regulation of osmotic and oxidative stress [69].
The localized cell wall stress caused by neutrophil attack provides an advantageous situation to dynamically model how C. albicans hyphae mobilize their cell wall machinery in response to neutrophil attack in vivo. Genetic deletion mutants show that Chs3p is responsible for the majority of this localized lateral cell wall chitin deposition and is important for eventual β-glucan unmasking, while Chs2p and Chs8p are not required for these changes. Time-lapse microscopy also demonstrated accumulation of Chs3-YFP at attack sites with chitin deposition, suggesting a new role for Chs3p in responding to neutrophil-mediated stress. The requirement of chitin synthases may be context dependent, as recent reports show Chs2p and Chs8p are involved in maintaining cellular integrity during some forms of stress in vitro [70]. This data, combined with that showing the importance of Hog1p signaling in responding with chitin deposition, supports previous observations that Hog1p can regulate and activate chitin synthesis [71]. The recruitment of cell wall regulatory and remodeling enzymes Sur7p and Phr1p suggests a multistep process of remodeling post-attack. While Phr1p is known to be recruited to apical growth sites and septa and to respond to other stresses, this is the first time it has been found enriched in a localized section of lateral cell wall [36, 72]. Sur7p is important in regulating cell wall organization and integrity [35]. Early Sur7p enrichment at sites of chitin deposition and β-glucan unmasking suggests it plays an early role in cell wall reorganization at sites of neutrophil attack, in contrast with the later role for Phr1p. Beyond providing detailed insight into how C. albicans responds to neutrophil attack, this model of localized cell wall stress offers a powerful new method to image the multistep dynamics of stress-stimulated cell wall remodeling.
β-glucan recognition is relevant beyond mammalian immunity, as it has been demonstrated that both invertebrates and plants sense and respond to β-glucan, with important implications for antifungal immunity [73, 74]. Indeed, dynamic host-pathogen interactions revolving around fungal β-glucan masking and host recognition also occur during infections in plants, suggesting parallels with important agricultural fungal infections [39, 41, 75].
The game of pathogen camouflage and host-mediated unmasking has been played out over generations throughout the animal, fungal and plant kingdoms. The localized fungal cell wall remodeling we observe upon immune-mediated stress represents a novel model to probe basic mechanisms of cell wall dynamics and may identify novel therapeutic targets or strategies especially relevant to the in vivo infection environment.
C. albicans strains used in this study are listed in the Table 1. C. albicans was maintained on YPD 37°C. Single colonies were picked to 5 mL YPD liquid and grown at 30°C overnight on a rotator wheel. For hyphal cells, a defined number of yeast cells were transferred into RPMI and grown in 5 mL tubes at 30°C overnight on a rotator wheel. JC94-2 was constructed from the JC94 parent [36]. PHR1-GFP, along with 1kb upstream and 0.5 kb downstream regulatory sequence, was amplified from JC94 genomic DNA with primers PHR1-CIp20-FOR (5’-ATATTCGACTGAAAGCTTGATTACAAGTGGGATGCAAAA-3’), and PHR1-CIp20-REV (5’-TCGTCGGGCTCAAAGCTTCGTTGAAAAAGCATAAGAAGG-3’) and cloned into CIp20 [76] with HinDIII, after which it was sequence verified. Integration of Stu1-cut CIp20-PHR1-GFP at the RP10 locus was confirmed by PCR and three independent clones had similar phenotypes.
Neutrophils from 6–12 week old female C57BL/6J or gp91phox-/- (B6.129S-Cybbtm1Din/J (Jackson Laboratories) [85]) mice were purified from bone marrow using biotin anti-Ly6G antibody (eBioscience) and AutoMACS separation (Miltenyi). Bone marrow from sex- and age-matched C57BL/6-backcrossed DPPI -/- and control mice was extracted after overnight shipment on cold packs [86]. C. albicans hyphae at a concentration of 3x108 cells/mL were labeled with Biotin-XX-SSE (Molecular Probes; 0.01 μg/μL). Cells were then labeled with Alexa Fluor 647-conjugated Streptavidin (Jackson Immunoresearch; 36 μg/mL). 3 x107 hyphal cells were then incubated with or without 7 x 106 neutrophils in RPMI + 5% FBS for 2.5 hours. Neutrophils were lysed with 0.02% Triton X-100, though all samples also received this treatment. C. albicans hyphae were then stained with sDectin-1-Fc (17 μg/ml) followed by donkey anti-human IgG Cy3 or Alexa Fluor 488 antibody (Jackson Immunoresearch; 0.8 mg/ml) and Calcofluor White (Sigma Chemicals; 25 ng/mL). Purified sDectin-1-Fc was prepared from stably transfected HEK293T cells as previously described [87]. Cells were visualized by optical sectioning fluorescence microscopy using a Zeiss Axiovision Vivotome microscope (Carl Zeiss Microscopy, LLC). Fields of view were chosen randomly and an equal number of images were obtained for each sample. Maximum image projections were used to score the percentage of live cells with increased chitin deposition, β-glucan exposure and the overlap of both phenotypes. The tips of hyphae were excluded from scoring to prevent confusion by any cell wall changes occurring during hyphal growth. For some experiments, cells were also scored for the absence of streptavidin labeling. Cells from the “no neutrophils” samples had a baseline level of cells without streptavidin labeling of approximately 20% due to growth post-labeling, while samples with neutrophils had little to no additional growth over the course of the experiment and therefore had a low baseline level of cells without streptavidin labeling. Cell viability was confirmed based on the characteristic cytoplasmic EGFP, dTom or FarRed670 expression of live cells, or by propidium iodide (250 ng/mL) exclusion in non-transformed strains.
For post-challenge labeling experiments, hyphae were biotinylated, and incubated with neutrophils; after neutrophil lysis and staining, Alexa Fluor 647-conjugated Streptavidin was included along with the secondary antibody. For chemical inhibition, neutrophils were pre-incubated with either the vehicle DMSO, 10 μM DPI, 300 μM Apocynin, 100 μM ABAH or 500 μM ABAH for 10 minutes before addition to C. albicans and samples were treated for the entire 2.5 hours. For UV inactivation experiments, hyphal cells were UV inactivated as described [7]. To ensure that lack of damage was not due to altered neutrophil attack rates, staining in experiments with the hog1Δ/Δ, cap1Δ/Δ, cek1Δ/Δ, mkc1Δ/Δ, chs2Δ/Δ chs8Δ/Δ and chs3Δ/Δ mutant strains was performed without neutrophil lysis with Triton X-100 treatment. To examine NETs, sDectin-1-Fc and CFW staining procedures were carried out as described above, with Sytox Green (Molecular Probes; 156 nM) added along with the secondary antibody and CFW. We used Anti-Histone H3 citrulline R2+R8+R17 (abcam; 0.014 mg/mL) and donkey anti-rabbit IgG Cy3 (Jackson Immunoresearch; 0.0075 mg/mL) as well as Anti-MPO (R&D Systems; 0.1 mg/mL) with Donkey anti-goat Cy3 (Jackson Immunoresearch; 0.007 mg/mL). In experiments with DNase 1, the RPMI used for the incubation was supplemented with 100 mM CaCl2 and 100 mM MgCl2.
Streptavidin-labeled hyphae of the indicated strain, at a concentration of 6x106 cells, were added to a Delta T imaging dish (Bioptechs Inc) with 8x105 neutrophils in 1 mL of Phenol red-free RPMI + 5%FBS (Lonza). The Chs3-YFP strain was not labeled and imaged in 1 mL of PBS with 5% FBS and 5.5 mM glucose. Imaging dishes were then either incubated at 37°C in an incubator for the indicated amount of time or immediately imaged on the Zeiss Axiovision Vivotome microscope (Carl Zeiss Microscopy, LLC) or Nikon PerfectFocus microscope (Nikon) with a heated stage (Bioptechs, Inc) at 37°C. Chs3-YFP timelapses were instead taken on a Nikon Ti-E PFS live cell microscope (Nikon, Inc). For staining in dishes, the sDectin-1-Fc and CFW staining was done as described in the above section except 4.25x106 neutrophils were added, they were not lysed and the process was carried out in the dish.
Six week old female Balb/cJ, C57BL/6J or gp91phox-/- mice (Jackson Laboratories) were infected via tail vein. WT, isotype controls (Rat IgG2a for 1A8 from Bio X cell) and RB6-8C5 (Bio X cell) treated mice received 1x105 cfu, those receiving 1A8 received 2.5x104 cfu and gp91phox-/- mice received 500 cfu. Mice were treated with isotype, RB6-8C5 or 1A8 antibody (Bio X cell; 100 μg in 200 μL of PBS) via i.p. injection on day 2 post-infection. On day 5 post-infection mice were sacrificed via CO2 inhalation followed by cervical dislocation. Neutropenia was confirmed by Wright staining of blood obtained by cardiac puncture. Organs were harvested and homogenized as described [12]. For some experiments, kidneys were bisected with a razor and half was processed for histology. Homogenates were stained with sDectin-1-Fc (17 μg/ml) then donkey anti-human IgG Cy3 (0.8 mg/ml) and Calcofluor White (25 ng/mL). Alternatively, homogenates were stained with anti-β-glucan antibody (Biosupplies, Inc., Australia; 1.7 mg/mL) then with goat anti-mouse Cy3 antibody (Jackson Immunoresearch; 3.8 mg/mL). Cells were visualized by optical sectioning fluorescence microscopy using a Zeiss Axiovision Vivotome microscope (Carl Zeiss Microscopy, LLC). Maximum projection images were quantified using Cellprofiler (www.cellprofiler.org) as described [12].
RAW-blue macrophages (Invivogen) were maintained in DMEM + 10% FBS supplemented with sodium pyruvate, gentamicin and zeocin. For detection of cytokines by ELISA, unlabeled 3.0x107 C. albicans hyphae were incubated with or without 7x106 neutrophils overnight in RPMI+5%FBS. A neutrophil alone group was also included. The next morning RAW-blue macrophages were harvested. Macrophages were resuspended at 2.77x106 cells/mL and pre-incubated with Anti-mDectin-1 (Bio-Rad; 10 μg) or IgG2b isotype control (Invivogen; 10 μg) for 90 minutes. Candida-neutrophil mixtures were treated with 0.05% Triton X-100 solution for 5 minutes. As control for the impact of neutrophil debris in the context of fungal stimulation, one of the neutrophil alone samples was added to the C. albicans alone sample just before lysis. Samples were washed extensively and then incubated with 200 units of DNase1 for 1 hour. Samples were then washed extensively again and resuspended in 500 μL PBS. 25 μL of sample was added to the wells of a 96 well plate in duplicate before UV inactivation by treatment with 5 x 100,000 μJ/cm2. For a positive control, depleted zymosan (Invivogen) was added and for a negative control sterile water was added. RAW-blue cells, either Anti-mDectin-1 treated or not, were then added to the UV-inactivated fungal cells at 5x105 cells per well. Supernatants were harvested after 6 hours. ELISA was performed using Mouse TNF-α and IL-6 DuoSets (R&D Systems) according to manufacturer’s instructions and detected using Supersignal ELISA Femto Maximum Sensitivity Substrate (ThermoFisher Scientific) using a Biotek Synergy 2 plate reader (Biotek Instruments, Inc).
Statistics were performed as described in figure legends. For normally distributed data, Student’s t test or one or two way ANOVA analysis with Tukey’s post-test were used. For non-parametric data, Kruskal-Wallis with Dunn’s post-test was applied. ANOVA and Kruskal-Wallis were done using Prism software (Graphpad Software). p < 0.05 was considered significant.
All animal studies were carried out in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All animals were treated in a humane manner according to guidelines of the University of Maine IACUC as detailed in protocol number A2014-02-01. The UMaine IACUC/Ethics Committee approved this protocol. Animals were euthanized by carbon dioxide inhalation. Infected animals were monitored twice daily for signs of infection and morbid animals were euthanized.
|
10.1371/journal.pcbi.1000122 | Memory Switches in Chemical Reaction Space | Just as complex electronic circuits are built from simple Boolean gates, diverse biological functions, including signal transduction, differentiation, and stress response, frequently use biochemical switches as a functional module. A relatively small number of such switches have been described in the literature, and these exhibit considerable diversity in chemical topology. We asked if biochemical switches are indeed rare and if there are common chemical motifs and family relationships among such switches. We performed a systematic exploration of chemical reaction space by generating all possible stoichiometrically valid chemical configurations up to 3 molecules and 6 reactions and up to 4 molecules and 3 reactions. We used Monte Carlo sampling of parameter space for each such configuration to generate specific models and checked each model for switching properties. We found nearly 4,500 reaction topologies, or about 10% of our tested configurations, that demonstrate switching behavior. Commonly accepted topological features such as feedback were poor predictors of bistability, and we identified new reaction motifs that were likely to be found in switches. Furthermore, the discovered switches were related in that most of the larger configurations were derived from smaller ones by addition of one or more reactions. To explore even larger configurations, we developed two tools: the “bistabilizer,” which converts almost-bistable systems into bistable ones, and frequent motif mining, which helps rank untested configurations. Both of these tools increased the coverage of our library of bistable systems. Thus, our systematic exploration of chemical reaction space has produced a valuable resource for investigating the key signaling motif of bistability.
| How does a cell know what type of cell it is supposed to become? How do external chemical signals change the underlying “state” of the cell? How are response pathways triggered on the application of a stress? Such questions of differentiation, signal transduction, and stress response, while seemingly diverse, all pertain to the storage of state information, or “memory,” by biochemical switches. Just as a computer memory unit can store a bit of 0 or 1 through electrical signals, a biochemical switch can be in one of two states, where chemical signals are on or off. This lets the cell record the presence/absence of an environmental stimulus, the level of a signaling molecule, or the result of a cell fate decision. There are a small number of published ways by which a group of chemical reactions come together to realize a switch. We undertook an exhaustive computational exploration to see if chemical switches are indeed rare and found, surprisingly, that they are actually abundant, highly diverse, but related to one another. Our catalog of switches opens up new bioinformatics approaches to understanding cellular decision making and cellular memory.
| Most chemical reaction systems have a single steady state, but a few interesting cases are known to oscillate [1], form spatial patterns [2], or have multiple stable states [3],[4]. Aside from their intrinsic mathematical and chemical significance, systems with multiple stable states are of particular biological interest because they can retain a “memory” of past inputs and cellular decisions [3],[4]. Bistability is a particularly interesting case of multi-stability, as it leads to switch-like behavior. Chemical stimuli can trigger a state change from one stable state to another. The current state of the chemical system is therefore a “memory” of this earlier stimulus.
A few biochemical switches have been extensively analyzed, including complex enzyme mechanisms [5],[6], kinase feedback [7],[8], dual phosphorylation [9], the cell cycle [10], triggering of caspases [11], and synaptic memory switches [12]–[14]. Two observations emerge from this set of known switches. First, relatively few switches are known. A recent computational exploration yielded only about 2% bistable models among those tested [15]. Furthermore, no entries are annotated as bistable in either KEGG (331 pathways) or BIOCARTA (355 pathways). Somewhat at odds with this absence of bistable pathways in pathway databases, kinetic models of bistable pathways are more common. There are several bistable models in the signaling model databases DOQCS (10/69; [16]) and BioModels.net (12/147; [17]), coming to about 10% of recorded models. This may be an overrepresentation, due to modeling interest in bistability. In particular, there are several signaling models that explore bistability as a basis for synaptic memory [12]–[14].
A second observation about the known bistable switches is that they are quite different in their chemical topologies. While feedback loops are a recurring motif [3],[18], there are some cases where enzyme saturation appears to play a role [13], and others where the balance between competing reactions itself generates bistability [9].
While signaling models tend to result in rather complex reaction systems, a distinct approach to the study of chemical bistability is driven from theoretical analyses of enzyme kinetics and flux reaction systems [3],[5]. These studies show that very few reactions are needed to achieve bistability. This raises the interesting question of whether there are core sets of reactions, or motifs, that are embedded in all bistable chemical reaction systems, despite their diversity. A corollary is whether such a set of motifs may help to detect bistable sub-systems in complex biological signaling networks.
Necessary conditions for bistability, such as positive loops in the system Jacobian, have been well characterized [18]. Earlier work by Clarke [19] parametrically defines all steady states of a given reaction system, but does not yield specific solutions when concentrations and rate constants are given. Recent studies detect chemical switches by testing for correlates of bistability [3],[5] or by looking for properties that frequently co-occur with bistability and, optionally, engineering bistability by minor modifications to such networks [15],[20],[21]. We sought to identify bistable systems without placing any “top-down” requirement on the mechanistic details, and use this unbiased search to reconstruct relationships between the switches.
Here we systematically explore chemical reaction space to show that bistable chemical switches are remarkably common. We show that all small bistable systems are related, and that larger ones frequently share motifs that may be predictive of bistability.
In our first phase of analysis, we began with a basis set of 12 reactions (Figure 1A) and systematically tested all reaction configurations involving 2 molecules, 3 molecules from 1 to 6 reactions, and 4 molecules from 1 to 3 reactions. In our second phase of analysis, we sampled a subset of possible reaction configurations involving 3 molecules from 7 to 15 reactions, 4 molecules from 4 to 5 reactions, and 5 molecules from 1 to 4 reactions. The number of possible configurations rose rapidly with the number of molecules and reactions (Figure 1D), and it took longer to test each configuration for bistability, hence we sampled a small subset of configurations for the second phase of analysis. For each configuration we generated ∼100 models using Monte Carlo assignment of concentration and rate parameters (see Methods section) and tested each for bistability. The propensity of a configuration for bistability was defined as the fraction of tested models for that configuration that exhibited two or more stable steady states.
We observed a large number of bistable systems even with our very sparse sampling of reaction parameter space (Figure 1E). 3,562 of the fully sampled configurations (∼10%) had at least one bistable model, and 918 of the larger reaction systems (∼5%) did. This large percentage was surprising for two reasons. First, known bistable configurations from biology are rare, as discussed above. Second, our sampling of parameter space was very sparse, so we would be likely to detect bistable configurations only if they remained bistable in a substantial portion of parameter space. Most known chemical bistable switches exhibit bistability in a relatively narrow range of parameters rarely exceeding a factor of two [9],[22]. While a factor of two may be substantial from a biological viewpoint, we required a 30-fold range to detect bistability. This was because even small models have a large number of parameters. For instance, a 3 molecule, 3 reaction system has 7 parameters. In order to obtain bistability in 1% of tested models of this configuration, bistability would have to be present over approximately half the sampling range (Figure 1F) for each parameter: (0.5)7∼0.01. Our logarithmic sampling spanned 3 orders of magnitude, so half this range is about 30-fold for each parameter. A few configurations had a propensity of over 50% (Figure 1G). This suggests that bistability in these systems is very robust.
Admittedly, due to our sparse sampling of parameter space, there could be undetected bistables in the space of systems sampled here. While a single configuration is sufficient to prove that a network has the capability for exhibiting bistability, our analysis methods do not support an impossibility proof for bistability. The range in which a system exhibits bistability can depend intricately on how the phase space is structured in terms of the system parameters such as molecule concentrations and rate constants. Bifurcation analysis can shed insight into parameter ranges feasible for realizing bistability. Nevertheless, even with the possibility of false negatives, it is significant that nearly 10% of explored systems are bistable and this percentage can only improve with greater analysis and exploration.
The simplest bistable system (3×2M101) involved 3 molecules and 2 reactions (Figure 2A). We tested its switch-like behavior by introducing perturbations from its stable states (Figure 2C). Small perturbations in 3×2M101 (small arrows in Figure 2C) caused transients which return to the originating stable state whereas large perturbations (large arrows) caused state flips. An intuitively appealing simpler system with only 2 molecules (Figure 2B) turned out not to be bistable with our mass-action formulation for enzymes (Text S1).
Positive feedback loops, such as autocatalysis and catalytic loops, have been implicated as a common motif leading to bistability in signaling [3],[18],[23]. In our study, autocatalysis (reactions D and L) was frequently present in bistable models, but it was not necessary. When we excluded autocatalysis from small reaction systems with only 3 molecules, there were fewer possible reaction configurations and a ten-fold reduction in percentage of bistable configurations. However, autocatalysis had little effect on the percentage of bistable configurations for 4 and 5 molecules (Figure 3).
In addition to autocatalysis, we found several cases where bistability arose from more subtle chemical interactions (e.g., Figure 2G and 2F and 3×3M40 in Figure 2E). Such reaction sets would have been difficult to identify as bistable by searching for similarities to published networks [24],[25]. Interestingly, all our switches had exactly two stable states; the lack of higher levels of multi-stability may simply be due to our sparse sampling of parameter combinations.
Are all discovered bistables distinct? Because isomorphisms were removed at the time of generating possible reaction signatures, we ensured that each discovered bistable mapped to a unique signature composed of the 12 basic reaction types. A remaining concern was that there might be equivalences in terms of the underlying dynamical system when the chemical systems were converted to mathematical models. We investigated this possibility by reducing all the composite reactions to approximate equivalences in the form of either a single reactant-single product reaction (type A) or a double reactant-single product reaction (type E) (see Methods section and Figure S2). We emphasize that these are “approximate” equivalences, for the following three reasons. First, many higher-order reactions required the inclusion of intermediate molecular species which were not present in the mathematical formulation of the original basis reactions. Second, the expanded reactions treated enzyme-substrate complexes as distinct molecular species having their own trajectories beginning from non-zero concentrations, whereas E-S complexes in the original reactions were initialized to zero in our modeling (see Methods). Third, backward rates from the E-S complexes to the reactants were assumed to be zero in our original modeling (e.g., for reactions C, D, J) whereas in the expanded modeling all reactions (forward and backward) have non-zero reaction rates. With these caveats in mind, we found situations where configurations were isomorphic according to our approximate mappings and both were bistable (Figure S2, frame B), and also cases where the configurations were approximately equivalent but one was bistable, and the others were not (Figure S2, frames C and D). These examples reveal that composite reactions such as are commonly used in biochemistry and in our study, complicate stability analysis in two ways. First, they may hide mechanistic similarities between systems. This can be addressed by expanding composite reactions into more basic steps, as we have done. Second, they may hide key assumptions such as intermediate species and fundamental reaction steps, which may cause major differences in the dynamical behavior of the reaction system. While this issue is important from a rigorous mathematical viewpoint, we point out that such approximations are inevitable when translating cellular biochemistry into idealized mathematical forms. We suggest that in many cases bistability is indeed preserved across approximations (e.g., Figure S2, frame B). Our study provides a framework for further systematic analysis of this question.
Does bistability “run” in families of related reaction topologies? To test this hypothesis, we constructed a directed acyclic graph (DAG) of configurations where each bistable configuration was a node, and each addition/removal of a reaction between nodes was an edge. We found that almost all bistable configurations from the first phase (3,415/3,562 = 95.9%) formed a single, highly interconnected set, i.e., a giant component. Most of the 147 “orphans” occurred at the boundaries of our sampling (98 at 3×6 and 47 at 4×3). These may simply represent novel ‘roots’ that connect further up in the reaction hierarchy. In Figure 4A, the DAG is represented as a multiply rooted “banyan-tree” like diagram where there is one main root (3×2M101) and multiple higher-order roots linked to the primary root through more complex, bistable “branches”. We may have missed low-propensity bistable configurations, so it is possible that isolated islands of lower propensity may be present. Conversely, it is also possible that finer sampling may uncover intermediate bistable systems that link orphan configurations into the DAG. We constructed a radial diagram restricted to those configurations that were derived from the simplest bistable configuration (3×2M101, Figure 4B). In Figure 4A and 4B, there was an apparent clustering of high-propensity nodes. We investigated this further by comparing projections of Figure 4A onto high-propensity nodes. A giant component persisted even when we increased the threshold for bistability propensity from >0 to ≥0.3 (Figure S1, frame A). This showed that highly bistable systems form a connected subgraph in the graph of all bistable systems. The much smaller non-autocatalytic subset of bistables was also multiply rooted with a giant component and a few separated nodes (Figure S1, frame B).
These graphs suggested that most bistable systems were derived from smaller ones. As reactions were added (Figure 4C), we encountered a decreasing number of novel bistables (i.e., cases that could not be derived by addition of a reaction to a smaller bistable configuration). This suggests that bistable systems involving small numbers of molecules may form the architectural core of more complex reaction networks that are also bistable.
We tested two implications of the “bistables are related” observation. First, we asked if we could take a large published bistable system and remove one reaction at a time without losing bistability. If we could continue this process till we ended up at a bistable configuration present in our dataset, then we had a continuous trajectory from our known DAG of bistables to the published model. Second, we asked if the large bistable system was a superset of a known bistable configuration, without requiring that there were intermediate bistables between the two. We performed this analysis on several known bistable reaction systems from published work (Figure 5). We found that in four of these cases, the published bistables were either already among our catalog, or had a subset of reactions that was bistable. In the remaining three configurations, there was neither a connection between the published models to the tree, nor was there a subset of reactions that was bistable in the DAG.
We therefore hypothesized that the DAG of bistables may be nearly complete for small systems, but the increasing degrees of freedom afforded by greater numbers of molecules and reactions helped realize bistability in new, unseen, ways. We developed two analysis tools that work in complementary ways to explore such larger configurations.
A suggestive observation from our first phase was that a large fraction of configurations (∼60%) contained saddle points and line solutions (Text S1). Most of these saddles (80.8% of the non-bistable set) occurred when the concentration of all but one molecule in the system was zero. A simple example of this is in Figure 2H, where molecule b catalyzes its own formation from a. When b is at zero, a does not change – it is metastable. However, the addition of a small amount of b causes the system to ‘fall’ into a truly stable state where all molecules have been converted into b. As has been previously analyzed [20] a rapid but saturating back-reaction is one way to convert this into a true stable system. This can be done using an enzyme to remove b more rapidly than it builds up, at low levels of b (Figure 2H). In this case, we reconstruct our original simplest bistable system (compare Figure 2A and 2H). We developed an algorithm that introduced several reactions to achieve bistability using this approach, in a general but not necessarily minimal manner. Due to the added reaction complexity, we generated bistables from non-bistable systems involving only 3 molecules and up to 5 reactions. The bistabilizer added at least 2 molecules, 3 enzymes and a reaction for each converted saddle point. We were able to generate a 5-fold higher proportion of bistables than were present in the source configurations in a sample of 70,000 models. This construction may usefully complement bifurcation discovery tools [26] to generate and refine bistable configurations.
Our second tool used motif matching. We analyzed the configurations of smaller bistables plus the sparsely sampled larger bistables to find frequently occurring groups of reactions, and then searched for these motifs in unexplored configurations. We analyzed bistables in each configuration class (3 molecules, 4 molecules, 5 molecules) separately for frequent motifs. A motif must occur with sufficient frequency in the given class to be detected (see Text S1 for frequency thresholds used). Because motifs are subsets of chemical reaction systems, they may not quite have the same number of molecules as the class of systems from which they are mined; for instance, a motif mined from 5-molecule bistable systems may not itself have 5 molecules. Furthermore, observe that while a motif is a subset of reactions that is well-represented in bistable systems, it may not be bistable. We found the greatest number of motifs (1615) from 3-molecule systems, and smaller numbers for 4 and 5 molecules (143 and 28, respectively), probably because our initial harvesting of bistables in the larger systems had yielded fewer confirmed bistables to scan for motifs.
The motifs were mostly independent and only one motif occurred in all three reaction classes. Coincidentally, this common, two-reaction, motif (composed of reactions DabX and Jbca) was identical to a bistable found both in our analysis and in previous work ([20]; Figure 2A and 5A). This motif/bistable also occurred in the top-5 motifs mined for each class when the motifs were ranked by their frequencies (Figure 6A–C). Interestingly, four of the top-5 motifs in the 3 molecule case contained at least one of reaction D or reaction J and all of the top-5 motifs in the 4 molecule case contained at least one of these two reactions. In the case of 5 molecules, only two of the top-5 motifs involved either D or J, though our much smaller sample set may have led to skewed results in this case. The remaining top-5 motif among the 3 molecule systems and the three other top-5 motifs in the 5-molecule systems utilized basic reversible reaction types such as A, B, F, G, H, and I that did not involve autocatalysis or even basic enzyme catalysis.
Just as a motif occurred in multiple bistable systems, a given bistable system could exhibit many distinct motifs. We used this property to advantage to help rank untested configurations for their potential to exhibit bistability. In each configuration class (3-molecule, 4-molecule, or 5-molecule systems), we searched for motifs specific to that class, and ranked the (untested) configurations in terms of the number of motifs they exhibited. We evaluated ≥100 of the top configurations for each class, exploring 120 parameter sets for each. We found bistability in 96% of the 3 molecule systems (214/222), 49% of the 4 molecule systems (49/100), and 13% of the 5 molecule systems (82/641). These numbers significantly improved upon the random sampling results of the second phase of analysis (Figure 6D).
Finally, we compared the motifs across the three classes (3, 4, and 5 molecule systems) to investigate whether there were any overlaps in mechanisms by which 3-molecule, 4-molecule, and 5-molecule systems exhibit bistability. On face value, there was little overlap between the motif sets taken pairwise (Figure 6E). To understand the distinctions better, we searched for all three sets of motifs in all three classes of bistable systems. For each motif, we counted its frequency in each class and identified the absolute value of the difference in frequencies across classes. The median difference in frequency is cataloged in Table 1 which shows that the motifs in 4-molecule systems were more similar to those in 5-molecule systems than either to 3-molecule systems. This suggests that there are qualitatively different mechanisms for bistability vis-à-vis 3 molecule systems and higher molecular systems.
Our study draws the first stability map of chemical reaction space. We find that bistables are common, especially in smaller reaction systems. They are also very robust, i.e., we find many configurations that are bistable over a very wide parameter range. Smaller bistables are all related to each other in a tree-like manner. While the overall configurations that support bistability are very diverse, there are frequently recurring motifs of reaction groupings in such configurations. These motifs serve to identify promising candidates in higher order systems.
Signaling motifs have been regarded as a good way to abstract out the chemical complexity of signaling [4],[24]. Specifically, positive feedback loops have long been considered good indicators for bistability. Our study shows that such broad network features are inadequate. The simplest form of positive feedback, that is, autocatalysis, is a good predictor for bistability only in very small reaction sets. In reactions with 4 or 5 molecules the proportion of bistables does not seem to depend on the presence or absence of autocatalysis (Figure 3). Instead we propose that our library of bistable configurations is a more complete and stronger approach. Our catalog, available from the DOCSS (Database of Chemical Stability Space) website at http://docss.ncbs.res.in, provides complete model descriptions in chemical reaction signature format, as well as selected bistables in SBML format. Together with recent methods to reduce chemical networks to their core reactions [27],[28], our catalog may open up chemical and bioinformatics approaches to searching for bistability in biochemical signaling pathways.
Bistable switches are important in biology in maintaining cellular history and decisions. Our study shows that there is a large repertoire of such switches for natural selection to draw upon, including many very simple switches. Furthermore, several of these switches are highly robust with respect to parameter variations. This has two implications for evolution. First, it is easy for evolving biochemical networks to stumble upon parameters that will give a switch. Second, such switches themselves will work effectively over a wide range of parameter conditions. The relatedness of the switches through addition or removal of individual reactions is also a good substrate for evolutionary modification. For example, a mutation that adds another enzyme regulator to a bistable switch is, by this argument, quite likely to retain the original bistability, along with the new regulatory properties. Overall, our survey of chemical topologies hints at an interconnected and rather well-populated terrain of bistability in a biologically biased region of chemical space.
We selected a set of 12 primary chemical reaction steps (Figure 1A) as the “alphabet” from which we performed our exhaustive search of chemical reaction space. In principle a small set of reactions may suffice to build up to arbitrary reaction schemes. For instance, using two reactions of type Ewe can realize the higher order enzymatic reactionwhich is reaction J in our system, by modeling it as a composite of
Our choice of primary chemical steps was biologically-inspired. In other words, we found a different proportion of bistables than we would see if we used, say, only the most elementary reactions such as type A and type E (Figure S2). Instead we reduced the parameter space by using biologically-inspired composite reactions, and hence sampled more completely in biological chemistry space. From this set of 12 primary chemical reaction steps, we constructed all possible reaction configurations involving 2 or 3 molecules, 4 molecules up to 5 reactions, and 5 molecules up to 4 reactions. This was a total of ∼2,800,000 configurations. These reaction configurations were topological: they defined the molecules and chemical steps, but did not specify concentrations or kinetic parameters.
We constructed reaction architectures involving m molecules as follows. We first selected one reaction out of the set of 12 reactions. We then assigned molecules to the slots of the reaction. For example, the reaction J has three slots, so we could assign molecules a, b, and c to this reaction. Having set up the first reaction, we then repeated the process n – 1 times to obtain a configuration of m molecules and n reactions (Figure 1). We eliminated all stoichiometrically invalid configurations by row-reducing the augmented stoichiometric matrix and checking for conserved moieties [29]. We repeated this entire process for all possible permutations of the m molecules. Similar approaches have been employed at a more elementary level of chemical reactive species to computationally analyze reaction systems [30],[31].
We signed each reaction with a terse unique 4-character string that completely specified all reactants and products, so that the first character of a reaction signature denotes one of the 12 reaction types (A–L), and the remaining two or three characters denote the molecular species participating in various roles in the reaction. The signature for a reaction architecture was obtained by concatenating the signatures for the constitutent reactions. We checked for isomorphic signatures (see Text S1) and only one signature per unique system was retained. The number of such unique, stoichiometrically valid reaction architectures was combinatorially large (Figure 1D). As further reactions were added, the number of possible configurations peaked and then declined because of stoichiometric constraints and symmetry (Figure 1D).
Our set of configurations did not deal with two cases that have previously been analyzed for bistability: continuous flux and buffered systems [21]. Instead our reactions required that there was mass conservation among the named molecules, but did permit the presence of ‘hidden’ molecules that were folded into the rate terms. Thus we represented a kinase reaction as an elementary enzymatic step, by “hiding” the ATP and ADP exchange: Substrate–Kinase → Product.
In this manner our reaction systems also accommodated steady state cases where continuous metabolic input was necessary to sustain stability. We stipulated that these ‘hidden’ molecules were stoichiometrically balanced within individual reactions, such as the enzymatic step above. We were able to approximate many cases of buffering simply by having a high concentration of the ‘buffered’ species.
In order to assess bistability we needed to work with specific models, with all parameters specified. We generated at least 100 models for each configuration we tested. Each model was generated from one of the configurations using Monte Carlo sampling to assign rate constants and concentrations. We chose concentrations using logarithmic sampling in the range 10 nM to 10 µM. This spans the concentration range of most biochemical reagents. We chose rate constants using logarithmic sampling in the range 0.01 µM−N s−1 to 10 µM−N s−1, where N was the order of the reaction from 0 to 4. Again, these rates were chosen to span the common range of biochemical reaction parameters.
Due to computational limitations we sampled only the smaller reaction sets completely for bistability. We completely sampled all configurations with 2 molecules, 3 molecules up to 6 reactions, and 4 molecules up to 3 reactions (Figure 1A–C and Text S1). This amounted to a total of 100,000 configurations and ∼20e6 models, sampling at least 100 models per configuration. We sampled the remaining reaction sets more sparsely, mostly 1 in 100, but we used 1 in 1,000 sampling for the very large 4×5 and 5×4 reaction sets.
We found steady states for each model using two distinct methods: homotopy continuation [32] and time course analysis (Text S1). Briefly, homotopy continuation finds steady states by tracking solution paths of systems of simultaneous equations. Time-course analysis simulates models from a number of distinct initial conditions toward steady states. Neither of these methods determined the global bifurcation behavior of the reaction system: they only identified fixed points of the fully parameterized model. We classified solutions as stable, saddle, or other using eigenvalue calculation and simulation around steady states (Text S1). |
10.1371/journal.ppat.1006120 | HTLV-1 bZIP Factor Enhances T-Cell Proliferation by Impeding the Suppressive Signaling of Co-inhibitory Receptors | Human T-cell leukemia virus type 1 (HTLV-1) causes adult T-cell leukemia-lymphoma (ATL) and inflammatory diseases. To enhance cell-to-cell transmission of HTLV-1, the virus increases the number of infected cells in vivo. HTLV-1 bZIP factor (HBZ) is constitutively expressed in HTLV-1 infected cells and ATL cells and promotes T-cell proliferation. However, the detailed mechanism by which it does so remains unknown. Here, we show that HBZ enhances the proliferation of expressing T cells after stimulation via the T-cell receptor. HBZ promotes this proliferation by influencing the expression and function of multiple co-inhibitory receptors. HBZ suppresses the expression of BTLA and LAIR-1 in HBZ expressing T cells and ATL cells. Expression of T cell immunoglobulin and ITIM domain (TIGIT) and Programmed cell death 1 (PD-1) was enhanced, but their suppressive effect on T-cell proliferation was functionally impaired. HBZ inhibits the co-localization of SHP-2 and PD-1 in T cells, thereby leading to impaired inhibition of T-cell proliferation and suppressed dephosphorylation of ZAP-70 and CD3ζ. HBZ does this by interacting with THEMIS, which associates with Grb2 and SHP-2. Thus, HBZ interacts with the SHP containing complex, impedes the suppressive signal from PD-1 and TIGIT, and enhances the proliferation of T cells. Although HBZ was present in both the nucleus and the cytoplasm of T cells, HBZ was localized largely in the nucleus by suppressed expression of THEMIS by shRNA. This indicates that THEMIS is responsible for cytoplasmic localization of HBZ in T cells. Since THEMIS is expressed only in T-lineage cells, HBZ mediated inhibition of the suppressive effects of co-inhibitory receptors accounts for how HTLV-1 induces proliferation only of T cells in vivo. This study reveals that HBZ targets co-inhibitory receptors to cause the proliferation of infected cells.
| Since HTLV-1 infects only through cell-to-cell transmission, increasing the number of infected cells is critical for transmission of HTLV-1. Proliferation of HTLV-1 infected cells is critical for development of leukemia and inflammatory diseases. In this study, we showed that HBZ promotes the proliferation of infected cells by targeting co-inhibitory receptors. Paradoxically, HBZ enhances the expression of the co-inhibitory receptors TIGIT and PD-1. We found that HBZ concurrently hampers the growth-inhibitory signal of TIGIT and PD-1, thereby leading to the enhanced proliferation of HTLV-1 infected cells in vivo. HBZ does this by interacting with THEMIS, which is expressed only in T cells. It is known that HTLV-1 infects different types of cells but increases only T cells. Functional impairment of co-inhibitory receptors by interaction of HBZ with THEMIS is a mechanism how HTLV-1 specifically induces proliferation of T cells.
| Human T-cell leukemia virus type 1 (HTLV-1) belongs to the delta type retrovirus group, which also includes bovine leukemia virus and HTLV-2. HTLV-1 causes adult T-cell leukemia-lymphoma (ATL) and inflammatory diseases [1–4]. This virus induces clonal proliferation of infected cells to enhance its transmission, since HTLV-1 is transmitted primarily by cell-to-cell contact [5–7]. It has been reported that an increased number of infected cells is correlated with a higher rate of transmission by breast-feeding [8]. Thus, increased numbers of HTLV-1 infected cells are beneficial for the transmission of this virus.
HTLV-1 encodes two regulatory genes, tax and rex, and three accessory genes (p12, p13, and p30) in the plus strand of the provirus [2]. Another regulatory gene, the HTLV-1 bZIP factor (HBZ) gene, is transcribed as an anti-sense transcript [9, 10]. It has been reported that HTLV-1 infected cells show higher susceptibility to antigenic stimulation. One mechanism of this hypersensitivity is due to Tax. Tax expression under control of the long terminal repeat (LTR) results in enhanced responsiveness to stimulation through the T-cell receptor (TCR)/CD3 complex [11, 12]. However, Tax expression is often lost in ATL cells and HTLV-1 infected cells [13–18]. Therefore, it is likely that another mechanism also promotes proliferation of HTLV-1 infected cells–perhaps a mechanism involving HBZ. Indeed, HBZ has been reported to promote proliferation of T cells in vivo and in vitro [19, 20].
The TCR recognizes cognate antigenic peptides presented by major histocompatibility complex molecules on antigen-presenting cells, and transduces a signal that is modulated by co-stimulatory and co-inhibitory receptors on the T cell [21, 22]. It has been reported that ATL cells and HTLV-1 infected cells express the co-inhibitory receptors PD-1 and T cell immunoglobulin and ITIM domain (TIGIT) on their surfaces [23–25]. Binding of one of these receptors to its ligand sends a suppressive signal through the ITIM or ITSM motif in the cytoplasmic region of the receptor [21]. However, ATL cells and HTLV-1 infected cells proliferate regardless of the higher expression of PD-1 and TIGIT on their surfaces. Until now, it has not been known how the suppressive signal from these co-inhibitory receptors is impaired.
In this study, we found that HBZ promotes T-cell proliferation mediated by TCR signaling. As a mechanism, HBZ interferes with the suppressive function of some co-inhibitory receptors and inhibits the expression of others. Thus, impairment of co-inhibitory receptors is a newly discovered mechanism by which HTLV-1 promotes the proliferation of infected T cells.
We have reported that HBZ promotes proliferation of a human T-cell line and HBZ knockdown inhibits proliferation of ATL cell lines [19]. Several mechanisms were identified for proliferation induced by HBZ [20, 26–31]. However, it remains unknown how HTLV-1 induces T-cell specific proliferation. We generated HBZ transgenic (HBZ-Tg) mice, in which HBZ is expressed under the control of the CD4 promoter/enhancer/silencer, so that only CD4+ T cells express HBZ [19, 32]. We also generated tax transgenic (tax-Tg) mice using the same promoter [33]. We isolated CD4+ T cells from HBZ-Tg and tax-Tg mice and evaluated their proliferation upon anti-CD3 stimulation. CD4+ T cells of HBZ-Tg mice proliferated much more than those of non-transgenic (non-Tg) mice, and the proliferation of CD4+ T cells was slightly enhanced in tax-Tg mice (Fig 1A and 1B). Co-culture of CD4+ T cells with dendritic cells (DC) further enhanced this proliferation (Fig 1A and 1B). However, the difference in proliferation between cells from HBZ-Tg and non-Tg mice was not observed in the presence of anti-CD28 antibody (0.3 μg/mL) (Fig 1C), indicating that CD4+ T cells of HBZ-Tg mice are hypersensitive to signaling via the TCR/CD3 complex.
To investigate whether the proliferation of CD4+ T cells of HBZ-Tg mice is increased in vivo, we induced experimental allergic encephalomyelitis (EAE) in HBZ-Tg and non-Tg mice by immunization with myelin oligodendrocyte glycoprotein (MOG)/complete Freund's adjuvant. Although disease severity was not different between HBZ-Tg mice and non-Tg mice (S1 Fig), the number of CD4+ T cells was increased only in the immunized HBZ-Tg mice compared with non-immunized HBZ-Tg and non-Tg mice (Fig 1D), suggesting that HBZ-expressing T cells have higher susceptibility to immune stimulation in vivo. HBZ-Tg did not show higher susceptibility to EAE regardless of impaired Treg function by HBZ [32]. It is speculated that partial inhibition of Treg functions by HBZ is not enough to increase incidence of EAE.
It has been reported that HTLV-1 infected cells and ATL cells express both co-stimulatory (OX40) and co-inhibitory receptors (PD-1 and TIGIT) on their surfaces [23, 25, 34, 35]. These findings suggest that HTLV-1 influences expression of co-inhibitory and co-stimulatory receptors. Therefore, we analyzed their expressions in HBZ-expressing T cells by real-time RT-PCR. As shown in Fig 2A, the expression of the co-inhibitory receptors TIGIT and PD-1 was enhanced, whereas transcription of other co-inhibitory receptors, BTLA and Lair-1, was suppressed in HBZ transduced T cells. HBZ suppressed somewhat the transcription of the co-stimulatory receptors CD28 and ICOS but did not influence that of OX40. In accordance with this finding, flow cytometric analyses showed that in CD4+ T cells from HBZ-Tg mice, cell-surface PD-1 and TIGIT were enhanced, while BTLA and LAIR-1 were decreased (Fig 2B and 2C). HBZ changed the cell-surface expression of co-stimulatory receptors only slightly or not at all (S2 Fig). On the other hand, expression of co-inhibitory receptors did not differ on CD4+ T cells between tax-Tg and non-Tg mice (S3 Fig). Expression of HBZ and tax in these transgenic mice was confirmed by RT-PCR (S4 Fig).
To study whether similar changes in levels of these co-stimulatory and co-inhibitory receptors are observed in ATL cells, we analyzed transcription and cell surface expression of these co-receptors. As shown in Fig 3A and 3C, TIGIT transcription and expression were significantly increased in ATL cases. PD-1 expression was upregulated in some ATL cases as reported previously [24]. BTLA transcription and cell-surface expression were not different in ATL cases compared with resting T cells, but suppressed compared with activated T cells (Fig 3A and 3C). Cell-surface expression of LAIR-1 was also suppressed in ATL cells. Transcripts of the HBZ and tax genes were measured by real-time RT-PCR in these cases (S5 Fig).
It makes sense that HBZ might decrease the expression of BTLA and LAIR-1, thus impairing their suppressive function and enhancing the proliferation of infected cells. However, enhanced expression of PD-1 and TIGIT would augment the suppressive function of these co-inhibitory receptors, leading to decreased proliferation of cells. This idea is not consistent with the observation that HBZ enhances proliferation of expressing T cells. Therefore, we speculated that even though HBZ increases the expression of TIGIT and PD-1, it may inhibit their suppressive function. On the other hand, expression of co-stimulatory receptors, ICOS and OX40 was decreased in ATL cases compared with control activated CD4+ T cells (S6 Fig).
The co-inhibitory receptors PD-1 and TIGIT possess ITIM or ITSM domains, and inhibit cell proliferation [21]. As described above, we hypothesized that HBZ may interfere with the T-cell inhibitory function induced by PD-1/PD-L1 and/or TIGIT/CD155 interaction. To study this possibility, we next analyzed the suppressive activity of TIGIT/CD155 interaction in the presence or absence of HBZ. CD4+ T cells were transduced with retroviruses expressing HBZ and stimulated with anti-CD3/CD155.Fc-coated beads or anti-CD3/control.Fc-coated beads. We then measured proliferation of the cells. As shown in Fig 4A, interaction with anti-CD3/CD155.Fc-coated beads suppressed the proliferation of CD4+ T cells transduced with empty vector, but not those transduced with HBZ, indicating that HBZ impairs TIGIT/CD155 mediated growth inhibition. Likewise, HBZ interfered with the suppressive effect of PD-1/PD-L1 interaction (Fig 4A). These data suggest that HBZ targets a common molecule(s) that is involved in mediating suppressive signals from both PD-1 and TIGIT. Furthermore, suppressive signal through BTLA was also inhibited by the presence of HBZ (S7 Fig). Thus, HBZ not only suppresses BTLA expression but also functionally inhibits suppressive signaling from BTLA.
Inhibitory signals through the ITIM and ITSM motifs of PD-1 and TIGIT are mediated by SHP-1 and SHP-2 [21, 36, 37], negative regulators of TCR signaling that dephosphorylate ZAP-70 and CD3ζ and suppress T-cell activation [38, 39]. We analyzed whether HBZ influences the interaction between the intracytoplasmic region of PD-1 and SHP-2. To study the interaction of SHP-2 with the cytoplasmic region of PD-1, we generated a chimeric molecule in which the intracytoplasmic region of human PD-1 was fused to the transmembrane and extracytoplasmic regions of mouse CD28 (mCD28/hPD1) [40]. Pervanadate induces tyrosine phosphorylation of intracellular proteins including PD-1, thus recruiting SHP-2 to the ITSM motif [40]. After treatment with pervanadate, SHP-2 was recruited to the chimeric molecule (Fig 4B, lane 2), while HBZ inhibited this interaction (lane 4). Thus, HBZ hinders recruitment of SHP-2 to the ITSM motif of PD-1.
Binding of PD-L1 induces a transient PD-1-TCR co-localization within microclusters–a co-localization that transiently associates with SHP-2 [41]. After stimulation by pervanadate, PD-1 formed TCR microclusters in Jurkat cells as reported previously (S8 Fig). Next, we analyzed the co-localization of PD-1 and SHP-2 after treatment with pervanadate to observe whether HBZ inhibits the interaction between PD-1 and SHP-2. As shown in Fig 5A, SHP-2 co-localized with PD-1 in Jurkat cells after pervanadate stimulation (stimulated Jurkat-mock). However, co-localization of these molecules was suppressed by the presence of HBZ (stimulated Jurkat-HBZ).
To quantitatively analyze the co-localization of PD-1 and SHP-2, we visualized PD-1 and SHP-2 using confocal microscopy. Captured raw images were analyzed using ImageJ software with the JACoP plug-in, and Pearson’s correlation coefficient [an index for the relationship of green (PD-1) and red (SHP-2) pixels] was calculated (Fig 5B). Co-localization of PD-1 and SHP-2 in Jurkat mock-transfected cells was minimal without pervanadate stimulation (the mean correlation coefficient value was 0.45), but these two molecules were highly co-localized upon pervanadate stimulation (correlation coefficient of 0.72). This value decreased dramatically, to 0.48, in the presence of HBZ (Jurkat-HBZ), indicating that HBZ strongly interferes with the co-localization of PD-1 and SHP-2. These data show that HBZ inhibits recruitment of SHP-2 to the cytoplasmic region of PD-1.
As shown above, HBZ inhibits the interaction between SHP-2 and PD-1. Indeed, phosphorylation of SHP-2 (Tyr580) was decreased in CD4+ T cells of HBZ-Tg mice (Fig 6A) and in HBZ-transduced murine primary T cells (Fig 6B). SHP-2 functions to dephosphorylate ZAP-70 and CD3ζ. After induction of phosphorylation by H2O2 [42], tyrosine phosphorylation of ZAP-70 lasted for a longer time in the presence of HBZ (Fig 6C). Similarly, tyrosine phosphorylation of CD3ζ persisted longer after it was induced by pervanadate (Fig 6D). These data indicate that HBZ interferes with the function of SHP-2, leading to suppressed dephosphorylation of ZAP-70 and CD3ζ.
PD-1 suppresses T-cell proliferation not only by interacting with SHP-1 and SHP-2, but also by interacting with PKCθ. Threonine phosphorylation (T538) of PKCθ is associated with its activation and IL-2 production by T cells. Signaling via PD-1 inhibits PKCθ T538 phosphorylation [43]. As shown in Fig 6E, HBZ did not enhance phosphorylation of PKCθ T538, in contrast to ZAP-70 and CD3ζ. These results suggest that HBZ mediated activation of TCR signaling is mainly through inhibition of the tyrosine phosphatase, SHP-2.
This study shows that HBZ inhibits the recruitment of SHP-2 to the cytoplasmic region of PD-1. However, it remains unknown how HBZ interacts with the complex containing SHP-2. Recently, THEMIS has been reported to interact with Grb2 and SHP-1 or 2, and inhibit T-cell activation [44]. We analyzed the interaction of HBZ with these host factors and found that HBZ binds to THEMIS, but not to Grb2 and SHP-2 (Fig 7A and S9 Fig), suggesting that THEMIS is a target of HBZ. Next, we analyzed whether HBZ affects the interaction between THEMIS and Grb2. We confirmed that THEMIS interacts with Grb2, and found that this interaction is hindered by the presence of HBZ (Fig 7B and 7C). These data demonstrate that HBZ interacts with THEMIS and partially impairs the association of THEMIS with Grb2. This interaction of HBZ with the complex containing SHP may hinder recruitment of SHP to the ITSM and ITIM motifs of co-inhibitory receptors such as PD-1.
Next, we analyzed whether HBZ interferes co-localization of PD-1 and THEMIS in the T cells. Stimulation by pervanadate induced co-localization of PD-1 and THEMIS, which was inhibited by HBZ (Fig 8A). Thus, HBZ interacts with THEMIS, which perturbs the complex containing SHP and impairs suppressive signal from PD-1 or TIGIT. To check whether suppressed THEMIS enhances T-cell proliferation through disrupted negative signal, we inhibited THEMIS expression by shRNA and found that suppressed THEMIS expression decreased T-cell proliferation (S10 Fig). It has been reported that proliferation of T cells from THEMIS knockout mice was suppressed [45], suggesting that THEMIS is also critical for T-cell proliferation in addition to suppressive signaling from co-inhibitory receptors.
THEMIS interacts with ITIM or ITSM domain of PD-1 and TIGIT in the cytoplasm whereas it has been reported that HBZ is primarily localized in the nucleus [46]. Therefore, localization of HBZ was analyzed in the presence of THEMIS. As reported previously, THEMIS existed in the cytoplasm (50 of 50 cells: 100%) whereas HBZ was mainly localized in the nucleus of 293T cells (67 of 74 cells: 90.5%)(Fig 8B). When both proteins were expressed, HBZ was co-localized with THEMIS in the cytoplasm (28 of 79 cells: 35.4%)(Fig 8B). Thus, THEMIS shifted localization of HBZ from nucleus to cytoplasm in 293T cells. When we analyzed localization of HBZ in T cells, HBZ was detected in both the nucleus and the cytoplasm (Fig 8C). Then, we analyzed THEMIS expression in 293T and Jurkat cells, and found that only Jurkat cells expressed THEMIS (Fig 8D), suggesting that THEMIS is responsible for cytoplasmic localization of HBZ.
To clarify the localization of HBZ in T cells in detail, we detected HBZ using antibody to the nuclear pore complex, and confirmed that HBZ was present in both nucleus and cytoplasm (Fig 8E). To study whether cytoplasmic localization of HBZ is attributed to THEMIS, we suppressed THEMIS expression using shRNA, and found that HBZ was present largely in the nucleus, suggesting that endogenous THEMIS contributes to changed localization of HBZ from the nucleus to the cytoplasm (Fig 9).
Co-stimulatory and co-inhibitory receptors control T-cell function and determine T-cell fate after a T cell is stimulated by TCR signaling [21]. In this study, we showed that HBZ enhances the susceptibility of expressing T cells to TCR-mediated signaling by perturbing signaling from co-inhibitory receptors. This study is the first to demonstrate that HBZ targets various co-inhibitory receptors by different mechanisms, and enhances proliferation. Expression of BTLA and LAIR-1 is decreased by HBZ, while HBZ impairs the suppressive function of PD-1 and TIGIT through inhibited recruitment of the SHP-2 containing complex to the cytoplasmic domain of PD-1. In contrast to BTLA and LAIR-1, expression of PD-1 and TIGIT are in fact upregulated by HBZ. Why does HBZ enhance expression of TIGIT and PD-1 among co-inhibitory receptors? Increased TIGIT expression competes with CD226, a co-stimulatory receptor, for binding with CD155, resulting in inhibition of T-cell activation [47]. In addition, HBZ suppressed CD226 expression [25]. Furthermore, our previous study indicated that TIGIT expressed on T cells is implicated in immune suppression through enhanced production of IL-10 from T cells and DC by reverse signaling [25]. Since reverse signal from PD-L1 and L2 on DC is also associated with suppressive phenotype of DC and moderate increase in IL-10 expression [48], PD-1 on T cells is also implicated in immune suppression. DC expresses the TIGIT ligand, CD155, and PD-1 ligands, PD-L1 and PD-L2, on the surface. DC-T-cell interaction plays a key role in immune responses to viral infections [49]. We have reported that increased expression of TIGIT on HBZ expressing T cells induces IL-10 production [25]. Furthermore, IL-12p40 production of DC cells was severely impaired in HBZ-Tg mice, which is likely caused by TIGIT on T cells [25, 47]. Thus HBZ suppresses host immune responses through enhanced PD-1 and TIGIT expression while simultaneously impairing SHP-2 mediated inhibitory signaling from these co-inhibitory receptors. In other words, HBZ modifies the functions of the co-inhibitory receptors PD-1 and TIGIT to allow the virus to evade the host immune system.
SHP-1 and 2 form complexes with Grb2 and THEMIS in T cells, and inhibit TCR mediated signaling [44]. Knockdown of THEMIS increased TCR-induced CD3ζ phosphorylation, a phenomenon that resembles the changes caused by HBZ. In this study, we show that HBZ interacts with THEMIS and weakens the interaction between THEMIS and Grb2. These data suggest that HBZ binding to THEMIS hinders recruitment of this complex to the ITSM motif of PD-1 and thus impedes suppressive signals. THEMIS is expressed only in the T-cell lineage [44, 50]. Therefore, it is thought that HBZ may not inhibit the co-inhibitory signal in non-T cells. Since the receptor for HTLV-1 is glucose transporter 1, HTLV-1 infects a variety of cells in vivo [51]. However, only infected T cells proliferate in vivo. Our observation that HBZ binds to THEMIS and impairs the growth-suppressive signal might account for this T-cell specificity of HTLV-1.
TCF-1 and LEF-1, transcription factors of classical Wnt signaling pathway, are critical for T-cell development in the thymus [52], and their expressions are suppressed in peripheral memory T cells. We have reported that TCF-1 and LEF-1 inhibit function of Tax, which may critically influence the peripheral T-cell tropism of this virus [53]. It remains unknown how this virus specifically promotes proliferation of infected mature T cells. This study reveals that interaction between HBZ and THEMIS impedes suppressive signal by interfered recruitment of SHP-2 to ITSM motif in T cells. This is thought to be a mechanism of T-cell specificity in HTLV-1 induced proliferation. Thus, HBZ and Tax determine specificity of HTLV-1 infected cells.
Analysis of the transcriptomes of 57 ATL cases using RNA-seq identified fusion gene products that contained five CD28-related in-frame fusions (CTLA4-CD28 or ICOS-CD28) that likely induce continuous or prolonged CD28 co-stimulatory signaling [54]. In addition, amplification of CD28 was frequently detected in ATL cases. Thus the CD28 co-stimulatory molecule is a frequent target of somatic changes in ATL cells. As shown in this study, HBZ perturbs co-inhibitory signaling, which enhances proliferation of not only ATL cells but also other HTLV-1 infected cells. Thus, in ATL cells, TCR signaling is a target of both somatic mutation and HBZ. We assume that HBZ perturbs TCR signaling and promotes proliferation beginning in individuals at the carrier state, and then somatic mutations potentiate and exacerbate the proliferative responses.
Recently it has been reported that structural variations in the 3’ untranslated region of PD-L1 enhanced PD-L1 expression in various cancer cells [55]. In particular, these structural variations were frequently observed in ATL (27% of 49 ATL cases). Overexpression of PD-L1 on ATL cells inhibits immune attack from CD8+ T cells through interaction with PD-1. Since ATL cells also express PD-1, PD-L1 might suppress the proliferation of ATL cells. However, since HBZ interrupts the suppressive signal from PD-1 through inhibited recruitment of the SHP containing complex to the ITSM motif, ATL cells avoid growth suppression while maintaining the immune suppressive effects of PD-1/PD-L1 interaction on CD8+ T cells.
HBZ is primarily localized in the nucleus [46], and interacts with transcription factors, which include p65, Smad3, and c-Jun, and other host factors such as p300 in the nucleus [56–59]. However, it has been reported that HBZ interacts with GADD34 in the cytoplasm, which activates the mammalian target of rapamycin (mTOR) signaling [60]. This finding suggests that HBZ also functions in the cytoplasm by interacting with host factors. This study showed that interaction of HBZ with THEMIS changes its localization and HBZ functions in the cytoplasm, and HBZ interferes suppressive function of co-inhibitory receptors. Thus, HBZ exerts the functions in both nucleus and cytoplasm.
In this study, we demonstrate that HBZ inhibits suppressive signaling from co-inhibitory receptors by decreased transcription or by inhibition of recruitment of SHP-2. Furthermore, HBZ enhances the expression of TIGIT and PD-1, which are associated with immune suppression. Thus, HBZ enables expressing T cells to survive and proliferate in vivo by utilizing and modifying the functions of co-inhibitory receptors.
C57BL/6J mice were purchased from CLEA Japan. Transgenic mice expressing HBZ (HBZ-Tg mice) under the control of the murine CD4-specific promoter/enhancer/silencer have been described previously [19]. All HBZ-Tg mice were heterozygotes for the transgene. Transgenic mice expressing tax (tax-Tg mice) under the control of the same promoter were generated as reported [33].
Jurkat cell line was provided by Dr. S. Sakaguchi (Osaka University, Japan). Jurkat cell lines stably expressing the spliced form of HBZ (Jurkat-HBZ) and control (Jurkat-mock) cells were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS) and 1 mg/mL G418 (Nacalai Tesque, Kyoto, Japan) [61]. The 293T cell line was purchased from ATCC (Manassas, VA, USA) and cultured in Dulbecco’s modified Eagle medium (DMEM) supplemented with 10% FBS. The 293FT cell line was purchased from Life Technologies and cultured in DMEM supplemented with 10% FBS and 0.5 mg/mL G418. The packaging cell line, Plat-E, was provided by Dr. T. Kitamura (Institute of Medical Science, The University of Tokyo, Japan) and cultured in DMEM containing 10% FBS, 10 μg/mL blasticidin and 1 μg/mL puromycin. These cell lines were grown at 37°C under a 5% CO2 atmosphere.
Peripheral blood mononuclear cells (PBMCs) of ATL patients and healthy donors were collected by Ficoll-Paque PLUS (GE Healthcare, Little Chalfont, UK). CD4+ T cells of healthy donors were isolated by Human CD4+ T Cell Enrichment Cocktail (STEMCELL Technologies, Vancouver, Canada) according to the manufacturer’s instructions. To obtain activated CD4+ T cells, CD4+ T cells were stimulated by 10 μg/mL phytohemagglutinin (PHA) (Sigma-Aldrich, St. Louis, MO, USA) for three days.
Animal experiments were performed in strict accordance with the Japanese animal welfare bodies (Law No. 105 dated 19 October 1973, modified on 2 June 2006), and the Regulation on Animal Experimentation at Kyoto University. The protocol was approved by the Institutional Animal Research Committee of Kyoto University (permit numbers D13-02, D14-02, D15-02, and A10-3). Experiments using clinical samples were conducted according to the principles expressed in the Declaration of Helsinki, and approved by the Institutional Review Board of Kyoto University (permit number G310). ATL patients and healthy blood donors provided written informed consent for the collection of samples and subsequent analysis.
pMX-IG, pMX-HBZ-IG and pMX-BTLA-IG were used for retrovirus production. The coding sequence of BTLA was amplified from cDNA of a wild type C57BL/6J mouse, and subcloned into pMX-IG. The HBZ expression vector, pcDNA3.1 HBZ-mycHis, was described previously [58]. The SHP-2 expression vector, pSP65SRa-SHP2-Flag, was kindly given by Dr. M. Hatakeyama (The University of Tokyo, Japan). The entire coding regions of Grb2 and THEMIS were amplified from cDNA prepared from resting human PBMCs or Jurkat cells. These PCR fragments were subcloned into pCMV-HA (Clontech Laboratories, Palo Alto, CA, USA) or pCAGGS-PA. The resulting plasmids were designated pCMV-HA-Grb2 and pCAGGS-PA-THEMIS, and they express HA (YPYDVPDYA)-tagged Grb2 and PA (GVAMPGAEDDVV)-tagged THEMIS, respectively. For generation of the chimeric mCD28/human PD-1 (hPD-1) expression vector, the mCD28 extracellular and transmembrane domains (bases 9–617 of NM_007642) were amplified from cDNA prepared from stimulated murine T cells, and the hPD-1 intracellular domain (bases 645–935 of NM_005018) was amplified from cDNA prepared from PHA-stimulated human PBMCs. These fragments were ligated and the resultant fragment was substituted for the GFP cording region of pMX-HBZ-IRES-GFP.
Anti-CD3 antibody (145-2X11, R&D systems, Minneapolis, MN, USA) together with recombinant mouse CD155.Fc or the control.Fc (Sino Biological, Beijing, China) or PD-L1. Fc or the control.Fc (R&D systems) or HVEM.Fc (R&D systems) were covalently attached to Dynabeads M450 Tosylactivated (Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA). Anti-CD3 antibody together with control.Fc was used for the control. For each 107 beads, 1 μg of anti-CD3 antibody (20% of total protein) and 4 μg of CD155.Fc, PD-L1.Fc, HVEM.Fc or control.Fc (80%) were used.
The PiggyBac-based shRNA expression vector, pB-CMV-GreenPuro-H1 (System Biosciences, Palo Alto, CA, USA), containing shRNA against THEMIS or luciferase (control), was introduced into Jurkat cells together with PiggyBac transposase expression vector by using Neon Transfection System (Invitrogen). Target sequences of each shRNA are shown in S1 Table.
To measure the proliferation of CD4+ T cells of HBZ-Tg mice, we isolated murine splenic CD4+ T cells using the CD4 T Lymphocyte Enrichment Set (BD Biosciences, San Jose, CA, USA). Murine splenic dendritic cells were isolated from collagenase-digested low-density cells using the Dendritic Cell Enrichment Set (BD Biosciences). Purified CD4+ T cells were labeled with 5-(and-6)-carboxyfluorescein diacetate succinimidyl ester (CFSE, Molecular Probes, Thermo Fisher Scientific, Waltham, MA, USA). Labeled CD4+ T cells of HBZ-Tg, non-Tg and tax-Tg mice (2×105 cells/well) were cultured with or without dendritic cells (1×104 cells/well) from non-Tg mice for three days with soluble anti-CD3 antibody (30 ng/mL) stimulation in round-bottomed 96-well plates. CFSE dilution was analyzed by flow cytometry. For TIGIT/CD155 and PD-1/PD-L1 proliferation assays, HBZ or empty vector transduced cells (see below) were labeled with CellTrace Violet (Invitrogen) and stimulated with anti-CD3/CD155.Fc or anti-CD3/PD-L1.Fc or anti-CD3/control.Fc-coated beads at a bead-to-cell ratio of 1:1 for three days. Dye dilution was analyzed by flow cytometry.
Transfection of the packaging cell line, Plat-E, was performed as reported [62]. Murine CD4+ T cells were activated by immobilized anti-CD3 (1 μg/mL) and soluble anti-CD28 (0.1 μg/mL) in 12-well plates. After 24 hours, activated T cells were transduced with virus supernatant and 4 μg/mL polybrene, and centrifuged at 3,000 rpm for 60 min. Cells were subsequently cultured for 48 hours.
Total RNA was isolated using Trizol Reagent (Invitrogen) and treated with DNase I to remove the genomic DNA. cDNAs were synthesized from 1 μg of total RNA using random primer and SuperScript III or IV reverse transcriptase according to the manufacturer’s instructions (Invitrogen). mRNA expression was analyzed by real-time PCR using FastStart Universal SYBR Green Master (Roche Diagnostics, Basel, Switzerland) and the StepOnePlus Real-Time PCR System (Applied Biosystems, Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer’s instructions. Primers used in this study are shown in S1 Table.
The following antibodies were used for flow cytometric analyses. Anti-mCD4 (GK1.5), mTIGIT (IG9), mCD28 (E18), mICOS (C398.4A), mOX40 (OX-86), hCD4 (RPA-T4), hPD-1 (29F.1A12), hBTLA (MIH26), hLAIR-1 (NKTA255), hCD28 (CD28.2), hICOS (C398.4A) and hOX40 (Ber-ACT35) antibodies were all purchased from BioLegend (San Diego, CA, USA). Anti-mBTLA (6F7), mLAIR-1 (113) and hTIGIT (MBSA43) antibodies were purchased from eBioscience (San Diego, CA, USA). Anti-mPD-1 (J43) was from BD Pharmingen (BD Biosciences). Anti-phospho-SHP-2 (Tyr580) antibody was from Cell Signaling Technology (Danvers, MA, USA). Anti-mouse IgG1, mouse IgG2b and rat IgG1 (BioLegend), Armenian hamster IgG (eBiosciences) and mouse IgG2a (BD Pharmingen) were purchased for isotype controls. For detection of SHP-2 phosphorylation, cells were permeabilized using BD Phosflow perm buffer II (BD Biosciences) according to the manufacturer’s instructions. Flow cytometric analysis was carried out using a FACSVerse with FACSuite software (BD Biosciences) and FlowJo (TreeStar, Ashland, OR, USA).
For the induction of EAE, five to seven-week old HBZ-Tg or control non-transgenic (non-Tg) C57BL/6J mice were immunized with an emulsion containing MOG peptide 35–55 (MEVGWYRSPFSRVVHLYRNGK). The emulsion was prepared by sonication, mixing 1 mL of MOG solution (1 mg/mL in PBS) with 1 mL of complete Freund’s adjuvant (Difco Laboratories, Detroit, MI, USA) containing desiccated Mycobacterium butyricum. The emulsion was injected subcutaneously in the area near the axillary lymph nodes and on both sides at the base of the tale of each mouse (50 μL/site, a total of 4 sites/mouse). On days 0 and 2 post-immunization, mice were injected intraperitoneally with 50 μL of a Pertussis toxin (Kaketsuken, Kumamoto, Japan) solution (4 μg/mL). Thereafter, mice were monitored daily for clinical signs of encephalomyelitis. A clinical score was assigned according to the following criteria: 0, no symptoms; 1, mild limp tail; 1.5, limp tail; 2, unilateral hind limb weakness or abnormal gait; 2.5, unilateral hind limb paralysis or bilateral hind limb weakness; 3, paraplegia; 3.5, unilateral fore limb weakness, with paraplegia; 4, unilateral fore limb paralysis or bilateral fore limb weakness; 4.5, bilateral fore limb paralysis; 5, moribund or dead.
For the immunoprecipitation studies of SHP-2 and PD-1, 293FT cells were transfected with the indicated expression vectors using Lipofectamine LTX (Invitrogen) according to manufacturer’s instructions. After 48 hours, cells were stimulated with 0.1 mM pervanadate solution for 5 min. The cell lysates were immunoprecipitated for 60 min at 4°C with 5 μg of anti-mCD28 (37.51), and immune complexes were incubated with Protein G-Sepharose (GE Healthcare) for 60 min at 4°C. For the immunoprecipitation studies of THEMIS, Grb2, SHP-2 and HBZ, 293FT cells were transfected as described above. After 48 hours, cells were stimulated with H2O2 for 5 min. The cell lysates were immunoprecipitated with 20 μg of anti-PA (NZ-1), anti-HA (HA-7) or anti-Flag (M2) antibodies, and immune complexes were incubated as described above. Normal mouse and rat IgG (Santa Cruz Biotechnology, Dallas, TX, USA) were used as controls.
The following antibodies were used for immunoblotting: anti-phospho-SHP-2 (Tyr580), phospho-ZAP-70 (Tyr319 and Tyr493), ZAP-70 (99F2), phospho-PKCθ (Thr538) and PKCθ (E1I7Y) antibodies were purchased from Cell Signaling Technology. Anti-phosphor-CD3ζ (Tyr83), CD3ζ and anti-THEMIS were purchased from Abcam (Cambridge, UK). Anti-PA (NZ-1) was purchased from Wako, Osaka, Japan. Anti-Flag-HRP (M2), HA-HRP (HA7), Myc (9E10), and tubulin (DM1A) antibodies were purchased from Sigma-Aldrich. Anti-mouse IgG-HRP, rabbit IgG-HRP and rat IgG-HRP antibodies were purchased from GE Healthcare. Mouse anti-HBZ monoclonal antibody (clone 1A10) was generated by immunizing C57BL/6 with using keyhole limpet hemocyanin (KLH)-conjugated HBZ peptide 97–133 (CKQIAEYLKRKEEEKARRRRRAEKKAADVARRKQEEQE).
To detect co-localizations of PD-1 with SHP-2, THEMIS or TCR, Jurkat-mock or Jurkat-HBZ cells were stimulated with 0.2 mM pervanadate solution for 2 min at 37°C. To evaluate the effect of THEMIS on the localization of HBZ, THEMIS-knocked down (KD) and control (luciferase KD) Jurkat cells were transfected with pcDNA 3.1 HBZ-mycHis or empty vector. The cells were washed with PBS and placed on MAS-coated glass slides (Matsunami Glass, Osaka, Japan). To detect HBZ, 293T cells cultured on type I collagen (Cellmatrix, Nitta Gelatin, Osaka, Japan)-coated coverslips, were transfected with pcDNA 3.1 HBZ-mycHis and/or pCAGGS-PA-THEMIS. The cells were fixed with 4% paraformaldehyde for 15 min, permeabilized with 0.2% Triton X-100 for 15 min, and blocked by incubation in 5% donkey serum (Jackson ImmunoResearch, West Grove, PA, USA) for 60 min. For immunostaining, the cells were incubated with anti-SHP-2 (sc-280), anti-PD-1 (sc-10297), anti-TCR β (sc-5277) (all Santa Cruz Biotechnology), anti-THEMIS (ab126771), anti-Nuclear Pore Complex Proteins (ab24609) (all Abcam), anti-myc (9E10, Sigma-Aldrich or Abcam) or anti-PA (NZ-1) antibodies for 60 min, followed by incubation with Alexa Fluor 488-conjugated donkey anti-goat IgG, Alexa Fluor 488-conjugated donkey anti-rat IgG, Alexa Fluor 594-conjugated donkey anti-mouse IgG, Alexa Fluor 594-conjugated donkey anti-rat IgG, Alexa Fluor 647-conjugated donkey anti-mouse IgG or Alexa Fluor 647-conjugated donkey anti-rabbit IgG antibodies (all Invitrogen), or DyLight 405-conjugated donkey anti-mouse IgG (Jackson ImmunoResearch) for 30 min. The stained cells were mounted with ProLong Gold Antifade Reagent or ProLong Gold Antifade Reagent with DAPI (all Molecular Probes, Thermo Fisher Scientific), imaged using an FV1000 confocal microscope (Olympus, Tokyo, Japan) or a Leica TCS SP8 (Leica Microsystems, Wetzlar, Germany), and analyzed with ImageJ.
For Figs 1, 2, 7, S2 and S10, statistical significance was determined by the two-tailed unpaired Student’s t-test. For Figs 3, 5 and S6, statistical analysis was performed using the one-way ANOVA with Tukey’s post hoc test (GraphPad Prism, GraphPad Software, La Jolla, CA, USA). Asterisks indicate the statistical significance as follows: *P < 0.05; **P < 0.01; ***P < 0.001; n.s., not significant.
|
10.1371/journal.pmed.1002855 | Operative versus non-operative treatment for 2-part proximal humerus fracture: A multicenter randomized controlled trial | Although increasingly used, the benefit of surgical treatment of displaced 2-part proximal humerus fractures has not been proven. This trial evaluates the clinical effectiveness of surgery with locking plate compared with non-operative treatment for these fractures.
The NITEP group conducted a superiority, assessor-blinded, multicenter randomized trial in 6 hospitals in Finland, Estonia, Sweden, and Denmark. Eighty-eight patients aged 60 years or older with displaced (more than 1 cm or 45 degrees) 2-part surgical or anatomical neck proximal humerus fracture were randomly assigned in a 1:1 ratio to undergo either operative treatment with a locking plate or non-operative treatment. The mean age of patients was 72 years in the non-operative group and 73 years in the operative group, with a female sex distribution of 95% and 87%, respectively. Patients were recruited between February 2011 and April 2016. The primary outcome measure was Disabilities of Arm, Shoulder, and Hand (DASH) score at 2-year follow-up. Secondary outcomes included Constant–Murley score, the visual analogue scale for pain, the quality of life questionnaire 15D, EuroQol Group’s 5-dimension self-reported questionnaire EQ-5D, the Oxford Shoulder Score, and complications. The mean DASH score (0 best, 100 worst) at 2 years was 18.5 points for the operative treatment group and 17.4 points for the non-operative group (mean difference 1.1 [95% CI −7.8 to 9.4], p = 0.81). At 2 years, there were no statistically or clinically significant between-group differences in any of the outcome measures. All 3 complications resulting in secondary surgery occurred in the operative group. The lack of blinding in patient-reported outcome assessment is a limitation of the study. Our assessor physiotherapists were, however, blinded.
This trial found no significant difference in clinical outcomes at 2 years between surgery and non-operative treatment in patients 60 years of age or older with displaced 2-part fractures of the proximal humerus. These results suggest that the current practice of performing surgery on the majority of displaced proximal 2-part fractures of the humerus in older adults may not be beneficial.
ClinicalTrials.gov NCT01246167.
| Proximal humerus fractures (PHFs) are among the most common fractures in older adults. The risks for suffering a PHF increase with age, especially after the age of 60 years.
Non-operative treatment of non-displaced PHF is uniformly approved, but the treatment of displaced 2-part PHF remains controversial. Recently, several studies have reported a significant increase in the surgical treatment of 2-part PHF, mainly due to the introduction of locking plates suitable for treating osteoporotic bone.
In this study, we investigated whether surgery with a locking plate is superior to non-operative treatment with a collar-cuff and early physiotherapy in the treatment of 2-part PHF involving the surgical neck.
We randomly allocated 88 patients aged 60 years or older with displaced 2-part PHF into 2 groups. One group underwent surgery with a locking plate and the other conservative (non-operative) treatment.
We recorded the Disabilities of Arm, Shoulder, and Hand (DASH) score, other measures of shoulder function and quality of life, and complications at 2-year follow-up.
We found no significant or clinically important between-group differences in all outcomes.
This trial provides no evidence that surgery is superior to non-operative treatment in the treatment of displaced 2-part PHF involving the surgical neck in older adults.
These results suggest that the current practice of performing surgery on the majority of displaced 2-part proximal fractures of the humerus in older adults may not be beneficial.
| Proximal humerus fractures (PHFs) are among the most common fractures in the older adult population [1,2]. In a Swedish nationwide study, the person-based incidence of PHF in adults was 175 per 100,000 person-years in women and 68 per 100,000 person-years in men [3]. The risk for having a PHF increases with age, especially after the age of 60 years [3,4].
The majority of PHFs are associated with a low-energy fall [4]. According to the published literature, minimally displaced or 2-part fractures constitute between 77% and 84% of all fractures [2,5]. Additionally, in another study, 76% are classified as AO Foundation/Orthopaedic Trauma Association (AO/OTA) A-class fractures [6], for which non-operative treatment can be considered. Fracture comminution and displacement, especially in 3- and 4-part fractures, on the other hand, are considered to be potential indications for operative treatment [7]. Recently, several studies have reported a substantial increase in surgery and claim it has now become current practice in the treatment of PHF, especially among the people aged over 60 years [3,8]. This increase has been mainly due to the introduction of locking plates, even though proper evidence of the superiority of surgery is lacking [8]. A few randomized controlled trials have compared operative treatment (plating or hemi-endoprosthesis) with non-operative treatment in displaced 3- and 4-part fractures [9–12]. The PROFHER trial, in which the majority of the fractures were displaced according to Neer criteria, also included 2-part fractures [12]. The study showed no differences between operative treatment and non-operative treatment based on a patient-reported outcome measure (PROM) or quality of life score. Based on the most recent Cochrane review as well as on other meta-analyses, no statistically or clinically significant difference in PROMs between operative and non-operative treatment has been observed [13–15].
It could be hypothesized that patients with displaced (defined by Neer criteria as more than 1 cm or 45 degrees) 2-part fractures [16] would benefit from operative treatment with a locking plate, but there is a scarcity of knowledge of the treatment outcomes associated with displaced 2-part fractures. We therefore conducted a multicenter, randomized, controlled efficacy trial that compared operative treatment with a locking plate with non-operative treatment in patients aged 60 years or over with 2-part displaced PHFs [17].
This PHF trial was a superiority, multicenter NITEP (Nordic Innovative Trial to Evaluate osteoPorotic fractures) group trial. The trial was conducted at 6 hospitals in 4 Northern European countries (Finland, Sweden, Denmark, and Estonia) [17]. All collaborating hospitals are emergency hospitals that routinely treat trauma patients. A specialized upper extremity team at each participating hospital provided the allocated treatment for the randomized patients, who were recruited between February 2011 and April 2016. The patients were informed about the treatment options during a pre-interviewing session, and all eligible patients gave their written informed consent before the randomization and initialization of the allocated treatment. The trial protocol was approved by the Regional Ethics Committee of Tampere University Hospital. In addition, the ethical committees of all participating hospital districts granted their ethical approval, and each hospital provided trial authorization before starting recruitment. Furthermore, independent steering and monitoring committees inspected the trial. The trial was designed by members of the protocol committee, and details of the trial design and methods have been published elsewhere [17].
In the protocol [17], 2 trial strata were described. Stratum I comprised 2-part fractures and stratum II comprised 3- and 4-part fractures. It was foreseen in the original protocol that stratum I would commence first, and therefore the reporting was planned separately for each protocol. At present, stratum II is still in the recruitment phase (159 of 218 patients recruited by 2 April 2019), which is scheduled to be completed during 2019. The analysis and reporting of these patients will start after the 2-year follow-up period. The trial protocol was signed by the principal investigator at each participating hospital in order to respect the approved treatment strategies.
The protocol was held in an Investigator Site File (ISF), and good clinical practice was the guiding principal in maintaining the study. The local investigators and the trial manager gathered the data. We used a prospective, randomized, open-label, blinded-endpoint design, where patients knew their allocation, but the persons who collected and analyzed the data and the authors were unaware of the study-group allocation of the patients. The authors vouch for the accuracy and completeness of the reported data and analyses, and the fidelity of the study to the protocol. The authors wrote the manuscript and made the decision to submit the manuscript for publication. This study is reported as per the Consolidated Standards of Reporting Trials (CONSORT) guideline (S1 CONSORT Checklist).
Patients aged 60 years or over with displaced 2-part low-energy PHF (in which the fracture line emerges through the surgical or anatomical neck) occurring less than 2 weeks before allocation and treatment onset were eligible for inclusion. Displacement was defined by Neer classification (displacement more than 1 cm or 45 degrees, with bony contact) [16]. Pre-allocation radiograph along with computed tomography (CT) confirmed the classification. Furthermore, 2 upper extremity trauma surgeons reviewed each patient before randomization to confirm the classification. Detailed inclusion and exclusion criteria are presented in S1 Appendix.
Randomization was performed at the patient level. Eligible patients were randomly assigned in a 1:1 ratio to undergo operative treatment with a locking plate or non-operative treatment. After enrollment, patients underwent randomization by means of a telephone call from the treating surgeon to the coordinating center’s research nurse. Sealed envelopes were used, and the pre-trial randomization sequence was generated with 10 blocks according to center and stratified by age (60 to 70 years, more than 70 years) due to the association between age and the measured outcomes. The trial was semi-blinded, and the outcome assessors—trained physiotherapists who otherwise did not take part in the study—were unaware of which treatment group patients belonged to, and patients were encouraged not to reveal their treatment group. In addition, patients wore a T-shirt to cover any scars on their shoulder.
The treatments in the operative and non-operative groups were standardized [17]. Patients who received operative treatment with a Philos locking plate (Synthes, Solothurn, Switzerland) were operated on by an upper extremity trauma surgeon. The operating surgeon had a minimum of 5 years’ experience in shoulder trauma, and thus any learning curve problems in the treatment were avoided. After treatment, both groups received the same physiotherapy described in more detail in S2 Appendix. A collar-cuff or sling was worn for 3 weeks to reduce pain, and pendulum movements were initiated from the first day possible (non-operative patients immediately and surgery patients from first post-op day). Elbow, wrist, and fingers were mobilized, and the use of the injured upper extremity in daily activities was encouraged. After 3 weeks, active range-of-motion exercises were initiated under the supervision of a physiotherapist. Both groups were offered supervised physiotherapy rehabilitation sessions. At the follow-up visits, the patients were asked about the number of sessions received.
The primary outcome was Disabilities of the Arm, Shoulder, and Hand (DASH) score measured at 2-year follow-up. The DASH is a PROM with 36 questions on coping in different everyday-life tasks [18]. The scale ranges from 0 to 100, with a low number indicating better function. The minimal clinically important difference (MCID) on the DASH has been estimated to be 10 to 15 [19–21]. Secondary outcomes included Constant–Murley score (CS) [22], the visual analogue scale for pain (VAS; 0 to 100 mm) [23], the quality of life questionnaire 15D [24], the EuroQol Group’s 5-dimension self-reported questionnaire EQ-5D(-3L) (with normal values from Finland) [25], and the Oxford Shoulder Score (OSS) [26]. The CS has a known wide interobserver variation, and therefore we arranged pre-trial training for the investigators in order to harmonize the measurements. The follow-up questionnaires included a separate section on adverse events (AEs), defined as untoward medical occurrences that may or may not have a causal relationship with the treatment administered. AEs were classified as serious AEs (SAEs) if they necessitated hospitalization or prolonged inpatient hospital care, or if they were life-threatening or resulted in death. Complications were also recorded (infection, nerve damage, bleeding, mal- or non-union, hardware problems). The data were collected during the research visits at 3 months, 6 months, 12 months, and 24 months.
The baseline and follow-up data acquired from the patients were stored in paper portfolios that were analyzed after the last patient’s 2-year follow-up visit in April 2018. The portfolios included the patients’ baseline characteristics, patient questionnaires, validated questionnaires, and case report files that contained information about allocation group, crossovers (treatment another than the randomized allocation), AEs, and SAEs.
Primary approval was received from the Regional Ethics Committee of Tampere University Hospital, ETL-code R10127. Participation was voluntary. All data are anonymous.
The trial was powered to detect a MCID in the DASH score of at least 10 points with a standard deviation (SD) of 15 (effect size d = 0.67) points. The initial planned sample size was 74 patients, which, at a 5% significance level, would provide 80% power to detect an effect size of 0.67 for the comparison of the 2 different treatment methods. Because the anticipated loss to follow-up was higher than the originally expected 10%, it was decided to raise the number of recruited patients from 81 to 88, resulting in 44 patients per group. This and other changes to protocol are summarized in S3 Appendix. The baseline characteristics were analyzed with descriptive statistics. Although we stated in the original trial protocol published in 2012 that subgroup analysis with respect to smoking, fracture type, and other diseases would be carried out, this was not done due to the rarity of these baseline characteristics in our sample. For the primary analysis, the primary and secondary outcomes at 2 years were compared between study groups with 95% confidence intervals. The principal analysis of the trial was performed after all participants had completed the 2-year follow-up. Trial analyses were conducted according to a prespecified statistical plan with the use of R software, version 3.5. All analyses were on an intention-to-treat basis and included all randomized patients in the groups to which they were randomized. Six patients were lost immediately after the allocation or were missing data from all the time points, and these patients were excluded from all the analyses. For the remaining patients, each outcome was treated as a time series in the individual patients, and the methods of linear interpolation and last observation carried forward were used to impute the missing data. A per protocol analysis was also conducted as a sensitivity analysis for the primary outcome. Continuous outcomes were compared using the Student t test. For each comparison, the homogeneity of variance was tested using Levene’s test meeting the assumption of the Student t test. Due to the skewness of the PROMs, we also used Mann–Whitney U test to compare the ranks between groups. All appropriate statistical tests were 2-sided.
Before revealing the randomization code, the writing committee (APL, VMM, MKL, and BOS) developed and recorded 2 interpretations of the results based on a blinded review of the primary outcome data (treatment A compared with treatment B). Thus, we aimed to decrease the possible researcher bias when interpreting the results. Only after the members of the writing committee had agreed that there would be no further changes in the interpretations, and unanimous consensus over the analysis reached, was the randomization code revealed, the correct interpretation chosen, and the manuscript finalized.
There was no patient involvement during the designing, recruiting, or conducting of the trial. However, after publication and dissemination of the results, patient organizations, health policy makers, and the general public will be informed of the important findings of the trial by means of congresses, social media, and general newsfeeds.
Between February 2011 and April 2016, a total of 88 patients with PHF with surgical neck involvement were randomly assigned to undergo either operative treatment with a locking plate or non-operative treatment. In total, 44 patients were assigned to undergo operative treatment with a locking plate and 44 to non-operative treatment (Fig 1). The characteristics of the study population at the time of enrollment are shown in Table 1.
The mean DASH score at 2 years was 18.5 (SE 3.1) points for the operative treatment group and 17.4 (SE 2.8) points for the non-operative group. The between-group difference was 1.1 points (95% CI −7.8 to 9.4, p = 0.81). Detailed results are shown in Fig 2 and Table 2. At 2 years, there were no statistical or clinically significant between-group differences in the study groups. Sensitivity analysis using the available data showed similar results (S4 Appendix).
When stratified by age, the between-group difference in mean DASH score (with t test) at 2 years in patients aged 60–70 years was −0.2 (95% CI −10.1 to 9.7, p = 0.97). The between-group difference in mean DASH score at 2 years in patients aged more than 70 years was 1.7 (95 CI −11.2 to 14.5, p = 0.79). An additional secondary sensitivity analysis, requested by an external reviewer, showed no significant between-group difference in the number of patients with a DASH score difference of 10 points or more between baseline and 24 months (p = 0.30); see S4 Appendix.
At 2 years, we found no statistical or clinically significant between-group differences in study groups for the CS (mean 68.0 points [SE 3.2] in the operative treatment group compared with 66.0 points [SE 3.3] in the non-operative group; between-group difference 2.0 [95% CI −5.6 to 9.6], p = 0.6). For the OSS, the mean was 40.2 points (SE 1.5) in the operative treatment group compared with 41.5 points (SE 1.4) in the non-operative group (between-group difference −1.3 [95% CI −5.3 to 2.8], p = 0.54). For the EQ-5D, the mean was 0.87 points (SE 0.02) in the operative treatment groups compared with 0.89 points (SE 0.02) in the non-operative group (between-group difference −0.02 [95% CI −0.08 to 0.02], p = 0.27). For the VAS, the mean was 11.5 points (SE 3.3) in the operative treatment group compared with 9.9 points (SE 2.7) in the non-operative group (between-group difference 1.6 [95% CI −6.8 to 9.9], p = 0.72). For the 15D, the mean was 0.862 points (SE 0.029) in the operative treatment groups compared with 0.879 points (SE 0.017) in the non-operative group (between-group difference −0.016 [95% CI −0.084 to 0.049], p = 0.6). For physiotherapy, the median number of visits was 4 in the operative group (range 0 to 25 visits) compared with 4 in the non-operative group (range 0 to 28 visits) (p = 0.38).
At 2 years, there were 3 complications. They all occurred in the operative group and required subsequent surgery. No other complications were detected. One patient fell and sustained a peri-implant fracture distal to the tip of the plate. This fracture was treated with open reduction and internal fixation by long anatomical locking plate. Two patients had implant failures because proximal screws migrated into the joint. These patients were treated with revision operation, and the screws were changed. The mean time to revision operation was 5.5 months (range 1.3 to 12.4 months). In the non-operative group, there were no complications; however, the difference between the groups was not statistically significant (3 versus 0, p = 0.24). One patient died during the study period. The death was not related to the trial. No patient withdrew from the study because of adverse effects.
The present investigation is, to our knowledge, the first prospective randomized controlled trial to compare non-operative and operative treatment with Philos plate in patients aged 60 years or older with displaced 2-part PHFs. This trial provides no evidence that surgery is superior to non-operative treatment in 2-year follow-up. Furthermore, we found no clinically or statistically significant between-group differences in any of the outcomes measured, including DASH (our primary outcome measure), CS, OSS, EQ-5D, 15D, VAS, complications, and mortality.
The DASH score, OSS, and CS are outcome measures commonly used to assess upper limb function after PHF. We acknowledge the limits of the DASH score as being a patient-reported scoring tool to assess symptoms and physical disability in the arm. However, it is also widely used in PHFs, it has a known MCID [19] and known average values for the general population, and it is validated in all Nordic countries.
The mean DASH scores in both of groups improved between 3 months and 6 months and between 6 months and 1 year. However, the scores did not improve after 1 year, when they reached the levels of baseline scores. The baseline values in our study are in line with the average values calculated from the general population, which constitutes both healthy individuals and people with disabilities of the upper limb [27]. Based on our results, we can conclude that the final function and satisfaction, measured by DASH, is obtained 1 year after the trauma.
We are not aware of any other published randomized trial that makes this specific comparison for displaced 2-part fractures of the surgical neck of the proximal humerus in older adults only. These are the most common displaced PHFs [5,28]. The PROFHER trial, which also included younger patients, is the only other trial to our knowledge that included these fractures. Although the study design, study settings, and age criteria differ between our study and PROFHER, both trials found no significant differences between surgical and non-operative treatment.
Although the difference in complications leading to revision operation in the present study did not reach statistical significance, the operative group had several complications, whereas the non-operative treatment group had none of any kind. All 3 complications leading to revision operation occurred within the first year (mean 5.5 months) after the surgery. In contrast, in the trial by Olerud et al., less than a quarter (7% from a total of 30%) of re-operations took place in the first year of follow-up [10]. In the PROFHER trial, 70% of secondary surgeries occurred within the first year [12].
One strength of the study lies in the fact that the trial participants had sustained injuries that are typically considered for surgical intervention. Second, both interventions and treatment protocols were representative of good practice, and this included the fact that all of the operations were performed by consultant surgeons. Additionally, the Philos locking plate used in this study is the most commonly used implant in current practice in Northern European countries as well as in many other countries. Moreover, only patients with 2-part PHFs involving the surgical neck were included in the trial. Previously, there has been disagreement over recognizing different fracture categories [29]. In our previous publication, however, we found substantial intra- and interobserver agreement in re-categorized Neer classification [30]. Finally, the dropout rate in our trial was 18%, which, according to Furlan et al., is acceptable, especially among patients with high mean age and numerous comorbidities [31].
The main weaknesses of the study were that patients were not blinded to the treatment, although outcome assessor physiotherapists were. Further, even though a priori power analysis was performed in order to establish the sample size required to adequately differentiate a true lack of clinically meaningful difference, there is the possibility the study was underpowered. Our main outcome variable, the DASH questionnaire, focuses on functional status and symptoms in general rather than on one specific anatomical region or specific disease entity. There is, however, no validated outcome measure specifically for shoulder fractures. In addition, as in the general PHF population, most patients in this study were women. Therefore, the results are representative and reliable for the female population.
The study group comprised patients with 2-part fracture of the proximal humerus with contact among the fragments. Hence, the results of this study are not applicable to fractures with no contact, head-split fractures, or dislocated fractures. For a better overview of patients included in the trial, we have uploaded anonymous plain radiographs to the NITEP group website along with 2-year DASH-values (http://www.nitep.eu/rtg).
The results of this superiority, semi-blinded randomized controlled trial show that surgery is not more beneficial than non-operative treatment in patients 60 years of age or older with displaced 2-part PHF. These results suggest that the current practice of performing surgery on the majority of displaced 2-part proximal fractures of the humerus in older adults may not be beneficial. Mean DASH scores already returned to the baseline level at 1 year, and therefore 1-year follow-up will be sufficient for future PHF trials with non-operative treatment and plating.
|
10.1371/journal.pcbi.1004534 | Bayesian Estimation of Conditional Independence Graphs Improves Functional Connectivity Estimates | Functional connectivity concerns the correlated activity between neuronal populations in spatially segregated regions of the brain, which may be studied using functional magnetic resonance imaging (fMRI). This coupled activity is conveniently expressed using covariance, but this measure fails to distinguish between direct and indirect effects. A popular alternative that addresses this issue is partial correlation, which regresses out the signal of potentially confounding variables, resulting in a measure that reveals only direct connections. Importantly, provided the data are normally distributed, if two variables are conditionally independent given all other variables, their respective partial correlation is zero. In this paper, we propose a probabilistic generative model that allows us to estimate functional connectivity in terms of both partial correlations and a graph representing conditional independencies. Simulation results show that this methodology is able to outperform the graphical LASSO, which is the de facto standard for estimating partial correlations. Furthermore, we apply the model to estimate functional connectivity for twenty subjects using resting-state fMRI data. Results show that our model provides a richer representation of functional connectivity as compared to considering partial correlations alone. Finally, we demonstrate how our approach can be extended in several ways, for instance to achieve data fusion by informing the conditional independence graph with data from probabilistic tractography. As our Bayesian formulation of functional connectivity provides access to the posterior distribution instead of only to point estimates, we are able to quantify the uncertainty associated with our results. This reveals that while we are able to infer a clear backbone of connectivity in our empirical results, the data are not accurately described by simply looking at the mode of the distribution over connectivity. The implication of this is that deterministic alternatives may misjudge connectivity results by drawing conclusions from noisy and limited data.
| Significant neuroscientific effort is devoted to elucidating functional connectivity between spatially segregated brain regions. This requires that we are able to quantify the degree of dependence between the signals of different areas. Yet how this must be accomplished—using which measures, each with their own limitations and interpretations—is far from a trivial task. One frequently advocated metric for functional connectivity is partial correlation, which is related to conditional independence: if two regions are independent, conditioned on all other regions, then their partial correlation is zero, assuming Gaussian data. Here, we use a probabilistic generative model to describe the relationship between functional connectivity and conditional independence. We apply this Bayesian approach to reveal functional connectivity between subcortical areas, and in addition we propose different variants of the generative model for connectivity. In the first, we address how a Bayesian formulation of connectivity allows for integration with other imaging modalities, resulting in data fusion. Secondly, we show how prior constraints can be incorporated in our estimates of connectivity.
| In the early days of neuroscience much attention was devoted to identifying the functional specialization of different brain areas [1]. More recently, this focus has shifted towards revealing how these areas are organized into networks and how these networks, rather than their individual constituents, are related to cognition [2–4] and neurological or psychological pathology [5–7]. The increasing interest in neuronal connectivity sprouted its own subdiscipline known as connectomics [8–10]. Within connectomics, one distinguishes between structural connectivity and functional connectivity. Structural connectivity is concerned with the anatomical white-matter fiber bundles that connect remote regions of the brain. It may be estimated in vivo by diffusion weighted MRI (dMRI), which measures the fractional anisotropy of the diffusion of water molecules [11]. Functional connectivity in turn expresses the (degree of) dependency between the neuronal activity of separate brain regions [6, 12] and is typically measured non-invasively via either functional MRI, electro- or magnetoencephalography (fMRI, EEG and MEG, respectively) [13].
Several measures to quantify (the degree of) functional coupling exist [14, 15], of which the most prevalent is covariance. When the activity signal is normalized to have zero mean and unit variance, covariance coincides with Pearson correlation. As the correlation matrix is easy to compute, it has become the de facto standard in operationalizing functional connectivity. It does however have an important drawback: it is unable to differentiate between direct and indirect effects. For example, if regions A and B are correlated, and similarly B and C show correlation, then correlation between A and C is induced [16, 17]. This poses a problem for functional connectomics, as it introduces type 1 errors. The problem may be remedied to some extent by using partial correlations instead. Its interpretation is similar to Pearson correlation, but it captures only direct effects as the influence from other regions is partialled out. In practical terms, the matrix of partial correlations may be obtained by taking the inverse of the covariance matrix, known as the precision matrix, and rescaling this. Assuming the data are normally distributed, both the precision matrix and the partial correlation matrix capture the conditional independence structure of the considered variables, i.e. when two regions are conditionally independent given all other regions, their precision and partial correlation are zero.
Ideally, the partial correlation matrix would correctly reflect the functional connectivity that generated the observed data. If this matrix is sparse, the corresponding conditional independence graph provides an intuitive representation of the interaction between different regions. In practice however, the obtained partial correlation matrices are not sparse, which makes the estimated connectivity more difficult to interpret. In addition, if the number of samples is small and the number of regions large, there is no unique inverse of the covariance matrix and consequently no unique matrix of partial correlations. Even when these conditions are met, the maximum likelihood solution is often ill-behaved, in which case the solution must be regularized [18]. A popular approximation of the precision matrix is acquired via the graphical LASSO (Least Absolute Shrinkage and Selection Operator), which regularizes the elements of the precision matrix using the ℓ1-norm [14, 16, 19]. This approach shrinks the partial correlations towards zero so as to create sparse solutions, which are easier to interpret. Although the graphical LASSO was found to be one of the must accurate methods in identifying connectivity in a comparative study [14], it introduces a bias that underestimates functional connectivity, thus creating type 2 errors [20]. In addition, both the original maximum likelihood solution as well as the LASSO estimate provide point estimates that do not quantify the reliability of their outcome. In earlier work, we have proposed a Bayesian alternative to the graphical LASSO that uses the G-Wishart distribution to restrict the partial correlation estimates to a previously defined conditional independence graph. We showed that structural connectivity provides an elegant candidate for this graph, and that this approach was able to outperform the graphical LASSO on simulated data [20]. Importantly however, we assumed that the conditional independence graph was available a priori. In the current contribution we take this line of reasoning a critical step forwards and learn both functional connectivity as well as its conditional independence structure simultaneously. Apart from estimating the degree to which two regions have correlated activity, we can now also express the probability of these regions being conditionally independent. As we will show, this results in a more effective approach to regularization than the graphical LASSO, while retaining the additional benefits of the Bayesian framework.
At the foundation of this contribution lies a probabilistic generative model that describes how a particular independence structure generates partial correlations that in turn generate observable data. Using a neurologically plausible simulation with several different conditions, as described by Smith et al. [14], we show that in many cases our Gaussian graphical model approach is favorable to both the maximum likelihood alternative and graphical LASSO regularized solutions. Subsequently, we apply the model to estimate functional connectivity between bilateral accumbens, amygdala, caudate, hippocampus, pallidum, putamen and thalamus using their blood-oxygenation level dependent (BOLD) signal time courses, measured using resting-state fMRI. Finally, we demonstrate how the advantages of a Bayesian approach can be put to practice by showing two extensions to our connectivity model. First, we show how the problem of data fusion for connectivity studies [21, 22] may be tackled by simply providing multiple likelihood terms; one for each imaging modality. This is demonstrated empirically by combining the fMRI time series with dMRI probabilistic tractography results. Second, we describe how further background knowledge on putative connections may be used to both constrain and inform functional connectivity.
From a methodological perspective, elucidating functional connectivity is often rephrased as a covariance selection problem. This boils down to finding a sparse partial correlation matrix associated with the time series (activity) of a set of variables (brain regions), a problem known as covariance selection. Here, the problem is approached using a Gaussian graphical model (GGM), where we assume that the data X = (x1, …, xn)T consist of n independent draws from a p-dimensional multivariate Gaussian distribution N ( 0 , K - 1 ), with zero mean and precision (inverse covariance) matrix K. Here, K ∈ P p, with P p the space of positive definite p × p matrices. The likelihood of K is given by
P ( X ∣ K ) = ∏ i = 1 n N ( x i ∣ 0 , K - 1 ) ∝ | K | n / 2 exp [ - 1 2 ⟨ K , Σ ⟩ ] , (1)
where Σ = XT X and ⟨⋅, ⋅⟩ the trace inner product operator. The assumption of Gaussianity is justified empirically, as BOLD data has been shown to follow a Gaussian distribution [23].
The precision matrix has the important property that zero elements correspond to conditional independencies, provided the data are normally distributed. In other words, Eq (1) specifies a Gaussian Markov random field with respect to a graph G = (V, E), with V = {1, …, p} and E ⊂ V × V, in which the absence of a connection indicates conditional independence, i.e. (i, j) ∉ E → kij = 0 [24, 25].
In order to estimate the precision matrix K of a zero-mean multivariate Gaussian density from data X one may maximize the log-likelihood which gives the maximum likelihood estimate (MLE):
K ^ = arg max K ∈ P p ( log | K | - ⟨ Σ K ⟩ ) (2)
where the maximization is constrained to precision matrices in the family of p × p positive definite matrices P p. If Σ is positive-definite, there exists a unique solution to Eq (2) in the form of Σ−1. However, if the number of samples is small compared to the number of variables, the solution does not exist, and even if n > p, the maximum likelihood estimate is often ill-behaved and requires regularization [18]. A frequently used method of regularization is called the graphical LASSO [26], which penalizes the magnitude of the elements of K. The LASSO approach gives the following MLE:
K ^ = arg max K ∈ P p [ log | K | - ⟨ Σ K ⟩ - λ ∥ K ∥ 1 ] , (3)
in which the shrinkage parameter λ determines the amount of penalization that is applied. Several studies have applied the graphical LASSO in order to estimate functional connectivity [14, 16, 19]. Alternative regularization schemes are available [27], such as ridge regression or elastic net [28], but we will not consider these methods in detail here. Rather, we emphasize that each of these regularization approaches provides only a point estimate, instead of a posterior distribution over K. This makes it impossible to quantify the uncertainty associated with the estimate, which can lead to incorrect conclusions about functional connectivity in light of finite data. Moreover, it has been shown that the graphical LASSO is not guaranteed to find the true graph even in the limit of infinite data [29]. In addition, solutions obtained through regularization tend to underestimate functional connectivity [20].
Recently, extensions of the (graphical) LASSO approach have been proposed that allow for statistical inference. For example, [30] introduce a significance test that can be applied to LASSO estimates while [31, 32] describe a desparsified LASSO that attempts to de-bias the results using a projection onto the residual space. However, these approaches make assumptions on the sparsity of K, which may not be warranted.
Alternatively, a Bayesian approach can be applied to the covariance selection problem, which dispenses with these assumptions. It requires that we specify a prior distribution on K. As we hope to identify conditional independencies between the considered variables, a convenient prior distribution arises in the form of the G-Wishart distribution [33]:
P ( K ∣ G , δ , D ) = W G ( δ , D ) = | K | ( δ - 2 ) / 2 Z G ( δ , D ) exp [ - 1 2 ⟨ K , D ⟩ ] 1 K ∈ P G , (4)
in which P G is the space of positive definite p × p matrices that have zero elements wherever (i, j) ∉ G, δ is the degrees of freedom parameter, D is the prior scaling matrix and 1 x evaluates to 1 if and only if x holds and to 0 otherwise. The G-Wishart distribution is conjugate to the multivariate Gaussian likelihood in Eq (1), so that
P ( K ∣ G , δ , D , X ) = W G ( δ + n , D + Σ ) = | K | ( n + δ - 2 ) / 2 Z G ( δ + n , D + Σ ) exp [ - 1 2 ⟨ K , D + Σ ⟩ ] . (5)
Note that the Wishart distribution is a special case of the G-Wishart distribution, with which it coincides if G is a fully connected graph.
It should be pointed out that in the limit of n → ∞, any prior will be fully dominated by the data. In theory, even when the true precision matrix K contains very small elements, the probability of a corresponding edge will go to 1 in the limit of an infinite amount of data. The interesting question is what happens if the magnitude of these elements scales as a function of n, e.g., as 1/n. Where asymptotic analyses have been successfully applied to better understand the behavior of regularization approaches such as the graphical LASSO [34, 35], such analyses of Bayesian procedures are complex and may lead to counterintuitive results [36]. For the G-Wishart prior in particular, similar analyses have, to the best of our knowledge, not yet been pursued.
The preliminaries described above allow us to specify the distribution that is central to this work, i.e. the joint posterior over both the conditional independence graph and the precision matrix (an illustration of the graphical model is provided in Fig 1A):
P ( G , K ∣ X ) ∝ P ( X ∣ K ) P ( K ∣ G ) P ( G ) . (6)
Note that the necessary hyperparameters are typically omitted for clarity.
In practice, functional connectivity is more intuitively understood in terms of partial correlations than as elements of the precision matrix. The partial correlation matrix R may be obtained from the precision matrix by applying the transformation
r i j = { 1 if i = j , - k i j k i i k j j otherwise. (7)
By transforming each element of K in Eq (6), the distribution P(G, R ∣ X) is constructed. When discussing our experimental results, we will focus on partial correlations rather than precision values, unless explicitly stated otherwise. Note that the relation between the dependency structure G and the precision matrix K, as discussed above, also holds between G and the partial correlations R. That is, absence of a connection in (i, j) ∈ G implies rij = 0.
The Bayesian generative model must be completed by specifying a prior distribution to draw G from. Here, we assume that a priori all edges are marginally independent and each have probability θ. That is, we have
P ( G ∣ Θ ) = ∏ i < j θ i j g i j ( 1 - θ i j ) 1 - g i j , (8)
with gij ∈ {0, 1}, gij = 1 ↔ (i, j) ∈ G and Θ = (θij)i < j. Initially we use θij = 0.5 ∀i, j to indicate that we have no a priori preference for either a dependence or an independence. The impact of different values for θij on the posterior estimates is discussed in S3 Text, where it is shown that the prior is to a large extent dominated by the likelihood.
One of the benefits of the Bayesian framework is that extensions to the generative model are straightforward to implement. In this section we use the distribution given in Eq (6) to provide two illustrations of such extensions for analyzing connectivity.
To analyze the performance of the Gaussian graphical model approach to functional connectivity, we compare our results to those presented in [14]. Here, realistic BOLD time series are generated according to the dynamic causal modeling (DCM) fMRI forward model [42], that makes use of the nonlinear balloon model [43], based on a known constructed network as its starting point. In total, 28 simulations with different parameters such as number of nodes, number of generated samples, sampling frequency and noise levels were constructed. For each simulation, 50 different time series are generated, simulating different ‘subjects’ (throughout we will refer to these pseudo-subjects as ‘runs’, to avoid confusion with the empirical data later on). The networks in the simulations were composed of 5, 10, 15 or 50 nodes and for each node between 50 and 10 000 samples were generated. For 15 of the 28 simulations, additional characteristics were introduced, such as shared input between a number of nodes, or mixing in timeseries between nodes (mimicking the effect of bad ROI definition) [14]. For the full description of the approach as well as the additional simulation parameters, we refer to the original description in [14] as well as the corresponding web page where the simulation may be downloaded (http://www.fmrib.ox.ac.uk/analysis/netsim/). In the simulation study, it was shown that using partial correlation (both maximum likelihood as well as LASSO regularized point estimates) resulted in the best (undirected) reconstructions of the ground truth. As these methods performed best, and are closely related to our approach, we use these to compare our results with.
The evaluation procedure is as follows: For each run of each of the 28 different simulations, the time series X of that run are used to compute P(G, R ∣ X). In addition, for each run the maximum likelihood estimate (MLE) is computed, as well as the graphical LASSO regularized point estimate using the same regularization as in [14] (i.e. λ ∈ {5, 100}). The quality of the reconstruction of the ground truth is quantified in three ways. Let R* be the ground truth functional connectivity that we are trying to recover and let T be a matrix that has 1 in its elements whenever the corresponding edge is present in the ground truth network, and 0 otherwise (ignoring directionality). Then Γ = ∣R* − R∣ gives the reconstruction error (where R is either a sample from P(G, R), or a point estimate). The total reconstruction error is η ( Γ ) = 2 p ( p - 1 ) ∑ i < j γ i j, the true positive error is η tp ( Γ ) = 1 N tp ∑ i < j γ i j δ t i j ≠ 0, where Ntp is the number of nonzero elements in the ground truth R*, i.e. the number of true present connections, and finally the true negative error is given by η tn ( Γ ) = 1 N tn ∑ i < j γ i j δ t i j = 0, where Ntn is the number of zero elements in the ground truth R*, i.e. the number of true absent connections. The indicator function δx evaluates to 1 if and only if its argument x holds true, and to 0 otherwise.
In [14], a null distribution is computed for each of the different methods, by randomly permuting the node labels in the different runs (to remove any influence between the different nodes), which is subsequently used to derive a z-score for an error measure similar to η. However, in the case of Bayesian functional connectivity, a distribution characterizing the uncertainty of the results is already available in the form of P(G, R). By applying η to each of the samples of this distribution, we obtain P(η). The standardized scores of a point estimate R relative to the BGGM distribution may be computed as z(R) = (η(R*, R) − μ)/σ, in which μ and σ are the mean and standard deviation of the distribution, respectively. The procedure is illustrated in Fig 2.
The Bayesian formulation of the model allows us to describe and compare the shapes of the different posterior distributions. We compute the entropy of the posterior distributions as
H = - ∑ G [ P ( G , K ∣ X ) log 2P ( G , K ∣ X ) ] , (13)
to indicate the diversity of models that have been encountered in the Markov chains. In addition, the posterior probability of the maximum a posteriori sample is derived, i.e.
P ( G ^ , K ∣ X ) = max G P ( G , K ∣ X ) , (14)
to quantify how much of the posterior distribution is dominated by its mode.
The Markov chain Monte Carlo (MCMC) scheme as described in S1 Text was used to approximate the posterior distributions of interest for each subject using either the simulated BOLD signal time series, the BOLD time series data for the fourteen subcortical regions (see Eq (6)), the combination of time series data and tractography output for the subcortical regions (see Eq (11)) or finally the BOLD time series data in combination with the informed prior. Throughout, a vague prior on the precision is used: P ( K ∣ G ) = W G ( 3 , I p ), cf. [44]. The parameters of the probabilistic streamline model are set to (α, β) = (1, 0.5), which expresses that high streamline counts are most likely associated with a structural connection, while still allowing for tractography noise [40]. Once convergence was established, the approximated distributions were uniformly thinned to T = 1 000 samples, to make subsequent analyses more manageable and to have an equal number of samples for all different settings. Details of convergence monitoring and computation speed are provided in S2 Text.
Below we discuss both the simulation results as well as the connectivity estimates obtained on empirical data. For readability, we refer to the probability of conditional dependence as ‘connection probability’ and to a pair of regions that are conditionally independent or not-independent, conditioned on all other variables, simply as an ‘independent’ (or ‘disconnected’) or ‘dependent’ (or ‘connected’) region pair, respectively.
Fig 3 shows the (smoothed) histograms of z-scores aggregated over the 50 runs per simulation, for the graphical LASSO approach with λ = 100 (the results for λ = 5 and the MLE are almost identical; the MLE results are shown in S1 Fig). In this figure, distributions of errors with high z-scores have substantially larger errors than the errors from the BGGM approach, while distributions with low z-scores have smaller errors. The significance threshold at p < 0.01 is indicated by the red dotted lines. The first row of Fig 3 shows the total scores (both true positives and true negatives) for each simulation, while the second and the third row split this score into the contributions for true positive connections and true negative connections, respectively. These results indicate that in terms of true positives, the LASSO approach typically has an equal to slightly better performance than our Bayesian alternative. However, the BGGM approach identifies true negatives at least as well as G-LASSO, and in several cases significantly outperforms it. On the whole, the proposed method is up to par with the graphical LASSO (for λ ∈ {5, 100}) and the MLE, while at times outperforming them greatly.
To obtain insight in the behavior that creates these results, we take a closer look at some of the simulation results. As an example, Fig 4A shows the ground truth network and the reconstruction by the graphical LASSO, as well as the expectation (i.e. posterior mean of the samples) using the BGGM approach. In addition, the figure shows for three different connections the estimated partial correlation in detail. The first, between nodes 1 and 5, is present in the ground truth network. Our approximation is (correctly) confident that this node pair is not independent, and assigns a posterior partial correlation distribution close to the ground truth. The graphical LASSO estimate is slightly closer to the ground truth than the mode of the distribution. For the second node pair, between nodes 3 and 5, a connection should be absent, but because of the limited number of data samples the signals of these nodes have become correlated. This time, the BGGM approach shows a bimodal distribution. The first mode is centered close to the graphical LASSO estimate, but the second mode is at zero, as there is non-negligible evidence for this pair of nodes being disconnected. This means that on the whole (i.e. the entire distribution), the BGGM approach correctly estimates this connection strength lower than the graphical LASSO. A similar observation can be made for the third node pair, between nodes 1 and 4, of which the BGGM estimate is fairly certain about their independence. Because of this, most of the partial correlation mass is at zero, rather than at the value indicated by the graphical LASSO estimate.
These results beg the question: what if we regularize the graphical LASSO even more? Although Smith et al. report no further improvement after λ = 100 [14], it is possible that more regularization brings the graphical LASSO estimate closer to the BGGM results. In Fig 4B, the same visualization is provided, but this time for λ = 10 000. This time, we see that indeed the graphical LASSO estimate is closer to the BGGM expectation than before. In particular for the connection between nodes 1 and 4, the graphical LASSO now correctly estimates the absence of this connection. However, for the connection between nodes 3 and 5, the results hardly change, which means that the BGGM estimate is closer to the ground truth still, as, conditioned on the absent connection, the estimated partial correlation is zero. Finally, for the true positive connection between nodes 1 and 5, we see that the strong regularization causes the graphical LASSO to underestimate the connection, which will only become worse when we increment λ even further.
These results may similarly be interpreted in terms of the (in)dependence graph. For weak regularization, the graphical lasso suggests false positives due to limited data. For more regularization, the same dependency structure is recovered as using (the mean of) the BGGM approach (see for example Fig 4B). Regularizing even stronger introduces false negatives. Note that these results follow from the results of the recovered partial correlation structures and are therefore not explicitly presented here.
In addition, we applied the extended BIC over the ‘graphical LASSO path’ (i.e. we applied the EBIC penalty to the graphical lasso estimates over a logarithmic range of λ, with the maximum penalty corresponding to the empty graph, as used in [54]) to a number of simulations. However, this analysis did not result in a λ will results significantly different than those already presented here, and has been omitted here.
The pattern of simulations in which the BGGM outperforms the graphical LASSO is not random. In [14], each of the simulations is based on a network consisting of 5 nodes, except for simulations 2, 3, 4, 6, 11, 12 and 17, which consist of networks of 10, 15, 50, 10, 10 and 10 nodes, respectively. Precisely these simulations benefit the most from the BGGM approach, as can be seen in Fig 3. As for these simulations the ratio N/p is smallest, it is here that the most improvement can be obtained from regularization, e.g. by the graphical LASSO [14]. As we have shown above, the BGGM provides further improvement still, because this approach conditions on conditional independencies.
We further analyzed the effect of sample size on recovery of the ground truth by taking the simulation with the most available samples (simulation 7 in [14]) and attempting to recover the ground truth using increasingly smaller subsets of the samples. We compared the BGGM results with the graphical LASSO with λ ∈ {5, 100, 1 000, 10 000}. The outcome of this experiment is shown in Fig 5, once again split into the total error, error in recovery of true positives and error in recovery of true negatives. The results indicate that for small sample size, the BGGM approach already outperforms the graphical LASSO in total error, although the differences become more pronounced as more samples are considered. Extremely strong regularization (i.e. λ = 10 000) does result in better estimation of absent connections (by simply forcing almost all connections to zero), but this comes at the cost of excluding connections that should be present. For weak regularization (i.e. λ = 5), small sample size appears to be somewhat beneficial in recovery of true positive connections, as here the performance of the graphical LASSO is similar to our approach. However, this effect diminishes as more samples are acquired (inducing more spurious connections). In terms of true negatives, weak regularization is clearly outperformed by the BGGM approach.
In addition, we analyzed the effect of small sample sizes on the estimates. We used simulation 3 (with p = 15) and repeated the procedure as before, but this time the number of samples was varied n ∈ {5, 10, …, 45, 50}, so that situations of n < p were included. The results of this experiment are shown in Fig 6. They show that, unsurprisingly, weak regularization (i.e. λ = 5) is insufficient to recover the ground truth when few samples are available. Strong shrinkage (i.e. λ = 10 000) results in a low recovery error, but this comes at the expense of significantly underestimating true positive connections. In general, the BGGM approach performs approximately equal to the graphical LASSO for small to moderate regularization, given this limited sample size scenario.
Below we discuss the connectivity estimates we obtained on the empirical data, for the original BGGM model, the data fusion variant and the effect of incorporating background information.
Functional connectivity may be quantified using different metrics. The most obvious approach is to use Pearson correlation, but this metric is sensitive to polysynaptic influences. An alternative that does not suffer from this drawback is partial correlation, which was further advocated for its ability to retrieve true connections and its capacity to deal with noise [14]. Partial correlation between two variables may be interpreted as Pearson correlation conditioned on all other variables. In practice, partial correlation can be computed by applying a simple transformation to the precision matrix of a multivariate Gaussian distribution. The precision matrix and, consequently, the matrix of partial correlations, has the interesting property that conditional independence between variables, given all other variables, appears as a zero value in the corresponding matrix element [20, 55], which may conveniently be collected in a conditional independence graph. Typically, this graph is mostly ignored, while the precision or partial correlation matrix is considered the quantity of interest. In this paper, we have provided a Bayesian generative model for functional connectivity in which the conditional independence graph plays a central role, as it is assumed to generate the precision matrix and thus functional connectivity. As opposed to regularized maximum likelihood estimates for the precision matrix, our approach characterizes the full posterior distribution of both conditional (in)dependencies and partial correlations. In addition to this model, we described a number of model variants that address specific issues with, and conceptual extensions to, connectivity.
We subjected our approach to the simulations that were presented in [14], and compared its performance to the maximum likelihood estimate as well as to the graphical LASSO. The latter of these two has been shown to be the most successful in recovering connectivity in these simulations [14]. The results of the simulation are encouraging. Although we observe that for true positive connections, our approach occasionally underestimates connections, it more than compensates for this in correctly estimating true negatives (i.e. the sparsity structure of the network). When true positives and true negatives are both taken into account, corrected for their respective numbers of occurrence, we find that our approach performs at least as well as the graphical LASSO, and significantly better for simulations with small sample size compared to the number of nodes in the network. A closer look at these results shows that when estimating partial correlations, conditioning on the presence or absence of a connection provides a considerable advantage over shrinkage. In particular for connections with a moderate probability of independence our method yields a bimodal distribution of partial correlations, differentiating between the conditionally dependent and independent node pairs.
In addition to our simulation results, we used our approach to approximate the posterior distribution of functional connectivity between subcortical areas for twenty participants. This allowed us to identify a connectivity backbone that consists of strong connections and partial correlations. At the same time, we see that a number of connections are strongly dependent, but foster only weak partial correlations. This emphasizes that a richer picture of connectivity is obtained by looking at both the structure of conditional independence, as well as the strength of these connections in terms of partial correlation.
Partial-correlation based methods are susceptible to common input effects that may induce spurious connections if they are not accounted for, for example when variables (i.e. brain regions) are missing [56, 57] from the analysis. If instead the full neural system is observed, it is straightforward that direct functional connections presuppose anatomical connections between the corresponding regions. This allows us to combine the generative model for functional connectivity with a similar model for structural connectivity [39] using probabilistic tractography obtained from diffusion weighted MRI. Conceptually, this results in a data fusion model in which an underlying model of anatomy drives both the observations for functional interactions, as well as for estimates of structural fibres. Compared to alternatives that, for example, weigh a regularization parameter by the strength of structural connectivity [22, 38, 58–61], our approach is based on a generative model in which data fusion is made possible by the use of different likelihood terms. Furthermore, in our model both sources of data affect both types of connectivity; structural connectivity regularizes functional connectivity and simultaneously functional dependencies influence the probability of structural connections. On empirical data the data fusion approach leads to sparser connectivity, in particular between hemispheres. However, some connections are conditionally dependent to such a degree that the model infers a connection regardless of the lack of support by the tractography data. This is helpful in estimating structural connectivity, as it is well known that structural connectivity based on diffusion weighted imaging suffers from a large number of false negatives [62]. In addition, data fusion lowers the variance for many of the partial correlations, indicating that combining both imaging modalities leads to more robust estimates [49, 62–64]. However, for a number of connections the data for functional and structural connectivity appear to contradict each other, which actually results in increased variance. Note that our data fusion approach has similarities to linked ICA [65], which also uses a Bayesian generative model to integrate different data modalities. However, whereas linked ICA assumes that each data modality may be decomposed into a number of (shared) components, our model assumes that anatomical connectivity is the variable that is shared across modalities.
Our final model variant uses an informative prior which encodes the assumption that between-hemisphere connections are restricted to those between functionally homologous regions (cf. for example [66]). This is only one of many prior distributions that, depending on the research question and available background information, may be used to inform the connectivity estimates. As expected, the prior removes the negative partial correlations that are visible for contralateral connections in the other model variants. Indirectly, the prior also affects the partial correlations within hemispheres, as they become slightly lower in magnitude across the board. These results touch upon an unresolved issue in connectomics concerning the interpretation of negative (partial) correlations. It has been suggested that a substantial number of negative partial correlations are due to global signal regression and are therefore artifactual in nature rather than biological [67–69]. On the other hand, it has been shown that even without global signal regression, negative connections exist and these may even have biological meaning [70]. Although it is outside the scope of this paper to resolve this matter, we have shown that an informed prior may be used to encode such assumptions or correct for biases.
As our approach is Bayesian it directly allows for statistical inference, so that the uncertainty associated with our estimates may explicitly be quantified. In terms of a binary graph that indicates conditional (in)dependence, this expresses itself by providing an expectation of a connection rather than a point estimate. For partial correlations, the approach provides the supported distribution instead of a single value. These posterior distribution shapes reveal that none of the model variants are dominated by their mode. In particular for the original model the distributions are very broad and contain many unique models. Although a number of connections is consistently present, the conditional independence graphs vary substantially across subjects. In contrast, the data fusion approach and the informed prior result in distributions that are more tightly centered around the maximum a posteriori connectivity, yet even here there remains substantial support for alternative models. This has important implications for connectomics studies. These are typically aimed at obtaining a point estimate (which can often be interpreted as the mode of an implicit posterior distribution), so a substantial number of connections with significant support from the data will be excluded and spurious connections will be suggested. The widths of the posterior distributions strongly advocate a Bayesian approach, or at the very least point-estimated connectivity results should be treated with great care, e.g. by applying a bootstrapping procedure [71].
The main limitation of our study is one of scale. Bayesian inference has the drawback of being computationally demanding in approximating the posterior distributions, and although state-of-the-art machinery has been applied to make this process efficient, it remains impossible to apply the same methods to a large number of variables. Applying the models to large-scale data sets requires either more efficient implementations, e.g. by using GPU programming, or additional efficiency gains in the field of Gaussian graphical models.
Finally, a fundamental assumption in Gaussian graphical model estimation is that the functional data are normally distributed. Should this assumption fail, it may prove difficult to interpret the estimated connectivity. However, as discussed by [23], BOLD time series do tend to be mostly Gaussian.
The most pressing issue for future work is, as mentioned above, improving the methodology to handle a larger number of variables. However, a number of interesting research questions may be addressed even with a limited number of regions. For example, a model may be constructed that defines the BOLD time series to be generated by a mixture of partial correlation matrices, instead of a single one. By applying appropriate constraints, such as that consecutive datapoints are likely to be generated by the same connectivity matrix, this setup can be applied to differentiate experimental conditions based on their connectivity distributions [72]. Similarly, subjects may be assigned to either patients or healthy controls by defining a shared conditional independence graph for either group.
The data fusion approach may be extended to incorporate any number of imaging modalities, provided that a forward model can be constructed that shares at least one variable with the other modalities. For example, structural connectivity may inform functional connectivity estimated from MEG instead of or in addition to fMRI data [59].
Additional information may also be incorporated into the prior. This may be explicit evidence for (the absence of) a connection, e.g. tracer studies that reveal the presence of a fiber bundle can make particular connections more likely or, conversely, knowledge about white-matter lesions may preclude connections. Alternatively one could construct a prior in which the probability of a connection is a function of the distance between the corresponding end points.
In conclusion, the proposed Bayesian approach to functional connectivity has demonstrated that connectivity may be meaningfully divided into structure and strength. Several model variants have been discussed, each with their own characteristics. Application of the models has shown convincingly that multiple unique structures are possible given the same data. This illustrates the advantages of a Bayesian approach to connectivity, and provides a word of caution for traditional (regularized) maximum likelihood estimators.
|
10.1371/journal.ppat.1003194 | Environmental Predictors of Seasonal Influenza Epidemics across Temperate and Tropical Climates | Human influenza infections exhibit a strong seasonal cycle in temperate regions. Recent laboratory and epidemiological evidence suggests that low specific humidity conditions facilitate the airborne survival and transmission of the influenza virus in temperate regions, resulting in annual winter epidemics. However, this relationship is unlikely to account for the epidemiology of influenza in tropical and subtropical regions where epidemics often occur during the rainy season or transmit year-round without a well-defined season. We assessed the role of specific humidity and other local climatic variables on influenza virus seasonality by modeling epidemiological and climatic information from 78 study sites sampled globally. We substantiated that there are two types of environmental conditions associated with seasonal influenza epidemics: “cold-dry” and “humid-rainy”. For sites where monthly average specific humidity or temperature decreases below thresholds of approximately 11–12 g/kg and 18–21°C during the year, influenza activity peaks during the cold-dry season (i.e., winter) when specific humidity and temperature are at minimal levels. For sites where specific humidity and temperature do not decrease below these thresholds, seasonal influenza activity is more likely to peak in months when average precipitation totals are maximal and greater than 150 mm per month. These findings provide a simple climate-based model rooted in empirical data that accounts for the diversity of seasonal influenza patterns observed across temperate, subtropical and tropical climates.
| Human influenza infections have a pronounced seasonal cycle in temperate regions. Recent laboratory and epidemiological evidence suggests that low humidity conditions in the winter may increase virus survival and enable the virus to transmit efficiently between hosts. However, seasonal influenza activity in some tropical locations occurs during the rainy season, whereas other tropical locations do not experience a well-defined influenza season. The primary goal of this study was to identify the relationship between the timing of seasonal influenza epidemics and climate variability across the globe. We show the importance of thresholds in humidity, temperature and precipitation that affect the epidemiology, and potentially the transmission route, of influenza. A better understanding of the environmental, demographic and behavioral drivers of influenza seasonality is important for optimizing intervention strategies, especially in low and middle-latitude regions.
| Influenza exerts a significant health burden on human populations across temperate, subtropical and tropical regions [1]. The striking seasonal pattern that characterizes influenza in temperate populations has long suggested a causal link between seasonal fluctuations in climatic and social factors and influenza transmission [2]–[4]. Temperate regions of the northern and southern hemispheres are characterized by highly synchronized annual influenza epidemics during their respective winter months [5], [6]. In contrast, influenza seasonal characteristics are more diverse in tropical and subtropical regions; some sites experience annual epidemics coinciding with the local rainy season [7]–[11], whereas others are characterized by semi-annual epidemics or year-round influenza activity without well-defined influenza seasons [7], [12], [13].
Recent epidemiological studies indicate that low levels of specific humidity are associated with the onset of pandemic and epidemic influenza in the US [14], [15], consistent with laboratory experiments and animal models suggesting that low specific humidity favors virus survival and aerosol transmission [16]–[17]. There are several alternative explanations for the winter seasonal transmission of influenza in temperate regions, including the inhibition of host immune function due to decreased exposure to solar radiation [18], [19], and the inhibition of mucociliary clearance by the inhalation of cold-dry air [20]. Person-to-person contact rates may also strengthen in the winter due to increased indoor crowding, modulated by school terms [21]. There are few biological explanations for the association between precipitations and influenza activity reported in some tropical and subtropical regions, although rainy conditions may also favor indoor crowding [4].
Although it is common for epidemiological studies to examine relationships between seasonal influenza activity and climatic factors for individual sites, few studies have assessed the consistency of these relationships across a broad range of temperate, subtropical and tropical sites. A recent study evidenced a link between influenza and temperature based on aggregate country-level data, but did not characterize the geographical and climatic boundaries that define regions experiencing different influenza seasonality patterns [22]. Here we investigate both relative and absolute associations between climatic factors and the timing of seasonal influenza epidemics for 78 individual sites sampled globally [23]. We develop models that predict the month of peak influenza activity for each study site as a function of climatic variables and identify climatic thresholds accounting for the diversity of influenza seasonality patterns observed globally.
Across the 78 sites, influenza peaks generally coincided with months of lower temperature, lower solar radiation and lower specific humidity than expected under the null hypothesis (mean rank = 4.3 [95% CI: 3.7, 5.0] for temperature, 4.5 [95% CI: 3.9, 5.1] for solar radiation and 4.8 [95% CI: 4.0, 5.6] for specific humidity). In contrast, relative humidity and precipitation did not significantly deviate from the null value (mean rank not significantly different than 6.5). The association between influenza peaks, temperature, solar radiation, relative humidity and specific humidity was most significant when the influenza peaks lagged 1-month behind the environmental predictors. We obtained similar results when the analysis was restricted to primary influenza peaks.
A similar analysis performed with a sliding geographical window revealed that the association between influenza peaks and climatic variables varied with latitude (Figure 2). The strongest association was found at high latitudes poleward of 25°N/S, with influenza peaks preferentially occurring in months with the lowest temperature, solar radiation and specific humidity. Influenza peaks occurred in the months with the highest levels of relative humidity and lowest levels of precipitation poleward of approximately 40°N/S. Primary influenza peaks equatorward of 10°N/S corresponded to the months with the highest annual levels of specific humidity and precipitation (p<0.05); whereas there was no association with temperature, solar radiation and relative humidity. In middle latitudes ranging between 12.5–25°N/S, there was no significant association between influenza peaks and climatic variables.
Temperature and specific humidity were the best individual predictors of influenza peaks. The model fits improved slightly when influenza peaks lagged 1-month behind these predictors, and accurately predicted 56–66% of the peaks in the global datasets, with highest accuracy at latitudes poleward of 25°N/S (Tables 1 and 2). The modeled relationship between specific humidity and all influenza peaks was U-shaped, with lowest probability of an influenza peak at 12 g/kg of specific humidity and increasing probabilities at lower and higher values (Figure 3). The analysis restricted to primary peaks revealed a similar relationship, with a minimum influenza probability at 11 g/kg. Unlike specific humidity the modeled relationship between temperature and influenza peaks was monotonic, with the greatest probability of a peak corresponding to low temperatures. Although the specific humidity and temperature models were the best predictors of the timing of influenza peaks across all sites, they were not significant predictors of influenza peaks equatorward of 25°N/S.
There was a strong inverse relationship between solar radiation and the probability of an influenza peak, especially when influenza peaks were lagged by 1-month. The solar radiation model outperformed the temperature and specific humidity models based on AIC, but it was not as strong a predictor of the timing of the influenza peaks (Tables 1 and 2).
Precipitation was a weak predictor of influenza peaks overall, but it was a strong predictor of influenza peaks equatorward of 12.5°N/S, particularly for primary influenza peaks (p<0.01) (Tables 1 and 2). Unlike the other climate variables, precipitation-based models performed slightly better when no lag was considered between precipitation and influenza activity.
Relative humidity was a strong predictor of influenza peaks globally, particularly when a 1-month lag was applied to the influenza peaks. There was a positive association between relative humidity and influenza peaks in high and low latitudes, but this model was a poor predictor of influenza peaks in middle latitudes. Further, the relative humidity model was not as strong as the specific humidity, solar radiation and temperature models (Tables 1 and 2).
Overall, the multivariate models most predictive of influenza peak timing included combinations of temperature and precipitation (all peaks, Table 1), and temperature and specific humidity (primary peaks, Table 2). These models accurately predicted peak influenza months in 78% and 89% of the 9 independent sites selected from the FluNet database, respectively (Figure 4). Further, the models predicted a nearly uniform probability of influenza peaks every month of the year in equatorial Colombia, a location that experiences minimal seasonal climatic fluctuation (Figure 4). Taken together, this analysis exploring the shape of the relationship between climatic variables and influenza highlights the covariability between specific humidity and temperature, and the significant predictive power of these variables at high latitudes. In contrast, precipitation and relative humidity were predictive of influenza peaks at low latitudes. Interaction terms describing monthly deviations from the annual average of each environmental predictor marginally improved some models but did not affect the main conclusions (Tables S1 and S2 in Text S1).
To further characterize the distribution of influenza peaks globally and identify the geographical and climatic boundaries defining influenza seasonality patterns, we categorized sites based on whether influenza epidemics occurred in months with low (cold-dry season) or high (humid-rainy season) levels of specific humidity relative to the local climatology (Figure 5). We found that the annual minimum level of specific humidity in a site was predictive of the seasonal characteristics of influenza activity locally. Sites characterized by cold-dry influenza peaks generally experienced annual minimum specific humidity values less than 12 g/kg when all influenza peaks were considered, and approximately 11 g/kg when the analysis was restricted to primary influenza peaks (Figure 5). The minimum specific humidity models were statistically significant, and classified 75% of the 96 total influenza peaks correctly (p<0.001, Figure 5C), and 82% of the 76 primary peaks correctly (p<0.001). Annual minimum temperature was a slightly better predictor of the type of influenza peak characterizing a site, correctly classifying 77% and 87% of all peaks and primary peaks, respectively (p<0.001). Sites characterized by cold-dry influenza peaks generally had annual minimum temperature values less than 21°C when all influenza peaks were considered, and approximately 18°C when the analysis was restricted to primary influenza peaks. The temperature and specific humidity models differentiated between the 6 cold-dry and 3 humid-rainy influenza peaks available in the independent FluNet sites with 100% and 78% accuracy, respectively.
Annual minimum solar radiation was a significant predictor of the type of influenza peaks, comparable to temperature and specific humidity, correctly classifying 71% and 79%of all peaks and primary peaks, respectively (p<0.001). Annual minimum relative humidity and annual monthly maximum precipitation were also significant predictors, but the models were significantly weaker than the other models. It should be noted that 12 sites had both a cold-dry and humid-rainy influenza peak, assuring that one peak would be classified incorrectly when all influenza peaks were considered.
Taken together, this analysis indicates that thresholds in specific humidity and temperature, and perhaps solar radiation, are associated with the timing of the influenza season and the occurrence of influenza activity in the dry-cold or humid-rainy months of the year. The specific humidity and temperature models were then used to predict the expected seasonal characteristics of influenza globally (Figure 5C–D). Both models suggest that seasonal influenza activity coincides with the humid-rainy season in large areas of Central and South America, and Southern Asia; while predictions were more uncertain in middle latitudes and there were inconsistencies between the two models for parts of Central Africa. In particular, the model driven by minimum temperature predicted the occurrence of humid-rainy influenza peaks in most of Central Africa, while the model driven by minimum specific humidity predicted a more restricted zone of humid-rainy peaks concentrated on the Western coast of this region. These discrepancies can be explained by a combination of warm year-round temperatures in this area, with low specific humidity values in parts of the year.
We explored the association between influenza seasonality and climate in a representative sample of 78 global sites, spanning an absolute latitudinal range between 1° and 60°. Our analyses revealed two distinct types of climatic conditions associated with influenza seasons observed globally: “cold-dry” and “humid-rainy”. In general, sites that experienced low levels of specific humidity and temperature (less than 11–12 g/kg and 18–21°C) for at least one month during the year were characterized by seasonal influenza activity during the months with minimal levels of specific humidity and temperature. In contrast, sites that maintained high levels of specific humidity and temperature were generally characterized by influenza epidemics during the most humid and rainy months of the year. The predictions of our climate-based models compared favorably to influenza epidemiological information collected independently of the dataset used for the model-building exercise.
The bimodal nature of the relationship, in both relative and absolute terms, between specific humidity and influenza peaks, and its strong relationship to other climate variables such as temperature and precipitation, makes specific humidity a useful gauge of the environmental favorability of influenza activity across all latitudes (Figure 5). However, although the specific humidity models were significant predictors of influenza peaks globally, this was primarily due to their performance in high latitudes. In low latitudes, precipitation was a stronger predictor of the timing of influenza activity, with peaks typically occurring in months with average precipitation greater than 150 mm (Figure 3).
Overall, although precipitation was strongly associated with influenza peaks in low latitudes, the timing of influenza peaks in this region was more difficult to predict than in high-latitude sites. Several sites in this region were not characterized by well-defined influenza season; rather, influenza activity was present year-round likely due to the limited seasonal environmental variation that characterizes much of the region. For example, equatorial sites such as Iquitos, Peru, and Singapore—where influenza seasonality is weak [27], [28]— experience limited fluctuations in precipitation, with monthly averages constrained to a narrow range of 150–300 mm year-round. In contrast, middle and low-latitude sites such as Fortaleza, Brazil and Yangon, Myanmar —which are noted for their well-defined influenza seasons [9], [11]—are characterized by large amplitude range in average monthly precipitation from 25 mm in the dry season to over 300 mm and 600 mm in the rainy season, respectively.
Model performances were particularly poor in a number of middle latitudes sites. Predicting influenza peaks in these sites may be complicated by large seasonal swings in climate that characterize the region, generating both cold-dry and humid-rainy seasons that are equally favorable for seasonal influenza activity, such as in Senegal (Figure 4). For these sites other factors might play a critical role in determining the timing of influenza activity, including population mixing (i.e., travel) with regions that do experience well-defined influenza seasons [29], [30], or the seasonal phasing with school cycles [21]. Moreover, the presence of both cold-dry and humid-rainy seasons could explain the occurrence of semi-annual influenza epidemics in some of these middle-latitude sites. For example, Hong Kong has a primary influenza peak in winter when average monthly specific humidity and temperature are less than 8 g/kg and 17°C, and a secondary influenza peak in summer when average monthly precipitation is near 400 mm.
Temperature was a strong predictor of influenza seasonality in high latitudes, suggesting that cold temperatures may drive seasonal epidemics in these regions. However, previous analyses of laboratory experiments have indicated that specific humidity is a more parsimonious predictor of virus survival and transmission than temperature [17]. Furthermore, individuals in temperate regions spend a vast majority of their time indoors where temperature is managed and does not correlate well with outdoor temperatures. Yet, temperature may affect the timing of influenza epidemics through mechanisms independent of virus survival; for example, low outdoor temperatures may promote indoor crowding, thereby increasing person-to-person contact rates [2]–[4]. It is also possible that even limited exposure to cold outdoor temperatures can have long-lasting physiological effects on hosts that make them more susceptible to infection or affect viral shedding [16]. Additional experimental and observational work is needed to disentangle the contribution of specific humidity and temperature on influenza seasonality; epidemiological information from Central Africa would be particularly useful in this respect as our climate-based predictive models disagreed in this region.
The findings that both cold-dry and humid-rainy conditions are associated with influenza peaks could be used to support the hypothesis that two distinct mechanisms account for influenza seasonality in temperate and tropical climates, perhaps due to changes in the dominant mode of transmission [31]. For example, specific humidity may drive the timing of influenza epidemics in high latitudes by increasing virus survival and enabling aerosol transmission; whereas direct transmission or transmission by fomites may dominate in low-latitude sites where rainy conditions favor indoor crowding. Middle latitudes may be a transition zone where influenza seasons are driven by low specific humidity or high levels of precipitation depending on local climate. Another intriguing possibility is that the relationship between specific humidity and virus survival underlies influenza transmission across all latitudes. For example, a few experimental studies have indicated a U-shaped relationship between relative humidity and influenza virus survival, suggesting a similar relationship for specific humidity given that experiments were held at constant temperature [32]–[34]. Other laboratory studies, however, have indicated that virus survival and transmission increase monotonically as specific humidity decreases [17], [35], [36]. Further, the hypothesis that specific humidity drives influenza transmission globally is inconsistent with the low predictive power of this climatic variable in middle and low-latitude sites in our study.
Relative humidity was a strong predictor of influenza peaks in high and low latitudes, but a poor predictor in middle-latitude. In high-latitude regions, relative humidity can vary significantly between indoor and outdoor environments, and it is typically minimal indoors during the winter when building air is heated. Our analysis relied on outdoor humidity and hence we cannot rule out that winter influenza epidemics in high latitudes could be related to low indoor relative humidity and associated changes to host physiology, such as reduced mucociliary clearance [20]. In low latitudes it is possible that relative humidity is confounding precipitation in our analysis. Disentangling these two factors will require more highly-resolved epidemiological data from equatorial regions, and further experimental and observational studies.
Solar radiation was also a significant predictor of influenza peaks in high latitudes suggesting that it may also underlie influenza seasonality in these regions, perhaps through variation in vitamin D intake [18]. However, solar radiation was not as strong a predictor of influenza peaks as were specific humidity and temperature. This corroborates recent studies indicating that specific humidity is a stronger predictor of seasonal influenza activity than solar radiation and vitamin D variability in the U.S. [14], [37]. Still, the potential seasonal forcing of solar radiation on influenza transmission warrants further experimental and observational investigation.
The power of this study was rooted in the large number of spatially diverse sites used to develop the epidemiological and climatic databases and associated models. However, the challenge of describing seasonal influenza activity consistently across a variety of data sources required a crude epidemiological measure, such as the average month of peak influenza activity. This measure of influenza activity has several key drawbacks. Foremost, all months with the exception of the peak influenza months were considered equal, whether they had substantial influenza activity or not. Second, the month of peak influenza activity may not be contemporaneous with the month in which transmission is under the most environmentally favorable conditions, since non-environmental factors such as viral seeding, population susceptibility, and person-to-person contact rates likely play a role in the timing of influenza epidemics [15], [21]. In this respect, it is reassuring that a 1-month lag maximized the association between influenza peak and most of the climatic variables, which is broadly consistent with the time scale of the ascending phase of a local epidemic. Third, we could not assess putative geographical variation in the transmission potential or intensity of influenza epidemics. For example, we may expect locations that have the most favorable environmental conditions to experience the greatest influenza annual attack rates and reproduction numbers, holding all other relevant variables equal. A further limitation relates to between-year variability in influenza timing and the limited temporal sampling of our dataset, which may have resulted in imprecise estimates of the average influenza peak in some sites, especially sites that had only one year of influenza data. However, sensitivity analyses limited to multi-year studies revealed similar relationships between climate predictors and influenza peaks, confirming the robustness of our results. Finally, we were unable to check whether between-year fluctuations in climatic variables may result in departures from average influenza seasonal characteristics in specific years. This question could be an interesting area for future research with more temporally refined epidemiological datasets.
A number of follow-up studies could help refine our understanding of the small and large-scale processes underlying influenza seasonality. For example, experimental infections in humans under controlled temperature and humidity conditions could determine which environmentally-mediated mechanisms are most important for human-to-human transmission. However, there are several ethical and methodological hurdles to overcome in such studies [38]. Seasonal fluctuation in contact rates could be monitored by wireless sensor technology, which has recently proved successful in estimating dynamic contact patterns in schools and at conferences [39]–[40]. On a broader spatial scale, determining regional differences in influenza transmission dynamics and attack rates would be most informative. A recent study has suggested that the reproduction number of seasonal epidemics was lower on average in Brazil than in temperate countries, which could be mediated by environmental factors [41]. Modeling of long-term influenza time series data could help assess the transmission impact of seasonal fluctuations in population mixing in different regions, such as those associated with school cycles [21] and transportation networks [29], [30], [42]. For example, epidemiological evidence indicates that influenza circulation was weakly seasonal in Iceland prior to the 1930s, presumably because of low connectivity with other populations, and epidemics only became fully synchronized with those in Europe and the USA following a dramatic increase in international travel in the 1990s [43]. Hence, efforts to collate multiyear influenza epidemiological information retrospectively and prospectively in various regions of the globe, especially from middle and low latitude regions, will be of tremendous help to further elucidate the environmental and population drivers of seasonality.
In conclusion, our study broadens our understanding of the relationships between seasonal influenza epidemics and environmental factors and provides a synthesis of epidemiological and climatic characteristics across temperate, subtropical and tropical regions. We have highlighted the importance of thresholds in specific humidity, temperature and precipitation that are associated with the epidemiology (and potentially the modes of transmission) of influenza. The results of this study could help improve existing influenza transmission models by providing a more accurate estimate of the environmental forcing on transmission processes, particularly in low and middle latitudes [14], [44]. Further, our models could be used to predict the seasonal timing of influenza activity in locations with little or no observational data on influenza activity, and help target surveillance efforts and optimize the timing of seasonal vaccine delivery, [45]. More broadly, we hope that our work will generate interest in testing the association between climatic patterns and infectious disease across a wide range of diseases and latitudes, particularly for respiratory and enteric pathogens that display marked seasonality [2], [46]. A better understanding of the environmental, demographic and social drivers of infectious disease seasonality is key for improving transmission models and optimizing interventions [47].
|
10.1371/journal.pgen.1006801 | C. elegans DAF-16/FOXO interacts with TGF-ß/BMP signaling to induce germline tumor formation via mTORC1 activation | Activation of the FOXO transcription factor DAF-16 by reduced insulin/IGF signaling (IIS) is considered to be beneficial in C. elegans due to its ability to extend lifespan and to enhance stress resistance. In the germline, cell-autonomous DAF-16 activity prevents stem cell proliferation, thus acting tumor-suppressive. In contrast, hypodermal DAF-16 causes a tumorous germline phenotype characterized by hyperproliferation of the germline stem cells and rupture of the adjacent basement membrane. Here we show that cross-talk between DAF-16 and the transforming growth factor ß (TGFß)/bone morphogenic protein (BMP) signaling pathway causes germline hyperplasia and results in disruption of the basement membrane. In addition to activating MADM/NRBP/hpo-11 gene alone, DAF-16 also directly interacts with both R-SMAD proteins SMA-2 and SMA-3 in the nucleus to regulate the expression of mTORC1 pathway. Knocking-down of BMP genes or each of the four target genes in the hypodermis was sufficient to inhibit germline proliferation, indicating a cell-non-autonomously controlled regulation of stem cell proliferation by somatic tissues. We propose the existence of two antagonistic DAF-16/FOXO functions, a cell-proliferative somatic and an anti-proliferative germline activity. Whereas germline hyperplasia under reduced IIS is inhibited by DAF-16 cell-autonomously, activation of somatic DAF-16 in the presence of active IIS promotes germline proliferation and eventually induces tumor-like germline growth. In summary, our results suggest a novel pathway crosstalk of DAF-16 and TGF-ß/BMP that can modulate mTORC1 at the transcriptional level to cause stem-cell hyperproliferation. Such cell-type specific differences may help explaining why human FOXO activity is considered to be tumor-suppressive in most contexts, but may become oncogenic, e.g. in chronic and acute myeloid leukemia.
| The transcription factor FOXO is a well-known tumor suppressor whose activity is controlled by nutrients and stress signaling. In the roundworm C. elegans, the activity of the FOXO protein DAF-16 is best known for its beneficial role in stress response and long lifespan. However, FOXO proteins may also promote tumor cell growth and maintenance in chronic and acute myeloid leukemia, suggesting that may have different roles in distinct contexts. Previously we have shown that selective activation of DAF-16 in the epidermis causes a tumorous growth in the stem cells of the C. elegans germline. Now we demonstrate that this oncogenic activity of DAF-16 is mediated by interactions with the transforming growth factor (TGFß)/Bone Morphogenic protein (BMP) signaling pathway. In the epidermis, direct binding of DAF-16 and R-SMAD proteins of the BMP pathway helps to activate genes involved in the mTORC1 signaling pathway that is frequently activated in tumors. We propose that the transcription factor DAF-16/FOXO may be controlled in different ways in the stem cells, in which its activity normally prevents tumor formation, and in other tissues, in which defects in controlling its activity may result in overwriting the beneficial stem cell activity to eventually promote tumor cell growth.
| The C. elegans FOXO transcription factor DAF-16 is one of the most intensively studied transcription factors due to its ability to extend lifespan and confer stress resistance [1]. In well-fed and stress-free animals, DAF-16 is inactivated by the insulin/IGF-1 signaling (IIS) through the IIS receptor DAF-2, phosphatidylinositol-4,5-bisphosphate 3-kinase (PI3K) and the AGC kinases AKT-1, AKT-2, and SGK-1 that promote its cytosolic localization [2–4]. Upon reduced IIS or stress stimuli DAF-16 becomes nuclearly localized and transcriptionally active. Active DAF-16 extends lifespan and increases stress resistance, indicating an overall beneficial effect of DAF-16 activation. DAF-16 activity in the germline, the source of stem cells in C. elegans, has a cell-autonomous inhibitory role in cell proliferation [5].
In a previous study we have shown that increasing DAF-16 activity by transgenic expression causes tumor-like germline hyperplasia that is accompanied by degradation of the extracellular matrix surrounding the gonad, and this eventually results in premature death [6]. Due to its similarity to several aspects of tumorigenesis in higher organisms, we suggest to term this phenotype a “C. elegans germline tumor”. The penetrance of this tumor-like phenotype is strongly enhanced in a loss-of-function mutant of shc-1, which encodes a DAF-2 interacting protein and homolog of the human adaptor protein Shc [6]. We had shown that DAF-16 activity exclusively in the hypodermis is sufficient to induce the tumor-like phenotype, arguing strongly in favor of a non-cell-autonomous mechanism. One of the surprising results of this study was that inactivation of daf-2, which should further increase DAF-16 activity according to the canonical IIS signaling pathway, rather suppresses DAF-16 mediated germline tumor phenotype. This suggests that the pro-oncogenic DAF-16 activity is positively rather than negatively regulated by IIS, providing a conundrum that could not be explained easily. Moreover, a loss-of-function mutation in daf-18/PTEN (Phosphatase and Tensin homolog), encoding a well-studied phosphatase with functions antagonistic to PI3K that, according to canonical insulin signaling, should enhance PI3K signaling and inactive DAF-16, further enhances this phenotype. Furthermore, the three AGC kinases AKT-1, AKT-2, and SGK-1 that in most previous studies have displayed redundant activities as negative regulators of DAF-16, have antagonistic roles in DAF-16 mediated tumor formation. AKT-1 acted as an inhibitor of germline tumor formation, as the shc-1;akt-1 double mutant shows tumor-like growth similar to that observed in shc-1;Is[daf-16::GFP] [6]. Strikingly, down-regulation of akt-2 or sgk-1, like daf-2, suppressed germline hyperplasia in shc-1;akt-1 animals, indicating that the activity of DAF-2, AKT-2 or SGK-1 in this context contributed to germline tumor formation. All these observations are inconsistent with the canonical model of the linear IIS pathway, in which DAF-2 negatively regulates DAF-16. DAF-16 in turn has been shown before to inhibit germ cell proliferation cell-autonomously, and also acts to inhibit gld-1 germline tumors [5, 7] that differ in certain aspects from the tumor-like phenotype observed here. These observations suggest a novel proliferation-promoting DAF-16/FOXO function in the presence of active IIS (a comparison of the novel versus canonical DAF-16 function and its regulation is summarized in the S1 Fig).
The C. elegans germline is the only tissue harboring stem cells and mitotic events throughout adulthood. The somatic distal tip cell (DTC) serves as the niche and maintains the germline stem cell fate by producing ligands to active Notch receptor on neighboring germ cells [8]. In addition, IIS promotes robust larval germline proliferation via inactivation of DAF-16 [5]. Moreover, the mammalian target of rapamycin complex one (mTORC1) signaling in the germline promotes cell proliferation and prevents differentiation [9]. Furthermore, the canonical TGF-ß/dauer signaling pathway has been shown to promote mitosis versus meiosis fate in parallel to Notch signaling [10]. Mutations causing hyperactive Notch signaling can lead to germline tumor due to excessive cell proliferation [11, 12]. In contrast, despite positive roles of IIS and mTORC1 signaling in regulating germline proliferation, active IIS contributing to germline tumor formation to our knowledge has only been reported once [6].
BMP signaling (also called the TGF-ß Sma/Mab pathway in C. elegans) is also highly conserved from invertebrates to humans. The BMP signaling cascade involves extracellular ligands like DBL-1, type I (SMA-6) and type II (DAF-4) receptors, R-Smads (SMA-2 and SMA-3), a co-Smad (SMA-4), and a transcription co-factor (SMA-9) [13–15]. BMP signaling in C. elegans positively regulates body size, male tail development and innate immunity [16–18]. In addition, reduction of BMP signaling and IIS delays reproductive aging by maintaining oocyte and germline quality through non-autonomous mechanisms [19].
Here, we uncover a novel interaction of BMP signaling and DAF-16 in promoting germline tumor formation. Our genetic and biochemical results suggest that the R-Smad proteins SMA-2 and SMA-3 bind directly to DAF-16 in the nucleus to regulate transcription of several common target genes of the mTORC1 pathway. R-Smad regulation of DAF-16 activity during germline tumorigenesis seems to be critical only in the hypodermis. Furthermore, we demonstrate that cell-autonomous and non-cell-autonomous activities of DAF-16 affect germline proliferation in opposite ways, suggesting that a functional balance of DAF-16 in distinct tissues may be important for rapid adaption to the diverse environmental change to ensure survival and reproduction.
We had shown previously that both shc-1 and daf-16 which link JNK and IIS signaling pathways contribute to tumorous germline formation [20]. In order to dissect the contribution of each pathway, we performed an RNA interference screen to identify kinases involved in germline hyperplasia. For this purpose, we screened the contribution of all kinases in the Ahringer lab [21] and the ORFeome library [22]. As a readout, we looked for suppression of the 49% penetrant sterility of shc-1;Is[daf-16::GFP] animals, a phenotypic aspect that is the consequence of the tumorous germline (S2 Fig) [6]. We found that RNAi clones of three kinase genes, sma-6, hpo-11, and rsks-1, not only strongly suppressed sterility (S1 Table), but also germline hyperproliferation and rupture of the gonadal basement membrane (Fig 1A–1E, S3 Fig and S2 Table). This suggests that the activity of all three genes contributes to germline tumor formation in shc-1;Is[daf-16::GFP] mutant animals.
Among these identified three genes, rsks-1 encodes the C. elegans S6 kinase homolog, a downstream target of mammalian target of rapamycin complex 1 (mTORC1) [23] which is activated by the GTPase RHEB-1/Rheb and the adapter protein DAF-15/Raptor [24–26]. This suggested that mTORC1 signaling might contribute to DAF-16 dependent germline tumor formation. We next asked whether inactivating other mTORC1 components in C. elegans, let-363/TOR, rheb-1/Rheb, and daf-15/Raptor, could also suppress the shc-1;Is[daf-16::GFP] tumor. RNAi knock-down of let-363, daf-15 or rheb-1, respectively, resulted in constitutive L3 larval arrest in the F1 generation of shc-1;Is[daf-16::GFP] animals, substantiating previous findings that mTORC1 is necessary for normal development of C. elegans [27]. However, this prevented examination of the tumor phenotype that we examined at mid-late L3 larval stage [6]. Therefore, we initiated RNAi knock-down at larval stage L1 since such short-term RNAi knock-down of mTORC1 did not cause L3 larval arrest. RNAi knock-down of daf-15, let-363 and rheb-1, respectively, strongly suppressed both germline hyperproliferation and disruption of the basement membrane of shc-1;Is[daf-16::GFP] mid-late L3 animals (Fig 1B and 1E). We conclude that mTORC1 in this genetic context also has oncogenic activity in C. elegans.
sma-6 encodes the TGF-ß/BMP type I receptor [28], suggesting a role of BMP signaling in germline tumor formation. We asked whether the other genes in the BMP pathway are also involved the germline phenotype. Crossing shc-1;Is[daf-16::GFP] with daf-4(m63) strongly enhanced the low penetrance of the constitutive dauer phenotype of daf-4, and prevented examination of this strain in stages later than L3. Inactivating the BMP receptor ligand gene dbl-1, the Co-Smad gene sma-4, the Smad genes sma-2, sma-3, or the Smad cofactor gene sma-9 by either mutation or RNAi fully suppressed the sterile phenotype of shc-1;Is[daf-16::GFP] and strongly decreased the percentage of animals with disrupted gonadal basement membrane in both L3 and day one adult animals (Fig 1F and 1G, S1 and S2 Tables). Therefore, in addition to mTORC1, also BMP signaling contributes to DAF-16 dependent germline tumor formation.
In C. elegans, a second TGF-ß/dauer signaling pathway regulates dauer formation [29, 30]. It had been suggested that neuronal TGF-ß dauer signaling promotes germline proliferation [10, 31]. We observed that RNAi knock-down of daf-1 encoding the type I TGF-ß receptor (specifically involved in dauer signaling) only slightly suppressed sterility, but failed to decrease the severity of the gonadal defect in shc-1;Is[daf-16::GFP] animals (S2 Table). Therefore, daf-1 RNAi behaved differently from sma-6 RNAi, suggesting that TGF-ß/dauer signaling probably affects the DAF-16 mediated germline phenotype only indirectly.
Since we discovered, to our knowledge, a novel functional cross-talk between BMP signaling and DAF-16, we asked whether they also interact to promote additional functions of each other. Active DAF-16 under reduced IIS leads to constitutive dauer formation at larval stage and lifespan extension during adulthood [32]. Therefore, we next tested whether BMP signaling is also required for DAF-16 to regulate lifespan extension or dauer formation. sma-6 RNAi knock-down did not shorten, but rather slightly extend the lifespan of daf-2(e1370) animals (Fig 2A and S3 Table). sma-6(wk7) also did not suppress constitutive dauer formation of daf-2(e1370) at 25°C (S4 Table). These data suggest that SMA-6/BMP function does not seem to be involved in other roles of DAF-16.
One of the best studied functions of BMP signaling in C. elegans is in body size regulation [28]. IIS also regulates body size, since daf-2 mutant animals are longer than wild type as a consequence of DAF-16 activation [34], results that we confirmed here (Fig 2B and S5 Table). However, sma-6(wk7) reduced body sizes of both daf-2 and daf-16;daf-2 animals, suggesting that BMP signaling positively regulates body size probably in parallel to DAF-16. Taken together, signaling to the germline is the only cooperative role of BMP and DAF-16 interaction that we found.
DAF-16 activity is regulated at least at two steps: via control of its nuclear translocation and, subsequently, via modifying its strength as a transcriptional activator through protein interactions and possibly protein modifications. The DAF-16(4A) variant shows constitutive nuclear localization and our previous study has indicated that both DAF-16(4A)::GFP and wild type DAF-16::GFP expressed in shc-1(ok198) background induce germline tumorigenesis [6, 35]. Nuclear localization of DAF-16(4A)::GFP was not altered by a sma-6 mutation (Fig 2C). However, sma-6(wk7) strongly reduced the penetrance of the germline phenotype in shc-1;Ex[daf-16(4A)::GFP] animals (Fig 2D and S2 Table). This suggests that SMA-6 may modulate the transcriptional activity of nuclear DAF-16.
Reporter gene studies have suggested that sma-6 is expressed in the hypodermis, intestine, and pharynx [28]. Using our own reporter genes, we also observed weak sma-3 expression in the somatic gonad (S4 Fig). It is not known whether sma-6 is expressed in the germline due to germline silencing of the transgenic reporter genes used in such expression analysis. To knock-down sma-6 exclusively in the germline, we performed sma-6 RNAi in the background of the rrf-1(ok589) loss-of-function mutant that limits the sensitivity against RNA interference to the germline [36]. sma-6 RNAi suppressed the gonadal integrity defects in shc-1;Is[daf-16::GFP], but not in shc-1;rrf-1;Is[daf-16::GFP] animals (Fig 3A, S1 and S2 Tables), indicating that somatic, but not germline activity of SMA-6 may contribute to DAF-16 dependent tumors formation. Since a recent report suggested that rrf-1(ok589) to some extent also might allow RNAi to work in the intestine and seam cells [37], we cautiously argue that neither seam cell nor intestinal SMA-6 activity seem to contribute to the DAF-16 dependent germline tumor phenotype.
In order to pinpoint the critical tissue for BMP activity, we expressed sma-6 under the control of tissue specific promoters in shc-1;sma-6;Is[daf-16::GFP] animals to rescue tumor suppression of sma-6(wk7) (rescue should result in reoccurrence of the tumor phenotype). Neither expression of sma-6 in the intestine, pharynx nor in the somatic gonad resulted in the reappearance of germline tumors in shc-1;sma-6;Is[daf-16::GFP] animals (Fig 3B). We tried in vain to receive stable transgenic lines of Pdpy-7:sma-6 in shc-1;sma-6;Is[daf-16::GFP] background, in which sma-6 should be expressed exclusively in the hypodermis. Although we obtained more than 200 F1 transgenic animals, we never observed sma-6 expression in the F2 generation, indicating either toxicity or silencing of this transgene. Therefore, we analyzed F1 transgenic animals with hypodermal-only sma-6 expression, but these did not show rescue. However, coexpression of sma-6 in both hypodermis and somatic gonad was sufficient to rescue sma-6 tumor suppression. In contrast, expressing sma-6 in all sma-6 positive tissues except the hypodermis (somatic gonad, intestine and pharynx), did not cause tumor reoccurrence. We conclude that the hypodermis and somatic gonad are the critical tissues for BMP signaling that promote DAF-16 tumor induction.
Since it has been shown that forkhead transcription factors can bind to Smad proteins directly [38, 39], a simple hypothesis was that DAF-16 might physically interact with R-Smad proteins SMA-2 and/or SMA-3 in the hypodermis. We tested R-Smad and DAF-16 interactions with three different assays.
First, we performed in vivo co-immunoprecipitation experiments from C. elegans cell extracts. Extracts from animals carrying functional GFP::sma-3 were precipitated with anti-GFP antibody and the resulting precipitates were subjected to Western blotting by using anti-DAF-16 antibody to detect endogenous DAF-16 (Fig 3C). We detected DAF-16 in GFP::SMA-3 but not in GFP precipitates, suggesting a specific interaction between DAF-16 and SMA-3. We were unable to express a functional version of tagged SMA-2 in C. elegans, since all fusion tags (GFP, Flag and V5 tags) we had tested resulted in loss of SMA-2 function, indicated by a failure of the respective transgene to rescue the small body size phenotype of sma-2(e502) animals.
Second, we performed IP with recombinant DAF-16::V5 and SMA-2::GFP/GFP::SMA-3 co-expressed in the human HEK293T cell line to verify the physical interaction. DAF-16::V5 protein was coimmunoprecipitated with both SMA-2::GFP and GFP::SMA-3 (Fig 3D), further supporting a direct interaction between the two R-Smad proteins and DAF-16.
Third, we tested whether DAF-16/R-Smad interactions occur in a yeast model. As positive control we included FTT-2, which is a C. elegans 14-3-3 homologue and has been shown to interact with DAF-16 [40]. We found that in a yeast-two-hybrid system, FTT-2, SMA-2, and SMA-3 all interact with DAF-16 (Fig 3E), corroborating our co-immunoprecipitation results. Even though it has been suggested previously that SMA-2 and SMA-3 form a heterodimer [41], we were unable to detect a direct interaction between SMA-2 and SMA-3 in this assay (S5 Fig).
Since both DAF-16 and its BMP interactors SMA-2 and SMA-3 encode transcription factors, we next sought to determine downstream genes influenced by both sets of proteins. Several reports have already documented candidate genes directly regulated by DAF-16 or BMP signaling, but genes co-regulated by DAF-16 and SMA-2/3 have not been reported so far. We therefore performed microarray assays to identify genes differentially regulated in shc-1(ok198);Is[daf-16::GFP], shc-1(ok198) daf-16(mu86) and shc-1(ok198);sma-6(wk7);Is[daf-16::GFP] strains. Hierarchical clustering of genes differentially regulated between any of the three strains suggested a substantial effect of BMP signaling on the transcriptional output of DAF-16 (S6 Fig). To analyze the consequences of this interaction in more detail, we used a linear model analysis to select genes with a statistically significant (FDR-corrected p-value < 0.01) and robust (at least two-fold expression change) response to DAF-16 or SMA-6 activity changes. We identified 2,006 genes as regulated by DAF-16 and 1,957 genes as regulated by SMA-6. Intriguingly, 777 of these genes respond to both DAF-16 and SMA-6, which corresponds to 39% and 40% of all genes regulated by DAF-16 and by SMA-6, respectively (Fig 4A). A more detailed investigation of these genes revealed that 79% (612/777) of the genes that at least partially depend on SMA-6 are genes that are activated by DAF-16, and another 108 genes are downregulated by DAF-16 in a SMA-6-dependent manner (Fig 4B). Together this indicates that 94% of the genes responsive to both factors are controlled by them in a cooperative manner.
Next we compared the genes identified in our analysis with previously reported consensus sets of genes that are activated (class I genes) and repressed (class II genes) by DAF-16 and that were derived from the combined analysis of different genome-wide expression studies. Typically, these compared daf-2(-) against daf-16(-);daf-2(-) backgrounds [42]. We found that 33% (162 of 498) of the genes we identify as upregulated by DAF-16 in a SMA-6 independent manner are known class I target genes. Similarly, 30% (218 of 731) of the genes that we see downregulated by DAF-16 independent of SMA-6 are known class II target genes. In contrast, only 8.6% (62 of 720) of the genes either up- or downregulated cooperatively by DAF-16 and SMA-6 are known class I or class II target genes, respectively (Fig 4C). This suggests that the common targets of DAF-16 and BMP signaling are typically distinct from the DAF-16 targets that were identified by comparing daf-2(-) and daf-16(-); daf-2(-) gene expression. To confirm this hypothesis, we also checked whether modulation of BMP signaling could alter expression of the frequently used DAF-16 target gene sod-3. In daf-2(e1370) mutant, sod-3 expression is increased in a DAF-16 dependent manner and this can be visualized in vivo by using an integrated transcriptional GFP fusion reporter [43]. We verified this observation and showed that daf-2 animals showed increased GFP expression, and this was fully dependent on DAF-16 (Fig 5A). Knock-down of sma-6 did not significantly affect the expression level of the sod-3::GFP reporter in daf-2 background, further supporting our microarray assay results.
Next, we searched for common targets of DAF-16 and BMP signaling to cause germline tumor formation. Inspecting the modENCODE database, we found reports of DAF-16 occupancy on the promoter regions of hpo-11, daf-15, rsks-1 and rheb-1 (S7 Fig). In addition, we identified consensus DAF-16 binding sites in each of their 5’ promoter sequences, making them candidates for being direct DAF-16 target genes. Next, we tested by mRNA quantification whether expression of these genes is affected by DAF-16 and BMP activities. Quantitative-PCR (q-PCR) results showed that the mRNA levels of rsks-1, rheb-1, daf-15, and hpo-11 in shc-1;Is[daf-16::GFP] animals were strongly increased when compared to wild-type, and the increase fully depended on DAF-16 (Fig 5B). Inactivation of sma-6 significantly reduced mRNA levels of rsks-1, rheb-1, daf-15, but not of hpo-11. In summary, combinatorial activation by DAF-16 and R-Smad transcription factors in the hypodermis may activate some common transcriptional target genes to cause tumor formation in the germline.
To show direct binding of DAF-16 and SMA-3 to the promoters of the candidate target genes, we performed ChIP-qPCR experiments using the promoter regions predicted by the modENCODE project (position of primer sets for ChIP-qPCR indicated in the S7 Fig). Our data revealed binding of both DAF-16 and SMA-3 in the rheb-1 promoter, and binding of DAF-16 in the hpo-11 promoter (Fig 5C). Quantification of DAF-16 and SMA-3 binding on both daf-15 and rsks-1 promoters failed, since we were unable to find primer sets suitable for q-PCR quantification (list of tested primer sets is provided in S9 Table).
One of the surprising results in Qi et al. (2012) was that DAF-16 dependent germline tumor formation was suppressed, rather than enhanced by additional inactivation of daf-2. This indicated an obvious discrepancy to the canonical model of the IIS pathway, which would predict that down-regulation of daf-2 should enhance, rather than suppress DAF-16 activities. We asked whether inactivation of daf-2 suppresses DAF-16 dependent tumor phenotype via regulating the identified downstream genes. Whereas all four candidate genes (rsks-1, rheb-1, daf-15, and hpo-11) are strongly activated in the tumor strain shc-1;Is[daf-16::GFP] that carries a daf-2(+) allele, they were differentially regulated in daf-2(-) animals (Fig 5D). In daf-2 mutant animals only hpo-11 mRNA level was up-regulated by DAF-16. In contrast, both rsks-1 and daf-15 transcripts were strongly reduced and only down-regulation of daf-15 mRNA level partially depended on DAF-16. This suggests the existence of both DAF-16 dependent and independent mechanisms to inactivate mTORC1 in daf-2(-) mutant background. In addition, our results indicate that the target specificity of DAF-16 can be differentially affected by upstream IIS.
The results so far suggest that DAF-16 mis-regulation as a consequence of distinct genetic backgrounds requires BMP signaling to become tumorigenic. The obvious question therefore was, whether BMP signaling also contributes to normal developmental functions of DAF-16 in wild-type strains, such as controlling germline proliferation. This could be examined by counting germline nuclei in the proliferative zone of the germline which is located between the DTC and the transition zone in which germ cell differentiation is initiated. daf-16 and sma-6 animals at day one of adulthood contained approximately 20% and 63% less mitotic germline nuclei, respectively, than wild-type animals (Fig 6A), suggesting that both DAF-16 and SMA-6 positively regulate distal germline proliferation. To address tissue specificity of DAF-16 and SMA-6, we tried rescuing daf-16 and sma-6 in a tissue-specific way. Transgenic expression of both wild-type DAF-16::GFP and constitutive nuclear DAF-16(4A)::GFP in the hypodermis completely rescued the decreased number of mitotic germline nuclei in daf-16 mutant animals. DAF-16 expression in the intestine also partially rescued the germline proliferation phenotype of daf-16 mutants whereas expression in the neuron or muscle had no effect. We noticed that, whereas in the majority of animals hypodermal DAF-16 expression of either wild-type daf-16::GFP or daf-16(4A)::GFP rescued the daf-16 mutant phenotype, in approximately 10% of these animals the numbers of mitotic germline nuclei even exceeded 300, something never observed in wild type animals. This sometimes led to enlargement of distal gonads (S8 Fig). Expression sma-6 in the somatic gonad robustly rescued germline proliferation defect while hypodermis expression sma-6 showed a minor, but significant, increase of proliferative germ cell number (Fig 6B). sma-6 expression in neither intestine nor pharynx had any effect on sma-6(-) animals. Taken together, these results suggest both hypodermal DAF-16 and BMP signaling are required for wild-type germline proliferation. And they also act in additional distinct tissues.
Do DAF-16 and BMP functions in the hypodermis depend on each other? Hypodermal expression of sma-6 failed to increase germ nuclei number in the proliferative zone of daf-16 animals (Fig 6A), and hypodermal daf-16 expression also did not increase the number of proliferative germ cells in a sma-6 mutant (Fig 6B), These results suggest that the DAF-16 and BMP signaling activities in the hypodermis to regulate germline proliferation depend on each other, consistent with our proposed model that DAF-16 and R-Smad protein interaction cooperate to control common target genes.
Next we asked whether selective down-regulation of the target genes of DAF-16/SMA-2/3 interaction in the hypodermis, namely rsks-1, rheb-1, daf-15 but also hpo-11, affects germ cell proliferations. For this purpose, we performed hypodermis-specific RNAi knock-downs of these four genes in an rde-1;Is[Plin-26::rde-1] strain. This strain harbors the argonaute/PIWI gene rde-1 required for the function of RNA interference exclusively in the hypodermis. rheb-1, daf-15, or hpo-11 RNAi resulted in an approximately 20% reduction of germline nuclei in the proliferative zone and rsks-1 RNAi led to an approximately 25% reduction (Fig 6C), indicating that RHEB-1, RSKS-1, DAF-15, and HPO-11 in the hypodermis positively regulate germline proliferation. We asked whether hypodermal specific RNAi knock-downs of these four genes could further reduce number of proliferative germ nuclei in either daf-16 or sma-6 background. rheb-1, daf-15, or hpo-11 RNAi failed to further reduce number of proliferative germ nuclei in either daf-16;rde-1;Is[Plin-26::rde-1] or sma-6;rde-1;Is[Plin-26::rde-1] animals (Fig 6D and 6E). rsks-1 RNAi further reduced number of proliferative germ nuclei in these two strains but to less extent compared with rde-1;Is[Plin-26::rde-1] (15% vs. 25% reduction). Taken together, these experiments indicate that rsks-1, rheb-1, daf-15 and hpo-11 are epistatic to both DAF-16 and BMP signaling in the hypodermis to promote germline proliferation. In addition, rsks-1 might also mediate additional pathway to regulate cell proliferation.
We identified protein interactions of DAF-1/FOXO and the Smad transcription factors of TGF-ß/BMP signaling that regulate target genes components of the mTORC1 pathway in C. elegans. This cross-talk promotes cell proliferation of germline stem cells non-cell-autonomously, and its hyper-activation results in a tumor-like germline phenotype. Another identified downstream target of IIS/DAF-16 axis encodes HPO-11, a pseudo-kinase with unknown function in C. elegans and ortholog of NRBP/MADM (nuclear receptor-binding protein, myeloid-leukemia factor-adaptor molecule), a regulator of germline stem cells and tumorigenesis in humans and Drosophila. Based on these functional similarities, we suggest that aspects of FOXO/R-Smad signaling to the germline stem cells may be conserved in evolution.
DAF-16/FOXO hyperactivity in the hypodermis results in a tumor-like germline phenotype that we showed to involve germ cell hyper-proliferation and disruption of the basement membrane [6]. Now we identified genes whose down-regulation suppressed this phenotype, among them sma-2, sma-3, sma-4, sma-6 and sma-9, encoding components of the TGF-ß/BMP pathway. A combination of genetic and biochemical experiments furthermore revealed that the BMP effectors SMA-2 and SMA-3 can physically interact with DAF-16 to promote germline proliferation at wild type conditions, but may cooperate with hyper-active DAF-16 to cause tumors. This interaction is selectively required for only a subset of DAF-16 functions, based on the following arguments: (I) We show that the well-known beneficial roles of DAF-16 in extending lifespan and in promoting dauer formation in daf-2(-) animals do not involve BMP signaling (Fig 2, S3 and S4 Tables). (II) The DAF-16 dependent transcriptional activation of sod-3 in daf-2 mutant animals is also independent of BMP (Fig 5A). (III) Moreover, both BMP and IIS/DAF-16 contribute to the regulation of body size, but do so independently of one another (Fig 2C and S5 Table). Taking together, these results suggest that the functional interaction between BMP signaling and DAF-16 has consequences that seem to be specific for the germline, even though this interaction takes place in the hypodermis, a tissue adjacent to the gonad. Although both DAF-16 and R-Smad proteins are capable of shuttling between nucleus and cytoplasm, we suggest that the relevant interaction described here occurs in the nucleus, since a sma-6 mutation could still suppress the phenotype caused by constitutively nuclear DAF-16(4A)::GFP. We therefore propose that BMP signaling promotes the activity of nuclear DAF-16, rather than its nuclear entry (Fig 2C and 2D), by modulating DAF-16 binding to the promoters of downstream genes. This claim was further justified by identifying DAF-16 and SMA-3 binding sites in the promoter of rheb-1 gene of the mTORC1 pathway (see below).
Hypodermal activity of BMP signaling to the germline has been described to regulate oocyte and germline quality maintenance, but this was suggested to occur independently of IIS [19]. Our newly discovered functional cross-talk between DAF-16 and BMP is distinct of the previously identified interactions of DAF-16 with the canonical TGF-ß/dauer signaling pathway involved in lifespan regulation [44]. TGF-ß/dauer signaling in the nervous system also promotes germline proliferation via preventing meiosis [10, 31]. Even though we observed that down-regulating the daf-1 gene, encoding a receptor type I homologue of the TGF-ß/dauer, but not the BMP signaling pathway, slightly decreased the severity of the DAF-16 dependent germline tumor phenotype, we consider the interaction between DAF-16 and canonical TGF-ß/dauer pathway most likely as being indirect, since DAF-16 and DAF-1 act in different tissues [6, 10].
We show that the R-Smad/DAF-16 interaction activates the expression of at least three genes (rsks-1, rheb-1, and daf-15) which encode components of the mTORC1 pathway. Down-regulation of each of them strongly suppressed tumorigenesis. In addition, DAF-16 activates hpo-11 possibly independent of BMP signaling, although the influence of hpo-11 on wild type mitotic germ cell numbers (Fig 6C) is fully lost in both daf-16(mu86) and sma-6(wk7) mutant backgrounds (Fig 6D and 6E). All four down-stream genes contain DAF-16 binding elements (DBE) or DAF-16 associate elements (DAE) in their 5’ promoter regions and hpo-11 has been reported to be a DAF-16 direct target [45]. Even though arguments were convincing that DAE may be a recognition site of PQM-1, the transcription factor acting antagonistically to DAF-16 [42], the C. elegans modENCODE project detected direct binding of DAF-16 onto the promoter regions of rheb-1 that contains DAE, rather than DBE [46], and we confirmed DAF-16 binding to rheb-1 by chromatin IP (Fig 5C). Our genetic data also do not find any evidence for a contribution of PQM-1 to DAF-16 dependent germline tumor formation, as pqm-1 RNAi in our hands did not modulate the tumor phenotype (S1 and S2 Tables). Moreover, pqm-1 was reported to be expressed exclusively in the intestine, whereas our data suggest that hypodermal expression of all four downstream genes promotes cell proliferation in the germline. Together, these data suggest that rsks-1, rheb-1, daf-15, and hpo-11 are candidates for direct transcriptional targets of R-Smad and/or DAF-16 to promote cell proliferation in the germline. Our results also indicate that DAF-16 can activate mTORC1 at the transcriptional level. This differs from the previous report that DAF-16 represses daf-15 transcription in daf-2(-) mutant animals [26], suggesting a context dependent regulation of DAF-16 on mTORC1. We also showed that presence or absence of DAF-2 determines whether DAF-16 activates or represses transcription of mTORC1 components (Fig 5D). Similarly, as we will discuss below, presence or absence of DAF-2 also determines whether DAF-16 acts oncogenic or as a tumor suppressor. Such antagonistic effect of FOXO is also known from human tissues [47]. Finally, our results indicate that similar to the widely accepted tumorigenic activity of mTORC1 in other organisms, its C. elegans counterpart may fulfill a comparable role [48, 49].
The complexity of IIS signaling and DAF-16 activities in distinct tissues was highlighted by the observation of germline tumors in distinct genetic contexts. Mutant genetic backgrounds resulting in germline tumors were shc-1;Is[daf-16::GFP], shc-1;akt-1, or shc-1;daf-18;akt-1 [6]. Common denominator is that DAF-16 is active in all strains, and shc-1 inactivation is a requirement for their solid phenotypic penetrance. However, both shc-1;Is[daf-16::GFP], and shc-1;akt-1 only resulted in tumors in the presence of daf-2(+), and daf-18/PTEN(-) exacerbated shc-1;akt-1 tumor penetrance, but did not cause germline proliferation defects in either shc-1 or akt-1 alone. These observations are, as we reported earlier [6], inconsistent with a simple model based on the canonical insulin signaling pathway, since daf-2(+) and daf-18(-) both are commonly known as negative regulators of DAF-16. Therefore, we hypothesize that down-regulation of daf-2 to suppress the DAF-16 induced germline phenotype may predominantly affect the germline rather than somatic functions of IIS [6]. It has been shown that daf-2 inactivation reduced mitotic rates of the germline stem cells via a cell-autonomous DAF-16 activity to arrest cell cycle [5]. We corroborated this result, since we counted a 40% reduction of the germline nuclei in the proliferative zone in daf-2(e1370). This reduction was almost completely dependent on DAF-16 (S9 Fig), suggesting a conserved function of IIS to tie nutrition to cell cycle control, and a cell-autonomous function of DAF-16 in the germline to prevent tumorigenesis. Our results, in contrast, suggest a distinct, germline proliferation-promoting and non-cell-autonomous role of DAF-16 which acts in the hypodermis and becomes dominant when the balance between somatic and germline IIS is perturbed. Since hypodermal expression of a daf-16 transgene in daf-16 mutant animals was sufficient to increase the number of germ cell nuclei in the proliferative zone, these two antagonistic activities of DAF-16 in the hypodermis versus the germline should function in parallel (Fig 7). In well-fed animals, DAF-2 is considered to be active, and DAF-16 therefore should be inactive. Consistently, DAF-16::GFP fusion proteins in somatic tissues have been shown to localize mostly to the cytosol. However, a pool of active DAF-16 must still be present, since daf-16 loss-of-function mutants can be phenotypically distinguished from daf-16(+) wild type animals due to their reduced number of proliferative germ cells (Fig 6A). Therefore, we propose that, in the presence of food, residual DAF-16 activities in both tissues are balanced with pro-proliferating hypodermal dominating over anti-proliferating germline DAF-16 to allow robust germline proliferation. These two opposing functions of DAF-16 allow a balanced response to diverse environmental challenges to ensure survival and optimal reproduction. Upon food limitation and subsequent reduction of IIS, DAF-16 activity in the germline becomes dominant and inhibits germline mitosis. Such different responses of somatic and germline cells, however, require differences in the composition of the IIS pathway in both tissues, for example, tissue-specific utilization of pathway components. Candidates mediating such differences are the AGC family kinases AKT-1, AKT-2, and SGK-1. Indeed, our previous experiments had already shown that the individual distribution of AKT kinases and SGK-1 differ between individual tissues, which might explain why shc-1;akt-1 germline tumor are potently suppressed by RNAi down-regulation of either akt-2 or sgk-1 [6]. However, in certain genetic backgrounds, in which hypodermal DAF-16 activation dominates and this balance is lost, the consequence may be germline tumor.
FOXO transcription factors play an essential role in maintaining hematopoietic stem cells [51]. An important function of FOXO3a was also proposed in the maintenance of cancer stem cells that are responsible for the reoccurrence of chronic myeloid leukemia [52]. In addition, FOXO 3/4 has been shown to promote acute myeloid leukemia (AML) via inhibiting myeloid maturation and apoptosis [47]. Due to a critical role of the PI3K/Akt/mTOR signaling pathway in survival and growth of malignant cells, mTOR inhibitors like rapamycin are used in treating AML. We showed here that mTORC1 can be activated by DAF-16 and BMP signaling to promote germline tumor and it remains to be tested whether this phenotype is sensitive to rapamycin treatment. Even more interesting is the identification of hpo-11 as a novel DAF-16 target gene, because its mammalian ortholog MADM (Mlf-1 adaptor molecule) physically interacts with myeloid leukemia factor 1 (Mlf1), which is involved in AML [53]. Interestingly, a recent publication had suggested that Drosophila MADM regulates the competition between germline stem cells and somatic cyst stem cells, for niche occupancy and also to control tumorigenesis [54]. This may suggest that the molecular mechanisms contributing to stem cell tumors that we identified here could be evolutionarily conserved. To our knowledge there is only one study suggesting that FOXO1/3 regulates follicle growth or death by interacting with the BMP pathways in granulosa cells [55]. It will be of great interest to test whether functional interaction between DAF-16 and BMP signaling is also conserved in Drosophila and in mammalian tumorigenesis, especially in AML.
Information about strains and constructs used is described in the S1 Materials and methods.
Scoring of disruption of the gonadal basement membrane at L3 larval stage was performed with MitoTracker staining which has been described to label basement membrane [56]. Basement membranes were stained by placing animals in a solution of 20 μM MitoTracker Red CMXRos (Molecular Probes) in M9 buffer at 23°C for 1 hr. The animals were allowed to recover for 30 min on NGM agar plate. The mitochondria staining with MitoTracker was photobleached using a 30 s exposure to fluorescence, leaving the basement membrane staining which is resistant to photobleaching.
For scoring disruption of the gonadal basement membrane in day one adults, animals were raised at 20°C and analyzed via DIC microscopy 24 hr after L4 stage. Germ cell outside of the gonad was indicator of gonad disruption. Per test at least thirty animals were examined and each test was performed at least three times.
For counting total numbers of the germ cell nuclei, the numbers of anti-PGL-1 positive cells were quantified at the mid-late L3 stage. Antibody staining was performed as described before [6]. For the whole worm staining, 1:200 anti-PGL-1 antibody was used (a gift from Dr. Susan Strome).
Lifespan assays were initiated at the L4 larval stage. Synchronized animals were raised at 15°C prior to lifespan analysis. In order to inactivate fer-15, L4 animals were transferred to 25°C for the assays and examined every day. Animals that showed no response to touch were scored as dead. All of the lifespan assays were performed three times.
The ‘number of nuclei in the proliferative zone’ included all the germ nuclei between the distal tip and the transition zone of day one adult animals (24 hr after mid-L4 larval stage). For visualizing and counting germ cell nuclei in the proliferating zone of the gonad, dissected gonads were fixed in methanol and resuspended in PBST containing 0.1% Tween 20 and 2 μg/ml 4’,6-diamidino-2-phenylindole (DAPI) before microscopy. Z-stack images of animals were collected and the numbers of the germ cells were counted by using a ImageJ Cell Counter plug-in originally written by Kurt De Vos at the University of Sheffield, Academic Neurology.
The cDNAs of interaction partners were cloned into pGBKT7 and pGADT7 yeast vectors. Competent cells of the yeast strain AH109 were cultured at 30°C until OD600nm value 0.8–1.2. After centrifugation, the pellets were resuspended in 1 ml LiAc-TE solution (100 mM LiAc, 10 mM Tris, pH 7.4, 1 mM EDTA). The transformation was conducted by adding 10 μg boiled salmon sperm DNA (2 mg/ml), 500 μl PEG (50% PEG MG3350) and 1 μg of each plasmid DNA to 100 μl yeast aliquot. After incubation at 30°C for 15 min, 50 μl DMSO were added. After heat shocking at 42°C for 15 min, the cells were spread onto plates without Trp or Leu. The interaction was double determined by spreading the transformed yeast onto the two selective plates (-Leu/-Trp/-His with 3 mM 3-AT and -Leu/-Trp/-Ade).
Total RNA was prepared from L3 animals raised at 20°C by using RNeasy Mini Kit (Qiagen, Venlo, The Netherlands). RNA integrity was analyzed by capillary electrophoresis using a Fragment Analyser (Advanced Analytical Technologies, Inc. Ames, IA). RNA samples had RNA quality numbers (RQN) between X and Y and were further processed with the Affymetrix WT Plus kit and hybridized to GeneChip C. elegans 1.0 ST arrays as described by the manufacturers. Partek Genomics Suite software was used for analysis (Partek Inc., St.Louis, MI). CEL files were imported including control and interrogating probes, pre-background adjustment was set to adjust for GC content and probe sequence, and RMA background correction was performed. Arrays were normalized using Quantile normalization and probe set summarization was done using Median Polish. Probe values were log2 transformed. For the statistical analysis of differentially expressed genes log-scale gene expression values were imported into R and analyzed using the limma package [57]. For GO term analysis we used g:Profiler (http://biit.cs.ut.ee/gprofiler)
Total RNA was prepared from L3 animals raised at 20°C by using RNeasy Mini Kit (Qiagen, Venlo, The Netherlands). act-4 was used as internal control. q-PCR primer sequences daf-15 for: GACAACCGCAAGGAATTATGA, daf-15 rev: CGCCGAGAAGTTGAAGGA, hpo-11 for: CACACGGTTTTATGTGTCAGC, hpo-11 rev: TTCCTGAACACCTTCTGCAAT, rsks-1 for: TCCACCAAATGTTCGTGTTG, rsks-1 rev: GTCGTTTTTCGCACTTGGA, rheb-1 for GAAAATCGGCGTTGGTACTT, rheb-1 rev: GGAACAACTTCTCGGGAAAAC. act-4 for: CCACCATGTACCCAGGAATC, act-4 rev: GTGGGGCGATGATCTTGA.
ChIP was performed as described in Kaletsky et al. [58] by using anti-GFP antibody in wild type (untagged), shc-1(ok198);Is[daf-16::GFP], daf-2(e1370);Is[daf-16::GFP] and a somatic exclusive GFP::sma-3 strain (CS119 sma-3(wk30)III;him-5(e1490)V;qcEx24[GFP::sma-3;rol-6]). The relative fold change of qPCR was normalized to the intern negative control which was daf-15 3’UTR sequence.
q-PCR primer sequences: Psod-3 for: ACAACAATGTGCTGGCCTTG, Psod-3 rev: AATGCATTTCGGGACGTTAG, Prheb-1 for: AATAACGCTTTCAACGCGGAG, Prheb-1 rev: ACCGTACCCAAGCAAACCTG, Phpo-11 for: CCCTTTGGCCGATTCTTGTC, Phpo-11 rev: GAGCCACATGAGACACACAC, daf-15 3’UTR for: AAAAGGCGCTTCATCATCCC, daf-15 3’UTR rev:CACATGAAATTGGTCCCCGC.
GraphPad Prism 6.0 software (GraphPad Software Inc., San Diego, USA) was used to analyze the data. One way analysis of variance followed by Dunnett’s multiple comparison test was used to evaluate statistical significance of multiple groups of samples unless stated otherwise. For all statistical tests the 0.05 level of confidence was accepted as a significant difference.
|
10.1371/journal.pcbi.1005376 | Elucidation of molecular kinetic schemes from macroscopic traces using system identification | Overall cellular responses to biologically-relevant stimuli are mediated by networks of simpler lower-level processes. Although information about some of these processes can now be obtained by visualizing and recording events at the molecular level, this is still possible only in especially favorable cases. Therefore the development of methods to extract the dynamics and relationships between the different lower-level (microscopic) processes from the overall (macroscopic) response remains a crucial challenge in the understanding of many aspects of physiology. Here we have devised a hybrid computational-analytical method to accomplish this task, the SYStems-based MOLecular kinetic scheme Extractor (SYSMOLE). SYSMOLE utilizes system-identification input-output analysis to obtain a transfer function between the stimulus and the overall cellular response in the Laplace-transformed domain. It then derives a Markov-chain state molecular kinetic scheme uniquely associated with the transfer function by means of a classification procedure and an analytical step that imposes general biological constraints. We first tested SYSMOLE with synthetic data and evaluated its performance in terms of its rate of convergence to the correct molecular kinetic scheme and its robustness to noise. We then examined its performance on real experimental traces by analyzing macroscopic calcium-current traces elicited by membrane depolarization. SYSMOLE derived the correct, previously known molecular kinetic scheme describing the activation and inactivation of the underlying calcium channels and correctly identified the accepted mechanism of action of nifedipine, a calcium-channel blocker clinically used in patients with cardiovascular disease. Finally, we applied SYSMOLE to study the pharmacology of a new class of glutamate antipsychotic drugs and their crosstalk mechanism through a heteromeric complex of G protein-coupled receptors. Our results indicate that our methodology can be successfully applied to accurately derive molecular kinetic schemes from experimental macroscopic traces, and we anticipate that it may be useful in the study of a wide variety of biological systems.
| Unraveling the lower-level (microscopic) processes underlying the overall (macroscopic) cell response to a given stimulus is a challenging problem in cell physiology. This has been a classic problem in biophysics, where the ability to record the activity of single ion channels that generate a macroscopic ion current has allowed a measure of direct access to the underlying microscopic processes. These classic studies have demonstrated that very different groupings of the microscopic processes can yield extremely similar macroscopic responses. Biologists in fields other than biophysics are frequently confronted with the same macroscopic-to-microscopic problem, usually, however, without any direct access to the microscopic processes. Thus, the development of computational methods to deduce from the available macroscopic measurements the nature of the underlying microscopic processes can be expected to substantially advance the study of many areas of cell physiology. Toward that aim, here we have derived and tested a hybrid computational-analytical method to extract information about the microscopic processes that is hidden in macroscopic experimental traces. Our method is independent of the particular system under study, and thus can be applied to new as well as previously-recorded macroscopic traces obtained in a wide variety of biological systems.
| In order to ensure their survival and correct physiological function, cells must respond appropriately to many different kinds of stimuli, such as the presence of various neurotransmitters or hormones in the extracellular fluid, the depolarization of the cell membrane, or stimuli such as mechanical force or light [1]. The cell’s response to any particular stimulus engages vast numbers of single molecules whose actions combine to perform various tasks within the cell.
To provide a framework for the study of such cellular responses to stimuli, biologists try to understand the actions of the single molecules in terms of elementary (“microscopic”) molecular processes, such as the opening or closing of ion channels in the membrane, the phosphorylation or dephosphorylation of intracellular proteins, or the nuclear translocation of transcription factors [2, 3]. Formally, the dynamics of these processes are typically described by molecular kinetic schemes or, mathematically, Markov-chain state models. Each state in the model represents, for example, a molecular conformation or ligand binding-site occupancy; rate constants govern the transitions between the states (e.g., Fig 1). These types of schemes have been successfully applied to describe many kinds of cellular systems, including the activity of ion channels [4, 5], pharmacology of receptors [6–8], and flux through metabolic pathways [2].
Clearly, it is desirable to study these microscopic processes directly. This has been possible, to some extent, in a few areas, notably the biophysics of ion channels, an exceptionally favorable case where the exquisitely sensitive patch-clamp technique [9, 10] is able to record the activity of single ion channels. As a result, this technique provides some measure of direct information about microscopic processes such as the opening and closing of the channels and allows the construction of molecular kinetic models of these processes [11, 12].
In most areas of cell biology, however, such direct access to the microscopic processes is not yet possible. Although new single-molecule techniques provide ways of understanding the properties of key biological molecules in isolation, monitoring the dynamics of single molecules inside cells is technically complex and in many experimental systems not practical (see [13, 14] for reviews).
The information that is routinely available from cells exposed to stimuli is the overall (“macroscopic”) cellular response (e.g., Fig 1A). This response combines the signal from a large number of molecules and multiple microscopic processes [15, 16]. The macroscopic response trace thus contains information about the microscopic processes, and, since it is relatively easy to record in most experimental systems, an attractive approach is to attempt to extract information about the microscopic processes, and the way in which they are combined in the molecular kinetic scheme, from the macroscopic trace.
This, however, is a challenging problem. Although, in theory, the information about all of the relevant microscopic processes is present in the macroscopic trace, the trace combines the information about the individual processes in a way that is often complicated and difficult to interpret immediately in terms of the underlying molecular kinetic scheme [17]. For example, the work with ion channels has demonstrated that very similar macroscopic responses can be generated by quite different arrangements of the microscopic processes in the molecular kinetic scheme [18] (as can be seen also in Fig 1B).
One way to simplify the problem is to assume a pre-existing molecular kinetic model of the system, based on other kinds of experimental data or simply educated guesswork. Thus, in the ion-channel field, elegant computational methods have been introduced to analyze macroscopic traces of ion currents [19, 20], but these methods rely strongly on specific biophysical models of the ion channels and thus are not immediately generalizable to the study of other biological systems.
In this study, we have developed a general method, the SYStems-based MOLecular kinetic scheme Extractor (SYSMOLE), to obtain molecular kinetic schemes from macroscopic traces. SYSMOLE is a hybrid computational-analytical method that, because it does not assume any pre-existing model of the system, can be applied to experimental traces recorded, in principle, in any biological system.
In practical terms, to successfully derive the molecular kinetic scheme underlying a macroscopic trace, a method must perform three distinct tasks: a) it must accurately model the dynamics of the trace, b) it must ensure that this model is unique, and c) it must convert this mathematical model into a biologically interpretable model. SYSMOLE includes a module to tackle each of these tasks (Fig 1C). The Identifier Module accurately captures the dynamics of the macroscopic trace and summarizes this information in the form of a transfer function. The Classifier Module derives a block diagram, a description of how different processes are combined, that is uniquely matched to the transfer function. Finally, the Molecular Kinetic Converter (MKC) Module utilizes both the transfer function and the block diagram to derive a molecular kinetic scheme that can be interpreted biologically in terms of conformational changes of proteins or chemical reactions.
We first describe the implementation of each of the three modules that comprise SYSMOLE. For didactic simplicity, we focus in our presentation on a second-order system, in which two first-order processes are combined in different arrangements; however, our approach is scalable to higher-order systems (S1 Text section 3). Next, we provide a rigorous evaluation of the performance of SYSMOLE in the presence of noise using synthetic traces. We then report on the performance of SYSMOLE on real experimental traces. We show that SYSMOLE was able to derive the correct molecular kinetic scheme of the inactivation of L-type calcium channels as well as the effects of the calcium-channel blocker nifedipine from macroscopic calcium-current traces. Finally, we describe the use of SYSMOLE to decipher the crosstalk mechanism, through the heteromeric receptor complex formed by the metabotropic glutamate receptor 2 (mGluR2) and the serotonin receptor 2A (5-HT2AR), of a new class of glutamate antipsychotics.
We decided to test first the ability of SYSMOLE to tackle a general class of problems in biology: distinguishing whether two similar macroscopic traces arise from different molecular kinetic schemes. Elucidating the molecular mechanisms of ion-channel inactivation from macroscopic current traces constitutes a classic biophysics problem in this class (see [4, 18, 31] for a review on the topic). Many of these molecular mechanisms were identified in the eighties and the nineties for different types of ion channels through single-channel electrophysiology recordings, fluctuation analysis, and analysis of gating currents [6, 10–12, 32]. However, analogous problems emerge in a great number of biological systems for which it is hard to design experiments to tease molecular schemes apart, or single-molecule experimental techniques are simply not yet available.
First we programmed an additional external module, the Synthetic Trace Simulator. This module generates a trace simulating each of the five molecular kinetic schemes associated with each configuration (Fig 4): first-order, second-order cascade, second-order feedback, second-order parallel addition, and second-order parallel subtraction. For practical purposes, in our implementation each molecular kinetic scheme is characterized by the parameters of two first-order processes a and b (ka, kb, τa and τb) and the kinetic parameters σk and γ derived from the equations detailed in the previous section. As a proof of concept, we used the Synthetic Trace Simulator to simulate voltage-clamp experiments and generate, in response to a step in voltage, the macroscopic current trace for molecular kinetic schemes that include first-order processes describing the activation and inactivation of ion channels (Fig 5). Emulating the classic biophysics problem, we compared two extremely similar macroscopic traces arising from different molecular kinetic schemes (compare left panels on Fig 5A and 5B). The first simulated current trace (Fig 5A) emerged from a molecular scheme in which inactivation affected both the closed and the open states of the channel (Fig 4D, black: closed state, blue: inactivated states, red: open state). The trace was generated with activation (a) and inactivation (b) processes defined by parameters ka = -5, kb = 2, τa = 5 ms, and τb = 200 ms. The second simulated current trace was generated with a molecular kinetic scheme where inactivation only acted on the open state (Fig 4C, black: closed state, red: open state, blue: inactivated state). This trace was obtained with a molecular kinetic scheme associated with first-order activation (a) and inactivation (b) processes with parameters ka = -5, kb = 3, τa = 5 ms, and τb = 100 ms. We then analyzed both traces with SYSMOLE: first we identified the dynamics of the macroscopic trace elicited by the voltage step and summarized them in a transfer function G(s), then we classified G(s) to obtain the underlying configuration, and finally we obtained the corresponding molecular kinetic scheme using the MKC Module (Fig 5).
Our results indicate that our methodology can accurately retrieve the molecular kinetic scheme of these two similar traces. Analogous tests with less similar traces were also performed with successful results. In order to control for a possible bias associated with using the same environments for simulation and analysis, two different implementations of the Synthetic Trace Simulator were tested with practically identical results (see Materials and Methods).
Our previous results had concluded that the system-identification approach presented in the first part of the Results section (SYSMOLE) could be successfully applied to accurately elucidate the molecular kinetic scheme underlying a given macroscopic trace. However, these studies had been performed in the absence of noise, an unrealistic situation in any biological experimental setting. We decided to test the effects of noise on the performance of SYSMOLE.
The robustness to noise of SYSMOLE will strongly depend on the features of the signals of interest and the type of noise existing on the trace. A useful approach to establish the reliability of SYSMOLE in the presence of noise is to simulate the types of experimental traces we are studying and test SYSMOLE when different levels of noise are added. We studied the effects of noise on the kinetic parameter values of the molecular kinetic scheme and the probability of error in classification of traces in which we observe a fast-activating process followed by a slower inactivation at the macroscopic level (Fig 5). These traces are representative of the types of traces obtained in voltage-clamp experiments with L-type calcium channels, and in the GPCR-heteromer signaling measurements presented in subsequent sections. By inspection of the traces (Figs 5 and 6A) one can easily conclude that the two processes are not combined in cascade, since this would be incompatible with a signal that reverses direction in the y axis (see Fig 2). Therefore, we focused our analysis on how noise affects the distinction between the feedback and parallel configurations, which based on our preliminary tests, is more prone to error due to the presence of noise in the trace. Noise tests were performed for all configurations and a wide range of combinations of first order parameters for the activation and inactivation processes with similar results.
Specifically, with the aid of the Synthetic Trace Simulator we generated synthetic traces involving two first-order processes characterized by parameters τa = 5 ms, τb = 100 ms, ka = −5, and kb = 3, using the molecular kinetic scheme associated with either the parallel subtraction (Fig 4E) or the feedback (Fig 4C) configurations. We generated multiple traces for each molecular kinetic scheme, to which we added increasing levels of white Gaussian noise (N = 100 simulations per level of noise). We analyzed each trace with SYSMOLE and measured (i) the relative error of the transition rate parameter values from the molecular kinetic schemes σ1, σ2 and σ3 (Fig 4D and 4E), and (ii) the probability of error in classification (Pe) defined as the ratio between the number of correct classifications and the total number of simulations at a given level of noise (Fig 6B). It should be noted that the scaling factor γ is an adjustable parameter that shapes the amplitude of the signal and its sensitivity to noise has less relevance than the transition rates or the configuration.
In our analysis, traces arising from molecular kinetic schemes associated with the parallel and feedback configurations had similar robustness to noise (Fig 6B). The relative errors in the transition rate kinetic parameters began to degrade at a signal-to-noise ratio (SNR) of 47 dB, but stayed within a relative error of 0.5 at an SNR = 30 dB and above (see Materials and Methods section for a description of how the SNR was calculated). The probability of error in classification increased rapidly for traces with a SNR lower than 25 dB for the parallel configuration and 22 dB for the feedback configuration, reaching a Pe = 1 at 18 dB for both configurations. It is worth noting that the Pe represents not only the instances in which a trace in feedback configuration is mistakenly classified as a parallel configuration or vice versa, but also errors in which the trace is classified as arising from a higher-order (≥2) or first-order system. These two types of errors quickly dominate the probability of error in detection at high levels of noise as the SNR degrades.
Together, our results indicate that for the traces of interest one can perform a noise analysis with similar synthetic traces and establish for a given SNR the expected probability of error in classification and relative error in the transition-rate parameters of the molecular kinetic scheme. For traces of the types obtained in our experimental systems (L-type calcium channels and GPCR Gi and Gq), our results indicated that a SNR of 25 dB or better should provide the right identification of the molecular kinetic scheme and a low relative error in the kinetic parameters. Analogous studies revealed that a similar robustness to noise of SYSMOLE can be achieved in the presence of Brownian noise, and that the effect of noise on Pe can be improved through pre-processing by filtering the signal prior to application of SYSMOLE (S1 Text sections 4.1 and 4.2). We also established the potential applicability of SYSMOLE to recognizing gene-regulatory mechanisms associated with a feedback or parallel configurations in the presence of cell-to-cell variability in gene induction noise (S1 Text section 4.3).
We decided to apply SYSMOLE to study the molecular mechanism of voltage-dependent inactivation of L-type calcium channels and the mechanism of action of nifedipine, a calcium channel blocker widely used clinically as an antianginal and antihypertensive drug [33, 34]. The inactivation mechanism of L-type channels has been extensively studied, and two major inactivation molecular mechanisms have been identified and characterized: a fast calcium-dependent inactivation mediated by the calcium ions entering inside the cells upon opening of the channel, and a voltage-dependent inactivation mediated by the voltage sensor located in the principal subunit and its movement upon depolarization to block the channel (see [35] for a review). Nifedipine, as part of the dihydropyridine class of channel blockers, binds to specific regions of the principal subunit to block the channel. The affinities of nifedipine for these binding regions differ depending on the state of the channel: nifedipine has a high affinity site for the inactivated state, a low affinity site accessible in the open state, and virtually no affinity sites for the closed state of the channel [36–39]. We tested whether we could derive molecular kinetic schemes consistent with the known molecular mechanism of voltage-dependent inactivation and nifedipine blockade of the L-type channels from experimental macroscopic current traces with the help of SYSMOLE.
We used macroscopic current traces from voltage clamp (VC) experiments performed in the accessory radula closer muscle of Aplysia californica previously obtained by our group [40]. Currents through these channels were isolated from potassium currents by adding selective blockers (see Materials and Methods) and obtained in response to a depolarization step from - 90 mV to 0 mV. Calcium ions were replaced by barium in the solutions to remove the contribution of calcium-dependent inactivation and isolate the voltage-dependent inactivation of the channels. The currents were measured in the absence or presence of nifedipine (100 nM and 1 μM) (see Fig 7A for representative current traces). Following the noise analysis described in the previous section, the SNR of all traces (> 38 dB for all cases) predicted a low probability of error in the identification by SYSMOLE of the correct molecular kinetic scheme as well as a low relative error for the transition rate parameters σk (Fig 6 and Table 1). In VC experiments the voltage is held constant while the current flowing through the ion channels in the cell membrane is measured based on the amount of current that an amplifier needs to supply to maintain the set voltage. We used the voltage as a function of time V(t) (a step from - 90 to 0 mV) as the input trace, and the recorded current i(t) as the output trace for SYSMOLE.
In biological terms, we can explain the current trace i(t) elicited by a step in voltage in a VC experiment using a molecular kinetic scheme as follows. Initially, the system is at steady-state and the fractional occupancies (i.e. the fraction of channels in any particular state) for each of the states are constant and the population of channels in dynamic equilibrium. At this point in time, the current readout is also constant (basal current) and proportional to the fractional occupancy in the open state. When the input steps to a higher voltage, the rate constants which govern the transitions between states change and the population of channels redistributes accordingly among the states. This redistribution is accompanied by a transient in the fractional occupancies, and thus in the current recorded. At the end of this transient a new steady-state level of current is reached. The value of the rate constants characterizing the transitions before the step in voltage are unknown, and therefore, the σk strictly represent the change in rate associated with the input, in this case a step in voltage.
The results of applying SYSMOLE to derive molecular kinetic schemes from macroscopic traces of L-type calcium channel currents obtained in the voltage-clamp (VC) experiments described, consistently yielded the molecular kinetic scheme associated with a feedback configuration (Fig 7B and Table 1). In the absence of nifedipine, inactivation of the channel is due to the voltage-dependent inactivation, which is engaged when the channel opens due to the depolarization of the membrane. As such, a molecular kinetic scheme in which the channels need to open to inactivate is consistent with this molecular mechanism (Fig 7B) [41]. Based on the known mechanism of action of nifedipine, and due to the high affinity binding of nifedipine to the inactivated state, we expected the presence of this drug to reduce the transition rate from the inactivated state to the open state (I→O). Similarly, due to the existence of a binding site for nifedipine in the open state of the channel, we also expected an increase in the transition rate between the open state and the inactivated state (O→I), which now includes two different states the one reached through voltage-dependent inactivation, and the inactivation due to the presence of nifedipine. Finally, we expected nifedipine not to affect the transition between the closed and the open state (C→O) since it does not bind to the channel in the resting closed state. All these effects were extracted by SYSMOLE as detailed in Table 1, and illustrated by the temporal evolution of the fractional occupancies in the obtained molecular kinetic scheme (Fig 7C).
Together, our results showed that SYSMOLE performed strongly with experimental traces and allowed us to derive the correct molecular kinetic scheme of the voltage-dependent inactivation of the L-type calcium channels and to predict the molecular mechanism of action of nifedipine.
G protein-coupled receptors (GPCRs) are membrane-bound receptors that transduce extracellular binding of molecules into intracellular signals by activating a class of heterotrimeric proteins called G proteins [42, 43]. Each subclass of G protein is associated with a signaling pathway with specific actions inside the cell in response to stimuli [44]. Classically, the signaling paradigm was considered to follow the rule that one ligand binds to one receptor which activates only one pathway. However, extensive biochemical and biophysical evidence has revealed the existence of GPCR homo- and hetero-dimers/oligomers that differentially alter G protein signaling. Furthermore, the regulation of these complexes is found to play a critical role in normal physiology and disease (see [45] for a review). Despite their importance, the molecular signaling mechanisms of GPCR heteromeric and homomeric complexes remain largely unknown.
The dysregulation of the heteromeric complex formed by the Gi-coupled metabotropic glutamate receptor 2 (mGluR2) and the Gq-coupled serotonin 2A receptor (5-HT2AR) has been linked to schizophrenia [46]. Preclinical and clinical studies suggest that activation of mGluR2 in the complex by allosteric and orthosteric agonists could represent a new therapeutic approach to treat schizophrenia and other disorders [47–49]. Our group has published work showing that psychedelic drugs upset the Gi/Gq signaling balance associated with the mGluR2/5-HT2AR complex by reducing Gi and increasing Gq signaling, while antipsychotic drugs restore the natural Gi/Gq balance. This is achieved through a crosstalk mechanism in which a ligand acting on one of the receptors, either 5-HT2AR or mGluR2, can change the signaling on their counterpart receptor [50]. However, the mechanism by which this crosstalk is achieved molecularly through the mGluR2/5-HT2AR complex remains to be elucidated.
Two possible molecular mechanisms occurring through cross-conformational changes have been postulated to explain the crosstalk observed between mGluR2 and 5-HT2AR through the mGluR2/5-HT2AR complex. One referred to as cis-activation, in which the ligand-free receptor is not able to activate G proteins unless the first one signals and is bound to G proteins. The second one referred to as trans-activation, in which the signal from the ligand-bound receptor is transmitted to the neighboring receptor which then signals by activating G proteins.
The cis-activation and trans-activation crosstalk theories could be distinguished in terms of molecular kinetic schemes (Fig 8A). We decided to apply SYSMOLE to derive the molecular kinetic scheme from macroscopic traces obtained from mGluR2/5-HT2AR in response to LY379268, a ligand belonging to the promising new class of glutamate antipsychotics [51], which has been shown to elicit a strong crosstalk between the mGluR2 and the 5-HT2AR [46, 50]. While glutamate and serotonin, the two endogenous neurotransmitters, activate Gi and Gq pathways respectively through the mGluR2/5-HT2AR complex, LY379268 can activate both Gi and Gq despite only binding mGluR2 (Fig 8A).
Ion channels have been extensively used to measure GPCR signaling activity. For our study, we expressed the mGluR2 and 5-HT2AR receptors together with a G protein sensitive potassium ion channel (GIRK4*) in Xenopus oocytes and measured using two-electrode voltage clamp the current flowing through the channel as a function of time (see Materials and Methods). The input signal for SYSMOLE was a step in concentration of LY379268, and the output signal the macroscopic current through the GIRK4* channel. Since Gi signaling increases current through the channel (downward direction) and Gq signaling decreases the current through the channel (upward direction) [52] the traces obtained (Fig 8B) were perfectly suited to apply SYSMOLE to understand the relationship between Gi and Gq signaling and derive a molecular kinetic scheme. It should also be noted that the noise level in these traces results in a SNR clearly below the cut-off for error-free detection (Fig 6). Should the traces be noisier or have a type of noise added other than Gaussian, prior to analyzing the experimental traces SYSMOLE should be tested with synthetic traces and its robustness to noise for that type of trace determined (see S1 Text sections 4.1 and 4.2 for a description of the robustness of SYSMOLE to Brownian noise, and pre-processing strategies to increase the SNR in the trace).
Given that both the opening and closing of the reporter channel are faster processes (milliseconds) compared to Gi and Gq activation (seconds) the trace does not capture them, and the resulting molecular kinetic scheme allows us to distinguish states resulting from combinations in which the Gi is OFF (no change in current) or ON (current through the channel is increasing), Gq is OFF (no change in current) or ON (current through the channel is decreasing) or combinations.
Application of SYSMOLE to different macroscopic traces obtained in response to LY379268 (Table 2) yielded the molecular kinetic scheme associated with the parallel subtraction (Fig 8B right) with four states, which can be related to different signaling states depending on whether the Gi or Gq signaling is ON or OFF in the complex. Additional controls were performed with injection of only one receptor (mGluR2 or 5-HT2AR) and endogenous ligands to ensure that the traces captured the crosstalk effects through mGluR2/5-HT2AR. From the molecular kinetic scheme, it can be inferred that Gi and Gq signaling are processes that occur in parallel upon LY379268 binding to mGluR2, and that activation of Gq does not require Gi activation. A close look at the fractional occupancy in each state as a function of time reveals that due to the strong Gi signal activation of mGluR2 by LY379268 (dominant agonist [50]), most of the complex molecules move from the Gi OFF/ Gq OFF state to Gi ON/ Gq OFF state and then transition directly to the Gi ON/ Gq ON state. Interestingly, there is also a fraction of the molecules that transition from the original Gi OFF / Gq OFF state directly to Gi OFF/ Gq ON state, a molecular state that had not been postulated before. This fraction, albeit small, amounts to approximately one third of the total fractional occupancy in the Gi ON/Gq ON state when the system reaches steady state Our results support the hypothesis of a trans-activation as the mechanism of crosstalk of the mGluR2/5-HT2AR heteromeric complex. It should be noted that the assignment of these ON and OFF states does not come directly from SYSMOLE, which is agnostic to the biological interpretation and is here used to generate hypotheses. The molecular existence of these states would need to be explicitly tested in further experiments (see Discussion).
Together, these results exemplify the usefulness of the SYSMOLE system-identification analytical tool in studying heteromeric signaling.
A correct characterization of the dynamics of different molecular processes in the cell is fundamental for understanding normal physiological and pathophysiological responses. In most biological systems the direct study of these microscopic molecular processes is not yet possible due to experimental limitations, and only overall macroscopic cell responses to stimuli (macroscopic traces) can be obtained. Here we introduced a method that combines computational and analytical approaches to extract information from the macroscopic trace regarding the molecular microscopic processes and the way in which they are combined. This method, SYSMOLE, conveniently describes the dynamics of these microscopic processes in the form of a molecular kinetic scheme analogous to those used in biophysics and pharmacology.
SYSMOLE is implemented in modular fashion. We first detailed the mathematical underpinnings of each of the three modules that are included in SYSMOLE: the Identifier Module, the Classifier Module, and the Molecular Kinetic Converter Module. Through rigorous and systematic evaluations, we explored SYSMOLE’s limitations and robustness to noise (see also S1 Text section 4). We validated the performance of SYSMOLE on experimental traces by correctly identifying the molecular kinetic scheme describing the activation and inactivation of L-type calcium channels, as well as the mechanism of action of nifedipine, a calcium channel blocker clinically used in patients with cardiovascular disease. Furthermore, we applied SYSMOLE to study the signaling crosstalk mechanism observed in the heteromeric complex formed by mGluR2 and 5-HT2AR in response to LY379268, a compound that belongs to the new class of glutamate antipsychotics [53]. A Matlab toolbox with all the functions that we have developed is available in Matlab Central (https://www.mathworks.com/matlabcentral/fileexchange/61465-sysmole). The SYSMOLE toolbox contains an implementation of the modules for second-order systems and third-order systems with three poles and two zeros. It contains the experimental data used in the manuscript, and also provides the user with templates to simulate and test noise models and apply SYSMOLE to traces from their biological system of interest.
In the broader context of systems biology, other powerful methods exist to extract network models from biological data in different domains (see reviews [54] and [55] for examples in allosteric networks and functional genomics, respectively). As one could expect, each method and approach has its strengths and is most adapted to deal with specific aspects or features of the data and the biological system. A large group of methods obtain possible networks from scarce and unhomogeneously-sampled measurements or traces [54–60]. SYSMOLE tackles a different set of questions and is best adapted to capture subtle features in adequately-sampled macroscopic traces that underlie essential differences in the biological system. As such, SYSMOLE is a complementary approach to other methods, and as the experimental techniques evolve to enable sampling more often and acquiring more data per unit time, tools like SYSMOLE will become particularly useful to have in the computational biologist toolbox.
Compared to other methods that deal with adequately-sampled data, SYSMOLE is a flexible method that can be generalized to the study of multiple biological systems and applied to different types of biological traces. Various methods have been developed to derive molecular kinetic schemes from macroscopic current traces in the ion channel biophysics field. The best results have been obtained utilizing maximum likelihood methods [19, 20] and covariance methods [61]. These methods strongly rely on underlying models to characterize ion channel current flow which, following decades of biophysics research, are usually available [4]. Unfortunately, these methods cannot be easily generalized to most biological systems, for which reliable models of the molecular microscopic processes are typically unavailable and large numbers of measurements are impractical. Furthermore, many of the available methods are not able to tackle stimuli (inputs) with different dynamics, a key feature of many biological systems and physiological responses. SYSMOLE overcomes these limitations. It does not rely on underlying models, it extracts information per input-output trace pair, and it incorporates information about the dynamics of the stimulus. Utilizing a system identification step based on a transfer function allows SYSMOLE to capture the dynamics without relying on a physical model. By extending the concept of a Markov chain and linking one of the states to the input, SYSMOLE can extract the molecular kinetic scheme from macroscopic traces elicited by inputs with various dynamics. Finally, the flexible properties of SYSMOLE have the added benefit to retrospectively allow leveraging of important biological information from the vast repository of macroscopic traces collected through the years by different labs.
The overall performance of SYSMOLE is determined by the module with the lowest performance, which in turn depends on the characteristics of each particular macroscopic trace and physiological process under study. The performance of the ARX (autoregressive) method used in the Identifier Module will degrade as the noise increases, or as the sampling frequency decreases and the input and output traces become insufficiently sampled in time. Based on our experience these limitations can be overcome through the application of classical preprocessing and filtering techniques to the traces [22] (see S1 Text section 4.1) together with an acquisition sampling frequency of 12T, with T being the time constant of the fastest process that we wish to capture. The Classifier Module will not perform properly if information regarding the differences in time constants of the processes under study is unavailable. Lacking this information will result in an improperly bound optimization problem which will not exclude the parameter region where the feedback and parallel solutions coincide (S1 Text section 1). This might hinder the applicability of the method to processes characterized by extremely similar time constants. Finally, the Molecular Kinetic Converter is an analytical conversion, and thus in principle will always perform correctly. To confirm the validity of the conversion for every case, internal checks are included along the mathematical derivation to ensure the rules presented are sound (S1 Text section 2).
For didactic simplicity, we focused in our presentation on a second-order system, in which two first-order processes are combined in different arrangements. However, each module in SYSMOLE can be scaled to multiple processes. The ARX implementation of the Identifier module can easily extract up to 6 to 10 processes and does not represent a bottleneck for scalability [24, 25]. The series of comparisons implemented in the Classifier Module can be updated to incorporate additional combinations of numbers of poles and zeros. The optimization problems needed to differentiate these additional combinations are also easily scalable to higher-order systems, and thus to multiple processes. Finally, the Molecular Kinetic Scheme can be scaled using a modular approach in which each combination of two processes is considered as a module, and the connection of that module to another module or process also occurs through three possible canonical configurations (see the Scalibility section in S1 Text for a description of the application of SYSMOLE to third-order systems with three poles and two zeros).
Application of SYSMOLE to the study of crosstalk through the mGluR2/5-HT2AR heteromeric receptor complex in Xenopus oocytes illustrates its strength in generating new hypotheses regarding molecular mechanisms. Our results indicate that LY379268, a glutamate antipsychotic that binds to mGluR2, can signal through Gq in the absence of any ligand bound to 5-HT2AR, which suggests trans-activation as the molecular mechanism of crosstalk. This analysis postulated the existence of a signaling state in which the receptor signals through Gq without signaling through Gi. This new state might have important implications in understanding the pharmacology of new glutamate antipsychotic compounds [53, 62, 63]. Since SYSMOLE strictly obtains a Markov-chain state model, the assignment of each of the states to a particular signaling conformation (Gi ON or OFF, Gq ON or OFF) is also postulated. Confirmation of the existence of these states will require further experiments in which it is possible to assume that the occupancies of the states of one or more states is constant or altered. One could for example perform similar crosstalk experiments (Fig 8) with complexes in which either of the receptors that form the heteromer are mutated such that it cannot bind Gi or Gq. Furthermore, relevance of this new postulated signaling state in neurons should also be assessed.
In conclusion, despite the extensive efforts devoted to the development single-molecule experimental techniques, alternatives are needed for the analysis of macroscopic traces. SYSMOLE, a hybrid computational and analytical tool designed to analyze macroscopic traces and derive molecular kinetic schemes, enables the user to extract information regarding the microscopic processes involved in the response to a given stimulus from the available macroscopic traces. SYSMOLE has also the added benefit to be able to leverage important biological information from the vast repository of macroscopic traces to generate new hypotheses.
The Identifier Module was implemented in Matlab (www.mathworks.com) using the System Identification Toolbox ARX functions. Details on loss-function minimization strategies in prediction-error-identifications can be found in [25].
The parallel and feedback optimization problems described in the main text were solved using a trust-region-reflective algorithm [64] included in the Optimization Toolbox of Matlab (www.mathworks.com).
Details on the derivation of the molecular kinetic schemes for each canonical configuration can be found in the supporting materials (S1 Text).
Two implementations were used: one in Matlab, the environment in which SYSMOLE is implemented, and one using COPASI [65], a software application for simulation and analysis of biochemical networks and their dynamics.
In order to obtain the signal-to-noise ratio (SNR) one needs to calculate the average power of the signal y(t) PY, as well as the noise power. The signal’s average power can be directly calculated in the time domain as:
PY=1N∑n=0∞y[n]2
(47)
where y[n] is the sequence resulting from discretizing y(t) with sample period Ts and sample number n = 1 … N.
The noise power depends on the noise characteristics and the bandwidth of y(t), which varies depending on the parameters of the first order processes that give rise to the signal (ka, kb, τa, and τb). All the necessary information needed to calculate this value is provided by the Identifier Module. On the one hand, in the Identifier Module we had included disturbances in the form of white Gaussian noise with variance λ filtered and added to the output (see Identifier). On the other hand, this module yields the transfer function G(s) that will allow us to calculate the bandwidth of the signal BW. In order to test whether a noise characterization as WGN is adequate for the type of experimental traces we analyzed, we confirmed that the λ value estimated by the Identifier Module matched the value of the variance of y(t) before stimulation both for the simulated synthetic traces as for the experimental traces. Once λ and BW are obtained, the noise power PN filtered to the signal’s bandwidth can be calculated as
PN=∫−∞∞|G(f)|2Sn(f)df=∫−∞∞|G(f)|2λdf
(48)
where G(f) is the frequency response of the transfer function G(s), and Sn(f) is the power spectral density of the noise. Assuming an ideal low-pass filter, the value of this integral can be approximated to
PN≃2λ.BW
(49)
For our analysis we used Eq (49) with a bandwidth calculated at 3 dB drop in gain. A design of a non-ideal low-pass filter with cutoff frequency equal to BW, 1 dB ripple in the pass band, and 60 dB attenuation in the stop band yielded practically identical results.
Digitized current and voltage values were obtained from previous voltage-clamp experiments performed in the accessory radula closer muscle of Aplysia californica. Solutions, drug delivery, and experimental conditions are described in detail in reference [66].
Oocytes were isolated and microinjected with equal volumes (50 nl), as previously described [67]. In all two-electrode voltage-clamp experiments (TEVC), oocytes were injected with 1 ng of mGluR2, 2 ng of 5-HT2AR, and 2 ng of GIRK4*, and were maintained at 18 ºC for 14 days before recording.
Whole-cell currents were measured by conventional two-electrode voltage-clamp (TEVC) with a GeneClamp 500 amplifier (Axon Instruments, Union City, CA, USA. A high-potassium (HK) solution was used to superfuse oocytes (96 mM KCl, 1 mM NaCl, 1 mM MgCl, 5 mM KOH/HEPES, pH 7.4) to obtain a reversal potential for potassium (Ek) close to zero. Inwardly rectifying potassium currents through GIRK4* were obtained by clamping the cells at - 80 mV. A solution of 3 mM of BaCl in HK solution was perfused at the end of each trace to ensure that the current measured corresponded to GIRK4*, as previously described [52].
|
10.1371/journal.pgen.1003555 | Distinct and Atypical Intrinsic and Extrinsic Cell Death Pathways between Photoreceptor Cell Types upon Specific Ablation of Ranbp2 in Cone Photoreceptors | Non-autonomous cell-death is a cardinal feature of the disintegration of neural networks in neurodegenerative diseases, but the molecular bases of this process are poorly understood. The neural retina comprises a mosaic of rod and cone photoreceptors. Cone and rod photoreceptors degenerate upon rod-specific expression of heterogeneous mutations in functionally distinct genes, whereas cone-specific mutations are thought to cause only cone demise. Here we show that conditional ablation in cone photoreceptors of Ran-binding protein-2 (Ranbp2), a cell context-dependent pleiotropic protein linked to neuroprotection, familial necrotic encephalopathies, acute transverse myelitis and tumor-suppression, promotes early electrophysiological deficits, subcellular erosive destruction and non-apoptotic death of cones, whereas rod photoreceptors undergo cone-dependent non-autonomous apoptosis. Cone-specific Ranbp2 ablation causes the temporal activation of a cone-intrinsic molecular cascade highlighted by the early activation of metalloproteinase 11/stromelysin-3 and up-regulation of Crx and CoREST, followed by the down-modulation of cone-specific phototransduction genes, transient up-regulation of regulatory/survival genes and activation of caspase-7 without apoptosis. Conversely, PARP1+-apoptotic rods develop upon sequential activation of caspase-9 and caspase-3 and loss of membrane permeability. Rod photoreceptor demise ceases upon cone degeneration. These findings reveal novel roles of Ranbp2 in the modulation of intrinsic and extrinsic cell death mechanisms and pathways. They also unveil a novel spatiotemporal paradigm of progression of neurodegeneration upon cell-specific genetic damage whereby a cone to rod non-autonomous death pathway with intrinsically distinct cell-type death manifestations is triggered by cell-specific loss of Ranbp2. Finally, this study casts new light onto cell-death mechanisms that may be shared by human dystrophies with distinct retinal spatial signatures as well as with other etiologically distinct neurodegenerative disorders.
| The secondary demise of healthy neurons upon the degeneration of neurons harboring primary genetic defect(s) is hallmark to neurodegenerative diseases. However, the factors and mechanisms driving these cell-death processes are not understood, a severe limitation which has hampered the therapeutic development of neuroprotective approaches. The neuroretina is comprised of two main types of photoreceptor neurons, rods and cones. These undergo degeneration upon heterogeneous mutations or environmental stressors and the underlying diseases present conspicuous spatiotemporal pathological signatures whose molecular bases are not understood. We employed the multifunctional protein, Ran-binding protein-2 (Ranbp2), which is implicated in cell-type and stress-dependent clinical manifestations, to examine its role(s) in primary and secondary photoreceptor death mechanisms upon its specific loss in cones. Contrary to prior findings, we found that dying cones can trigger the loss of healthy rods. This process arises by the immediate activation of novel Ranbp2-responsive factors and downstream cascade events in cones that promote extrinsically the demise of rods. The mechanisms of rod and cone demise are molecularly distinct. Collectively, the data uncover distinct Ranbp2 roles in intrinsic and extrinsic cell-death and will likely contribute to our understanding of the spatiotemporal onset and progression of diseases affecting photoreceptor mosaics and other neural networks.
| The disintegration of neuronal networks owing to the non-autonomous death of neurons without primary damage is a hallmark manifestation of many neurodegenerative diseases and contributes determinately to their onset or progression [1]–[3]. Cone or rod photoreceptor neurons employ cell type-specific spectrally tuned and highly homologous phototransduction cascades. Neurodegenerative disorders affecting these neurons serve as excellent models to understand autonomous and non-autonomous cell death processes. First, early studies with chimeric mice with a mixture of healthy and unhealthy rod photoreceptors owing to the expression of rod-specific degenerative mutations by the latter showed that damaged rod photoreceptors promote the non-autonomous death of healthy rod photoreceptors [4]–[7], but the analogous event does not appear to occur between neighboring healthy and damaged cone photoreceptors [8]. Second, rod photoreceptor-specific mutations causing the death of rod photoreceptors promote ultimately the non-autonomous death of cone photoreceptors [3], [9]–[12]. This secondary loss of cone photoreceptors has the greatest impact on human vision, because cone photoreceptors mediate daylight and high acuity vision as well as color perception. For example, rod photoreceptor-specific mutations affecting phototransduction components of rod photoreceptors, such as the catalytic subunit of cGMP phosphodiesterase, rhodopsin and cyclic nucleotide-gated (CNG) channel subunits, lead to the degeneration of damaged rod and healthy cone photoreceptors [4], [9], [11]–[16]. By contrast, the cellular effects of cone photoreceptor-specific mutations causing cone degeneration are much less clear, but they are thought to spare the viability of rod photoreceptors. For example, cone-specific mutations impairing genes homologous to those of rod phototransduction cause the death of cone photoreceptors only [17]–[26].
A number of distinct models have been put forward to explain the secondary loss of healthy cone photoreceptors upon the primary degeneration of damaged rod photoreceptors. These include oxidative stress and metabolic imbalance caused by increased oxygen tension [27]–[29], loss of paracrine neurotrophic [30] or vasculotrophic support [31], microglia activation [32], [33] and release of rod-derived toxic byproducts [34]. Distinguishing what mechanisms trigger extrinsic-elicited cell death pathways is further complicated by our limited knowledge of the intrinsic and primary cell-death pathways affecting rod photoreceptor themselves, and whether these are consistent across rod degeneration models [35]–[45]. Likewise, the knowledge about the molecular and subcellular events underlying the primary demise of cone photoreceptors is very limited [19], [46]. Ascertaining the spatiotemporal processes causing autonomous and non-autonomous neural death is critical to our understanding of the pathogenesis of a variety of human photoreceptor dystrophies, such as retinitis pigmentosa (RP) and age-related macular degeneration (AMD), that harbor hallmark spatiotemporal manifestations presumably driven by distinct intrinsic and extrinsic factors.
The pleiotropic protein, Ran-binding protein-2 (RanBP2), is essential for organism viability and energy metabolism [47], [48]. Prior studies on RanBP2 indicate that it plays critical cell-type-dependent physiological roles in mediating gene-environment interactions. In this regard, distinct disease stressors, such as phototoxicity [49], [50], Parkinsonian toxic insults [51], and carcinogens [48], trigger a variety of cell-context-dependent clinical and pathophysiological manifestations upon partial deficits or mutations of Ranbp2. Further, semi-dominant mutations in human RANBP2 cause either acute necrotizing encephalopathies (ANE1) or acute transverse myelitis (ATM) upon exposure to a variety of infectious agents [52]–[54]. In this study we set out to determine the intrinsic and extrinsic effects of lack of Ranbp2 function in the survival of cone or rod photoreceptor neurons upon selective ablation of Ranbp2 in cone photoreceptors, where Ranbp2 is highly expressed [55]. We show that cone-specific ablation of Ranbp2 promotes the autonomous non-apoptotic death of cone photoreceptors and the cone-dependent apoptotic demise of rod photoreceptors by distinct cell-type death mechanisms. Hence, a primary impairment of cone photoreceptors can promote the secondary death of healthy rod photoreceptors, a paradigm-shift observation with implications to our understanding of human neurodegenerative diseases affecting distinct photoreceptor cell types with hallmark regional distributions in the retina and other neural networks of the central nervous system.
To determine the physiological role of Ranbp2 in cone photoreceptors and uncover the effects of its genetic ablation in autonomous and non-autonomous molecular and cellular events affecting targeted cones and healthy rod photoreceptors, Ranbp2 was selectively targeted in mouse cones by Cre-mediated recombination of floxed Ranbp2 [48], [56], [57] (Figure 1A). Hemizygous transgenic mice expressing Cre under control of the R/G cone opsin promoter (Tg-HRGP-cre) [56], [57] were crossed with Ranbp2Flox/+ mice [48] to produce Tg-HRGP-cre:Ranbp2Flox/+. These mice were then crossed to Ranbp2Flox/+ or Ranbp2Gt(pGT0pfs)630Wcs/+ , which harbors a constitutively disrupted allele of Ranbp2 [47] (Figure 1B), to generate the lines, Tg-HRGP-cre:Ranbp2Flox/Flox and Tg-HRGP-cre:Ranbp2Flox/Gt(pGT0pfs)630Wc. These lines were morphologically and functionally indistinguishable from each other and they are hereafter designated as HRGP-cre:Ranbp2−/−. An out-of-frame Ranbp2 transcript comprising the fusion of exons 1 and 3 is produced at P7 upon Cre expression at P6 (Figure 1C). Cre was specifically expressed in cell bodies of M- and S-cone photoreceptors (Figure 1D, Figure S1).
We examined the temporal effects of loss of Ranbp2 expression in the morphology and survival of cones by comparing the immunostaining of retinal sections between HRGP-cre:Ranbp2−/− and HRGP-cre:Ranbp2+/− mice with anti-Cre and cone-specific anti-arrestin-4 (Arr4) antibodies [58] at P9, P13, P20 and P27 of age (Figure 2). In both genotypes, the cell bodies of Cre-expressing cone photoreceptors migrated to the distal (outer) region of the outer nuclear layer (ONL) by P13, where the majority of cone cell bodies are typically localized, and the outer segment (OS) and synaptic pedicles developed properly. However, few Cre+-cell bodies appeared displaced in the proximal (inner) ONL of HRGP-cre:Ranbp2−/− mice (Figure 2D″). By P20, cones of HRGP-cre:Ranbp2−/− mice presented prominent swelling of the synaptic pedicles (Figure 2F′) and retraction of some cell bodies to the proximal region of the ONL (Figure 2F″). By P27, only a very few surviving cones were present in HRGP-cre:Ranbp2−/− (Figure 2H′–2H″); by 6 weeks resilient cones lacking outer segments were rarely present (Figure S2), whereas no cones were present in 12-week old mice (data not shown).
The degeneration of cone photoreceptors was quantitatively monitored in retinal flat mounts. We compared the number of M-cones and OS length between HRGP-cre:Ranbp2−/− and HRGP-cre:Ranbp2+/− mice (Figure 3). No differences were seen in the number of M-cones or in the length of their OS at P15. By P20, both measures were significantly decreased in HRGP-cre:Ranbp2−/− mice, and very few M-cones remained at P27 (Figure 3A–3C). Notably, the few surviving M-cones retained in P27 HRGP-cre:Ranbp2−/− mice still had OS which appeared of comparable length to those of HRGP-cre:Ranbp2+/− littermates (Figure 3A, 3C). At P20, the number of S-cones and the length of their OS were significantly decreased in HRGP-cre:Ranbp2−/− mice and prominent clumps of S-opsin were observed in the OS (Figure S3). We then examined the position of the Cre+-cell bodies within the proximal and distal ONL at peripheral and central regions of dorsal retinas (Figure 3D, 3E). Unlike HRGP-cre:Ranbp2+/− littermates, P13 HRGP-cre:Ranbp2−/− mice presented displaced Cre+-cell bodies in the peripheral and central regions of the proximal ONL. At P20, the number of Cre+-cell bodies had strongly decreased across the central and peripheral regions of the HRGP-cre:Ranbp2−/− retina, but this decrease was much more pronounced in the central retina (Figure 3E). By P27 and 3-months of age, respectively, very few and no Cre+ cells were observed in any regions of the retina (Figure 3E, data not shown).
OS shortening is thought to precede cell death of rod and cone photoreceptors [13], [28]. Hence, we employed multiple cell death markers to dissect out the molecular and subcellular processes underpinning the activation and progression of cell death between cone and rod photoreceptors of HRGP-cre:Ranbp2−/− mice. We used TUNEL staining to determine whether degenerating cone photoreceptors underwent apoptosis. As shown in Figures 4A and 4B, we did not identify any cell bodies that were TUNEL+ and Cre+. Instead, we found that all TUNEL+-cell bodies were Cre−, most likely rod photoreceptors, because these neurons comprise 97% of all photoreceptor cell types [59]. Morphometric analyses showed a drastic increase of TUNEL+-cell bodies at P20 (Figure 4B). Akin to the localization of Cre+ cells, this increase was significantly more pronounced also in central and peripheral regions of the distal ONL than in the counterpart proximal regions (Figure 4B).
To further differentiate TUNEL+-apoptotic cell bodies from necrotic cells among cone and rod photoreceptors, we carried out morphometric analysis of co-localization of TUNEL+ and cone Arr4+-cell bodies of retinal explants incubated with the membrane-impermeable and DNA-binding fluorescent dye, ethidium homodimer III (EthD-III). There were no Arr4+Tunel+ or Arr4+EthD-III+ cells in either genotype (Figure 4C). Instead, we found that HRGP-cre:Ranbp2−/− at P20 had about 70 and 55% of EthD-III+ Arr4− and TUNEL+Arr4− cells, respectively, with 30% of these cells being EthD-III+TUNEL+Arr4− (Figure 4C, 4D). The presence of any of these apoptotic or necrotic cell bodies were negligible in HRGP-cre:Ranp2+/− (Figure 4C, 4D). These data support that the TUNEL+Arr4−, EthD-III+ Arr4− and EthD-III+TUNEL+Arr4− cells represent different cell death stages of rod photoreceptors.
To establish unequivocally that TUNEL+Cre−-apoptotic cell bodies are rod photoreceptors neurons, retinal sections were co-immunostained for Nr2E3, a transcription factor specifically expressed in cell bodies of rod photoreceptors [60]. As shown in Figure 5A (a–b″″), all Cre+-cell bodies of either genotype were Nr2E3−, whereas many TUNEL+-cell bodies were Nr2E3+ in HRGP-cre:Ranbp2−/−. We also identified a small subpopulation of TUNEL+-cell bodies with Nr2E3 aggregation at discrete perinuclear foci (Figure 5A, c–d″″), suggesting that Nr2E3 localization changes and its expression decreases during rod photoreceptor death. Quantitative morphometric analysis of triple-immunostained retinas showed that about 3.2% of the total cells in HRGP-cre:Ranbp2−/− were TUNEL+, while 52% and 44% of these TUNEL+ cell bodies were Cre−NR2E3+ or Cre−Nr2E3−, respectively; the remaining 4% were dying rods with discrete perinuclear aggregation of Nr2E3 (Figure 5B). Further examination of the TUNEL+Nr2E3− and TUNEL+Nr2E3+ cells indicated that they were TUNEL+EthD-III+ or TUNEL+EthD-III− (Figure S4). These data support that TUNEL+Nr2E3+ and TUNEL+Nr2E3− cells also represent different stages of cell death of rod photoreceptors.
We next examined whether cone and rod photoreceptor degenerations were accompanied by the activation of caspases, a cardinal feature of cell death [61], [62]. We screened whole retinal extracts of HRGP-cre:Ranbp2−/− mice at P20 with substrates against specific caspases and found strong activation of caspases 3/7, mild activation of caspases 8 and 9, and no activation of caspases 1, 2 and 6, when compared to HRGP-cre:Ranbp2+/− mice (Figure S5). To define the spatiotemporal profile of caspase activation, we examined the activities of caspases 3 and 7 in retinal extracts and carried out morphometric analyses of retinal sections immunostained with antibodies against cleaved (activated) caspase 3, 7 or 9 of age-matched mice of both genotypes. In HRGP-cre:Ranbp2−/− mice, we found that the activities of caspase 3, caspase 7, or both, peaked at P13 (Figure 6A), well before the rise in TUNEL+-cell bodies in the ONL (Figure 4B), and that these activities were negligible by P27, when most cones had died (Figure 6A). To distinguish caspase 3 and 7 activities, we immunostained retinal sections of different ages for active caspase 3 and 7 and performed morphometric analysis (Figure 6B, 6C). These experiments showed that the majority of active caspase 3+-photoreceptor cell bodies were Cre−TUNEL− (Figure 6B), but there was also a small but significant fraction of active caspase-3+Cre+-cell bodies at P13 in HRGP-cre:Ranbp2−/− mice (Figure 6B, 6C). All types of active caspase 3+-photoreceptor cell bodies were drastically decreased in HRGP-cre:Ranbp2−/− mice and their presence were no different from HRGP-cre:Ranbp2+/− by P20 (Figure 6B, 6C a–a′″, d–d′″). By contrast, caspase 7+-cell bodies became prominent only at P20 and most were Cre+ (Figure 6C, b′–b′″, e′–e′″). Further, the temporal profiles of caspase 3 and 7 activities paralleled their transcriptional up-regulation (Figure S6). Collectively, the data support that activation of caspase 3 in rods and caspase 7 in cones contribute to the rise of activities of caspases 3 and 7 at P13 and P20, respectively.
Next, we examined the activation of Parp1 (Parp1+; Figure 6C, Figure S7), which is cleaved during apoptosis into 89 and 24 kDa fragments by caspase 3 or 7 [63], [64] and has been linked to rod photoreceptor degeneration [65]. We found that Parp1+cells were never Cre+ and that their appearance was biphasic with activity peaks at P9 and P20 (Figure 6C, Figure S7A, a–b″″). These temporal peaks of Parp1+-activities in rod photoreceptors coincided with the activation of caspase 9 in rods at P9 (Figure S7A, c–d″″) and caspase 7 in cones at P20 (Figure 6C). No Parp1+-cell bodies could be identified at P13 (Figure 6C). At P9, caspase 9+TUNEL+ and Parp1+TUNEL+ cell bodies were always cre− rods, while TUNEL+caspase 9−cre− and TUNEL+Parp1−cre− rods could also be observed (Figure S7A). These events indicate that caspase 9 and Parp1 activation in rod photoreceptors is short-lived. These manifestations were accompanied by transcriptional up-regulation of caspase 9 and Parp1 between P9 and P20, but not of caspase 8, which is typically activated by extrinsic cell-death mechanisms (Figure S7B) [61], [62].
The formation of rod apoptotic cell bodies prompted us to examine whether apoptosis-inducing factor (AIF) or cytochrome c was released from the mitochondria, as these are also hallmark events of the apoptotic cascade [66]–[68]. Subcellular fractionation of retinas showed no sign of AIF or cytochrome c in the cytosol fraction of either genotype at the peak of apoptosis (Figure S8A). Changes in AIF levels were also not observed in the nuclear-enriched fraction of either genotype (Figure S8B). In addition, the expression levels of the apoptotic protease activating factor-1 (Apaf-1) remained unchanged (data not shown). Finally, we examined markers for necroptosis, such as members of the receptor-interacting proteins, RIP1 and RIP3, and macroautophagy, such as the autophagosomal membrane marker, light chain 3B II (LC3B II), since they are thought to be induced upon photoreceptor degeneration [69]–[71]. Immunoassays of retinal extracts found no differences in these markers between HRGP-cre:Ranbp+/− and HRGP-cre:Ranbp2−/− mice at P13 (Figure S8C, S8D).
Histological examination of semi-thin retinal sections showed that retinas of HRGP-cre:Ranbp2−/− developed prominent interstitial spaces between photoreceptors cell bodies and across the ONL by P20, a phenotype not present in HRGP-cre:Ranbp2+/− mice (Figure 7A). The interstitial spaces could always be traced to prominent euchromatic nuclei typically localized at the distal (outer) edge of the ONL (Figure 7A), hallmark morphological and topographic features of cell bodies of cone photoreceptors. Detailed examination of the ultrastructure of the ONL showed that the interstitial spaces reflect degenerating lower fibers of cone photoreceptors that were dilated and very lucent (Figure 7B). This striking phenotype led us to hypothesize that ablation of Ranbp2 in cones promotes the activation of metalloproteinase(s) (MMPs) causing the weakening and degradation of the extracellular matrix, which normally organizes photoreceptor cell bodies within the ONL [72], and may contribute to the retraction of Cre+-cell bodies from the distal (outer) to the proximal (inner) region of the ONL (Figure 2). Hence, we screened retinal extracts for each the eleven MMP activities. In comparison to HRGP-cre:Ranbp2+/−, retinal extracts of HRGP-cre:Ranbp2−/− mice presented ∼3-fold higher activity of MMP11, but not of any other MMPs (Figure S9A). MMP11 activity was elevated at P13 and P20 and returned to control levels at P27 when most cones have degenerated (Figure 7C). The increase of MMP11 activity was also accompanied by a ∼3-fold increase of the active form of MMP11 (Figure 7D, 7E). The transcriptional up-regulation of Mmp11 in HRGP-cre:Ranbp2−/− mice as early as P9 followed its transient down-regulation at P7 and preceded the activation of MMP11 at P13, whereas the transcriptional levels of Timp3 remained largely unchanged across different ages (Figure S9B, 7C). To assess the cellular origin of MMP11 expression and activity, we performed immunohistochemistry of MMP-11 in retinal sections from P20 mice, and found that MMP-11 was localized prominently around cell bodies, inner segments and lower fibers of cone Arr4+-cells (Figure 7F). Collectively, these data confirm that ablation of Ranbp2 in cones promotes the up-regulation of MMP11 expression and activity in these neurons, an event which likely contributes to the development of interstitial spaces in the ONL and retraction of cones cell bodies to the proximal ONL in HRGP-cre:Ranbp2−/− mice.
To ascertain in greater detail the morphological changes of degenerating cone photoreceptors, we examined the ultrastructure of HRGP-cre:Ranbp2+/− and HRGP-cre:Ranbp2−/− retinal sections at P20. The subcellular subcompartments of cone and rod photoreceptors of HRGP-cre:Ranbp2+/− mice had normal morphologies (Figure 8A, 8A′, 8F, 8F′, 8I), whereas cone photoreceptors of HRGP-cre:Ranbp2−/− mice exhibited distinct morphological changes across multiple subcellular compartments. Across different photoreceptor cells of HRGP-cre:Ranbp2−/−, we found that cone OS membranes were extended, disorganized and collapsed into large, amorphous and lucent areas (Figure 8B, 8B′, 8C, 8C′), that electrodense material accumulated at the connecting cilium (Figure 8D, 8D′), and that prominent electron lucent areas were present in the inner segments (Figure 8E, 8E′). Additional abnormalities were seen at the synaptic pedicles including accumulation of multilamellar bodies (Figure 8G, 8G′), mitochondria with widespread electron lucent matrix areas without cristae and formation of cytosolic lucent areas without a limiting membrane around subcellular debris (Figure 8H, 8H′). Albeit less extensive, cristae of some mitochondria in the rod synaptic spherules were also disrupted resulting in the formation of lucent areas within the matrix (Figure 8J).
We examined the effect of Ranbp2 ablation selectively in cones on the expression of cone photoreceptor-specific and other pertinent genes by quantitative real time-PCR (qRT-PCR) (Figure 9, Table S1). We found that Ranbp2 ablation led to rapid declines of M-opsin (Opn1mw) and S-opsin (Opn1sw) mRNAs as early as P13 (Figure 9A), when there are yet no prominent cellular changes in M-opsin and Cre+-neurons between genotypes (Figure 3B, 3E), and such declines occurred without concomitant changes in rhodopsin (Rho) mRNA (Figure 9A). Other cone-specific genes showed a similar pattern of down-regulation, including Pde6h, Pde6c and Gnat3 (Figure 9B). By P27, when cones have degenerated, the expression of all cone-specific genes was at or below the detection limit in HRGP-cre:Ranbp2−/− retinas (Figure 9A, 9B). By contrast, the expression of the pan-photoreceptor markers, Rcvn and Osgep, increased transiently at P13 (Figure 9C).
We also examined the genetic variants of miR-124a (miR-124a1/Rncr3, miR-124a2, miR-124a3), which mediate cone survival, and its downstream target transcript, Lhx2, whose translation is suppressed by miR-124a [73]. We found strong up-regulation of miR-124a in the HRGP-cre:Ranbp2−/− retina, which was accompanied by increased levels of Lhx2, albeit of a lesser magnitude (Figure 9D). The time course of these changes was similar, beginning at P13, peaking at P20 and returning to HRGP-cre:Ranbp2+/− levels at P27 when cones have degenerated (Figure 9D). We also assessed the expression of Trß2, Otx2 and Crx, encoding transcription factors critical to the maturation and maintenance of cone photoreceptors (Figure 9E) [74]–[78]. Crx expression was decreased at P7, when Cre-mediated excision of Ranbp2 occurs (Figure 1C), and then rose steadily up to P13, and then decreased to HRGP-cre:Ranbp2+/− levels at P27. In comparison, expression of Trß2 and Otx2 increased later, with a sharp peak at P20, and then declined to basal levels at P27 (Figure 9E). Like Crx, CoREST (also known as Rcor1), a cofactor of REST (repressor element 1 silencing transcription factor), was transiently down-regulated at P7 and then rose at P9 until P20, when there was a transient up-regulation of neuropilin-1 (nrp1), a receptor whose transcriptional expression is modulated by miR-124 and CoREST (Figure 9E) [79]. The expressions of CoREST and nrp1 also returned to basal levels at P27 (Figure 9E).
Because Ranbp2 ablation induced the expression and activation of MMP11, a metalloproteinase which is known to cleave selectively the α3 chain of collagen VI (Col6α3) [80], we examined the time course changes of Col6α3 expression in the HRGP-cre:Ranbp2−/− retina. A selective rise in Col6α3 expression, but not α1 chain of collagen I (Col1α1), was detected as early as P9, shortly after a transcriptional increase in Mmp11 was also detected (Figure 9F, S9B). Cola6α3 expression peaked at P20 and then became indistinguishable from HRGP-cre:Ranbp2+/− mice at P27. We did not observe any transcriptional changes in let-7c, a miRNA reported to promote the down-regulation of Mmp11 (Figure 9F) [81]. Finally, we examined transcriptional changes in markers associated with neurodegenerative mechanisms including with autophagy (e.g. cathepsin S, lysozyme and culsterin), glycolysis (6-Pfk), hypoxia (Hif1α) and inflammation (Gfap). Among these, we found changes only in Hif1α and Gfap with Hif1α rising from P9 until P20 and Gfap transiently spiking at P20 when most cones are undergoing degeneration (Figure 9G).
Ranbp2 has multifaceted roles in biology and pathology across tissues, cell types and cell-stages. This complexity reflects the interaction of the diverse structural modules of Ranbp2 with multifunctional partners. Hence, we employed the Ingenuity pathway analysis (IPA) to identify and delineate connectivity maps linking Ranbp2 with the regulation of expression of genes/proteins identified by this study and genetic, protein and metabolic networks. The top network hit generated by IPA (score 22) was associated to “Cellular Development, Nervous System Development and Function and Carbohydrate Metabolism” (Figure 10). This network comprised thirty-two gene products and three endogenous chemicals, D-glucose, sn-glycero-3-phosphocholine and tretinoin (all-trans retinoic acid) (Figure 10). Remarkably, this connectivity map revealed links between transcription factors, many of which belong to the orphan nuclear receptor family and are known to be modulated by Ranbp2 levels, and metabolites, whose levels are also affected by Ranbp2 and regulate transcriptional activities of nuclear factors [47], [49], [51], [82], [83]. Among other novel points of interest, the IPA revealed central roles of i) huntingtin cross-talk with nuclear factors modulated by Ranbp2, a mechanism which is thought to be disrupted in Huntington's disease (HD) [84], and ii) two secreted and extracellular signaling proteins, wingless-type MMTV integration site family, member 2 (WNT2B) and brain-derived neurotrophic factor (BDNF), intersecting multiple nodes modulated by Ranbp2 and other factors. WNT2B and BDNF are known to play important roles in regulation of cell growth, differentiation, tumorigenesis, and to support the survival of existing neurons, respectively [85]–[87].
Genetic excision of Ranbp2 is detectable by P7, which coincides with immediate transcriptional changes in the levels of Mmp11, Crx and CoREST (Figure 9E, S9B). Reduced levels of cone-specific transcripts, such as those encoding phototransduction proteins, were not detectable until P13 and changes in cone morphology were not observed until P20. These observations, and the non-autonomous molecular and cellular effects of cones on rod photoreceptors, prompted us to determine the onset and progression of cone and rod physiological dysfunction caused by ablation of Ranbp2 in cones. We measured cone and rod function by light- and dark-adapted electroretinograms (ERGs), respectively. Figure 11 summarizes the ERG data obtained from mice between P13 and 150 days. At P13, dark-adapted ERGs of HRGP-cre:Ranbp+/− and HRGP-cre:Ranbp2−/− mice were comparable (Figure 11A, left; Figure 11B), indicating equivalent retention of rod-mediated outer retinal activity at this age. In comparison, light-adapted ERGs, reflecting cone activity [88], [89] were already reduced significantly at P13 (Figure 11A, right; Figure 11C). By P22, the amplitude of the light-adapted ERGs were markedly reduced in HRGP-cre:Ranbp2−/− mice (Figures 11D), and cone ERGs were extinguished in HRGP-cre:Ranbp2−/− mice at P29 (Figure 11A, 11E). The amplitude of the dark-adapted ERG a-wave, which reflects phototransduction in the outer segments of rod photoreceptors, was not significantly different between HRGP-cre:Ranbp2+/− and HRGP-cre:Ranbp2−/− mice at any age examined (Figure 11A, 11B, 11F, 11G). In mice aged P29 and P150, the dark-adapted ERG b-wave was reduced, but in a luminance-dependent fashion (Figure 11F, 11G). At low stimulus luminances, where the b-wave reflects synaptic transmission from rod photoreceptors to rod bipolar cells, there was no significant reduction in amplitude. At higher stimulus luminances, where both rod- and cone-mediated synaptic activity contribute to the b-wave, a significant amplitude reduction was observed (Figure 11F). The synaptic localization of post-synaptic density 95 (PSD95) protein was comparable between genotypes at P20 (Figure S10), indicating that the b-wave reductions noted at high flash luminances reflects a loss of the cone pathway contribution to the ERG b-wave at these stimulus levels, and not to an alteration in synaptic transmission between rod photoreceptors and bipolar cells. Figure 11G plots the amplitude of HRGP-cre:Ranbp2−/− light- and dark-adapted ERG components relative to those of control littermates (HRGP-cre:Ranbp2+/−) ranging from P13 to 21-weeks in age. HRGP-cre:Ranbp2−/− mice have a selective reduction in the light-adapted ERG indicating that the dark-adapted ERG is not sensitive to the limited cone-induced rod loss seen by other cell biological measures previously described in this study. In agreement with the electrophysiological observations, the complete loss of cone photoreceptors and the limited cone-induced loss of rod photoreceptors did not cause significant changes in ONL thickness and cell body density when analyzed by light microscopy in mice as old as 12-weeks of age (Figure S11).
The results of this study demonstrate: i) cone-specific ablation of Ranbp2 triggers the demise of cone photoreceptors with concomitant non-autonomous cell death of rod photoreceptors, ii) the death of rod photoreceptors is contingent on the presence of cone photoreceptors, and iii) the cell death mechanisms of cone and rod photoreceptors are intrinsically distinct with the former and latter undergoing atypical features of necrosis and apoptosis, respectively [61], [62]. We found that cone photoreceptors undergo necrotic changes including massive erosive destruction of their intracellular contents and late caspase 7 activation, but without apparent changes in loss of membrane permeability. In comparison, rods undergo apoptosis, caspase 3 and Parp1 activations, and loss of membrane permeability.
This study unveils early molecular events triggered by and concomitant with the ablation of Ranbp2 in cone photoreceptors, including changes in Mmp11, Crx and CoREST expressions at P7 and of MMP11 activity at P13, followed by the activation of caspase 7 and a compensatory burst in the expression of the MMP11 substrate, Cola6α3, at P20, when the degeneration of cones is well underway. These events are accompanied by the up-regulation of cone survival genes, such as genetic variants of miR-124a and its downstream target transcript, Lhx2, at P13, when the down-regulation of cone-specific genes, such as M- and S-opsins, becomes significant. miR-124 promotes the translational suppression of Lhx2, a process which has been linked to the suppression of apoptosis during cone photoreceptor differentiation and aberrant sprouting of hippocampal neurons [73]. In contrast to these observations, up-regulation of miR-124a does not prevent the demise of mature cone photoreceptors and occurs concomitantly with a rise of Lhx2 levels. Further, changes in CoREST levels, a target of miR-124 [79], preceded those of all miR-124 variants. These observations suggest that the miR-124 variants may act on other substrates, such as Foxa2 [90], and that may modulate the survival of cones (or rods) independently of Lhx2 and CoREST suppression [79]. Finally, even though cones account for only ∼3% of all photoreceptor types of the mouse retina [59], the return of cone-derived caspase 7 and MMP11 activities and rod-derived caspase 3 activity to control levels after cones have degenerated strongly support these events are triggered solely by the Ranbp2-dependent dysfunction of cones.
The early and selective activation of MMP11 followed by the development of prominent interstitial spaces between photoreceptor cell bodies, swelling of the lower fibers of cones, and the loss of membrane permeability of rods supports that MMP11 plays an important role in the development of autonomous and non-autonomous photoreceptor degeneration. These roles may include autocrine, paracrine and even intracrine actions, whereby intracrine and paracrine functions contribute to intracellular erosion of cones and non-autonomous death of rods, respectively. In comparison to other MMP proenzymes, MMP11 shows important functional differences. While most MMPs are activated extracellularly, MMP11 is secreted in the active form [91]. This exposes intracellular substrates to active MMP11, a potential intracrine signal, and its extracellular proteolytic functions may stimulate autocrine and/or paracrine signaling. Regardless of its molecular mechanisms of action(s), increased levels of MMP11 expression is reported to modulate cell survival [92]–[96], to act as a negative prognostic of cancer patient survival [97]–[99], and to promote oncogenic homing, tumorigenesis and metastasis [92], [96], cardinal manifestations triggered also by haploinsufficiency and hypomorphism of Ranbp2 [48].
MMP11 up-regulation has been reported to promote the suppression of apoptotic and necrotic cell death, rather than stimulation of cell proliferation [95], [96]. These cellular manifestations appear at odds with the physiological phenotypes of our study. It is possible that the distinct pathological outcomes produced by the induction of MMP11 upon loss of Ranbp2 reflect intrinsically distinct tissue/cell-type-dependent signaling cues produced by MMP11 substrates. Hence, etiological distinct disorders, such as cancer and neurodegeneration, may share pathomechanisms with distinct clinical outcomes. The identification of pathophysiological substrates of MMP11 will be critical to uncover the scope of its pathobiological roles and aid toward the design of tissue-selective MMP11 inhibitors. It will be interesting to explore whether paracrine factors and players within the network uncovered by the IPA presented in this work, such as Wnt2b and BDNF, their receptors or nuclear transcriptional factors, are substrates of MMP11 or exert regulatory effects on MMP11 expression or activity. Although a specific pharmacological inhibitor of MMP11 is not available [100], [101], the genetic interactions between Mmp11 and Ranbp2 may be revealed by assessing the effect of genetic ablation of Mmp11 on the pathophysiological manifestations observed in HRGP-cre:Ranbp2−/− (e.g. loss of membrane permeability of rods and rod apoptosis) and to define a potential role for MMP11 in neuroprotection against autonomous and non-autonomous cell death mechanisms affecting cone or rod photoreceptor survival.
Another critical outcome of this work is the finding of non-autonomous apoptotic death of rod (Nr2E3+) photoreceptors upon cone dysfunction and death. TUNEL+ cones were never identified at any age, while dying rods comprised a mixed population of TUNEL+, EthD-III+ or both. Thus, rods undergo a programmed cell death that is triggered by the demise of cones. Important, rod cell death ceases at P27, when cone degeneration is complete based on the lack of apoptotic bodies, the absence of Parp1+ and caspase3/7+ cells, and the preservation of the ONL and rod photoreceptor function at P27 and later ages.
Our results also indicate the presence of atypical mechanisms of cell death. While classical cell death features were identified during cone or rod degeneration, the following observations do not match previously recognized canonical paradigms. First, caspase 7 activation, typically observed in apoptosis, was found in dying cones lacking cleaved Parp1, a substrate of caspase 7. Instead, cone death was characterized by the rampant disintegration of critical subcellular structures, such as outer segments and mitochondria, swelling of lower fibers and cone pedicles, features that are consistent with necrosis. However, EthD-III+ cones were not identified as would be expected from the typical loss of membrane permeability caused by necrosis. Second, although rod photoreceptor death is consistent with apoptosis as supported by the presence of caspase 3+, Parp1+ and TUNEL+-cell bodies, classic necrotic features were also present, including a loss of plasma membrane permeability (EthD-III+ cell bodies) and limited formation of electron lucent areas in the matrix of mitochondria at synaptic spherules. These observations indicate that activation of caspase 3 and caspase 7 does not determine the mode of rod or cone cell death, respectively. This conclusion is supported by comparable rates of rod death in rd1 mice on a wild-type or caspase3−/− background [43]. Third, we did not observe a release of cytochrome C and AIF to the cytoplasm, as typified in apoptosis [66] and observed in other photoreceptor degeneration models [35], [45], an induction of RIP1, RIP3, or caspase 8 activation as observed in necroptosis [61], [69] and lipidation of microtubule-associated protein 1 light chain 3 (LC3/Atg8) that is typical of autophagy [70], [71]. Altogether these data support the existence of complex, shared, unique and thus atypical cell death mechanisms between rod and cone photoreceptors. These manifestations are likely determined by the cell-type dependent pleiotropic molecular and metabolic activities of Ranbp2 [47]–[54]. Emerging mouse models of Ranbp2 harboring losses in selective domains and functional activities of Ranbp2 will aid in parsing the contribution of such activities to cellular functions and intrinsic or extrinsic cell death modalities unique or shared by several diseases.
Finally, our data show that the non-autonomous death of rods upon loss of Ranbp2 in cones is not promoted by the absence of cones or loss of rod-cone cell contacts. Instead, a likely mechanism is the release of a diffusible factor by cone photoreceptors upon their dysfunction that is deleterious to rod photoreceptors. MMP11 activation is an excellent candidate to play a critical role in such paracrine and death signaling. Our data predict that the topographic density of cone and rod photoreceptors in the retina will play a determinant role in the level of rod photoreceptor degeneration. This issue is of crucial significance to human retinal dystrophies, because the unique and heterogeneous topographic distributions of photoreceptor types across the human retina often mirror retinal pathologies with regional tissue and clinical hallmarks, such as RP, cone-rod dystrophies and age-related macular degeneration (AMD). It is likely that loss of biological activities regulated by RANBP2 will exert prominent pathological outcomes in regions of the retina, such as the macula, where the topographic density of cone and rod photoreceptors is similar.
HRGP-Cre mice (kindly provided by Yun-Zheng Le, University of Oklahoma Health Sciences Center) [56], [57] were crossed to Ranbp2+/flox mice (kindly provided by Jan M. van Deursen, Mayo Clinic College of Medicine) [48] to produce HRGP-cre:Ranbp2+/flox. HRGP-cre:Ranbp2+/flox were then crossed to Ranbp2+/flox or Ranbp2Gt(pGT0pfs)630Wcs/+ [47] to generate HRGP-cre:Ranbp2+/− (+/−) and HRGP-cre:Ranbp2−/− (−/−). Mice were in the following mixed genetic background: 129 SvJ,C57BL/6J,FVB/N,129olaHsd. Mice were screened also for rd1 and rd8 alleles. Mice were raised in a pathogen-free transgenic barrier facility at <70 lux and given ad libitum access to water and chow diet 5LJ5 (Purina, Saint Louis, MO). Animal protocols were approved by the Institutional Animal Care and Use Committees at Duke University and Cleveland Clinic, and all procedures adhered to the ARVO guidelines for the Use of Animals in Vision Research.
Primary antibodies used this study were: mouse anti-Cre (Covance, Princeton, NJ), rabbit anti-Cre (Novagen, Gibbstown, NJ), rabbit anti-Arr4 (Millipore/Pel-Freez Biologicals , Billerica, MA), rabbit anti-L/M opsin (#Ab 21069) [55], rabbit monoclonal anti-cleaved caspase3 Asp175 (Cell Signaling, Danvers, MA), rabbit polyclonal anti-cleaved-caspase 7 Asp353 (Cell Signaling), rabbit anti-cleaved caspase 8 Asp391 (Cell Signaling), rabbit anti-cleaved caspase 9 Asp 330 (Cell Signaling), rabbit anti-cleaved Parp1 Asp214 (Cell Signaling), mouse anti-full-length Parp1 (BD Bioscience, San Jose, CA), mouse anti-LC3 B (Cell Signaling, Danvers, MA), mouse anti-cytochrome C (BD Bioscience, San Jose, CA), mouse anti-mitochondrial heat shock protein 70 (Affinity Bioreagent, Golden, CO), rabbit-cytosolic heat shock protein 70 (Assay Design, Farmingdale, NY), rabbit anti-GFAP (DAKO, Carpinteria, CA), mouse anti-glutamine synthase (Sigma Aldrich, Saint Louis, MO), mouse anti-RIP1 (BD Bioscience, San Jose, CA), rabbit anti-RIP3 (Sigma Aldrich, Saint Louis), rabbit anti-GAPDH and goat anti-AIF1 (Santa Cruz Biotechnology, Santa Cruz, CA), rabbit anti-Apaf-1 (LSBio, Seattle, WA), mouse anti PSD-95 (Affinity BioReagent), mouse anti-MMP11(Thermo Scientific, Waltham, MA), rabbit anti-S-opsin (Millipore/Pel-Freez Biologicals), rabbit anti-Nr2E3 (Proteintech, Hayward, CA) and Peanut Agglutinin TRTIC conjugate (Sigma-Aldrich). Alexa-conjugated secondary antibodies (408, 488, 568 and Cy5) were from Invitrogen (Carlsbad, CA).
For total RNA and pre-miRNA isolation, retinas were homogenized with the TRIZOL Regent (Invitrogen) using Bullet Blender BBX24 (Next Advance Inc., Averill Park, NY) in the presence of 0.5 mm zirconium oxide beads (Next Advance Inc., Averill Park, NY) for 3 min at 8,000 rpm. RNA was reverse transcribed into cDNA using SuperScript II reverse transcriptase (Invitrogen). For RT-PCR, the ∼380 bp amplicon encompassing fused exons 1 and 3 of recombinant Ranbp2 mRNA was amplified from 250 ng of retinal cDNA. PCR was performed using primer 1 (Pr1:CGCCCCGAGAGTACATTTCTA) and primer 2 (Pr2:AAGTTTATTCCATCCATCTTCA) with GoTaq Green Master Mix (Promega, Madison, WI) under the following cycling conditions: 5 min/95°C of initial denaturation, 35 cycles/94°C (30 s), 55°C(30 s) and 72°C(30 s) with final elongation step at 72°C for 3 mins. Same cycling conditions were applied for Cre (CTAATCGCCATCTTCCAGCAGG, AGGTGTAGAGAAGGCACTTAGC) and Gapdh (GCAGTGGCAAAGTGGAGATT, GAATTTGCCGTGAGTGGAGT). qRT-PCR reactions were carried out with 8 ng of cDNA, 800 nM forward and reverse primers, 10 µl of 2×SYBR® Green PCR Master Mix (Applied Bioscience, Warrington, MA) in a 20 µl final volume in 48-well plates using the ECO™ Real-Time PCR system (Illumina, San Diego, CA). The relative amount of transcripts was calculated by the ΔΔ CT method using Gapdh as reference (n = 3–4). Primer sequences and designations are provided in Table S1.
The superior region of mouse cornea was burned using Low Temperature Cautery (Bovie Medical Corporation, St. Peterburg, FL) immediately after mice were killed. For immunohistochemistry, eyeballs were removed and fixed with 2% paraformaldehyde/phosphate-buffer saline (PBS), pH 7.4 for 4 hr after small incisions were made in the anterior portion. Upon removal of the lens, eyecups were infiltrated with 5% sucrose/100 mM PBS, pH 7.4, for 5 hr followed by 30% sucrose/100 mM PBS for 12 hours, embedded in Tissue-Tek O.C.T. compound (Sakura, Torrance) and stored at −80°C. 12 µm thick retinal cryosections along the vertical meridian of the eyecup were mounted on glass slides. For flat mounts, retinas were removed from fixed eyeballs, cut in a four-quadrant cloverleaf pattern using the caruncle as an orientation landmark and fixed for an additional 15 min in a 24-well plate. Specimens were incubated in blocking buffer (PBS, pH 7.4, containing 0.1% Triton X-100, 10% normal goat serum) for 1 hr at RT (for sections) or 12 hr at 4°C (flat mounts), followed by incubation with primary antibodies in incubation buffer (PBS, pH 7.4, containing 0.1% Triton X-100, 5% normal goat serum) overnight at RT (for sections) or for 3 days at 4°C (flat mounts). Specimens were washed thrice with washing buffer (PBS/0.1% Triton X-100) for 10 min, and incubated in incubation buffer for 2 hr with Alexa-conjugated secondary antibodies (1∶1,000; Invitrogen). For Nr2E3 immunostaining, eyeballs were fixed instead with 1% paraformaldehyde for 1 hr at room temperature. 6 µm-thick retinal cryosections were incubated in blocking buffer for 1 hr at room temperature followed by treatment with proteinase K (20 µg/ml, Promega, Madison, WI) for 9 min and standard immunostaining protocols as described earlier. Specimens were washed again thrice and mounted on glass slides for visualization and image acquisition. Images were acquired with a Nikon C1+-laser scanning confocal microscope coupled with a LU4A4 launching base of 4-solid state diode lasers (407 nm/100 mW, 488 nm/50 mW, 561 nm/50 mW, 640 nm/40 mW) and controlled by the Nikon EZC1.3.10 software (v6.4).
TUNEL assays were performed with the DeadEnd Fluorometric TUNEL System (Promega, Medison) with the following modifications from the manufacturer's instructions. Briefly, specimens were incubated with 20 µg/ml proteinase K for 15 min at RT followed by fixation with 4% paraformaldehyde for 15 min, and incubations with primary and secondary antibodies along with DAPI (Invitrogen, Carlsbad, CA). Specimens were equilibrated for 5 min in manufacturer's buffer before undergoing TdT reactions for 60 min at 37°C. Reactions were stopped with 2×SSC, washed three times with PBS and mounted.
Fresh P20 retinal explants were incubated with 5 µM EthD-III (Biotium, Hayward, CA) in the Neurobasal−A Medium/B-27 Supplement (Invitrogen) for 4 hr at 37°C in a humidified 5% CO2 atmosphere, washed vigorously 3 times with 100 mM PBS for 10 min and fixed with 2% paraformaldehyde/100 mM PBS for 20 min at room temperature. Specimens were then processed as described for immunostaining and TUNEL procedures.
Morphometric analyses of M-cone photoreceptors were performed from 127×127 µm image fields captured with a Nikon C1+-laser scanning confocal microscope. Optical slices were 3D-reconstructed for the whole length of outer segments (∼25 µm, step size of 0.5 µm) from retinal flat mounts immunostained with an L/M-opsin antibody [55]. M-cone photoreceptors and the length of their outer segments were then tallied and measured from three image fields for each retina with the post-acquisition Nikon Elements AR (ver3.2) software. Cre+, c-casp3+ and TUNEL+-cells bodies were imaged from flat mounts and three image fields of 127×127 µm and collapsed for the whole depth of ONL (∼60 µm, step size of 3 µm) from the central or peripheral retinal regions. For TUNEL+, EthD-III+ and Arr4+-cells bodies, three image fields of 127×127×5 µm from retinal flat mounts were randomly selected and quantitatively analyzed. For tallying of DAPI+, TUNEL+,−Nr2E3+or Cre+-cell bodies, whole vertical meridian sections were counted and averaged to generate pie graphs (Excel, Microsoft, Seattle, WA). The distal region of the ONL was arbitrarily defined as the 15 µm distal segment of the ONL up to the external limiting membrane and the remaining segment was defined as proximal. 3D-reconstruction of collapsed images and morphometric analysis was performed with Nikon Elements AR software (ver. 3.2). Two-tailed equal or unequal variance t-test statistical analysis was performed. p≤0.05 was defined as significant.
Retinas were homogenized at 4°C with Bullet Blender BBX24 (Next Advance Inc., Averill Park, NY) in the presence of 0.5 mm zirconium oxide beads (Next Advance Inc.) and RIPA buffer containing complete protease inhibitors (Roche Applied Bioscience, Penzberg, Germany) and 10 mM iodoacetamide (Sigma Aldrich). Protein concentration was measured by the BCA method using BSA as a standard. Samples (55 µg) were resolved on 11% SDS-polyacrylamide gel electrophoresis (SDS-PAGE), immunoblotted and developed using the SuperSignal Pico West (Thermo Scientific) as described previously [50]. Blots were probed with mouse anti-MMP-11 antibody (Thermo Scientific). Densitometry analysis of immunoblots was performed with Metamorph v7.0 (Molecular Devices, Sunnyvale, CA). Two-tail t-test statistical analysis was performed for validation of significant change (p<0.05).
Retinas were homogenized using BulletBlender BBX24 (Next Advance Inc.) with 0.5 mm zirconium oxide beads (Next Advance Inc.) and RIPA buffer. Retinal homogenates were centrifuged at 10,000 g for 15 min at 4°C. Supernatants were collected and protein concentrations determined by the BCA method using BSA as standard. Retinal homogenates were diluted in caspase assay buffer and caspase profiling assays were performed with the Sensolyte AFC Caspase sampler kit according to company's protocol (AnaSpec, Fremont, CA). The following caspase substrates were used for screening: Ac-YVAD-AFC (SB1) and Ac-WEHD-FAC (SB2) for caspase 1, Ac-VDVAD-FAC (SB3) for caspase 2, Ac-IETD-FAC, (SB4) for caspase 8, Ac-DEVD-FAC (SB5) and Z-DEVD-AFC (SB6) for caspases 3/7, Ac-LEHD-FAC (SB7) for caspase 9 and Ac-VEID-AFC (SB8) for caspase 6. Assays were first optimized for substrate dilution, extract concentration and reaction time. Substrate screenings were carried with three different protein concentrations. Analytical assays of caspase 3/7 (SB5) were performed with diluted 50 µl of retinal homogenate (280 ng) and 50 µl of substrate (1∶200 dilution in caspase assay buffer), mixed in the 96 well-plate, incubated with shaking for 1 hr under dark at room temperature. Measurements of fluorescence were performed at excitation/emission/cutoff = 380/500/495 nm with SpectraMax M5 (Molecular Devices, Sunnyvale, CA). Control reactions without extracts were subtracted from the samples' readings.
MMPs' screening assays were performed with Sensolyte 520 generic MMP assay kit per company's protocol (AnaSpec, Fremont, CA). Retinas extracts were prepared as described previously with the exception that RIPA buffer was replaced with MMP assay buffer (AnaSpec). Briefly, MMP assays are based on the dequenching of fluorescence intensity of 5-FAM upon proteolytic cleavage of 5-FRAM/QXL520 peptide substrate by MMPs. MMPs screenings, except MMP11, were performed upon activation with 1 mM 4-aminophenylmercuric acetate (APMA) at different time intervals (MMP2, 7, 8 and 13: 1 hr activation; MMP9, 12 and 14: 2 hr activation; MMP1: 3 hr activation; MMP3 and 10: 24 hr activation) at 37°C followed by 1 hr incubation with 5-FRAM/QXL520 peptide substrate. For MMP11, retinal extract was directly mixed with substrate and incubated for 1 hr before fluorescence reading. 50 µl of activated/non-activated protein extract at 56, 112 or 224 ng and 50 µl of substrate (final substrate dilution 1∶200) were mixed in a 96 well-plate, incubated with shaking for 1 h at room temperature at dark. Measurements of fluorescence were performed at excitation/emission/cutoff = 490/520/495 nm with SpectraMax M5 (Molecular Devices, Sunnyvale, CA). Control measurements without retinal extracts under the same conditions were subtracted from the sample readings. Analytical assays with MMP11 were performed with 56 ng of retinal extract.
Mitochondrial and cytosolic fractions of the retina were isolated with the Mitochondria Isolation Kit for Tissue (Abcam, Cambridge, MA) per manufacturer's instruction. Briefly, retinas were washed with washing buffer, homogenized with a Kontes Microtube Pellet Pestle Rod with motor in isolation buffer, centrifuged at 1,000× g for 10 mins, the pellet was saved (nuclear-enriched fraction) and supernatant was centrifuged again at 12,000× g for 15 min. The supernatants (cytosolic fraction) were saved, the pellets (mitochondrial fraction) washed with Isolation buffer twice and re-suspended with isolation buffer with complete protein inhibitor cocktail (Roche Applied Bioscience, Penzberg, Germany).
Eyeballs were removed and fixed with 2% glutaraldehyde:paraformaldehyde/0.1% cacodylate buffer, pH 7.2, overnight at 4°C. For semi-thin histological sections, 0.5 µm sections along the vertical meridian were mounted on glass slides and stained with 1% methylene blue. Light images of the retina sections were acquired with a Axiopan-2 light microscope controlled by Axovision Rel 4.6 and coupled to a AxioCam HRc digital camera (Carl Zeiss, Germany). For electron microscopy, specimens were post-fixed in 2% osmium tetraoxide in 0.1% cacodylate buffer and embedded in Spurr resin. 60 nm-thick sections were cut with Leica Ultracut S (Leica Microsystems, Waltzer, Germany), stained with 2% uranyl acetate/4% lead citrate and imaged with JEM-1400 transmission electron microscope (JEOL, Tokyo, Japan) coupled with an ORIUS 1000CCD camera.
The Ingenuity pathway analysis (IPA, Ingenuity Systems, Redwood City, CA) was used to examine the gene dataset modulated by loss of Ranbp2 and to define connectivity maps and networks. Network of genes are algorithmically produced based on their connectivity. Significant network scores reflect the negative logarithm of a P value associated with the likelihood of connectivity of a set of genes in a network. The network with highest score (score of 22) was chosen for further analysis.
After overnight dark adaptation, mice were anesthetized (ketamine: 80 mg/kg; xylazine: 16 mg/kg) and eyedrops were used for pupil dilation (1% tropicamide; 2.5% phenylephrine HCl; 1% cyclopentolate HCl) and corneal anesthesia (1% proparacaine HCl). The active electrode was a stainless-steel wire active electrode that contacted the corneal surface through 1% methylcellulose; needle electrodes placed in the cheek and tail served as reference and ground leads, respectively. Responses were differentially amplified (0.3–1,500 Hz), averaged, and stored using a UTAS E-3000 signal averaging system (LKC Technologies, Gaithersburg, MD). Strobe flash stimuli were initially presented in darkness within a ganzfeld bowl. Flash luminance ranged from −3.6 to 2.1 log cd s/m2 and stimuli were presented in order of increasing luminance. A steady adapting field (20 cd/m2) was then presented in the ganzfeld. After a 7 min light adaptation period cone ERGs were evoked by strobe flash stimuli superimposed upon the adapting field. Flash luminance ranged from −0.8 to 1.9 log cd s/m2 and stimuli were presented at 2 Hz in order of increasing luminance. The amplitude of the a-wave was measured 8 ms after flash onset from the prestimulus baseline. The amplitude of the b-wave was measured from the a-wave trough to the peak of the b-wave or, if no a-wave was present, from the pre-stimulus baseline.
|
10.1371/journal.pbio.2000420 | Bordetella bronchiseptica exploits the complex life cycle of Dictyostelium discoideum as an amplifying transmission vector | Multiple lines of evidence suggest that Bordetella species have a significant life stage outside of the mammalian respiratory tract that has yet to be defined. The Bordetella virulence gene (BvgAS) two-component system, a paradigm for a global virulence regulon, controls the expression of many “virulence factors” expressed in the Bvg positive (Bvg+) phase that are necessary for successful respiratory tract infection. A similarly large set of highly conserved genes are expressed under Bvg negative (Bvg-) phase growth conditions; however, these appear to be primarily expressed outside of the host and are thus hypothesized to be important in an undefined extrahost reservoir. Here, we show that Bvg- phase genes are involved in the ability of Bordetella bronchiseptica to grow and disseminate via the complex life cycle of the amoeba Dictyostelium discoideum. Unlike bacteria that serve as an amoeba food source, B. bronchiseptica evades amoeba predation, survives within the amoeba for extended periods of time, incorporates itself into the amoeba sori, and disseminates along with the amoeba. Remarkably, B. bronchiseptica continues to be transferred with the amoeba for months, through multiple life cycles of amoebae grown on the lawns of other bacteria, thus demonstrating a stable relationship that allows B. bronchiseptica to expand and disperse geographically via the D. discoideum life cycle. Furthermore, B. bronchiseptica within the sori can efficiently infect mice, indicating that amoebae may represent an environmental vector within which pathogenic bordetellae expand and disseminate to encounter new mammalian hosts. These data identify amoebae as potential environmental reservoirs as well as amplifying and disseminating vectors for B. bronchiseptica and reveal an important role for the Bvg- phase in these interactions.
| Bordetella species are infectious bacterial respiratory pathogens of a range of animals, including humans. Bordetellae grow in two phenotypically distinct “phases,” each specifically expressing a large set of genes. The Bvg+ phase is primarily associated with respiratory tract infection (RTI) and has been well studied. The similarly large set of genes specifically expressed in the Bvg- phase is poorly understood but has been proposed to be involved in some undefined environmental niche. Recently, we reported the presence of Bordetella species in many soil and water sources, indicating extensive exposure to predators. Herein, we show that the Bvg- phase mediates B. bronchiseptica interactions with the common soil predator D. discoideum. Surprisingly, the bacterium not only can evade predation but can propagate and disseminate via the complex developmental process of D. discoideum. After multiple passages and over a million-fold expansion in association with D. discoideum, B. bronchiseptica retained the ability to efficiently colonize mice. The conservation of the genes involved in these two distinct phases raises the possibility of potential environmental sources for the frequently unexplained outbreaks of diseases caused by this and other Bordetella species.
| Bordetella species are gram-negative bacteria that infect the respiratory tracts of mammals. The highly genetically conserved classical Bordetella species comprise B. pertussis and B. parapertussis, the etiological agents of whooping cough in humans [1], as well as B. bronchiseptica, which infects a variety of mammals and immunocompromised humans [1–3]. The major virulence genes in the classical bordetellae are regulated under the Bordetella virulence gene (BvgAS) two-component system, which senses environmental cues and controls transcription of over 100 virulence-associated factors [4,5]. The “Bvg positive (Bvg+) phase” refers to the activated state of the BvgAS system [5,6] in which the expression of genes that have been shown to be necessary for mammalian respiratory tract infection and survival are induced [6–9]. In contrast, at lower temperatures, in the “Bvg negative (Bvg-) phase,” the expression of virulence factors is repressed, and a similarly large set of genes, including those that enable flagella-mediated motility and growth in dilute nutrients, are specifically expressed [6,8,10]. Mutants that are locked in the Bvg- phase are rapidly cleared from inoculated animals, revealing the critical role of Bvg+ “virulence factors” during infection [11]. In contrast, bacteria locked in the Bvg+ phase efficiently infect hosts, indicating that the Bvg- phase is not required for successful interactions with the host. Explanations for the conservation of the large set of Bvg- genes include speculated roles for the Bvg- phase in survival in some unknown extrahost environment, so far supported by anecdotal evidence [12–15]. We have recently described a search of the National Center for Biotechnology Information (NCBI) nucleotide database that revealed evidence of Bordetella species in a large number of soil and water samples [15]. Phylogenetic analyses suggested that Bordetella species from these environments are the ancestral source from which modern respiratory pathogens emerged. To be successful in these environments, bordetellae are expected to be well adapted to interact with other bacteria and environmental predators. Thus, we hypothesize that Bordetella species have evolved mechanisms to successfully interact with predators and that these are associated with the Bvg- phase.
Amoebae are common environmental protists that feed on bacteria and have been isolated from soil, air, water, and nasal mucosa of both healthy and sick human volunteers [16–18]. When food (e.g., bacteria) is plentiful, amoebae survive and proliferate as single-celled amoebae. However, once the food source has been depleted from an area, some species of amoebae cooperate to spread to new, more fertile hunting grounds. In the case of D. discoideum, this cooperation involves a cyclic adenosine monophosphate (cAMP) signal that triggers aggregation of amoebae to ultimately form a multicellular fruiting body comprising a stalk and a sorus containing amoeba spores [19]. Sori can be disseminated in various ways, such as by wind shifting leaf litter or by the shuffling of passing animals, allowing the spores a chance to be deposited onto new food sources where they can germinate and once again feed as single-celled organisms. While many species of bacteria serve as a food source for amoebae, some bacteria, including several human pathogens such as Legionellae pneumophila and Francisella tularensis [20], have evolved means of surviving amoeba predation by persisting in single amoeba cells and blocking their host’s ability to differentiate into mature fruiting bodies [21–23]. Since amoebae and immune cells share similarities in their mechanisms used to phagocytize and kill bacteria [24], the ability of these pathogens to survive intracellularly during in vivo infection may be linked to an evolved mechanism for avoiding amoeba predation. Moreover, the relatively frequent isolation of amoebae from healthy human nasal mucosa [25] indicates that persistent nasal colonizers such as Bordetella spp. frequently encounter amoebae in vivo, raising the possibility that complex interactions between these organisms may have evolved over time.
We have previously shown that B. bronchiseptica can occupy an intracellular niche within macrophages during infection [3], an ability shared with other organisms that survive amoeba predation. Here, we show that B. bronchiseptica not only survives amoebic predation but also successfully infects and persists within amoeba cells. Unlike other bacteria that block fruiting body development [21–23], we show that B. bronchiseptica permits the complete D. discoideum life cycle and even localizes to the sori of the amoeba fruiting bodies for further propagation. Importantly, B. bronchiseptica is sequentially carried along with amoebic spores to new locations through many passages on other bacterial “food,” providing sustainable expansion/dissemination during a viable life cycle outside of a mammalian host. We show that the Bvg- phase is advantageous for B. bronchiseptica survival in the amoeba sori and therefore identify a role for the Bvg- phase in this potential ex vivo life cycle. When associated with the amoeba sori, B. bronchiseptica can be transferred by flies or ants and can efficiently infect mice, suggesting that amoebae can act as amplifying and transmission vectors for B. bronchiseptica in addition to being environmental reservoirs. Together, these data suggest a role for the Bvg- phase in a life cycle that does not require a mammalian host, which may explain the complexity and high conservation of genes specifically expressed in the Bvg- phase.
We and others have previously documented the ability of B. bronchiseptica to survive within phagocytic mammalian cells in vitro [3,26–31] and in vivo [2,3,32,33] during infection. Amoebae, which are similar to macrophages both morphologically and structurally [24], serve as environmental hosts for some intracellular human pathogens [34–37]. As the likely ancestors of pathogenic Bordetella species are environmental species found in soil and water where amoebae are prevalent [15], we hypothesized that B. bronchiseptica may survive within phagocytic amoeba cells. In order to determine whether B. bronchiseptica can survive intracellularly in amoeba cells, we performed a gentamicin protection assay and enumerated intracellular bacterial numbers at 1 and 24 h post gentamicin (p.g.) (Fig 1 and S1 Fig). While Klebsiella pneumoniae (previously referred to as K. aerogenes) failed to be recovered intracellularly at 1 h post infection, high numbers of B. bronchiseptica (3.1 x 107 colony-forming units [CFU] or approximately 10% of the inoculum) were recovered intracellularly from amoeba cells (p < 0.00005) (Fig 1). In fact, substantial numbers of B. bronchiseptica persisted for at least 24 h after gentamicin treatment (S1 Fig), suggesting that B. bronchiseptica has the ability to invade and survive in amoeba cells.
To visualize the association, D. discoideum were exposed to mCherry-expressing B. bronchiseptica. Following treatment with gentamicin that killed extracellular bacteria, D. discoideum were stained with endolysosomal (p80) and lysosomal (vatA and lamp-1) markers (Fig 2A). The association of these intracellular markers with the presence of fluorescent B. bronchiseptica further supports the location of the bacteria within D. discoideum (Fig 2A).
Visual examination of images of D. discoideum exposed to RB50 expressing mCherry taken at various time points revealed that ~90% of amoebae contained at least one B. bronchiseptica bacterium at 1 h, 2 h, and 4 h post gentamicin treatment (S2 Fig). Furthermore, we have imaged 65-nm sections of an amoeba population at 1 h post gentamicin treatment using electron microscopy. This imaging showed the majority of amoebae harboring intracellular bacteria with numbers ranging from 1 to 8 intracellular bacteria (average of 2.3 per amoeba) in these sections (Fig 2B & S3 Fig). While the intracellular bacterial numbers from thin microscopy sections do not accurately represent the total number of bacteria per amoeba cell, it is clear from the images that B. bronchiseptica are intracellular and that amoebae are able to contain multiple bacteria. These fluorescent confocal and electron microscopy images, in conjunction with the intracellular recovery data, support the ability of B. bronchiseptica to survive within a large proportion of the amoeba population for an extended period of time.
We hypothesized that the mechanism that allowed for the intracellular survival of B. bronchiseptica in D. discoideum would enable it to survive in other amoeba species. Therefore, we tested the ability of B. bronchiseptica to survive in Acanthamoeba castellanii, a free-living amoeba known to cause keratitis and encephalitis in humans. A gentamicin protection assay was performed similar to the one described above, and the number of bacteria that survived intracellularly was enumerated. B. bronchiseptica was recovered at 4 h post gentamicin treatment (S4 Fig), demonstrating its ability to survive within multiple species of amoebae for extended periods of time. Thus, in addition to the common soil organism D. discoideum, other amoebae may provide an environmental niche for these important mammalian respiratory pathogens. Future studies should assess whether the association of B. bronchiseptica with amoebae that naturally infect mammals could contribute to dissemination and transmission.
Several bacterial species that resist amoeba predation do so by surviving in single-celled amoebae [20]. Interestingly, these amoeba-resistant bacteria disrupt feeding, motility, and/or other behavior, effectively inhibiting the differentiation of D. discoideum into fruiting bodies [21,38,39]. The small number of bacterial species that do not prevent fruiting body formation actually aid the amoeba by serving as food, increasing amoeba spore counts over time, or producing metabolites that negatively affect competitor amoeba species [40–42]. We therefore investigated whether B. bronchiseptica prevents D. discoideum fruiting body formation. When added to lawns of B. bronchiseptica, D. discoideum spores were able to form mature fruiting bodies in the area where amoeba spores were deposited, indicating that B. bronchiseptica does not kill the amoeba or inhibit D. discoideum aggregation and fruiting body formation (S5A Fig). Additionally, similar numbers of amoeba spores were recovered from sori of D. discoideum grown on lawns of B. bronchiseptica and K. pneumoniae over time, suggesting that B. bronchiseptica does not negatively affect amoeba spore formation or persistence (S5B Fig).
B. bronchiseptica was able to survive intracellularly within the amoebae (Figs 1 and 2) yet permitted the full amoeba life cycle (S5 Fig); therefore, we hypothesized that B. bronchiseptica survives throughout the amoeba life cycle and may be carried to the fruiting body sorus. To determine whether bacteria are carried to the sorus and remain viable thereafter, amoeba sori grown on B. bronchiseptica or K. pneumoniae lawns were harvested at various time points, and the bacteria were enumerated (Fig 3). By day 9 post inoculation, the amoebae had formed fruiting bodies, which contained large numbers of B. bronchiseptica (approximately 5 x 103 CFU/sorus). The number of B. bronchiseptica recovered from sori increased ~400% by day 16 (~2 x 104 CFU/sorus) and doubled again by day 23 (~5 x 104 CFU/sorus), indicating that B. bronchiseptica are able to survive and multiply in amoeba sori over time. In comparison, K. pneumoniae was not recovered from sori throughout the time course (Fig 3). Imaging the sori grown on B. bronchiseptica RB50 harboring pLC003, a mCherry-containing plasmid, revealed fluorescence within the fruiting body (Fig 4 and S6 Fig). These results show that, in contrast to bacterial species that merely serve as food for amoebae, B. bronchiseptica are able to evade D. discoideum predation and travel with the amoebae to the fruiting body. Furthermore, these data indicate that B. bronchiseptica are able to survive, persist, and replicate in the fully formed amoeba fruiting bodies.
In order to determine if B. bronchiseptica are located intracellularly within the spores of the amoeba sori, we treated the sori with gentamicin (S1 Table). In contrast to our data above demonstrating B. bronchiseptica survival within a single-cell amoeba (Fig 1), B. bronchiseptica in the sori were not protected against gentamicin killing. Furthermore, confocal microscopy of sori grown on B. bronchiseptica RB50 pLC003 with calcofluor-stained spores showed that while B. bronchiseptica localizes to the amoeba sorus, it is outside of the spores (Fig 5 and S7 Fig). Together, these results suggest that B. bronchiseptica travels to the sorus but escapes D. discoideum cells, persisting and growing in their periphery.
The ability to associate with and replicate in amoeba sori suggests that B. bronchiseptica can take advantage of the amoeba strategy for geographic dissemination [19]. In order to determine whether B. bronchiseptica can be transported along with D. discoideum, we conducted a sorus passaging assay. Sori formed from amoebae grown on a B. bronchiseptica lawn for 16 d were collected, quantified, and diluted, and a fraction was then delivered to a fresh plate of K. pneumoniae. After the growth cycle was completed on that plate of K. pneumoniae, sori were collected and transferred to a fresh plate of K. pneumoniae. This process was repeated through seven passages (Fig 6). At each passage, B. bronchiseptica was recovered at high numbers from amoeba sori; from the fourth passage through the seventh, the bacterial load recovered per plate averaged ~7 x 105 CFU (Fig 6). Interestingly, while B. bronchiseptica was not observed intracellularly in amoeba spores, it maintained its association with the sori through multiple passages, despite the overwhelming abundance of an alternate food source, K. pneumoniae.
The high number of bacteria recovered at each passage highlights the ability of B. bronchiseptica to utilize D. discoideum as a vector for expanding its numbers. At each passage, the sori containing B. bronchiseptica were diluted 10-fold when transferred to a new lawn of K. pneumoniae. Yet, B. bronchiseptica proliferated such that high CFUs of B. bronchiseptica were recovered at each passage. Thus, by the seventh passage, B. bronchiseptica had expanded approximately 10,000,000-fold within sori. These data indicate that B. bronchiseptica are able to use D. discoideum as an expansion vector and can disseminate and grow along with the amoeba through consecutive life cycles.
The virulence genes up-regulated in the Bvg+ phase have been shown to be necessary for the infection of a variety of mammalian hosts [7,9,13], while the genes associated with the Bvg- phase have been hypothesized to be important for environmental survival outside of the mammalian host [5,9]. In order to determine whether the Bvg two-component system regulates genes involved in interactions with amoebae, we grew D. discoideum on lawns of wild-type B. bronchiseptica (RB50) or RB50 derivatives locked either in the Bvg- (RB54) or Bvg+ (RB53) phase. When sori from these plates were collected 10 d later, ~70% fewer Bvg+ phase-locked mutants were recovered than either wild-type or Bvg- mutants (Fig 7). Since wild-type B. bronchiseptica is expected to be in the Bvg- phase at the amoeba growth temperature (21°C) [4], these data suggest that the genes expressed in the Bvg- phase mutant are important for B. bronchiseptica survival in amoeba sori, while the Bvg+ phase is less conducive to B. bronchiseptica transport to or survival in sori. Notably, in three independent experiments, the small number of RB53 that were recovered included a substantial proportion of spontaneous Bvg- mutants (S2 Table), supporting a strong selective advantage for the Bvg- phase during interactions with amoebae. These data suggest that genes expressed in the Bvg- phase mediate successful interactions with amoebae that are ubiquitous in the environment.
The apparent advantage of the Bvg- phase bacteria during interactions with amoebae suggests that the expression pattern of B. bronchiseptica genes within the sori will be similar to the Bvg- phase. Moreover, D. discoideum survive and grow at 21°C, indicating that B. bronchiseptica within the sori (B. bronchisepticasori) may be in the Bvg- phase. Therefore, we compared the expression of B. bronchiseptica genes in the sori to Bvg- and Bvg+ phase-lock mutants grown under standard liquid culture conditions. The genes chosen for comparison are up-regulated under either Bvg- (cheZ, flhD) or Bvg+ (cyaA, fimC, fhaB) conditions [4]. Consistent with our hypothesis, B. bronchisepticasori expressed the chemotaxis protein gene cheZ similarly to the Bvg- mutant and significantly (p < 0.001) higher than the Bvg+ mutant (Fig 8A). In contrast, B. bronchisepticasori expression of flhD, a regulator of flagellum assembly, was significantly different from either Bvg- (p = 0.035) or Bvg+ (p = 0.002) mutants, potentially reflecting altered flagella-based motility within sori (Fig 8A).
The expression of the adenylate cyclase gene cyaA in B. bronchisepticasori was similar to the Bvg- mutant (Fig 8B), supporting our hypothesis that B. bronchiseptica in the sori resembles the Bvg- phase. However, B. bronchisepticasori had significantly higher expression of both the fimbriae biogenesis gene fimC (p = 0.002) and adhesion gene fhaB (p < 0.001) compared to the Bvg- mutant (Fig 8B). Altogether, we compared the expression of five genes, two of which (cheZ and cyaA) were supportive of our hypothesis that B. bronchisepticasori is in the Bvg- phase, while one gene (fimC) suggests a Bvg+ phenotype. Notably, the genes (fimC, flhD, and fhaB) that disagree with our initial hypothesis are involved in motility and adherence, which could affect bacteria persisting in the sori in a variety of ways.
The sticky sorus adheres to passing objects or animals to mediate the physical dispersal of D. discoideum spores. We therefore hypothesized that localization to the sori may similarly allow B. bronchiseptica to spread geographically. In order to demonstrate whether B. bronchiseptica in sori can be transmitted to a new location via an intermediary, we used flies to mechanically disperse fruiting body contents. To rigorously test the possibility, we coated SM/5 agar in a 50-mL conical tube with a lawn of K. pneumoniae (estimated >100,000,000 CFU) and introduced the contents of a sorus containing D. discoideum spores and a relatively small number (estimated <100,000) of B. bronchiseptica. Therefore, the only B. bronchiseptica present on the agar were those delivered in association with the amoeba sori. After fruiting bodies spread across the plate, flies were added for 1 min and then transferred to new plates either containing a lawn of K. pneumoniae (to assess amoeba transmission) or Bordet-Gengou (BG) agar with streptomycin (to assess B. bronchiseptica transmission) (Fig 9A). Transmission of amoeba spores to a new location via flies was demonstrated by the formation of plaques and fruiting bodies on amoeba-specific plates; plaques formed along the path walked by the fly, evidently growing where spores were deposited at each fly footstep (Fig 9B). Similarly, B. bronchiseptica colonies formed on BG plates corresponding to the fly’s path, demonstrating that B. bronchiseptica within the sori can travel with amoeba spores to seed colonies in new locations (Fig 9C).
To confirm the apparent association between the mechanical distribution and subsequent bacterial growth, we also tested the ability of ants to act as dissemination vectors (Fig 10). The ants’ progression across the BG plates was filmed, and their movements were analyzed by tracking software (Fig 10). When the positional data of the ant’s thorax were overlaid with the BG plate, it became evident that the growth of B. bronchiseptica correlates with the ant’s tracks across the plate (Fig 10). Thus, B. bronchiseptica can be transmitted with amoeba spores to new locations by environmental mechanical vectors including insects.
The ability of B. bronchiseptica to successfully disseminate along with amoebae is a compelling explanation for the high conservation of the genes expressed in the Bvg- phase. However, the value of maintaining two distinct life cycles associated with two different hosts, rather than specializing to one, should be dependent on the ability to switch between these distinct ecological niches. To assess the ability of B. bronchiseptica to move from the amoeba to the mammalian host, mice were challenged either with B. bronchiseptica grown in culture or with sori containing B. bronchiseptica at matched inocula of 5 x 105 CFU in a volume of 50 μL. Bacteria from amoeba sori and bacteria grown in culture similarly colonized the lungs and tracheas of mice by day 3 post inoculation (Fig 11). In order to rigorously test how efficiently B. bronchiseptica passaged on amoebae can colonize mice, we administered a very low dose of bacteria to the mouse (25 CFU in 5 μL). Even this very small number of bacteria, a tiny fraction of those present in individual sori, was sufficient to efficiently colonize mice (S8 Fig). Survival in amoeba therefore does not inhibit B. bronchiseptica transmission to mammalian hosts. Furthermore, B. bronchiseptica from amoeba sori that had been serially passaged on lawns of K. pneumoniae four consecutive times still retained the ability to efficiently colonize the mammalian respiratory tract even when administered at a low-volume and a low-dose inoculum of 25 CFU in 5 μL (S8 Fig). Together, these data suggest that while B. bronchiseptica association with amoebae involves the ability to modulate to the Bvg- phase, it does not inhibit the ability to modulate back to the Bvg+ phase in order to colonize a mammalian host.
Herein, we describe for the first time the ability of the respiratory pathogen B. bronchiseptica to thrive and disseminate outside the mammalian host while also identifying a novel role for the enigmatic Bvg- phase. B. bronchiseptica is shown to survive both within amoeba cells (Figs 1 and 2) and in association with amoeba sori (Figs 3, 4 and 5). While the ability of B. bronchiseptica to survive within the amoeba cells appears to be transitory, it is sufficient for the bacteria to evade amoebic predation and localize to the amoeba sori. Moreover, B. bronchiseptica forms a persistent relationship with the amoebae such that the two can be disseminated to new locations together via environmental vectors, such as insects (Figs 9 and 10). Once in the amoeba sori, B. bronchiseptica expand in number and disseminate along with D. discoideum as the latter feeds on other bacterial species, and B. bronchiseptica are repeatedly incorporated into the new fruiting bodies formed once the other bacterial food is depleted (Fig 6). Even after repeated passaging, B. bronchiseptica retains the ability to shift to the Bvg+ phase and efficiently infect a mammalian host (Fig 11 and S8 Fig). Thus, amoebae can act as environmental reservoirs and amplification vectors as well as modes of dissemination/transmission for B. bronchiseptica (Fig 12).
For a century, D. discoideum has served as a model organism for studies of cell migration, cell signaling, cytokinesis, cellular development, altruism, and phagocytosis [43–48]. Relatively recently, several important human pathogens were shown to survive amoeba predation, and amoebae are now studied as potential environmental reservoirs for these pathogens [21,49,50]. Interestingly, many of these amoeba-resistant bacteria have been shown to interfere with the amoeba life cycle—for example, preventing D. discoideum differentiation into mature fruiting bodies [21]—and such pathogens have only been shown to survive in single-celled amoebae. Meanwhile, other amoeba-resistant bacteria such as Burkholderia spp. have been shown to decrease spore production even when grown on an abundant food source [41]. Here, we have shown that B. bronchiseptica are able to utilize the amoeba life cycle such that they can be recovered from amoeba sori and disseminate with the amoeba to new geographical locations through multiple passages (Figs 3, 4, 8 and 9) while having no obvious detrimental effect on the amoeba (S5 Fig). This work reveals novel B. bronchiseptica-amoebic interactions that involve not only the utilization of the amoeba mechanism for dissemination but successful and successive growth and propagation along with the amoebae through multiple life cycles.
The ability of D. discoideum to form symbiotic relationships with bacteria has previously been described [42,51], as many amoebae (nonfarmers) consume all of their bacteria prey, while others (farmer amoebae) can carry a bacterial food source (i.e., K. pneumoniae) to novel locations. Recent work has shown that Burkholderia spp. have two distinct clades that are able to associate with both farmer and nonfarmer amoebae [51,52]. The farmer-associated Burkholderia clade promotes growth of farmer amoebae and inhibits nonfarmer amoebae while allowing farmers to carry both food and nonfood bacteria [51,53]. In contrast, the clade that colonizes nonfarmer amoebae imbues the amoeba with farmer characteristics, such as carriage of bacteria and association through multiple generations of growth [52] and decreased spore production in food-rich areas by excreting small molecules. Our data indicate that B. bronchiseptica has an apparent symbiotic relationship with D. discoideum that permits growth and expansion of both amoebae and bacteria. This relationship is with a nonfarmer amoeba strain as shown by the inability of K. pneumoniae to evade D. discoideum predation in the single-cell stage (Fig 1) or localize to the sori (Fig 3). Therefore, the Burkholderia spp. clade able to interact with nonfarmers would be most relevant to contrast with our work. Similar to those Burkholderia spp., B. bronchiseptica is able localize to the amoeba sori (Fig 3) and continue to associate with the amoeba through multiple life cycles even when grown on another viable and more plentiful food source (Fig 6). In contrast to observations with Burkholderia spp., however, we did not observe a reduction in spore production from the fruiting bodies of amoebae grown on B. bronchiseptica relative to those grown on K. pneumoniae (S5 Fig). Although Burkholderia spp. and Bordetella spp. are relatively closely related and are likely to share some aspects of their ability to successfully interact with amoebae, these dissimilarities suggest B. bronchiseptica may have novel mechanisms of interaction with amoebae. This work also demonstrates advantages for the bacterial traveler along with the D. discoideum dissemination mechanism, as well as the contribution of this interaction to the natural history of the bordetellae. Importantly, the ability to successfully transmit in these two independent settings, potentially regulated in part by the BvgAS two-component system, is uniquely demonstrated here for Bordetella spp. but may be observed in other soil-adapted organisms.
The enigmatic Bvg- phase of the Bvg regulon has previously been hypothesized to be important for temporary, short-term ex vivo survival either during transmission from host to host or in a hypothetical and undefined environmental reservoir [12–14]. Here, we show that the Bvg- phase contributes to a previously uncharacterized life cycle outside of the mammalian host. Based on these results, we propose a novel perspective of the Bvg+ and Bvg- phases that allows for independent but interconnected life cycles in distinct niches (Fig 12). The virulence genes associated with the Bvg+ phase allow for colonization of the mammalian host and the complex processes involved in dissemination between mammalian hosts (the Bvg+ life cycle). In contrast, the Bvg- phase not only enables survival in an extrahost environment but also contributes to a complete life cycle that includes propagation and dissemination in association with amoebae in the environment. The stable association through multiple modes of amoeba dissemination, as well as through several generations of the amoeba life cycle, demonstrates a well-adapted association. This novel life cycle provides the first potential explanation for the many genes that are highly conserved and specifically expressed in the Bvg- phase. During their apparently independent adaptation to a closed life cycle within humans, B. pertussis and B. parapertussis have lost hundreds of genes [54–56]. It may be that the role for many of these lost genes involves environmental survival and interactions with predators primarily encountered outside the mammalian host. An example consistent with this view is the Pseudomonas fluorescens strain that became an amoebic food source due to a genetic mutation in an inedible ancestor [40]. Future work will determine whether the ability to interact with amoebae is similarly affected by loss of genes in these and other Bordetella species.
In addition to surviving intracellularly in multiple species of amoebae (Fig 1 and S3 Fig), B. bronchiseptica recovered from amoeba sori was able to efficiently reinfect mice (Fig 11). Therefore, the widely prevalent D. discoideum, or some other amoeba, could serve as a transmission vector for B. bronchiseptica. Large-scale studies testing the prevalence of amoebae have shown that 9% of healthy human volunteers have amoebae present in the nasal mucosa, indicating that amoebae can colonize humans and appear to comprise part of the microbiota in healthy mammals [57]. Also, the high levels of amoebae recovered from water systems and the transmission of amoeba-associated bacteria via air handling systems [57–59] indicate that transmission of B. bronchiseptica to healthy and immunocompromised individuals via amoebae is possible. These findings have important implications for management strategies to control the spread of Bordetella species, which may require taking amoebae into account as an environmental reservoir and transmission vector.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Institutional Animal Care and Use Committee at the Pennsylvania State University at University Park, Pennsylvania (#46284 Bordetella-Host Interactions), or at the University of Georgia at Athens, Georgia (Bordetella-Host Interactions A2016 02-010-Y2-A3). Mice used in these experiments were humanely killed by using carbon dioxide inhalation.
B. bronchiseptica strains RB50 (wild-type), RB53 (Bvg+ phase-locked), and RB54 (Bvg- phase-locked) and K. pneumoniae (previously known as K. aerogenes) have been previously described [9,60]. B. bronchiseptica was grown and maintained on BG agar (Difco) supplemented with 10% defibrinated sheep’s blood (Hema Resources) and 20 μg/ml streptomycin (Sigma). K. pneumoniae was grown and maintained on Luria Bertani (LB) Media agar (Difco). For inoculation with culture-grown bacteria, B. bronchiseptica was grown overnight at 37°C to mid-log phase in Stainer Scholte (SS) liquid broth [61], and K. pneumoniae was grown at 37°C to mid-log phase in liquid LB Media Broth.
The plasmid pLC002 was constructed by cloning a gentamicin resistance gene into pBBR1- mcs2 [62]. The gentamicin cassette was amplified from pBBR1-mcs5 [62] using the primers 5′-AAAAAGCTTATGTTACGCAGCAGCAACG-3′ and 5′-ATAGAATTCTTAGGTGGCGGTACTTGG-3′. PCR products were purified using the Zymo DNA Clean and Concentrator kit (Irvine, California, US) prior to a double digestion with HindIII and EcoRI (cut sites in the primers are indicated by italics). The amplicon was then ligated into pBBR1-mcs2, which was similarly digested with HindIII and EcoRI. Sequence analysis was used to confirm the plasmid pLC002.
Inserting a mCherry gene into pLC002 created the plasmid pLC003. The following primers were used to amplify mCherry from pSCV26 [63]: 5′-AAGGGATCCATGGTGAGCAAGGGCGAG-3′ and 5′-AGCACTAGTTTACTTGTACAGCTCGTCC-3′. BamHI and SpeI sites, shown in italics, were designed within the primers to facilitate cloning of the mCherry gene into pLC002. PCR products were purified using the Zymo DNA Clean and Concentrator kit (Irvine, California, United States) prior to double digestion with BamHI and SpeI, as necessary. pLC002 was similarly digested with BamHI and SpeI. The digested mCherry amplicon was ligated into this plasmid and confirmed with sequence analysis to generate pLC003.
The plasmid pLC018 contains the mCherry gene cloned into pBBR1-mcs4 [62]. The following primers were used to amplify mCherry from pSCV26: 5′-AAGGGATCCATGGTGAGCAAGGGCGAG-3′ and 5′-ACCGAATTCTTACTTGTACAGCTCGTCC-3′. BamHI and EcoRI sites, shown in italics, were designed within the primers to facilitate cloning of the mCherry gene into pBBR1-mcs4. PCR products were purified using the Zymo DNA Clean and Concentrator kit (Irvine, California) prior to double digestion with BamHI and EcoRI. The PCR products were ligated into pBBR1-mcs4, which was also digested with BamHI and EcoRI. Sanger sequencing confirmed the proper ligation of the digested mCherry PCR product into pBBR1-mcs4 to generate pLC018.
Fluorescent images of B. bronchiseptica in association with D. discoideum were taken using a Nikon A1 Confocal Laser Microscope.
|
10.1371/journal.pntd.0001627 | Regulatory T Cells in the Pathogenesis and Healing of Chronic Human Dermal Leishmaniasis Caused by Leishmania (Viannia) Species | The inflammatory response is prominent in the pathogenesis of dermal leishmaniasis. We hypothesized that regulatory T cells (Tregs) may be diminished in chronic dermal leishmaniasis (CDL) and contribute to healing during treatment.
The frequency and functional capacity of Tregs were evaluated at diagnosis and following treatment of CDL patients having lesions of ≥6 months duration and asymptomatically infected residents of endemic foci. The frequency of CD4+CD25hi cells expressing Foxp3 or GITR or lacking expression of CD127 in peripheral blood was determined by flow cytometry. The capacity of CD4+CD25+ cells to inhibit Leishmania-specific responses was determined by co-culture with effector CD4+CD25− cells. The expression of FOXP3, IFNG, IL10 and IDO was determined in lesion and leishmanin skin test site biopsies by qRT-PCR. Although CDL patients presented higher frequency of CD4+CD25hiFoxp3+ cells in peripheral blood and higher expression of FOXP3 at leishmanin skin test sites, their CD4+CD25+ cells were significantly less capable of suppressing antigen specific-IFN-γ secretion by effector cells compared with asymptomatically infected individuals. At the end of treatment, both the frequency of CD4+CD25hiCD127− cells and their capacity to inhibit proliferation and IFN-γ secretion increased and coincided with healing of cutaneous lesions. IDO was downregulated during healing of lesions and its expression was positively correlated with IFNG but not FOXP3.
The disparity between CD25hiFoxp3+ CD4 T cell frequency in peripheral blood, Foxp3 expression at the site of cutaneous responses to leishmanin, and suppressive capacity provides evidence of impaired Treg function in the pathogenesis of CDL. Moreover, the concurrence of increased Leishmania-specific suppressive capacity with induction of a CD25hiCD127− subset of CD4 T cells during healing supports the participation of Tregs in the resolution of chronic dermal lesions. Treg subsets may therefore be relevant in designing immunotherapeutic strategies for recalcitrant dermal leishmaniasis caused by Leishmania (Viannia) species.
| The immune inflammatory response is a double edged sword. During infectious diseases, regulatory T cells can prevent eradication of the pathogen but can also limit inflammation and tissue damage. We investigated the role of regulatory T cells in chronic dermal leishmaniasis caused by species of the parasite Leishmania that are endemic in South and Central America. We found that although individuals with chronic lesions have increased regulatory T cells in their blood and at skin sites where immune responses to Leishmania were taking place compared to infected individuals who do not develop disease, their capacity to control the inflammatory response to Leishmania was inferior. However, healing of chronic lesions at the end of treatment was accompanied by an increase in the number and capacity of regulatory T cells to inhibit the function of effector T cells that mediate the inflammatory response. Different subsets of regulatory T cells, defined by the expression of molecular markers, were identified during chronic disease and healing, supporting the participation of distinct regulatory T cells in the development of disease and the control of inflammation during the healing response. Immunotherapeutic strategies may allow these regulatory T cell subsets to be mobilized or mitigated to achieve healing.
| Dermal leishmaniasis (DL) caused by species from the Viannia subgenus is characterized by a paucity of parasites in lesions associated with a robust inflammatory response and frequently follows a chronic course [1]. Both cutaneous and muco-cutaneous presentations of chronic dermal leishmaniasis (CDL) caused by Leishmania (Viannia) are associated with elevated cellular immune responses [2], [3]. Human studies and a recent murine model of chronic dermal disease have shown that a mixed Th1/Th2 cytokine pattern occurs in CDL caused by L. (Viannia), with prominent secretion of IFN-γ, IL-13, IL-10 and TNF-α [4]–[6]. Even though regulatory mechanisms are likely to have a significant role in the development of such an immune response, little is known about their impact in determining susceptibility and shaping disease outcome.
Regulatory T cells (Tregs) maintain tolerance to self tissues by inhibiting the action of auto-antigen reactive lymphocytes in an antigen-specific manner. In infectious diseases, Tregs also regulate the intensity and duration of immune responses, limiting damage to self tissues [7]–[8]. In the murine model of DL caused by L. major, Tregs have been shown to promote parasite persistence and reactivation in the resistant strain C57Bl/6 while inhibiting pathology in the susceptible Balb/c strain [9]–[12]. Tregs also suppressed pathology in mouse strains susceptible to disease caused by L. amazonensis [13]. These observations indicate that regulation of Leishmania-specific responses by Tregs may differentially alter the outcome of infection according to the susceptibility phenotype dictated by the host response. In resistant individuals, excessive Treg action may interfere with elimination of infection, whereas in susceptible individuals deficient Treg function may lead to excessive inflammation and dermal pathology.
Studies in humans infected with species from the Viannia subgenus have demonstrated that T cells with regulatory phenotype and function are present in cutaneous lesions [14]–[17]. An association between increased Foxp3 expression and unresponsiveness to treatment and chronic disease has been reported in human DL caused by L. guyanensis infection [16]–[17]. In contrast, no differences were found in the frequency of Tregs in peripheral blood between asymptomatically infected individuals (AI) and patients with CDL in L. braziliensis infection [14]. Hence, the role of Tregs in the pathogenesis of DL and their participation in the therapeutic response remain unclear.
The purpose of this study was to evaluate the role of Tregs in CDL caused by species of the Viannia subgenus and in the resolution of chronic lesions following treatment with pentavalent antimony. We found that lack of regulation of IFN-γ secretion by Tregs was associated with development of chronic disease, while an increase of Treg function after treatment was associated with lesion healing.
In this study, asymptomatic infection was considered to approximate clinical resistance to natural infection, and chronic disease to define a clinically susceptible phenotype, analogous to the healing and non-healing phenotypes in murine models of cutaneous leishmaniasis. Because AI and DL patients evidently remain infected indefinitely [18]–[23], exposure to Leishmania antigens would persist in both clinical outcomes. Since DL is generally a self-resolving disease, discrimination of spontaneous healing and chronic disease is not reliably determined during early or intermediate stages of evolution. However, longer times of evolution (chronic disease) have been shown to be associated with increased immune reactivity to Leishmania antigens including significantly higher antibody titers and DTH responses [2]. Furthermore, Th1/Th2 transcription factor expression and inflammatory cytokine responses distinguished asymptomatic clinical outcome and chronic cutaneous disease [5]. The rationale for analyzing the Treg response in asymptomatic infection and chronic disease was, therefore, that these phenotypically distinguishable outcomes are natural expressions of clinical resistance and susceptibility to human dermal leishmaniasis.
All participants provided written informed consent. The study protocol, consent forms and all procedures were approved by the CIDEIM Institutional Review Board for the ethical conduct of research involving human subjects.
Participants were residents of endemic areas for L. panamensis and L. braziliensis located within the southwestern Pacific coast region of Colombia (Departments of Valle del Cauca and Nariño) [24]. We included cases caused by both of these species of the Viannia subgenus because there is significant overlap of clinical presentations and broad measures of immune responses (DTH, lymphocyte proliferation and antibody titer) in patients with dermal disease caused by these species [2], [25]. AI had a positive LST and no evidence or history of dermal lesions. CDL patients had dermal lesions of ≥6 months duration, parasitological diagnosis by microscopic examination of tissue samples from lesions, culture or biopsy, and had not received anti-leishmanial treatment before enrollment. All subjects had negative serology for HIV and HTLV-1. A LST was performed in all participants and evaluated at 48 hours as previously described [26]. Leishmanin (Instituto Nacional de Salud, Colombia) was composed of equal amounts of L. panamensis and L. amazonensis promastigote proteins at a concentration of 5 µg/mL. Peripheral blood samples (100 mL), and skin biopsies from a leishmaniasis lesion and LST reaction site were obtained upon entry into the study. The species of parasite strains isolated from patients with a positive culture were determined as previously described [27]. CDL patients were treated with meglumine antimoniate at a dose of 20 mg Sb/kg/day. After treatment, a second 100 mL blood sample and a biopsy from the same lesion site were obtained and clinical responses were evaluated. Complete healing was defined as total re-epithelialization and absence of any evidence of inflammation for all lesions. Partial healing was estimated as the percentage of reduction of ulcer/plaque area or nodule volume. Skin biopsies from four healthy volunteers were obtained for normalization of gene expression data. Biopsies were obtained using a 4 mm disposable punch, embedded in OCT, frozen and stored in liquid nitrogen until processed.
PBMCs were isolated by centrifugation over Histopaque-1077 (Sigma-Aldrich, St. Louis, MO). CD4+CD25− and CD4+CD25+ cells were isolated by MACS using the CD4+CD25+ Regulatory T cell isolation kit (Miltenyi Biotec, Bergisch-Gladbach, Germany) following the manufacturer's instructions. Purity assessed by staining with anti-CD4 and anti-CD25 was ≥90% for both populations. Monocytes were isolated by allowing PBMCs to adhere to plastic for 2 hours followed by three washes with PBS. Purity assessed by staining with anti-CD14 was ≥65% and contaminating CD4hi lymphocytes were <5%.
PBMCs from 12 AI, 14 CDL patients before treatment, and 11 CDL patients after treatment were suspended in FACS buffer (1× PBS, 1% BSA, 0.1% NaN3) and incubated for 30 minutes with anti-CD4-APC (BD Biosciences, San Jose, CA), anti-CD25-PE (Miltenyi Biotec) and anti-CD127-FITC (eBioscience, San Diego, CA) or anti-CD4-FITC (BD Biosciences), anti-glucocorticoid-induced TNF receptor family-related protein (GITR)-PE and anti-CD25-APC (eBioscience). For evaluation of forkhead box p3 (Foxp3), cells were stained with anti-CD4-FITC and anti-CD25-APC, fixed and permeabilized using the Fixation and Permeabilization kit (eBiosciences), and incubated with anti-Foxp3-PE or anti-Rat IgG2a-PE (eBioscience). After washing, cells were analyzed in a Navios flow cytometer (Beckman Coulter, Brea, CA) and data was analyzed using FlowJo 7.6 software (Tree Star, Inc., Ashland, OR). Gates used for analysis were set using isotype controls.
Killed promastigotes of the L. panamensis strain MHOM/COL/81/L13 were prepared by the freeze/thaw method, as previously described [28]. CD4+CD25− cells from 12 AI, 14 CDL patients before treatment, and 11 CDL patients after treatment were incubated with 5 µM carboxyfluorescein diacetate succinimidyl diester (CFSE, Invitrogen, Carlsbad, CA) for 10 minutes at 4°C, washed three times with PBS and resuspended in RMPI 1640 (Sigma-Aldrich) with 10% FBS (Gibco, Carlsbad, CA), 2 mM L-glutamine, penicillin (100 U/mL) and streptomycin (100 mg/mL). 1×106 CD4+CD25− cells per well were distributed in 24 well plates with 2×105 L. panamensis promastigotes, 5×105 monocytes from the same subject (used as antigen presenting cells, APCs), or phytohaemagglutinin (PHA, Sigma-Aldrich) 10 µg/mL, and CD4+CD25+ cells at 1∶0, 4∶1 or 1∶1 CD4+CD25−: CD4+CD25+ cell ratios, in a final volume of 1 mL. After 5 days of incubation at 37°C with 5% CO2, cells and supernatants were harvested for evaluation of proliferation and cytokine secretion, respectively. Cells were stained with anti-CD4-APC and analyzed by flow cytometry. CFSE fluorescence in CD4+CD25− cells was evaluated in the CD4+ gate. Unlabeled CD4+CD25+ cells and CD4+ lymphocytes contaminating the APC preparation were gated out of the analysis (Figure S1A). Regions were drawn for each proliferation peak induced by PHA (Figure S1B) and the proliferation index (PI) was calculated with the formula PI = Σ(% in region×2n−1), where n is the region number [29]. Interferon-γ (IFN-γ) and interleukin-10 (IL-10) were measured in supernatants by ELISA as previously described [28]. Co-cultures were considered to have positive proliferation or IFN-γ secretion when the value in wells with CD4+CD25− cells, APCs and L. panamensis or PHA was higher than the mean+2SD of the control wells without APCs, L. panamensis or neither. The percent inhibition of proliferation or IFN-γ secretion by CD4+CD25+ cells was calculated in positive co-cultures using the formula: (value without CD4+CD25+ cells - value with CD4+CD25+ cells)/(value without CD4+CD25+ cells) ×100. Negative results were considered 0% inhibition.
RNA was isolated from LST biopsies from 5 AI and 7 CDL patients and lesion biopsies from 11 CDL patients both before and after treatment using the RNeasy mini kit (QIAGEN, Valencia, CA) according to the manufacturer's instructions. For cDNA synthesis, 100 ng total RNA was transcribed with the High-capacity cDNA reverse transcription kit (Applied Biosystems, Foster City, CA), following the manufacturer's instructions. Gene expression was measured in real-time with the CFX966 Real Time System (Bio-Rad, Hercules, CA) using Taqman Universal PCR master mix and Taqman gene expression assays for GAPDH (Hs99999905m-1), FOXP3 (Hs01085835-m1), IFNG (Hs00174143-m1), IL10 (Hs00174086-m1) and IDO (Hs00158027-m1) genes (Applied Biosystems), following the manufacturer's instructions. The expression levels of FOXP3, IFNG, IL10 and IDO relative to healthy skin were calculated using the ΔΔCT method, as previously described [30], using GAPDH as endogenous reference gene and four healthy skin biopsies as calibrators.
The Kolmogorov-Smirnov test was used to determine parametric or non-parametric distribution of the data. Thereafter, for comparisons between AI and CDL patients, parametric data were analyzed using student t-test, and the Mann-Whitney test was applied for non-parametric data. For comparisons between CDL patients before and after treatment, parametric and non-parametric data were analyzed using the paired t-test and the Wilcoxon signed-rank test, respectively. Correlation analyses were conducted using the Spearman coefficient. Statistical significance was defined as p<0.05. All data were analyzed using Prism 5 software (GraphPad Software, Inc., La Jolla, CA).
Twelve AI and 14 CDL patients participated in this study. There were no significant differences in age, gender or size of LST reaction between AI and CDL patients (data not shown). The clinical characteristics of CDL patients are summarized in Table 1. Three patients were lost to follow-up. The other eleven patients were evaluated at a second visit within 24 days of completing treatment. Five patients showed complete healing, and lesions in six patients had substantially improved though not completely healed (Table 1).
High expression of CD25 has been described as a reliable marker for Tregs [31]–[34]. Lack of expression of CD127, expression of GITR, or expression of Foxp3 have also been described as Treg markers [34]–. Therefore, to evaluate the frequency of Tregs in the peripheral blood of AI and CDL patients, we measured the frequency of CD25hi CD4 T cells that expressed each of these markers. We found that CD4+CD25hiFoxp3+ cells were more abundant in the peripheral blood of CDL patients (0.9±0.4% in CDL patients vs. 0.4±0.2 in AI, p = 0.0003; Figure 1A), while significant differences were not observed in the frequency of CD4+CD25hiCD127− and CD4+CD25hiGITR+ cells between the two groups (Figures 1B and C).
We evaluated the functional capacity of Tregs to suppress Leishmania-specific proliferation and IFN-γ secretion using co-cultures of CD4+CD25− cells (effectors) and CD4+CD25+ cells (Tregs) from peripheral blood. In the absence of Tregs, proliferation and IFN-γ secretion were induced by L. panamensis antigens in effector T cells from all CDL patients and 9 of 12 AI. Proliferation was significantly higher for CDL patients in relation to AI (p = 0.037, Figure 2A), while no statistically significant differences were detected for IFN-γ secretion (Figure 2B). No significant effector functions were observed in the absence of either APCs or L. panamensis antigens (Figures 2A and B), indicating that they were the result of antigen-specific APC-T cell interactions. L. panamensis induced similar levels of proliferation and IFN-γ secretion in patients infected with L. panamensis or L. braziliensis and in patients in which infecting species was undetermined (data not shown), confirming recall of T cell responses across species. High levels of proliferation were induced by PHA in all subjects (Figure 2A), while IFN-γ secretion was low with this stimulus under the conditions of assay (Figure 2B), being positive only for 6 AI and 3 CDL patients.
Suppression assays with L. panamensis stimulation were performed at effector∶Treg ratios of 4∶1 and 1∶1 in subjects that had positive effector functions, except for three AI from whom the number of CD4+CD25+ cells isolated was insufficient to evaluate the 1∶1 ratio. At a 4∶1 effector∶Treg ratio, no statistically significant differences in antigen-specific suppression were detected between AI and CDL patients (Figures 2C and D). At a 1∶1 effector∶Treg ratio, suppression of IFN-γ secretion was significantly higher in AI compared to CDL patients (80±6.1% vs. 48±9.2%, p = 0.044; Figure 2D) whereas suppression of proliferation was not significantly different between the two groups (Figure 2C). To evaluate suppression after a polyclonal stimulus, inhibition of proliferation by Tregs at a 4∶1 ratio was evaluated after PHA stimulation. No significant differences in suppression of proliferation induced by PHA were detected between AI and CDL patients (mean inhibition 19.5±4.1% and 25.9±5.8%, respectively; p = 0.313; Figure 2C).
We next examined the influence of Tregs in situ by measuring the transcription of genes related to immune regulation and inflammation at the site of injection of leishmanin antigen from 5 AI and 7 CDL patients. We evaluated the relative expression of four genes: FOXP3, IFNG, IL10, and 2,3 indoleamine deoxygenase (IDO), an enzyme expressed by APCs that is involved in immune regulation through the catabolism of the essential aminoacid tryptophan [38], [39]. We found that the relative expression of FOXP3 was significantly higher in CDL patients in relation to AI (relative expression 12.7±4.3 vs. 0.7±0.8, p = 0.046, Figure 3A). No statistically significant differences were found for the other genes, although several individuals with chronic disease showed upregulation of IFNG and IDO (Figures 3B–D).
At the end of treatment of CDL, the proportion of CD4+CD25hiCD127− cells increased significantly, from 1.5±0.2% to 2.3±0.3% (p = 0.0009, Figure 4B), whereas the frequency of CD4+CD25hiFoxp3+ and CD4+CD25hiGITR+ cells did not change significantly (Figure 4A and C). Proliferation and IFN-γ secretion in the absence of Tregs did not change significantly at the end of treatment, although significant proliferation was no longer observed in one patient (data not shown). In the presence of Tregs at a 1∶1 effector∶Treg ratio, 8 of 10 CDL patients demonstrated increased inhibition of proliferation by CD4+CD25+ cells at the end of treatment and 9 of 11 showed an increase in inhibition of IFN-γ secretion (Figures 4D and F). Inhibition of both parameters increased significantly for the group as a whole, with the mean percent inhibition of proliferation increasing from 21.9% to 46.5% (p = 0.025, Figure 4D) and the mean inhibition of IFN-γ secretion rising from 47.2% to 81.3% (p = 0.007, Figure 4F). Suppression of proliferation induced by PHA did not change significantly after treatment (p = 0.99; Figure 4E).
The relative expression of FOXP3 and IL10 did not change significantly after treatment of CDL (Figures 5A and C). The relative expression of IFNG decreased in 8 out of 11 patients at the end of treatment, with the mean value declining from 127.0±80.3 to 24.3±7.9. However, this difference was not statistically significant (p = 0.148, Figure 5B). Finally, although the expression of IDO in chronic lesions at the start of the study varied widely between individuals, at the end of treatment the expression of this gene was significantly downregulated from 1087±601 to 154±50 (p = 0.037, Figure 5D). Because IDO expression in APCs is upregulated by both inflammatory signals (particularly IFN-γ) [40], [41] and Tregs (through CTLA4-B7 interactions) [42], we examined whether IDO expression was correlated with that of IFNG or FOXP3 in CDL lesions. Our analysis revealed a significant positive correlation between IFNG and IDO expression. In contrast, FOXP3 expression was not significantly correlated with IDO expression (Figure 5E). Thus, IDO expression in the skin of CDL patients was most likely upregulated by IFN-γ.
To determine whether suppression of effector functions was attributable to IL-10 secretion by CD4+CD25+ cells, we evaluated IL-10 in the supernatants of the effector-Treg co-cultures. The presence of CD4+CD25+ cells in the co-cultures did not result in higher IL-10 concentrations in any of the study groups (Figure 6). To the contrary, analysis of co-cultures at the 1∶1 ratio (n = 31) for all groups revealed significantly lower IL-10 secretion in the presence of CD4+CD25+ cells (Figure 6). Therefore, inhibition of effector CD4 T cell functions in this in vitro system was not attributable to IL-10 secretion by Tregs. Rather, these results indicate that IL-10 secretion by CD4+CD25− effector T cells and/or APCs was also inhibited by Tregs.
The regulation of adaptive immune responses is indispensable for the effective clearance of antigen without harm to self tissues. Although the importance of Tregs in this process is well established, evaluation in human disease is challenging because of the increasing recognition of the complexity of their phenotype and function. To clarify the role of Tregs in the pathogenesis and healing of DL caused by species of the Viannia subgenus, we studied a cohort of subjects from an area in Colombia endemic for L. panamensis and L. braziliensis. Comparisons of the frequency and suppressive function of cells expressing Treg markers in peripheral blood and the expression of genes related to inflammation and regulation in the skin of AI and CDL patients, and CDL patients before and after treatment, yielded evidence that Tregs participate in clinical resistance/susceptibility and lesion healing.
Patients with chronic dermal disease had a significantly higher frequency of CD25hiFoxP3+ CD4 T cells in their peripheral blood than infected individuals who did not develop disease. Additionally, FOXP3 expression was upregulated at skin sites of challenge with leishmanin antigen only in CDL patients. These results are consistent with several examples of chronic disease in which Foxp3 is upregulated in T cells [43]–[46] and previous reports of infiltration of dermal lesions caused by L. guyanensis and L. brazilienisis by Foxp3+ cells [15], [16], [47] and upregulation of FOXP3 in lesions from post Kala Azar DL [48]. Foxp3 may therefore constitute a marker for chronic DL caused by L. (Viannia) species.
Although CDL patients had higher numbers of CD25hiFoxp3+ cells in peripheral blood than AI, the capacity of their CD4+CD25+ cells to inhibit pro-inflammatory IFN-γ secretion by effector T cells elicited by Leishmania antigens was lower. This suggests that the CD25hiFoxp3+ cells that accounted for the higher frequency in CDL patients are not specific for Leishmania antigens. Since Foxp3 is a marker of natural Tregs generated in the thymus after interaction with self antigens [37], [51], these cells may be a subset of Tregs specific for self antigens that are released during the inflammatory response [7], [52], [53]. Alternatively, the higher proportion of CD25hiFoxp3+ cells may reflect chronic activation in CDL patients, since FoxP3 can be transiently expressed by activated T cells [49], [50]. However, we do not favor this explanation because high expression of CD25 has been consistently proven to be a marker of Tregs [31]–[34] and effector cells expressing FoxP3 are mostly not CD25hi [50]. Thus, we postulate that CD25hiFoxp3+ Tregs, although more frequent in CDL, are not responsible for the antigen specific suppression demonstrated in the co-culture assays. Rather, this suppression is attributable to a Leishmania-specific, Foxp3− subset. However, we cannot rule out the possibility that Leishmania-specific effector cells acquired resistance to regulation by Tregs after chronic antigenic stimulation in CDL.
Clinical evaluation at the end of treatment with pentavalent antimony showed that all patients had initiated a healing response. Half had already healed completely while the other half showed significant improvement (≥50%, Table 1). Clinical follow-up was possible for 7/11 patients at ≥13 weeks after starting treatment and confirmed that all seven had completely healed (data not shown). Thus, no evidence of treatment failure was seen in this study cohort. The healing responses in this CDL cohort were associated with significant increases in the frequency of CD25hiCD127− CD4 T cells in peripheral blood and in Leishmania-specific inhibition of CD4 T cell effector functions by CD4+CD25+ cells. In fact, of the 11 patients, 10 were able to inhibit IFN-γ secretion by ≥75% at the end of treatment and 9 showed an increase in CD4+CD25hiCD127− cells. Furthermore, transcription of IFNG at the lesion site was not significantly upregulated at the end of treatment, but rather decreased in 8 of the 11 patients. Interestingly, we found that Foxp3 expression did not change after treatment in CD4+CD25hi blood cells or at the lesion site, suggesting that a Foxp3+ subset of Tregs was not responsible for the heightened suppression that was associated with healing. Rather, these findings support the participation of a CD127−, Leishmania-specific subset of Tregs in the healing of chronic dermal lesions caused by L. (Viannia).
Induction of Tregs during chronic infections is the result of antigen presentation in a particular cytokine environment [7], [53]. In the current study, parasite killing mediated by the anti-leishmanial drug would presumably lead to release of antigens, new antigen presentation events and Treg induction and/or activation in the peripheral lymphoid organs [54]. Newly induced Tregs would home to the lesion site to mitigate the inflammatory response and aid in tissue repair. Homing of Tregs to the skin is essential for the maintenance of skin homeostasis and for suppression of Th1 cell-mediated responses [55], [56]. Furthermore, Tregs have been shown to contribute substantially to tissue repair by providing regulation at sites of healing in many experimental models. In an in vitro wound healing model, Tregs were shown to counteract Th17 cell-mediated inhibition of fibroblast migration into the wound [57], and tissue regeneration after kidney injury was shown to depend on inhibition of pro-inflammatory cytokine secretion by Tregs [58]. In another in vitro model, intact extracellular matrix components that gradually replace inflammation-promoting fragmented components during tissue repair induce Tregs and activate their function [59]. Therefore, healing of dermal lesions such as those present in CDL would likely benefit from immune regulation by Tregs. Since all patients had initiated a healing response at the end of treatment, our results are consistent with the participation of a subset of Foxp3− CD127− Leishmania-specific Tregs in lesion resolution.
The limited absolute numbers of CD4+CD25+ cells allowed us to test only two effector∶Treg ratios. Significant differences in antigen-specific suppression were detected at the 1∶1 ratio, but not at the 4∶1 ratio. Since suppression by Tregs is activated through cognate peptide/MHC-TCR interactions [60], the inability to detect suppression at a higher ratio is probably due to the expected low frequency of Leishmania-specific Tregs at the outset of co-culture and the absence of proliferation by these cells [61] (as evidenced by lack of variation in forward scatter, Figure S1A). Therefore, even though the number of Leishmania-specific CD25hi T cells in the co-culture is unknown, the observation of significant suppression indicates that the number reached an effective threshold under the assay conditions at a 1∶1 ratio. In addition, the CD4+CD25+ preparation includes activated effector T cells that upregulate CD25 and, therefore, the number of Tregs in the culture would have been diluted by these cells. For these reasons, we believe that the suppression observed in our system at a 1∶1 ratio reflects the presence of circulating antigen-specific Tregs that can contribute to the mitigation of the inflammatory response upon homing to the lesion site.
We examined the participation of IL-10 in the inhibition of CD4 T cell effector functions in our co-cultures because this cytokine is secreted by T cells with an effector phenotype in intracellular parasitic infections [62]–[64] as well as by Tregs. Furthermore, we have previously determined in a population from the same region that IL-10 is expressed in susceptible individuals and related to both development of CDL and IFN-γ secretion [4]. Hence, it was conceivable that inhibition was due to IL-10 secretion by a CD25+ activated effector population present in the CD4+CD25+ preparation and not by Tregs. The diminished production of IL-10 in co-cultures including CD4+CD25+ cells demonstrated that IL-10 was not responsible for suppression by these cells, supporting the interpretation that the observed inhibition was indeed mediated by bona fide Tregs. We also found that IL10 transcription was not significantly changed at the lesion site immediately following treatment, suggesting that this cytokine is not prominent in the healing process.
IDO is an enzyme expressed by APCs that metabolizes tryptophan, an essential aminoacid for both lymphocyte proliferation and parasite survival [38], [39]. Hence, the expression of IDO in leishmaniasis lesions may regulate the immune response and aid in parasite eradication. Its expression is induced by inflammatory cytokines, including IFN-γ [40], [41], and by Tregs through interaction of CTLA-4 and B7 molecules [42]. Therefore, IDO transcription in the skin may reflect both the inflammatory environment and Treg function. In this study, we found that the expression of IDO was correlated with that of IFNG and not to FOXP3 in CDL patients. This indicates that IDO induction in APCs is not a likely mechanism of suppression used by Tregs that infiltrate sites of ongoing immune responses to Leishmania in CDL patients. Rather, IDO expression reflects the reduced inflammatory environment present at these sites, which is consistent with increased function of a subset of Tregs after treatment.
The relatively small sample size limited our ability to detect statistically significant differences in the expression of the genes IFNG and IDO between AI (healing/clinical resistance) and patients with chronic disease (non-healing/clinical susceptibility) even though several of the latter showed marked upregulation. Furthermore, although we did not determine the mechanism of regulation by CD4+CD25+ cells, the results of functional as well as phenotypic analyses provide compelling evidence of the participation of distinct Treg subsets in both pathogenesis and resolution of CDL caused by Leishmania species of the (Viannia) subgenus. Further characterization of these subsets is warranted since immunotherapeutic strategies targeting these regulatory cells could promote healing of recalcitrant presentations of leishmaniasis.
|
10.1371/journal.ppat.1003098 | Teleost Fish Mount Complex Clonal IgM and IgT Responses in Spleen upon Systemic Viral Infection | Upon infection, B-lymphocytes expressing antibodies specific for the intruding pathogen develop clonal responses triggered by pathogen recognition via the B-cell receptor. The constant region of antibodies produced by such responding clones dictates their functional properties. In teleost fish, the clonal structure of B-cell responses and the respective contribution of the three isotypes IgM, IgD and IgT remain unknown. The expression of IgM and IgT are mutually exclusive, leading to the existence of two B-cell subsets expressing either both IgM and IgD or only IgT. Here, we undertook a comprehensive analysis of the variable heavy chain (VH) domain repertoires of the IgM, IgD and IgT in spleen of homozygous isogenic rainbow trout (Onchorhynchus mykiss) before, and after challenge with a rhabdovirus, the Viral Hemorrhagic Septicemia Virus (VHSV), using CDR3-length spectratyping and pyrosequencing of immunoglobulin (Ig) transcripts. In healthy fish, we observed distinct repertoires for IgM, IgD and IgT, respectively, with a few amplified μ and τ junctions, suggesting the presence of IgM- and IgT-secreting cells in the spleen. In infected animals, we detected complex and highly diverse IgM responses involving all VH subgroups, and dominated by a few large public and private clones. A lower number of robust clonal responses involving only a few VH were detected for the mucosal IgT, indicating that both IgM+ and IgT+ spleen B cells responded to systemic infection but at different degrees. In contrast, the IgD response to the infection was faint. Although fish IgD and IgT present different structural features and evolutionary origin compared to mammalian IgD and IgA, respectively, their implication in the B-cell response evokes these mouse and human counterparts. Thus, it appears that the general properties of antibody responses were already in place in common ancestors of fish and mammals, and were globally conserved during evolution with possible functional convergences.
| In humans and mice, a broad and robust antibody response is the most effective system of defense against pathogens. While all vertebrates but the most ancient use an antibody system, it remains unknown if the clonal selection rules are conserved across vertebrates. Using high-throughput sequencing and antibody repertoire analysis, we show in teleost fish that virus infection induces a typical antibody response with clonal expansions. In spleen, the antibody response mainly consists in IgM, which is found across vertebrates, with a significant contribution of the fish-specific mucosal antibody IgT. IgD does not contribute as already observed in mammals. Although antibody molecules can be of different nature and structure across species, the general features of adaptive immunity were already in place in common ancestors of fish and mammals, and globally conserved during evolution while the anatomy and physiology of the immune system diverged.
| The immune system of mammals is characterized by the presence of B and T lymphocytes, each carrying a single receptor for antigen generated through somatic rearrangements of V-(D)-J genes. Upon infection specific B and T lymphocytes develop protective responses against the intruding pathogen, and after disease resolution they maintain an increased resistance against the eliciting agent through persistence of memory cells. The two properties of specificity and memory proceed from the fundamental fact that a few T and B cell clones expressing receptors specific for the eliciting antigen expand and carry out the adaptive response, as formulated in the clonal theory of immunity [1], [2].
The origin and development of this remarkable defense system during the evolution of vertebrates remain poorly understood. The primordial adaptive immune system of extinct vertebrates is no longer accessible, but it can be approached indirectly through comparative analyses of B and T cell responses in distant contemporary species such as teleost fishes and mammals, which diverged more than 350 million years ago. Indeed, common fundamental features like typical B and T lymphocytes were already present in their last common ancestor [3]. However, it is important to note that the organization of lymphoid tissues, where lymphocytes develop, encounter with antigen, and get activated, is profoundly different between fish and mammals [4]. While in mammals B lymphopoiesis occurs in bone marrow, fish lack this tissue and their B cells differentiate in the anterior kidney or pronephros [5], [6]. Fish also lack lymph nodes, so that B and T cell responses occur mainly in spleen and mucosal territories. Due to these anatomical differences, it might be expected that fish and mammal T and B cell responses show distinct functional properties.
We previously showed that T cell diversity was extensive in central lymphoid organs of fish, and that antiviral T cell response was comparable to that of mammals, with typical public and private components [7]. However, no T cell response was observed at 10°C in trout, and at 16°C the kinetics of the response was slower compared to mammals. The TCRβ (TRB) repertoire of the gut was highly diverse even in adult fish, in contrast to the one of mouse and human [8]. Taken together, these studies indicate that some, but not all, features of T cell immunity are similar between fish and mammals.
Regarding B cell diversity and responses, the lack of germinal centers and lymph nodes could have more dramatic consequences than for T cells since B cells require a proper microenvironment for differentiation and maturation [9], [10]. While B cell responses have been observed after immunization with antigens in many fish species, allowing efficient vaccination against viral diseases, some species like Atlantic cod (Gadus morhua) do not show specific antibodies (Ab) responses (probably due to the lack of MHC class II molecules) despite high amount of serum “natural” Abs [11], [12]. In rainbow trout (Oncorhynchus mykiss), different types of Ab secreting cells have been distinguished, as in mouse and human [13]: plasmablasts, which produce low amount of Ab, replicate, and express low level of B cell receptor (BCR), and plasma cells (long- or short-lived), which produce high level of Ab, do not replicate nor express BCR [5], [6], [14]. Importantly, while B cells encounter their specific target in spleen or kidney, and differentiate into Ab secreting cells in these tissues, mature plasma cells migrate and persist in the anterior kidney [15], as observed in mouse or in human for the bone marrow [16]. However, fish B cells lack efficient affinity maturation, as observed also in amphibians [9], [10].
Teleost fish have three heavy chain isotypes: μ and δ that correspond to the IgM and IgD classes found in all vertebrates with jaws (gnathostomata), and τ that encodes the IgT class, which is specific to fish [17]. IgM and IgT are never co-expressed by the same B cell, which identifies two distinct lineages of B cells in fish [18]. In fact, the configuration of the IG locus ensures the exclusive expression of either IgM/IgD or IgT; isotypic commutation and switch recombination do not occur in fish [19]. It is currently thought that IgM is primarily a systemic immunoglobulin. The function of fish IgD is still elusive, yet it may have a role in innate immunity. In the channel catfish (Ictalurus punctatus), secreted IgD lacking the antigen-specific V domain could bind to basophils to induce proinflammatory cytokines [20]. IgT is specialized in gut mucosal immunity in rainbow trout: a gut protozoan parasite elicited a local IgT response, while the IgM response was restricted to the serum [21]. Additionally, intestinal bacteria were coated by IgT, suggesting that this IG class might play a role in the interactions between the host intestinal mucosa and microflora [21].
While the diversity of antibody response was initially considered lower in fish and amphibians compared to mammals, the recent discoveries of fish genomics revealed that the potential combinatorial Ab repertoire is probably bigger in many fish species than in humans and mice [22]. The first complete description of the V domain repertoire expressed by a vertebrate organism was recently achieved from whole healthy zebrafish [23], with the usage of 454 GS FLX high-throughput pyrosequencing. Focused on the naive IgM repertoire, this study showed that most of the possible combinations of IGH V, D and J genes were expressed in all individuals with a similar frequency distribution. A network analysis of these data found a common architecture of sequence diversity in different fish, beyond the individual differences [24], [25].
The implication of the different isotypes in the defense against pathogens is still largely unknown in fish, as well as the clonal complexity of their respective responses. Here, we have performed a comprehensive study of the clonal complexity of B cell repertoire of healthy fish as well as after systemic infection, for all expressed isotypes. To investigate if the clonal nature of the B cell response and the various isotype contributions were conserved across species, we characterized not only the IgM, but also IgT and IgD responses of the rainbow trout (Oncorhynchus mykiss) against a rhabdovirus, the Viral Hemorrhagic Septicemia Virus (VHSV). We considered the available repertoire expressed by the cells that can respond to the antigens at a given moment in a given tissue of the individual. In this line, the available repertoire is not a mere list but also refers to the frequency of the different specificities. As in a recent report by Ademokun et al. [26], we used a combination of CDR3 spectratyping and 454 GS FLX pyrosequencing. Our data reveal a highly diversified available VH domain repertoire, and demonstrate that fish Ab response presents features typical of adaptive responses previously described for mammals, as we previously observed for T cells. Remarkably, we observed a broad polyclonal response implicating a large number of VH subgroups.
As a start, we characterized the expressed Ab repertoire from three healthy homozygous isogenic fish. To this end, we used a spectratyping approach that determines the profile of CDR3 length distribution for each VHC combination, based on PCR reactions between Ig constant (C) and variable (V) regions [27]. First, heavy chain rearranged transcripts (IGH V-D-J-C) were amplified using a set isotype-specific primers for IgM, IgD and IgT, respectively, and a set of IGHV subgroup-specific primers that can amplify all members of the 11 known IGHV groups (Figure 1A and Figure S1A) [28]. Of these 11 VH groups, we found that seven were used by all three isotypes (Figure S1B), suggesting that a large fraction of the potential repertoire was expressed in peripheral B cells from healthy fish. To assess the diversity of these VH (V-D-J) rearrangements, each VHC PCR product was then subjected to a run-off reaction using a labeled internal C-specific primer. The resulting products, which reflect the variable numbers of nucleotides present in the CDR3 region, were analyzed on a sequencing apparatus to determine their size distribution. These analyses produced bell-shaped profiles for all VHC primer combinations in healthy fish (Figure 1B), which is a typical feature of immune repertoire expressed by naïve lymphocytes in mammals [27]. The profiles consisted of 4–10 peaks for VHCμ (IgM) and VHCδ (IgD), and 5–13 peaks for VHCτ (IgT), separated by intervals of three nucleotides as expected from in-frame transcripts (Figure 1B; profiles for all expressed VH subgroups are shown in Figure S2A). IgT junctions were therefore longer than those associated with IgM or IgD, as previously noted by others [18]. The only rearrangement that displayed a non-bell shaped profile in all naïve fish was VH4-Cδ, which had an extra peak associated with a unique CDR3 (ARGTEYYFDY) (Figure 1B).
To analyze the variability of these profiles in a quantitative manner we used the ISEApeaks statistical analysis software [29], [30]. ISEApeaks computes the “perturbation” of a given VHC spectratype profile in comparison with a reference as described in Material and Methods in the section “CDR3 length spectratyping analysis”. Taking the average of control fish profiles as a reference, we quantified the inter-individual variability of VHC spectratype profiles among healthy fish for each isotype. These results were represented by a matrix of distance in red color scale (Figure 1C-D-E, left panels). The perturbation scores calculated in this way were low, as evidenced by their appearance in grey or light pink, indicating that the CDR3 length profiles were remarkably similar for the different healthy fish as expected for truly naïve animals belonging to the same fish clone.
We also compared the CDR3 length diversity of IgM, IgD, and IgT through an index based on the concept of Shannon entropy that takes into account both the number of peaks and the evenness of their area for each VHC combination (Figure S2B). This analysis did not reveal any significant difference of CDR3 length diversity between IgM, IgD, and IgT in the spleen of naïve fish, reflecting that these expressed repertoires were all diverse.
We then further compared the expressed IgM, IgD, and IgT repertoires to each other using ISEApeaks. For this, we computed the perturbation scores of each individual fish for a given isotype (e.g. IgM), and compared it to the average score of the three fish for a distinct isotype (e.g. IgD or IgT) (Figure S2C). This analysis revealed that VHCμ and VHCδ profiles were similar to each other for most VH, while VHCτ rearrangements were clearly distinct from both VHCμ and VHCδ profiles. These findings likely reflect the fact that μ and δ IGH transcripts can be produced through alternative splicing of the same primary transcripts, while IGHτ rearrangements utilize distinct D and J genes. This is also consistent with the fact that trout naïve B cells either co-express IgM and IgD, as known for mammalian B cells, or only IgT as previously reported [18], [21].
To analyze the B cell response to a viral infection, four clonal fish (genetically identical to the healthy fish described above) were challenged two times with VHSV, on days 1 and 21. After three more weeks (day 42), all fish contained neutralizing VHSV antibodies in serum (see below). Also, FACS analysis indicated that the spleen B cells of these clonal fish contained about 75–90% IgM+IgT− and 25–10% IgM−IgT+ B cells as previously reported. This ratio was not significantly modified after infection (Figure S3A). The B cell response was characterized at this time point after challenge (day 42) by spectratyping analysis of the expressed IgM, IgD, and IgT repertoires.
To characterize the molecular diversity of this anti-viral B cell response at the CDR3 sequence level, we performed deep sequencing analyses of a number of VHC combinations involved in major (VH4 and VH5.1 for IgM; VH4 for IgT), moderate (VH1.1 for IgM; VH5.1 for IgT) or weak (VH5.4 for IgM and IgT) responses at the mRNA level. IgD was not analyzed further because of its minor contribution to the response. The sequence reads obtained through 454 pyrosequencing were analyzed by IMGT/HighV-QUEST.
Sequences encoding different V-D-J rearrangements were assembled into junction sequence types (JST) for statistical analysis (Figure S4). We hereafter refer to JST in our analysis, defined as a CDR3 amino acid sequence associated to a given (V, J) pair. As a preliminary study, we estimated the error rate to be around f = 3×10−3 per base pair using a known VH sequence (see methods), which was close to the estimations previously reported varying between 0.4 and 1% [23], [31]. Using this average error rate per site, we corrected the number of reads for each junction nucleotide sequence (JNS) by adding the sequences lost and subtracting those gained due to erroneous mutations introduced by the PCR and sequencing procedure (Figure S5). These corrected datasets were then translated and aggregated into corrected JST datasets, which were analyzed in parallel to the observed, unprocessed JST datasets.
For each VHC sequencing analysis, we first classified the observed JST according to their CDR3 size to produce “virtual spectratypes”. These “virtual spectratypes” were consistent with the PCR-based spectratypes (Figure S6), indicating that the deep sequencing analysis of the PCR products did not add biases in IgM and IgT CDR3 length distributions, as previously noted for human IgM and IgA in [26].
An important question about large clonal IgM and IgT expansions found in the spleen concerned the capacity of the corresponding B cells to produce secreted Abs. To clarify this issue, we selectively amplified either membrane-bound or secreted transcripts using reverse primers located in the transmembrane exon or in the 3′ end sequence specific for the secreted isoform. First, we used primers specific for JST found expanded in our 454 sequencing after infection (Figure 6). The secreted isoform was always expressed at a greater magnitude than the membrane isoform for μ chains of IgM and for τ chains of IgT. Secreted μ transcript for the VH5.1J5 public response was also amplified from all infected fish tested, while the VHx4J1 private response was only detected in some individuals. No amplification was observed with primers targeting JST that were not expanded in the sequence datasets, or from control animals.
We next quantified the whole population of secreted versus membrane IGH μ (or τ transcripts – potentially including all non-responsive clones and sterile transcripts – using forward primers specific of the 3′ exon for each isotype, and the reverse primers described above. Our QPCR results showed that even at the B cell whole-population level, the fold change after VHSV infection is higher for transcripts encoding secreted IGH μ and τ than for those encoding membrane-bound IG (2.0+−0.66, and 3.7+−0.94 times more, respectively) (Figure S12). With regards to the different τ subtypes, τ1 and τ3 were involved in the response while τ2 was not expressed. No induction was seen for IGH δ transcripts (Figure S12). Overall, these results confirm that the expansions of VHCμ or VHCτ observed in the spleen following infection mainly represent transcripts encoding secreted IGH chain. Hence, the spleen contains virus specific Ab producing cells at this stage of the response.
Finally, we estimated the serum concentration of IgM and IgT in control and infected animals (Figure 6B). After infection, fish serum contained in average 7853 µg IgM per ml and 85 µg IgT per ml, representing an increase of 15 and 117 times, respectively, in comparison with healthy fish. These results confirm that both IgM and IgT participate to the Ab response against the virus. However, the serum IgM concentration was ∼100 fold higher than that of IgT, thus, IgM overwhelmingly represents the main responding isotype.
Neutralising antibodies are of critical importance for the fish defense against VHSV infection, and their presence is highly correlated to the protection after vaccination. To address the question of the respective importance of serum IgM and IgT in virus neutralization, we performed inhibition assays of VHSV neutralization using either anti-IgM or anti-IgT Abs. For this purpose, we used the anti-IgM monoclonal Ab 1.14 [35], and a polyclonal anti-IgT antiserum [21]. As shown in Table 3, we found that the neutralizing activity of the serum from vaccinated fish was mainly due to IgM, while IgT had only a minor contribution if any. The inhibition of neutralization by the anti-IgM Ab was dose dependent, and reached a maximum at the dilution of the serum in the assay (data not shown).
In this work, we characterized the clonal structure of the rainbow trout IgM, IgD and IgT response against a systemic viral infection. IgM represented the main response in the spleen, with high clonal complexity and all IGHV subgroups involved. A striking result of our study is the implication of B cells expressing the mucosal isotype IgT in the spleen response to the virus, with distinctive clonal expansions identified by amplified junctions in IGH τ transcripts. In contrast, the IgD expressed repertoire was modified only to a modest extent, suggesting that either activated B cells loose IgD expression as in mammals [36] or that IgD+ responding B cells leave the spleen and re-locate in other territories. Based on combined CDR3 length spectratyping and deep sequencing of IGH transcripts at one time point during the response to secondary virus infection, our approach does not provide a kinetic description, nor directly addresses the specificity of the antibodies corresponding to CDR3 expansions identified here. However, the correlation of such expansions with viral infection as well as their frequency and distribution among infected fish provide a first comprehensive view of a spleen antibody response to a virus in fish, and more generally in a vertebrate.
A global understanding of the mechanisms that determine the structure of the B cell expressed available repertoire requires a comprehensive survey of Ab diversity during the maturation of the immune system and along immune responses. A widely used method developed for large-scale IG and TcR V domain repertoire analysis consists in CDR3 length profiling, leading to 6–15 peak spectratypes for each V/C or V/J combination [27]. This approach of individual repertoire variations can be coupled with statistical analysis [37], and has been frequently used to characterize T and B cell responses [33], [37]–[39]. However, it does not provide a full description of the junctions. An important breakthrough towards a complete description of immune repertoires was recently made with the usage of 454 GS FLX high-throughput pyrosequencing for the direct analysis of whole B cell diversity of individual zebrafish [23]. A few studies based on the same approach have been published on human B and T cells [24], [40]–[42]. While it is well-known that pyrosequencing induces artefacts such as mutations in poly C/G stretches, and artificial duplications of sequences [43], we show here that 454 pyrosequencing and CDR3 length spectratyping lead to fully consistent and comparable results from the same samples. These results are in agreement with the observations of Ademokun et al. [26], who followed a similar approach to analyze the modifications of the human IgM and IgA expressed repertoires after vaccination against pneumococcal pneumonia in young and elderly. Importantly, our analysis assessing for each JST the number of sequences lost or gained due to technical error confirmed that the structure of our datasets was in the same line as in previous reports. Also, our sequencing error rate allowed the identification of V and J genes by the IMGT/VQUEST annotation pipeline. To our knowledge, this study is the first comprehensive description of a B cell response to a viral pathogen.
Combining CDR3 spectratyping and pyrosequencing, we show here that the fish Ab response primarily involves IgM. A few very large IGHμ clones were preeminent in the response to the virus, likely targeting a small number of dominant epitopes. These large sets of amplified JST are difficult to explain by sampling artifacts of the deep sequencing. Indeed, our model of repertoire description by sequencing indicates that at least the largest JST represent true clonal expansions. These JST expansions were either private – ie found in only one fish- or public – ie found in (almost) all individuals. An IgM response was identified, involving VH5.1 rearranged to J5, which represented a typical public response consisting in a set of highly similar junctions present in all fish. The observed public responses were strongly expanded and could be inferred from the spectratyping, corresponding to high peaks found in all infected fish. These results suggest the presence of a common pool of pre-existing spleen B cells among which the public IgM response to VHSV could be recruited, a notion that is supported by findings in other models. Thus, in the large-scale sequencing of the naïve zebrafish IgM VH domain repertoire, convergences have been identified in the normal repertoire, with identical VHC sequences retrieved in different individuals at significant frequencies [23].
Strikingly, all VHCμ profiles were significantly altered after infection, indicating that the IgM response to the virus comprised diverse components likely targeting many epitopes. Since VHSV is a natural pathogen of rainbow trout, a possible explanation of this observation is that natural selection has favored multicomponent Ab responses to this highly variable negative strand RNA virus. Another non-exclusive explanation of this observation would be a non-specific polyclonal B cell stimulation by the viral infection. Such polyclonal stimulation could be mediated by different mechanisms including direct activation by viral components through pathogen recognition receptors, bystander activation by activated T cells, or viral superantigens. In the mouse, a number of mitogens from different origins [44]–[46] are very effective in inducing B cell polyclonal activation [47], [48]. Moreover, it has been known for a long time that salmonids B cells stimulated by mitogens not only proliferate, but also increase their global production of antibodies [49], [50]. Interestingly, the nucleocapsid of the Rabies virus, a rhabdovirus related to VHSV, acts as a B cell superantigen in human and induces polyclonal B cell activation [51]. Whether such polyclonal B cell stimulation has deleterious or beneficial impact on the host is still a matter of debate [47].
For a comparable number of sequences analyzed, no public response was found for IgT, not even for the VH4-Cτ combinations showing highly skewed spectratypes in all fish. In contrast to IgM, IgT CDR3 length spectratypes were found significantly skewed after viral infection for only two VH. These results indicate that the number of pre-existing clones directed against the virus was higher among spleen IgM+ B cells compared to IgT+ B, which is in accordance with the respective frequency of IgM+ and IgT+ B cells in the spleen of control fish. Perhaps, if IgT has been evolutionarily selected mainly for mucosal defense, “pre-existing” specificities in the available repertoire may be focused on mucosal pathogens. However, the biochemical properties of the V-D-J junctions analysis in silico in the whole set of IgM and IgT sequences from healthy and infected fish identified no clear difference of isoelectric point, charge or gravy between the two isotypes (Figure S13).
Since IgT mainly protect gut mucosa, the question of the role of IgT expansions in the spleen of virus-infected animals is intriguing. The dominant expression of the secreted form of the IgHτ transcripts in infected animals and the increase of serum IgT concentration upon infection indicate that such spleen IgT+ cells likely produce Abs. A first possibility would be that Ab secreted into the blood by such spleen IgT+ clonal expansions would indeed contribute to the antiviral defense of the whole organisms including mucosa. Alternatively, virus specific IgT+ B cells may get activated and proliferate before migrating to the mucosal territories, and may be found in the spleen as transiting cells. These observations are reminiscent of the complex roles of IG classes and subclasses in humans. While IgA is the most abundant Ab class in the mucous and other secretions from epithelia, it is also present – mostly the subclass IgA1- in the serum where it can bind to antigens and mediate Ab-dependent cellular cytotoxicity (ADCC), respiratory burst, degranulation and phagocytosis through binding to the FcαRI receptor [52]. Reciprocally, similar to IgA, human IgM pentamers can be secreted through the gut epithelium into the gut lumen, whereas IgG can be found in colostrum and milk. Also, IgA+ B cells recirculate from the gut lymphoid structures, joining the bloodstream and homing back to the gut mucosa [53]. Thus, the classical division of IG classes and subclasses into systemic or mucosal Abs cannot be viewed in absolute terms in mammals, and this apparently applies to fish too.
In healthy fish, we found B cell expansions in the spleen of control trout as reported in zebrafish [23]. While it has been reported that rainbow trout spleen contains mostly resting B cells, and few Ab producing cells [54], our results indicate that spleen comprises indeed a small proportion of such clones, expressing either IgM or IgT. The distributions of amplified JST indicate that clones responding to the virus are generally larger than the plasma cell clones present in the spleen of control fish, which is also in good accordance with the spectratyping profiles.
A tentative estimation of the relative proportion of large responding IgM+ and IgT+ clones in the B cell population was made from sequence data produced in healthy and infected fish. Considering that an Ab-producing cell produces 100 to 1000 more mRNA than a resting B cell [55], [56], the proportion of expanded cells in healthy fish was estimated from the proportion of sequences present 5 times or more in the 454 datasets, for a given number of available reads (Table 4, Figure S14). In pyrosequencing, the emulsion-based PCR from individual molecules makes possible that a sequence may appear a few times instead of only one, making clonality comparisons comparative rather than absolute [40]. We therefore considered that the observation of 1–5 occurrences – instead of 0–1 – did not reflect the precise frequency of a JST, but could rather represent variable amplification from a unique sequence, due to a side effect of the technique. We therefore considered that this represented resting cells, and that higher numbers of occurrences denoted Ab-producing cells (Figure S14). The estimated frequency of Ab-producing cells in the spleen of control animals varied with the rearrangement considered, but for a given V, was lower for IgT+ cells. Thus, our calculation predicted ≈100 times more IgM-producing cells than and IgT- producing cells in control fish for VH4, and around 10 times more for VH5.1 and VH5.4 (Table 3). Calculations from corrected datasets led to similar results (Figure S14). After infection, the predicted frequency of Ab-producing cells increased generally 2 to 9 times, with the exception of VH4Cτ for which the final frequency reached the same order as for the other combinations but from a very low initial value (Table 3). Overall, these estimations suggest that the viral infection significantly increases the total frequency of large B cell clones, and are in accordance with the idea that IgM represents the dominant class produced in the spleen in control as well as in infected animals [27]. Since all expressed VH subgroups were involved in the response to the virus, this represents a considerable shift of the activated B cell subset. It has been reported that in naïve fish IgM+ and IgT+ B cells represent respectively 75–90 and 10–25% of the whole spleen B cell population, and that IgT concentration in plasma of naïve fish is approximately 1000 fold lower than that of IgM [21]. This would suggest an even greater difference of frequency of IgM+ versus IgT+ Ab-producing cells at the scale of the whole organism. In the clonal fish used in this study, we found the same proportion of IgM+ and IgT+ B cells, and it was not significantly modified by the infection. Considering that plasmablasts and plasma cells express membrane Ig at very low level [21], this observation is well consistent with an important contribution of these subsets to the response seen at the transcript and serum Ab levels. At any rate, inhibitions of viral neutralization by fish serum using anti IgM or anti IgT antibodies clearly showed that IgM is of foremost importance for the antibody-mediated protection against the viral infection.
Previous studies of trout Ab response to TNP keyhole limpet hemocyanin (TNP-KLH) using affinity-based immune-partitioning assays have shown that a first low affinity followed by an intermediate/high affinity population that persists longer while high affinity Abs appears much later after week 15 post-immunization and are expressed at higher concentrations. These results indicate that an increase of Ab affinity indeed occurs in fish, being slow but very significant [57], [58]. While the existence of somatic hypermutation in fish is now well established [59], its importance for affinity maturation is still not fully appreciated. In fact, it is generally accepted that in the absence of refined modes of selection of late-developing clones, B cells in which somatic mutants lead to a higher Ab affinity are not quickly amplified within the population [4]. Jiang et al. have analyzed the secondary IGH repertoire of the developing zebrafish using deep sequencing and found that average numbers of mutation in highly abundant IGH sequences increases with age [60]. Our results are difficult to compare directly with the simple antigen immunization since we analyzed antibody repertoires during the response to secondary infection. However, the large IGH expansions we observe in the spleen likely represents maturating plasma cells producing intermediate/high affinity antibodies, before migration into the anterior kidney. Somatic hypermutation analysis from our sequence data was hampered by the lack of knowledge about the VH genomic repertoire of the clonal fish we used and by the intermediate sequencing depth of our datasets. With well-annotated igh loci in the coming rainbow trout genome, further studies combining affinity assessment and deep sequencing of IGH transcripts will make it possible to correlate affinity maturation and somatic hypermutation.
Regarding the IgD, complete IGHD transcripts were expressed at very low level in the trout spleen, even after virus infection. Importantly, the few perturbations observed in IgD profiles after infection were not statistically significant. Since amplified peaks were absent in VHCδ profiles of infected fish even when they were present in the corresponding VHCμ profiles, we conclude that the great majority of responding clones found in spleen do not express IgD. Hence, either the majority of responsive IgD+ cells migrate out of the spleen, or pathogen-specific B cell activated by their specific antigen differentiate into IgD− plasma blasts/cells. While it has been recently reported that B cells expressing IgD are found in significant numbers in pronephros and gills [61], we did not find a higher expression of IgD in gut, gill or pronephros after virus infection and we would therefore favor the second hypothesis. In fact, resting B cells co-express IgM and IgD in human, mouse and catfish, correlating with the conserved respective locations of the IGHM and IGHD genes in the IGH locus. While human IgD+ B cells have been described in the respiratory tract of human, and an IgM−IgD+ B subset is found in channel catfish [20], our work confirms that such cells are apparently not present in the spleen of rainbow trout [21]. In fact, the only hint of an IgD clonal expansion we found was the presence of an extra peak in VH4Cδ profiles in all tested animals, containing the amplified CDR3 ARGTEYYYFDY. This recurrent expansion present even in control animals could be due either to the recognition of an environmental pathogen/antigen, or may constitute an invariant receptor following a non-classical selection pathway. In fact, IgD expression both in control and virus infected fish appears to be almost subsidiary. The discovery of pathogens/antigens specifically eliciting IgD response would greatly improve the knowledge of its biological significance. While the structure of fish IgD is quite divergent from mammalian IgD, the absence of IgD clonal expansion triggered by VHSV infection reminds the down regulation of this isotype on activated B cells in mouse and human.
While differences in microenvironments in which B cell responses develop in fish and mammals apparently do not enforce dramatic changes in the structure of B cell response, our observations are quite parallel to what was observed in human and mice. As a rhabdovirus, the VHSV is a typical highly cytopathic virus with a surface densely packed with multiple copies of a unique glycoprotein (G), as for VSV or rabies virus. Such viruses elicit high affinity, protective Ab responses with V regions frequently encoded in the germline [62]. In human, IgG antibodies produced against rabies virus express diverse VH genes and targets both G and the ribonucleoprotein complex, while the pre-immune IgM binding the virus express mainly VHIIIb [63], [64]. Interestingly, this response not only comprises IgG but also the mucosal isotype IgA [63], [64].
This work demonstrates the appearance of large clones of IGH junctions after viral infection, that likely represent the generation of typical polyclonal IgM and IgT B cells responses to a virus infection in fish. In the spleen, IGHμ shows a strikingly diversified response involving all VH subgroups with some very large public and private clones. IGHτ encoding mucosa-specialized IgT also show clonal expansions expressing secreted isoform, but at a lower scale. IgD does not seem to be affected by the response. Further characterization of B cell response to simple antigens will help understanding the respective importance of specific clonal expansions and bystander polyclonal activation. Ig transcripts from sorted B cells that bind a fluorescent antigen could be subjected to deep sequencing or single cell (RT)PCR to link directly junctions to the fine specificity of the antibodies.
While new variations of fish IG have been recently described like catfish V-less IgD [20] and IgT/IgM chimeric molecules in cyprinids [65], our results emphasizes the capacity of repertoire studies to uncover functional features of antibodies. Additionally, fast progress in sequencing technology and IG annotation open the way to individual and comprehensive descriptions of Ab repertoires, and monitoring using high-throughput sequencing will likely become paramount for health management in aquaculture as well as human medicine.
Rainbow trout homozygous clone B57 [66] were raised in the fish facilities of Institut National de la Recherche Agronomique (INRA, Jouy en Josas, France). Two years old adult fish were placed in individual aquaria kept at 16°C. Immunization and virus challenge were performed using the attenuated 25–111 variant of strain 07–71 of VHSV [7] through intramuscular injection. A first sub-lethal dose of 105 PFU/fish was applied to each fish. This infection usually leads to a good protection against a subsequent lethal infection. Three weeks later, fish received a second injection of the 25–111 virus (5×107 PFU/fish) and samples were collected 3 weeks later. Control fish were left untreated. Trout were sacrificed by overexposure to 2-phenoxyethanol diluted 1/1000. Blood was extracted and let to clot at 4°C overnight for serum extraction. Tissues were removed, frozen in liquid nitrogen and stored at −80°C to use in RNA preparation. Serum extraction was performed by centrifugation at 200× g for 10 min, supernatants were collected and centrifuged at 1000× g for 20 min. Serum was frozen at −20°C to use in titration assays and Ig concentration measurement. For flow cytometry analysis of spleen leukocytes, fish were sacrificed by overexposure to 2-phenoxyethanol diluted 1∶1000. The spleen was removed aseptically and cells from the spleen of a single fish were deposited on a Ficoll solution (Lymphocyte separation medium [d = 1.077]; Eurobio, Les Ullis, France) and centrifuged 10 min at 900 g. The leukocyte fraction was collected at the Ficoll-medium interface.
All animals were handled in strict accordance with good animal practice as defined by the European Union guidelines for the handling of laboratory animals (http://ec.europa.eu/environment/chemicals/lab_animals/home_en.htm) and by the Regional Paris South Ethics committee, and all animal work was approved by the Direction of the Veterinary Services of Versailles (authorization number 78–28).
Total RNA was individually prepared from spleen by disruption in TRIzol reagent (Life Technologies, Cergy-Pontoise, France) using 1/1.2 mm ceramic beads (Mineralex, France). Disruption protocol was 2 pulses of 15 sec at 6000 rpm in a Precellys tissue homogenizer (Bertin Technologies, France). The whole spleen was used for mRNA preparation. We thus ensured to analyze all B/plasmablast/plasma cells subsets and to avoid any biases due to regional concentration of responding B cells or to differential density of B cell subsets, which may affect the outcome of density gradient. Total RNA was purified and DNase treated using QIAgen RNA extraction kit. RNA (2 µg) was reverse transcribed into cDNA using Superscript II Reverse Transcriptase (Invitrogen Life Technologies) with 2.5 µM oligodT25 primer in a final volume reaction of 20 µl.
The spectratyping of TcRB CDR3 length (Immunoscope analysis) first developed for mouse or human [27] was previously adapted for rainbow trout [7]. The diversity of IG transcripts was studied following a similar approach. A first amplification (PCR1) using a forward primer specific for a subgroup or a set of IGHV genes in combination with a reverse primer Cμ, Cδ or Cτ specific for IGHM, IGHD or IGHD genes was performed as follows: 1 µl cDNA was used as template for PCR1 using 0.4 mM of each dNTP, 0.4 µM of each primer (forward: VHfamily specific, reverse: Cisotype specific), and 0.025 u µl−1 of GoTaq DNA polymerase (Promega) in 1× reaction buffer with 2 mM of MgCl2 (95°C for 5 min; 40 cycles of 95°C 45 s, 60°C 45 s, 70°C 45 s; 70°C 10 min) (see Figure S1 for primer sequences and correspondence with the IMGT nomenclature of gene names based on available sequences, IMGT Repertoire, http://www.imgt.org). Specific IGHV and C primers amplify VHC sequences with a given IGHV but with different IGHJ content and diverse CDR3 lengths. In a second step, VHC PCR products were subjected to run-off reactions (PCR2) using 5′ 6-FAM- fluorescent C internal, isotype specific, reverse primers. Two µl of PCR1 product were used as template using 0.4 mM of each dNTP, 10 pmoles of the fluorescent reverse primer, and 0.025 u µl−1 of GoTaq DNA polymerase (Promega) in 1× reaction buffer with 2 mM of MgCl2 (95°C for 5 min; 5 cycles of 95°C 1 min, 60°C 1 min, 70°C 2 min; 70°C 10 min). Two µl of run-off product were mixed with 8 µl deionized formamide (Applied Biosystems) and 0.5 µl of the internal standard (GeneScan 500XL ROX, size standard, Applied Biosystems). Mix was denatured at 95°C for 5 min and placed on ice before analysis in an ABI 3730HT sequencer (Applied BioSystems) at GeT-PlaGe core facility, Toulouse, France. CDR3 length distributions were analyzed using GeneMapper (Applied BioSystems) and ISEApeaks software [30], [37] to extract and analyze spectratype data for each VHC combinations. Each spectratype or profile is composed of several peaks (typically 4 to 10 for VHCμ and VHCδ, and 5 to 13 for VHCτ) separated according to their corresponding length of run-off products, spaced by 3 nucleotides as expected for in-frame transcripts. Each peak corresponds to a given CDR3 length. For each profile associated with a VHC combination, the area under each peak was calculated and stored in a peak database. These values were then computed to quantify the differences between spectratypes. As the intensity of CDR3 peaks is not comparable between different profiles/amplifications, we considered the percentage of use of each CDR3 length (i.e. the “relative area”), obtained by dividing the area of CDR3 peaks by the total area of all peaks within the profile. In a given context, the “perturbation” for a given VHC profile was calculated as follows: Where: pi,an and pi,ref are the relative areas of the peak #i from the analyzed and reference profiles respectively; n is the number of peaks detected in the reference profile. The reference profile is the average profile of individuals from a homogeneous group chosen as reference group. Perturbation scores are computed using the formula above for all VHC profiles and across all analyzed samples including the reference group. VHC perturbations range from 0 (identical profiles), to 100 (non-overlap profiles). To assess a significant perturbation between infected and control groups, we performed statistical tests at level α = 0.01 (type I error) on the perturbation index. Statistical significance of the difference in perturbation scores for two different groups was determined using empirical Bayes test from the limma packages (R/Bioconductor). We chose this test because it outperforms the classical Student t-test or Mann-Whitney-Wilcoxon non parametric test for small samples [67]. As recommended in [68] for a multiple testing problem (several tests done simultaneously), we used Benjamini-Hochberg procedure [69] to control the False Discovery Rate (FDR) which is the expected proportion of false positives among positive tests. Agglomerative Hierarchical clustering (Euclidean distance, K = 3) was performed using the software TM4 MultiExperiment Viewer (TMEV) with these data to group the different repertoires into three classes. The complete linkage method was chosen to aggregate similar clusters.
VHC or VHJ PCR products were cloned into pCR2.1 vector (TOPO TA Cloning Kit, Invitrogen, San Diego, CA). The PCR product was concentrated and purified using the PCR purification and gel extraction kits (QIAgen) and 1.5 µl of the purified product was used in the cloning process, following manufacturer's instructions. Clones were picked at random, plasmidic DNA were purified (plasmid miniprep spin kit, Nucleospin, Macherey-Nagel, Durin, Germany) and sequenced using M13 universal primers.
454 pyrosequencing applied on selected VHC combinations involved in major (VH4 and VH5.1 for IgM; VH4 for IgT), moderate (VH1.1 for IgM; VH5.1 for IgT) or weak (VH5.4 for IgM and IgT). PCR products amplified from the same individuals as for the spectratype analysis were analyzed, providing 1500 to 3000 useful sequences per PCR product. Sequencing libraries were produced from relevant PCR products mixed in equal amounts after picogreen quantification. To identify the origin of sequences, tagged PCR products were created using a nucleotide based barcode system (or “molecular identifier”, MID). IGH V, J or C MID-primers were designed by adding a MID sequence of 10 bp upstream to the primer sequence to analyse selected VHC or VHJ PCR products. The forward and the reverse primers of a given PCR product contained the same MID. Ten different MID sequences were used, allowing the co-processing of the same VHC or VHJ product from different individuals or samples. Primers used were: forward VH1.1, VH4, VH5.1, VH5.4, VH6, and reverse Cμ1, Cτ1, JH6. PCR conditions were as follows: 1 µl cDNA was used as template, using 0.4 mM each dNTP, 0.4 µM each MID-primer (forward: V-specific, reverse: J-specific or C-specific), and 0.08 u µl−1 of BIOTAQ DNA polymerase (Bioline) in 1× reaction buffer with 3 mM MgCl2 (95°C for 5 min; 40 cycles of 95°C 45 s, 60°C 45 s, 70°C 45 s; 70°C 10 min). MID-tagged PCR products were cleaned by gel extraction (QIAgen gel extraction kit), quantified, and aliquots of 100 ng of each PCR product were pooled for 454 library preparation. Library preparation and 454-pyrosequencing runs were done at GeT-PlaGe core facility, Toulouse Midi-Pyrénées, France, using GS FLX Titanium General DNA Library Preparation kit from Roche. Briefly, double stranded DNA was end-polished and ligated to sequencing adaptors. Library immobilization, fill-in reaction and single-stranded template DNA (sstDNA) library isolation were then performed following manufacturer's instructions. sstDNA library was then amplified and processed for sequencing using the GS FLX Titanium emulsion PCR protocol. Emulsion PCR reactions were prepared with a ratio of 0.12 molecule per DNA capture bead. 454 pyrosequencing libraries were constructed from around 20 pooled MID-labeled PCR products and sequenced using 1/8 of the chip which corresponds to a total of 70000 to 100000 reads.
A high proportion of unproductive, i.e. mainly out-of-frame, junctions (overall 40% in naïve animals) was observed. In fact, a defect in elimination of transcripts with STOP codons is well known in fish and has been repeatedly observed for TcRB transcripts of which more than 30% are unproductive due to out-of-frame V-D-J rearrangements. We verified that this high proportion of out-of-frame junction was observed among IG sequences as well, which were produced by Sanger sequencing of cloned VHC PCR products (unpublished data). Hence, this observation did not indicate a dramatic error rate of the pyrosequencing and we focused our analysis on productive rearrangements.
A script (described in Figure S4) was developed for sequence quality control, re-classification using barcodes, and preparation of sequences for annotation. Sequence annotation was performed by the IMGT/V-QUEST and IMGT/JunctionAnalysis programs of IMGT/HighV-QUEST [46], the IMGT web portal for high-throughput and deep IG and TR sequencing, ensuring a rigorous identification of V, D and J sequences, an exhaustive and detailed analysis of the junctions as well as filtering of unproductive junctions. For the time being, rainbow trout IGH loci have not been fully sequenced and annotated. Hence, our definition of V and J refers to subgroups rather than to defined genes/alleles. VH groups and J sequences mentioned in this study correspond to the current content of IMGT gene tables. CDR3 sequences are identified and extracted following the IMGT numbering. Thus, 1500 to 3000 useful sequences was generally obtained per PCR product. While this order of magnitude was far too small to produce a full (saturated) description of the junction repertoire associated to these VHC (or even VHJ) combinations, it provided a synthetic and cost-effective overview of the most prominent frequency changes induced by acute responses. Depending on sequencing run 45–61% of reads were checked for length, quality, presence of MID and V/C sequences, and sent to IMGT for annotation; 26–42% were finally considered. Based on the whole data set used in the project, on average 33% of reads were used in the final analysis.
Sequences were assigned to each analyzed sample from their primer-MID content. Briefly, primers and MID used in a run were identified in each 454 sequence. Sequences with a structure inconsistent with the primer-MID combinations used in the run were excluded. In the next step, sequences that did not contain complete junction (CDR3) region with flanking sequences long enough to identify the IGHV and IGHJ involved in the rearrangement, were discarded. The remaining sequences were then classified into sets corresponding to a unique combinations of [MID; V; C or J; sequencing direction] where the MID specified the original sample (tissue or individual fish).
To assess the error rate of 454 pyrosequencing, we chose a 140-bp region of the IGHV4 with no variation due to polymorphism among the reads, and we counted the indels and point substitutions in a dataset representing 1797091 nucleotides.
Sequences that passed these quality control criteria were then annotated using IMGT/HighV-QUEST (http://www.imgt.org/HighV-QUEST), which provided a full annotation including identification of the IGHV, IGHJ, IGHC, CDR3 length, location, and protein sequence. This annotation was then used for further specific analysis of the VH domain repertoire amplified in each sample. For each VHC or VHJ combination, the number of sequences having the same CDR3 amino acid sequence (i.e. its “number of occurrence”), and the number of distinct CDR3 found once, twice,… were determined. The physicochemical properties of the peptide encoded by the CDR3 were determined using ProtParam [70].
For real time PCR, 3 µl of cDNA (1∶3 diluted) was used as a template for amplification using gene specific primers (Figure S1). PCR amplification was performed in a Mastercycler ep realplex (Eppendorf), using ready prepared 2× master mix (Power SYBR Green PCR master mix, Applied Biosystems) with a final PCR volume of 25 µl, in white 96-well plates (Eppendorf). PCR conditions were 95°C for 10 min followed by 95°C for 30 sec, 60°C for 30 sec and 72°C for 30 sec. The fluorescence signal output was measured and recorded at 80°C during each cycle for all wells for 40 cycles. A negative control (no template) reaction was also performed for each primer pair. A sample from the serial dilution was run on a 2% agarose gel and stained with ethidium bromide and viewed under UV light to confirm a band of the correct size was amplified.
A melting curve for each PCR was determined by reading fluorescence every degree between 60°C and 95°C to ensure only a single product had been amplified. Trout elongation factor-1α (EF-1α) and β-actin were used for normalization of expression. The relative expression level of the genes was determined using the Pfaffl method [71]. Efficiency of the amplification was determined for each primer pair using serial 10 fold dilutions of pooled cDNA, performed in the same plate as the experimental samples. The efficiency was calculated as E = 10 (−1/s), where s is the slope generated from the serial dilutions, when Log dilution is plotted against ΔCT (threshold cycle number).
IgM and IgT concentration in fish serum was estimated as previously reported in [21]. VHSV neutralization assays were performed as described in [7]. For the inhibition of neutralization, an excess of anti-IgM mab (1.14) or anti-igT polyclonal antibody [21] was pre-incubated with the serum 3 hours at room temperature, and this mix was then used for virus neutralization. While only partial inhibition was observed at low dilutions of serum, inhibition reached a reproducible plateau for higher dilutions. Typically, maximum inhibition was observed when using 10 µl of serum diluted 100 times or more, with 5 µg of anti IgM and 1 µg of anti IgT antibodies. A mouse IgG1 was used as an isotype matched control, and did not inhibit virus neutralization.
|
10.1371/journal.ppat.1001157 | In Vitro and In Vivo Studies Identify Important Features of Dengue Virus pr-E Protein Interactions | Flaviviruses bud into the endoplasmic reticulum and are transported through the secretory pathway, where the mildly acidic environment triggers particle rearrangement and allows furin processing of the prM protein to pr and M. The peripheral pr peptide remains bound to virus at low pH and inhibits virus-membrane interaction. Upon exocytosis, the release of pr at neutral pH completes virus maturation to an infectious particle. Together this evidence suggests that pr may shield the flavivirus fusion protein E from the low pH environment of the exocytic pathway. Here we developed an in vitro system to reconstitute the interaction of dengue virus (DENV) pr with soluble truncated E proteins. At low pH recombinant pr bound to both monomeric and dimeric forms of E and blocked their membrane insertion. Exogenous pr interacted with mature infectious DENV and specifically inhibited virus fusion and infection. Alanine substitution of E H244, a highly conserved histidine residue in the pr-E interface, blocked pr-E interaction and reduced release of DENV virus-like particles. Folding, membrane insertion and trimerization of the H244A mutant E protein were preserved, and particle release could be partially rescued by neutralization of the low pH of the secretory pathway. Thus, pr acts to silence flavivirus fusion activity during virus secretion, and this function can be separated from the chaperone activity of prM. The sequence conservation of key residues involved in the flavivirus pr-E interaction suggests that this protein-protein interface may be a useful target for broad-spectrum inhibitors.
| Enveloped viruses infect cells by fusing their membrane with that of the host cell. Dengue virus (DENV) is an important human pathogen whose membrane fusion is triggered by low pH during virus entry into the cell. However, newly synthesized DENV must also transit through a low pH environment during virus exit. DENV is believed to escape premature fusion in the exit pathway via the small viral protein pr, which is processed and associates with virus after biosynthesis, and is released from the virus particle in the neutral pH extracellular environment. Here we have reconstituted the interaction of pr with the DENV fusion protein E using soluble protein components. The interaction has a low pH optimum and inhibits membrane insertion of the fusion protein. The recombinant pr peptide can “add back” to fully infectious mature DENV and block virus fusion and infection. We found that mutation of a critical conserved histidine on the fusion protein inhibits the interaction of E and pr, and makes the virus susceptible to low pH-induced inactivation during exit. This work characterizes the mechanism of pr protection, and suggests that the conserved multifunctional pr-E interaction may be an important target for anti-viral strategies.
| The emergence and resurgence of human viral pathogens can be traced to a complex variety of causes including increased urbanization, human contact with animal reservoirs, a decrease in effective public health systems, and the spread of insect vectors that disseminate some viral infections [1], [2], [3]. Flaviviruses are a genus in the Flaviviridae family and include important emerging and resurgent human pathogens such as dengue virus (DENV), West Nile virus (WNV), tick-borne encephalitis virus (TBEV) and yellow fever virus [2], [4]. Flaviviruses are transmitted by insects such as mosquitoes and ticks, and can cause severe human diseases characterized by encephalitis, meningitis, and hemorrhages [2], [3]. More than one third of the world's population lives in dengue fever endemic areas, and there are an estimated 50–100 million cases of dengue infection and 500,000 cases of the more lethal complication, dengue hemorrhagic fever, per year [5], [6], [7], [8]. There are currently no antiviral therapies for flaviviruses. DENV vaccine development is underway but is problematic due to the presence of four DENV serotypes and the potential for antibody-dependent enhancement of infection [2], [6], [9], [10]. Antiviral therapies could thus be an important alternative for DENV and for viruses such as WNV in which the cost and potential side effects of vaccination must be weighed against the relatively low number of human cases [2].
Flaviviruses are small, highly organized enveloped viruses with a spherical shape [4], [11]. They contain a positive-sense RNA genome packaged by the viral capsid protein. The nucleocapsid is surrounded by a lipid bilayer containing the viral membrane protein E. Flaviviruses infect cells by receptor engagement at the plasma membrane, endocytic uptake, and a membrane fusion reaction triggered by the low pH of the endosome compartment [12], [13]. The viral E protein binds the receptor and drives the fusion of the viral and endosome membranes to initiate virus infection. The pre-fusion structure of the E protein ectodomain (here referred to as E′) shows that E contains three domains composed primarily of β-sheets: a central domain I (DI) connecting on one side to the elongated domain II (DII) with the hydrophobic fusion loop at its tip, and connecting via a flexible linker on the other side to the immunoglobulin-like domain III (DIII) [14], [15], [16], [17], [18], [19] (Figs. 1A, S1). Although these regions are not present in the truncated E′ ectodomain, DIII connects to a stem domain and C-terminal membrane anchor (TM). The E protein in mature infectious flavivirus is organized in homodimers that lie tangential to the virus membrane [20]. Within each dimer the E proteins interact in a head to tail fashion, with the fusion loop of each E protein hidden in a hydrophobic pocket formed by DI and DIII of the dimeric E partner.
The E protein mediates virus-membrane fusion by refolding to a hairpin-like E homotrimer with the fusion loops and TM domains at the same end [21], [22]. This reaction involves low pH-triggered dissociation of the homodimer, fusion loop insertion into the endosome membrane, formation of a core trimer composed of DI and DII, and the foldback of the DIII and stem regions towards the target membrane and their packing against the core trimer. The prefusion and postfusion conformations of the flavivirus E fusion protein are structurally and functionally similar to those of the E1 fusion protein from the alphavirus Semliki Forest virus (SFV) [23], [24], [25], and these fusion proteins are often referred to as “class II” [26], [27], [28]. In addition to the ectodomains whose trimer structures are described above, truncated fusion proteins composed of domains I and II (DI/II) can reconstitute SFV and DENV core trimer formation on target membranes [29], [30]. Such core trimers act as specific targets for DIII binding, thus recapitulating the protein-protein interactions during class II trimerization and hairpin formation.
Flaviviruses bud into the endoplasmic reticulum (ER) and are transported as virus particles through the secretory pathway and released by exocytosis [4]. Given the low pH that is present in the Golgi complex and trans-Golgi network (TGN) [31], how do flaviviruses avoid inactivation during their transport? The particles are assembled in the ER as immature non-infectious viruses containing heterodimers of the precursor membrane protein (prM) and E protein [4], [26], [32]. Subsequent exposure to low pH in the secretory pathway triggers a dramatic rearrangement to E homodimers and makes the prM protein accessible to furin cleavage [33], [34]. Processing of prM by cellular furin results in mature infectious virus in which E homodimers are poised to mediate fusion [33]. Important recent studies describe the structure of pr peptide in complex with E, and indicate that processed pr remains associated with the virus at low pH and can inhibit virus-membrane interaction [34], [35], [36]. Thus, pr on the virus could protect E protein from low pH in the secretory pathway.
The flavivirus prM/pr protein plays multiple roles in the virus life cycle (reviewed in [26]). prM acts as a chaperone for E protein folding [37] and associates with the tip of E [34]. prM also appears to respond to low pH to permit E rearrangement on the virus surface and allow furin access for prM processing [34], [38]. Following cleavage, the pr peptide may prevent premature virus fusion through bridging interactions that stabilize the E homodimer and thereby prevent dissociation to E monomers, a key fusion intermediate [35], [36]. To better understand these multiple roles of prM/pr, separation of its chaperone and pH-protection functions and characterization of the pr-E interaction are needed.
Here we developed a system to produce DENV pr peptide and reconstitute the pr-E interaction in vitro. At low pH pr bound to both monomeric and dimeric forms of E and blocked their membrane insertion and trimerization. Addition of exogenous pr to mature DENV particles inhibited virus fusion and infection. Mutation of a key histidine residue in the pr-E interface, E H244, reduced pr's binding and inhibitory activity, and reduced DENV secondary infection and particle production. The defect in particle production could be partially rescued by neutralization of exocytic low pH, indicating the important role of pr in protecting DENV from premature fusion during transport to the plasma membrane.
A number of truncated E proteins have been successfully produced by co-expression with prM (e.g., references [30], [39]), while the pr-E structural studies were based on a secreted hybrid protein containing truncated prM linked to truncated E [34]. Previous studies indicated that full-length TBEV prM could fold correctly when expressed in the absence of E protein [37], suggesting that production of pr peptide alone might be possible. We generated a construct based on residues 1–86 of DENV2 prM, truncating pr just before the start of the furin cleavage recognition site at residue 87 (Fig. 1A). This sequence was linked to a mammalian signal peptide at the N-terminus and to an affinity tag at the C-terminus, and expressed in 293T cells. The protein was isolated in a highly purified form by affinity chromatography and gel filtration (Fig. 1B), and was recognized by mAb prM-6.1 against prM [40] (data not shown). The pr peptide migrated at a position of ∼17 kDa in reducing SDS-PAGE, in keeping with its predicted size of 13 kDa plus the presence of carbohydrate due to the glycosylation site at position 69. This carbohydrate was removed by Peptide N-glycosidase F (PNGase F) to give a peptide of the predicted size. The protein was largely resistant to Endoglyosidase H (Endo H) digestion, indicating maturation of the carbohydrate chain as the protein transited through the Golgi complex. A mobility shift was observed upon reduction of pr, in keeping with the presence of 3 disulfide bonds in the structure of pr [34].
We also produced and purified a dimeric ectodomain form of DENV2 E protein containing all three domains (E′), a monomeric form containing E domains I and II (DI/II), and E domain III (DIII) (Fig. 1A and 1C), all as previously described in detail [30], [41].
As a first test of in vitro pr-E binding, we coupled pr to sepharose beads and tested its ability to pull-down truncated E protein containing only domains I and II. This form of E protein is monomeric and the tip of DII is thus accessible even at neutral pH. Previous studies showed that this and other DENV DI/II proteins are active in membrane insertion and trimerization at both neutral and low pH [30]. We observed efficient pull-down of DI/II protein by pr-sepharose (Fig. 2A), but in spite of the accessibility of the pr binding site on DI/II at neutral pH, pull-down was low pH-dependent. The pull-down of DI/II protein by pr was specific, as it was blocked by inclusion of mAb 4G2 against the E fusion loop at the DII tip, and did not occur with BSA-sepharose beads. These data suggested that the recombinant pr peptide could bind to the tip of DI/II in a low pH-dependent reaction.
For more detailed studies of pr-E binding, we performed surface plasmon resonance (SPR) assays using our various forms of recombinant E protein with immobilized pr peptide. Compared to the pull-down assay, SPR can detect low levels of protein-protein interactions as binding is detected in real time and does not require removal of unbound E. The E′ protein is a dimer at neutral pH and dissociates to monomers at low pH [30]. When SPR was performed with E′ protein buffered at pH 8.0 there was very low binding (low signal response) (Fig. 2B). As the buffer pH was decreased, the signal gradually increased, with maximal response observed at ∼pH 6.25 and no further increase at pH 6.0. A rapid decrease in signal was observed when the samples were shifted to protein-free buffer, indicating rapid dissociation of the pr-E interaction. Similar results were obtained using monomeric DI/II, with the lowest binding at pH 8.0, highest binding at pH 6.25, and a slight decrease at pH 6.0 (Fig. 2C). Thus, the dimeric E′ and monomeric E DI/II proteins bound pr peptide with similar pH-dependence. Binding to pr was specific, as little interaction was observed using the structurally similar E1 DI/II protein of SFV (Fig. 2D). In addition, binding of DENV E DI/II protein to pr was inhibited by preincubation with mAb 4G2 against the fusion loop (molar ratio 1∶1) (data not shown). Determination of the affinity of pr-E binding was not performed as the data did not fit to a simple Langmuir model of 1∶1 binding, presumably because of E protein aggregation at low pH.
Previous studies showed that retention of endogenous pr peptide on the furin-processed DENV particle inhibits virus interaction with liposomes at low pH [35]. Structural considerations suggested that this inhibition occurs primarily by blocking low pH-triggered dissociation of the E dimer, a required first step in the fusion reaction. To test this mechanism, we evaluated the effect of pr on the membrane interactions of dimeric and monomeric forms of E protein. The E′ dimer was preincubated with pr peptide or an unrelated protein with the same affinity tag for 5 min at pH 8.0, and then treated at pH 5.75 in the presence of target liposomes. Membrane-associated proteins were separated by liposome floatation on sucrose gradients. There was no liposome co-floatation when E′ protein was incubated with liposomes at neutral pH (Fig. 3A). About 70% of the total E′ floated with liposomes in the top part of the sucrose gradient after treatment at pH 5.75 in the presence (Fig. 3A, top panel) or absence (data not shown) of a control protein. In contrast, when E′ was preincubated with pr peptide (pr∶E′ molar ratio 12∶1) and treated with low pH, only ∼2% of E′-ST floated with the liposomes (Fig. 3A, middle panel). Inhibition by pr was not observed when it was added after E′ was treated at low pH in the presence of liposomes for 30 min (Fig. 3A, bottom panel), and thus pr needed to be present during the membrane insertion step. Inhibition was concentration-dependent, with 22% E′ co-floatation at a pr∶E′ molar ratio of 3∶1, 8% at 6∶1, and 0.4% for 24∶1 (data not shown; see also Fig. 3E).
We then tested the effect of pr on the DENV E DI/II protein. This protein is monomeric and its stable membrane interaction requires DIII to “clamp” the core trimer [30]. As shown in Fig. 3B, ∼25% of DI/II co-floated with liposomes at low pH in the present of DIII, while no co-floatation was detected when BSA was substituted for DIII protein. The addition of pr peptide blocked membrane interaction of DI/II when added prior to liposome incubation (Fig. 3B, 3rd panel), but not after liposome incubation (Fig. 3B, bottom panel).
The structurally related alphavirus protein SFV E1 DI/II is monomeric and efficiently interacts with membranes at low pH (80% cofloatation, Fig. 3C, middle panel). No inhibition occurred when pr peptide was added prior to liposome addition (Fig. 3C, bottom panel), in keeping with the lack of pr-SFV DI/II binding in the SPR experiments discussed above. Thus, pr peptide specifically inhibits target membrane interaction of both monomeric and dimeric forms of the DENV E protein.
E′ protein efficiently inserted into membranes over a wide range of pH values from 6.25-4.5 (Fig. 3D–E). However, pr's inhibition of E membrane insertion was less efficient in the pH range (pH 5.0) present in the late endocytic pathway (Fig. 3D–E). This loss of pr inhibition at more acidic pH may be relevant to recent studies of infection by immature DENV [42], as mentioned in the discussion section below.
All of the results above were obtained with soluble forms of the E protein. In order to test the ability of exogenous pr peptide to interact with and inhibit intact DENV, we took advantage of a previously described assay that monitors low pH-triggered fusion of DENV with cells [41]. In this fusion-infection assay, virus is pre-bound to target cells on ice, and then treated at 37°C for 1 min at low pH to trigger virus fusion with the plasma membrane. This fusion reaction is then quantitated by detecting the infected cells by immunofluorescence. We tested the effect of pr peptide during this 1 min low pH treatment using DENV1 WP and DENV2 NGC. The sequence of E DI/II is 68% identical between these two serotypes. Both serotypes showed efficient fusion and infection after treatment at pH 6.0, with about a 10-fold increase compared to samples treated at pH 7.9 (Fig. 4). The addition of pr peptide during the 1 min low pH treatment strongly inhibited DENV fusion and infection. Inhibition was dose-dependent, with 45–49% inhibition at 6 µM pr and 81–85-% inhibition at 30 µM pr. In contrast, pr did not inhibit low pH-triggered fusion by the alphavirus SIN (Fig. 4). Thus, exogenous DENV2 pr peptide can specifically interact with mature DENV1 and DENV2 to block virus fusion and infection. We did not observe inhibition when DENV was preincubated with 30 µM pr at pH 7.0 and then added to target cells in a standard infection assay, suggesting that under these conditions an inhibitory concentration of pr was not present during low pH-triggered fusion reaction in the endosome. This result also indicates that the presence of pr did not affect virus-cell binding.
Although the interaction of pr with DENV can clearly prevent virus-membrane interaction and fusion (this study and [35]), the importance of pr in protecting DENV during exocytic transport has not been defined. The binding interface between prM and E contains three complementary electrostatic patches containing 11 residues [34] (see also Fig. S1). Sequence analysis shows that these 11 residues (Fig. 5A, numbered residues) are highly conserved among the 4 DENV serotypes, and that D63 and D65 of pr, and the complementary H244 on E protein are conserved among all reported flavivirus sequences [34]. Optimal pr-E binding in vitro occurred at ∼pH 6.25 (Fig. 2), suggesting that protonation of H244 could be involved in this pH-dependence. To test this we substituted alanine for H244 in the DI/II protein. DI/II H244A was produced in highly purified form with electrophoretic mobility similar to that of the wild type (WT) protein in reducing and non-reducing SDS-PAGE (Fig. 1C).
We first tested the effect of the H244A mutation on pr-E binding. In agreement with our earlier results, WT DI/II protein was efficiently pulled-down by pr-sepharose (Fig. 5B). Pull-down was low pH-dependent and blocked by mAb 4G2 against the E fusion loop at the DII tip. In contrast, almost no H244A DI/II protein was pulled-down by pr-sepharose at either low pH or neural pH (Fig. 5B). SPR analysis of WT DI/II protein showed most efficient binding at pH 6.0, and binding was blocked by pre-incubating the DI/II protein with mAb 4G2 (molar ratio 1∶1) before dilution into SPR buffer (Fig. 5C, upper panel). Equivalent concentrations of H244A DI/II protein showed greatly reduced binding to pr compared to that of WT protein (Fig. 5C, lower panel). Although H244A binding was decreased, the residual binding was still blocked by mAb 4G2 and had an acidic pH optimum. This suggests that binding also involves other residues in the pr-E interface, such as the complementary residues identified in the structural studies and shown in Fig. 5A.
We then asked if the H244A DI/II protein was still active in binding to target liposomes. WT or mutant DI/II proteins were mixed with liposomes at low pH in the presence of DIII protein to stabilize the core trimer. Both proteins efficiently bound liposomes in a DIII-dependent reaction (Fig. 6), indicating that the mutant protein retains its ability to insert into target membranes and form a core trimer. In agreement with the results in Fig. 3C, floatation of the WT protein was blocked by inclusion of pr during the membrane insertion step (Fig. 6). In contrast, the efficiency of floatation of the H244A mutant protein was 43% in the absence of pr and 47% in the presence of pr. Thus, the H244A mutation did not inhibit E-membrane interaction but made that interaction insensitive to the presence of pr.
Since the E H244A mutation disrupts E protein's interaction with pr, we used this mutation to address the importance of pr in protecting DENV during transport through the exocytic pathway. We introduced the E H244A mutation into the infectious clone of DENV1 WP. WT and mutant viral RNAs were prepared by in vitro transcription and were electroporated into BHK cells. After culture for 3 d at 37°C, both WT and mutant RNA-electroporated cells expressed abundant E protein as detected by immunofluorescence microscopy (Fig. 7). Parallel cultures were incubated for 6 d and progeny virus in the culture media was detected by infectious center assays on indicator BHK cells. WT-infected cells produced infectious progeny virus with a titer of ∼1.5×105 IC/ml. However, two independent infectious clones of the H244A mutant produced no detectable progeny virus, even though the viral RNAs mediated efficient primary infection as shown in Fig. 7. This agrees with previous studies indicating lethal effects of an H244A mutation on DENV2 [43].
The absence of secondary infection by the H244A DENV1 mutant could be due to decreased virus particle production and/or production of particles that are non-infectious. Efficient DENV particle production is dependent on E protein folding, particle budding into the ER, and subsequent particle egress through the secretory pathway. To investigate these issues, we took advantage of the ability of the flavivirus prM and E proteins to assemble into virus-like particles (VLP) in the absence of other viral components or virus infection [44], [45], [46]. The VLP system avoids complications arising from selection of revertants of deleterious virus mutations such as H244A. Flavivirus VLP bud into the ER in the immature prM form, undergo furin maturation during transport through the secretory pathway, and display similar low pH-dependent fusion activity as infectious virions [44], [47]. The VLP system has been used extensively to follow the process of flavivirus particle production and the role of prM in this process [37], [44], [45], [48].
We established stable HEK 293 cells that inducibly express the DENV1 WT or H244A prM-E proteins. After 36 h induction with tetracycline, both WT and H244A cells show abundant intracellular expression of the DENV1 E protein as detected by immunofluorescence, while the parent cell line is negative for E expression (Fig. 8A). To evaluate whether WT and H244A E proteins were correctly folded, cells were induced for 36 h, lysed, and immunoprecipitated with a rabbit polyclonal antibody to E DIII, and with two conformation-specific mAbs. mAb 4E11 recognizes a discontinuous epitope on DENV E DIII and requires proper DIII disulfide bond formation for recognition [49], [50]. mAb 4G2 recognizes the fusion loop at the tip of flavivirus E DII and its epitope is sensitive to reduction [51]. Expression studies have shown that the 4G2 epitope is not formed if the E protein is expressed in the absence of prM [52], indicating that this epitope is particularly useful for diagnostic tests of prM's chaperone interaction with E (see also reference [37]). As shown in Fig. 8B, lysates from cells induced to express prM plus WT or H244A E proteins showed strong reactivity with all three antibodies. Quantitation of multiple experiments confirmed that WT and H244A E proteins were comparably recognized by the 4E11 and 4G2 mAbs. Thus, by these criteria H244A E protein interacts with prM protein and is correctly folded. This result also agrees with our finding that truncated H244A E protein expressed with prM in the S2 cell system was fully active in low pH-dependent membrane binding and trimerization, suggesting correct folding (Fig. 6).
We then used the inducible cells to examine VLP production. Expression was induced for 36 h. The cells were then lysed and the E proteins immunoprecipitated, and the VLP in the culture media were pelleted by ultracentrifugation. Analysis by western blotting showed strong E protein expression in both WT and H244A cells, and no expression in the parent cells (Fig. 8C). The WT cells released E protein in VLP, but VLP release from cells expressing the H244A mutant E protein was greatly reduced (Fig. 8C, - media samples). This result is in keeping with the hypothesis that the H244A cells assemble VLP in the neutral pH environment of the ER but that VLP release is inhibited by the lack of pr protection from the low pH of the secretory pathway. To test this idea, we induced WT and H244A prM-E expression and cultured the cells in the presence of 20 mM NH4Cl to neutralize the acidic pH in the Golgi and TGN compartments (Fig. 8C, +NH4Cl lanes). The cellular expression level of either E protein was not significantly affected by NH4Cl treatment, and WT VLP production was similar in NH4Cl-treated cells and untreated cells. However, production of VLP containing the H244A mutant E protein was increased 4–7 fold in NH4Cl-treated cells. While H244A VLP production was still significantly decreased compared to that of WT, it was selectively rescued by NH4Cl treatment.
During translation of the flavivirus polyprotein, prM is the first protein translocated into the ER lumen, where it acts as a chaperone during the folding of the subsequently translocated E protein [4], [37], [44]. In addition to this important role of prM during E protein synthesis, a variety of data suggest that the interaction of pr peptide with the viral E protein protects flaviviruses from low pH during their transport through the exocytic pathway [34], [35], [36]. Here we showed that a recombinant pr peptide was efficiently folded, glycosylated, and secreted from 293T cells in the absence of its normal prM context and furin processing. Recombinant pr bound to soluble E proteins at low pH, inhibited E-membrane insertion, and interacted with mature dengue virus to block fusion and infection. Alanine substitution of the conserved E H244 within the pr-E interface disrupted pr-E binding in vitro and blocked secondary virus infection. VLP production was inhibited by the H244A mutation and partially rescued by pH neutralization with NH4Cl. Together our data demonstrate the critical role of pr in protecting DENV from exocytic low pH.
The in vitro interaction of pr with various truncated forms of E protein was strongly pH-dependent, with a pH optimum of ∼6.25. In situ measurements indicate that the pH of the TGN is ∼6 [53], while the pH optimum of DENV2 NGC fusion is ∼6.2 [41]. The low pH of the TGN is critical for the rearrangement of immature DENV to allow furin cleavage, but once the virus is processed it becomes fusion-active in this same pH range. Thus the pH dependence of the pr-E interaction appears optimized to protect DENV during its continued transit through the secretory pathway. Pr's inhibition of E membrane insertion was less efficient at a pH value (pH 5.0) similar to that in the late endocytic pathway (Fig. 3D–E). This loss of pr inhibition at more acidic pH could help to explain the recent finding that infection by immature DENV is enhanced by antibodies to prM [42]. The antibody-bound immature virus is likely to be endocytosed and processed by cellular furin in the endocytic pathway [54]. The lower pH conditions of the late endocytic pathway could then cause the loss of pr inhibition and allow virus fusion.
The structure of furin-cleaved DENV at pH 6.0 shows that pr is bound to the virion through interactions with the DII tip of one E protein and DI on the neighboring E monomer [35], [36]. This suggested that pr might primarily block virus-membrane interaction by preventing dissociation of E dimers, a required first step in the fusion pathway [55]. Our results show efficient binding of pr to the dimeric form of the DENV E protein, but also to the monomeric DI/II form. We do not know if the E′ protein dimer is stabilized by pr interaction or if the dimer dissociates prior to interaction with pr, and experiments to address these points were inconclusive (data not shown). The similar pH dependence of pr binding to monomeric and dimeric E proteins suggests that pr may bind the same site in both cases. mAb 4G2 against the fusion loop inhibited pr interaction with E DI/II, confirming that pr was binding to the DII tip rather than to other sites on expressed E proteins. In keeping with its binding site in the vicinity of the fusion loop, pr peptide blocked the membrane insertion and liposome co-floatation of E′ and DI/II proteins. Prior studies showed that a monomeric DI/II protein with a single Strep affinity tag stably inserts into liposomes at either neutral or low pH [30], and pr blocked this insertion even at pH 8.0 where its interaction with DI/II was suboptimal (data not shown). Thus, while the pr-E interaction is strongly low pH-dependent, its functional inhibition of membrane insertion can still be observed at neutral pH in the presence of excess pr.
Several other studies have addressed the role of E H244 in the flavivirus lifecycle. Experiments in TBEV evaluated particle production and membrane fusion activity using a VLP system [56]. Mutation of H248 (TBE numbering) to A or I blocks VLP secretion, in agreement with our results. However, an H248N mutant efficiently produces VLP, and these particles show WT levels of fusion activity. WNV E H246A or Q mutations inhibit release of infectious reporter virus particles from cells, as do a number of other substitutions at this position [57]. Replacement of H246 with aromatic residues such as phenylalanine allows both particle release and infectivity. An H244A mutation in DENV2 NGC inhibits infectious virus production [43]. E H244 and its interacting partners D63 and D65 on pr are conserved within the flaviviruses, and thus these data from several flaviviruses plus our DENV results support an important role for the E 244 position. However, a histidine residue at this position does not seem to be strictly required for particle production, suggesting that substitutions such as 244F and 244N can support the interaction of E with pr.
In contrast to the block in production of H244A VLP, the H244A DI/II protein was efficiently secreted from cells. Mutant protein secretion was somewhat reduced, with the final yield of DI/II H244A about half that of the WT protein in two separate preparations (data not shown), suggesting some effects of non-optimal pr interaction. However, unlike the E protein in virus or VLP, the truncated DI/II protein lacks the TM region and does not mediate membrane fusion, and thus may be relatively independent of the pH-protection function of pr. The purified WT and mutant DI/II proteins were able to bind liposomes and form core trimers that were stabilized by DIII (Fig. 6). Thus, the mutant protein is correctly folded and active in membrane insertion. Studies with conformation-specific mAbs also provided evidence for the correct folding of H244A E protein (Fig. 8B). Together, these results suggest that the H244A E protein is still able to access the chaperone functions of prM, while its decreased pr binding indicates that it can no longer utilize the pH protection functions of pr.
These data are consistent with the idea that, similar to WT E, the mutant protein is assembled with prM into VLP in the ER. The membrane insertion and trimerization activity of H244A suggest that the full-length mutant protein would be fusion-active on such VLP once they are transported from the neutral pH of the ER to the low pH of the Golgi and TGN [31]. Thus, the decreased release of H244A VLP and its partial rescue by neutralization of the exocytic pathway support a critical role for pr in protecting DENV from exocytic low pH, and suggest that virus/VLP fuses in the TGN in the absence of pr-E interaction. Rescue of H244A VLP production by NH4Cl was clearly incomplete. This may be due to complex aspects of both virus and cell, such as direct effects of the H244A mutation on particle assembly in the ER, or difficulties in blocking fusion of a virus with the relatively high pH threshold of DENV.
Several strategies have been used to block flavivirus and alphavirus fusion reactions and thus inhibit virus infection. SFV and DENV fusion are specifically blocked by exogenous DIII, which binds to the core trimer and prevents the foldback of endogenous DIII and hairpin formation [41]. A later stage in DENV fusion is targeted by a stem-derived peptide, which binds to the ectodomain trimer in which DIII has folded back but stem packing has not yet occurred [58]. These virus protein-protein interactions can be reconstituted in vitro [29], [30], [58], opening the possibility of using them as screens for small molecule inhibitors of virus fusion and infection.
The in vitro reconstitution of the pr-E interaction using soluble components could also act as a screen for small molecule inhibitors of this important flavivirus protein-protein interaction. Such inhibitors could act at multiple points in the virus lifecycle. During virus protein biosynthesis, an inhibitor could block the chaperone interaction of prM with E, leading to misfolding of E and its elimination by the ER quality control pathway. An inhibitor of pr interaction could make E protein susceptible to premature fusion in the TGN and could thus block virus production similar to the H244A mutation. It is also possible that small molecule inhibitors of pr-E binding could interact directly with the DII tip on mature virus particles, perhaps stabilizing the dimer and/or blocking membrane insertion of the fusion loop, thereby blocking virus fusion. Thus the in vitro system we describe here has the potential to identify molecules that could aid in the study of the flavivirus lifecycle and that could act to inhibit specific steps.
Previous studies showed that after cleavage endogenous pr is retained on the virus particle if the virus is maintained at acidic pH [35]. Under these conditions, the virus-pr complex does not bind target membranes, while virus from which pr is first released at neutral pH efficiently binds membranes upon shift to acid pH. Thus, the bound endogenous pr inhibits virus-membrane interaction and presumably blocks virus fusion [35]. Our results demonstrated that even after maturation to fully infectious DENV particles, exogenous pr could add back to the virus and inhibit low pH-triggered virus fusion and infection. The flavivirus membrane fusion reaction is very rapid, occurring within seconds of low pH treatment [47]. Recombinant DENV2 pr peptide inhibited fusion by both DENV1 and DENV2, suggestive of a fairly broad spectrum inhibition in agreement with the strong sequence conservation at the pr-E interface [34].
The structure of the flavivirus E protein in its pre-fusion and post-fusion conformations defines the dramatic conformational changes between these two states. Many questions about the intermediates that connect the pre- and post-fusion conformations remain. In particular, it will be important to define the membrane protein rearrangements in the context of the highly organized flavivirus particle. For example, a neutralizing E mAb that blocks virus fusion was used to trap a West Nile virus fusion intermediate [59]. It will be interesting to evaluate if exogenous pr peptide could also be used as a novel probe to capture intermediates in the flavivirus fusion pathway.
BHK-21 cells and C6/36 mosquito cells were cultured as described previously [60]. 293T cells and T-REx™-293 cells were cultured as previously described using tetracycline-deficient fetal calf serum for the latter cells [61]. The DENV2 New Guinea C (NGC) strain and the DENV1 Western Pacific (WP) strain were propagated in C6/36 cells in DMEM containing 2% heat-inactivated fetal calf serum and 10 mM Hepes, pH 8.0, as previously described [41], [62]. Sindbis virus expressing green fluorescent protein was obtained as an infectious clone (a kind gift from Dr. Hans Heidner) and propagated in BHK cells [63].
4G2 is a mouse monoclonal antibody (mAb) that recognizes the fusion loop of flavivirus E proteins [51], [64]. mAb prM-6.1 recognizes a linear epitope on prM, and was a kind gift of Drs. Chunya Puttikhunt and Nopporn Sittisombut [40]. 4E11 is a mouse mAb that recognizes DIII of DENV E protein and neutralizes all 4 serotypes of dengue virus [49], [50], and was a kind gift of Dr. Fernando Arenzana-Seisdedos (Institute Pasteur, Paris). The anti-DIII polyclonal antibody Sango was raised by immunization of a rabbit with purified DENV2 DIII protein [30]. Western blot detection of truncated E proteins used 4G2 or Sango antibodies. A mAb to β-actin was obtained from Sigma and used to confirm equivalent loading of cell lysate samples. Immunofluorescence detection of DENV-infected cells used the antibody to DIII or mouse polyclonal anti-DENV2 hyperimmune ascitic fluid (obtained from Robert B. Tesh, University of Texas Medical Branch), with Alexa Fluor 488 or rhodamine-conjugated secondary antibodies (Molecular Probes).
The sequence encoding residues 1–86 of pr was amplified by PCR of an expression plasmid for DENV2 NGC prM-E DI/II [30]. The PCR product was ligated into the pPUR vector (Clontech), with the 21-residue TPA signal peptide [65] fused at the N-terminus and a tandem Strep tag at the C terminus (Fig. 1). The plasmid, referred to as pPUR-TPA-pr-STST, was transfected into 293T cells using polyethylenimine (PEI, Polysciences). For optimal protein production, 3.5×106 cells were plated per 10 cm dish and cultured for 24 h in 10 ml of complete medium. 7.5 µg plasmid in 1 ml DME was mixed with 30 µg PEI, incubated 10 min, then added drop wise to the cell culture medium. After 12 h, the medium was changed to 10 ml DME plus 2% serum. The culture medium was collected after 48h and again after 72h. Pr was purified by affinity chromatography on a Strep-Tactin column from IBA BioTAGnology and by gel filtration using a Sephadex G75 column [30]. Final yields were ∼2 mg purified protein/1 liter culture supernatant.
Truncated DENV E proteins (Fig. 1) were obtained by inducible expression in Drosophila S2 cells and purified by affinity chromatography as previously described in detail [29], [30]. The H244A mutation was introduced into the DI/II protein by in vitro mutagenesis, and S2 cell expression and purification were performed as above. DENV2 NGC DIII (Fig. 1) was previously referred to as LDIIIH1CS [30], and contains domain III, the linker between domain I and domain III, and the H1 and CS regions of the stem domain. DIII was expressed in E.coli and refolded as previously described [41]. SFV E1 DI/II protein was produced as previously described [29]. All purified proteins were stored in TAN buffer (20 mM Triethanolamine[TEA], pH 8.0; 130 mM NaCl) at −80°C.
SDS-PAGE analysis was performed using 10–12% acrylamide gels with a Bis-Tris buffer system (Invitrogen). Western blots were performed with Alexa Fluor 688-conjugated secondary antibodies (Molecular Probes), and were quantitated using an Odyssey Infrared Imaging system and Odyssey InCell Western software (LI-COR Biosciences) [30]. Standard curves with purified E proteins confirmed the linearity of this analysis (data not shown).
Pr or BSA was coupled to NHS-activated sepharose 4 fast flow (GE Healthcare) as described in the manual. In brief, sepharose was washed with 1mM HCl, and incubated with 660 µg pr or BSA/ml in 0.2 M NaHCO3, 0.5 M NaCl, pH 8.3 at room temperature for 1.5 hr. The reaction was quenched with 0.1M Tris-HCl pH 8.5 for 30 min and free protein removed by washing with PBS. About 1mg of protein was coupled to 1ml beads. For the pull-down assay, 3 µg DI/II protein was pre-incubated where indicated with 24 µg 4G2 (molar ratio 1∶2) or control mAb for 10 min at room temperature, and then incubated for 1 h on a rocker at room temperature with 10 µl of pr- or BSA-sepharose in a buffer containing 20 mM MES, 20 mM TEA, 130 mM NaCl, 0.2% Tween 20 at pH 8.0 or 6.25. The beads were then washed twice with the corresponding buffer and the bound DI/II was analyzed by SDS-PAGE and western blot.
SPR studies were performed on a BIAcore 2000 instrument (GE Healthcare). Purified recombinant pr was immobilized on a CM5 biosensor chip by primary amine coupling as described in the manual. In brief, pr peptide was diluted to 10 µg/ml in 10 mM sodium acetate pH 4.7 and pre-concentrated on the chip surface. The chip was then activated by a mixture of 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide and N-hydroxysuccinimide, followed by quenching with 1M ethanolamine at pH 8.5. Under these conditions, pr was immobilized to a final density of 600 or 1000 response unit (RU). A control cell was mock-coupled with protein-free solutions. To test interaction, truncated E proteins were diluted to 1.2 mM in a MES/TEA buffer (20 mM MES, 20 mM TEA, 130 mM NaCl) at a pH range of 6.0 to 8.0, and flowed over the chip for 300 s at 0.3 µl/min, followed by buffer alone at the same flow rate. After each round, the chip was regenerated by washing with 50 mM NaOH in 1 M NaCl. The pr chip showed undiminished E binding activity for at least 50 rounds.
Liposomes were prepared by freeze-thaw and extrusion through 200 nm polycarbonate filters [66], and were stored at 4°C in TAN buffer under N2 and used within 2 weeks of preparation. Liposomes were composed of a 1∶1∶1∶3 molar ratio of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE), sphingomyelin (bovine brain) (Avanti Polar Lipids; Alabaster, AL), and cholesterol (Steraloids, Inc.; Wilton, NH), plus trace amounts of 3H-cholesterol (Amersham; Arlington Heights, IL).
Protein-membrane interaction was monitored using a liposome co-floatation assay [29], [30]. E′ or DI/II proteins at a final concentration of 50 µg/ml were incubated in TAN buffer (pH 8.0) for 5 min at 28°C in the presence of 200 µg pr peptide/ml as indicated. Liposomes were then added to a final concentration of 1mM lipid and the samples were adjusted to pH 5.75 by the addition of 0.3 M MES or maintained at pH 8.0, and the incubation continued at 28°C for 30–60 min. The samples were then adjusted to 20% sucrose and loaded on top of a 300 µl cushion of 40% sucrose, then overlaid with 1.2ml 15% sucrose and 200 µl 5% sucrose. All sucrose solutions were at the same pH as the samples, and were wt/wt in TAN buffer at pH 8.0 or in MES buffer (50 mM MES, 100 mM NaCl) at pH 5.5. Gradients were centrifuged for 3 hr at 54,000 rpm at 4°C in a TLS55 rotor, and fractioned into the top 700 µl, middle 400 µl and bottom 1 ml. The 3H-cholesterol marker was quantitated by scintillation counting. 200 µl of each fraction were precipitated with 10% trichloroacetic acid and analyzed by SDS-PAGE and western blotting [29]. Purified human secreted placental alkaline phosphatase with a ST affinity tag (Seap) was used as a control protein [67], and was a kind gift from Yves Durocher, Biotechnology Research Institute, Montreal.
The fusion-infection assay was performed essentially as described previously [41]. In brief, BHK cells grown on 96-well plates were washed twice with ice cold binding medium (RPMI without bicarbonate, 0.2% BSA, 10 mM Hepes, and 20 mM NH4Cl, pH 7.9). Virus stocks were diluted in binding medium and incubated with cells on ice for 3 h with gentle shaking. Cells were washed twice with binding medium to remove unbound virus and pulsed for 1 min at 37°C in 100 µl RPMI without bicarbonate, containing 0.2% BSA, 10 mM Hepes and 30 mM sodium succinate at pH 6.0 or 7.9, containing the indicated concentration of pr peptide. Infected cells were incubated in MEM plus 2% FCS and 50 mM NH4Cl for 4 h at 37°C, and then at 37°C for 2 d in the presence of 20 mM NH4Cl. The number of infected cells was quantitated by immunofluorescence using mouse polyclonal anti-DENV2 antibody. Infection observed at pH 7.9 represents virus that is endocytosed and fuses during 1 min at this pH.
The DENV1-WP infectious clone (reference [68], a kind gift from Dr. Barry Falgout) was digested with KpnI and a 3.3kb fragment including the E sequence was sub-cloned into the pGEM3Z vector to generate pGDENV1 3.3. pGDENV1 3.3 was used as a template to generate the E H244A mutation, using circular mutagenesis as previously described [69]. A 2.6kb BstB1/XhoI fragment containing the H244A mutation was sub-cloned into the DENV1-WP infectious clone to obtain DENV1-E H244A. The mutation was confirmed by restriction analysis and sequencing of the complete prM-E region. Two independent infectious clones were used to confirm the phenotype.
The WT and the mutant infectious clones were linearized by Sac II digestion and used as templates for in vitro transcription [70]. RNAs were electroporated into BHK cells and cells were cultured overnight at 37°C followed by 6 d at 28°C in MEM containing 2% FBS and 10 mM HEPES, pH 8.0. Progeny virus in the medium was quantitated by infectious center assay on indicator BHK cells, using mouse polyclonal anti-DENV2 antibody. To detect primary infection, aliquots of the electroporated cells were plated on coverslips, cultured 3 d at 37°C, and processed for immunofluorescence microscopy as above.
WT and E H244A mutant DENV1 prM-E sequences were PCR-amplified from the pGDENV1 3.3 subclones described above, and cloned into pcDNA4/TO (Invitrogen). These constructs were transfected into T-REx™-293cells using Lipofectamine 2000 (Invitrogen) and selected in T-REx HEK medium containing 125 µg/ml Zeocin, all as previous described [61].
To test E protein folding and expression, 1×106 WT and mutant E expressing cells were seeded in 10 cm plates, cultured for 24h, and then E protein expression was induced by culture for 36 h in 1.5 µg/ml tetracycline in DME medium with 10% FCS at 37°C. Cells were lysed in RIPA buffer (50 mM Tris-HCl pH 7.4, 150mM NaCl, 1% NP40, 0.5% Na-deoxycholate, 0.1% SDS, 1mM PMSF, 1× Roche complete protease inhibitor cocktail) on ice for 1 hr. The cell lysates were cleared by centrifugation for 30 min at 10,000×g and protein concentrations were quantitated and normalized. E proteins were immunoprecipitated from cell lysate samples (500 µg total cellular protein) using 20 µg purified mAb 4G2 or mAb 4E11 and 20 µl protein-G sepharose, or 30 µl Sango antibody and 20 µl protein-A sepharose. 4E11 and 4G2 immunoprecipitated samples were blotted with Sango. Sango immunoprecipitated samples were blotted with mouse anti DENV2 serum.
For VLP secretion studies, 2–3×106 cells were seeded in 10 cm plates, cultured for 24h, and then induced by culture for 36 h in 1.5 µg/ml tetracycline in DME medium with 10% FCS at 37°C. The culture media were centrifuged at 10,000×g for 30 min to remove cell debris. VLPs were then pelleted through a 0.5 ml sucrose cushion by centrifugation at 54,000 rpm for 2 h at 4°C using a TLS55 rotor. To test the effect of neutralizing the pH of acidic cellular compartments, cells were seeded and induced as above. After 2 h of induction the media were changed to DME medium containing 20 mM HEPES pH 8.0, 2% FCS, and 1.5 µg/ml tetracycline plus 20 mM NH4Cl as indicated, and the incubation continued for a total of 36 h. E proteins in the cell lysates were immunoprecipitated using mAb 4G2. VLP and lysate samples were then analyzed by SDS-PAGE and western blot using Sango.
|
10.1371/journal.pcbi.1005224 | A Graph-Centric Approach for Metagenome-Guided Peptide and Protein Identification in Metaproteomics | Metaproteomic studies adopt the common bottom-up proteomics approach to investigate the protein composition and the dynamics of protein expression in microbial communities. When matched metagenomic and/or metatranscriptomic data of the microbial communities are available, metaproteomic data analyses often employ a metagenome-guided approach, in which complete or fragmental protein-coding genes are first directly predicted from metagenomic (and/or metatranscriptomic) sequences or from their assemblies, and the resulting protein sequences are then used as the reference database for peptide/protein identification from MS/MS spectra. This approach is often limited because protein coding genes predicted from metagenomes are incomplete and fragmental. In this paper, we present a graph-centric approach to improving metagenome-guided peptide and protein identification in metaproteomics. Our method exploits the de Bruijn graph structure reported by metagenome assembly algorithms to generate a comprehensive database of protein sequences encoded in the community. We tested our method using several public metaproteomic datasets with matched metagenomic and metatranscriptomic sequencing data acquired from complex microbial communities in a biological wastewater treatment plant. The results showed that many more peptides and proteins can be identified when assembly graphs were utilized, improving the characterization of the proteins expressed in the microbial communities. The additional proteins we identified contribute to the characterization of important pathways such as those involved in degradation of chemical hazards. Our tools are released as open-source software on github at https://github.com/COL-IU/Graph2Pro.
| In recent years, meta-omic (including metatranscriptomic and metaproteomic) techniques have been adopted as complementary approaches to metagenomic sequencing to study functional characteristics and dynamics of microbial communities, aiming at a holistic understanding of a community to respond to the changes in the environment. Currently, metaproteomic data are largely analyzed using the bioinformatics tools originally designed in bottom-up proteomics. In particular, recent metaproteomic studies employed a metagenome-guided approach, in which complete or fragmental protein-coding genes were first predicted from metagenomic sequences (i.e., contigs or scaffolds), acquired from the matched community samples, and predicted protein sequences were then used in peptide identification. A key challenge of this approach is that the protein coding genes predicted from assembled metagenomic contigs can be incomplete and fragmented due to the complexity of metagenomic samples and the short reads length in metagenomic sequencing. To address this issue, in this paper, we present a graph-centric approach that exploits the de bruijn graph structure reported by metagenome assembly algorithms to improve metagenome-guided peptide and protein identification in metaproteomics. We show that our method can identify much more peptides and proteins, improving the characterization of the proteins expressed in the microbial communities.
| Microbiome studies have produced massive metagenomic data, and more recently other meta-omics including metatranscriptomic and metaproteomic data [1]. Analyses of these data reveal insights into the composition, function and regulatory characteristics of the microbial communities associated with different ecosystems, habitats and hosts [2–8]. While metagenomic sequencing reveal important properties of microbial communities, other meta-omic (e.g., metatranscriptomic [9, 10] and metaproteomic [11–13]) techniques can provide additional insights, in particular on functional characteristics, such as gene activities, their regulation mechanisms and the dynamics of microbial communities, to understand how microbial organisms work as a community to respond to the changes in their environment, e.g., the health condition of the host of human microbiome [14–16]. Current metatranscriptomic and metaproteomic studies often directly adopt protocols originally developed for the transcriptomic and proteomic studies of model bacterial species; for examples, many metatranscriptomic projects exploited the bacterial RNA-seq protocol [17, 18], while most metaproteomic studies applied the common bottom-up proteomics approach, in which the proteins extracted from community samples are first tryptically digested and then analyzed by using one-dimensional or two-dimensional liquid chromatography tandem mass spectrometry (LC-MS/MS) [19–24].
Similarly, metaproteomic data are analyzed using the bioinformatics approaches used in bottom-up proteomics. Specifically, the first step of metaproteomic data analysis is the peptide identification, achieved by searching MS/MS spectra from an LC-MS/MS experiment against the tryptic peptides in silico digested from a target database of proteins that are potentially present in the metaproteomic sample. Many peptide search engines have been developed for this purpose in the proteomics field, including commonly used tools such as Mascot [25], Sequest [26], X!Tandem [27], InSPEct [28] and MSGF+ [29]. Their applications in metaproteomics rely on the pre-assembly of a protein database. Early metaproteomic studies used the collection of proteins encoded by fully sequenced bacterial genomes that likely live in the environment (e.g., human gut [11]) as the target database. This collection may be largely incomplete, e.g., a large fraction (10%-34%) of genes from HMP [30] or MetaHIT [31] shotgun sequencing are completely novel [6]. As a result, more recent metaproteomic studies employed a metagenome-guided approach, in which complete or fragmental protein-coding genes were first predicted from metagenomic sequences (i.e., contigs or scaffolds), acquired from the matched community samples, and predicted protein sequences were then used in peptide identification [24]. Several software tools have been developed for protein coding gene prediction from metagenomic sequences, including MetaGeneMark [32] and our own software FragGeneScan [33, 34]. A key challenge of this approach is that the protein coding genes predicted from assembled metagenomic contigs can be incomplete and fragmented due to the complexity of metagenomic samples and the short reads length in metagenomic sequencing. As the linear representations of contigs and scaffolds in metagenome assembly do not capture their putative connections, the short contigs contain only gene fragments, and even long contigs contain broken genes at their ends. As a result, the target peptides collected in this manner may miss many full-length tryptic peptides that are potentially observed in the metaproteomic experiments.
To alleviate the peptide/protein identification problem caused by incomplete/fragmental reference proteins, we propose a graph-centric approach to improving metagenome-guided peptide identification in metaproteomics. Many short read assemblers, including those commonly used for metagenome assembly such as Velvet [35], SOAPdenovo [36], MegaHIT [37] and SPAdes [38], employed the de Bruijn graph [39, 40] as the core data structure, in which each edge represents an assembled unique sequence from metagenomic reads (i.e., the contigs), and the graph structure represents the ambiguous connections between contigs that cannot be resolved by using sequencing reads. Some assemblers including SOAPdenovo [41] and metaSPAdes [42] report the de Bruijn graph of the assembly along with contigs. As demonstrated in our previous work, by exploiting the de Bruijn graph structure in metagenome assembly, we can reconstruct longer and more complete transcript sequences from short metatranscriptomic reads than the straightforward approach based solely on contigs [43]. Here, we attempt to predict protein coding genes directly from the sequences in the de Bruijn graph, including the proteins that span multiple edges in the graph, to expand the target protein database for metaproteomic data analysis. We implemented an algorithm that takes as input the de Bruijn graph of a metagenome assembly, traverses the graph in a depth-first search (DFS) fashion, and outputs a target database consisting of the tryptic peptides in all putative open reading frames (ORFs) encountered during the traversal. In the following step, the identified tryptic peptides were used to retrieve potential protein sequences by traversing the graph for the second time. Using three metaproteomic datasets with matched metagenomic sequencing data, we show that much more peptides and proteins can be identified when the targeted database is constructed from graph structures of matched metagenomic sequences than those from the database only consisting of proteins predicted from contigs, indicating the metagenome-guided graph-centric approach can improve the peptide and proteins identification in metaproteomics.
As illustrated in Fig 1, we developed a pipeline for protein identification from metaproteomic data when metagenomic and metatranscriptomic data are acquired from matched samples. The pipeline exploits the maximum information available when both metagenomic and metatranscriptomic data are obtained from matched samples, and attempts to address the objective of protein identification in metaproteomics.
The pipeline is particularly useful when the depth of metagenomic and metatranscriptomic sequencing are not sufficiently high, and thus they complement to each other to provide a comprehensive coverage of the whole set of genes encoded in the metagenome. In this pipeline, we first assemble the metagenomic and metatranscriptomic sequencing data together (note that because there is no split gene structures in bacterial genes, metatranscriptomic sequencing reads represent contiguous segments in corresponding bacterial genomes in the same manner as metagenomic reads), and the resulting assembly (denoted as Assembly-Combined) are used to construct the target protein database for protein identification.
We emphasize that in the pipeline, the metagenome/metatranscriptome assembly is represented as de Bruijn graphs instead of a collection of contig sequences as used in conventional methods. As a result, peptide/protein sequences are extracted from the de Bruijn graphs, and thus may span multiple edges (contigs) in the graph. In order to retain the de Bruijn graph representation in the assembly, we take the SOAPdenovo assembly algorithm [41] as an example in this paper, which reports the de Bruijn graph structures in addition to the contig sequences in the assembly. Other assemblers can also be used in our pipeline, as long as they report graph structures of the assembly. Below we will present software tools (Graph2Pep and Graph2Pro) to extract peptides/proteins from the de Bruijn graphs of metagenome and/or metatranscriptome assembly.
To utilize de Bruijn graphs of metagenome assembly for protein identification, we use a two-step strategy: in the first step, all putative tryptic peptides are predicted from the de Bruijn graph, while in the second step, full-length protein sequences are predicted to cover the whole set of tryptic peptides identified from the initial database searching results of the metaproteomic data. This way, we will not overburden the MS/MS spectra identification with excessive and potentially error-prone reference protein sequences that could be predicted from the graphs.
As illustrated in Fig 2, our core algorithms (Graph2Pep and Graph2Pro) for peptide/protein identification from graphs both take as input a contracted de Bruijn graph, a directed graph reported by a fragment assembly algorithm (such as SOAPdenovo), in which each vertex represents a k-mer, and each edge represents a DNA sequence resulting from the collapse of the one-in-one-out k-mers between the two terminal vertices. Because both DNA strands are represented in the graph, it has a symmetric property: each edge (and vertex) has a counterpart that represent the reverse complement of the DNA sequence represented by the edge (and vertex); when an edge represents a palindromic sequence, its counterpart is itself.
The combined set of tryptic peptides, including those predicted from long edges and those extracted from one or more short edges in the graph (by Graph2Pep), are used as the target database for peptide identification in the metaproteomic data by using a peptide search engine (such as MSGF+ as used here). Note that this step is not going to generate the final report of protein identification; instead, it will produce a collection of tryptic peptides that are encoded in the de Bruijn graph assembly, and are likely to be present in the sample. Therefore, we can use a less stringent criterion to filter peptide identifications (i.e., by using a relatively low FDR threshold 5% in this work) so that more putative peptides can be taken into consideration when we attempt to construct the target database of potential proteins in the sample (using Graph2Pro; see below) for the second (and final) step of protein identification.
We implemented the Graph2Pep and Graph2Pro algorithms in C++ and incorporated them into a pipeline for metaprotomics data analysis. We also included in our pipeline open source software tools (e.g., FragGeneScan and MSGF+) released by us and others previously, and several wrapper scripts in Python. These programs have been assembled in a streamline, and thus can be conveniently used for peptide/protein identification in metaproteomics when matched metagenomic and/or metatranscriptomic data are available. The package is available as open source software at https://github.com/COL-IU/Graph2Pro. In this study, we only consider the fully tryptic peptides in Graph2Pep algorithm. However, the program has one parameter allowing for adjusting the maximum number of mis-cleavages (default = 0). Note that it takes longer time to run the Graph2Pep program when mis-cleavages are allowed. In addition, the users can adjust another parameter of length threshold (default = 6 as used in this study) in the Graph2Pep program to filter peptides shorter than the threshold to be used in the first round of database searching. In a test case, the de Bruijn graph contains 18,523,653 edges and 37,047,308 vertices, from which 44,798,054 putative tryptic peptides are generated by Graph2Pep. The programs runs in 11 minutes and 22 seconds on a single CPU of Intel(R) Xeon(R) E5-2670 0 @ 2.60GHz.
We implemented our graph-centric algorithms Graph2Pep and Graph2Pro in C++, and incorporated them into a pipeline for protein identification from metaproteomic MS/MS spectra data. We applied our pipeline to the waste water microbiome data, and the results show that our pipeline can significantly improve the identification of proteins from MS/MS spectra. Detailed information of identified proteins and their functional annotations are available in the supplementary data.
For each sample (SD3, SD6 and SD7), we assembled the combined datasets of metagenomic and metatranscriptomic sequences. The statistics of the assembly results and the protein-coding genes predicted from the contigs in the assemblies are summarized in Table 1. There are 19,553 contigs from the assembly of SD3 dataset with the N50 contig length of 840 bps, while more and longer contigs are assembled in SD6 and SD7 datasets. FragGeneScan predicted 32,760 protein-coding genes in the SD3 dataset, 113,135 genes for SD6 and 111,849 genes for SD7. Based on the graph structures of the assemblies, Graph2Pep output ∼ 16 million, ∼ 35 million and ∼ 33 million peptides in SD3, SD6 and SD7 datasets, respectively.
The assembly results (both the contigs and the assembly graph) of the combined metagenomic and metatranscriptomic datasets were used to predict peptides/proteins for MS/MS spectra identification. 603,867, 150,216, 148,310 MS/MS spectra in the samples of SD3, SD6 and SD7, respectively, were given as the input to the database search by MSGF+. The peptide identification results at the false-discovery rate of 1% are summarized in Table 2. We also showed MS/MS spectra identification based on proteins predicted from contigs for comparison. In SD3 dataset, we identified 18,498 spectra (PSMs, or peptide spectrum matches) using proteins predicted from contigs (by FragGeneScan), and 43,946 PSMs using peptides predicted from the assembly graph (by Graph2Pep) both at 1% FDR. In total, the first round of database searching identifies 13,928 unique peptides from 52,498 spectra, including 2,354 unique peptides and 9,496 spectra identified in both sets of FragGeneScan-predicted proteins and Graph2Pep-predicted peptides. The Venn diagrams of overlap between the identified unique peptides predicted by FragGeneScan and those predicted by Graph2Pep in the SD3, SD6 and SD7 datasets are shown in S1 Fig. Following the initial database searching, the identified unique peptides were mapped back to the assembly graph using Graph2Pro, and a total of 14,174 proteins were retrieved covering all identified peptides. To be noted here, in this step, we used the peptides of 5% FDR in order to increase the coverage of potential proteins in the sample.
Proteins generated by Graph2Pro were then used as the new target database for a second round of peptide identification using MSGF+ on the same set of MS/MS spectra in the SD3 dataset, which identified a total of 18,162 unique peptides from 73,527 PSMs, corresponding to 12.18% of the whole input set of MS/MS spectra at 1% FDR. Comparing to the conventional protein identification procedure that identified 3.06% of MS/MS spectra from the proteins predicted in the contigs, the proposed pipeline identified about four times (398%) PSMs and unique peptides (383%). In particular, the second round of database search identified 21,029 (40.06%) more PSMs and 4,234 (30.40%) more unique peptides comparing with the first round of search, indicating the second traversal of the de Bruijn graph substantially increased the coverage of the target metaproteome. Similar levels of improvement were achieved on the other two datasets (252% and 321% for SD6 and SD7 datasets, respectively).
Our results showed that using assembly graphs of metagenome also significantly improved the identification of proteins from MS/MS spectra (Table 3). We take SD3 dataset as an example. A total of 2,043 proteins (that contains one or more identified peptides) can be identified using only the contigs. Out of 2,043 proteins, there are 1,065 proteins with at least two identified peptides. We note this number is comparable to the original results reported in Muller et al. [24], which reported 1,815 identified proteins. By contrast, 13,431 proteins can be identified when the assembly graph is used, while 3,245 proteins have at least two identified peptides. We clustered the combined set of 15,474 protein sequences based on a similarity cutoff of 0.8 by using CD-HIT [50], resulting in 11,209 clusters. Only 290 out of these 10,996 clusters contain proteins identified without using assembly graph, while 9,338 protein clusters contain only proteins identified by using using the assembly graph (and thus is rescued by the graph-centric approach). Similar results were obtained on the other two datasets (SD6 and SD7).
We studied the impact of the expanded set of identified proteins by the graph-centric approaches on the downstream analysis. We focused on the functional categories of identified proteins and the metabolic pathways they are involved in.
Our graph-centric approaches enabled the identification of more proteins from the MS/MS spectra, revealing a more comprehensive functional profile of the microbial communities (with more eggNOG families identified). For the SD3 datasets, 8,706 out of 13,431 (64.82%) proteins in our expanded collection of identified proteins share similarity with eggNOG proteins, resulting in the identification of 1,206 COG families. By contrast, only 626 COG families were predicted using the 2,043 proteins identified by MS/MS spectra search against predicted proteins from contigs only (1,555 proteins share sequence similarities with eggNOG proteins). Table 4 lists the additional families predicted from our expanded collection of identified proteins, each supported by at least 10 proteins, and their annotations. Fig 3 shows the numbers of proteins in the top 20 eggNOG families with most proteins identified. Clearly, each of the functional categories is supported by considerably more proteins identified by the graph-centric approaches. We also conducted the functional analysis for the other two datasets (SD6 and SD7) and observed similar results (see S1 and S2 Tables and S2 and S3 Figs for details).
Next we show that a more comprehensive coverage of metabolic pathways can be achieved by using our extended collection of proteins identified from metaproteomics data. A total of 213, 203, 223 MetaCyc metabolic pathways were reconstructured from SD3, SD6 and SD7 datasets, respectively, when proteins predicted from contigs only were used for MS/MS spectra identification. These numbers were increased to 328, 262, 294, respectively, when additional proteins were identified by our graph-centric approaches. In addition, our expanded collection of identified proteins provide a higher coverage of the pathways. Below we show two interesting pathways to demonstrate the importance of improved protein identification.
Table 5 shows the number of enzymes we identified in the wastewater datasets that are involved in the Rubisco shunt pathway (MetaCyc ID: PWY-5723; see the diagram at http://metacyc.org/META/NEW-IMAGE?type=NIL&object=PWY-5723). The results suggest that using assembly graph helps to increase the coverage of the pathway across all three datasets, SD3, SD6 and SD7. Rubisco shunt was first found in developing embryos of Brassica napus L. (oilseed rape), in which Rubisco (ribulose 1,5-bisphosphate carboxylase/oxygenase) acts without the Calvin cycle and increases the efficiency of carbon, resulting in 20% more acetyl-CoA and 40% less loss of carbon as CO2 [51]. We found MS/MS data supporting eight out of the nine enzymes involved in the Rubisco shunt. The eight enzymes we identified are EC.2.2.1.1, EC.2.2.1.2, EC.2.7.1.19, EC.2.7.1.40, EC.4.1.1.39 (Rubisco), EC.4.2.1.11, EC.5.1.3.1, and EC.5.3.1.6. For example, we identified a putative Rubisco in the SD3 dataset. The protein (Sequence ID: Protein12587; see the sequence in the FASTA file SD3.hybrid.fgsdbgraph.protein.fasta available at our website) contains 186 amino acids, which shares 94% sequence identity with a putative Rubisco identified in an uncultured bacterium (Sequence ID: gb∣AIF32007.1) according to the NCBI BLAST search. Strikingly, only three out of these enzymes were identified in the SD3 dataset when only the contigs were used (see Table 5). The second example (Fig 4) involves 2-chlorobenzoate degradation pathway (MetaCyc ID: PWY-6221) and catechol degradation to 2-oxopent-4-enoate I pathway (MetaCyc ID: P183-PWY). Enzymes involved in the degration of 2-chlorobenzoate degradation were detected persistently in all SD3, SD6 and SD7 samples. Chlorobenzoates are a group of compounds that occur in the environment either because of their release as herbicides or as products of bacterial degradation of polychlorinated biphenyls (PCBs; classified as a persistent organic pollutant, due to their environmental toxicity [52]). The reaction that converts 2-chlorobenzoate to catechol was first identified in Burkholderia cepacia 2CBS, which was shown to be able to grow with 2-chlorobenzoate as the sole source for carbon and energy [53]. Two key functions involved in the 2-chlorobenzoate degradation, i.e., EC.1.14.12.24 and EC.1.13.11.2, were supported by identified proteins in our collection (highlighted in purple in Fig 4; both enzymes were identified in SD3 dataset by our approach, but none were identified if only contigs were used to predict reference genes; and in SD7 dataset, EC.1.13.11.2 was rescued by using assembly graph).
In this paper, we presented two algorithms (Graph2Pep and Graph2Pro) for metaproteomic data analysis based on a graph-centric approach, in which the de Bruijn graph representation of the assembly of metagenomic sequences (or of the combined set of metagenomic and metatranscriptomic sequences) is used to produce the target protein database subject to the protein identification using metaproteomic data. We tested the algorithms on the metaproteomic datasets from a wastewater study in which matched metagenomic and metatranscriptomic data were also acquired. Comparing with the conventional method where the target protein database was constructed from the proteins predicted from the assembly contigs, our graph-centric approach significantly improved the protein identifications. Notably, although in this study, we consider the trypsin as the digestion enzyme, which is used by most metaproteomics projects, our algorithms can handle data collected by using other digestion enzymes, where the users need to define a different set of amino acid residues as the cleavage sites (e.g., the glutamyl and aspartyl residues when Glu-C is used) in our programs. We also note that more proteins can be identified when the assembly of combined metagenomic and metatranscriptomic datasets is used, when both metagenomic and metatranscriptomic datasets are available.
The graph-centric approach presented here relies on the de Bruijn graph representation of the sequence assembly (either from metagenomic sequences or from the combined metagenomic and metatranscriptomic sequences). In our pipeline, we utilized the output of SOAPdenovo that contains the topology of the de Bruijn graph in addition to the contig sequences (each corresponding to an edge in the graph). Many other metagenome assembly algorithms (e.g., metaVelvet [55] and meta-IDBA [56]) are based on the data structure of de Bruijn graph, which, however do not output the graph structure explicitly. As a successor of SOAPdenovo for metagenome assembly, the MegaHIT algorithm [37] can output the de Bruijn graph topology as temporary files in FASTG format, which is designed to incorporate allelic polymorphism and assembly uncertainty in an assembly graph [57]. The recently released metaSPAdes assembler [42] adapted the core SPAdes algorithm for metagenome assembly, and also output the assembly graph in FASTG format. Our current algorithms of Graph2Pep and Graph2Pro can support the input assembly graph in FASTG format, but has not been tested for its performance using the output from the other metagenome assemblers. Here, we would like to encourage the de Bruijn graph based assembly algorithms to allow users to generate explicit output of de Bruijn graphs (e.g., in FASTG format) that will be valuable for downstream analysis (such as the metatranscriptomic and metaproteomic analysis guided by metagenome assembly, as presented here).
Our graph-centric approaches are shown to be effective for improving the protein identification from metaproteomic MS/MS data. However, considering the fact that complex microbial communities contain hundreds or even thousands of species with highly uneven abundances, it will be both experimentally and computationally challenging to detect all proteins produced by the species, especially the proteins produced by the rare species in the community.
The ultimate goal of metaproteomics is not only to identify proteins expressed in the microbial community, but also to estimate their abundances (i.e., their expression levels) under different conditions. Nevertheless, a protein can be quantified only if it can be identified by using the metaproteomic data. Therefore, the methods presented here that increase the coverage of protein identification will also help the subsequent steps for protein quantification. We plan to implement the functionality of protein quantification based on label-free quantification approaches in the future release of our software.
|
10.1371/journal.pbio.1001184 | Structure and Evolution of Streptomyces Interaction Networks in Soil and In Silico | Soil grains harbor an astonishing diversity of Streptomyces strains producing diverse secondary metabolites. However, it is not understood how this genotypic and chemical diversity is ecologically maintained. While secondary metabolites are known to mediate signaling and warfare among strains, no systematic measurement of the resulting interaction networks has been available. We developed a high-throughput platform to measure all pairwise interactions among 64 Streptomyces strains isolated from several individual grains of soil. We acquired more than 10,000 time-lapse movies of colony development of each isolate on media containing compounds produced by each of the other isolates. We observed a rich set of such sender-receiver interactions, including inhibition and promotion of growth and aerial mycelium formation. The probability that two random isolates interact is balanced; it is neither close to zero nor one. The interactions are not random: the distribution of the number of interactions per sender is bimodal and there is enrichment for reciprocity—if strain A inhibits or promotes B, it is likely that B also inhibits or promotes A. Such reciprocity is further enriched in strains derived from the same soil grain, suggesting that it may be a property of coexisting communities. Interactions appear to evolve rapidly: isolates with identical 16S rRNA sequences can have very different interaction patterns. A simple eco-evolutionary model of bacteria interacting through antibiotic production shows how fast evolution of production and resistance can lead to the observed statistical properties of the network. In the model, communities are evolutionarily unstable—they are constantly being invaded by strains with new sets of interactions. This combination of experimental and theoretical observations suggests that diverse Streptomyces communities do not represent a stable ecological state but an intrinsically dynamic eco-evolutionary phenomenon.
| Soil harbors a diverse spectrum of bacteria that secrete small molecules such as antibiotics. Streptomyces bacteria, considered the most prolific producers, have been mined for decades for novel products with therapeutic applications, yet little is known about the properties of the interaction networks these compounds mediate. These networks can hold clues about how the diversity of small molecules and of Streptomyces strains with different production and resistance capabilities is maintained and promoted. To explore the network properties, we developed a high-throughput platform for measuring pairwise phenotypic interactions mediated by secreted metabolites, and used it to measure the interaction network among 64 random Streptomyces isolates from several grains of soil. We found many strong but specific interactions that are on average determined more by metabolite production than by metabolite sensitivity. We found reciprocity between strains, whereby if one strain inhibits or promotes the growth of a second strain, it's likely that the second strain affects the first strain in a similar manner. These interactions are not correlated with phylogeny, as very closely related strains exhibit different interaction patterns. We could explain these findings with a mathematical model requiring interplay between ecological dynamics and evolution of antibiotic production and resistance, suggesting that the bacterial and small molecule diversity of these communities is maintained by constant evolutionary turnover of interaction phenotypes.
| Sampling DNA from diverse ecosystems has revealed a breathtaking diversity of microbial life [1],[2], especially in soil [3]–[5]. But we have barely begun to explore, both experimentally and theoretically, how these complex communities coexist and function. We know that microbes can interact via secretion of a wide array of small molecules, most notably antibiotic compounds. But how prevalent, diverse, and specific are such interactions? How is the incredible diversity of microbes and their natural products maintained and promoted by complex and spatially structured networks of interactions?
To tackle these questions, we isolated bacterial strains from individual grains of soil and systematically measured all pair-wise interactions among them. We measured compound-mediated interactions, where a “sender” strain affects a “receiver” strain by secreting metabolites, antibiotics, or other compounds (Figure 1AB). We focused on bacteria from the genus Streptomyces, which are the most prolific producers of small molecules, are abundant in soil [6], and exhibit diverse production and resistance capabilities [6]–[8] that are modular and prone to Horizontal Gene Transfer (HGT) [9]. Sixty-four Streptomyces from four individual grains of soil were isolated (Figure 1C), phenotyped for all possible pair-wise interactions, and genotyped for 16S rRNA. We explored the statistical properties of the resulting network and juxtaposed them with those emerging from a simple ecological model of bacteria evolving production of and resistance to antibiotics [10],[11].
We developed a high-throughput platform for measuring directional pairwise interactions by observing how the products of one bacterial strain affect the colony growth of another. A fine-pored filter is placed on a nutrient agar surface, a lawn of the sender strain is grown on top, and the filter is removed—leaving behind sterile agar that has been altered or conditioned by the sender strain. The conditioned agar is then resupplied with concentrated liquid nutrients to compensate for the nutrients consumed by the donor. A receiver strain is point-inoculated onto the sterile conditioned agar and time-lapse movie of the growing colony is taken. A high-throughput implementation of this assay allowed us to acquire 11,500 movies along 15 d at 4 h time resolution, covering all pairwise interactions within our collection of 64 strains in duplicate (Figure 1D, Materials and Methods). By comparing colony growth of the receiver strains on conditioned and non-conditioned agar, we identify interactions between strains. We quantified the first time point in which each colony becomes visible on the images (appearance time) to identify growth inhibitory and growth promotion interactions. In addition, we visually scored instances of inhibition of aerial mycelium formation (and subsequent sporulation).
We found a rich and complex interaction matrix among our collection of strains with multiple cases of growth inhibition, growth enhancement, and inhibition of aerial mycelium (Figure 2A; and see also all the 11,500 time lapse movies underlying this matrix in Figure S8). This matrix showed two immediately apparent special properties: it is “balanced” and it is “sender determined,” as we now explain.
The first noteworthy property of the matrix is that the frequency of interactions (a.k.a. connectance) is balanced: the probability that two random isolates interact is neither close to zero nor close to one. While there are many strong interactions, the matrix is far from the limit in which interactions are non-specific because everyone interacts with everyone else: of the 64 isolates, there are at least 42 different interaction profiles. We found 45% of growth or aerial mycelium inhibitory interactions (25% complete inhibitions of growth) and 19% growth promotion (see Materials and Methods for additional details). The frequency of inhibitory interactions is significantly higher (and more balanced) than previous estimates based on zones of inhibition [12]. The balanced frequency of interactions makes this network more highly connected than most known ecological networks (only for a few food webs the density approaches 30%) [13]–[16].
The second striking property of the measured matrix is that it is very different along the sender and receiver axes, with characteristic stripes of inhibition and non-inhibition running along the receiver (vertical) axis. This asymmetry is surprising because the existence of an inhibitory interaction is, in general, influenced by both the sender, which needs to produce a toxic compound, and the receiver, which needs to be sensitive to the compound produced. In the extremes, a matrix that is determined purely by the properties of the sender would exhibit perfect stripes in the vertical direction, while a receiver-determined matrix would exhibit stripes along the horizontal (sender) axis. Thus, the network we observed is more sender-determined. This sender-receiver asymmetry can be quantified by comparing the distribution of the fraction of isolates that each isolate inhibits (sender degree) with the distribution for the fraction of isolates that inhibit each isolate (receiver degree) (see Figure 2B). The sender degree is broad and peaks near its extreme values, while the receiver degree is narrower and unimodal. The difference of the variances of the receiver and sender distributions is a measure of the sender-receiver asymmetry (Q = −0.37; Figure 2C). The negative value of this quantity means that information gathered about a sender from a few interactions let us predict far better the rest of its interactions than the corresponding information about a receiver. The sender-receiver asymmetry is pronounced but not extreme, indicating the importance of resistance to antibiotics that strains do not themselves produce. The bimodality of the sender degree distribution makes this network very different from networks with nodes randomly connected with a fixed probability (Erdös–Renyi random graphs), scale-free (social) networks [13], and food webs with exponential-tailed distributions [17].
It is unclear how coexistence between strains that inhibit almost everyone and strains that inhibit almost no one is maintained. One possibility is the presence of an ecological tradeoff between ability to inhibit and ability to resist, which would imply a positive correlation between the sender and receiver degrees. But no such correlation exists; on the contrary, isolates that inhibit most are also among the most resistant (Figure S1A, p<10−4). There is also no correlation between growth rate on non-conditioned media and the sender or receiver degree (Figure S1B).
We decided to look for hints about the maintenance of a diverse sender-determined network in the network evolution. We sequenced the 16S rRNA of all isolates and found that closely related isolates are less likely to inhibit each other (Figure 3B), but there is a poor overall correlation between phenotypic and phylogenetic distances (Figure 3A,C). Even isolates with identical 16S rRNA sequences can have very different interaction profiles. This lack of strong correlation between phylogeny and inhibition profiles is consistent with previous work [18]. To further exploit this phylogenetic signal, we compared the phenotypic divergence of sender and receiver profiles for isolates with the same 16S, and contrasted it with the null expectation of isolates with different 16S (Figure 3D). Interestingly, the sender profiles diverge disproportionately more than the receiver profiles for closely related strains even after controlling for the overall sender-dominated nature of the matrix (P = 2⋅10−4). So it seems that the Streptomyces community is in a state in which frequent evolutionary changes in production (mediated for example by transfer of plasmids carrying antibiotic production genes) cause dramatic changes to ecological interactions. The coupling between ecology and evolution is therefore important for understanding the network properties.
Is the balanced frequency of interactions and sender-determined nature accidental or a natural outcome of the ecological and evolutionary dynamics of interacting Streptomyces communities? Can we account for the large changes in interaction patterns over short evolutionary distances?
We consider a simple in silico model of communities of strains producing and resisting a set of antibiotics. A strain inhibits another if it produces at least one antibiotic to which the other is sensitive. Communities of strains with randomly assigned production of and resistance to antibiotics exhibit a diverse set of qualitatively different interaction matrices, depending on the frequency of production and resistance (Figure 4A). With many antibiotics, matrices similar to the observed—balanced frequency of interactions and moderately sender-determined—occupy a small region of the parameter space, and require low frequency of production and high frequency of resistance. This raises the question of whether introducing evolution into the model can inherently direct it into the regime of balanced and sender-determined interactions.
We imposed simple non-spatial ecological dynamics that implicitly incorporates the importance of spatial relations between bacteria over short time scales (the antibiotics stay near their producers). The fitness of each strain depends on the weighted sum of its interactions with all other strains, incorporating the following contributions: (i) a negative effect of being inhibited by others, (ii) an advantage of inhibiting others, and (iii) a reduced ability to inhibit if being inhibited (protection by inhibition). A cost for production or resistance of any antibiotic is also added. The resulting mathematical structure is that of a discrete time Lotka-Volterra model with coefficients derived from the pairwise interaction matrix (Figure 4B and Materials and Methods).
For simplicity, the model ignores that antibiotics might also function as a “common good” reducing competition from non-resistant Streptomyces and non-Streptomyces strains. In addition, it ignores the possibility of resistant neighbors extending their protection to non-resistant strains [19]. We also ignore positive interactions and primary metabolism differences (utilization of different resources and cross-feeding), which are potentially important. Some of these effects can be incorporated by adding terms with higher order interactions (for several model extensions, see Materials and Methods and Figure S7).
To capture the long-term effects of the interplay between ecology and evolution on the statistical properties of interactions, we added mutations to the above model. Mutations allow acquisition or loss of production and resistance to any of the antibiotics. Turnover of production and resistance capabilities is indeed expected to be important for Streptomyces, as evidenced by the modular nature of antibiotic production and the vectors through which it spreads.
The simulation starts from a single strain that is sensitive to all antibiotics and follows the dynamics until a statistical steady state is reached (Figure 4C). We systematically explored the behavior of the model for a range of costs of production and resistance (Figure 4D). The results show a maximum cost of production above which no antibiotics are produced (Figure 4D, white area). Strikingly, below this threshold we see balanced sender-determined matrices (Figure 4D, green shades), as long as the production costs are higher than the resistance costs (above the dashed blue line). There is an inherent feedback that keeps the frequency of interactions from becoming too low or too high: an increase of interaction frequency selects for an increase in resistance levels, which then leads to a decrease of the interaction frequency. This qualitative picture holds provided that the level of protection by inhibition is below a certain threshold (Figures S2, S3A). On the other hand, if inhibition is an effective defense, then when resistance cost is high the system collapses into a state in which most strains inhibit each other (Figures S2, S3A). While different outcomes are possible in the model, we observe balanced and sender-determined matrices over a large region of the parameter space (Figure S3).
We also explored the relation between interaction and phylogeny in the simulations. In agreement with our experimental observation, in the balanced and sender-determined region, we find that in the resulting interaction matrices (Figure S4A) strains are more likely to interact when they are phylogenetically distant (Figure S4B), and there is a weak overall correlation between phylogenetic and phenotypic distance (Figure S4C).
Community diversity requires both ecology and evolution. The functional diversity of the system increases sharply with both the evolutionary rate and the population size, and turning off the ecological interactions or reducing the mutation rate leads to a loss of diversity (Figure 4E and Figure S5). The community steady state is characterized by a continuous turnover of different interaction phenotypes (Figure S4D), indicating its evolutionary instability.
To investigate statistical properties beyond those captured by the degree distributions, we followed an established procedure for identifying interaction motifs—local patterns of interactions that are more frequent than expected by chance [20]. We discovered that the simulated interaction networks are strongly enriched for mutual inhibition when compared with random networks with the same sender and receiver degrees for each isolate (Figure 5A). This is not surprising since mutual inhibition is an important mechanism for ecological balance. We, therefore, looked for reciprocity in the experimental data. In the experimental data, unlike the model, there is an extra complexity due to positive interactions (growth promotions are not included in the model). With both positive and negative interactions there are six two-isolate motifs (Figure 5B,C). We compared the six motif frequencies with those for random matrices that have the same sender and receiver degrees for each isolate, and which preserve the corresponding degrees for the growth promotion interactions. Since an obvious source of reciprocity structure is the presence of identical isolates, we excluded strains with identical 16S and interaction profiles from this analysis. As we observe in the model, the analysis of the experimental data revealed statistically significant enrichment for reciprocal interactions—there are more mutual inhibitory interactions and mutual growth promotions than expected and fewer asymmetric relationships (Figure 5B).
If reciprocity of interactions among pairs of strains is a property of coexisting communities, we may expect that it will be more enriched in strains coming from the same soil grain than for strains isolated from different grains. We found that while the frequency and strength of positive and negative interactions does not differ within and between grains, interactions of pairs of strains within grains do indeed tend to be more reciprocal than interaction of strains from different grains (Figure 5C). This result is significant only if we include the inhibitions of aerial mycelium. The motif distributions are also sensitive to the choice of thresholds for defining interactions. A threshold independent analysis of the continuous data shows again enrichment for reciprocity (Figure S6, p = 0.001). The apparent enrichment for reciprocity remains if we control for a tendency to have isolates with more similar 16S within a grain. A larger dataset will be required to distinguish between different underlying causes for the patterns of interactions within and between soil grains.
We find that Streptomyces isolates from soil grains exhibit diverse and rich interaction patterns. The interaction matrix they form has a balanced frequency of interactions—the probability that two random strains interact is neither close to zero nor to one. The sender-degree distribution is broad and bimodal—isolates tend to inhibit almost everyone or almost no one, which makes the interactions statistically controlled more by the properties of the sender than the receiver. This sender-receiver asymmetry, while pronounced, is not extreme, indicating the importance of resistance to compounds produced by others. These properties make this network very different from other ecological networks, which have monotonic degree distributions, and typically exhibit much lower interaction frequency. Finally, the community is enriched in reciprocal interactions—interaction pairs are enriched in mutual inhibitory interactions and mutual growth promotions, while it is rare to find cases in which one strain promotes a second, but this second strain inhibits the first. This reciprocity is further enriched among strains derived from the same grain of soil, thus revealing spatial structuring of interactions.
These properties of the interaction network have emerged from a long evolutionary process, which we probed by juxtaposing interactions and phylogeny. We found that the interactions of an isolate can change dramatically even over short evolutionary time (indicated by very close 16S sequences), with evolution changing the production profiles more than the resistance profiles. Incorporating such fast evolution in a dynamic ecological model of antibiotic interactions, we find that most of the observed properties of the network are reproduced under a broad range of parameters. The community compositions are not static—increase in production of an antibiotic promotes resistance, which promotes sensitivity, and invites production again. As the community undergoes cycles with respect to different antibiotics, different combinations of production and resistance become favorable, which makes it evolutionary unstable. In our model both ecological interactions and continuous turnover of interaction phenotypes are required to maintain functional diversity.
Our work has several important limitations. Perhaps the main one is that interactions are measured in the lab and actual interactions in the soil may be more complex or different. We were also limited to studying only the interactions among Streptomyces strains; interactions between Streptomyces and other microbes could be of major importance. Higher order interactions, such as synergy or antagonism between natural products, or induction of small molecule production by other small molecules, are not captured by the pairwise measurements. Many of these shortcomings are inherent to most current studies of microbial species interactions. However, the systematic and high-throughput nature of the current study allows us to ask questions at the statistical level, and might therefore be less prone to some of these difficulties. Furthermore, the high-throughput interaction platform developed here and the simulations offer a natural foundation for many subsequent studies of microbial communities, which will address some of the above concerns, potentially yielding important biological insights. For example, it is now possible to probe how the statistical properties of networks, such as the relative significance of positive and negative interactions, are affected by media composition and the presence of other small molecules. This enables investigations of the regulatory roles of and epistatic effects between small molecules. It would also be interesting to see whether the effects of a sender on a receiver will be modified if the sender is co-incubated with the receiver. Finally, the interaction platform can be used to follow the evolutionary and ecological dynamics of synthetic laboratory communities of interacting microbial strains.
The observed network properties do not seem to correspond to an ecologically stable state maintained by antibiotic interactions alone. Instead, the model and observations suggest that they are supported by a constant evolutionary change. The distribution of production and resistance in the community is poised so that simple changes in production capabilities of a strain can alter its interactions with many other strains potentially to a great ecological advantage. This evolutionarily unstable ecological state seems complemented by the modular nature of the secondary metabolite gene clusters, which enable such changes and, thus, lead to turnover of interaction phenotypes of different strains and species. This continuous turnover might in turn be important for the emergence and maintenance of the modularity and clustering of small molecule production and resistance genes and their recruitment to mobile genetic elements [21]. This reasoning suggests a unified view of network structure, network evolution, and modularity of secondary metabolism to be further explored.
We sampled four soil grains of soil by touching the soil with a dry needle tip, and lifting particles of less than 1 mg of wet weight. Three of the grains were 1 cm away from each other in one soil core, and the fourth grain was 10 cm away from a second soil core. The depth was approximately 2 cm below the surface. The sampling was performed in December from foliage-covered soil away from visible roots. Each grain was dried for 2 d, then suspended in dH2O, vortexed, sonicated, diluted, and plated on Streptomyces Isolation Media [7]. Plates containing five colonies or fewer were sampled in order to minimize potentially biasing interactions between emerging colonies. Isolates that exhibited the characteristic aerial mycelium pattern of Streptomyces were selected at random after 2 wk, and their genus identity later verified by sequencing. Five of the isolates were classified as genus Kitasatospora within the family Streptomycetaceae by the Ribosome Database Project [22]. Each isolate was restreaked once, then grown in TSB for 3 d, and 300 µl/plate was spread on four petri dishes containing Bennett's agar [7]. Plates were incubated for 14 d at 28°C. Spore lawns were harvested in 12 ml of 0.01% Tween 80, vortexed for 2 min, and filtered through 5 µm syringe filter to separate the spores from mycelium. The filtrate was centrifuged at 1,000 g for 10 min, and the spore pellet was resuspended in 1.1 ml of 20% glycerol, aliquoted, and frozen at −80°C. Each spore stock that we used was thawed only once. During stock preparation, tubes were kept on ice.
Bulk soil was sent to the Soil and Plant Tissue Testing Lab at the University of Massachusetts at Amherst. The soil pH is 5.5. The texture is loam with 46.7% sand, 42.1% silt, and 11.2% clay. Organic matter, 12.6%. NO3-N, 0 ppm. Mineral content: P, 7 ppm; K, 230 ppm; Ca, 1,511 ppm; Mg, 157 ppm. Micronutrients: B, 0.3 ppm; Mn, 7.1 ppm; Zn, 9.3 ppm; Cu, 0.3 ppm; Fe, 32.4 ppm; S, 28.8 ppm. Cation Exch Cap, 21.7 Meq/100 g.
Media for interactions: 15 g purified agar in 1 L d H2O, 2 g potato starch, 0.8 g casein, 1 g KNO3, 0.4 g K2HPO4, 0.2 g MgSO4, 30 mg CaCl2·2H2O, pH 7.2. All components were autoclaved separately in concentrated form, and all agar plates were made from the same autoclaved stocks. Resupply media was 18× concentrated interaction media with the exception of KHPO4, which was 36× concentrated, pH 7.0.
Black 96-well agar plates were robotically over-filled with agar, and before solidification a glass plate was lowered to 1.5 mm above the plate to flatten the agar meniscus. The glass plate was slid sideways upon solidification of the agar. The resulting agar columns were flat on top (to ensure good filter contact and high image quality), protruded above the edge of the plate (to ensure good contact with filter during conditioning), and well separated from neighboring wells (to prevent cross-talk). Since high pipetting accuracy was required, the aspirated amount was automatically adjusted based on the instantaneous agar temperature (∼50°C), care was taken to dip the pipette tips to the same depth in the agar reservoir, and room ventilation was turned off to prevent asymmetric cooling of the agar in the tips. Rectangular filters—polycarbonate 0.03 µm pore size—were placed over the agar plates. Each well was inoculated with 8 µl of spore stock. Due to the hydrophobicity of the filters, droplets above neighboring wells were well separated.
After 8 d of incubation, growth on each filter was imaged, and the filter removed. Filter images were used to discard data from defectively conditioned or contaminated wells. Plates were resupplied with a 20 µl droplet of resupply media, and dried in a fume hood for 90 min. Each plate was pinned from a source 96-well plate containing 100 µl/well of spore stock (∼107 spores/ml). Source plates were kept between 4 and 8°C during pinning. Pins were sterilized between plates to prevent contamination of the source plate due to accidental contamination of the agar plates. The time of pinning of each plate was recorded, and it was placed upside down on a flatbed scanner so that the agar surface is 2 mm away from the scanner glass surface. The focusing plane of the scanners was correspondingly adjusted. To minimize agar drying, plates were sealed to the scanners with packing tape. Colonies were scanned approximately every 4 h. Temperature was maintained at 28°C, but jumped temporarily by 1°C after each scan. Plates were scanned for at least 15 d.
The appearance time for each colony (the first time point at which a colony becomes visible on the images) was manually determined using custom interactive software. Colonies associated with agar defects or contaminations were discarded. Aerial mycelium is apparent on the images as a fuzzy texture on top of the colonies (Figure 1D). Aerial mycelium inhibition was scored if there was no or very little (in comparison to non-conditioned) aerial mycelium coverage of the colony after 15 d.
Twenty-five percent of the interactions are complete inhibitions, i.e. no visible growth of receiver colonies. One isolate inhibits itself. An additional 10% of interactions are partial inhibitions with colonies appearing at least 1 d later on conditioned media (for a total of 35%). The fraction of inhibitory interactions is 45%, if inhibitions of aerial mycelium formation are included.
Colonies were grown in TSB for 3 d, centrifuged at 1,000 g for 10 min, and resuspended in dH20 three times. Cells were then resuspended in lyses buffed (PrepMan) and heated to 100°C for 10 min, centrifuged, and the supernatant was frozen at −20°C. 3 µl of this supernatant was added to 60 µl PCR mix containing 12 µl Qiagen Q-solution, 2.4 µl of 10 µM forward primer GAG AGT TTG ATC CTG GCT CAG, and reverse primer CGG CTA CCT TGT TAC GAC TTC. Samples were PCR amplified (95°C for 3 min, 35 cycles of 95°C for 1 min, 55°C for 1 min, 72°C for 1:30 min, and final extension at 72 deg for 7 min), and PCR products were sent for sequencing upon confirmation of existence of a product of the expected size (∼1.5 kb). Sequences from the forward and reverse primers had a significant overlap. Sequences are available through Genbank, accession numbers: JN020489–JN020551. The grain and isolate number within a grain is specified in the description for each sequence; e.g. G4_6 is the sixth isolate from grain four.
We considered the profiles of two isolates distinct if they differed by more than 2 d in appearance time (the first time point in which a colony becomes visible on the images) for both replicates in at least three sender or three receiver positions. According to this measure, there are 42 distinct phenotypic profiles.
For a N×N binary interaction matrix (one indicates an interaction, and zero no interaction), the frequency of interactions is , and the sender-receiver asymmetry is defined as . Matrices with negative Q are sender-determined, and with positive Q are receiver-determined. We obtain negative Q independently of how we threshold the inhibitory interactions and of whether or not we include aerial mycelium inhibitions.
The fraction of differences between profiles was calculated (after discarding defective and inconsistent replicas). The profiles were taken from a binary interaction matrix in which inhibitions were defined as delays in colony appearance time of more than 1 d. Increasing the threshold to 3 d (i.e. strong inhibitions) did not change the qualitative findings of Figure 3. However, inclusion of aerial mycelium inhibitions renders the statistics of Figure 3D insignificant.
Sequences were aligned to a universal 16S rRNA template using the Ribsomal Database Project website [22]. For each pair of sequences, only positions for which both sequences have high-quality values from the sequencing trace were considered; the rest were treated as missing values. Phylogenetic distance was computed as the fraction of differences (in high-quality positions). Alignment gaps were counted as normal differences.
For the measured network of positive and negative interactions (without weights), we generated an ensemble of random networks that have the same number of ingoing and outgoing arrows of positive and negative interactions for each isolate. Networks were randomized by taking random pairs of single arrows (between different isolates) and swapping the isolates on which they end, provided the two arrows created by the swap do not exist already or correspond to missing or defective experimental values. (In this way, the missing or defective values of the matrix were kept in place.) Each cycle consisted of swapping one pair of positive and one pair of negative arrows. This operation was performed thousands of times before selecting each random ensemble representative. For each random network the frequency of each of the six pairwise motifs was calculated, without counting any of the diagonal matrix elements. The motif significance (p value) was calculated as the fraction of random networks that have more extreme motif frequency than that for the observed network. The protocol was analogous for the matrices resulting from the eco-evolutionary model, which had only negative interactions and no missing or defective values.
Ecology: Each strain i is characterized by an array Ziα, specifying whether it is producer (P), resistant (R), or sensitive (S) to antibiotic α. Let Ai←j be the binary matrix of inhibitory interactions. Strain j inhibits strain i, i.e. Ai←j = 1 , if Zjα = P and Ziα = S for any α. Let ni be the fractional abundances of different strains (summing to one). The “fitness” of i is , where . At each time step N individuals are drawn from different species with relative probafobilities . λ and η are (positive) ecological parameters controlling the direct benefit of inhibiting neighbors and the consequence of mutual inhibition (the level of protection by inhibition is 1−η), and is the intensity of selection within an ecological cycle (which we specify through by ). λ = 0 means that inhibition is a zero-sum game; in the other extreme λ = 1 means complete spite (no direct benefit for the inhibitor). Production or resistance of an antibiotic incurs a multiplicative fitness cost, so that is the bare fitness reflecting the costs and of production and resistance (antibiotic dependent), and δ is the Kronecker delta. Evolution: each antibiotic position in each of the N individuals mutates within the SRP space of possibilities with probability specified by a set of transition rates: , , , , , .
N = 106, = 0.05, λ = 0.15, η = 0.7, 40 antibiotics. , , , and . The relative mutation rates assume that loss of function is more likely than gain of function, and gain of resistance is easier than gain of production. The no interaction case in Figure 4E corresponds to = 0. Figures S2 and S3 explore the behavior of the model for other parameters.
We examined the behavior of the model when different antibiotics have different production costs rather than identical costs. The production costs were uniformly distributed in the interval ranging from the resistance cost up to the maximal cost for which a producer can invade a sensitive strain. We discovered that this extends the region over which we observed balanced interaction matrices, and leads to receiver-determined matrices at large resistance costs (Figure S7A). We also added an evolutionary operator that mimics more closely within-population HGT—rates of change towards production and resistance of an antibiotic are proportional to the abundance of production and resistance to that antibiotic in the population (rather than being constants). With probability of 10−4 an organism pairs with another random organism and gains a production or resistance for an arbitrary antibiotic of the donor. Adding within-population HGT (while keeping the mutations) did not qualitatively change the results (Figure S7B).
|
10.1371/journal.pgen.1005392 | Molecular Clock of Neutral Mutations in a Fitness-Increasing Evolutionary Process | The molecular clock of neutral mutations, which represents linear mutation fixation over generations, is theoretically explained by genetic drift in fitness-steady evolution or hitchhiking in adaptive evolution. The present study is the first experimental demonstration for the molecular clock of neutral mutations in a fitness-increasing evolutionary process. The dynamics of genome mutation fixation in the thermal adaptive evolution of Escherichia coli were evaluated in a prolonged evolution experiment in duplicated lineages. The cells from the continuously fitness-increasing evolutionary process were subjected to genome sequencing and analyzed at both the population and single-colony levels. Although the dynamics of genome mutation fixation were complicated by the combination of the stochastic appearance of adaptive mutations and clonal interference, the mutation fixation in the population was simply linear over generations. Each genome in the population accumulated 1.6 synonymous and 3.1 non-synonymous neutral mutations, on average, by the spontaneous mutation accumulation rate, while only a single genome in the population occasionally acquired an adaptive mutation. The neutral mutations that preexisted on the single genome hitchhiked on the domination of the adaptive mutation. The successive fixation processes of the 128 mutations demonstrated that hitchhiking and not genetic drift were responsible for the coincidence of the spontaneous mutation accumulation rate in the genome with the fixation rate of neutral mutations in the population. The molecular clock of neutral mutations to the fitness-increasing evolution suggests that the numerous neutral mutations observed in molecular phylogenetic trees may not always have been fixed in fitness-steady evolution but in adaptive evolution.
| Mutations that have little influence on biological function are referred to as neutral mutations and frequently appear in molecular phylogenetic analyses. The fixation of neutral mutations in populations has been attributed to genetic drift in fitness-steady evolutionary processes or hitchhiking in adaptive evolution. We examined the fitness-increasing evolution of Escherichia coli for thermal adaptation to observe the fixation dynamics of genome-wide mutations. In the adaptive evolution, all genomes in the population equally accumulated neutral mutations by replication errors. The infrequent occurrence of an adaptive mutation on one of the genomes by chance resulted in the fixation of the neutral mutations that had pre-accumulated in the same genome by hitchhiking. Via successive hitchhiking events, the neutral mutations were fixed in the population linearly over generations at the same rate as the spontaneous mutation accumulation rate in the genome. The molecular clock of neutral mutations thus functions even in adaptive evolution. The evolutionary period characterized by the accumulation of numerous neutral mutations observed in molecular phylogenetic trees may not be specific to neutral evolution but may occur in adaptive evolution as well.
| The dynamics of adaptive evolution are more intricate than a simple sum of mutation and selection due to the entanglement of several evolutionary events [1], which include rare adaptive mutations [2,3,4,5,6,7,8], epistasis [9,10,11] and hitchhiking [12,13,14] at the genome level and clonal interference [15], frequency-dependent selection [16] and genetic drift [17] at the population level. Recent genomic sequencing analyses of yeast evolution experiment have confirmed that the rise and fall of adaptive genotypes are complicated due to the concurrence of hitchhiking and clonal interference [14].
Despite these recently uncovered complicated dynamics, a constant mutation fixation rate in a population is considered a simple rule in phylogenetic analysis. Based on fossil records, the molecular clock, which was first proposed in the 1960s, suggests that mutations accumulate over time [18]. Intensive studies largely developed the concept of molecular clock for broader applications [19,20,21]. Although the constancy of the mutation fixation rate may vary to some extent, the molecular clock has become a simple tool for evolutionary researchers to convert mutational differences into phylogenetic trees to trace past evolutionary events [19,22].
There are two possible mechanisms for the molecular clock. The first mechanism is the genetic drift of neutral mutations, also known as neutral evolution [23]. Mutations with no fitness contribution can drift to the majority by stochastic effects due to a limited population size. Because mutations occur at neutral sites at the spontaneous mutation accumulation rate and propagate equally, the mutational difference among independent populations derived from a common ancestor increases over generations in a clockwise manner at the spontaneous mutation accumulation rate. In selective environments, neutral mutations drifting toward fixation are swept out by adaptive mutations. The other mechanism is hitchhiking, in which neutral mutations accumulating on single genomes are fixed through the propagation of adaptive mutations that occurred last on the same genomes. No matter which of the genomes in a population gain an adaptive mutation, the number of neutral mutations that take hitchhiking does not vary greatly because all the genomes accumulate neutral mutations equally at the same spontaneous mutation accumulation rate, setting the molecular clock for neutral mutations [24,25,26] Thus, the fixation rate of neutral mutations by hitchhiking is expected to be identical to the spontaneous mutation accumulation rate, which is the same as the fixation rate by genetic drift. The main difference between the two mechanisms in neutral mutation fixation is the time scale. The mutation fixation that mediated by genetic drift requires much longer time than that mediated by hitchhiking. Theoretically, the number of generations is close to the population size in genetic drift, but is as small as the inverse of selection coefficient in hitchhiking [27].
While both genetic drift and hitchhiking occur in laboratory population dynamics, direct evidence that these mechanisms fix neutral mutations in proportion to the number of generations has not been obtained. Pioneering experimental population dynamics studies on fruit fly demonstrated that genetic drift fixes neutral alleles [28], and many studies have demonstrated that a substantial fraction of the mutations found in phylogenetic trees and in natural populations are attributable to fixation by genetic drift [27]. However, the process by which neutral mutations are successively fixed by genetic drift has not directly been observed. While hitchhiking has been observed in experimental populations [12,13,14], experimental evidence supporting a role of successive occurrences of hitchhiking in the linear fixation of neutral mutations over generations is limited. If hitchhiking fixes neutral mutations at a rate that is equal to the spontaneous mutation accumulation rate, the concept of a molecular clock could be extended to fitness-increasing evolution, which is not covered by the neutral theory of molecular evolution of Kimura.
Here, we report the repeated appearance of hitchhiking in a severe selective environment, leading to clock-like molecular evolution of neutral mutations. We previously conducted an evolution experiment in which E. coli was thermally adapted to 44.8°C [29]. Genome-wide analysis over the course of this evolution revealed that the majority of the accumulated mutations shifted from a positive to neutral fitness contribution. In the present study, we duplicated and prolonged the previous evolution experiment for further thermal adaptation to 46°C. The dynamic process of the genome-wide mutation fixation between the generation of the previous evolution, at which the fitness contribution of mutations shifted from positive to neutral, to the final generations of the duplicated evolutionary lines revealed that hitchhiking was responsible for the constant fixation rate of neutral mutations, even though the population dynamics of the genotypes were complicated due to the stochastic concurrence of adaptive mutations and clonal interference. The possibility that at least some of the neutral mutations that accumulated in the past were fixed by hitchhiking in adaptive evolution is discussed.
To investigate the dynamics of genomic molecular evolution in a selective environment, we prolonged the previous thermal evolution experiment in which an E. coli cell population evolved to grow exponentially at 44.8°C [29]. The final population at 7580 generations was duplicated (Line1 and Line2) and challenged at the 0.2°C increment (Fig 1a). The increment was repeated whenever the exponential growth rate exceeded 0.4/h for two consecutive days until the cells of the duplicated lines were able to grow at 46.0°C. In terms of fitness increase, the selection pressure in the prolonged thermal adaptation evolution with the sequential increments of 0.2°C (from 7581 to 8382, and 7581 to 8829 generations) was equivalent to that in the previous report from 5212 to 7580 generation at 44.8°C [29], because the rates of fitness increase (the increase in growth rate per generation) were both at the order of 10−4, which was one order higher than the fitness increase (~10−5) observed in the nearly static phase at 37°C [29,30].
Although the reduction and recovery of the growth rate after every temperature increment were basically the same for the two lines, Line1 required more generations to reach 46.0°C than did Line2. The temperature dependence of the growth rate revealed that the final population of Line1 but not Line2 was able to grow at 0.5°C higher than the final temperature of 46.0°C (Fig 1b). The two lines, which originated from the same population and were subjected to the same rules for temperature increment, evolved to develop different thermal tolerances at different speeds.
Whole-genome mutation analysis of the final populations of Line1 and Line2 revealed difference in the number of accumulated mutations. A next-generation sequencer (Applied Biosystems SOLiD 3 system) was used to estimate the ratio of substituted bases to original bases in the two populations. We conducted Sanger sequencing to confirm possible single nucleotide polymorphism (SNP) candidates based on SOLiD data, except those SNPs in rRNA and tRNA. The ratios of SNPs determined from the SOLiD data were consistent with those determined by Sanger sequencing. The standard deviation of the ratio of SNPs was 6.2% for Sanger sequencing (S1 Fig). The proportion based on Sanger sequencing was used in the subsequent analysis for a precise and costless evaluation. In addition to the mutations that were identified in the final generation in the previous study, 82 new substitutions were confirmed, of which 39 and 20 were exclusive to Line1 and Line2, respectively (S1 Table). The greater number of mutations in Line1 compared to Line2 may be attributable to the greater number of generations of Line1 compared to Line2.
SNP temporal change from generation 5212 of the previous evolution experiment to the final generations of Line1 and Line2 revealed that the mutations were clustered in fixation dynamics, as predicted by hitchhiking. We determined the fixation processes of 46 mutations that appeared between generations 5212 and 7580 and the newly discovered 82 mutations with the estimated ratios of substituted bases to original bases by Sanger sequencing as previously described. The estimated frequencies (%) of the mutations for the intermediate and two final populations were listed in S2 Table. The 5212th generation is the time point after which the spontaneous mutation accumulation rate accelerated. Several mutations simultaneously appeared over the detection limit of 5% for Sanger sequencing and proceeded together toward fixation (Fig 2).
Based on the concurrent frequency changes toward fixation, we clustered the 128 mutations into 22 clusters (S2 Table). To determine whether the mutations belonging to the same cluster were on single same genomes as predicted by hitchhiking, the eight mutations that belonged to the first cluster, A1, which appeared at generation 5504 and became the majority at generation 5802, were further examined by randomly isolating ten clones from the population at generation 5504. Among the ten clones, four possessed all eight mutations, while the other six clones did not exhibit mutations at the eight sites. Thus, these eight mutations were present on the same genome at generation 5504. The clone ratio of 4:6 is within the standard deviation from the average frequency of 29% over the eight mutations that was estimated from the peaks of Sanger sequencing at generation 5504 (S2 Table). Similarly, all 10 mutations (the fourth cluster, A4) that exhibited an average frequency of 84% at generation 6448 were observed at a clone ratio of 9:2 between clones with all 10 of the mutations and those with no mutations at the corresponding sites, and all three mutations (the fifth cluster, A5) that displayed an average frequency of 81% at generation 6780 were observed at a clone ratio of 7:3 between the clones possessing all or none of the three mutations. These results indicate that the clustering in the fixation dynamics is attributable to the accumulation of the mutations on the same genomes before becoming the majority in the population
The synthesis and growth rate analysis of the recombinant genotypes with the mutations in the first cluster, A1, which became the majority at generation 5802, indicated that the nonsense mutation in mutH hitchhiked on at least one of the other mutations in the same cluster. Mutations on mismatch repair genes that induce mutator phenotypes are fixed together with beneficial mutations that compensate for the genetic load due to high spontaneous mutation accumulation rates [13,31,32]. To determine if the mutH mutation hitchhiked on beneficial mutations, we genetically constructed ten E. coli strains (genomes), based on two different genetic backgrounds of CloneA and CloneB (Fig 3). The CloneA and CloneB genomes carried either none or all of the five mutations (four nonsynonymous and one non-coding region mutations, i.e., mutH, helD, cyaA, nadR and phoU/bglG) in the A1 cluster, respectively. Five genomes out of ten were derived from CloneA and were constructed by adding one of the five mutations to the genome before the fixation of the A1 cluster. The other five genomes derived from CloneB were constructed by back-mutating one of these five mutations to the genome holding the A1 cluster mutations.
Growth assay of these strains (genomes) revealed both the negative contribution of the mutH mutation and the epistasis between the mutations. The addition of the mutH mutation to CloneA reduced the growth (0.05 h-1 decrease) and its subtraction from CloneB rescued the growth (0.03 h-1 increase) (Fig 3), indicating that the negative contribution of the mutH mutation to the growth fitness was independent of the genetic background. The contribution of the other mutations to the growth fitness were somehow dependent on the genetic background. For instance, the growth rate of CloneB was higher (0.04 h-1 increase) than that of the helD mutation supplied CloneA (Fig 3). That is, the accumulation of the other four mutations positively contributed to the genome that carrying the helD mutation, nevertheless these four mutations were harmful to the genome without the helD mutation (CloneA) when occurred individually. The results strongly suggested the epistasis between these mutations. The growth deficiency caused by the mutH mutation might be also partially compensated by the other mutations on the same genome.
We confirmed this compensation by randomly isolating 94 clones from the population at generation 5358, the time point just before the A1 cluster rose above the lower detection limit (5%). Among the 94 clones, the mutH mutation appeared only on the single clone that was accompanied by the mutations in helD, nadR and phoU/bglG (Clone_1/94). Clone_1/94 harboring the four mutations (mutH, helD, nadR and phoU/bglG) showed the same growth rate as that of the genetically constructed clone of the same genetic background (cyaA back-mutation from CloneB), and showed slightly higher growth rate than that of CloneA. It indicated that the negative contribution of the mutH mutation to the growth fitness in the genetic background of CloneA was neutralized by the helD, nadR and phoU/bglG mutations (Fig 3). The growth fitness rescued by the three mutations might allow the cell (genome) carrying the mutH mutation to escape from extinction and to propagate till a substantial number of cell population, as the detected Clone_1/94. Because the growth rate of Clone_1/94 was lower than that of CloneB, a plausible order for the accumulation of the five mutations is mutH, (helD, nadR and phoU/bglG), and cyaA. Although all of the combinations of the five mutations and three synonymous mutations in the A1 cluster should be constructed to examine all possible orders of mutation accumulation, it is reasonable to propose that the mutH non-synonymous mutation could take hitchhiking by neutralizing its negative fitness contribution with at least one of the other mutations in the first cluster A1.
The fixation process of the mutated genomes exhibited complicated dynamics due to the stochastic appearance of adaptive mutations and clonal interference. Theoretically, if adaptive mutations are introduced into populations excessively by a high spontaneous mutation rate or a large population size, these mutations will interfere with each other so that only a fraction of them will be fixed in populations [33]. Clonal interference has been reported in the experimental evolution of large populations [9,34,35,36]. In the experiential evolution of yeast, hitchhiking and clonal interference were simultaneously observed [14]. Consistent with these results for yeast, complicated dynamics were observed when we estimated the frequencies of the single-mutated genomes by averaging the frequencies over the mutations in the same clusters (Fig 4). Until generation 7134, the adaptive genomes stochastically appeared and simply increased their frequency, with the exception of a small fluctuation at generation 6128 for the genome corresponding to the A3 cluster. The fixation processes of the adaptive genomes overlapped as the frequency-increasing genomes gained other adaptive mutations before complete fixation. This overlap is a consequence of the accelerated spontaneous mutation accumulation rate. If adaptive mutations occur on different genomes, the adaptively mutated genomes may interfere with each other. In fact, from generation 7134 to 7580, the most dominant genome on Lineage Blue was interfered with by the other genome on Lineage Red. Lineage Blue survived to the end of Line2, while Lineage Red was inherited at the end of Line1.
Interestingly, only those genotypes that are 34 Hamming distances away from each other occupied the population at generation 7580 without any possible intermediate genomes (70% for the genome with the R1 cluster in Lineage Red, 28% for the genome with the B1~B3 cluster in Lineage Blue, see S2 Table). Generation 7580 is the time point at which the final population of the previous experimental evolution was duplicated into Line1 and Line2. By randomly isolating 20 clones at generation 7580, the coexistence of the two long-distanced genomes was confirmed; 14 clones belonged to the genotype on Red, while 6 clones were on Blue. The coexistence of the distanced genotypes can be attributed to the concurrence of hitchhiking and clonal interference. After generation 7580, further complication was observed in Line1: Lineage Blue, once pushed down by Red at generation 7580, returned to the majority at generation 7741 but became extinct with the domination of Red, which gained the cluster R2. A completely different fate was observed in Line2, where Lineage Blue ultimately eliminated Lineage Red. Thus, the genomic evolution demonstrated complicated population dynamics with the entanglement of the stochastic appearance of the adaptive mutations triggering hitchhiking and clonal interference.
Despite the complicated fixation dynamics of genomic mutation, the molecular evolution rate exhibited a simple clock-like constancy. Synonymous and non-synonymous mutations accumulated monotonously over generations (synonymous and non-synonymous mutations are indicated by middle dots and top dots, respectively, while mutations in non-coding regions are indicated by the bottom dots in Fig 5). Because each of the hitchhiking events propagated a single genome with various numbers of non-deleterious mutations at various rates, the data points deviated to some extent from lines of constant slope. The fixation rates in Line1 and Line2 were approximately equal to each other, with the exception of the specific data points for the synonymous mutations in Linkage Red and Blue after generation 7134 (Lineage Blue exhibited a slightly higher rate than Red, with n = 9 and 5, P<0.05). Even with these different population dynamics, the molecular clock rates of the two lineages were approximately the same order.
The synonymous mutations accumulated at a rate of the same order as the spontaneous mutation rate. By dividing the slope for the unified data of Line1 and Line2 by the synonymous nucleotide sites of 0.96 Mbp of E. coli, we estimated the fixation rate of synonymous mutations as 0.7×10−8 per generation per site. The rate is on the same order as the previously reported accelerated spontaneous mutation accumulation rate of 1×10−8 [29]. Because the synonymous mutations are close to neutral in fitness contribution, hitchhiking in the evolution experiment allowed neutral mutations to be fixed at the same rate as if genetic drift were involved. Because every genome in the population accumulated neutral mutations at the spontaneous mutation accumulation rate, regardless of which genomes were propagated by hitchhiking, the expected fixation rate of neutral mutations should be equal to the spontaneous mutation accumulation rate as theoretically discussed [25,26]. This is the first experimental demonstration of the molecular clock with the same rate as the spontaneous mutation accumulation rate.
Non-synonymous mutations also accumulated at a constant rate, as expected for hitchhiking. Theoretically, strong hitchhiking generated near-neutral fixation probabilities for mutations, although the mutations contributed to the fitness of varied magnitudes [24,37]. In the regimes of emergent neutrality, beneficial and deleterious non-synonymous mutations might affect each other and average out their individual fitness to be nearly neutral. Here, we assumed for simplicity’s sake that only a single non-synonymous mutation that occurred on a genome immediately before its propagation was adaptive, while the other non-synonymous mutations that pre-accumulated on the same genome, each of which could be positive or negative in fitness contribution to some extent, were averaged to be neutral. In fact, complementation of the fitness contribution among the non-synonymous mutations on the same genome was observed in the hitchhiking of the mutH mutation as described above. Based on this assumption, among the 82 non-synonymous mutations fixed through 20 independent propagation events (excluding, for simplicity’s sake, the mutations of cluster R4 and R5 that decreased at the end in Line1) in Line1 and Line2, the majority (62 non-synonymous mutations) should be “averaged” neutral mutations. Because of this neutrality, the hitchhiking events ensured a constant fixation rate of the non-synonymous mutations in the same manner as for synonymous mutations.
Under the severe selection pressure of thermal adaptation, we demonstrated that hitchhiking ran the molecular clock, despite the complicated dynamics of mutation fixation with the stochastic appearance of adaptive mutations and clonal inference. As neutral mutations accumulate on all genomes in a population spontaneously with equal probability, regardless of which genome gains an adaptive mutation to propagate or when, the fixation rate of the neutral mutations should coincide with the spontaneous mutation accumulation rate, suggesting that hitchhiking can be a mechanism for the molecular clock commonly observed in phylogenetic trees in addition to the genetic drift proposed by Kimura [23].
The effect of hitchhiking on molecular evolution rates has been investigated with sophisticated models and computer simulation [25,26]. In the present study, we proposed a simple model to elucidate the dependency of the hitchhike-driven molecular clock rate on the experimentally estimated parameters. To acquire an easy accessible and clear view on the dynamics of the molecular clock, we adopted only a few parameters to propose the model as follows. We classified the mutations in three categories, adaptive, harmful, and neutral, to avoid the complicacy of the epistasis between the mutations. The mutation that appeared lastly on a genome and dominated the population (e.g., > 10%) was categorized as the adaptive mutation. In the present case, the cyaA mutation was the adaptive mutation. As shown in Fig 3, the cyaA mutation positively contributed to the growth fitness of CloneB. It determinatively triggered the fixation of the other four mutations (mutH, nadR, phoU/bglG and helD, as Clone_1/94) and the domination of the population. In comparison, although the helD mutation positively contributed to the growth fitness when it appeared alone on CloneA, it was not the adaptive mutation but was categorized as the neutral mutation, as same as the rest three mutations. Thus, in this model, the neutral mutation could be either deleterious or beneficial when solely appeared but must be irrelevant at dominating the population when appeared with other mutations simultaneously on the identical genome. In addition to this cancelling effect between the beneficial and deleterious mutations, weakened efficacy of selection on non-synonymous mutations might be involved, because of clonal interference [24,37]. The harmful mutation was the deleterious mutation finally disappeared in the population.
To understand how the three categories of mutations participate in the dynamics of the molecular clock, we firstly assumed that every genome spontaneously accumulated the neutral mutations at the following rate:
R_ Neutral_Genome =μL(1−α−β)
(1)
where R_Neutral_Genome is the average number of neutral mutations per genome per generation; μ is the spontaneous mutation accumulation rate (per bp per generation); L is the length of the genome (bp); and α and β are the fractions of the adaptive and harmful mutations. Here, α corresponds to the fraction of the adaptive mutations that become dominated and is smaller than the probability of the beneficial mutation. Thus, the population gains adaptive mutations at the following rate:
R_Adaptive_Population=μLαN
(2)
where R_Adaptive_Population is the number of adaptive mutations per generation in the population, and N is the population size (the number of genomes in a prokaryotic population). The adaptive mutation mediated population domination takes place at the interval (generations) of the inverse of R_Adaptive_Population in average. The neutral mutations accumulate equally on each genome at the rate of R_Neutral_Genome, till an adaptive mutation appears and takes hitchhiking. Therefore, the average number of neutral mutations per hitchhiking (N_Neutral_Hichhike) is:
N_Neutral_Hitchhike =1μLαN×μL(1−α−β)=(1−α−β)/αN
(3)
As the adaptive mutation causes the fixation of the neutral mutations of N_Neutral_Hichhike, the number of the neutral mutations fixed in the population per generation (R_Neutral_Population) is:
R_Neutral_Population=μLαN×(1−α−β)/αN=μL(1−α−β)
(4)
Consequently, the fixation rate of the neutral mutations by hitchhiking at the population level is equal to the spontaneous neutral mutation accumulation rate at the genome level, that is, R_Neutral_Population = R_Neutral_Genome. Noted that the fitness increase attributed to the adaptive mutations does not influence the rate of hitchhike-driven molecular clock of neutral mutations, according to the theoretical demonstration [25,26].
Based on the model, we depicted how the hitchhiking events occurred during the experimental evolution by estimating α and β. A total of 20 clusters of mutation accumulation (i.e., 20 genomes) were identified in the evolution. Note that the R4 and R5 clusters were excluded from the 22 clusters (S2 Table) in this analysis, because they interfered at the end of Line1. As these 20 genomes propagated independently for 4419 generations (5212 to 7580 in common, 7581 to 8829 in Line1, 7581 to 8382 in Line2), the adaptive mutations were supposed to occur at an interval of 221 generations in average, i.e., 1/R_Adaptive_Population = 221. If the synonymous mutations were all nearly neutral compared with the adaptive or harmful mutations at non-synonymous sites, the modification of R_Neutral_Genome for the neutral mutation accumulation rate for synonymous and non-synonymous sites turned to be as follows:
R_Neutral_Synonymous=μLs
(5)
R_Neutral_Non-synonymous=μLn(1−α−β)
(6)
where Ls and Ln are the total lengths of the synonymous and non-synonymous genomic sites. R_Adaptive_Population can be estimated with μLnαN, because the adaptive mutations are supposed to occur at the non-synonymous site. The average numbers of neutral mutations per hitchhiking (N_Neutral_Hitchhike) at the synonymous and non-synonymous sites are as follows:
R_Neutral_Synonymous/R_Adaptive_Population=Ls/LnαN
(7)
R_Neutral_Non-synonymous/R_Adaptive_Population=(1−α−β)/αN
(8)
Because 32 synonymous mutations were fixed through the propagation of the 20 genomes, the average number of neutral mutations on synonymous genomic sites per hitchhike Ls/LnαN was 1.6. With a population size N of 107 and a ratio of synonymous sites over non-synonymous sites on the E. coli genome of Ls/Ln of 0.96 Mb/3.2 Mb = 0.3 [29], we estimated α as 2×10−8. The estimated value for the fraction of adaptive mutations α is small compared to previous estimates, even after considering clonal interference [33,38]. The small estimate occurs in part because only the adaptive mutations that triggered the hitchhikes are counted in α, while the other beneficial mutations that appeared and compensated for the deleterious mutations on the same genome were regarded as neutral in this study. In addition, as the 62 “averaged” neutral non-synonymous mutations became fixed during the 20 propagations, the average number of neutral mutations on non-synonymous genomic sites per hitchhike (1−α−β)/αN was 3.1, leading to an estimated β of 0.42. Approximately 42% of the mutations at non-synonymous sites were excluded from the population as harmful mutations, most likely due to the severe selection pressure from the successive temperature increments of 0.2°C. Based on the estimated values, the depicted dynamics for hitchhike occurrence were that the neutral mutation accumulation rates for synonymous and non-synonymous regions at the genomic level, R_Neutral_Synonymous = 0.01 and R_Neutral_Non-synonymous = 0.02, respectively, were faster than the adaptive mutation fixation rate at the population level, R_Adaptive_Population = 0.006. The excess accumulation of the neutral mutations hitchhiked on the rare adaptive mutations as theoretically proposed [25,26].
Regardless of the involvement of mutator phenotypes, the hitchhike-driven fixation rate for neutral mutations should coincide with the spontaneous mutation rate [25,26]. Obviously, the accelerated spontaneous mutation rate induced by the mutation of mutH resulted in the accumulation of many mutations in the genomes in the limited generations, resulting in the observation of the constant mutation fixation rates (Fig 5). Without the accelerated spontaneous mutation rate, we would have observed few neutral mutations, as we observed only one synonymous mutation until generation 5212 [29]. In long-term experimental evolution conducted by Lenski’s group, many synonymous mutations were observed after a mutator phenotype emerged [30]. The fixation rate of those synonymous mutations approximately coincided with the accelerated spontaneous mutation rate, suggesting that some of these mutations, if not all, might have been fixed by hitchhiking as observed in this study. These authors also reported that the synonymous mutations were not fixed before the appearance of the mutator, implying no hitchhike events. However, this result does not deny the independency of the hitchhike-driven fixation rate of neutral mutations from the mutator phenotypes. If the spontaneous mutation rate had remained without the two-order acceleration but at an ordinary value of 10−10/site/generation over 105 generations, one could have observed the hitchhike-driven fixation of approximately ten synonymous mutations. Adaptive mutations also accumulated in a clock-wise manner in Lenski’s experimental evolution. The rate of adaptive mutation fixation has theoretically been investigated [39]. Long-term experimental evolutions would verify the molecular clock for neutral and adaptive mutations. As the coincidence of the fixation rate of neutral mutations at the population level with the spontaneous mutation accumulation rate at the genome level occurs in two fixation mechanisms, hitchhike and genetic drift, the dominant mechanism remains unclear. Genetic drift has been intensively investigated and has provided the theoretical backbone for the molecular clock of neutral evolution, in which organisms gain little fitness [23]. As the fixation of neutral mutations are supposed to be governed by genetic drift before the adaptive mutations take hitchhiking, it would be worthwhile to determine under what conditions genetic drift works to fix neutral mutations without the disturbance by hitchhiking in adaptive evolution.
Here, we discuss the question with a simple model which requires only the parameters that were estimated above, although there are sophisticated investigations on the fixation process of neutral mutations [12,27]. First, suppose that a neutral mutation appears on a genome in a population of size N and propagates through genetic drift, which requires many generations of typically the order of N for asexual organisms. To avoid sweeping out or fixing the neutral mutation by hitchhiking in its drifting process, no adaptive mutations should occur during N generations. The probability that no adaptive mutation occurs in a population of size N for N generations or longer is (1−μLNα)N. In the case of the evolution experiment in this study, the probability is virtually zero with the estimated parameters μ = 10−8, L = 0.32×107, α = 2×10−8, N = 105. In other words, for a neutral mutation to be fixed by genetic drift at a 50% probability (1−μLNα)N = 0.5 or higher, the population size should be approximately on the order of 104 or less. If one applies to the equation the typical spontaneous mutation rate of 10−10 for the non-mutator phenotype but an adaptive mutation fraction α of 2×10−6 because of the relief from clonal interference, the same population size of 104 will be obtained. To apply the equation to eukaryotes, if the effective genome length L for adaptive mutations is replaced with the inverse of the recombination rate (i.e., 10−8/site/generation) multiplied by the sparseness of the coding region or non-neutral sites (for instance, 0.01), the threshold will be 105. A similar threshold value above which hitchhiking dominates over genetic drift to fix neutral mutations has been proposed for a different reason [12]. Because a significant fraction of species have a population size of greater than 104~5, some if not all of the neutral mutations that were observed in the phylogenetic trees might have been fixed by hitchhiking [40].
We experimentally demonstrated that hitchhiking ensured that the molecular clock for neutral mutations in fitness-increasing evolution occurred at the same rate as expected for genetic drift in neutral or fitness-steady evolution. The extension of the molecular clock, which has been verified for neutral evolution, to adaptive evolutions can provide insights into the past environment. Genetic drift, which requires the suppression of adaptive mutations over a long generation time, has led us to suppose that the past environment was sufficiently steady that the fitness contributions of fixed mutations could quickly change from positive to neutral by diminishing returns, leading to fitness-steady or neutral evolution. However, if the hitchhike-driven molecular clock runs for neutral mutations, adaptive mutations may occur frequently. Thus, it is reasonable to propose that the past environment was not always so steady but underwent dynamic changes that allowed organisms to adaptively evolve according to the hitchhike-driven molecular clock.
A previously evolved E. coli strain 45L[29] (DH1ΔleuB:: (gfpuv5-kmr)) was used for prolonged thermal adaptive evolution. The generations were calculated according to the growth datasets as summarized in S3 Table in the present study and previously reported in Table S5. The generations of 45L (the final population) and 45A (the intermediate population in which the fitness contributions of accumulating mutations shifted from positive to neutral) were 7580 and 5212 and were incorrectly reported as 7560 and 5191 in the previous study [29]. The corrected generations are consistent with the raw data in supplemental Table S5 of the previous study.
Cell culture was performed as previously reported with minor modifications [29]. Culture temperatures of 44.8°C, 45.0°C, 45.2°C, 45.4°C, 45.6°C, 45.8°C and 46.0°C were confirmed using a Platinum Resistance Thermometer 5615 (Fluke). Serial transfers of Line1 and Line2 were performed daily by diluting the exponential phase culture with pre-warmed (37°C) fresh medium. The dilution rate was determined to maintain cell growth within the exponential phase, i.e., the cell concentration was controlled at OD600 = 0.05–0.2 after 24 h of culture based on the growth rate estimated the previous day. During the temperature increase periods and the first one or two days of the culture restarted from the glycerol stock, the cell cultures were usually kept at a relatively high concentration, e.g., OD600>0.5. If the OD600 of the 24-h culture was lower than 0.05, the cell culture was extended for 24 h. The growth rate was calculated as described previously [29], and the details are summarized in S3 Table.
Cells were inoculated from a glycerol stock and cultured at the corresponding adaptive temperature (44.8°C and 46.0°C, respectively). Following pre-cultures for 24 h, the cell cultures were transferred to fresh medium and incubated at 20.0°C, 37.0°C, 45.0°C, 46.0°C, 46.5°C and 47.0°C for 24 h. The growth rate at each culture temperature was calculated as the average value of triplicate cultures.
The spontaneous mutation rate was determined by the fluctuation test as previously described [29]. The initial and final cell concentrations in all tests were controlled at approximately 1,000 and 5×107−1×108 cells/mL per tube, respectively. The cell concentrations were measured by flow cytometry (FACSCalibur, Becton, Dickinson and Company). A total of 35 tubes of 5 mL culture were used for each test. The cells in the culture were collected and transferred to plates containing 25 μg/mL streptomycin. After two days of incubation at 37°C, the numbers of plates without colonies were counted. The resultant probability, that is, the ratio of the number of plates without colony formation to the total number of plates, was used to calculate the spontaneous mutation rate.
The cell populations at the beginning of prolonged evolution and at the ends of Line1 and Line2 were subjected to next-generation resequencing analyses. Cells grown until stationary phase were collected as previously described [29]. Genomic DNA was purified and fragmented using the Covaris system as previously described [41]. Whole-genome resequencing was performed using the Applied Biosystems SOLiD 3 system according to the manufacturers’ instructions. Replicates of 50-base mate pair libraries (two quarters of a slide) were prepared for each sample. The overall procedure of library preparation, resequencing reaction and base calling was performed according to the manufacturer’s recommendations. The sequence reads were assembled and aligned using Bioscope v2.0 (Applied Biosystems). The reads in each dataset were mapped to the ancestor genome DH1 [41], to acquire the results of single nucleotide substitutions. The further confirmation of the detected mutations, the temporal change and heterogeneity of the mutations in the cell populations were performed using the Sanger method, as described in the following section.
All single nucleotide substitution mutations identified by genome resequencing were further confirmed by Sanger methods. The mutation fixation dynamics were evaluated by Sanger sequencing of the stocked cell populations that were acquired in the evolution experiments. The cell stocks were directly subjected to Sanger sequencing to avoid the possible biased enrichment of heterogeneity within the populations. The results of the Sanger sequences were visualized using the software VectorNTI (Invitrogen) or 4Peaks (Mek&Tosj). The changes in mutation fixation dynamics were determined according to the ratio of the peak values representing the wild type and the substituted nucleotides in each population for each mutation. The analyses and graphics of the mutation fixation dynamics were generated using the software Mathematica v9.0 (Wolfram).
Single-cell isolation of the cells that were stocked during the evolution experiments was performed to purify the genotypes among the populations. The glycerol stocked cells were precultured at 44.8°C, and the cell concentrations of the cultures were determined by flow cytometry (FACSCalibur, Becton, Dickinson and Company). The cell populations were subsequently diluted with fresh media and inoculated into 96-deep-well plates at concentrations of 0.1 cells/well in 200 μL per well. The plates were incubated at 37°C with shaking at 1,100 rpm (Deepwell Maximizer MBR-022UP, TAITEC) for two or three days. The wells that became turbid were a result of single-cell-derived growth. The cells were checked for growth at 44.8°C and finally harvested for further mutation analyses.
|
10.1371/journal.pntd.0006192 | Polymorphisms in voltage-gated sodium channel gene and susceptibility of Aedes albopictus to insecticides in three districts of northern West Bengal, India | The control and prevention of dengue largely depends on vector control measures, environmental management, and personal protection. Dengue control programmes are facing great challenges due to development of insecticide resistance among vector mosquitoes. Information on susceptibility status to different insecticides is important for national programmes to formulate vector control strategies.
We have studied the larval susceptibility of Aedes albopictus to temephos and adult susceptibility to 4% DDT, 0.05% deltamethrin, and 5% malathion as per WHO protocols in the northern districts of West Bengal. Polymorphisms in the VGSC gene were studied by direct sequencing of PCR products.
The Ae. albopictus larval population showed sensitive [Resistance Ratio (RR99)<3] to moderate levels of resistance (5<RR99>10) to temephos at different study sites. Adult bioassay results revealed that Ae. albopictus was highly resistant to DDT [Corrected Mortality (CM) < 90%] in all the study sites and susceptible to deltamethrin and malathion (CM > 98%), except in Dhupguri where a low level of resistance to deltamethrin (CM = 96.25%) was recorded. None of the six important kdr mutations (S953P, I975M/V, L978, V980G, F1474C, D1703Y) were found in the VGSC of studied mosquitoes, but we identified 11 synonymous and 1 non-synonymous mutation in the VGSC gene.
The higher susceptibility level to deltamethrin and malathion, along with the absence of important kdr mutations indicates that these two insecticides are still effective against Ae. albopictus in the study areas. The susceptibility status of temephos should be monitored closely as low to moderate levels of resistance were observed in few sites. A similar study is recommended for monitoring and early detection of insecticide resistance in other parts of the country.
| Dengue is one of the most important arboviral infections in India, and transmitted by Aedes mosquitoes. Control of dengue largely depends on effective vector control measures as no specific drugs or vaccines are available, to date. The knowledge of insecticide susceptibility status for local vector mosquitoes is essential for the formulation of effective vector control measures. Therefore, regular monitoring of insecticide susceptibility is essential for the early detection of resistance. In the present study, sensitive to moderate levels of resistance to temephos were recorded among the Aedes albopictus larval populations of different study sites. Adult Ae. albopictus were highly resistant to DDT and susceptible to deltamethrin and malathion in all the study sites, except one site where a low level of resistance to deltamethrin was observed. By sequencing the VGSC gene we observed that the studied Ae. albopictus population does not contain any of the important kdr mutations which are associated with DDT and pyrethroid resistance in other insects. We found only one non-synonymous mutation at codon S1485L, but its role in pyrethroid resistance is yet to be established.
| Dengue is a mosquito-borne flavi-viral disease and a major public health problem in more than 120 countries [1, 2]. In recent years, dengue transmission has increased predominantly in urban, semi-urban areas and has even extended to the rural areas, becoming a major public health concern globally. A recent estimate showed 390 million new dengue infections throughout the world, of which, 96 million cases manifested the severe form of the disease [2] and almost half of the world’s population are at risk of dengue infection [3]. In India, dengue is spreading into new areas and emerging as a major public health problem. In 2016, a total of 129166 dengue cases and 245 deaths were reported from India, of which 22865 cases and 45 deaths were reported from West Bengal [4]. Aedes aegypti and Aedes albopictus are the vectors of dengue along with three other important human viral diseases: yellow fever, chikungunya, and Zika. No effective vaccine against dengue is available to date. Vector control and personal protection from mosquito bites are suggested to reduce its transmission. For proper formulation and implementation of vector control strategies, thorough information about vector species distribution and their susceptibility to available insecticidal agents are necessary [5].
Four different classes of insecticides are in use as adulticides against Aedes mosquitoes: organophosphates, pyrethroids, organochlorines, and carbamates [6, 7]. Among these, pyrethroids and organophosphates are widely used throughout the world [8, 9, 10]. Pyrethroids are used as indoor residual treatment and impregnation of bed nets whereas organophosphates are used as larvicides and space treatments [6]. The National Vector Borne Disease Control Programme (NVBDCP) of India recommends different insecticides for vector management, such as temephos (50 EC) as a larvicide, DDT and synthetic pyrethroids (recently introduced) for indoor residual spray (IRS), deltamethrin (pyrethroid) for impregnation of bed nets, and malathion for ultra low volume (ULV) spray. In India, Aedes mosquito control is mainly based on anti-larval measures and the use of insecticides by space spraying of pyrethrum and fogging of malathion during a disease outbreak to kill adults. The development and spread of resistance by the vector mosquitoes against all available insecticides is a great challenge to prevent the transmission of mosquito-borne diseases. Ae. albopictus and Ae. aegypti showed resistance to DDT [11, 12, 13], but were susceptible to malathion and deltamethrin [11, 12, 14, 15] in different parts of India. Pyrethroids are synthetic analogues of naturally occurring pyrethrum from the extracts of the Chrysanthemum flower and represent the most widely used insecticide against insect vectors [16]. Unfortunately, pyrethroid efficacy is being threatened due to rapid development of resistance by the vector mosquitoes [8, 17]. The World Health Organisation (WHO) formulated standard diagnostic bioassay test kits to monitor the susceptibility of mosquitoes against different insecticides [18].
Exposure to pyrethroids and DDT results in “knockdown” (i.e., rapid paralysis) due to prolonged-activation of sodium channels. Pyrethroids and organochlorines cause overstimulation of the mosquito nervous system by repeated action potentials form the opening of the sodium channel [19, 20, 21]. Knockdown resistance (kdr) is the major mechanism of pyrethroid resistance, caused by mutations in the voltage-gated sodium channel gene (VGSC gene) [22, 23]. In insects, the voltage-gated sodium channel is an integral transmembrane protein which is composed of four homologous domains (I-IV). Each domain consists of six subunits (S1-S6) which are connected by loops. The segments S5, S6, and the P-loop between them form a central aqueous pore, and the S1-S4 segments of each domain unite to form four independent voltage-sensitive domains [24, 25]. Insects have only one functional sodium channel gene [19]. There are two receptor sites in the four-domain sodium channel for simultaneous binding of pyrethroids [26].So far, ten different mutations at eight codons comprising fifteen haplotypes have been reported in Ae. aegypti. The frequency of these mutations varies geographically [27, 28] but such reports from India are very rare.
Periodical monitoring of insecticide resistance among the prevailing vector population in a given geographical region will be helpful to formulate vector control strategies by the NVBDCP. The present work was designed to study the susceptibility status of Ae. albopictus to temephos, DDT, deltamethrin, and malathion, as well as polymorphisms in the VGSC gene in dengue endemic areas of northern West Bengal.
This study was carried out in one municipality and two blocks of Darjeeling, two blocks of Jalpaiguri, and one block of Uttar Dinajpur districts of West Bengal during June 2016 to September 2016. The study locations were Siliguri Municipal Corporation (SMC), Matigara, and Khoribari of the Darjeeling district; Malbazar, Dhupguri of the Jalpaiguri district, and the Itahar block of Uttar Dinajpur. Most of the study sites were sub-urban except Siliguri Municipal Corporation (urban) and Khoribari (rural) (Fig 1).
The aquatic stages (larvae and pupae) of Aedes sp. were collected from the seventeen localities of three districts. For each collection site, larvae and pupae were collected from domestic, peri-domestic and natural breeding places. The collected immature stages of mosquitoes were stored in plastic containers containing water from the same breeding habitat and transferred to the laboratory. In the laboratory, the wild caught mosquito larvae and pupae were transferred into a larvae rearing tray along with water collected from the field and supplied with food for ornamental fishes available in the local market along with yeast. The mosquito larvae and pupae were reared to the adult stages in the laboratory under controlled conditions (temperature 25°C ± 2°C; relative humidity 80% ± 10%). After emergence, the adults were identified by using the standard identification keys of Barraud, 1934 [29] and Tyagi et al., 2012 [30]. The identified Ae. albopictus were allowed to breed under laboratory conditions. The larvae and adults of the F1 generation were used for larval and adult insecticide bioassays.
Susceptibility of larvae to temephos (50EC; Nitapol Industries Pvt Ltd., Kolkata) was estimated using the standard WHO bioassay protocol [31]. The stock temephos solution of 1 ppm concentration and other subsequent dilutions were prepared in 95% ethanol and stored at +4°C for use in the susceptibility bioassay. Bioassays were conducted using 20–25 third instar to early fourth instar larvae (wild caught strain and laboratory strain) in disposable paper cups filled with the required concentration of insecticide solution and double distilled water at room temperature (25°C ± 2°C). Eight different concentrations (0.0005, 0.001, 0.005, 0.01, 0.05, 0.1, 0.5 and 1.0 ppm) were used as per WHO recommendation [32, 33] and each experiment was replicated at least three times. Each set of the bioassay was accompanied by two sets of controls (equal concentration of 95% ethanol). Larval mortality was recorded after 24 h of exposure. The larvae that were motionless or convulsive upon a sharp stimulation were counted as dead [31]. Larval mortality was determined by dividing the number of dead larvae by the total number tested. A test was considered as invalid if pupation rate was greater than 10%, or mortality rate in the control was greater than 20% [31]. The degree of resistance was determined by the resistance ratio (RR99), which is calculated by comparing the lethal concentration (LC99) value for a population with the LC99 value for the insecticide for a laboratory colony. The RR99 ≤3 was considered as susceptible, and 3 < RR99 ≤ 5 as low resistance, 5 < RR99 ≤ 10 as moderate resistance, and RR99> 10 as high resistance [34].
Two to three day old laboratory emerged unfed female Ae. albopictus mosquitoes were used for the insecticide susceptibility bioassay as per WHO protocol [18]. The tested insecticides were 4% DDT, 0.05% deltamethrin, and 5% malathion. The insecticide-impregnated papers were procured from the Vector Control Research Unit (VCRU), Universiti Sains Malaysia, Malaysia. Five different holding tubes were used for each set of the experiment of which four were a test and one was a control. In each holding tube,15–20 adult female mosquitoes were kept for one hour. After one hour of holding, mosquitoes from four tubes marked as test were exposed to insecticide-impregnated papers. The control tests were performed using silicone oil, olive oil, and risella oil pre-impregnated papers for deltamethrin, malathion, and DDT, respectively. Mosquitoes were allowed in the exposure tube for one hour and cumulative knock down was recorded after 10, 15, 20, 30, 40, 50, and 60 minutes. After 60 minutes of exposure, the mosquitoes were transferred to holding tubes and fed on a 5% sucrose solution for the next 24 h. Mortality was scored after 24 h to determine the susceptibility status as per WHO recommendation [18]. Mosquitoes were considered dead if they were motionless, when they were mechanically stimulated, following the method of Gonzalez Audino [35]. The live and dead mosquitoes obtained from the adult bioassays were stored at -20°C and used for molecular biological assays.
Genomic DNA was isolated from both live and dead mosquitoes (individually) by using the DNeasy Blood & Tissue Kit (Qiagen, Germany), as per the manufacturer’s instructions. Before initiation of DNA isolation, the wings of the mosquitoes were removed and the remaining part of the mosquito was carefully homogenised by a Tissue Ruptor (Qiagen, Germany). Extracted DNA was stored at -20°C until further study.
PCR was done using three different primer pairs targeting six amino acid loci (S953P, I975M/V, L978, V980G of domain II, F1474C of domain III and D1703Y of domain IV) of the voltage-gated sodium channel gene (VGSC) of Ae. albopictus, which is responsible for knockdown resistance (kdr). The details of primers and PCR conditions are given in Table 1 as described earlier by Kasai et al., 2011 [36].
The quality of PCR products was ascertained by 2% agarose gel electrophoresis following ethidium bromide stain. The PCR product was gel purified using the Qiagen gel extraction kit (Qiagen, Germany) and sequencing was outsourced to Chromous Biotech, Bangalore. Four different primers i.e., aegSCF3, aegSCR22 (forward and reverse primer for domain II), aegSCR8 (reverse primer for domain III), and albSCF7 (forward primer for domain IV) were used for sequencing of the PCR products.
In the present study, we numbered the codons of the VGSC gene according to the sequence of Ae. albopictus. The sequences were analysed using the software BioEdit Sequence Alignment Editor version 7.0.9.0. The sequences were aligned with the reference sequence for Ae. albopictus (GenBank accession no. AY663384.1), using an online multiple sequence alignment tool.
Before initiation of the work, the objectives of the study were explained to the local population of each study site. Permission was taken from the owners of private houses/lands before collection of immature stages of mosquito. The study did not involve any endangered and protected species. Mosquitoes were maintained under optimal conditions such as temperature, humidity, and adequate food supply in the laboratory. The study protocol was approved by the Institutional Ethics Committee of Calcutta School of Tropical Medicine, Kolkata.
Larval bioassay data were analyzed using Log dose probit (Ldp) Line computer software (Ehabsoft, Cairo Egypt; available at: http://www.ehabsoft.com/ldpline) according to the Finney’s method [37]. Chi-squared (χ2) test was used to estimate the goodness of fit, while linear regression was used to evaluate the data linearity. Lethal concentrations (LC10, LC50, and LC99) along with the slope were estimated at 95% confidence intervals (CI). For adult bioassays, observed mortality was calculated by the formula: observed mortality (%) = (Total no. of dead mosquitoes / Total mosquitoes exposed) x 100. The observed mortality was corrected using Abbott’s formula when the mortality rate of control was within 5% - 20%. Corrected Mortality (CM) (%) = [(% of observed mortality—% of control mortality) / (100 - % of control mortality)] x 100. For adult bioassays, resistant/susceptibility status was defined according to WHO recommendations [18]. Mosquitoes were considered susceptible (S) if the corrected mortality (CM) rate was greater than 98% and resistant (R) if mortality rate was less than 90%. Mortality rate between 90–98% was considered as possible resistance (PR) and needs verification by alternative methods like enzyme bioassay and molecular marker studies [18]. The cumulative knock down rates (KDR) were calculated by observing the number of knocked down mosquitoes after 10, 15, 20, 30, 40, 50 and 60 minutes during the hour-long exposure period. Knockdown time (KDT10, KDT50, and KDT95) is the time required for knockdown of a particular proportion of mosquitoes following exposure to any insecticide. KDTs were determined using Log dose probit (Ldp) Line computer software (Ehabsoft, Cairo Egypt; available at: http://www.ehabsoft.com/ldpline) programme according to the Finney’s method [37].
The study was conducted in one municipality and 5 different blocks of 3 districts in the northern part of West Bengal during June 2016 –September 2016. The study sites of Dhupguri and Itahar blocks were surrounded by paddy fields, whereas the presence of both paddy fields and tea gardens were characteristic of the remaining study sites except Siliguri Municipality Corporation (SMC) and Matigara. Most of the study sites were suburban in nature except the Siliguri municipality area (Urban) and Khoribari (rural) (Fig 1). Storage water tanks, discarded tyres, tree holes, construction sites, flower pots, plastic cups, coconut shells, and discarded containers were the different seasonal breeding sites found in the study area. The climatic conditions of all study sites were humid and sub-tropical in nature and the temperature varies from 8°C in winter to 40°C in summer.
The summary of larval bioassay results is presented in Table 2. The LC10, LC50, and LC99 values of different study sites did not follow a normal distribution for mortality to the log dose (χ2 ≥ 16.08; p ≤ 0.01). The LC50 values ranged from 0.0009 to 0.0015 mg/L and LC99 from 0.1565 to 0.3343 mg/L. The calculated RR50 and RR99 values in different study sites were ranged from 1.0 to 2.5 and 2.45 to 5.24, respectively.
The results of the adult susceptibility bioassay for Ae. albopictus are given in Table 3. After 24 hours of exposure, the corrected mortality rates for 4% DDT were 23.75% to 85.53% in different study sites. The obtained mortality rates were well below the WHO recommended 90% mortality rate for resistance. So, results suggested that the Ae. albopictus population of the study areas was highly resistant to DDT. In all of the study sites, the corrected mortality rate for 0.05% deltamethrin ranged from 98.08% to 100%, except in Dhupguri where the corrected mortality was 96.25%. So, Ae. albopictus population of all the study sites was susceptible to deltamethrin except Dhupguri. The corrected mortality rate for 5% malathion was >98% in all the study sites indicating susceptibility to malathion.
The knock down time (KDT10, KDT50, KDT95) for DDT, deltamethrin, and malathion showed a linear probit for knock-down rates with time in most of the study sites (Table 3). The observed KDT50 values were 23.62 to 51.39 mins for DDT, 10.14 to 13.82 mins for deltamethrin, and 17.52 to 25.31 mins for malathion. The KDT95 values for DDT were 80.10 to 212.11 mins, for deltamethrin 22.85 to 43.28 mins and for malathion 32.39 to 77.04 mins. The survival rate of Ae. albopictus against DDT, deltamethrin, and malathion over an exposure time of 1 hour is given in Fig 2A–2C. During 1 hour of exposure, the knock down rate (KDR) varies from 68.75% - 93.75% for DDT, 100% for deltamethrin, and 95.00% - 100% for malathion.
DNA was isolated from 30 dead and 10 alive, deltamethrin-exposed Ae. albopictus mosquitoes and used for PCR amplification. For detection of kdr mutations three DNA fragments of 480 bp, 740 bp, and 280 bp for domain II, III, and IV of VGSC gene were amplified, respectively. None of the six important kdr mutations (i.e., S953P, I975M/V, L978, V980G, F1474C, D1703Y) were found among the studied mosquitoes. We detected 3 synonymous mutations in domain II, 1 non-synonymous and 3 synonymous mutations in domain III, and 5 synonymous mutations in domain IV. The frequencies of observed mutations are presented in Table 4. The DNA sequences have been submitted to GenBank under accession nos.MF776970 and MF774494.
Emergence and spread of insecticide resistance is the biggest challenge to control vector-borne disease transmission [38]. In Aedes mosquitoes there are two major mechanisms for pyrethroid resistance: increased detoxification and mutation in the VGSC gene. To date more than 50 different VGSC mutations have been identified in different insect species [19]. The six non-synonymous amino acid substitutions: S989P, I1011M, L1014F, V1016G in domain II, F1534C in domain III and D1763Y in domain IV of house fly are found to be associated with pyrethroid resistance. These codons are orthologous to the codons 953, 975, 978, 980, 1474 and 1703, respectively of Ae. albopictus. The involvement of other mutations in pyrethroid resistance remains to be investigated. The L1014F, at S6 subunit of domain II was the first pyrethroid-resistance-associated mutation identified in the house fly and German cockroach [39, 40, 41]. I1011M was identified in domestic house fly from Brazil, Guyana, whereas V1016G was identified from Indonesia and Thailand [42]. Later, different substitutions, I1011V and V1016I, were found in Ae. aegypti populations from Latin America [43]. The most significant F1534C, located in S6 subunit of domain III was discovered in DDT/permethrin-resistant Ae. aegypti in Thailand and Vietnam [44, 45]. The adult insecticide susceptibility bioassay is applied to determine the lethal dose of different insecticides by direct exposure. Additional tests, such as polymorphisms in marker genes and biochemical assays of different enzymes are used as supplementary evidence to clarify the results of bioassays and potential mechanisms.
In the present study, we determined the susceptibility status of Ae. albopictus against DDT, deltamethrin, and malathion. The results showed that Ae. albopictus is significantly resistant to DDT with a higher KDT and KDR and a lower mortality rate. Similar observations have also been reported from other parts of the country [11, 12, 14]. Though DDT is not in use against Aedes vector mosquitoes, this compound is still in use for control of malaria vectors. The present study areas have been highly endemic for malaria for a long time, with the exception of Itahar. Thus, Aedes mosquitoes have been exposed to DDT for many generations which might be the cause of the high level of resistance that has developed to DDT. Ae. albopictus from the present study areas were susceptible to deltamethrin and malathion. Pyrethroid resistance in adult Aedes sp. is a problem worldwide. The level of resistance varies from region to region. A lower level of resistance is found in Asian, African, and Northern American countries, [46, 47, 48, 49] whereas higher levels of resistance are found in South American countries [50, 51]. In the present study, lower values of knock down time and knock down rate were observed in Ae. albopictus against deltamethrin and malathion. The KDT values recorded in the present study did not follow a normal distribution pattern which indicates that the prevailing Ae. albopictus population is susceptible to these insecticides.
In India, temephos is used as larvicidal agent. In contrast to adult susceptibility, higher levels of larval resistance have been found in Asian, African, and North American countries [49, 52, 53, 54, 55]. In the present study we found that the Ae. albopictus larvae were sensitive to temephos in Khoribari and Dhupguri (RR99<3); showed a low level of resistance in Matigara, Malbazar, and Itahar (3<RR99>5), and moderate resistance in Siliguri (5<RR99>10) [33].A similar type of observation was also reported from the north eastern part of India [11, 12]. The Siliguri Municipal Corporation is the only urban site in the present study, where temephos has been in use for a long time. A longer duration of exposure to temephos might be the cause of the observed moderate level of resistance against it in Ae. albopictus from Siliguri. In contrast, a recent report from the northern part of West Bengal showed susceptibility of Ae. albopictus larvae to temephos assessed by larval susceptibility and bioassay of detoxifying enzymes [56].
The KDR is a mechanism of DDT and pyrethroid resistance. Mutations at codons 953, 975, 978, 980, 1474, 1703 of the VGSC gene of Ae. albopictus have been found to be associated with reduced susceptibility to both DDT and pyrethroids [22, 23]. As per our present study, the only previous report from India did not reveal any mutation in the VGSC gene of Ae. albopictus [8] but two other reports reveal mutations at codon F1534C [57] and at codon T1520I + F1534C of the Ae. aegypti VGSC gene [13]. In the present study, we detected only one non-synonymous mutation at S1485L in three samples. Interestingly, all three mosquitoes were susceptible to deltamethrin. So, the role of this mutation in pyrethroid resistance cannot be explained. We also detected 11 synonymous mutations among both dead as well as live deltamethrin-exposed mosquitoes.
We did not assess the detoxifying enzyme levels associated with DDT and deltamethrin resistance. The higher susceptibility level in deltamethrin with absence of important kdr mutations and higher susceptibility to malathion indicate that these two insecticides are still effective in the study areas. The susceptibility status of temephos as a larvicide should be monitored closely as moderate and lower levels of resistance were observed in mosquitoes from a few study sites. A similar study is highly recommended for monitoring and early detection of pyrethroid and malathion resistance in other parts of the country.
|
10.1371/journal.ppat.1002256 | The Human Papillomavirus E6 Oncogene Represses a Cell Adhesion Pathway and Disrupts Focal Adhesion through Degradation of TAp63β upon Transformation | Cervical carcinomas result from cellular transformation by the human papillomavirus (HPV) E6 and E7 oncogenes which are constitutively expressed in cancer cells. The E6 oncogene degrades p53 thereby modulating a large set of p53 target genes as shown previously in the cervical carcinoma cell line HeLa. Here we show that the TAp63β isoform of the p63 transcription factor is also a target of E6. The p63 gene plays an essential role in skin homeostasis and is expressed as at least six isoforms. One of these isoforms, ΔNp63α, has been found overexpressed in squamous cell carcinomas and is shown here to be constitutively expressed in Caski cells associated with HPV16. We therefore explored the role of p63 in these cells by performing microarray analyses after repression of endogenous E6/E7 expression. Upon repression of the oncogenes, a large set of p53 target genes was found activated together with many p63 target genes related to cell adhesion. However, through siRNA silencing and ectopic expression of various p63 isoforms we demonstrated that TAp63β is involved in activation of this cell adhesion pathway instead of the constitutively expressed ΔNp63α and β. Furthermore, we showed in cotransfection experiments, combined with E6AP siRNA silencing, that E6 induces an accelerated degradation of TAp63β although not through the E6AP ubiquitin ligase used for degradation of p53. Repression of E6 transcription also induces stabilization of endogenous TAp63β in cervical carcinoma cells that lead to an increased concentration of focal adhesions at the cell surface. Consequently, TAp63β is the only p63 isoform suppressed by E6 in cervical carcinoma as demonstrated previously for p53. Down-modulation of focal adhesions through disruption of TAp63β therefore appears as a novel E6-dependent pathway in transformation. These findings identify a major physiological role for TAp63β in anchorage independent growth that might represent a new critical pathway in human carcinogenesis.
| High-risk human papillomavirus infection can cause cancer of the uterine cervix. The viral proteins leading to transformation of the infected keratinocytes are the E6 and E7 oncogenes which interact with and induce degradation of the cell cycle regulators p53 and pRB. In cervical carcinoma cells, repression of E6/E7 stabilizes the p53 transcription factor leading to activation of a large group of cellular p53 target genes. Here we show that repression of E6/E7 also induces transcriptional activation of an additional large set of genes involved in cell adhesion including previously described p63 target genes. Indeed, we further demonstrated that these p63 target genes are activated by TAp63β and not by p53 or by the ΔNp63α or β isoforms, even though these transcription factors are also expressed in these cells. In cervical carcinoma cells, E6 expression therefore leads to TAp63β degradation thereby allowing anchorage independent growth. Our work describes a new E6-dependent transformation pathway in HPV-associated carcinogenesis. TAp63β inhibition may also represent a common pathway to activate anchorage independent growth in cancers.
| Infection of the anogenital mucosal epithelium with high risk Human Papilloma Virus (HPV) is linked to 99% of cervical carcinomas [1]. Cell lines derived from these cervical carcinomas remain associated with HPV and contain part of the viral genome integrated in the cellular genome. However, not all viral genes are retained in this integration; the E6 and E7 oncogenes remain, while the open reading frames encoding viral proteins E1 and E2, necessary for viral DNA replication, are disrupted [2], [3]. We have previously used the HPV18-associated HeLa cell line to study transcriptional modulation of viral and cellular genes following repression of the E6 and E7 oncogenes, and found that a large number of cellular genes were in fact modulated via E6 and E7 [4], [5]. Of particular interest was the discovery that genes targeted by either p53 or E2F were respectively activated or repressed through repression of E6 and E7 [4].
We now wish to develop and extend these findings. In particular we are interested in the potential effect of HPV E6 and E7 on other less well defined members of the p53 family. The p73 and p63 transcription factors are more recently discovered p53 family members, and although they share structural homology with p53 and are able to interact with similar DNA binding motifs, they modulate different regulatory pathways [6]–[9]. While p53 is a tumor suppressor and does not obviously participate in embryonic development, p63 and p73 on the contrary are strongly linked to embryonic development in mice [10], [11].
The critical developmental role of p63 is illustrated in null mice, which do not survive beyond few days after birth and exhibit limb truncation and a striking absence of skin [10], [11]. Human genetic syndromes with comparable phenotypes have also been linked to p63, including EEC syndrome (Ectrodactyly Ectodermal dysplasia Clefting), ADULT (Acro-Dermato-Ungual-Lacrimal-Tooth malformations), LMS (Limb-Mammary syndrome), Hay Wells syndrome, SHFM (Split-hand/foot malformations), and Rapp-Hodgkin Syndrome, reviewed in [12]. These findings of a crucial role of p63 in skin development were supported by its specific expression pattern in epithelia [13]. There is also evidence that p63 may play a role in cancer since it has been shown to be overexpressed in squamous cell carcinomas [14]. In addition, the p63 proteins are transcription factors that modulate a large cluster of genes, which differ from p53 target genes [15]–[19]. One specific cluster of p63 target genes is related to cellular adhesion and contains either cell: matrix or cell: cell adhesion genes such as integrins, and components of the ECM, desmosomes or tight junctions.
Interpreting the potential role of p63 in cancer is complicated by the existence of six isoforms of the transcription factor. The p63 gene contains two promoters driving the production of distinct proteins, one with an N-terminal transactivation domain (TAp63) and one without (ΔNp63). Moreover, each of these two proteins may be expressed as α, β and γ isoforms, exhibiting particular expression patterns and characteristics. For example, ΔNp63α is abundantly expressed in proliferating keratinocytes of the epithelial basal layers, while the TA isoforms are not detectable in skin [20]. The ΔNp63α isoform is also increased in squamous cell carcinomas [13], thus supporting its proliferative functions. In contrast, TAp63 has recently been described as preventing premature aging by promoting adult stem cell maintenance in mice, linking its activities to DNA repair and wound healing rather than to differentiation [21]. Although this isoform of p63 is able to functionally interact with ΔNp63, p73 and p53, it also displays a specific role as transcriptional modulator of its own target genes such as the kinase inhibitor p57kip2 [21]. Roles of the various p63 isoforms as well as p53 and p73 in skin proliferation and differentiation are not yet fully understood. Additional complexity comes from the fact that they are able to compensate each other to some extent, as demonstrated for example by the increase of p63 expression in the skin of the p53 -/- mice [22]. They are also able to regulate each other, although these cross regulations are not yet fully understood and involve transcriptional as well as post transcriptional mechanisms [23].
Here we show that repression of the E6/E7 oncogenes induces transcriptional activation of a large set of potential p63 target genes in the HPV16-associated cervical carcinoma Caski cell line. We demonstrate, through siRNA silencing and direct over-expression of various p63 isoforms, that some of these genes are indeed p63 targets, mostly responsive to TAp63β and not to the constitutively expressed ΔNp63α and β isoforms. Furthermore we showed that the stability of the TAp63β isoform is down-modulated by E6, thus functionally linking repression of p63 target genes to expression of the E6 oncogene in cervical carcinoma cells. Over-expression of TAp63β or repression of E6, both induce increase in focal adhesion in cervical carcinoma cell lines. These data emphasize a new role of the TAp63β isoform in activating cell adhesion in epithelial cells and of the E6 oncogene in down-modulating this pathway in cervical cancer.
We wished to understand whether a p63-mediated adhesion pathway might be important in the HPV16-associated cervical carcinoma Caski cell line and how these cells express p63 compared to HeLa cells and non transformed human keratinocyte cell lines. Levels of mRNA for the ΔN and TA isoforms of p63 were measured by Real Time PCR in HeLa and Caski cancer cell lines, compared to the non HPV-associated keratinocyte cell lines HaCaT and N-Tert. Low levels of transcripts (around 28 ct) of the TAp63 isoforms were detected in all four cell lines, while transcript levels of the ΔNp63 isoforms varied from very low in HeLa cells (30 ct) to markedly higher levels in the HaCaT, N-Tert (around 18 ct) and Caski cells (around 20 ct) (Figure 1A). In HeLa cells, no endogenous p63 isoforms were detected at the protein level, as expected from the low levels of transcription (data not shown). We took advantage of this and used HeLa cells transfected with expression vectors of the different p63 isoforms as a comparison to study endogenous p63 protein levels in the other cell lines. Only 2 isoforms of ΔNp63 (α and β) were consistently detected in N-Tert, HaCaT and Caski cell extracts with an expression level of ΔNp63α between 1,5 and 2 times higher in non HPV-associated keratinocytes than in Caski (Figure 1B). Under similar conditions, TAp63 isoforms were not detectable in any of these cell extracts. However, low levels of TAp63 were detected with the use of a purified antibody against the TA domain of p63. Caski cells express mainly the TAp63β isoform while HaCaT cells express both β and γ isoforms and N-Tert and primary keratinocytes mainly the γ isoform with very low levels of the β isoform (Figure 1C).
In a previous study we described the modulation of putative p63 target genes in HeLa cells following HPV18 E6/E7 repression [5]. This occurred in the absence of detectable endogenous p63 expression; therefore we asked whether the range of p63 target genes modulated might be greater in Caski cells, which express significant levels of endogenous p63. A truncated form of the HPV protein E2 efficiently represses E6/E7 transcription in HeLa cells [4], and thus may be exploited as a reagent to study the effects of their inhibition. We used Affymetrix microarrays to study gene modulation in Caski cells infected by recombinant adenoviruses expressing GFP or the GFP-E2 protein truncated of its transactivation domain (E2C). As revealed by Real Time PCR, E2C repressed E6/E7 transcription by more than 90% under conditions where 100% of Caski cells were infected (Figure 2A). We also verified the activation of the endogenous p53 target gene p21 (8-fold) and the repression of the cell cycle gene cenpa (5-fold), that have previously been shown to be modulated by repression of E6/E7 in cervical carcinoma cells (Figure 2A) [5]. Gene expression analysis in E6/E7-repressed Caski cells revealed modulation of 825 genes: 463 repressed genes, including mitotic genes and E2F target genes (data not shown), and 362 genes whose activation was increased by more than 1.5-fold, with a p-Value less than 0.01 (data not shown). Using the TRANSFAC Professional data base (Biobase) we assigned 52 of these activated genes to the p53 pathway (Table S1) while 159 could be assigned to the p63 pathway (Table S2), with 36 genes overlapping with p53 (Figure 2B). In all, 123 exclusive p63 putative target genes were regulated following HPV16 E6/E7 repression. Of these, approximately 20% have previously been allocated to adhesion pathways [5], [15] (Table S3 in Text S1).
To confirm up-regulation of an adhesion pathway upon repression of E6 and E7 transcription, Caski cells infected with the adenovirus encoding GFP-E2C compared to the GFP control were subjected to gene expression analyses using a Human extracellular matrix and adhesion molecules PCR array (SABiosciences). Repression of E6/E7 transcription by E2C induced a strong response in genes involved in cellular adhesion in Caski cells, with 34% of the genes present in the PCR-arrays activated between 1.8 and 7-fold and 6% of genes repressed (Figure S1A in Text S1). Silencing E6/E7 with siRNA was also efficient, activating 53% of the PCR-arrays genes while only 3% were repressed (Figure S1B in Text S1). Among the genes contained within these arrays, 18 are known p63 targets. Half of these genes were activated in both conditions of E6/E7 repression.
As siRNA was also effective in repressing E6/E7 transcription in Caski cells (Figure S1B in Text S1), we used this technique to further appreciate the effects of the HPV oncogenes on the p63 pathway. We used E6/E7 siRNA to silence mRNA expression of both viral oncogenes together, since they are expressed from a single polycistronic mRNA and cannot be silenced individually [5]. Silencing of E6/E7 by about 60% did not affect levels of endogenous ΔNp63 mRNA (Figure 3A). This was confirmed by western blots, using a pan-p63 antibody (4A4), with pan-p63 siRNA (which silenced all p63 isoforms) and p53 siRNA as positive and negative controls respectively (Figure 3B upper panel). The p63 siRNA decreased p63 protein expression by about 50% while the p53 siRNA decreased p53 expression by 80% as quantified from the western blots shown in Figure 3B. In contrast, E6/E7 siRNA induced stabilization of the p53 protein, which was stabilized even in the presence of cotransfected p63 siRNA (Figure 3B lower panel). We then selected a group of the putative p63 target genes identified as activated in the microarrays from the list of genes given in (Table S3 in Text S1) and assessed their levels of transcription following siRNA silencing of E6/E7. As expected, these genes were activated between 3 and 4-fold following silencing of E6 and E7 transcription, thus confirming the microarray data (Figure 3C). Interestingly however, these experiments lead to the conclusion that HPV16 E6/E7 cause down regulation of putative p63 target genes without modulating ΔNp63.
The fact that E6/E7 silencing did not affect ΔNp63 expression while it stabilizes p53 (Figure 3B) led us to ask whether the cellular target genes were indeed being modulated by p63. Direct silencing of p63 by a pan-p63 siRNA induced up to 50% repression of the set of target genes, except for the p53 target gene p21 (Figure 3D), while silencing p53 did not strongly influence their expression except in the case of notch1 and p21 (Figure 3D). We decided to confirm whether activation of these genes following E6/E7 silencing occurred through p63 by concomitantly transfecting the two siRNA. Cotransfection of Caski cells with the E6/E7 and p63 siRNA induced a reversion of the gene activation seen with the E6/E7 siRNA alone, thus indicating that these genes are indeed mainly controlled by p63 (Figure 3E). Activation of p21 and both notch and ceacam1 was either not or partially reversed by cotransfection of p63 siRNA, confirming that p21 is mainly activated by p53 in Caski cells, while notch1 and ceacam1 are activated by both p63 and p53 as previously published for notch1 [24], [25]. Therefore, although this panel of genes is indeed activated by p63, the ΔN isoforms constitutively expressed in Caski cells may not be directly involved in this regulation, since their expression is not modulated upon E6/E7 silencing (Figure 3A and 3B upper panel). As the p63 siRNA was designed to repress both ΔN and TA isoforms of p63, we therefore examined a possible role for the TAp63 and specifically of the beta isoform that is expressed in Caski cells (Figure 1C).
We first measured the effect of the six over-expressed p63 isoforms [26] or p53 on the transcription of previously described groups of endogenous p63 and p53 target genes in transfected HeLa cells [5]. The endogenous p63 target genes were activated by ectopic expression of the longer TA α and β isoforms and not by the shorter γ isoform (Figure S2A in Text S1). The ΔN isoforms were less active, although the β isoform did exhibit some level of activation (Figure S2B in Text S1). None of these genes were activated by p53, but p21 was activated around 10-fold (Figure S2C in Text S1). While p53 did not activate p63 target genes, we found that the TAp63γ isoform efficiently activates p53 target genes including p21 (Figure S2D in Text S1). These results demonstrated that a cross-activation of the p53 target genes by p63 can occur, although with marked preference for the shorter TAp63γ isoform, as previously reported [27].
We then decided to confirm which p63 isoforms regulated the endogenous set of p63 target genes in Caski cells. We used recombinant adenoviruses expressing both α and β isoforms of TA and ΔNp63β that were previously found to be the most efficient in activating p63 target genes in transfected HeLa cells (Figure S2A and S2B in Text S1). Infection with an increasing multiplicity of infection (m.o.i.) of the recombinant adenovirus expressing the TAp63β isoform led to increased expression of the transcription factor (Figure 4A and 4B, upper panels), coupled with corresponding increases in mRNA expression of the endogenous set of genes (Figure 4C). Activation ranges from 2-fold for notch1 to 20-fold for ceacam1 with TAp63β (Figure 4C) while in contrast the ΔNp63β isoform (Figure 4A and 4B lower panels) did not induce significant changes in the basal mRNA levels of the p63 target genes, a maximum 2-fold change being seen for the itga2 and lamb3 gene expression (Figure 4D). The TAp63α long isoform was similarly inactive in this assay (Data not shown). Interestingly, we found that overexpression of the TAp63β isoform, but not of the ΔNp63β isoform, led to a decrease of about 3-fold of the endogenous constitutive level of the ΔNp63α isoform through an unknown mechanism (Figure 4B, upper panel). These results unambiguously establish that the set of genes modulated in our microarrays is composed of p63 target genes, likely activated most efficiently by the TAp63β isoform in Caski cells.
We then wanted to extend our finding to other cervical carcinoma cell lines and used SiHa also associated with HPV16. We first demonstrated that repression of E6/E7 transcription by E2C induced the same p63 target genes as in Caski cells as well as the p53 target gene p21 (Figure S3A in Text S1). We also infected SiHa cells with the recombinant adenoviruses expressing TA and ΔN p63β isoforms using conditions where 100 % of the cells were infected (Figure S3B in Text S1). This showed that expression of TAp63β and not ΔNp63β, activated the p63 target genes (Figure S3C in Text S1). Altogether these data strongly suggest that this new p63 pathway is conserved in cervical squamous cell carcinoma and that the adhesion genes are modulated by TAp63β preferentially with no clear involvement of the ΔNp63 isoforms.
We have shown that TAp63β is a transcriptional activator whose activity is repressed by the HPV oncogenes in Caski cells. Since expression of the viral oncogenes occurred through polycistronic mRNA as previously discussed, we could not separately repress E6 or E7 in our model system. However, the putative down modulation of TAp63β described here is reminiscent of the down modulation of p53 by E6 in cervical cancer [28]–[30], and we decided to verify the simpler hypothesis incriminating E6 rather than E7 in p63 modulation. We tested whether HPV18 E6 could degrade TAp63β in our system by cotransfection with an expression vector for HPV18 E6 into HeLa cells [29]. In the same experiment we also cotransfected p53 as a positive control and ΔNp63β to verify whether the TA isoform is preferentially degraded. We performed these experiments in the presence of cycloheximide to prevent protein synthesis. This revealed that expression of HPV18 E6 accelerated the degradation of TAp63β although often less efficiently than p53 (Figure 5A). The ΔNp63β isoform remained quite stable, at least for the first hour of cycloheximide treatment (Figure 5A). Thus E6 can induce degradation of the TAp63β isoform, whilst the ΔN isoform seems more resistant. Together with our finding that the TAp63β isoform is an activator of p63 target genes in Caski cells, its destabilization by E6, leading to transcriptional repression of the target genes, thus mimics the situation previously reported for p53 in cervical cancer.
The next step was to ask whether this degradation of TAp63β was mediated through interaction between TAp63β and E6 and via the ubiquitin ligase E6AP, as demonstrated for p53 [30]. Interaction between Myc-tagged E6 and TAp63β was readily detectable by coimmunoprecipitation in cotransfected HeLa cells treated with lactacystin to inhibit protein degradation. P53 was also coimmunoprecipitated with Myc-tagged E6, while the Myc-tagged Cyclin B, used as a negative control, did not coimmunoprecipitate with TAp63β (Figure 5B). We used an E6AP siRNA [31] to silence E6AP in HeLa cells cotransfected with E6 and TAp63β. Interestingly silencing of the E6AP ubiquitin ligase did not lead to stabilization of TAp63β (Figure 5C) while in the same cells, endogenous p53 was stabilized about 10-fold, back to its basal level in untreated cells (Figure 5C). This suggests the existence of another pathway for E6-mediated degradation of TAp63β which was however stabilized 3.5-fold by the proteasome inhibitor lactacystin as expected while p53 was stabilized 5-fold in the same experiment (Figure 5D).
We then needed to verify whether repression of E6 and E7 and subsequent stabilization of TAp63β could be detected in cervical carcinoma cell lines SiHa and Caski. We infected the two cell lines with the adenovirus expressing E2, as the best repressor of oncogenic transcription (repression of more than 90% compared to 60–70% of E6/E7 with siRNA) or adenovirus expressing GFP as a negative control. Extremely efficient repression of E6 and E7 induced a strong stabilization of 3-fold of the p53 protein, as expected, altogether with stabilization of 2 and 3.4-fold of the TAp63β isoform in Caski and SiHa respectively (Figure 5E). Ectopically expressed TAp63β appeared as a slightly lower migrating band due to insertion of a shorter version of the protein in the original expression vector as described [26] (Figure 5E). These data show that the low level of the TAp63β protein is due to its continuous degradation by E6, a situation reminiscent of p53 degradation. In the same Caski cells, we also observed slightly reduced expression (about 20% reduction) of the endogenous ΔNp63α isoform after E6/E7 repression and stabilization of TAp63β (Figure 5E) as demonstrated earlier in cells infected with the TAp63β recombinant adenovirus (Figure 4B, upper panel). However, this was not observed in SiHa cells where endogenous ΔNp63α is not detectable (Figure 5E). From these data we could draw a model whereby p63 target genes are activated by TAp63β which is targeted for degradation by E6; while ΔNp63α would modulate other target genes that are not studied here (Figure 5F). The overall effect of E6 expression in Caski is summarized by repression of the adhesion genes, represented by notch1, and activation of ΔNp63α, both by indirect mechanisms resulting from the pivotal degradation of TAp63β by E6 (Figure 5F).
The next question was whether activation of adhesion genes would stimulate phenotypic changes of TAp63β expressing cells. Cervical carcinoma SiHa cells infected with TAp63β or E2 recombinant adenoviruses acquired star-like structure together with enlargement of their cytoplasms (Figure 6A). In E2-expressing SiHa cells, these morphological changes correlated with stabilization of endogenous TAp63β as shown by immunofluorescence (Figure 6B) and western blot (Figure S5A in Text S1). Immunofluorescent labeling of vinculin and paxillin, markers of focal adhesion, showed an increase in number and size of the labeled foci at the membrane of E2-infected SiHa (Figure 7). In Caski cells, labeling of vinculin clearly showed an increase in focal adhesion in TAp63β and E2 expressing cells which was drastically reduced by silencing of p63 in E2 expressing cells, thus indicating that this increase in focal adhesion is mediated by TAp63 (Figure 8). An increase in focal adhesion was also seen in primary keratinocytes infected by the recombinant adenovirus expressing TAp63β, although these cells already present higher basal levels of focal adhesion than Caski (Figure S4 in Text S1), while in contrast no activation of TAp63β expression was detected in E2 expressing keratinocytes (Figure S5A in Text S1). Increase of focal adhesion and reorganization in cell membrane of E2 expressing SiHa and Caski cells were not accompanied by an increased expression of paxillin and involucrin, as shown by western blots (Figure S5B in Text S1). These results therefore point to an indirect role of TAp63β in modulating focal adhesion, which is in line with our previous findings since paxillin and vincullin were not found among the p63 transcriptional target genes. More work is needed however to fully characterize these phenotypic changes associated with expression of the TAp63β isoform and to understand the mechanism by which focal adhesion is modulated. Altogether, we have demonstrated here for the first time a strong involvement of the TAp63β isoform in formation of focal adhesion in epithelial cells and its down modulation by expression of the HPV E6 oncogene.
The p63 transcription factors are either activators or repressors of transcription depending on the isoforms expressed, the cell system and the target genes. However, repression of the viral oncogenes E6 and E7 in cervical carcinoma modulates p63 target genes in a unique way since an entire p63 target gene set was activated while none were repressed. It should be noted that activation of the p63 target genes in our system was generally lower than that of p53 target genes, as if the regulatory pathways leading to modulation of the p63 targets were less homogeneous. Indeed, p63 is expressed as various isoforms which can interact with each other and which share similar binding sites together with p53. This has led to an intricate situation where activators and repressors are not well defined. For instance, ΔNp63 is usually recognized as a repressor although it also contains a specific activator domain [32], [33], while TAp63 is mainly described as an activator due to the presence of the long N-terminal transactivation domain. However, the longest TAp63α isoform also contains a C-terminal domain which is a repressor of the TA domain, thus leading to inefficient transcriptional activation [34]. The most efficient activator so far described in the literature has been the shortest isoform TAp63γ but this was demonstrated with a group of p53 target genes [27] as confirmed in our system. In contrast, we demonstrated here that TAp63β can activate p63 target genes in a specific manner in Caski and SiHa cervical carcinoma cell lines. This finding is unexpected since the TAp63 isoforms have not been detected in human epithelial cells and had not been assigned specific roles in skin development [35], although TAp63 was recently shown to play an important role in regulating cellular senescence and genomic instability [21]. Its depletion in skin induces premature aging due to uncontrolled multiplication of skin stem cells which is in line with a putative oncogenic role of TAp63β depletion in cervical carcinomas.
Another major finding of our work is the fact that the HPV18 E6 oncogene can accelerate degradation of the TAp63β isoform, although independently of the E6AP ubiquitin ligase which is implicated in degradation of p53 [30]. TAp63 is known to be degraded through the proteasomal pathway via its amino-terminal TA domain [36], [37], independent of MDM2 [36]. The Itch ubiquitin ligase has been suggested as one of the major factors controlling the stability of both the TA and ΔN p63 isoforms [38] although it degrades the ΔN isoforms more efficiently [39], [40]. As for the putative roles of viral oncoproteins, it has been reported that they are unable to interact with p63 or to noticeably modify p63 activity [41]. Our work presents evidence that the high risk HPV18 E6 oncoprotein can bind TAp63β and accelerates its degradation. However and significantly, E6 uses a different pathway than for the degradation of p53 and the ubiquitin ligase involved in TAp63β degradation has yet to be identified. A striking consequence of this finding is that repression of the endogenous oncogenes induces stabilization of endogenous TAp63β together with p53, in all cervical carcinoma cell lines that we have examined including HeLa, Caski and SiHa. Another consequence of our findings is that ΔNp63 does not seem implicated in regulation of the anchorage independent growth in cervical cancer which is surprising in light of its pivotal role in skin homeostasis. In cervical carcinoma cells, stabilization of TAp63β induces an increase in focal adhesion. In keratinocytes, we could also show that overexpression of the β isoform increased focal adhesion. However the basal level of focal adhesion is higher in keratinocytes compared to cervical carcinoma cells and the best expressed TAp63 in these cells is the γ isoform. We infer from these data that the γ isoform is also probably able to modulate focal adhesion but more work is needed to clarify this point.
The p53/p63 family of transcription factors also includes p73 and its various isoforms [42], which share identical binding sites with p63 and can form heterotetrameric factors to concomitantly bind the same sites if expressed at the same time [43]. The p73 transcription factors however, are not expressed in skin and have no known roles in its homeostasis, while they play roles in neurogenesis, inflammation, sensory pathways and osteoblastic differentiation [44] as well as in genomic stability and tumor suppression [45]. Since the present study mostly describes modulation of a group of target genes involved in adhesion pathways that are therefore directly linked to skin maintenance and differentiation, it is presumed that p73 should not play a dramatic role here. We cannot exclude however that some cross-regulation could take place.
We have shown here that TAp63β activates genes involved in cell adhesion and induces phenotypic changes in focal adhesion in keratinocytes upon its ectopic expression. The bovine papillomavirus E6 protein has previously been shown to modulate focal adhesion through direct interaction with paxillin [46], [47]. We cannot exclude that HPV E6 could also directly influence cell adhesion in cervical carcinoma cells. However, we show here that E6 induces a strong modulation of focal adhesion through degradation of TAp63β. This novel role of E6 thereby participates in oncogenic transformation of cervical carcinoma cells through induction of anchorage independent growth. TAp63β therefore appears as a tumor suppressor in our system, which is in agreement with its recently reported roles in mouse models [21], [48], [49]. This is, to our knowledge, the first time that TAp63β is implicated in a human cancer, cervical carcinoma, which is one of the leading causes of death by cancer in women worldwide.
Recombinant adenoviruses expressing green fluorescent protein (GFP) alone, GFP fused to HPV18 E2 protein deleted of its transactivation domain (E2C) or E2 full-length were previously described [50]. Recombinant adenoviruses expressing TAp63β or ΔNp63β isoforms were prepared from the pIRES expression plasmids containing the TAp63β or ΔNp63β gene followed by the IRES sequence and the GFP gene. Recombinant adenoviruses expressing the TA and ΔNp63α isoforms were a kind gift of Barry Trink [26].
Cells were infected with purified adenoviruses expressing GFP, GFP-E2 or GFP-E2C at a multiplicity of infection (m.o.i.) of 100 or 200 PFU/cell or with adenoviruses expressing GFP, TAp63β or ΔNp63β at an m.o.i. of 20, 50 or 100 PFU/cell, as described previously [4].When needed, 24 hours after infection, cycloheximide was added at 100 µg/ml for up to 2 h and/or lactacystin was administered at 10 µM for up to 6 hours.
HeLa, Caski and SiHa cells were grown using standard procedures in a 37°C humidified incubator with 5% CO2. All cell lines were grown in high-glucose Dulbecco's modified Eagle's medium (Invitrogen) with 10% heat-inactivated fetal bovine serum.
Plasmids were transfected using FuGENE HD (Roche) according to the manufacturer's instructions. Expression plasmids for the six p63 isoforms were kindly provided by Barry Trink.
For siRNA transfection, DharmaFECT 1 (Thermo Scientific) was used according to the manufacturer's instructions.
The siRNA sequences used were:
HPV18 E6/E7 siRNA: GCAUGGAGAUACACCUACAdTdT
HPV16 E6/E7 siRNA: GAGCUGCAAACAACUAUACdTdT
Control siRNAs used in HPV16 positive cells (Caski) and in HPV18 positive cells (HeLa) were HPV18 E6/E7 siRNA and HPV16 E6/E7 siRNA respectively.
For p63 siRNA (which targets all p63 isoforms) and p53 siRNA, we used ON-TARGETplus SMARTpools (Thermo Scientific), which are pools of 4 different siRNAs for each gene with the following sequences:
p63 siRNAs: (UCUAUCAGAUUGAGCAUUA; GCACACAGACAAAUGAUU; CGACAGUCUUGUACAAUUU; GAUGAACUGUUAUACUUAC)
p53 siRNAs: (GAAAUUUGCGUGUGGAGUA; GUGCAGCUGUGGGUUGAUU; GCAGUCAGAUCCUAGCGUC; GGAGAAUAUUUCACCCUUC).
The high density microarrays used in this study were the Human Gene 1.0 ST array from Affymetrix, including 28,869 well-annotated genes with 764,885 distinct probes. A total of 200 ng of RNA for each experiment was amplified, terminally labeled and used for hybridization according to the manufacturer's instructions (Affymetrix). The microarrays were washed and stained using Fluidics Station (GenChip) and scanned using a GeneArray 25000 Scanner. Data collected from each hybridization experiment were analyzed and statistical analyses carried out using Partek Genomics Suite software.
Total RNA was extracted using RNAEasy Mini Kit (Qiagen) and a total of 2.5 µg of RNA were reverse transcribed with Superscript II (Invitrogen) according to the manufacturer's instructions. From the resulting synthesized single-stranded cDNA, 1/100 was used for each Real-time PCR in the presence of 1 µM of specific primers and Syber Green master mix (Applied Biosystems). Quantitative PCR were performed with an MX3005P sequence detection system (Stratagene). Each cDNA was normalized to histone deacetylase 1 (HDAC1), glyceraldehyde-3-phosphate dehydrogenase (GADPH), and 18S ribosomal RNA. PCR were performed in duplicate and fitted to standard curves, providing mean cycle threshold values that were translated into arbitrary units corresponding to mRNA levels. Data were analyzed with MxPro v.4.00 software (Stratagene).
Primers:
GADPH: 5′-TCCATCACCATCTTCCAGG-3′; 5′-CATCGCCCCACTTGATTTTG-3′
HDAC: 5′-TTTTCAAGCCGGTCATGTCC-3′; 5′- CCGCACTAGGCTGGAACATC-3′
18S: 5′-TGCGAATGGCTCATTAAATCAGT-3′; 5′-AGAGGAGCGAGCGACCAA-3′
16E6E7: 5′-GCTCAGAGGAGGAGGATGAAATAG-3′; 5′-TCCGGTTCTGCTTGTCCAG-3′
18E6E7: 5′-CCCCAAAATGAAATTCCGGT-3′; 5′-GTCGCTTAATTGCTCGTGACATA-3′
pan-p63: 5′-GACAGGAAGGCGGATGAAGATAG-3′; 5′-TGTTTCTGAAGTAAGTGCTGGTGC-3′
TAp63: 5′-AAGATGGTGCGACAAACAAG-3′; 5′-AGAGAGCATCGAAGGTGGAG -3′
ΔNp63: 5′-TCAATTTAGTGAGCCACAGTAC-3′; 5′-CTGTGTTATAGGGACTGGTG-3′
p53: 5′-TTCGACATAGTGTGGTGGTGC-3′; 5′-AGTCAGAGCCAACCTCAGGC-3′
CENPA: 5′-TCAGAGTAGCCTCACCATTAGTGG-3′; 5′-AGCTACACATCCGTTGACAAGC-3′
CTGF: 5′-CCCTGCATCTTCGGTGGTA-3′; 5′-AGGCACGTGCACTGGTACTTG-3′
ITGA2: 5′- CTTTGGTTAGCCTTGCCTTAGG-3′; 5′- CCCCTGCAAGGAAGAATCAC-3′
LAMB3: 5′-GGGCACTCAGAGACATGTCACTT-3′; 5′- TGACACCGCTCACAGTTCTTG-3′
CEACAM1: 5′-CCAAATCAAAGCCAGCAAGACCAC-3′; 5′-TCATTTGTGGAGCAGGTCAGGTTC-3′
Notch1: 5′- GAGTGGGACCAACTGTGACAT-3′; 5′-CCGTTGACACAAGGGTT-3′
FAT: 5′- CAACCTTCCCTACTACGCCG-3′; 5′- ACATGGCCCACCTCAGTGTC-3′
FN1: 5′-GTGTGTTGGGAATGGTCGTG-3′; 5′-AAGCTGCGAGTAGGCAATGC-3′
p21: 5′-GCGACTGTGATGCGCTAATG-3′; 5′-CGGTGACAAAGTCGAAGTTCC-3′
All Real time PCR data are represented as mean ± SD. Data were analyzed using paired t test for comparison between two groups. P Values under 0.05 were considered significant. All experiments were done at least three times.
Infected or transfected cells were collected 24 or 42 h post-infection or post-transfection, and protein extracts were subjected to electrophoresis before transfer onto nitrocellulose membranes. The membranes were then incubated with antibodies specific for: p63 (4A4, BD Pharmingen), TAp63 purified antibody (Biolegend Poly 6189) or the monoclonal TAp63, a kind gift of Franck McKeon, [35], p53 (DO-1, Santa-Cruz), GFP (TP401; Torrey Pines Biolabs) or beta-actin (A2066, Sigma); and either mouse or rabbit secondary antibodies coupled to peroxidase. Protein bands were revealed using the Amersham ECL plus kit. Quantification of western blots was done by measuring the relative intensity of the bands compared to internal controls (GFP and or actin), the values given are in arbitrary units.
Hela cells were co-transfected with TAp63β and Myc-E6 HPV18 or Myc-CyclinB2, as a negative control. As positive control, HeLa cells were cotransfected with Myc-E6 HPV18 and p53 expressing vectors. 20 h post-transfection, cells were treated with 10 µM of lactacystin for 6 h. Cells were harvested and proteins were extracted in an extraction buffer (300 mM NaCl, 50 mM Tris-HCl [pH 8], 0.5 % Nonidet P-40 [NP-40], 1 mM EDTA, protease and phosphatase inhibitors) for 30 min at 4°C, followed by centrifugation. Extracts containing 1 mg of total proteins were immunoprecipitated with the rabbit anti-myc antibody (A-14; Santa Cruz) 5 h at 4°C. After centrifugation, beads were washed four times with 1 ml of extraction buffer. Proteins from the beads and input (1/40 of total proteins) were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and analyzed in Western blot experiments. For Western blot experiments, mouse anti-Myc (9E10; Santa Cruz), mouse anti-p53 (DO-1; Santa Cruz) and mouse anti-pan-p63 (4A4, BD Pharmingen) antibodies were used.
Cells grown on cover slips were fixed at 4°C for 30 min with 2% paraformaldehyde (v/v) and washed in PBS. Cells were then permeabilized for 30 min at room temperature in PBS, 2% serum, 0.1% triton. After washing with PBS, the cells were incubated with primary antibodies for 1 h at room temperature, washed 3 times with PBS, 2% serum, followed by secondary antibodies for 1 h at room temperature. After washing with PBS, cell nuclei were stained with DAPI (Invitrogen). Antibodies against paxillin (clone 5H11, Abcam ab3127) and vinculin (clone hVIN-1, Sigma) and TAp63 purified antibody (Biolegend Poly 6189) were used. Secondary antibodies were fused to Alexa fluor 568 (Invitrogen). Confocal fluorescent images were obtained by a Zeiss confocal laser scanning microscope with a 40X objective.
p63: *603273, p53: *191170, pRb: +180200, p73: *601990, E2F1: *189971, p57kip2: *600856, p21: *116899, CENPA: *117139, CTGF: *121009, ITGA2: +192974, CEACAM1: *109770, Notch1: *190198, COL5A1: *120215, LAMB2: *150325, FAT: *600976, FN1: +135600, FBN1: *134797, GDF15: *605312, Sestrin: *606103, RRM2B: *604712, E6AP: *601623, Paxillin: *602505, Vinculin: *193065, MDM2: +164785, GADPH: *138400, HDAC: *601241, 18S: *180473.
|
10.1371/journal.ppat.1005082 | An siRNA Screen Identifies the U2 snRNP Spliceosome as a Host Restriction Factor for Recombinant Adeno-associated Viruses | Adeno-associated viruses (AAV) have evolved to exploit the dynamic reorganization of host cell machinery during co-infection by adenoviruses and other helper viruses. In the absence of helper viruses, host factors such as the proteasome and DNA damage response machinery have been shown to effectively inhibit AAV transduction by restricting processes ranging from nuclear entry to second-strand DNA synthesis. To identify host factors that might affect other key steps in AAV infection, we screened an siRNA library that revealed several candidate genes including the PHD finger-like domain protein 5A (PHF5A), a U2 snRNP-associated protein. Disruption of PHF5A expression selectively enhanced transgene expression from AAV by increasing transcript levels and appears to influence a step after second-strand synthesis in a serotype and cell type-independent manner. Genetic disruption of U2 snRNP and associated proteins, such as SF3B1 and U2AF1, also increased expression from AAV vector, suggesting the critical role of U2 snRNP spliceosome complex in this host-mediated restriction. Notably, adenoviral co-infection and U2 snRNP inhibition appeared to target a common pathway in increasing expression from AAV vectors. Moreover, pharmacological inhibition of U2 snRNP by meayamycin B, a potent SF3B1 inhibitor, substantially enhanced AAV vector transduction of clinically relevant cell types. Further analysis suggested that U2 snRNP proteins suppress AAV vector transgene expression through direct recognition of intact AAV capsids. In summary, we identify U2 snRNP and associated splicing factors, which are known to be affected during adenoviral infection, as novel host restriction factors that effectively limit AAV transgene expression. Concurrently, we postulate that pharmacological/genetic manipulation of components of the spliceosomal machinery might enable more effective gene transfer modalities with recombinant AAV vectors.
| Mammalian cells have developed diverse innate/intrinsic immune strategies to counteract viral infections. Post-entry infection steps of a single-strand DNA virus, adeno-associated virus (AAV), are subject to such restrictions. Here, we screened an siRNA library to identify a novel cellular factor involved in AAV restriction. We found PHF5A, a component of the U2 snRNP mRNA splicing factor, blocks expression from recombinant AAV vectors. Disruption of PHF5A expression specifically enhanced AAV vector performance. Moreover, genetic and pharmacological inhibition of other U2 snRNP proteins, but not spliceosome proteins involved in other splicing steps, strongly increased transgene expression from AAV vectors. Further study demonstrated that U2 snRNP proteins recognize incoming AAV capsids to mediate this cellular restriction at the step after second-strand synthesis. In summary, we identify the U2 snRNP spliceosome complex as novel host factors that effectively restrict recombinant AAV vectors. Considering frequent reorganization of host splicing machinery in DNA virus infections, it is conceivable that U2 snRNP plays a role as a broad spectrum antiviral factor and helper viruses have evolved to counteract this restriction through sequestration of snRNP proteins.
| Viral pathogens are known to reorganize different components of the host cell machinery during the course of infection. For instance, adenoviruses have been shown to induce nuclear reorganization of host splicing factors and mislocalization of the DNA damage response machinery [1]. Similarly, herpesviruses can induce sequestration of cellular chaperone proteins and the 26S proteasome in nuclear foci to facilitate quality control during replication [2]. Adeno-associated viruses (AAV) are helper-dependent parvoviruses that have evolved strategies to replicate efficiently by exploiting host cell co-infection by adenoviruses or herpesviruses [3]. The infectious pathway of wild type AAV and recombinant AAV vectors consists of multiple stages starting with cell surface receptor binding, followed by endocytosis, endosomal escape, nuclear import, second-strand synthesis, and subsequent expression of the vector-encoded transgene [4,5]. The post-entry steps leading to AAV transduction are particularly subject to restriction by cell intrinsic factors [6,7,8]. Studies have identified impaired AAV vector transduction due to inefficient nuclear import [6], uncoating of vector genomes [7], or second-strand synthesis [8,9]. Treatment with proteasome inhibitors has demonstrated improved transduction by AAV vectors [10,11], suggesting the involvement of proteasomal degradation pathways in restricting AAV transduction. Nevertheless, modest increases in accumulation of viral DNA following proteasomal inhibition cannot solely account for substantial increases in AAV transduction [12], and the underlying mechanism remains elusive. Other host factors such as the FKBP52 [13], Mre11/Rad50/Nbs1 complex [14,15], APOBEC3A [16] and more recently, TRIM19/promyelocytic leukemia protein (PML) [17] have been shown to inhibit AAV replication by blocking second-strand synthesis.
AAV has emerged as a promising vehicle to achieve long-term gene expression with low toxicity. Recombinant vectors based on naturally occurring AAV serotype capsids and libraries of engineered capsid mutants have demonstrated unique receptor usages and tissue tropisms, providing versatility for tissue-targeted gene expression [18,19,20,21,22]. For instance, AAV vectors with AAV serotype 9 (AAV9) capsid efficiently transduce cardiac tissues, while vectors with AAV2 capsid show efficient transduction of kidney cells [23,24,25]. Importantly, recent phase I and phase II clinical trials using AAV vectors have established their safety, in some cases, with notable clinical benefits [26,27,28,29,30], opening the door to AAV vector gene therapies for various human disease conditions. Currently, however, efficient gene transduction by AAV vectors typically requires high doses of vectors. This presents a major barrier for the widespread use of AAV vectors in the clinic, due to potential challenges in manufacturing clinical grade vectors for high dose studies as well as the increased risk of eliciting host immune responses or inducing insertional mutagenesis at high vector doses [31,32,33]. Improving AAV vector transduction efficiency would reduce vector doses required for efficient gene delivery, minimizing the risks associated with high dose AAV vectors. The current report is focused on the identification of novel host restriction factor(s) that limit expression from AAV vectors as well as proof-of-principle studies that would enable effective gene therapy with lower vector doses in clinical trials.
We screened an siRNA library, which covers 600 known and putative human genes in the ubiquitin and proteasome pathways, for AAV vector transduction. We identified 12 candidate genes (Fig 1A). Disruption of those genes in HeLa cells increased luciferase expression by an AAV9 vector, AAV9 CMV-Luc, over 10-fold (Fig 1A). Further verification with distinct siRNAs and lenti-shRNA vectors found disruption of PHF5A, RAB40A and PRICKLE4 reproducibly increased AAV9 transduction. Treatment of HeLa cells with two PHF5A siRNAs led to over 80% reduction in PHF5A transcripts (Fig 1B) and increased the transduction by AAV9 vectors up to 12-fold (Fig 1C). In contrast, disruption of PHF5A expression did not strongly enhance luciferase expression of adenoviral or HIV-based lentiviral vectors (Fig 1D). Similar results were observed upon disruption of RAB40A and PRICKLE4 (S1 Fig).
To rule out possible off-target effects of siRNA, we generated a lentiviral vector expressing an siRNA-resistant, HA-tagged PHF5A mutant, PHF5A-HA-Escape (Fig 1E). When endogenous PHF5A expression was disrupted by the PHF5A siRNA, two independent HeLa cell lines with stable PHF5A-HA-Escape expression (Fig 1E, right panel) did not show enhanced AAV9 vector transduction (Fig 1F). Thus, the increased expression from AAV9 vector by the PHF5A siRNA is PHF5A-specific, but not due to off-target effects.
The AAV CMV-Luc vector construct used in the library screening contained a human beta globin intron. To rule out the possibility of PHF5A modulating the CMV promoter activity or the intronic unit, we first replaced the CMV promoter and intron sequence in the AAV vector genome with an intron-less retroviral SFFV promoter. Disruption of PHF5A increased transduction by multiple AAV serotypes (Fig 2A), indicating that the PHF5A-mediated restriction was independent from internal promoters or receptors used by AAV vectors. Likewise, knocking down PHF5A was effective at increasing AAV vector transduction in other cell types, including A375 melanoma cells and primary cardiac fibroblasts (Fig 2B).
Next, we examined the influence of PHF5A ablation on multiple stages of AAV vector transduction. No notable effects were observed on AAV cellular or nuclear entry (Figs 2C and S2). Additionally, approximately 30% of total AAV DNA detected was DNase-resistant at 24 hours post infection (p.i.) (Fig 2D), indicating that PHF5A does not affect uncoating process of AAV vectors. Southern blot analysis demonstrated no notable increase in double-stranded-monomers in cells pretreated with the PHF5A siRNA (Fig 2E). Upon transduction with a GFP-expressing self-complementary AAV (scAAV) vector, which does not rely on second-strand synthesis for transgene expression, we found significant increases in GFP-expressing cell populations in HeLa cells treated with the PHF5A siRNA (Fig 2F). These results indicate that PHF5A blocks the process of AAV vector transduction after second-strand synthesis. We then explored the effects of PHF5A disruption on the transcription of AAV9 CMV-Luc vector. Northern blot analysis showed that pretreatment with the PHF5A siRNA increased the levels of luciferase-specific transcripts (Fig 2G and 2H), suggesting that PHF5A affects the step before translation. When HeLa cells were transfected with the AAV vector genome plasmid, pAAV CMV-Luc, or single-stranded AAV vector genomic DNA from purified AAV vector particles, PHF5A ablation caused no increase in luciferase expression of transfected viral genome (Fig 2I). Together, this suggests that PHF5A acts to restrict AAV vector transduction somewhere between AAV second-strand synthesis and the transcription of the AAV vector transgene. It also appears that PHF5A does not directly target AAV vector genome. Additionally, introduction of disruptive mutations in any of the three GATA-type zinc finger motifs in PHF5A led to the loss of anti-AAV activity (S3 Fig).
PHF5A has been reported to interact with various proteins, including the U2 snRNP proteins, SF3B1, SF3B2, SF3B3 [34,35,36], U2AF1, ATP-dependent helicases EP400 and DDX1, and arginine-serine-rich domains of splicing factor SFRS5 [36]. Additionally, through co-immuno-precipitation of HA-tagged PHF5A, we identified potential PHF5A-interacting proteins, including FUS, EEF1, EEF2 and HIST1H4B. To further understand the underlying mechanism, we assessed the effects of disrupting those proteins on expression from AAV vectors. After verification of reduction in corresponding transcripts upon transfection of specific siRNAs (S4A Fig), siRNA-treated cells were infected with AAV9 CMV-Luc vectors at 24 hours post transfection, with luciferase activity assayed 48 hours p.i. Ablation of U2 snRNP components and U2 snRNP-associated factor (U2AF1) resulted in a substantial increase in luciferase activity relative to HeLa cells pre-treated with a control siRNA (Figs 3A and S4B). Disruption of HIST1H4B, one of histone H4 genes, also showed a modest increase, while ablation of other factors showed no notable effect. Of note, disruption of spliceosome proteins involved in other splicing steps, including SNRNP200 and PRPF31, essential factors for U4/U6-U5 formation and function, did not increase the AAV vector transduction (S4A and S4B Fig). These results suggest that PHF5A blocks AAV vector transduction through an interaction with U2 snRNP proteins and associated U2AF1, independently of cellular RNA spliceosome function. Similar to the effects of PHF5A knockdown, disruption of U2 snRNP components or U2AF1 did not enhance the luciferase expression from an adenoviral vector or a transfected AAV vector plasmid, pAAV CMV-Luc (Fig 3B and 3C). Taken together, we conclude that infectious AAV particles and all steps in intracellular trafficking pathway are essential for the restriction of transduction by U2 snRNP and associated proteins.
To further confirm the role of U2 snRNP proteins in the restriction of AAV vectors, we assessed the influence of pharmacological inhibition of U2 snRNP on expression from AAV vectors. One drug we employed was meayamycin B, a potent SF3B1 inhibitor, synthesized according to the literature [37]. When HeLa cells were pre-treated with this drug at an increasing dose 3 hours before AAV9 vector infection, dose-dependent increases (up to 49-fold) in relative luciferase activity were seen (Fig 3D). As we reported previously, treatment of HeLa cells with over 20 nM of meayamycin B for two days showed cytostatic effects [37]. A related SF3B1 inhibitor, meayamycin, also demonstrated a substantial increase in AAV transduction (S4C Fig). In contrast, other drugs reported to block other splicing steps, including isoginkgetin at the prespliceosome/A complex stage [38] and 3-Aminophenylboronic acid (ABA) at the second stage (excision of the lariat intron), did not show notable increases in AAV vector transduction (Figs 3E and S4D) although isoginkgetin blocked cellular mRNA processing (Fig 3F). Of note, a low dose (5 nM) meayamycin B treatment substantially enhanced AAV vector transduction without strongly affecting mRNA splicing (Fig 3D and 3F), indicating that inhibition of the general splicing process is not necessary to enhance AAV vector transduction. Additionally, we found that pretreatment with the drug is not needed in order for it to enhance AAV vector infection (Fig 3G). The largest increase in luciferase activity (464-fold) was observed when cells were treated by meayamycin B 9 hours p.i. In contrast, treatment at 33 hours post p.i. showed relatively weak effects. To further map the optimal timing of U2 snRNP inhibition for AAV vector transduction, we treated AAV2 and AAV9 vector-infected HeLa cells with 5 nM meayamycin B at various time points and duration, and assessed luciferase activity at 3 days p.i. Treating with meayamycin B 3 hours p.i. and washing cells 1, 2 or 3 days after receiving the drug resulted in similarly high levels of enhanced luciferase expression (S4E Fig). Washout of meayamycin B for 48 hours after 3–24 hours of treatment did not strongly compromise its effects on AAV vector transduction. In contrast, the effects of meayamycin B on AAV vector transduction were impaired when drug was added either 24 or 48 hours p.i. (S4E Fig). Thus, optimal enhancement of AAV vector transduction requires initiation of U2 snRNP inhibition prior to 24 hours post AAV vector infection. This indicates that U2 snRNP blocks AAV vectors at a particular post-entry step of viral infection, likely occurring before 24 hours p.i. Similar to PHF5A disruption, meayamycin B also enhanced the transduction by both single-stranded AAV and scAAV vectors through increasing the number as well as the fluorescent intensity of GFP-positive cells (Fig 3H). In addition, dual treatments with meayamycin B and PHF5A or SF3B1 siRNAs showed no additional impact on the AAV vector transduction (Figs 3I and S4F), verifying that meayamycin B and PHF5A/SF3B1 target a common pathway. On the other hand, dual treatment with a proteasomal inhibitor MG132 and meayamycin B showed a synergistic effect (Fig 3J), indicating that the U2 snRNP proteins block AAV restriction independently from the proteasomal pathway. We also tested the influence of adenovirus co-infection on the U2 snRNP-mediated restriction of AAV vectors. Although human adenovirus 5 (Ad5) co-infection alone enhanced AAV2 transduction 30-fold (Fig 3J), dual treatments of Ad5 co-infection and SF3B1 knockdown showed minimal additive effects on AAV transduction (Fig 3K). Similar results were observed with meayamycin and Ad5 dual treatments (S4G Fig). These results strongly suggest that Ad5-mediated activation of AAV vector transduction is through U2 snRNP inhibition. Indeed, when the influence of Ad5 infection on subcellular localization of U2 snRNP was determined, Ad5 infection showed notable PHF5A and SF3B1 displacement in HeLa cells (Figs 3L and S4H). On the other hand, meayamycin B treatment failed to increase AAV vector production, or rescue AAV vector production in the absence of the adenovirus helper plasmid (S4I Fig), suggesting that U2 snRNP inhibition is not sufficient to provide adenoviral helper function during AAV production.
Next, we assessed the interaction between PHF5A and AAV vector components. Upon pull-down of HA-tagged PHF5A from AAV vector-infected cells (Figs 4A and S5), total DNA in the pellets was assessed for AAV vector genomes. Quantitative real-time PCR detected significantly more AAV vector genome DNA in the HA pulldown from PHF5A-HA-over-expressing cells as opposed to control cells (Fig 4B). We also tested the influence of heat-mediated conformational changes of viral capsids, which lead to the exposure of the hidden VP1 N-terminal and viral genomic DNA [39,40], on the interaction with PHF5A. A three-fold increase in AAV genome copies was detected in the HeLa-PHF5A-HA pulldown when the cell lysates were incubated with pre-heated AAV2 CMV-Luc vector than non-heated vector (Fig 4C). Those data indicate interaction of PHF5A with AAV vector genome, directly or indirectly.
Next, immunohistochemistry was performed in order to identify the subcellular localization of PHF5A and AAV capsid. The A20 anti-AAV2 capsid antibody was used to detect AAV capsids (Fig 4D). The majority of AAV capsid signals were found in the cytoplasm at 4 and 12 hours p.i. (Fig 4D). Although endogenous PHF5A was predominantly found in the nucleus, especially in the nucleoli, of uninfected cells (Fig 4D), a notable increase in cytoplasmic PHF5A signals was found in AAV2 vector-infected cells at 1 and 4 hours p.i. (S6A Fig). Of note, AAV2 capsid signals frequently co-localized with the cytoplasmic PHF5A body signals (Fig 4E). When HeLa cells were exposed to an empty AAV2 vector, similar cytoplasmic recruitment of PHF5A to AAV2 capsids was also evident (Fig 4E, lower panels), suggesting that the PHF5A translocation (or de novo recruitment in the endosome) to AAV2 capsids is independent of AAV vector genome. Analysis of Z-stack images of AAV-infected cells at 4 hours p.i. showed comparable perinuclear and nuclear accumulation of AAV capsids between control and the PHF5A-siRNA treatment, further supporting that PHF5A does not affect AAV vector trafficking and nuclear import (Figs 4F and S7).
When the subcellular localization of the major U2 snRNP component, SF3B1, was assessed, nuclear speckles in the nucleoli and a diffuse cytoplasmic signal were observed (S8A Fig). Although the widespread SF3B1 signals often overlapped with the AAV2 capsid signals, the diffuse cytoplasmic signal made it difficult to verify co-localization with AAV2 vector particles (S8B Fig). To validate the AAV2 capsid and SF3B1 interaction, we employed the gradient technique to separate free-SF3B1 forms from particulated AAV capsid-associated SF3B1 (Fig 4G), which was used to determine the interaction between retroviral capsids and a cytoplasmic retroviral restriction factor TRIM5alpha [41]. Upon ultracentrifugation through the 25% iodixanol layer, AAV2 capsid proteins VP1, VP2 and VP3 (87, 72, 62 kDa) were detected in the pellets (Fig 4H). When the same samples were probed for SF3B1 (175 kDa), endogenous SF3B1 was seen in the input and upper layer samples of untreated HeLa cells (Fig 4H). In AAV2-treated lysates, additional SF3B1 bands were seen in the pellet (Fig 4H, S8C and S8D Fig). When the co-precipitation of PHF5A with AAV2 capsid was assessed, no intact PHF5A protein bands (15 kDa) were detected after incubation at 4°C for 1 hour, likely due to its instability (Fig 4H). Instead, multiple high molecular weight signals were detected in the input and top layer samples by anti-PHF5A and anti-HA antibodies (Figs 4H and S8D). Notably, high molecular weight PHF5A signals were detected in the pellet of AAV2-treated samples but not in the pellet of untreated samples. It is possible that PHF5A is modified (or modifies other proteins) upon interaction with AAV components. To further map the responsible region for the interaction between SF3B1 and AAV2 capsid, we performed the same experiments using empty AAV capsid, made with the VP3 protein alone. The VP3-only capsid was able to enrich SF3B1 in the pellets (Fig 4I). In contrast, multiple attempts to enrich SF3B1 or PHF5A through pulling down non-assembled VP1 proteins were unsuccessful (Fig 4J). These results suggest that U2 snRNP proteins interact with the AAV2 capsid structure within the VP3 region, but not AAV2 vector genomic DNA or non-assembled AAV2 VP1 protein.
Finally, we tested the ability of meayamycin B to boost AAV transduction in various cell types, relevant to gene therapy applications. When primary pancreatic islets were transduced with AAV8 CMV-GFP and treated with 2 nM meayamycin B 3 hours p.i., there were increased numbers of GFP expressing cells in drug treated mouse islets (Fig 5A). When primary human pancreatic islets were infected with AAV2 or AAV9 CMV-Luc vectors and treated with 0, 2, 5, or 20 nM meayamycin B at 7 hours p.i., we found dose-dependent increases in luciferase expression in AAV2 and AAV9 infected cells (Fig 5B). Likewise, meayamycin B treatment increased AAV2 and AAV9 transduction of primary neonatal rat cardiomyocytes as well as porcine hepatocytes (Fig 5C and 5D). These results demonstrate that meayamycin B enhances AAV vector transduction of a variety of cell types from different host species. Although we typically observed no notable toxicity in primary cells treated with 5 nM meayamycin B, prolonged treatment with over 10 nM meayamycin B showed anti-proliferative effects as we reported previously [42]. Since meayamycin B is rapidly cleared from circulation by unknown mechanism(s) [42], we were unable to evaluate drug doses high enough to test the impact on AAV vector transduction in vivo.
Here, we have demonstrated that PHF5A and U2 snRNP proteins, such as SF3B1, restrict AAV vector transduction through recognition of incoming AAV capsids. Of particular relevance to gene therapy applications, genetic and pharmacological inhibition of PHF5A or U2 snRNP-associated proteins strongly increased the transduction efficiency of AAV vectors. Thus, transient suppression of U2 snRNP or designing AAV vectors to avoid this restriction can provide a novel strategy to achieve efficient AAV vector transduction with reduced vector doses, which in turn could lead to improved safety profiles for AAV-mediated gene therapy applications.
Several strategies have demonstrated the potential to enhance AAV vector transduction efficiency, including treatments with genotoxic agents [43,44,45], adenoviral E1b55k/E4orf6 proteins [14], a specific EGFR protein tyrosine kinase inhibitor (Tyrphostin-23) [46], and proteasome inhibitors [10,11]. A major effect of the genotoxic treatments, such as hydroxyurea and topoisomerase inhibitors, is improved double-strand synthesis of the input vector genome [8,9,47]. In contrast, the adenoviral proteins degrade the cellular Mre11 repair complex (MRN) to promote AAV vector transduction as well as provide crucial helper functions for wild-type AAV replication [14], although a recent study suggests a role for MRN in gene expression [15]. Tryphostin-23 dephosphorylates FKBP52, a protein binding to the viral single-strand DNA, and improves viral second-strand DNA synthesis [46] and intracellular trafficking of AAV vectors [48]. Importantly, Tryphostin-23 does not show a synergistic effect with the proteasome inhibitor MG132, suggesting that both drugs target a common step of AAV vector transduction [12,48]. We found U2 snRNP inhibition had no notable effect on AAV vector second strand synthesis. Although we initiated the screening of the study using a commercially available siRNA library targeting known and putative proteasomal pathway proteins, MG132 treatment showed an additive effect with SF3B1 inhibition, suggesting U2 snRNP inhibition and MG132 work on distinct pathways. In contrast, dual treatments with Ad5 co-infection and SF3B1 inhibition showed no additive effect, suggesting a common target shared by Ad5 and SF3B1 inhibition. Since Ad5 infection induced PHF5A and SF3B1 displacement, it is plausible that Ad5 co-infection increases expression from AAV vectors, at least in part, through U2 snRNP inhibition.
Dissection of the mechanism by which PHF5A and U2 snRNP components block AAV vector transduction might allow rational design of next generation AAV vectors that can potentially circumvent this host restriction machinery. Based on the following observations, we conclude that U2 snRNP restricts AAVs at an early stage of infection. First, optimal enhancement of AAV vector transduction required U2 snRNP inhibition at an early time point post AAV vector infection (3–24 hours p.i.), while washout of a U2 snRNP inhibitor for the following 2 days did not impair the effects. Thus, short-term U2 snRNP suppression appears to change the fate of AAV vectors up to 2 days post-incubation. Secondly, U2 snRNP inhibition showed no enhancing effect when purified AAV vector genomic DNA or AAV vector genome plasmids were introduced by transfection. Third, although PHF5A was predominantly found in the nuclei in uninfected HeLa cells, we found frequent co-localization of AAV vector particles and PHF5A signals, both in the nucleus and the cytoplasm of cells exposed to AAV vectors. We speculate that newly synthesized PHF5A likely interacts with incoming AAV capsids in the cytoplasm. Finally, pull-down and capsid co-precipitation assays using AAV vectors and empty AAV particles indicate that PHF5A and SF3B1 interact with the capsid, likely mediated by domains within the VP3 region. Of note, the use of heated AAV particles, which leads to exposure of the hidden VP1 N-terminal and viral genome release [39,40,49], increased co-precipitation of capsid-associated AAV genome by PHF5A pull-down, or SF3B1 co-precipitation by the capsid. Consistent with these observations, previous studies have demonstrated the majority of viral DNA can remain associated with the capsid upon thermally induced DNA release [40]. A recent study has also implicated AAV capsid proteins in playing a role in second strand synthesis as well as the transcription of vector genomes [50], supporting prolonged association of AAV capsid proteins with vector genomes at the time of transcription in the nucleus. Taken together, our results strongly support the notion that direct interaction of PHF5A and U2 snRNP components in a cooperative fashion with conformationally altered AAV capsids and the exposed vector genome blocks subsequent transcription.
Although the exact mechanism is currently under investigation, one potential path being explored hinges on the involvement of U2 snRNP proteins in chromatin regulation. For instance, Isono et al. [51] have reported the essential role of SF3B1 (and likely other U2 snRNP proteins) in mammalian polycomb-mediated epigenetic silencing of homeotic genes. Sudemycin E, a U2 snRNP inhibitor, has also been shown to cause changes in histone modifications [52]. SF3B1 and SF3B2 are also found to associate with the histone H3 tails [53]. The ability of PHF5A and SF3B proteins to recruit additional factors to the AAV capsid and its associated vector genome is currently unknown, but if true these findings would provide further insight into the mechanism of host restriction. Further insight into the latter mechanism can potentially be derived from earlier reports that suggest that the splicing machinery is significantly remodeled during host cell infection by helper viruses such as adenoviruses [54]. As outlined earlier, the spatial organization of host splicing factors into distinct clusters within the nucleus appears to be regulated during adenoviral infection [1,55]. Similarly, herpes simplex virus infection induces snRNP-containing bodies [56] through interaction between IE63 protein (ICP27) with SF3B2 (SAP145) [57]. Thus, it is tempting to speculate that wild type AAV might have evolved to exploit the mislocalization/sequestration of splicing factors during helper virus co-infection, while recombinant AAV are unable to evade such host restriction factors in the absence of helper viruses. It is also possible that U2 snRNP plays a role as a broad spectrum antiviral factor, while helper viruses have evolved to counteract this restriction through sequestration of snRNP proteins.
Based on the aforementioned reasons, our current working model is the U2 snRNP recognition of incoming AAV capsid, leading to subsequent block of AAV transcription. However, some observations suggest potential U2 snRNP-mediated AAV restriction at a late stage of transduction. For instance, at very late time points in infection in cell cultures, there was still a substantial enhancement in AAV vector (100-fold at 24 hr and 20-fold at 33 hr p.i.). At this late time point, most of the genomes are considered to be in the nucleus, and it is less likely that U2 snRNP can target incoming AAV capsid. One plausible explanation is that U2 snRNP can also target AAV genome-associated capsid in the nucleus for blocking AAV vector expression. Another point is on our Northern blot analysis of vector transcripts upon PHF5A knockdown. We found a notable increase in cytoplasmic AAV transcripts, but lesser degree in nuclear transcripts. Thus it remains possible that U2 snRNP can also target the nuclear export/cytoplasmic accumulation of AAV transcripts. Another caveat of our experimental system was the use of rapidly dividing cells, where some, or even the majority, of vector genomes can be lost at later time points.
In addition to mechanistic analysis, we have compared FR901464 analogs and herboxidiene, and have identified that meayamycin B is the most potent SF3B1 inhibitor [58]. Importantly, treatment with meayamycin B substantially enhanced AAV vector transduction in various clinically relevant cell types, including primary cardiomyocytes, pancreatic islets and hepatocytes. Thus, pharmacological inhibition of U2 snRNP components may provide a novel strategy to improve AAV vector transduction. Unfortunately, however, intravenous administration of meayamycin B leads to rapid clearance, likely due to absorption, distribution, metabolism and/or excretion [42]. Additionally Meayamycin B also has a potent anti-proliferative effect at higher doses [37,58]. These features present a barrier to immediate in vivo applications of meayamycin B for improved AAV gene delivery. Nevertheless, since a low dose meayamycin B substantially increased AAV vector transduction without strongly affecting host RNA splicing in vitro, designing a novel U2 snRNP inhibitor with reduced cytostatic effects and in vivo stability may allow co-administration of the inhibitor with AAV vectors for improved AAV gene therapy with reduced vector doses.
In conclusion, we demonstrate that the U2 snRNP spliceosome inhibits AAV vector transduction and genetic/chemical modulation of this machinery improves transduction efficiency. This finding may lead to approaches that might help reduce AAV vector doses in clinical applications. Further understanding the underlying mechanism would provide novel insights into host-virus interactions and could inform the rational or combinatorial design of next generation AAV vectors with improved transduction efficiency and safety profile.
Primary human islets were obtained through the Integrated Islet Distribution Program (IIDP) and the use of the cells was approved by the Mayo Institutional Review Board (IRB10429). All animal experiments were conducted according to the National Institute of Health guidelines and approved by the Institutional Animal Care and Use Committee (IACUC A33214 and IACUC A9014).
pAAV-CMV-Luc vector genome construct, which drives firefly luciferase expression by a CMV internal promoter followed by the human beta globin intron, was described previously [24]. pAAV-SFFV-Luc was generated by replacing the CMV promoter and the beta globin intron region by Mlu1-BamHI with the intron-less SFFV retroviral promoter from a lentiviral vector plasmid, pSIN-Luc. The HA-tagged wildtype PHF5A-expressing lentiviral vector, pSIN-PHF5A-HA, was constructed by amplifying the human PHF5A ORF (GenBank accession number BC075808) in pCMV-SPORT6 (OpenBiosystems, MHS1010-97228317) by primers with a 3’ hemagglutinin (HA) tag, followed by cloning into the BamHI and NotI sites of a pHR-SIN CSGW PGK Puro (gift from Prof. Paul J. Lehner). Site-directed mutagenesis was performed to generate pSIN-PHF5A-Escape, with three point mutations in the PHF5A siRNA #1 targeting site. Further site-directed mutagenesis was performed to generate zinc finger mutant vectors, pSIN-PHF5A-HA-Esc-C46A/C49A, C58A/C61A and C72A/C75A (zinc fingers 3, 1 and 2 mutants, respectively). Primers used in the cloning are 5’-BamHI-F, 5'-GTCGGATCCGCCACCATGGCTAAACATCATCCTGA; 3’-NotI-HA-R, 5'-GGAGCGGCCGCTCAGGCGTAGTCAGGCACGTCGTAAGGATACCTCTTCTTGAAGCCGTATT; PHF5A-Escape-F, 5'-GCCGCAAGCAGGCAGGGGTGGCCATCGGAAG; PHF5A-Escape-R, 5'-CTTCCGATGGCCACCCCTGCCTGCTTGCGGC; PHF5A-ZF1m-F, 5'-ACCAGGGGCGCGCTGTGATCGCTGGAGGACCTGGGG; PHF5A-ZF1m-R, 5'-CCCCAGGTCCTCCAGCGATCACAGCGCGCCCCTGGT; PHF5A-ZF2m-F, 5'-GGTCTCTGATGCCTATTATGCTAAGGAGGCCACCATCCAGG; PHF5A-ZF2m-R, 5'-CCTGGATGGTGGCCTCCTTAGCATAATAGGCATCAGAGACC; PHF5A-ZF3m-F, 5'-GGTGCGCATAGCTGATGAGGCTAACTATGGATCTTACCAG; PHF5A-ZF3m-R, 5'-CTGGTAAGATCCATAGTTAGCCTCATCAGCTATGCGCACC.
HeLa (ATCC), 293T (ATCC), A375 (ATCC) and primary human cardiac fibroblast cells (ScienCell Research Laboratories) were cultured in Dulbecco’s modified Eagle’s medium containing 10% fetal bovine serum (FBS) (GIBCO) and antibiotics (penicillin 100 U/mL and streptomycin 100 μg/mL) (Corning Cellgro). HeLa cells stably expressing a series of PHF5A mutants were generated by transduction of corresponding lentiviral vector, followed by puromycin selection. Human islets were obtained through Integrated Islet Distribution Program and cultured in RPMI1640 medium supplemented with 10% FBS and antibiotics. Murine islets were harvested through intraductal collagenase perfusion and enzymatic digestion of the pancreas as previously described [59], and maintained in RPMI1640 medium supplemented with 10% FBS and antibiotics. Porcine hepatocytes were isolated from 15–20 kg pigs by a 2-step collagenase perfusion technique as previously described [60], and cultured in DMEM medium supplemented with 10% FBS, 10mM HEPES and antibiotics. Primary cardiomyocytes were isolated from newborn Dahl salt-sensitive rats using the Neonatal Cardiomyocytes Isolation System (Worthington, Lakewood, NJ) according to the manufacturer’s instruction. Beating cardiomyocytes were plated in gelatin/fibronectin-coated plates in DMEM medium supplemented with 10% FBS.
Helper-free AAV vectors were produced by transfection of three plasmids as described previously [61]. Briefly, 293T cells were transfected with three plasmids, including pHelper (Stratagene), one of the RepCap-expression plasmids (pRep2Cap2, pRep2Cap6, pRep2Cap9, or pRep2Cap8, kindly provided by Dr. James Wilson) and a transfer vector plasmid (pAAV-CMV-Luc, pAAV-SFFV-Luc, pAAV-CMV-Emerald GFP, or pScAAV-CMV-GFP [62]. pScAAV-CMV-GFP plasmid was kindly provided by Dr. R Jude Samulski through the National Gene Vector Biorepository. The resulting vectors were gradient purified using iodixanol (Optiprep Density Gradient Medium, SigmaAldrich), desalted and concentrated using Amicon Ultra-15 100k filtration (Amicon, Billerica, MA, USA) and resuspended in PBS. The genome copies (gc) of concentrated AAV vector stocks were determined by quantitative PCR as described previously [24]. Luciferase- or shRNA-carrying lentiviral vectors were produced as described previously [63]. Human adenovirus 5 (ATCC VR1516) was purchased from ATCC. Unless otherwise stated, no helper virus co-infection was used during AAV vector transduction.
Human siGENOME Ubiquitin Conjugation Subsets #1 (89 genes), #2 (115 genes) and #3 (396 genes), a SMARTpool siRNA Library in Reverse Transfection Format (RTF) covering 600 gene targets, were purchased from Thermo Fisher Scientific. According to the provided RTF protocol, 5,000 cells/well HeLa cells were seeded, followed by AAV9 CMV-Luc infection at a multiplicity of infection of 100 (gc/cell). 48 hours after infection, luciferase assay was performed using the ONE-Glo Luciferase Assay System (Promega).
HeLa cells were seeded in a 96-well plate at 5,000 cells/well for one day. Cells were then transfected with 0.5 μL of 10 μM siRNA using DharmaFECT Transfection Reagents (Thermo Fisher Scientific) according to the manufacturer’s instruction. Following siRNAs were used; control siRNA (siKrt1 5 SI02636732 from Qiagen), siPHF5A#1 and #2 (PHF5A 6 SI04210892 and 7 SI04310621) from Qiagen, siGenome Smart Pool siRNAs for PHF5A-interacting proteins;—siHIST1H4B (NM_003544, cat# M-011463-00), siU2AF1 (NM_001025203, cat# M-012325-01), siSF3B1 (NM_001005526, cat# M-020061-02), siSF3B2 (NM_006842, cat# M-026599-03), siSF3B3 (NM_012426, cat# M-020085-01). Twenty four hours post transfection, cells were infected with luciferase- or GFP-expressing vectors for 2 days.
Meayamycin B was described previously [58]. Isoginkgetin was purchased from Millipore and resuspended in DMSO. 3-Aminophenylboronic acid was purchased from Sigma and resuspended in DMSO.
cDNA synthesis was performed with one μg RNA using RNA to cDNA EcoDry Premix (Clontech). Primers used were as follows: Fig 1B PHF5A (cat# Hs00754435_s1, Invitrogen); Fig 2C luciferase (cat# Mr03987587_mr, Invitrogen); Fig 4B and 4C AAV polyA (Forward 5’-CCTGGGTTCAAGCGATTCTC-3’, Reverse 5’-AGCTGAGCCTGGTCATGCAT-3’, Probe 5’-/FAM/TGCCTCAGCCTCCCGAGTTGT, IDT).
Western blotting was performed as described previously [64]. Following primary antibodies were used: rabbit anti-PHF5A (Sigma HPA028885-100UL) 1:50, rat anti-HA clone 3F10 (Roche 11867423001) 1:250, rabbit anti-VP1, 2, 3 (American Research Products, Inc. 03–61084) 1:250, mouse anti-SAP155 (SF3B1) (MBL International D221-3) 1:250. ImageJ software was used to quantify Western blots from immunoprecipitations.
DIG High Prime DNA Labeling and Detection Starter Kit II (Roche) was used for Southern blotting to detect the luciferase DNA in the AAV vector genome. A luciferase DNA fragment from pSIN-Luc was labeled according to the manufacturer’s instruction. HeLa cells were seeded in a 6-well plate at 200,000 cells per well, followed by transfection with control or PHF5A siRNAs. 24 hours post transfection, AAV9 CMV-Luc (MOI 8 x 104) was added for 1, 3 or 6 hours. Cells were harvested in lysis buffer for nuclear fractionation. Nuclear lysates were purified as in “Cell fractionation and analysis of nuclear rAAV genomes” section. 2 μg of DNA sample was run on a 2% agarose gel without ethidium bromide at 50V for 1.5 hours. The gel was prepared for transfer in the following washes with rocking: 0.25N HCl 10 min, rinse ddH20, denaturation buffer (0.5N NaOH, 1.5M NaCl) 15 min, denaturation buffer 30 min, neutralization buffer (0.5M Tris, 1.5M NaCl) 15 min, and neutralization buffer 30 min. The gel was then blotted overnight by capillary transfer with 10x SSC on a positively charge nylon membrane (Roche). DNA was fixed to the membrane by UV-crosslinking and the luciferase probe was hybridized to the DNA overnight at 43.5°C. The membrane was washed and developed according to the protocol (Roche).
Cells were prepared for Northern blot by seeding at 200,000 cells per well in a 6-well plate, transfected with control and PHF5A siRNAs, followed by transduction with AAV9 CMV-Luc vector (MOI 4 x 105). 36 hours post transduction cells were harvested and nuclear and cytoplasmic RNA was isolated using the PARIS kit (Ambion). 1 μg RNA was run on a formaldehyde gel, washed, and blotted by capillary transfer overnight according to the DIG Northern Starter Kit (Roche). The RNA was fixed to the membrane by UV-crosslinking and the DIG-labeled luciferase DNA probe from Southern blotting was incubated with the pretreated membrane overnight at 50°C. The membrane was washed and developed according to the manufacturer’s instructions.
Cells were resuspended in cytoplasmic lysis buffer (1.3M sucrose, 20mM MgCl2, 4mM Tris, 4.2% Triton X-100) and incubated on ice for 10 min. The lysates were homogenized using a 21-guage needle and syringe, spun at 14,000 rpm for 15 min at 4°C and the supernatant (cytoplasmic fraction) was collected. The pellet (nuclear fraction) was washed with PBS. RNA was eliminated by RNaseA (200U/mL) treatment. A Qiagen QIAamp DNA Mini kit was used to further purify the DNA. Quantitative real-time PCR was performed to determine AAV luciferase genomic copy numbers. To assess encapsidated AAV genomes in the nucleus, the above procedure was followed with the addition of DNase (Invitrogen) treatment at 37°C for 30 min (both control and DNase treated samples) at the beginning of DNA purification and quantitative real-time PCR detection.
AAV genomic DNA was isolated using the QIAamp DNA mini kit following the provided Protocols for Viral DNA (Qiagen). The genomic DNA was isolated from 4 x 1011 vector genomes of purified AAV9 CMV-Luc vector. HeLa cells transfected with control or PHF5A siRNAs were transfected with the purified AAV genomic DNA (0.1 μg/well) by FuGENE6 (Promega), and luciferase expression was analyzed 48 hours after viral DNA transfection.
Semi-confluent HeLa cells with or without stable overexpression of the HA-tagged PHF5A were infected with AAV2 or AAV9 CMV-Luc (MOI 4 x 105) in a 6-well plate for 6 hours at 37°C. Cells were then harvested on ice in RIPA buffer containing protease inhibitor, followed by pull-down with 20 μL anti-HA agarose beads (Pierce). After 15 cycles of washing, pellets were resuspended in 0.5 mL PBS and split into 2 aliquots. One aliquot was used for Western blotting of AAV capsid proteins, and the other was used for the isolation of total DNA by QIAamp DNA Mini kit for RT-qPCR detection of AAV genomic DNA. For pull-down assay using heated AAV particles, 100 μL cell lysate was combined with 3 x 1010 gc of AAV2 CMV-Luc vector particles that were unheated or pre-heated for 30 min at 65°C, followed by precipitation by anti-HA agarose as above.
The Lab-TekII 8-well chamber slides (Thermo Fisher Scientific) were pretreated for 5 min with poly-d-lysine (Sigma, 0.1 mg/mL). HeLa cells were plated at 1 x 104 cells/well and were infected with AAV2 CMV-Luc (4 x 1010 gc/well) or empty AAV2 (25 μl/well) at 37°C for 5 min, 4 hours, or 12 hours. Cells were fixed in 4% paraformaldehyde for 20 min at room temperature, permeabilized with 0.3% Triton X-100 for 15 min, and blocked with 5% FBS/PBS for 30 min. Primary antibody was added, and cells were incubated for 1.5 hours at room temperature in a humidified chamber. Secondary antibody followed according to the same procedure. Then cells were washed three times with PBS, treated with DAPI (Sigma, 1:2000) for 1 min, washed three times with PBS, and mounted with Dako fluorescent mounting media. Confocal microscopy was performed on an LSM 780 confocal microscope (Zeiss). The following primary and secondary antibodies were used for immunocytochemistry of uninfected and AAV infected HeLa cells: anti-AAV particles (A20) mouse monoclonal antibody (American Research Products) at 1:100 followed by FITC-conjugated donkey anti mouse IgG (H+L) (Jackson 715-095-151) 1:500; rabbit anti-PHF5A (Sigma HPA028885-100UL) 1:250 and Alexa Fluor 594-conjugated donkey anti-rabbit IgG (H+L) (Invitrogen A-21207) 1:2000; rabbit anti-phospho-SF3B1 (MBL International PD043) 1:500 and Alexa Fluor 594-conjugated donkey anti-rabbit IgG (H+L) 1:500.
AAV2 CMV-Luc vectors (5 x 1010 gc/tube) and purified AAV2 empty, VP3 only particles were left unheated or pre-heated at 65°C for 30 min and placed on ice. HeLa and HeLa-PHF5A-HA cells were harvested by incubating 1 well of a 6-well plate with 500 μL RIPA buffer supplemented with protease inhibitors on ice for 10 min. Cells were harvested by scraping, homogenized using a 21-guage needle and syringe, and spun for 5 min at 13,200 rpm. 400 μL of the cell lysate was added to the virus and samples were rotated for 1 hour at 4°C. The 25% iodixanol solution was prepared using 3.2 mL 1x PBS, 2.8 mL 9:1 Optiprep to 10x PBS, and 0.15% phenol red. 30 μL of the lysate virus mixture was removed and used as an input. The remaining lysate-virus mix was layered on top of 0.5 mL 25% iodixanol. These samples were spun at 4°C for 1 hour at 14,000 rpm. After spinning the clear upper layer, red lower layer, and pellet were harvested for Western blot. The following antibodies were used in this experiment: rabbit anti-PHF5A (Sigma HPA028885) 1:100, rat anti-HA clone 3F10 (Roche 11867423001) 1:500, rabbit anti-VP1, 2, 3 (American Research Products, Inc. 03–61084) 1:250, mouse anti-SF3B1 (MBL International D221-3) 1:500, rabbit anti-phospho-SF3B1 (MBL International PD043) 1:400, rabbit anti-histone H2B (Cell Signaling #8135) 1:1000, and rabbit anti-histone H3 (Cell Signaling #4499) 1:1000 in blocking buffer. To note, the membrane that received phospho-SF3B1 primary antibody was blocked in 5% BSA/PBS + 0.05% Tween-20. Antibodies were diluted in 1x TBS + 0.001% Tween-20.
HeLa cells at 80% confluency were treated with pre-mRNA splicing inhibitors at the following concentrations: 3-Aminophenylboronic acid (5mM and 1mM), Isoginkgetin (25uM and 12.5uM), and meayamycin B (10nM and 5nM). Eight hours post drug RNA was isolated using TRIzol (Invitrogen), and cDNA synthesis was performed with one μg RNA using RNA to cDNA EcoDry Premix (Clontech). cDNA was amplified by KOD Hot Start DNA Polymerase (EMD Millipore) using primers for MAPT exon 10 5’-AAGATCGGCTCCACTGAGAA-3’ and 5’-ATGAGCCACACTTGGAGGTC-3’.
|
10.1371/journal.pbio.1000040 | The Atonal Proneural Transcription Factor Links Differentiation and Tumor Formation in Drosophila | The acquisition of terminal cell fate and onset of differentiation are instructed by cell type–specific master control genes. Loss of differentiation is frequently observed during cancer progression, but the underlying causes and mechanisms remain poorly understood. We tested the hypothesis that master regulators of differentiation may be key regulators of tumor formation. Using loss- and gain-of-function analyses in Drosophila, we describe a critical anti-oncogenic function for the atonal transcription factor in the fly retina, where atonal instructs tissue differentiation. In the tumor context, atonal acts by regulating cell proliferation and death via the JNK stress response pathway. Combined with evidence that atonal's mammalian homolog, ATOH1, is a tumor suppressor gene, our data support a critical, evolutionarily conserved, function for ato in oncogenesis.
| During embryonic development, cells become more and more specialized, and this process is referred to as differentiation. In contrast to normal adult cells, cancer cells—like embryonic cells—display fewer differentiated properties. It has been postulated that the acquisition of terminal differentiation helps inhibit tumor formation; however, no direct evidence for this hypothesis was available. The development of the eye in the fruit fly, Drosophila melanogaster, has long been used as a model for studying genetic factors controlling differentiation. More recently, eye development has also been used to study how tumors can form and progress. In this study, we used this model to show that genes, such as atonal, that instruct the differentiation of specific tissues can act as tumor suppressers and inhibit the formation and progression of tumors in those tissues. Losing such genes can generate tumors, whereas activating them can strongly inhibit these tumors.
| Cell fate commitment in neural and neuroendocrine lineages of the peripheral nervous system (PNS) as well as secretory epithelia is controlled by genes of the basic helix-loop-helix (bHLH) superfamily of transcription factors. One of the most structurally and functionally conserved groups within this family is the Atonal (Ato) group proteins [1,2]. Drosophila ato (CG7508) and mammalian ATOH1 (Ensembl accession number: ENSG00000172238) are required for cell fate specification and the induction of differentiation in the PNS and the secretory lineages in all animal species. In Drosophila, ato is necessary for cell fate specification and differentiation of mechano- and photoreceptors [3–5].
The acquisition of differentiated cell fate endows cells with two key features. First, it allows them to become distinct from one another and, accordingly, functionally specialized. Second, it inhibits further cell division under physiological conditions, thus controlling tissue size. When the regulation of cell division fails, cancer may develop. Cancer, however, is the result of a selective process in which cells accumulate several genetic and epigenetic mutations giving them a growth advantage over surrounding cells by, for example, the inhibition of apoptosis, induction of angiogenesis, and growth factor–independent survival [6]. More than one mutation is needed for cancer to arise, and it is therefore thought that mutations occur in undifferentiated cells that are proliferative. As such, oncogenesis might select for cells that have lost their capacity to induce differentiation. In this context, it has been a long-standing postulate that cancer is a disease of loss of differentiation [7,8]. Work in the seventies and eighties by Harris and colleagues shows that hybrids of malignant and diploid cells only become malignant again after losing chromosomal loci required for differentiation (e.g., [9]). More recently, the interplay between differentiation and cancer has gained renewed attention through the study of a pool of undifferentiated cells in tumors, the so-called cancer stem or tumor-initiating cells [10]. A major theme emerging from these studies is the importance of the maintenance of an undifferentiated state in this niche for tumor growth to occur. Furthermore, the fact that signals implicated in regulating differentiation across various lineages, such as the WNT and Notch pathways, also have been implicated both in the promotion and suppression of cancer [11,12] suggests a mechanistic link between the regulation of differentiation and tumor progression. Importantly, however, these pathways are also implicated in stem cell or progenitor cell maintenance and do not act in a lineage-restricted fashion. As such, the definition of their role in tumor progression vis-a-vis differentiation is unclear. For this hypothesis to be correct, at least one key prediction should hold true: master control genes that instruct cell fate commitment in specific lineages should act as brakes on the oncogenic process, either by preventing uncontrolled proliferation or by inducing cell death when a differentiated state can no longer be maintained. Thus, we hypothesize that such master control genes suppress both tumor formation and progression.
To test this prediction in lineages in which ato is the key regulator of cell fate commitment, we asked two experimental questions. First, does ato loss of function contribute to tumor initiation or progression in tissues where ato instructs differentiation, such as the Drosophila retina? Second, can ato gain of function inhibit the formation or progression of these tumors? Finally, we examined the genetic pathway by which ato suppresses tumor formation.
We find that loss of ato strongly enhances the formation and progression of tumors in flies. Conversely, gain of ato function strongly inhibits tumor formation and metastasis. Finally, we describe a highly conserved anti-oncogenic genetic pathway that links ato activity to the stress sensor Jun N-terminal kinase (JNK) pathway. Combined with genetic and molecular evidence from mouse and human cancer models [13], these data support a key role for ato in a very early step of the oncogenic process and suggest that mutations in master control genes of cell fate commitment may be pivotal switches during tumorigenesis.
We took advantage of the genetic power of Drosophila melanogaster to investigate whether the gain and loss of function of Drosophila ato suppresses and enhances tumor formation, respectively. ato instructs differentiation in the Drosophila eye [4]. We therefore turned to a well-established in vivo eye cancer model, namely “eyeful” flies, which has been used to study the mechanisms of Rb and the PTEN-AKT pathway in cancer [14,15]. The eyeful flies have activated Notch signaling in the developing eye due to overexpression of the Notch ligand Delta (Dl, CG3619), combined with overexpression of lola (CG12052) and psq (CG2368). Flies overexpressing only Dl, leading to an increase in eye size but no tumors, will henceforth be called “sensitized” flies.
To analyze the tumor burden, each eye was scored separately. Eyes were counted as hyperplastic when the eye showed at least one fold. Metastasis can be seen as masses of amorphous red-pigmented cells outside of the eye field and are observed on the head and in the thorax and abdomen (Figure 1J–1L). Consistent with previous data, eyeful flies display excessively enlarged eyes, and eye tumors occur in 57% of the eyes, with 3% of the flies showing macroscopically visible metastases derived from the developing retina (n = 102; Figures 1A, 1I, S1A, and S1B). Overexpression of ato, or its mouse ortholog Atoh1 (Ensembl: ENSMUSG00000073043)—but not a green fluorescent protein (GFP) control transgene—in the eyeful background almost completely suppresses the formation of eye tumors (Figure 1B (ato): 2%, p < 0.0001, n = 118; Figure 1C (Atoh1): 1%, p < 0.0001, n = 98; GFP (unpublished data): p = 0.25; Figure 1I). More importantly, reduction of endogenous ato expression in the eye using an ato RNA interference (RNAi) construct (a kind gift from A. Jarman), which leads to loss of differentiated eye tissue in wild-type flies (Figure S2B), results in a dramatic increase in both tumor incidence (90%, p < 0.0001, n = 165; Figure 1D and 1I) and the number of flies with metastases (17%, p = 0.0003). These effects were not due to the overexpression of a double-stranded RNA, per se, because the expression of an RNAi construct for GFP did not change the tumor burden (unpublished data; tumors: p = 0.59, metastasis: p = 0.21).
Since Ato is known as a transcriptional activator, we asked whether its role in eye tumors is mediated by its ability to activate gene expression. We constructed a repressor form of Ato by fusing it to an Engrailed repressor domain (AtoERD) [16]. Expression of AtoERD in the developing eye leads to loss of differentiated eye tissue mimicking ato loss-of-function mutations and the atoRNAi construct (Figure S2A–S2C). Expression of AtoERD in eyeful flies results in both the loss of differentiated eye tissue in 30% of the eye fields (Figure 1J, open arrow), as well as in 100% tumors in the remaining eyes (p < 0.0001, n = 48; Figure 1E and 1I). Importantly, these tumors include large patches of undifferentiated tissue, showing that loss of ato's differentiation function is linked to its anti-oncogenic function (Figure 1E and 1M, dotted line). Expression of AtoERD in the eyeful flies also results in 75% of the flies showing metastasis (Figure 1J, black arrow, and Figure 1I; p < 0.0001). These data suggest that ato is a key regulator of tumor progression in Drosophila and that it may perform this function by regulating the differentiation status of the transformed tissue.
Loss of ato in a wild-type background abrogates retinal differentiation and causes subsequent loss of the entire tissue [4]. If loss of differentiation is an early causal event in cancer, a key anti-oncogenic role for ato requires that its loss act as a switch for tumor initiation in a pre-oncogenic background. To this end, we used the sensitized genetic background that was used to generate the eyeful model, namely eye-specific Dl overexpression [14]. This genotype results in an increase in proliferation, leading in turn to a slight overgrowth of the eye, but no tumors are observed (n = 478; Figure 1F) [17]. Inhibition of ato function by AtoERD leads, as it does in wild-type flies, to loss of retinal differentiation (36% of the eye fields, empty arrow, Figure 1L), and to a 9% de novo tumor incidence in the remaining eye fields (p < 0.0001, n = 76; Figure 1H, 1L, and 1L′), with 6% of the flies showing metastasis (p = 0.0003). Similarly, ato knockdown using atoRNAi in this sensitized background leads to eye tumors in 0.5% of the eyes (p = 0.0333, n = 379; Figure 1G and 1K), and 0.3% of the flies have metastasis (p = 0.1953). The metastases in the sensitized flies upon loss of ato function are mostly present in the thorax and on the head. The metastases on the head (atoRNAi = 75%; atoERD = 43%) show a high resemblance to ectopically induced eyes as seen by Kurata and colleagues upon overactivation of Notch signaling [18]. The loss of ato might increase Notch signaling and as such interfere with patterning and determination. We note, however, that we never see any head metastasis when overexpressing Dl, indicating that loss of ato creates new phenotypes including the metastasis in the thorax, which cannot be explained by an increase in Notch activity alone. In either case, these data support the hypothesis that loss of ato function is sufficient to transform a sensitized lesion into a metastatic tumor, possibly by interfering with patterning and determination.
ato loss- and gain-of-function analyses suggest a decisive role in tumor formation in the fly retina. The growth of tumors is a balance between cell proliferation and cell death. We asked whether ato regulates either or both of these processes in the context of eyeful tumors, during the development of these tumors. To this end, we examined third instar larval eye discs, the Drosophila eye anlage, for markers of apoptosis and proliferation. Overexpression of Ato in the eyeful fly results in dramatically increased levels of the apoptotic regulator caspase-3 (FlyBase ID: FBgn0028381) in eyeful eye discs (3-fold, p = 0.011; Figure 2A–2D). This explains, at least in part, the suppression of the eyeful tumors in the adult flies.
Next, we examined proliferation in eyeful eye discs under gain and loss of ato function conditions, using phospho-HistoneH3 (FlyBase ID:FBtr0071345) as a marker. Proliferation in the third instar eye disc normally occurs anterior to the morphogenetic furrow, where all the cells are still undifferentiated. Additionally, approximately two rows of undifferentiated cells posterior to the furrow, called the second mitotic wave (SMW), also proliferate. In the eyeful discs, proliferating cells are not restricted to these two domains but are also present posterior to the SMW (Figure 2E). Expression of ato in the eyeful disc reduces this ectopic proliferation (Figure 2F), whereas inhibition of ato activity increases ectopic proliferation (Figure 2G). As suppression of Ato activity can initiate de novo tumor formation in a sensitized background, we examined proliferation upon expression of AtoERD in the Dl-sensitized background. In the sensitized eye discs, proliferating cells are mostly restricted anterior to the furrow and the SMW (Figure 2H), whereas loss of ato function leads to the appearance of proliferating cells in the posterior region of the disc (Figure 2I).
The total number of cell divisions in a tissue determines the overall size of that tissue. We therefore quantified the total number of phospho-HistoneH3–positive cells per disc. Overexpression of Ato in eyeful eye discs results in a significant decrease in number of cells expressing the mitotic marker phosphorylated HistoneH3 (p = 0.00004; Figure 2J). Conversely, expression of the dominant-negative AtoERD leads to a significant up-regulation of proliferation in the eyeful eye discs (p = 0.004; Figure 2J). Thus, Ato limits number of cell divisions in the eyeful tumors. Expression of AtoERD in the Dl-sensitized eye discs results in a significant increase in phosphorylated HistoneH3 expression in the developing eye discs (p = 0.002; Figure 2J), explaining the induction of tumors by loss of ato.
Our analysis suggests that ato regulates both proliferation and death of retinal precursors during tumor formation in the Drosophila eye.
During normal development, Ato is required for the correct differentiation of retinal cells and the proper patterning of the eye disc. If Ato's function in suppressing eye tumors is related to its activity as a differentiation factor, we might expect to observe Ato-dependent alterations in tissue differentiation and organization upon manipulation of Ato activity in a tumor context. To test this prediction, we examined the expression of the early differentiation and R8 marker Senseless (Sens, CG32120), the general retinal photoreceptor marker embryonic lethal, abnormal vision (ELAV, CG4262), and the epithelial marker Armadillo/β-Catenin (Arm, CG11579) following manipulation of Ato activity.
In wild-type and Dl-sensitized eye discs Arm, ELAV, and Sens reveal the regular and stereotypical differentiation and epithelial organization of the developing retina, although the Dl-sensitized discs are clearly larger (Figure 3A and 3B). Loss of Ato activity in the Dl-sensitized eye discs (Figure 3C) results in the disruption of the regular pattern of Arm expression, suggesting defects in the organization of the retinal epithelium. This is accompanied by severe reduction in Sens and ELAV staining, suggesting lack of differentiated photoreceptors. The proportion of undifferentiated to differentiated cells is increased, indicating that the initial steps of retinal differentiation, namely the specification of the Ato-dependent R8 cell, are compromised (Figure 3C, white arrows). In some instances, lobes of proliferative and undifferentiated tissue are observed in these eye discs (Figure 3C, open arrow), correlating with the appearance of tumors in the adult flies. In the eyeful eye discs, disorganization of the epithelium as well as defects in the pattern of differentiated cells are apparent (Figure 3D). Overexpression of Ato in the eyeful eye discs restores both the size and all three markers to essentially wild-type patterns of expression (Figure 3E), explaining the appearance of normal adult eyes in this background. Conversely, expression of AtoERD severely disrupts retinal patterning and the expression pattern of all three markers (Figure 3F). Differentiation markers are not only reduced, but also appear in a highly disruptive pattern to the extent that the morphogenetic furrow is difficult to discriminate (Figure 3F).
In summary, loss- and gain-of-function analyses in Drosophila support a critical and early role for the loss of ato in tumor initiation and progression. This effect is likely mediated by alteration in the expression of downstream genes required for retinal differentiation, as such perturbing proliferation, apoptosis, and tissue organization.
To better understand the role of that Ato plays in tumor formation, we sought to determine the genetic mechanism by which it acts to suppress the formation and progression of tumors. Gain- and loss-of-function analysis indicated an ato-dependent regulation of proliferation in the Drosophila eye. Recently, the Drosophila ortholog of the gene encoding the cell cycle inhibitor p21waf1, dacapo (dap, CG1772), was reported to be a target gene of ato in the eye [19]. Consistent with this, overexpression of wild-type ato in the eye disc leads to significant up-regulation of Dap mRNA (∼80%, p = 0.018), whereas expression of AtoERD leads to significant down-regulation of Dap mRNA (28%, p = 0.014; Figure 4A). Ato expression also results in earlier onset and elevated Dap levels in the eyeful and wild-type eye discs, in agreement with the reduction in pH3 levels observed in the same discs (Figures 4B–4D and S3).
We have shown that Ato regulates apoptosis and that it restores proper differentiation in the eyeful eye discs. We reasoned that tumorous eyeful cells may interpret the Ato differentiation signal as cellular stress and, as a result, commit suicide. A major regulator of cell death in response to stress is the JNK pathway. We therefore examined the expression of phosphorylated (i.e., activated) form of the Drosophila JNK (pJNK), Basket (Bsk, CG5680). Eyeful discs show reduced pJNK levels. Overexpression of Ato in this background, as well as wild-type eye discs, results in dramatic up-regulation of pJNK levels (Figures 4B–4D and S3).
These data suggest that Ato regulates the expression and activity of major regulators of cell proliferation and death. We therefore tested whether these genes also play a role in the eyeful tumors. Dap overexpression leads to a significant inhibition of tumor occurrence (22%, p < 0.0001, n = 107; Figure 5C and 5I), but only a partial reduction in metastasis (1%, p = 0.1145). Thus, whereas Dap regulation appears to contribute to tumor suppression by Ato, it is unlikely to explain the full effect of ato expression.
To analyze whether the elevated activity of JNK signaling upon ato expression is functionally relevant, we inhibited JNK signaling using a dominant-negative form of Bsk (BskDN). This partially mimics down-regulation of ato in the eyeful model and results in tumors in 61% of the eyes (p = 0.574; Figure 5D and 5I) and an approximately7-fold increase in metastasis (p = 0.0003). Furthermore, expression of BskDN in the Dl-sensitized background leads to the induction of tumors (2%, p = 0.0113, n = 57; Figure 5H and 5I) and metastasis (3.5%, p = 0.0112). Conversely, overexpression of Djun (Jra, CG2275), the transcriptional effector of the JNK pathway, leads to reduction of the tumor burden (38% tumors, p = 0.0036, n = 40; 0% metastasis, p = 0.559; Figure 5F and 5I), partially mimicking overexpression of Ato expression.
Next, we tested genetic epistasis between ato and JNK by overexpressing Ato while simultaneously inhibiting JNK signaling. This leads to a suppression of the inhibitory effects of ato on the eyeful flies and restores tumor formation (48%, p < 0.001), as well as enhances the metastatic phenotype (15%, p = 0.001, n = 27; Figure 5E and 5I). This indicates that JNK signaling is downstream of ato and that ato requires active JNK signaling to inhibit cancer formation.
Our data support a function for ato in oncogenesis. Loss of ato promotes tumor formation and progression and might, as such, be selected for during oncogenesis. This indicates that tumor formation and progression might not only require maintenance of self-renewal capacity, but also loss of the capacity to induce cell fate commitment and differentiation. Therefore, genes that act precisely at the junction of the transition from a proliferating progenitor to a committed cell ought to show anti-oncogenic behavior. Losing ato in the absence of any other compounding factor is neutral towards tumor formation. However, loss of ato in a sensitized background is sufficient to initiate and enhance tumor formation. In our experiments, we used activation of the Notch signaling pathway as a sensitizing factor, but other pathways also lead to the formation of tumors when ato is lost [13]. Therefore, loss of differentiation factors might “tip the balance” towards malignancy, regardless of what the additional oncogenic event may be. It will be interesting to investigate what the different pathways are that interact with loss of ato to enhance cancer formation and how they switch an ato mutation from neutral to tumor progression to oncogenic.
The induction of cellular differentiation acts on two levels: first, the cell cycle is inhibited by the expression of cell cycle inhibitors; and second, gene expression is modulated to instruct a specific fate and function. Several lines of evidence suggest that both levels of ato activity are important in its anti-oncogenic function. First, ato regulates the expression of dap—itself a direct target gene of ato during normal differentiation—during eye tumor formation. Second, loss of ato leads to more proliferation in the sensitized and cancerous tissue in a Drosophila model. Third, loss of ato leads to the disruption of retinal differentiation and patterning, correlating with the formation of tumors that include overgrowth of undifferentiated tissue in the fly eye. Together, these data support the idea that Ato exerts its anti-oncogenic function by activation of its developmental target genes and pathways. Finally, earlier reports suggest that, under certain conditions, proliferation can be uncoupled from the induction of differentiation as double-mutant cells for retinoblastoma and dacapo in the developing Drosophila eye keep proliferating although they start to differentiate [20]. Our data suggest that the inhibition of proliferation is not the only mechanism by which differentiation factors might suppress tumor formation, as ato is also able to induce apoptosis in an eyeful eye disc.
The function of JNK in the Drosophila eye has been described as both tumor promoting and anti-oncogenic. Igaki and colleagues describe a role for JNK in invasion upon loss of cell polarity [21], and Uhlirova et al. describe how JNK cooperates with Ras to induce tumor overgrowth in the eye [22]. In the overgrowth-sensitized setting of scribble mutant cells, however, JNK is necessary to remove these cells by apoptosis [22]. This shows that the molecular environment in which JNK acts decides the outcome. We propose that the status of differentiation might be an important factor in the decision of the outcome of JNK activity. ato function might divert JNK from an oncogenic function to a tumor suppressor function in which JNK will reduce the size of the overgrowth and, as such, reduce the number of metastases. Our data indicate that although JNK is necessary for the anti-oncogenic function of Ato, it is not sufficient, because inhibition of JNK signaling does not completely mimic the loss of ato function in the eyeful flies. This suggests JNK as a permissive, rather than instructive, factor for ato's function and indicates that ato might also modulate tumor formation by JNK-independent mechanisms.
In summary, we present the first evidence that a master regulator of tissue-specific differentiation is a key regulator of tumor initiation and progression. The evidence that the human ortholog of Ato is a tumor suppressor gene in colorectal cancer, the largest cause of cancer deaths world-wide [13], as well as the absolute functional conservation between fly and mouse Ato [23] underscore the importance of understanding the fundamental molecular and genetic mechanisms of the function of this group of key developmental regulators.
Fly strains used were ey-GAL4, GS88A8, UAS-Dl/Cyo (called eyeful flies in the text) and ey-GAL4, UAS-Dl/Cyo flies (a gift from M. Domiguez), UAS-atoRNAi3B and UAS-atoRNAi 4E (gift from A. P. Jarman), UAS-ato, UAS-Atoh1, w1118 P{UAS-bsk.DN}2, UAS-Djun, UAS-dacapo (a gift from A. Hidalgo), CantonS, and yw. All flies were raised at 25 °C on standard fly food.
Eye discs of wandering third instar larva were dissected and processed as described [24]. ato antibody (kind gift from A. Jarman and P. zur Lage), Dap antibody (Developmental Studies Hybridoma Bank), P-JNK (Cell Signaling Technologies), phospho-HistoneH3 (Upstate Biotechnologies), and cleaved caspase-3 (Cell Signaling Technology).
Uas-atoERD was generated by fusing the full-length Atonal ORF to a fragment encoding a Myc-tagged Engrailed repression domain (amino acids 2–298) [25] using the pUAST vector [26]. Seven uas-atoERD transgenic lines were obtained using standard Drosophila transformation protocols.
Crosses between ato-GAL4 (P{GawB}NP6558 obtained from Drosophila Genetic Resource Center, Kyoto) or Gal4–7 and UAS-ato or UAS-atoERD were performed at 18 °C and shifted to 28 °C at third larval instar stage. Eye-antennal discs were dissected in RNA later (Ambion). RNA extraction was performed with Mini RNA Isolation kit (Zymo Research). act79B, gadph, and Rpl32 were used as control housekeeping genes (ΔCT), and Canton S and UAS-Ato as control RNA (ΔΔCT).
The number of proliferating cells per eye disc was quantified using the “analyse particle” function in ImageJ with the parameters 5 to 60 for size and 0.5 to 1.0 for circularity.
|
10.1371/journal.pcbi.1005893 | Non-linear auto-regressive models for cross-frequency coupling in neural time series | We address the issue of reliably detecting and quantifying cross-frequency coupling (CFC) in neural time series. Based on non-linear auto-regressive models, the proposed method provides a generative and parametric model of the time-varying spectral content of the signals. As this method models the entire spectrum simultaneously, it avoids the pitfalls related to incorrect filtering or the use of the Hilbert transform on wide-band signals. As the model is probabilistic, it also provides a score of the model “goodness of fit” via the likelihood, enabling easy and legitimate model selection and parameter comparison; this data-driven feature is unique to our model-based approach. Using three datasets obtained with invasive neurophysiological recordings in humans and rodents, we demonstrate that these models are able to replicate previous results obtained with other metrics, but also reveal new insights such as the influence of the amplitude of the slow oscillation. Using simulations, we demonstrate that our parametric method can reveal neural couplings with shorter signals than non-parametric methods. We also show how the likelihood can be used to find optimal filtering parameters, suggesting new properties on the spectrum of the driving signal, but also to estimate the optimal delay between the coupled signals, enabling a directionality estimation in the coupling.
| Neural oscillations synchronize information across brain areas at various anatomical and temporal scales. Of particular relevance, slow fluctuations of brain activity have been shown to affect high frequency neural activity, by regulating the excitability level of neural populations. Such cross-frequency-coupling can take several forms. In the most frequently observed type, the power of high frequency activity is time-locked to a specific phase of slow frequency oscillations, yielding phase-amplitude-coupling (PAC). Even when readily observed in neural recordings, such non-linear coupling is particularly challenging to formally characterize. Typically, neuroscientists use band-pass filtering and Hilbert transforms with ad-hoc correlations. Here, we explicitly address current limitations and propose an alternative probabilistic signal modeling approach, for which statistical inference is fast and well-posed. To statistically model PAC, we propose to use non-linear auto-regressive models which estimate the spectral modulation of a signal conditionally to a driving signal. This conditional spectral analysis enables easy model selection and clear hypothesis-testing by using the likelihood of a given model. We demonstrate the advantage of the model-based approach on three datasets acquired in rats and in humans. We further provide novel neuroscientific insights on previously reported PAC phenomena, capturing two mechanisms in PAC: influence of amplitude and directionality estimation.
| The characterization of neural oscillations, which are observed in the mammalian brain at different temporal and spatial scales, have given rise to important mechanistic hypotheses regarding their functional role in neurosciences (e.g. [1, 2]). One working hypothesis suggests that the coupling across neural oscillations may regulate and synchronize multi-scale communication of neural information within and across neural ensembles [2, 3]. The coupling across different oscillatory activity is generically called cross-frequency-coupling (CFC) and has started to receive much attention [4–8]. The most frequent instance of CFC consists in the observation that the power of high frequency activity is modulated by fluctuations of low-frequency oscillations, resulting in phase-amplitude-coupling (PAC). Other instances of CFC include phase-phase coupling [9, 10], amplitude-amplitude coupling [11, 12], and phase-frequency coupling [4, 8]. By far, PAC is the most reported CFC in the literature.
Seminally, PAC was described in local field potential (LFP) of rodents displaying a modulation of gamma band power (40–100 Hz) as a function of the phase of their hippocampal theta band (5–10 Hz) [13, 14]. Parallel recordings in different brain areas in behaving rodents have also highlighted differences in PAC between brain areas (e.g., hippocampus and striatum) at specific moments during a goal-oriented behavior, both in terms of which high-frequency range and how narrow-band the low frequency is [14]. PAC may promote cellular plasticity underlying memory formation [15]. In humans, theta (4–8 Hz)/gamma (80–150 Hz) PAC was described in human auditory cortex during speech perception [6]. In more recent work, theta/gamma PAC was reported during the processing of auditory sequences in both humans and monkeys [16], during working memory maintenance in human hippocampus [17], and during serial memory recall using non-invasive human magnetoencephalography (MEG) [18]. PAC has been proposed to support the maintenance of information and to play an important role in long distance communication between different neural populations, considering that slow oscillations can propagate at larger scales than fast ones [4, 5, 8, 19, 20]. Consistent with this notion, PAC has also been reported across distinct brain regions [21]. In sum, PAC has been proposed as a canonical mechanism for neural syntax [3].
Given the growing interest in CFC, and in PAC more specifically, developing adequate and unbiased tools to quantify the posited signatures of neural computations has motivated a number of contributions [6, 22, 23]. While these works have met some success, they have also pointed out numerous methodological challenges [24, 25] illustrated and discussed in several simulation studies [23, 26]. Some important limiting factors and potential pitfalls of current used techniques are related to the impact of incorrectly choosing the bandwidth of bandpass filters [24, 27], the consequences of applying Hilbert transform on wide-band signals [27, 28], and the potential misidentification of CFC [29, 30]. The direction we advocate to address these issues is to use a signal model that would fit the data in the sense of the mean squared error, and from which a measure of PAC or CFC could be derived. What all existing metrics have in common is that they do not make use of a signal model, and therefore setting the parameters of the method, such as filtering parameters, can only be driven by how much they lead to a strong PAC/CFC metric. As a consequence, even though current metrics give reasonable estimates of PAC, a legitimate and controlled comparison of methods and parameters, and therefore of the results, is impossible. Additionally, while simulations provide better control, they do not fully solve this issue faced by experimentalists since a simulation may approximate at best, or miss at worst, the real structure of neurophysiological signals. Hence, with the present contribution, we hope to propose a major improvement in CFC/PAC estimation by adopting a principled modeling approach.
To initiate this statistical model-based approach, we propose to consider a signal model which is rich enough to capture time-varying phenomena such as PAC. A model-based approach allows computing the likelihood of a recorded neural signal, which can be interpreted as a measure of the goodness of fit of the model. In the present case, the goodness of fit corresponds to the classical measure of explained variance. Such an evaluation metric is a natural criterion to compare models, and a first step towards an automatic model parameters selection on brain data. Additionally, with a model-based approach, it becomes possible to answer a large variety of practical questions: From a statistical standpoint, one can define which parametrization or which hyper-parameter should be selected, while from a neuroscientific standpoint, one could ask, for instance, whether the instantaneous amplitude of slow oscillations contributes to PAC, or whether the amplitude modulation affects the entire neural spectrum identically. Importantly, our statistical signal modeling approach is not a biophysical one and is thus distinct from previous work [8, 31]. Our goal here is to better explain and describe the empirical data themselves in the absence of any assumption regarding the neural mechanisms that have generated them.
To capture PAC in univariate time-series, we propose to use auto-regressive (AR) models, which are stochastic signal models as opposed to deterministic. A canonical deterministic signal model consists in modeling a time-series as a pure sinusoid corrupted by some additive noise. As recently discussed, considering neural oscillations as pure sinusoidal brain responses is an oversimplification of the complex neurophysiological reality [32]. AR models do not make the strong assumption of sinusoidality and are rather based on the ability of the model to forecast the signal values. AR models have been successfully used to address multiple problems in neurophysiology such as spectral estimation [33], temporal whitening [34], and connectivity measures like Granger causality [35–37] or coherence [38].
In the present work, we show how driven auto-regressive (DAR) models [39, 40] can be extended in order to study PAC. Standard AR models are statistically efficient given their low number of parameters, but they are linear, and therefore they cannot directly model non-linear phenomena like PAC. In order to extend AR models to cope with such situations of non-linearity and non-stationarity in signals, various advanced AR models have been proposed in other research fields such as audio signal processing and econometrics [41–43] (see related non-linear AR models section below). Here, we consider the slow oscillation as the exogenous driver so as to allow the coefficients of the AR model to be time varying, thereby making the instantaneous power spectral density (PSD) of the signal be a function of a slow exogenous time series. By doing so, DAR models are statistical signal models (and not biophysical ones) that do not model PAC explicitly but which are able to capture it.
In what follows, we first present how PAC is commonly analyzed in neural time series while pointing out limitations. We then review the current literature on non-linear AR models beyond neuroscience. We then detail the methodological aspects of DAR models first generically and then specifically for brain data. We describe how we can derive from a DAR model a power spectral density parametrized by an external signal and how it allows to compute so-called comodulograms and estimate directionality of coupling; this enables a fine-grained analysis of spectral modulations observed in PAC. Finally, results show how the comparison of goodness of fit between different models allow to conclude on optimal parameters (e.g. filtering parameters), leading to new neuroscientific insights, such as the wideband or the asymetric spectral properties of the driver. We also report that taking into account the instantaneous amplitude of the driver improves the goodness of fit, which suggests that this information is relevant to the coupling phenomenon and should not be discarded as in most PAC metrics. Last but not least, directionality estimation results reveal some delays between high frequency bursts and the driving low frequency oscillation.
We note y the signal containing the high-frequency activity, and x the signal with slow frequency oscillations, also called the exogenous driver. When a signal x results from a band-pass filtering step, we note the central frequency of the filter fx and the bandwidth Δfx. The value of the signal x at time t is denoted x(t).
To estimate PAC, the typical pipeline reported in the literature consists in four main processing steps:
The Modulation Index (MI) described in the pioneering work of [6] is the mean over time of the composite signal z = a y e ϕ x. The stronger the coupling between ϕx and ay, the more the MI deviates from zero. This index has been further improved by Ozkurt et al. with a simple normalization [26]. Another approach [23, 44] has been to partition [0, 2π] into smaller intervals to get the time points t when ϕx(t) is within each interval, and to compute the mean of ay(t) on these time points. PAC was then quantified by looking at how much the distribution of ay differs from uniformity with respect to ϕx. For instance, a simple height ratio [44], or a Kullback-Leibler divergence as proposed by Tort et al. [23], can be computed between the estimated distribution and the uniform distribution. Alternatively, it was proposed in [11] to use direct correlation between x and ay. As this method yielded artificially weaker coupling values when the maximum amplitude ay was not exactly on the peaks or troughs of x, this method was later extended to generalized linear models (GLM) using both cos(ϕx) and sin(ϕx) by Penny et al. [22]. Other approaches employed a measure of coherence [45] or the phase-locking value [46]. All these last three approaches offer metrics which are independent of the phase at which the maximum amplitude occurs. The methods of Tort et al. [23], Ozkurt et al. [26], and Penny et al. [22] will be considered for comparison in our experiments.
As one can see, there is a long list of methods to quantify CFC in neural time series. Yet, a number of limitations which can significantly affect the outcomes and interpretations of neuroscientific findings exist with these approaches. For example, in typical PAC analysis, a systematic bias rises where one constructs the so-called comodulogram. A comodulogram is obtained by evaluating the chosen metric over a grid of frequency fx and fy. This bias emerges from the choice of the bandpass filter, which involves the critical choice of the bandwidth Δfy. It has been reported several times that to observe any amplitude modulation, the bandwidth of the fast oscillation Δfy has to be at least twice as high as the frequency of the slow oscillations fx: Δfy > 2fx [27, 47]. As a comodulogram uses different values for fy, many studies have used a variable bandwidth, by taking a fixed number of cycles in the filters. The bandwidth is thus proportional to the center frequency: Δfy ∝ fy. This choice leads to a systematic bias, as it hides any possible coupling below the diagonal fy = 2fx/α, where α = Δfy/fy is the proportionality factor. Other studies have used a constant bandwidth Δfy; yet this also biases the results towards the low driver frequency fx, considering that it hides any coupling with fx > Δfy/2. A proper way to build a comodulogram would be to take a variable bandwidth Δfy ∝ fx, with Δfy > 2fx. However, this is not common practice as it is computationally very demanding, because it implies to bandpass filter y again for each value of fx.
Another common issue arises with the use of the Hilbert transform to estimate the amplitude and the phase of real-valued signals. Such estimations rely on the hypothesis that the signals x and y are narrow-band, i.e. almost sinusoidal. However, numerous studies have used this technique on very wide-band signals such as the entire gamma band (80-150 Hz) [6] (see other examples in [28]). The narrow-band assumption is debatable for high frequency activity and, consequently, using the Hilbert transform may yield non-meaningful amplitude estimations, and potentially poor estimations of PAC [27, 28]. Note also that, in this context, wavelet-based filtering is equivalent to the Hilbert transform [48, 49], and therefore does not provide a more valid alternative option.
Besides these issues of filtering and inappropriate use of Hilbert transforms, Hyafil [29] also warned that certain choices of bandwidth Δfy might mistake phase-frequency coupling for PAC, or create spurious amplitude-amplitude coupling; see also the more recent work in [24] for discussion and more practical recommendations for PAC analysis.
Here we advocate that the DAR models detailed in the next sections address a number of the limitations just mentioned. They do not use bandpass filter or Hilbert transform on the high frequencies y. They introduce a measure of goodness of fit, through the use of a probabilistic signal model whose quality can be assessed by evaluating the likelihood of the data under the model. In practice, the likelihood quantifies how much variance of the signal can be explained by the model, and is similar to the R2 coefficient in generalized linear models (GLM). To the best of our knowledge, the only related model-based approach to measure PAC used GLM [22]. With GLM, however, the modeling part is done independently on each signal yf, which is the band-pass filtered version of y around frequency f. For each of these frequencies f a different model is fitted. By doing so, a GLM approach cannot model the wide-band signal y as it is limited to multiple estimations frequency by frequency bin. This largely limits the use of the likelihood to compare models or parameters. On the contrary, we propose to model y globally, without filtering it in different frequency bands.
To conclude this section, and to position this work in the broader context of modeling approaches for neuroscience data, we would like to stress that our proposed method can be considered as an encoding model for CFC, as opposed to a decoding model [50–52]. Indeed, our model reports how much empirical data can be explained and by doing so enables us to test neuroscience hypotheses in a principled manner [51].
The literature on the use of non-linear auto-regressive (AR) models is quite large and covers fields such as audio signal processing and econometrics. For instance, AR models with conditional heteroskedasticity (ARCH [53], GARCH [54]) are extremely popular in econometrics where they are used to model signals whose overall amplitude varies as a function of time. Here, however, in the context of CFC and PAC, one would like to model variations in the spectrum itself, such as shifts in peak frequencies (a.k.a. frequency modulations) or changes in amplitude only within certain frequency bands (a.k.a. amplitude modulations). To achieve this, one idea is to define a linear AR model, whose coefficients are a function of time and change slowly depending on a non-linear function of the signal.
The first models based on this idea are SETAR models [41], which switch between several AR models depending on the amplitude of the signal with respect to some thresholds. To get a smoother transition between regimes, SETAR models have inspired other models like EXPAR [55] or STAR [42], in which the AR coefficients change continuously depending on a non-linear function of the past of the signal.
These models share the same underlying motivation as the DAR models described below but, crucially, DAR models can be designed and parametrized to capture PAC phenomena independently of the phase in the driving signal at which the high frequency content is the strongest. In other words, DAR models can work equivalently well if the high frequency peaks are in the troughs, the rising phase, the decreasing phase or the peaks of the low frequency driving signal. Moreover, as DAR models do not require to infer the driving behavior from the signal itself and rather rely on the prior knowledge of the slow oscillation, the inference is significantly faster and more robust.
In this section, we first define our statistical signal model, explain how we estimate its parameters on a signal y and its driver x, and demonstrate how one can infer hyper-parameters by comparing the likelihood of several models. Then, we detail how to use this model on neurophysiological signals by presenting the preprocessing steps and showing how to make the model invariant to the phase of the coupling. We also detail how to express the power spectral density (PSD) conditionally to the driver’s phase, which allows for a fine-grained analysis of the signal’s spectral properties and to build comodulograms. Finally, we present our protocol to simulate signal with CFC and the empirical datasets used to validate our method.
We now illustrate the benefits of DAR models using simulated data and neurophysiological recordings in rats and humans. Our primary goal is to demonstrate that assessing the validity of the model through its likelihood is a statistically principled approach that allows to reveal new insights on empirical data with interesting neuroscientific consequences.
In this section, we present the outcome of using the model selection procedure to estimate the best filtering frequency fx and bandwidth Δfx to extract the driver x. We first describe the outcome on simulated signals (ground truth) and then on empirical datasets.
Given that DAR models are parametric with a limited number of parameters to estimate, less time samples may be needed to estimate PAC as compared to non-parametric methods. We tested this assumption using simulated signals of varying duration. We computed their comodulograms (as in Fig 7) and selected the frequencies of maximum coupling. For each duration, we simulated 200 signals, and plotted the 2D histogram showing the fraction of time each frequency pairs corresponded to a maximum. We then compared the same four methods: DAR models with (p, m) = (10, 1), the GLM-based model [22], and two non-parametric methods [23, 26]. Results shown in Fig 9 show that parametric approaches provided a more robust estimation of PAC frequencies with short signals (T = 2 sec) than non-parametric methods.
The robustness to small sample size is a key feature of parametric models, as it significantly improves PAC analysis during shorter experiments. When undertaking a PAC analysis across time using a sliding time window, parametric models should therefore provide more robust PAC estimates. Note that the specific time values in these simulations should not be taken as general guidelines as they depend on the simulation parameters such as the signal-to-noise ratio. However, across all tests, parametric methods consistently provided more accurate results than non-parametric ones.
One can note that in DAR models, the driver contains not only the phase of the slow oscillation, but also its amplitude. As the driver is not a perfect sinusoid, its amplitude fluctuates with time. On the contrary, most PAC metrics discard the amplitude fluctuations of the slow oscillation and only consider its phase. To evaluate these two options, we compared two drivers using DAR models: the original (complex) driver x(t), and the normalized driver x ˜ ( t ) = x ( t ) / | x ( t ) |. This normalized driver only contains the phase information, as in most traditional PAC metrics. Using cross-validation, we compared the log-likelihood of four fitted models, and found a difference always in favor of the non-normalized driver x(t), as it can be visualized in S3 Fig. This result shows that the coupling phenomenon is associated with amplitude fluctuations, a kind of phase/amplitude-amplitude coupling, as it was previously observed in [25]. Indeed, the GLM parametric method [22] was improved when taking into account the amplitude of the slow oscillation. Here, we use our generative model framework to provide an easy comparison tool through the likelihood, to validate this neuroscientific insight from the signals.
In this section, we report the results of the directionality estimates using both simulations and neurophysiological signals. It is noteworthy that in DAR models, we arbitrarily call driver the slow oscillation although the model makes no assumption on the directionality of the coupling.
Cross-frequency coupling (CFC) and phase-amplitude coupling (PAC) more specifically have been proposed to play a fundamental role in neural processes ranging from the encoding, maintenance and retrieval of information [3–5, 8, 17, 81, 82], to large-scale communication across neural ensembles [7, 19, 83, 84]. While a steady increase in observations of PAC in neural data has been seen, how to best detect and quantify such phenomena remains difficult to settle. We argue that a method using DAR models, as described here, is rich enough to capture the time-varying statistics of brain signals in addition to provide efficient inference algorithms. These non-linear statistical models are probabilistic, allowing the estimation of their goodness of fit to the data, and allowing for an easy and fully controlled comparison across models and parameters. In other words, they offer a unique principled data-driven model selection approach, an estimation strategy of phase/amplitude-amplitude coupling based on the approximation of the actual signals, a better temporal resolution of dynamic PAC and the estimation of coupling directionality.
One of the main features of PAC estimation through our method is the ability to compare models or parameters on non-synthetic data. On the contrary, traditional PAC metrics cannot be compared on non-synthetic data, and two different choices of parameters can lead to different interpretations. There is no legitimate way to decide which parameter shall be used with empirical data using traditional metrics. The likelihood of the DAR model that can be estimated on left-out data offers a rigorous solution to this problem.
We presented here results on both simulated signals and empirical neurophysiological signals. The simulations gave us an illustration of the phenomenon we want to model, and helped us understand how to visualize a fitted DAR model. They also served a validation purpose for the bandwidth selection approach that we performed on real data. Using the data-driven parameter selection on non-synthetic signals, we showed how to choose sensible parameters for the filtering of the slow oscillation. All empirical signals are different, and it was for example reported in the neuroscience literature that peak frequencies vary between individuals [85] and that this should not be overlooked in the analysis of the data. The parameter selection based on fitted DAR models makes it possible to fit parameters on individual datasets. Our results also shed light on the asymmetrical and wide-band properties of the slow oscillation, which could denote crucial features involved in cognition [32].
The second novelty of our method stands in considering the amplitude fluctuations of the slow oscillation in the PAC measure and not only its phase. Using the rodent and human data, we showed that the instantaneous amplitude of the slow oscillation influences the coupling in PAC, as it was previously suggested in [25]. The amplitude information should therefore not be discarded as it is done by existing PAC metrics. For instance, the measure of alpha/gamma coupling reported during rest [86, 87] should incorporate alpha fluctuations when studied in the context of visual tasks [88], as an increase of alpha power is often concomitant with a decrease of gamma power [89]. The comparison between DAR models considering or not these low-frequency power fluctuations would inform on the nature of the coupling: purely phase-amplitude, or rather phase/amplitude-amplitude. In Tort et al. [14], both theta power changes and modulation of theta/gamma PAC were reported in rats having to make a left or right decision to find a reward in a maze. The use of our method could decipher whether the changes in coupling were related to the changes in power, informing on the underlying mechanisms of decision-making. Moreover, as our method models the entire spectrum simultaneously, a phase-frequency coupling could potentially be captured in our models. Therefore, our method is not limited to purely phase-amplitude coupling, and extends the traditional CFC analysis.
Furthermore, in those types of experiments, changes in PAC can be very fast depending on the cognitive state of the subject. Therefore, the need for dynamic PAC estimates is growing [14]. We showed with simulations that DAR models are more robust than non-parametric methods when estimating PAC on small time samples. This robustness is critical for time-limited experiments and also when analyzing PAC across time in a fine manner, typically when dynamic processes are at play.
Last but not least, likelihood comparison can also be used to estimate the delay between the coupled components, which would give new insights on highly debated questions on the role of oscillations in neuronal communication [90, 91]. For example, a delay close to zero could suggest that the low and high frequency components of the coupling might be generated in the same area, whereas a large delay would suggest they might come from different areas. As an alternative interpretation, the two components may come from the same area, but the coupling mechanism itself might be lagged. In this case, a negative delay would suggest that the low frequency oscillation is driven by the high frequency oscillations, whereas a positive delay would suggest that the low frequency oscillation drives the high frequency amplitude modulation. In any case, this type of analysis will provide valuable information to guide further experimental questions.
A recent concern in PAC analysis is that all PAC metrics may detect a coupling even though the signal is not composed of two cross-frequency coupled oscillators [30, 92–95]. It may happen for instance with sharp slow oscillations, described in humans intracranial recordings [68]. Sharp edges are known not to be well described by a Fourier analysis, which decomposes the signal in a linear combination of sinusoids. Indeed, such sharp slow oscillations create artificial high frequency activity at each sharp edge, and these high frequencies are thus artificially coupled with the slow oscillations. This false positive detection is commonly referred to as “spurious” coupling [96]. Fig 12 shows a comodulogram computed on a simulated spurious PAC dataset, using a spike train at 10 Hz, as described in [94]. The figure shows that all four methods, including the proposed one, detect some significant PAC, even though there is no nested oscillations in the signal. Even though our method does not use filtering in the high frequencies, it does not solve this issue and is affected in the same way as other traditional PAC metrics. Indeed, our work shed light on the wide-band property of the slow oscillations, but DAR models cannot cope with full-band slow oscillations, which contain strong harmonic components in the high frequencies. However, we consider that such “spurious” PAC can also be a relevant feature of a signal, as stated in [68]. In their study, they show that abnormal beta oscillations (13-30 Hz) in the basal ganglia and motor cortex underlie some “spurious” PAC, but are actually a strong feature associated with Parkinson’s disease. A robust way to disentangle the different mechanisms that lead to similar PAC results remains to be developed.
The method we presented in this paper uses univariate signals obtained invasively in rodents or humans. As a lot of neurophysiological research uses non-invasive MEG or EEG recordings containing multiple channels, a multivariate analysis could be of high interest. One way to use data from multiple channels is to estimate a single signal using a spatial filter such as in [97]. Such a method is therefore complementary to univariate PAC metrics like ours which can be applied to the output of the spatial filter. The method from [97] builds spatial filters that maximize the difference between, say, high-frequency activity that appears during peaks of a low-frequency oscillation versus high-frequency activity that is unrelated to the low-frequency oscillation. Again, from the signal obtained with the spatial filter, it is straightforward to adapt most PAC metrics such as our method.
Neurophysiological signals have all the statistical properties to make them a challenge from a signal processing perspective. They contain non-linearities, non-stationarities, they are noisy and they can be long, hence posing important computational challenges. Our method based on DAR models offer novel and more robust possibilities to analyse neurophysiological signals, paving the way for new insights on how our brain functions via spectral interactions using local or distant coupling mechanisms.
Inline with the open science philosophy of this journal, our method is fully available as an open source package that comes with documentation, tests, and examples: https://pactools.github.io.
|
10.1371/journal.pgen.1003463 | Functional Dissection of the Drosophila melanogaster Condensin Subunit Cap-G Reveals Its Exclusive Association with Condensin I | The heteropentameric condensin complexes have been shown to participate in mitotic chromosome condensation and to be required for unperturbed chromatid segregation in nuclear divisions. Vertebrates have two condensin complexes, condensin I and condensin II, which contain the same structural maintenance of chromosomes (SMC) subunits SMC2 and SMC4, but differ in their composition of non–SMC subunits. While a clear biochemical and functional distinction between condensin I and condensin II has been established in vertebrates, the situation in Drosophila melanogaster is less defined. Since Drosophila lacks a clear homolog for the condensin II–specific subunit Cap-G2, the condensin I subunit Cap-G has been hypothesized to be part of both complexes. In vivo microscopy revealed that a functional Cap-G-EGFP variant shows a distinct nuclear enrichment during interphase, which is reminiscent of condensin II localization in vertebrates and contrasts with the cytoplasmic enrichment observed for the other EGFP-fused condensin I subunits. However, we show that this nuclear localization is dispensable for Cap-G chromatin association, for its assembly into the condensin I complex and, importantly, for development into a viable and fertile adult animal. Immunoprecipitation analyses and complex formation studies provide evidence that Cap-G does not associate with condensin II–specific subunits, while it can be readily detected in complexes with condensin I–specific proteins in vitro and in vivo. Mass-spectrometric analyses of proteins associated with the condensin II–specific subunit Cap-H2 not only fail to identify Cap-G but also the other known condensin II–specific homolog Cap-D3. As condensin II–specific subunits are also not found associated with SMC2, our results question the existence of a soluble condensin II complex in Drosophila.
| The accurate duplication and segregation of chromosomes during cell divisions are prerequisites for ensuring genetic stability within an individual organism and in entire populations. Among the many components involved in regulating these processes, a protein complex called condensin plays a crucial role in shaping mitotic chromosomes, so that they can be faithfully distributed. Many organisms contain two of these condensin complexes (condensin I and II), which both have been shown to be required for accurate chromosome distribution. In the fly Drosophila melanogaster, condensin II appears to lack one of its components, called Cap-G2. We have tested the hypothesis whether the corresponding component of condensin I (Cap-G) might also participate in the assembly of condensin II. Careful analyses of complexes formed in the living organism or in the test tube argue against Cap-G being part of condensin II. Moreover, our results question the very existence of a soluble condensin II complex in flies, as opposed to other organisms. Surprisingly, a substantially truncated variant of the essential Cap-G still supports development of living and fertile flies. As this variant localizes within the cell differently from full-length Cap-G, our results show that localization of a protein does not always determine its function.
| Chromosome condensation is a critical process ensuring faithful distribution of the replicated genetic information onto the daughter cells. While the exact mechanism underlying the longitudinal compaction of the dispersed interphase chromatin into the rod-like and sturdy metaphase chromosomes is still subject of intense research, the participation of the condensin complexes in this process has been thoroughly demonstrated (for review see [1]–[3]). However, while condensin is clearly required and sufficient for compaction of sperm chromatin incubated in Xenopus laevis egg extracts [4], [5], the phenotypes observed after condensin depletion in other systems suggest the existence of alternative mechanisms mediating chromatin compaction. Condensin depletion in vertebrate cells, worms and flies does affect the structure of mitotic chromosomes, but compaction of chromatin is only slightly impaired. The extent of this compaction phenotype varies by the organism studied and the experimental system used (for review see [3]). However, in all cases, persistent interconnections of chromatin fibres can be observed in anaphase (so-called anaphase bridges), resulting in severe problems during chromatid segregation in mitosis. Thus, condensin has a role in resolving chromatin bridges present between the replicated chromatids.
Plants and animals harbour two condensin complexes, both containing the structural maintenance of chromosomes (SMC) proteins SMC2 and SMC4, but differing in their non-SMC regulatory subunits. Condensin I complexes contain the subunits Cap-D2, Cap-G and Cap-H (also called Barren in Drosophila), while condensin II complexes contain the related subunits Cap-D3, Cap-G2 and Cap-H2. Cap-H and Cap-H2 belong to the kleisin family of proteins which are characterized by their ability to bind to the head domains of SMC protein dimers [6]. Cap-G, Cap-G2, Cap-D2 and Cap-D3 contain in their N-terminal parts extended regions of Huntingtin, elongation factor 3, A-subunit of protein phosphatase 2A, TOR1 lipid kinase (HEAT) repeats, which are thought to mediate protein-protein interactions [7]. In vertebrates, both condensin complexes play essential roles and collaborate in structuring of mitotic chromosomes and in ensuring their unperturbed segregation. Interestingly, the two complexes fulfil non-overlapping functions as exemplified by distinct phenotypes upon depletion of either condensin I or condensin II-specific subunits [8]–[10], by their alternating association with mitotic chromosomes [11], [12], or by their different localization in interphase cells: Condensin I-specific subunits are enriched in the cytoplasm, while condensin II-specific subunits can be found primarily in the nucleus [9]–[11]. Within the eukaryotic kingdom, the composition of the condensin complexes found in different species is not uniform. Fission and budding yeast harbour homologs only for condensin I, as do e.g. ciliates and kinetoplastids (for review see [3]). C. elegans, on the other hand, contains three condensin complexes, one of which (condensin IDC) has specialized to function in dosage compensation in hermaphrodites [13]. In Drosophila melanogaster, condensin I is present, but for condensin II only the subunits Cap-H2 and Cap-D3 can be identified. No gene encoding the condensin II-specific subunit Cap-G2 is apparent in the genome. This has led to the speculation that Drosophila Cap-G might be a component of both complexes, just as SMC2 and SMC4 [14]–[16]. The essential role for all condensin I-specific subunits in mitotic proliferation is well established [14], [17]–[22]. On the other hand, loss-of-function mutations of the Drosophila genes encoding Cap-H2 and Cap-D3 are viable, indicating that their function is dispensable for mitotic proliferation [18], [23], [24]. However, Cap-D3 and Cap-H2 mutant males are sterile, and cytological as well as genetic evidence clearly indicates a role during male meiosis for these two subunits [23]. Interestingly, mutations in Cap-H2 have also been shown to prevent the dispersal of nurse cell polytene chromosomes, which are present for a short developmental period during oogenesis, and to enhance transvection phenomena. Conversely, Cap-H2 overexpression leads to dispersal of the polytene chromosomes in larval salivary glands and in addition suppresses transvection [24]. These results suggest that Cap-H2 negatively regulates chromosome associations and additional genetic evidence indicates that this function is dependent on Cap-D3 [24]. Moreover, Cap-D3 has been shown to interact with the Drosophila Retinoblastoma (Rb)-protein homolog Rbf and the two proteins colocalize on the regulatory regions for transcription of the antimicrobial peptide (AMP) genes, thereby influencing innate immunity [16], [25]. Thus, the Drosophila condensin II subunits Cap-H2 and Cap-D3 perform roles in regulating gene expression, as has been demonstrated for condensin complexes in other studies [20], [21], [26], [27]. However, whether these functions are performed in the context of a physical protein complex containing SMC2, SMC4, Cap-H2, Cap-D3 and possibly Cap-G is unknown. While biochemical evidence for the existence of a soluble condensin I complex has been published [18], the existence and protein composition of a soluble condensin II-like complex in Drosophila is uncertain.
Here, we have analyzed in detail the localization behaviour and complex formation capabilities of Drosophila Cap-G in vivo and in vitro to test the hypothesis, whether it might be a common component of both condensin complexes in Drosophila. The comparison of the localization and dynamics of various fluorescently tagged, functional condensin subunits highlights the fact that Cap-G indeed behaves differently from other condensin I-complex components. However, complex formation studies strongly argue against Cap-G being associated with condensin II-specific components. Furthermore, immunoprecipitation analyses consistently provide evidence for soluble condensin I complexes, but fail to support the presence of native soluble condensin II complexes in vivo and indicate a strongly reduced complex formation potential in vitro. Thus, while we cannot exclude the assembly of condensin II-like complexes specifically on chromatin in specialized cell types, our data argue against the existence of an abundant and stable soluble condensin II complex in Drosophila.
In interphase, vertebrate condensin I subunits are primarily cytoplasmic, while condensin II subunits are primarily nuclear [9]–[11]. Consistently, Drosophila Barren/Cap-H and Cap-H2 have also been found to be cytoplasmic or nuclear enriched, respectively [24], [28]. Towards a comparative description of the localization behavior of Drosophila condensin subunits in the living organism, we have generated EGFP-fused variants of the condensin subunits Cap-D2, SMC2 and Cap-G (Figure S1A). EGFP-Cap-D2 should label exclusively condensin I-complexes, while SMC2h-EGFP is expected to occur in both condensin I and condensin II. As no condensin II-specific Cap-G2 subunit has been identified in Drosophila, Cap-G has been hypothesized to be also part of both condensin complexes [14]–[16]. Thus, Cap-G localization may provide a hint as to whether it is part of only condensin I or both condensin complexes in Drosophila.
All three transgene constructs are expressed under control of the flanking genomic regulatory sequences and quantification of the expression levels in early embryogenesis reveal a ratio of transgene products of approximately 1∶4∶8 (Cap-G-EGFP∶SMC2h-EGFP∶EGFP-Cap-D2; Figure S1B). Despite these differences, all transgenes encode biologically functional products as the presence of single copies of the transgenes can complement the lethality associated with loss-of-function-mutations in the respective genes (Table S1 and data not shown). Analysis of living embryos progressing through the divisions of the syncytial blastoderm revealed that during interphase, SMC2h-EGFP and EGFP-Cap-D2 are enriched in the cytoplasm, as has been reported previously for the condensin I - specific subunit Cap-H/Barren (Figure 1A; Videos S1 and S2; [28]). In contrast, Cap-G-EGFP is nuclear enriched in interphase, reminiscent of condensin II localization in vertebrates (Figure 1A, Video S3). All three EGFP-fused subunits rapidly associate with condensing chromatin at early stages of mitosis. However, Cap-G-EGFP associates with chromatin slightly earlier than EGFP-Cap-D2 and SMC2h-EGFP, which might be due to its preferential nuclear localization in interphase. All condensin subunits leave chromatin during late anaphase/early telophase (Figure 1A; Videos S1, S2, S3). As the different condensin subunits exhibit distinct localization patterns during interphase, and differ in their chromatin association kinetics, we scrutinized the dynamics of mitotic chromatin association of these subunits during cycle 12 of the syncytial divisions. To this end, we performed quantitative measurements of the EGFP fluorescence signals and normalized them to the simultaneously recorded fluorescence measurements of the mRFP1-fused histone variant His2Av, which was also expressed in these embryos. The data revealed that Cap-G-EGFP is loaded maximally already at nuclear envelope breakdown, a time-point when the EGFP-fused subunits Cap-D2 and Cap-H/Barren (data from [28]) are just beginning to associate with chromatin (Figure 1B). Interestingly, SMC2h-EGFP loading appears even more delayed (half-maximal association of SMC2 is −2.5 min before anaphase onset vs. −3.5 min for Cap-D2 and Cap-H/Barren; Figure 1B). Similar loading kinetics are observed, when SMC2h-EGFP chromatin association was determined in an SMC2 mutant background, ruling out the possibility that the presence of endogenous SMC2 significantly delays incorporation of the EGFP-fused variant (Figure S2). For all analyzed subunits, maximal chromatin association levels are achieved during late metaphase/early anaphase. During exit from mitosis, the four condensin subunits delocalize from chromatin with almost identical kinetics (Figure 1B).
To assess, which regions of Cap-G mediate the subcellular localization during the cell cycle, we expressed various EGFP-fused deletion constructs under GAL4/UAS-control in early embryos and analyzed the localization behavior of the fusion proteins while cells were progressing through epidermal mitosis 14 (Figure 2A). Computational analyses predict nuclear localization signals (NLS) at positions 1072, 1162, and 1210. Consistently, a C-terminal Cap-G fragment (Cap-GC; aa 958–1351) encompassing these signals is strongly nuclear enriched in interphase. At nuclear envelope breakdown, the fusion protein distributes throughout the cell (Figure 2B, Video S4). During early to mid mitosis, Cap-GC-EGFP associates only very weakly with chromatin. However, beginning with late anaphase, Cap-GC-EGFP accumulates on the segregating chromatids (Figure 2B, Video S4). The construct Cap-GNM-EGFP (aa 1–977) lacks the C-terminal region with the NLS, but retains an extended region predicted to form HEAT-repeats and it displays a complementary localization behavior when compared to Cap-GC-EGFP. In interphase, this Cap-G variant is primarily localized in the cytoplasm, but approximately 20–40 sec after nuclear envelope breakdown, it associates rapidly and efficiently with mitotic chromatin (Figure S3). Starting with anaphase, Cap-GNM-EGFP dissociates from chromatin similar to full length Cap-GFL-EGFP (Figure 2B, Video S5) and as was observed for the other condensin subunits (Figure 1B; [28]).
To assess, whether the mitotic localization behavior of Cap-GNM-EGFP reflects its potential to form complexes with the other condensin subunits, we performed immunoprecipitation analyses. Extracts were prepared from embryos expressing various EGFP-fused Cap-G variants followed by precipitation using anti-EGFP antibodies. Proteins bound to the beads were eluted in two steps, with the second being more stringent. Four prominent protein bands in the high molecular weight range can be detected on silver stained gels in the first round eluates of both Cap-GFL-EGFP and Cap-GNM-EGFP-coupled beads (Figure 2C). The identity of two of the bands was confirmed as Cap-D2 and Cap-H/Barren by immunoblot analysis (Figure S4). Based on their migration behavior, the first and third bands were suspected to correspond to SMC4 and SMC2, respectively. This assignment was corroborated by mass-spectrometric analyses of Cap-GFL-EGFP immunoprecipitates (see below). The antibody-bound EGFP-fused Cap-G variants were primarily eluted under more stringent conditions (Figure 2C). Cap-GC-EGFP immunoprecipitates did not contain the other condensin I subunits in significant amounts, as did not the precipitates of a Cap-GNM-EGFP- variant with a further N-terminal truncation of 242 amino acids (Cap-GNM4-EGFP). This latter variant does not localize to mitotic chromatin and it is distributed in interphase throughout the cell (Figure S5). The HEAT repeats predicted to form in the N-terminal region of Cap-G are implicated in protein-protein interactions [7]. Thus, binding of Cap-G to the condensin complex may be mediated via the HEAT-repeat motifs, since Cap-GNM4-EGFP lacking a large part of this domain is not able to precipitate Cap-D2 or Cap-H/Barren. However, the N-terminal 242 amino acids are not sufficient for efficient association with mitotic chromatin, since the variant Cap-GNM1-EGFP, which encompasses the region of aa 1–848, is primarily cytoplasmic in interphase like Cap-GNM-EGFP, and associates only very weakly with chromatin during mitosis (Figure S5). We conclude that the C-terminal third of Cap-G contains nuclear localization sequences, but it is dispensable for mitotic chromatin association. Moreover, the HEAT-repeat region as well as the stretch encompassing aa 848–977 within the N-terminal two-thirds of Cap-G are required for binding to mitotic chromatin, most probably by virtue of their mediating the assembly into condensin complexes.
We have noticed that during interphase, Cap-GFL-EGFP and Cap-GC-EGFP are not homogeneously distributed in the nucleoplasm. As the patchy appearance of Cap-G signals is reminiscent of heterochromatin distribution in these nuclei, we analyzed embryos expressing EGFP-fused Cap-GFL or Cap-GC concomitant with a red fluorescently labeled variant of heterochromatin protein 1 (mRFP1-HP1) (Figure 3A). HP1 binds to histone H3 methylated at lysine 9 and is thus a marker for heterochromatin distribution in interphase cells [29]. In vivo microscopy of embryos progressing through epidermal cycle 14 revealed that the two Cap-G variants indeed largely co-localize with mRFP1-HP1 during interphase, indicating heterochromatin association of Cap-G (Figure 3A). During mitosis, mRFP1-HP1 dissociates from chromatin, as has been previously observed with fixed material (arrowheads in Figure 3A; [30]). This observation, together with the fact that Cap-GC associates with chromatin in late mitosis when mRFP1-HP1 is still absent, indicates that Cap-G chromatin association does not depend on the presence of HP1. While Cap-G clearly co-localizes with heterochromatin in interphase, it does not appear to be physically associated with HP1 in a common protein complex as HP1 cannot be co-precipitated with Cap-G (Figure S6).
Embedded within the heterochromatin are the centromeres. As enrichment of other condensin subunits in centromeric regions has been demonstrated [11], [17], [31]–[33] and a genetic and physical interaction of Cap-G with the centromeric H3 variant Cid has been established [14], we scrutinized the dynamics of Cap-G chromatin association. To this end, we analyzed the localization behavior of Cap-GFL-EGFP in comparison with Cid-mRFP1 in embryos progressing through cycle 14. Indeed, early chromatin accumulation of Cap-GFL-EGFP occurs in nuclear regions where Cid-mRFP1 signals can be detected (Figure 3B). Similar dynamics are observed when embryos progress through syncytial cycle 12, and quantitation reveals an approximately twofold enrichment of Cap-GFL-EGFP in centromere-proximal vs. centromere-distal regions in early stages of Cap-G chromatin association (Figure S7). Thus, our observations are consistent with a model in which Cap-G first binds to centromeric regions and then spreads into the adjacent heterochromatin.
The C-terminus of Cap-G is required for nuclear localization and sufficient to confer heterochromatic enrichment during interphase. The N-terminal two-thirds of Cap-G, on the other hand, are sufficient for efficient chromatin localization during mitosis and for assembly within the condensin I holocomplex. To assess the relevance of the functional features contributed by the Cap-G C-terminus, we generated individuals expressing Cap-GNM or Cap-GNM-EGFP as sole source for this condensin subunit in a Cap-G1/Cap-G6 trans-heterozygous mutant background. Loss-of-function mutations in Cap-G are embryonic lethal [14], [20]. Expression of Cap-GFL-EGFP, either under control of the genomic regulatory sequences or under GAL4/UAS control using the ubiquitous da-GAL4 driver, gave rise to viable and fertile adults demonstrating the biological functionality of these constructs (Table S1). Surprisingly, adult flies were also obtained with high efficiency by ubiquitous expression of two independent pUAST-based UAS-Cap-GNM-EGFP-transgene insertions in the same trans-heterozygous Cap-G mutant background. As pUAST does not direct expression in the female germline, female fertility could not be assessed in these cases. However, expression from the Cap-GNM-EGFP transgene contained in a pUASP-based vector, which also allows expression in the female germline [34], restored fertility in both sexes (Table S1). Immunoblot analysis confirmed that these animals lacked expression of endogenous Cap-G and survived solely due to the expression of the C-terminally truncated Cap-G variants (Figure S8A). To assess, whether the C-terminally truncated Cap-GNM variant also fails to localize to interphase nuclei in the absence of competing full-length Cap-G, we analyzed Cap-GNM-EGFP localization in the rescue situation. Cap-GNM-EGFP is excluded from the nuclei in interphase also in a Cap-G mutant background, and it does not bind to chromatin in prophase, ruling out the possibility that the presence of competing full-length Cap-G might prevent early chromatin association of the Cap-GNM variant (Figure S8B and see also Figure S10C). Not all Cap-GNM transgenes complemented the Cap-G mutant phenotype efficiently. Fertility was only observed after crosses of rescued individuals with wild type flies, and many eggs laid by Cap-GNM rescued mothers displayed developmental defects (data not shown). Therefore, it was not possible to establish stable rescue stocks. We conclude nevertheless that the C-terminal 374 amino acids of Cap-G are not absolutely critical for condensin function required for development from the fertilized egg to a fertile adult. While the full-length protein rescues with higher efficiency than the C-terminal truncated version when expressed at comparable levels (Table S1; genomic transgenes), the development of fertile adult animals is still possible when the C-terminal domain of Cap-G is lacking. As this C-terminal part contains the NLS, nuclear enrichment of Cap-G during interphase is dispensable for condensin function in the cell cycle and during development.
Due to the lack of an obvious Cap-G2 homolog encoded in the Drosophila genome, Cap-G has been hypothesized to be part of both condensin subunits, just as SMC2 and SMC4 [14]–[16]. In the anti-Cap-GFL-EGFP immunoprecipitates shown in Figure 2C, four prominent high molecular weight bands are evident, which were assigned to the condensin I-specific subunits and the two SMC's. As the condensin II-specific subunits Cap-H2 and Cap-D3 might not have been abundant enough in the analyzed extracts to be detected by silver staining, we performed additional immunoprecipitation experiments followed by sensitive mass spectrometric (MS) analysis of the precipitates. We have used a variety of strains expressing condensin subunits fused with fluorescent proteins, which were precipitated with the appropriate antibodies (Figure 4A). First, we prepared extracts from early embryos or from ovaries isolated from individuals expressing Cap-GFL-mRFP1 under the control of the genomic regulatory sequences. Like the EGFP-fused Cap-G variant, mRFP1-fused Cap-G is biologically functional as it rescues Cap-G mutants to vitality and fertility (data not shown). After immunoprecipitation using anti-mRFP1 antibodies, aliquots of the eluates were separated on an SDS-polyacrylamide gel and stained with silver to visualize the precipitated proteins (Figure 4B). In a parallel experiment, lanes with the eluates were stained with colloidal Coomassie Blue, cut into seven slices each and processed for MS. This procedure allowed a comprehensive evaluation of the proteins associated with the precipitated bait. As a negative control, an extract from w1-ovaries not containing mRFP1-fused proteins was treated identically. From the list of identified proteins all non-Drosophilid proteins were removed, and then sorted according to the cumulative intensities of the identified peptides. In both the ovary and the embryo extracts, among the top eleven most abundant proteins, SMC2, SMC4, as well as the condensin-I specific subunits Cap-H/Barren and Cap-D2 were identified (Figure 4B). The majority of the peptides specific for SMC2 or SMC4 were detected in gel slices containing proteins of molecular weights corroborating our assignment of the SMCs in the silver stained IP-eluates shown in Figure 2C. However, in the complete list of identified proteins (189 for the embryonic extracts and 537 for the ovary extract), neither Cap-D3 nor Cap-H2 were found, not even represented by a single peptide (Tables S2 and S3). In a complementary approach, we expressed EGFP- and mCherry-fused variants of the condensin II-specific subunit Cap-H2 in ovaries using the GAL4/UAS-system. These variants were shown to be functional as they I) rescue the phenotypic consequences described for Cap-H2 mutants in ovarian nurse cell nuclei and II) trigger a dispersal of polytene chromatin when expressed in the nuclei of larval salivary glands (Figure S9; [24]). Anti-EGFP-Cap-H2 and anti-mCherry-Cap-H2 precipitates from ovarian extracts were separated by SDS-PAGE, stained with colloidal Coomassie Blue, and analyzed by MS (Figure 4C). Within the lists of identified proteins, SMC2 and SMC4 can be found in both experiments. However, the SMCs were ranked much lower in this experiment when compared to the Cap-G immunoprecipitates, indicating that they are of relatively low abundance in the Cap-H2-specific precipitates. Significantly, within the complete list of more than 1200 proteins in both cases, neither Cap-G nor Cap-D3 could be found (Table S4). The N-terminal EGFP- and mCherry-fusions in our Cap-H2 constructs may preclude efficient complex formation. Therefore, we also performed immunoprecipitations of SMC2, from protein extracts of wild type or SMC2h-EGFP expressing individuals, using either anti-SMC2-antibodies or anti-EGFP antibodies, respectively. In these experiments, we would expect to precipitate both condensin I and condensin II complexes. Again, we could identify the components of the condensin I complex in all cases, but in none of the three experiments, the condensin II-specific components Cap-H2 or Cap-D3 were detected (Figure 4D, Tables S5, S6, S7, S8).
As Cap-GFL is nuclear during interphase, like condensin II subunits in other systems, one might expect condensin II-like phenotypes in Cap-G mutant animals rescued by Cap-GNM, which is cytoplasmic in interphase. A prominent phenotype in Drosophila Cap-D3 and Cap-H2 mutants is the perdurance of nurse cell chromosome polyteny in developing egg chambers [24]. However, in Cap-GNM rescued females, the nurse cell chromosomes disperse on time, arguing against nuclear Cap-G fulfilling a condensin II-like function (compare Figure S10A and S10B). We have ascertained that in the rescue situation in this tissue, Cap-GNM is also excluded from the nuclei (Figure S10C).
Taken together, the phenotypic analysis of nurse cell chromosomes in Cap-GNM rescued females, as well as our immunoprecipitation analyses argue against Cap-G being incorporated into a soluble condensin II-like complex in Drosophila. Furthermore our MS results also speak against the presence of soluble condensin II-like complexes in the analyzed extracts in significant amounts.
The analysis of condensin subunit associations described above involved immunoprecipitations from complexes present in soluble extracts from Drosophila tissues. To allow the assessment of direct protein-protein interactions in a more simple system, we analyzed complex formation of various condensin subunits produced in an in vitro transcription/translation (IVT) system. In case the molecular mass of the synthesized proteins was sufficiently different, they were co-translated in the presence of [35S]methionine, subjected to immunoprecipitation using antibodies against fused epitope-tags, separated by SDS-PAGE, and detected by autoradiography. Otherwise, proteins were translated in different reactions only one of which contained [35S]methionine. After mixing the extracts and subsequent immunoprecipitation, the components were detected after SDS-PAGE both by autoradiography and immunoblot.
To validate our system, we first wanted to demonstrate the physical interactions between the condensin I-specific non-SMC subunits. We used a C-terminally His-FLAG-epitope-tagged Cap-H/Barren (Barren-HFHF) construct as bait. A C-terminally extended Cap-H/Barren variant has been shown to be biologically functional in the fly [28]. As a negative control, we prepared human securin analogously tagged at its C-terminus with His-FLAG (hSecurin-HFHF). Both Cap-G and Cap-D2 can be specifically co-immunoprecipitated with Barren-HFHF, but not with hSecurin-HFHF (Figure 5A). Thus, the Drosophila HEAT-repeat containing condensin I subunits interact with the kleisin subunit Cap-H/Barren like their human counterparts [35]. If Cap-G is also part of condensin II, one would expect that it forms a complex with the condensin II-specific kleisin subunit Cap-H2. However, while Cap-G can be readily detected in immunoprecipitates of Barren-HFHF, it is not present in Cap-H2-HFHF immunoprecipitates (Figure 5B). This result once more argues against Cap-G being a condensin II component.
The human kleisin subunits were shown to interact with SMC4 [35]. Consistently, Drosophila SMC4 can be precipitated with Barren-HFHF, in low amounts with Cap-H2-HFHF, but not with hSecurin-HFHF (Figure 5C). This result reveals on the one hand a reduced binding efficiency between Drosophila Cap-H2 and SMC4, which is consistent with the results from our immunoprecipitation analysis of ovarian extracts containing ectopically expressed Cap-H2-variants (Figure 4C). On the other hand, it demonstrates that in the IVT-system Cap-H2-HFHF is produced in a conformation competent for complex formation, ruling out the possibility that the lack of interaction between Cap-H2-HFHF and Cap-G is due to mis-folded Cap-H2-HFHF. Next we asked whether we could reconstitute the condensin II-specific interaction between Cap-D3 and Cap-H2. To this end, we synthesized a Cap-D3 variant fused at its N-terminus with six copies of the human c-myc-epitope (myc-Cap-D3). In these experiments, we used as negative control the catalytic (C)-subunit of human protein phosphatase 2A, also with an N-terminal myc6-tag (myc-hPP2(A)C). Cap-H2 could be identified in myc-Cap-D3 immunoprecipitates, but not in myc-hPP2(A)C precipitates (Figure 5D). However, the co-precipitation efficiency was again very low. Cap-H/Barren was also detected in myc-Cap-D3 immunoprecipitates, but this protein was also precipitated by myc-hPP2(A)C, arguing for non-specific associations. To underscore the biological relevance of these in vitro studies, we attempted to form ternary complexes. Based on the geometry of the human condensin complexes, Cap-D2 does not directly interact with the SMC subunits, but the kleisin subunit Cap-H/Barren is expected to bridge Cap-D2 and SMC4. Indeed, SMC4 can be precipitated together with myc-Cap-D2 when Cap-H/Barren is present, but not in its absence (Figure 5E, compare lanes 9 and 11). When Cap-H2 was included in an analogous reaction instead of Cap-H/Barren, Cap-H2 was precipitated with low efficiency, but SMC4 could not be detected (Figure 5E, lane 10). In an effort to reconstitute an analogous condensin II subcomplex, we precipitated myc-Cap-D3 in the presence of both Cap-H2 and SMC4 or just SMC4. In this case, no ternary complex could be detected and only inefficient co-precipitation of Cap-H2 with myc-Cap-D3 was observed (Figure 5F, lane 10). Cap-H/Barren did not co-precipitate with Cap-D3 above background. Taken together, our in vitro complex forming studies confirm the predicted interactions among the Drosophila condensin I-specific subunits. However, the complex forming potential between condensin II-specific subunits is limited and we find again no evidence for incorporation of Cap-G in a condensin II-like subcomplex.
We set out to test the hypothesis that in Drosophila, Cap-G might be part of both condensin I and condensin II. This hypothesis is based on the facts that i) no condensin II-specific Cap-G2 homolog can be identified in the Drosophila genome and ii) that SMC2 and SMC4 are also part of both condensin complexes.
The localization pattern of Cap-G-EGFP in interphase initially suggested its participation in a condensin II-like complex since it was found to be nuclear like vertebrate condensin II subunits [9]–[11]. At least, a functional importance was suggested by the preferential nuclear localization of Cap-G and its different dynamics in chromatin association when compared to the other EGFP-fused condensin I subunits. However, the intriguing observation that flies are viable and fertile, when they exclusively express a C-terminal truncation variant of Cap-G, which is nuclear excluded in interphase and gains access to chromatin only around NEBD, suggests that its nuclear localization is dispensable for proliferation and development, at least under laboratory conditions. Furthermore, the observed heterochromatic enrichment of Cap-G and its initiation of loading at the centromeric regions are obviously not essential. It is possible that the Cap-G C-terminus, which contains many predicted phosphorylation sites in Drosophila and other organisms [36] may fine tune Cap-G activity. This fine-tuning is probably required for the restoration of full fertility in both sexes and early syncytial development, as shown by the defects when no full length Cap-G is provided by the mother. In this respect, the C-terminus might be required for full length Cap-G to be sequestered into the nucleus to avoid any dominant negative effects in the cytoplasm.
SMC2h-EGFP and EGFP-Cap-D2 localize like Cap-H/Barren-EGFP [28] in the cytoplasm during interphase and rapidly associate with chromatin during early stages of mitosis. Intriguingly, these subunits associate significantly later with chromatin than Cap-G-EGFP, indicating that Cap-G has the potential to bind to chromatin in the absence of the other condensin subunits. This notion is supported by the observation that Cap-GC can associate with chromatin in late anaphase, at a time point when the other subunits dissociate. Recently, it has been shown in human tissue culture cells and fission yeast that Cap-H binds to the N-terminal tail of histone 2A and the variant histone 2A.Z. In vitro studies have revealed that this binding can occur independent of other condensin subunits [37]. While these results are consistent with chromatin targeting of condensin via Cap-H in these systems, our findings suggest that in Drosophila, Cap-G may direct chromatin targeting of condensin. The target molecule on chromatin, which is recognized by Drosophila Cap-G, remains to be identified.
While our study is the first report on the dynamics of SMC2 localization in Drosophila during the cell cycle, our data on Cap-D2 appear to be at odds with studies on fixed S2 tissue culture cells using anti-Cap-D2-antibodies [18]. In this study, Cap-D2 was reported to be primarily nuclear. This discrepancy can be explained by the different tissues analyzed. Nuclear import may be slow for Cap-D2, as, in fact, Savvidou et al. [18] observe increasing nuclear concentration of Cap-D2 when the cells progress through G1-S-G2. During the rapid syncytial divisions, nuclear import of Cap-D2 may not be efficient. Analysis of other tissues of EGFP-Cap-D2 expressing animals indeed showed nuclear localization, for example in ovarian follicle cells (data not shown). Interestingly, nuclear localization of Cap-H2 has also been described to progressively increase in more advanced ovarian nurse cell nuclei when compared with nuclei at younger stages [24], own unpublished observation).
Condensin complexes have been initially identified and characterized in the biochemically tractable Xenopus egg extract system [4]. In mitotic extracts, soluble 13S heteropentameric holocomplexes as well as 8S SMC2/SMC4 dimers were readily detected. Besides this initial identification of the complex later termed condensin I, condensin II was also detected in high-speed supernatants of Xenopus egg extracts [12], as well as in HeLa cell lysates [12], [38]. Quantification revealed that in the Xenopus egg extract system condensin I is present in roughly five-fold excess over condensin II while in HeLa cells both complexes occur in approximately equimolar amounts [12]. These differences in abundance are paralleled by a different appearance of condensed chromosomes. While in HeLa cells, metaphase chromosomes appear short and thick, the condensed chromosomes in the Xenopus egg extract system are rather long and thin. Intriguingly, experimentally shifting the ratio of condensin I∶condensin II in Xenopus egg extracts from ∼5∶1 to ∼1∶1 resulted in shorter and thicker chromosomes [31]. As metaphase chromosomes in Drosophila are also short and thick, one would expect a roughly balanced abundance of the two condensin complexes, if condensin I and II play comparable roles in the fly. As we did not detect any soluble endogenous condensin II complexes in our immunoprecipitation analyses, this is apparently not the case. We have analyzed extracts from ovaries and embryos. Cap-H2 mutants display a phenotype in ovarian nurse cell nuclei suggesting that Cap-H2 is expressed at this stage [24]. Also, the temporal expression data provided by the modENCODE project reveal expression of both Cap-H2 and Cap-D3 in ovaries and in early embryos, albeit at only low to moderate levels [39]. In fact, these levels are significantly lower than those reported for condensin I-specific subunits in most tissues indicating that condensin II-like complexes must be of low abundance. Our analysis of ovarian extracts derived from females overexpressing Cap-H2-fusion proteins circumvented the issue of low endogenous expression levels. Indeed, in these experiments, SMC2 and SMC4 were found to be associated with overexpressed Cap-H2, but peptide intensities and unique peptide numbers were significantly lower than in the experiment, in which proteins in association with Cap-G-mRFP1 in ovaries were assessed. Also, as our in vitro interaction assays revealed only weak affinities of Cap-H2 towards Cap-D3 and SMC4 in solution, a condensin II-like holocomplex in Drosophila may be functionally assembled in an efficient manner only on chromatin, unlike the situation found in vertebrates. Published studies on the phenotypic consequences of the loss of Cap-D3 or Cap-H2 have shown that these phenotypes can be modified by mutations in other condensin subunit genes (namely Cap-H2, Cap-D3 and SMC4), thus revealing genetic interactions [23], [24], [40]. However, it remains to be shown, whether these genetic interactions are based on a physical interaction of these subunits bound to the chromatin. Furthermore, such a chromatin-associated condensin II-like holocomplex is unlikely to play a mitotic role, given the absence of mitotic phenotypes in Cap-H2 and Cap-D3 mutants [18], [23], [24], which is also consistent with the failure of EGFP-Cap-H2 to load onto mitotic chromatin (data not shown).
Cap-G was not found in association with overexpressed Cap-H2, even though Cap-G would be expected to bind to the kleisin component if it was part of a condensin II-like complex [35]. The direct binding assays of in vitro translated proteins also did not produce any indication of an association of Cap-G with Cap-H2, rendering the proposal of the participation of Cap-G in a condensin II-like complex highly unlikely. So the question remains whether a second HEAT-repeat containing protein besides Cap-D3 is part of a putative condensin II complex in Drosophila. BLAST analyses do not produce Cap-G2 homologs encoded in the D. melanogaster genome or in any of the sequenced genomes of dipterans. It is possible that a Cap-G2 homolog does exist in Drosophila, but has escaped detection using the BLAST algorithms because it might have diverged significantly during evolution. Therefore, we have scrutinized the list of proteins identified in the Cap-H2 immunoprecipitates for possible Cap-G2 candidates by the virtue of a size above 100 kDa, and an extended stretch of predicted HEAT repeats in the N-terminal region, but with dissimilarity to importins/exportins which also have blocks of HEAT repeats in their N-termini. However, none of the proteins contained in the list of immunoprecipitated proteins qualifies as a Cap-G2 homolog based on these criteria (data not shown). Thus, the possibility remains that condensin II has diverged in dipterans to function as a mainly chromatin-bound heterotetrameric complex lacking a Cap-G2 subunit. Moreover, in combination with the facts that Cap-H2 and Cap-D3 loss-of-function mutants have no obvious mitotic phenotype [18], [23] and that these two subunits have been shown to participate in such diverse processes as transvection, the regulation of AMP-expression or chromosome territory formation [23]–[25], [40], our results support a model in which a Drosophila condensin II-like complex has functionally specialized beyond regulation of chromatin structure during nuclear divisions.
Fly stocks were obtained from the Bloomington Drosophila Stock Center at Indiana University, unless indicated otherwise. Expression constructs for condensin subunits were generated by cloning genomic fragments isolated from bacterial artificial chromosomes (BACs) obtained from CHORI BacPac Resources into appropriate vectors, or cDNAs obtained from the Drosophila Genomic Resource Center (DGRC) into the vectors pUAST or pUASP1 [14], [41]. Appropriate restriction sites for cloning were introduced by PCR with primers containing the recognition sequences for the respective enzymes. The integrity of coding regions amplified by PCR was verified by subsequent DNA sequence analysis. Transgenic flies were generated by using established germ line transformation protocols for microinjection into w1 embryos (pUAST, pUASP1 and pBac-constructs) or into embryos expressing the PhiC31 integrase and containing an attP landing site at specific genomic sites [42].
For the construction of fly stocks expressing an EGFP-fused variant of SMC2, a 5.2 kb fragment containing SMC2 including its flanking genomic regions was amplified from the BAC clone CH321-59P12 as template and cloned into the pattB vector [42]. A 1370 bp internal PstI/MluI SMC2-fragment was subcloned into the pSLfa1180fa vector [43] and fused with the EGFP-coding sequence using a BspEI site introduced by inverse PCR. The EGFP tag was fused internally between amino acid residues G582 and S583 of SMC2 within the hinge region (SMC2h-EGFP). Internal fusions within the hinge region of yeast SMC1 and SMC3 have been shown to be functionally tolerated [44]. The modified fragment was cloned back into the pattB-SMC2 vector. Transgenic flies were generated via injection of the pattB-SMC2h-EGFP plasmid into y1, w1, M[vas-int]ZH2A; M[3x3P-RFP,attP′]ZH96E embryos [42].
For the construction of fly stocks expressing an EGFP-fused variant of Cap-D2 under control of the genomic regulatory sequences, a 6.8 kb genomic fragment encompassing Cap-D2 and 600 bp upstream of the transcriptional start site as well as 1,600 bp downstream of the poly(A) site was cloned via recombineering [45] into pattB using the BAC CH321-26K05 as sequence source. A 1.5 kb NotI/Acc65I fragment of the 5′-terminal Cap-D2 region was isolated from pattB-Cap-D2 and subcloned into the pBluescriptSK vector (Stratagene). The naturally occurring NcoI site at the Cap-D2 translational initiation codon was used to insert a PCR-amplified fragment encoding EGFP, flanked by PciI sites, which are compatible with NcoI. The 2.2 kb EGFP-fused NotI/Acc65I 5′-terminal Cap-D2 fragment was cloned back into the NotI/Acc65I cleaved pattB-Cap-D2. Transgenic flies were generated via injection of the pattB-EGFP-Cap-D2 plasmid into y1, w1, M[vas-int]ZH2A; M[3x3P-RFP,attP′]ZH22A embryos [42].
For the construction of fly stocks expressing EGFP- and mRFP1-fused variants of Cap-G under control of the genomic regulatory sequences, a 1.2 kb XhoI fragment encompassing the 3′-terminal region of the Cap-G reading frame and downstream regulatory sequences was cloned from a genomic Cap-G pBac rescue construct [14] into the vector pLitmus 28 (New England Biolabs). After introduction of a NotI restriction site immediately upstream of the translational stop codon by inverse PCR, PCR-amplified fragments encoding either EGFP or mRFP1 flanked by NotI sites were cloned into this newly generated site. The modified 1.9 kb XhoI fragments were excised from the pLitmus 28 constructs and cloned back into the pBac Cap-G rescue constructs. Transgenic flies were generated via injection of the pBac-Cap-G-mRFP1 and pBac-Cap-G-EGFP plasmids into w1 embryos using established procedures [43]. The genomic region encoding Cap-G-EGFP was also cloned into the pattB vector and transgenic lines were established after injection into y1, w1, M[vas-int]ZH2A; M[3x3P-RFP,attP′]ZH96E embryos.
For the construction of pUAST-Cap-G-EGFP vectors containing various Cap-G fragments, the corresponding Cap-G coding regions were PCR-amplified from the cDNA clone SD10043 and cloned into pUAST-MCS-EGFP [46]. Fragments encoding the following Cap-G-variants were amplified: Cap-GFL (full length, aa 1–1351); Cap-GNM (aa 1- 977); Cap-GNM1 (aa 1–848); Cap-GNM4 (aa 243- 977); Cap-GC (aa 958–1351). For the construction of pUASP1-Cap-GNM-EGFP, the Cap-GNM-EGFP-fragment was transferred from pUAST-Cap-GNM-EGFP into pUASP1 [14]. The constructs were used for P-element-mediated germ line transformation by injection into w1 embryos following established procedures. For all experiments, the following established lines were used: UAST-Cap-GFL-EGFP II.2, UAST-Cap-GFL-EGFP III.2, UAST-Cap-GNM-EGFP III.2, UAST-Cap-GC-EGFP II.3, UAST- Cap-GC -EGFP III.2, UAST-Cap-GNM1-EGFP II.1, UAST-Cap-GNM4-EGFP II.1, UASP1-Cap-GNM-EGFP III.4, UASP1-Cap-GNM III.2. Cap-GNM –EGFP and Cap-GNM were also cloned into the pattB vector containing the flanking Cap-G genomic regulatory elements ensuring expression at physiological levels. Transgenic lines were established after injection into y1, w1, M[vas-int]ZH2A; M[3x3P-RFP,attP′]ZH96E embryos.
For the construction of pUASP1-EGFP-Cap-H2 and pUASP1-mCherry-Cap-H2, the Cap-H2 coding region (based on the Cap-H2-RE annotation) was isolated using NcoI/XhoI from the cDNA clone SD18322 and subcloned into pLitmus28. The resulting plasmid pLitmus28-Cap-H2 was cleaved with AvrII/NcoI and PCR-fragments encoding mCherry and EGFP were inserted as AvrII/PagI fragments. The EGFP-Cap-H2 and mCherry-Cap-H2 cassettes were finally transferred as SpeI/Asp718-fragments into pUASP1 to generate pUASP1-EGFP-Cap-H2 and pUASP1-mCherry-Cap-H2, respectively, which were used for P-element-mediated germ line transformation. For all experiments, the transgene insertions UASP1-EGFP-Cap-H2 II.4 and UASP1-mCherry-Cap-H2 II.1 were used.
For expression of UAS-transgenes, we used da-GAL4 G32 [47], F4-GAL4 [48], maternal α4tub-GAL4-VP16 [49] and tubP-GAL4.
Rescue experiments were performed using trans-heterozygous mutant allele combinations of the respective genes, simultaneously expressing our transgenes either under control of the flanking genomic regulatory regions or under UAS-control driven by the ubiquitous active GAL4-driver da-GAL4 G32 or by α4tub-GAL4-VP16 in the case of Cap-H2. The following alleles were used: Cap-G1 and Cap-G6 [14], Cap-D2f03381, Cap-D2 Df(3R)01215, SMC2jsl2, SMC2f06842, SMC2Df(2R)BSC429, Cap-H2Df(3R)Exel6159 , Cap-H2EY09979 and Cap-H2TH2 [24]. For Cap-G, Cap-D2 and SMC2, complementation of the lethality associated with the trans-heterozygous mutant situation was assessed. For Cap-G, rescued trans-heterozygous individuals could be readily identified by the recessive markers al, b, c and sp present on the Cap-G1 and Cap-G6 chromosomes [14]. For Cap-H2, suppression of the delayed dispersal of nurse cell chromatin observed in Cap-H2 mutant ovarioles [24] was monitored upon transgene expression. Furthermore, the phenotype upon overexpression of EGFP-Cap-H2 and mCherry-Cap-H2 in larval salivary glands was compared with the phenotype obtained after the GAL4 dependent Cap-H2 overexpression using the allele Cap-H2EY09979, which is an UAS containing P-element inserted upstream of Cap-H2.
To drive expression of Cap-GNM1-EGFP or Cap-GNM4-EGFP together with His2Av-mRFP1, individuals of the corresponding UAS-lines were crossed with w*, α-tub-GAL4-VP16, gHis2Av-mRFP1 II.2 flies (generously provided by C. Lehner, University of Zurich).
To express HP1-mRFP1 together with Cap-GFL-EGFP or Cap-GC-EGFP, we generated recombinant chromosomes containing either UAST-Cap-GFL-EGFP II.2 or UAST-Cap-GC-EGFP II.3 together with gmRFP1-HP1 II.1 [50] using standard genetic techniques.
To co-express Cap-GFL-EGFP with Cid-mRFP1, both under control of the flanking genomic sequences, lines were generated by classical genetic techniques containing the gCap-GFL-EGFP III.1 and gCid-mRFPII.1 [28] transgenes.
For chromatin loading analyses, chromosomes carrying a transgene allowing expression of His2Av fused with mRFP1 [51] were combined with gCap-G-EGFP III.1, or gSMC2h-EGFPΦX-96E or gCap-D2-EGFPΦX-22A.
Antibodies against the human c-myc epitope [52], Drosophila Cap-H/Barren [22] and Drosophila Cap-D2 [18] have been described previously.
Rabbit-anti-Flag (Sigma), mouse-anti α-Tubulin (Sigma) as well as secondary antibodies (Jackson laboratories) were obtained commercially. Antibodies against EGFP and mRFP1 were raised in rabbits using bacterially expressed full length proteins as antigen. The anti-mRFP1 antibodies also recognize and precipitate mCherry-fused proteins. Mouse monoclonal antibodies against EGFP were purchased from Roche Biochemicals or were a gift from D. van Essen and S. Saccani (MPI Freiburg, Germany).
Antibodies against Cap-G and SMC2 were raised in rabbits using bacterially expressed N-terminal protein fragments of Cap-G (aa 1- 553) and SMC2 (aa 1–313), respectively. The antisera were affinity purified using standard procedures [53]. For immunoblotting, the antibodies were used at a 1∶3000 dilution. A mouse monoclonal antibody directed against HP1 was obtained from the Developmental Studies Hybridoma Bank (clone C1A9; dilution 1∶1000 for immunoblotting).
For in vivo microscopy, embryos at the desired developmental stage were collected and processed as previously described [54]. Single-stack confocal images were acquired every 18 or 20 sec using a Leica SP5 confocal microscope (Leica Microsystems, Germany), equipped with a 63× oil-immersion objective, a 458–514 nm Ar laser and a 561 nm DPSS laser for the excitation of EGFP and mRFP1, respectively. For fixed samples stained with Hoechst 33258, a 405 nm UV- diode laser was used in addition, and confocal images were acquired with a 40× oil-immersion objective.
Images were processed using ImageJ 1.46 (National Institute of Health, USA) and Adobe Photoshop CS4 (Adobe Systems Inc.). In some images, shot noise was decreased with a Gaussian filter.
Quantitative fluorescence measurements to determine chromatin association of the EGFP-fused condensin subunits was done as described in [28] with the exception that a Leica SP5 confocal system was used for analysis of EGFP-Cap-D2 and SMC2h-EGFP. The analyzed genotypes were gCap-G-EGFP III.1, gHis2Av-mRFP1 III.1/TM3, Ser or gHis2Av-mRFP1 II.2; gSMC2h-EGFPΦX-96E or gCap-D2-EGFPΦX-22A; gHis2Av-mRFP1 III.1. or SMC2f06842/SMC2Df(2R)BSC429; gSMC2h-EGFPΦX-96E .To quantify Cap-G-EGFP in centromeric regions, embryos co-expressing Cap-G-EGFP and Cid-mRFP were analyzed by laser scanning time lapse microscopy while progressing through the syncytial cycle 12. Small circular regions of interest (R.O.I.s) were defined in the channel for Cid-mRFP fluorescence, one encircling a centromere (cen-proximal) and one of the same size in a region within the nucleus but not encircling a centromere. The identical R.O.I.s were applied to the channel for Cap-G-EGFP fluorescence and the ratio of the cen-proximal fluorescence intensity:cen-distal fluorescence intensity was calculated. For each time point, 62 pairs of R.O.I.s from three different embryos were evaluated.
For immunoblotting experiments, ovaries of 4–5 days old females were dissected in 1× PBS and homogenized in sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) sample buffer. Protein samples corresponding to 5 ovaries were loaded on Tris-glycine based polyacrylamide gels and blotted onto nitrocellulose membranes.
For the immunoprecipitation experiments, 5–8 h old embryos expressing fluorescently tagged Cap-G variants were collected on apple-juice agar plates and dechorionized. Alternatively, we dissected ovaries in 1×PBS from females expressing epitope-tagged condensin subunits. These tissues (150 µl embryos or 300 ovaries) were homogenized in 4 volumes of lysis buffer (50 mM HEPES at pH 7.5, 60 mM NaCl, 3 mM MgCl2, 1 mM CaCl2, 0.2% Triton X-100, 0.2% Nonidet NP-40, 10% glycerol) including protease inhibitors (2 mM Pefabloc, 2 mM Benzamidin, 10 µg/ml Aprotinin, 2 µg/ml Pepstatin, A, 10 µg/ml Leupeptin). In the experiment shown in Figure S6, aliquots of the raw extracts were treated with a mixture of DNaseI and nuclease S7 for 45 min at 4°C to solubilize chromatin. The extracts were cleared by centrifugation (20 min, 14000×g; 4°C) and 200–400 µl of the supernatants were used for immunoprecipitation with anti-EGFP-, anti-mRFP1-, or anti-SMC2-antibodies bound and covalently cross-linked using dimethyl pimelimidate to Protein A-Sepharose (Affi-Prep Protein A, BIORAD; 25 µg of affinity purified antibodies bound to 30 µl of Protein A-Sepharose slurry). In the experiment shown in Figure S6, mouse monoclonal antibodies (Roche) were coupled to Protein G-Sepharose (GE Healthcare). After 3–4 h incubation at 4°C with gentle agitation, the Sepharose was washed for four times with 1 ml of lysis buffer. Bound polypeptides were eluted by incubation with 40 µl of elution buffer (50 mM Tris/HCl at pH 6.8; 2% (w/v) SDS) for 10 min at 37°C and/or by addition of 40 µl SDS-PAGE sample buffer and subsequent incubation at 95°C for 5 min (“hot elution”).
The immunoprecipitates were subjected to SDS–PAGE followed by silver staining (“PageSilver Silver Staining Kit”, Fermentas) or by western blot analysis.
For mass spectrometric analysis, immunoprecipitates were separated by SDS-PAGE on precast gradient gels (Serva, Heidelberg) and the proteins were visualized by staining with colloidal Coomassie Blue according to [55]. Entire gel lanes containing immunoprecipitates were cut into slices. Proteins were extracted from the gel pieces, digested with trypsin, separated via on-line nanoLC and analyzed by electrospray tandem mass spectrometry at an LTQ Orbitrap mass spectrometer. The complete lists with the identified proteins are available in the supplementary information.
DNA fragments encoding different regions of the condensin subunits were amplified by PCR and inserted into the vectors pCS2(F/A), pCS2(F/A)-HFHF (allowing a fusion of a C-terminal His6 Flag His6 Flag epitope tag), and/or pCS2-myc6(F/A) (allowing a fusion of an N-terminal myc6-epitope tag) , which all contain FseI/AscI-restriction sites within their MCS. Condensin coding regions were amplified from the cDNA clones SD10043 (Cap-G), LD40412 (Cap-D2), RE48802 (Cap-H/Barren), SD18322 (Cap-H2, based on the Cap-H2-RE annotation) and RE74832 (Cap-D3, based on the Cap-D3-RA annotation).
To generate pCS2-Cap-G-EGFP, the Cap-G-EGFP fragment was transferred from UAST-Cap-GFL-EGFP into pCS2(F/A). To generate pCS2-SMC4, the corresponding coding region was amplified using first strand cDNA derived from reverse transcription of mRNA extracted from w1-embryos, using the “RNeasy Mini Kit” and the “Omniscript RT Kit” (Qiagen), and inserted into pCS2(F/A).
For controls, the plasmids pCS2-hSecurin-HFHF and pCS2-myc6-hPP2A(C) (generously provided by O. Stemmann) were used, which contain the coding DNA sequences for human securin and the catalytic subunit of the human protein phosphatase 2A, respectively.
Coupled in vitro transcription/translation reactions (IVT) were performed using the “TNT SP6 Coupled Reticulocyte Lysate System” or the “TNT SP6 Quick Coupled Transcription/Translation System” (Promega) according to the manufacturer's instructions. Up to 3 different plasmids (final amount of 2 µg DNA total) were included in 25 µl reaction mixtures. For radioactive labeling, 0.4 µM [35S]methionine (1000 Ci/mmol) was added to the reaction mix. In some instances, the produced proteins migrated at almost the same position during SDS-PAGE. In these cases, only the components without an epitope tag were translated in the presence of [35S]methionine. The epitope tagged variants were translated in a separate reaction in the absence of radioactive label. Afterwards, the reactions were mixed and subjected to immunoprecipitation using 5 µl of mouse-anti-Flag-Agarose-slurry (Sigma, A1080) or 5 µl Protein-A-Sepharose beads to which monoclonal mouse antibodies against the myc-epitope had been covalently crosslinked with dimethyl pimelimidate [53]. After 3 h incubation at 4°C with gentle agitation and a subsequent brief centrifugation, the supernatants were removed and immunoprecipitates were washed 3 times with 1 ml of lysis buffer. Bound polypeptides were eluted by addition of 40 µl SDS-PAGE sample buffer and subsequent incubation at 95°C for 5 min. Precipitated polypeptides as well as samples derived from the input and supernatant fractions were resolved by SDS-PAGE and analyzed by immunoblotting and/or autoradiography (FLA 7000 Phosphoimager, Fuji Corp.)
|
10.1371/journal.ppat.1000930 | Requirement of NOX2 and Reactive Oxygen Species for Efficient RIG-I-Mediated Antiviral Response through Regulation of MAVS Expression | The innate immune response is essential to the host defense against viruses, through restriction of virus replication and coordination of the adaptive immune response. Induction of antiviral genes is a tightly regulated process initiated mainly through sensing of invading virus nucleic acids in the cytoplasm by RIG-I like helicases, RIG-I or Mda5, which transmit the signal through a common mitochondria-associated adaptor, MAVS. Although major breakthroughs have recently been made, much remains unknown about the mechanisms that translate virus recognition into antiviral genes expression. Beside the reputed detrimental role, reactive oxygen species (ROS) act as modulators of cellular signaling and gene regulation. NADPH oxidase (NOX) enzymes are a main source of deliberate cellular ROS production. Here, we found that NOX2 and ROS are required for the host cell to trigger an efficient RIG-I-mediated IRF-3 activation and downstream antiviral IFNβ and IFIT1 gene expression. Additionally, we provide evidence that NOX2 is critical for the expression of the central mitochondria-associated adaptor MAVS. Taken together these data reveal a new facet to the regulation of the innate host defense against viruses through the identification of an unrecognized role of NOX2 and ROS.
| The understanding of the mechanisms allowing the host to mount a rapid and efficient innate immune response to RNA viruses has been the subject of intensive research in recent years. Major groundwork allowed the identification of key sensors of virus nucleic acids, including RIG-I and Mda5, which through association with the MAVS adaptor initiate the signaling cascade required for activation of the IRF-3 transcription factor and downstream antiviral genes. Mechanisms of activation and degradation of key signaling molecules allow a tight control of the intensity and duration of the response. Our knowledge of how redox processes regulate signaling cascades is a fast moving field of research. Particularly, the identification of non-phagocytic reactive oxygen species-producing NADPH oxidase (NOX) enzymes revealed new insights into their function in innate immunity. Our endeavor in characterizing the role of NOX in the antiviral response reveals a new facet to the overall picture of antiviral response regulation. Here, we demonstrate that NOX2 is essential for MAVS expression in airway epithelial cells, thereby controlling the capacity of the cell to mount an efficient innate antiviral response following recognition of viruses.
| The capacity of the host to rapidly respond to virus infection is essential to establish an antiviral state that restricts virus replication and spreading, and to permit the production of proinflammatory chemokines and cytokines that attract and activate immune cells to the site of infection. Although major breakthroughs have recently been made, much remains unknown in our understanding of the molecular mechanisms involved in virus recognition and how it is transmitted via signaling messengers to the expression of antiviral and proinflammatory genes.
Initiation of these innate immune responses is achieved through recognition of invading viruses by pattern recognition receptors (PRR) that specifically recognize pathogen-associated molecular patterns (PAMPs). Virus-derived nucleic acids are considered major PAMPs that activate various PRRs, including members of the membrane-bound Toll-like receptors (TLRs) family, TLR-3, -7 and -9, and the recently identified cytoplasmic RIG-I-like receptors (RLRs), including RIG-I and Mda5 [1]. Following recognition of viral RNAs, RIG-I/Mda5 elicit signaling cascades via a caspase recruitment domain (CARD)-mediated interaction with the mitochondria-associated adaptor MAVS, also known as CARDIF/IPS-1/VISA [1], which in turn interacts with the TNFR-associated death domain (TRADD) protein [2]. At this level, the signaling cascades diverge due to specific interactions either with the FADD and RIP1 adaptors or with the E3 ubiquitin ligase TRAF3 and the adaptor protein TANK, to elicit activation of the NF-κB and IRF-3 transcription factors, respectively [2].
Activation of the ubiquitously expressed IRF-3 transcription factor is central to the development of an antiviral state, mainly through the rapid and robust expression of type I Interferons (IFNs) genes, a prerequisite for the induction of numerous antiviral proteins that modulate protein synthesis, growth arrest and apoptosis [3]. Moreover, IRF-3 also has the capacity to directly regulate a subset of these antiviral genes, including IFIT1, which encodes for the ISG56 translation regulator, thereby establishing an early IFN-independent antiviral response [4]. Virus-induced IRF-3 activation relies on a complex set of phosphorylation events mediated at least by the IκB-kinase (IKK)-related kinases, TANK binding kinase-1 (TBK1) and IKKε, that regulates its dimerization, nuclear accumulation and transactivation capacities [5], [6].
Reactive oxygen species (ROS), such as hydrogen peroxide and superoxide anion, are now well appreciated to act as cellular switches for signaling cascades leading to gene regulation involved in physiological processes, including cell proliferation, apoptosis and immune and proinflammatory responses [7], [8]. Amongst the enzymatic systems that produce ROS, the family of NADPH oxidase/Dual oxidase (NOX/DUOX) enzymes is now considered predominant in various cell types. Seven members of this family have been identified, named NOX1-5 and DUOX1-2, each with tissue- and cell-type specific expression patterns [9]. NOX2, which is mainly expressed in, but not restricted to, neutrophils and macrophages, is well known to play pivotal roles in host defense against bacterial and fungal pathogens, through production of superoxide in the phagosome [10]. Interestingly, recent functional data have emerged that suggest the involvement of NOX/DUOX members in the innate host defense to invading microorganisms in non-phagocytic cells [11], [12]. Particularly, a crucial role of NOX1 and NOX4 in the regulation of TLR-mediated intracellular signaling via MAPK and NF-κB has previously been highlighted [11], [13]. More recently, NOX2 interaction with TLR2 was shown to be required for efficient innate immune responses to Mycobacteria [14]. Additionally, we recently reported for the first time that NOX2 plays an essential role during Paramyxoviridae virus infections through the regulation of the NF-κB-mediated proinflammatory response in airway epithelial cells (AEC) [15]. Here, we add a new facet to the regulation of the antiviral response. Our data demonstrate that NOX2 and ROS are critical for the ability of the host cell to trigger an efficient RIG-I-mediated IRF-3 activation and downstream antiviral genes through regulation of MAVS expression.
To start evaluating the potential implication of NOX-derived superoxide in IRF-3-mediated antiviral responses, the effect of antioxidants and pharmacological inhibitors on Sendai virus (SeV)-induced IFNβ- and ISG56-promoter activities was investigated in A549 cells. As shown in Figure 1A, SeV-induced IFNβ-promoter activity was significantly reduced in the presence of Tempol, a cell-permeable superoxide dismutase mimetic. Consistent with an implication of superoxide-dependent IRF-3 regulation, Tempol also inhibited the activity of the ISRE-containing ISG56-promoter in response to SeV (Figure 1A). Further analyses revealed that pretreatment with diphenyleneiodonium (DPI) or apocynin (Apo), two inhibitors classically used to target NADPH oxidase activities, also effectively inhibited SeV-induced IFNβ- (Figure 1B) and ISG56-promoter activities (Figure 1C and D) in a dose-dependent manner, while the same inhibitors did not similarly impact the activity of the unrelated pEF1 promoter (Figure S1 and Text S1).
To provide further evidence on whether NADPH oxidase-derived ROS are required for the regulation of IRF-3-dependent antiviral genes, the expression profile of IFNβ and IFIT1 (encoding for ISG56) genes was monitored in vivo by real-time PCR during SeV infection in the absence or presence of DPI. While IFNβ and ISG56 mRNA levels were markedly induced following SeV infection, the mRNA levels exhibited 1.8 log and 1.6 log reduction, respectively, in the presence of DPI (Figure 1E). Altogether these results suggest that efficient IFNβ and IFIT1 genes expression in response to SeV infection requires the production of ROS produced by a NADPH oxidase.
To determine whether the observed antioxidant-mediated inhibition of SeV-induced IFNβ and IFIT1 antiviral genes expression could be linked to an effect on NOX2, the effect of interference with NOX2 expression on their respective promoter activities was assessed. Immunoblot for NOX2 (Figure 2A) confirmed that NOX2 specific RNAi oligonucleotides significantly decreased NOX2 protein expression in A549, as previously validated and confirmed at the mRNA level [15], without affecting cellular viability (Figure S2 and Text S1). Reduction of NOX2 expression effectively altered the gene transactivation capacity of endogenous IRF-3. First, the stimulation of IFNß- and ISG56-promoter activities by SeV were dramatically decreased in NOX2-depleted cells compared to cells transfected with CTRL RNAi (Figure 2B). Second, ectopic expression of NOX2 significantly enhanced SeV-induced activation of the IFNβ promoter (Figure 2C). Additionally, to further establish the role of NOX2 in vivo, the expression of endogenous IFNβ was analyzed by real-time PCR following SeV infection of CTRL- and NOX2-RNAi-transfected cells. SeV-induced IFNβ mRNA level was significantly reduced by 53% in the absence of NOX2 as compared to control cells, while SeV replication quantified by real-time PCR analysis of the nucleocapside (N) RNA revealed a 1.6 fold increase (Figure 2D). Taken together, these results demonstrate that NOX2 contributes to the regulation of SeV-induced IFNβ- and ISG56-promoter activities, thereby suggesting that NOX2 is an essential component of the signaling pathway triggering IRF-3 activation following virus infection.
In uninfected cells, IRF-3 is predominantly present as two forms corresponding to an unphosphorylated (form I) and a N-terminus hypophosphorylated form (form II) [16]. The C-terminus of IRF-3 contains three clusters of virus-induced phosphoacceptor sites, Ser 385/386 (Cluster I), Ser 396/398 (Cluster II) and Ser 402/405 and Thr 404 (Cluster III) that are detected as two dimeric active forms (form III and IV) in SDS-PAGE [17], [18]. Thus, to provide further evidence that NOX2 and ROS are essential for IRF-3 activation, the effect of Tempol or NOX2 knock down through RNAi on IRF-3 phosphorylation and dimerization was analyzed. As shown in Figure 3A and B, SeV-induced phosphorylation of IRF-3 at Ser 396, and to a lesser extent at Ser 398, was significantly reduced in Tempol-treated compared to vehicle-treated cells. In the same line, formation of the active dimeric form of IRF-3, evaluated by native-PAGE, was also effectively impaired (Figure 3B). Phosphorylation at Ser 386 has previously been reported to correlate with IRF-3 in its dimeric form [18]. Accordingly, detection of Ser 386 was decreased in cells treated with Tempol (Figure 3B). Importantly, interference with NOX2 expression similarly inhibited SeV-induced IRF-3 Ser 396, 398 and 386 phosphorylation and dimerization (Figure 3C and D). Impairment of IRF-3 phosphorylation through interference with NOX2 was also confirmed in the context of SeV infection of primary cells, using normal human bronchial epithelial cells (NHBE) (Figure 3E). Altogether, these data demonstrate that NOX2 and ROS are essential for the efficient activation of IRF-3 during virus infection.
We and others have previously demonstrated that IKK-related kinases, IKKε and TBK1, are part of the kinase activity that is essential for IRF-3 phosphorylation and subsequent activation of IRF-3 in the context of virus infection [5], [6]. To start depicting the role of NOX2 in the upstream signaling pathways leading to IRF-3 activation, the role of NOX2 and ROS in SeV-induced TBK1/IKKε expression and activity was assessed. Initial analysis of kinases expression during SeV infection in the absence or presence of DPI or in CTRL RNAi vs NOX2 RNAi-transfected cells revealed that basal and SeV-induced expression of IKKε is dramatically decreased by DPI treatment or depletion of NOX2 expression (Figure 4A and C). Furthermore, quantification of SeV-induced TBK1 activity demonstrated that it is inhibited in a dose-dependent manner by DPI-treatment reaching around 52% inhibition at a concentration of 30 µM DPI compared to vehicle-treated cells (Figure 4B). Finally, in NOX2 RNAi-transfected cells, SeV-induced TBK1 activity was diminished by about 55% compared to the activity measured in CTRL RNAi-transfected cells (Figure 4D). These data provide important evidence for the requirement of NOX2 in the activation of IRF-3 kinases, TBK1 and IKKε, in response to SeV infection.
RLRs play unique and redundant roles in RNA virus recognition and appear to function in both cell- [19] and virus-specific manners [20], [21]. In order to further investigate the role of NOX2 in virus-mediated IRF-3 activation, we first thought to confirm, in our A549 model, the essential role of RIG-I in SeV recognition that was previously highlighted in embryonic fibroblasts (MEFs), lung fibroblasts, dendritic cells (DCs) and 293 cells [19], [20], [21], [22], [23]. RNAi specifically targeting RIG-I were used to determine the role of RIG-I in IRF-3 phosphorylation. As shown in Figure 5A, interference with RIG-I expression completely abolished IRF-3 Ser 396 phosphorylation following SeV infection, demonstrating that RIG-I is essential for downstream signaling to IRF-3 in the early time points following SeV infection in A549 cells.
Kato and collaborators recently reconciled previous conflicting results concerning the role of RIG-I and Mda5 in the recognition of poly I:C by showing that RIG-I and Mda5 selectively recognize short and long dsRNA, respectively [24]. Using lipid-mediated transfection of sheared poly I:C (see experimental procedures section for details on poly I:C preparation), we were able to specifically trigger IRF-3 Ser 396 phosphorylation by a RIG-I-dependent/Mda5-independent pathway (Figure 5B). Thus, activation of IRF-3 through the RIG-I-dependent pathway during SeV infection can be mimicked by poly I:C treatment in A549 cells.
In order to further evaluate the potential role of NOX2 in the RIG-I-dependent signaling pathway, stimulation of A549 cells with poly I:C was performed in the absence or presence of antioxidants or NADPH oxidase inhibitors. As shown in Figure 6A and B, induction of the ISG56- promoter activity was significantly reduced when poly I:C treatment was performed in the presence of Tempol, DPI or Apo. Similarly, inhibition of NOX2 expression by RNAi also resulted in a dramatic diminution of the capacity of poly I:C to stimulate the IFNβ- and ISG56- promoters (Figure 6C). To further establish the role of NOX2 in IRF-3-dependent antiviral genes expression in vivo, the expression profile of IFNβ and IFIT1 genes was analyzed by real-time PCR following poly I:C stimulation of CTRL and NOX2 RNAi-transfected cells. As illustrated in Figure 6D, poly I:C-induced IFNβ and ISG56 mRNA levels were significantly reduced in the absence of NOX2, as compared to control cells.
Finally, to confirm that the observed inhibition of the poly I:C-induced IFNβ and ISG56 mRNA levels was mediated by alteration of IRF-3 activation, WCE derived from A549 transfected with CTRL or NOX2 RNAi and further stimulated with poly I:C, were analyzed for IRF-3 phosphorylation. In cells where NOX2 expression is significantly downregulated by RNAi, poly I:C-induced IRF-3 phosphorylation at Ser396 was barely detectable, while it was significantly induced in CTRL RNAi transfected cells (Figure 7A and B). Consistently, poly I:C-induced IRF-3 dimerization was strongly impaired in NOX2 RNAi vs CTRL RNAi transfected cells (Figure 7C). Altogether these results provide strong evidence that NOX2 and ROS are essential for RIG-I-induced, IRF-3-mediated antiviral gene transcription.
Based on the observation that NOX2 downregulation and antioxidant treatments reduced the ability of cells to mount an efficient RIG-I mediated antiviral response through inhibition of IRF-3 phosphorylation, the regulation of signaling molecules upstream of IRF3 kinases was evaluated. Analysis of the expression of known upstream signaling molecules by immunoblot revealed that MAVS level is dramatically reduced in NOX2 vs CTRL RNAi-transfected A549 (Figure 8A) and NHBE (Figure 8B) cells. On the other hand, expression of RIG-I, of the ubiquitin ligase TRIM25 that is known to regulate RIG-I activity [25], and of the ubiquitin ligase TRAF3 that interacts with MAVS to trigger virus-induced IRF-3 activation were similar in both conditions. Moreover, expression of the TRAF6 ubiquitin ligase involved in MAVS-mediated NF-κB activation, which is also placed under NOX2 control [15], was also found to be equal in both conditions (Figure 8A). A specific reduction of MAVS protein level was also observed in Tempol-treated A549 cells compared to control cells (Figure 8A). Further analysis of MAVS expression by real-time PCR demonstrated that NOX2 downregulation by RNAi resulted in a 55% reduction of MAVS mRNA compared with control cells transfected with CTRL RNAi (Figure 8C). As previous reports highlighted a key role of MAVS localization in the mitochondria outer membrane to its function [26], and anticipating that NOX2 might regulate MAVS at different levels, the localization of the remaining amount of MAVS was also visualized by confocal microscopy. MAVS staining colocalized with mitochondrial marker in both control and NOX2-depleted A549 cells (Figure 8D), thus excluding an effect of NOX2 on MAVS function through modulation of its subcellular localization. Taken together these data demonstrate that in AEC, NOX2 and ROS promotes MAVS mRNA expression.
NOX enzymes are the main source of deliberate cellular ROS production in response to various stimuli. Numerous studies over the past years have increasingly clarified the function of NOX in various biological processes, including cell proliferation, apoptosis, proinflammatory response and host defense, notably through their role as cellular switches that regulate signal transduction pathways [7], [8]. The importance of NOX enzymes in the innate host defense is exemplified by the role of NOX2 in the generation of high amount of ROS, known as oxidative burst, in phagocytic cells as part of their armory of anti-bacterial mechanisms [27]. In this study, we reveal a novel essential function of NOX2-derived superoxide in the innate immune antiviral response triggered following recognition of invading virus by the RIG-I cytoplasmic sentinel. Our data demonstrate that efficient IRF-3 activation and downstream antiviral genes, IFIT1 and IFNβ, expression in response to SeV or RIG-I stimulation are impaired by NOX2 knock down, pharmacological inhibition of NOX or treatment with superoxide dismutase mimetic.
Few reports previously identified a role of NOX2 in other aspects of the cellular response to virus infections. NOX2 was recently shown to mediate HIV Tat-induced JNK activation and cytoskeletal rearrangement in HUVEC [28]. Moreover, Paramyxoviridae virus-induced NF-κB activation and downstream proinflammatory cytokines production in AEC were shown to be dependent on NOX2 [15]. Reduced inflammation of lung parenchyma has also been noticed during influenza infection in mice lacking NOX2 murine homolog [29]. However, this study does not distinguish between a role of NOX2 in phagocytes recruited to the lung parenchyma from a potential role of NOX2 in non-phagocytic cells. Earlier reports suggested NOX-dependent IRF-3 activation in response to Respiratory Syncitial Virus (RSV), but this conclusion was only based on an inhibitory effect of DPI or Apo treatment [30]. Based on the previous demonstration that RSV-induced IRF-3 activation is triggered by a RIG-I-dependent recognition mechanism [31] and our recent observation that NOX2 is involved in RSV-mediated NF-κB activation in AEC [15], the role of NOX2 in RIG-I-mediated activation of IRF-3, presented in this study, provides a likely mechanism for this yet unexplained redox-dependent activation of IRF-3 during RSV infection.
Beside RLH receptors, viral nucleic acids are also sensed by members of the Toll-like receptor (TLR) family, including TLR-3 that binds extracellular or endosomal dsRNA [1]. Exogenous H2O2 treatment was recently shown to enhance TLR-3 mediated NF-κB, but not IRF-3, activation [32]. This result suggests that redox-dependent regulation of IRF-3 following recognition of viral RNA is not a universal mechanism, but depends on the PRR engaged following virus infection. However, it is also important to consider that the use of exogenous H2O2 constitute a major difference with our study. Indeed, not only the subtype of ROS, but also the localization of the ROS signal at specific subcellular compartment, is considered essential for activating specific redox signaling events [33], [34]. Thus, the effect of exogenously added H2O2 most likely differs from the role of endogenous NOX2-dependent superoxide studied here. Interestingly, IRF3 activation following TLR-4 stimulation with LPS stimulation in U373/CD14 appears to be dependent on the activity of another member of the NOX family, NOX4 [35]. In this study, NOX4 appears to play a role in an ASK1/p38 axis that leads to IRF-3 nuclear accumulation. However, whether NOX4 is also involved in the TBK1/IKKε pathway that was shown to be involved in LPS-induced IRF-3 activation in macrophages [36] has not yet been investigated. In the present study, a role of NOX4 in the RIG-I-mediated IRF-3 activation was excluded based on the absence of detection of NOX4 expression in the A549 cell model used, as previously described [15].
Recently, a role of ROS in RLR signaling was documented in Atg5−/− MEF cells that are defective in autophagy process. ROS associated with the accumulation of dysfunctional mitochondria in these cells enhanced RLR-mediated cytokines production [37]. In the same study, rotenone treatment, which artificially induces accumulation of mitochondria-associated ROS, was sufficient to enhance RLR signaling [37]. However, it is not yet clear how this ROS-dependent mechanism is involved in RLR signaling regulation in normal cells exhibiting functional autophagy and if the effect of mitochondria-associated ROS is associated with increased IRF-3 activation. This production of ROS at a non-physiological level is considered harmful, and therefore might represent more a deleterious mechanism involved in virus pathogenicity than a physiological regulation of RLR signaling as illustrated by our results. Other studies also presented evidence of a cross-talk between NOX2 and mitochondria-associated ROS in the regulation of signaling cascades. In human umbilical vein endothelial cells and human alveolar macrophages, TNFα-induced NF-κB activation was shown to be dependent on both NOX2 and the mitochondrial respiratory chain activity [38], [39]. Thus, one may not exclude at this point that both the NOX2-dependent and a mitochondria-associated mechanism might cooperate in the regulation of RIG-I-mediated activation of the antiviral response.
Since NOX2 knock down alters phosphorylation of multiple phosphoacceptor sites, including Ser 386, Ser 396 and Ser 398 in the C-terminal clusters of IRF-3, it likely plays a role in several pathways converging to IRF-3 activation downstream of RIG-I. Our data demonstrate that NOX2 is required for TBK1 catalytic activation and IKKε expression. Recent studies strongly suggest that TBK1 and IKKε specifically phosphorylate Ser 402 and Ser 396 of IRF-3 [40], [41], [42], thereby implying that yet unidentified kinases that are responsible for the phosphorylation of other critical phosphoacceptor sites might also be controlled by NOX2. Recently, Protein kinase C-α [43], PI3 kinase [44], IKKα [45] and JNK [46], were shown to be involved in IRF-3 phosphorylation, but whether these kinases directly target IRF-3, and if so, on which specific phosphoacceptor sites, has not yet been established.
The observation that NOX2 knock down and antioxidant treatment abrogated RIG-I mediated IRF-3 phosphorylation raised the question about the identity of the molecular target(s) in the RIG-I-induced signaling cascade. Our data demonstrate that in AEC, NOX2 is essential for expression of the MAVS adaptor, which acts as a central platform to catalyze the formation of the mito-signalosome containing, among other signaling molecules, RIG-I and TRAFs involved in the antiviral cascade [47]. ROS and NADPH oxidases are known to regulate mRNA expression by several means, including regulation of redox-sensitive transcription factors, such as NF-κB and AP-1 [48], [49] and modulation of mRNA stability as reported elsewhere for TLR-4, IL-1 or p53 mRNA [26], [50], [51]. Interestingly, NOX2 was specifically shown to regulate cell cycle via induction of p21cip1 mRNA expression in endothelial cells following nutrient deprivation [52]. The transcriptional and post-transcriptional mechanisms involved in MAVS mRNA expression have not yet been elucidated. Further studies are required to identify these mechanisms and uncover how NOX2 is involved in their regulation.
Function of NOX2-derived ROS under basal conditions was previously documented in other non-phagocytic cells [53], [54], [55]. Particularly, an active NOX2-containing NADPH oxidase is well documented to contribute to endothelial cells proliferation [53], [54]. It is worth mentioning that during the course of our study, Takemura and Collaborators identified a role of NOX2 in the regulation of basal and EGF-induced ENac channel activity in alveolar epithelial cells [56]. Our data describe the essential role of NOX2 and ROS in basal MAVS mRNA and protein expression. However, during virus infection, MAVS is proteolysed by a proteasome-dependent PCBP2/AIP4 axis [57]. Thus, it is an interesting working hypothesis that NOX2-dependent mechanism might permit expression of MAVS during virus infection to regenerate the pool of available MAVS in the cell to mount a sustained antiviral response. Studies are underway to characterize in detail basal and virus-induced NOX2 activity. To date, NOX2 subcellular localization in non-phagocytic cells appears to vary from cell type to cell type, being either at the plasma membrane, in the endosomes responsible for early receptor-mediated signaling, known as “redoxisomes”, or in the perinuclear region [58]. An interesting aspect of these studies will be to determine the localization of NOX2 and ROS production involved in MAVS expression, as this is now considered an important aspect in the specificity of redox-dependent functions [59].
DPI, DMSO and BSA were purchased from Sigma-Aldrich. Apo was purchased from Calbiochem. Tempol was from Biomol International. Oligonucleotide primers were purchased from Invitrogen (Carlsbad, USA) or Alpha DNA (Montreal, Canada). RNAi oligonucleotides were from Dharmacon. Target sequences of the RNAi used against the different genes are as followed: CTRL, 5′cauagcguccuugatcaca3′; NOX2, 5′gaagacaacuggacaggaa3′; RIG-I, 5′aacgauuccaucacuauccau3′; Mda5, 5′ggugaaggagcagauucag3′.
The pRL-null reporter plasmid was obtained from Promega. ISG56-pGL3 and IFNβ-pGL3 luciferase reporter constructs and GST-IRF-3(aa387-427)-pGex-KG expression plasmid were previously described [4], [5]. Plasmids used to establish Real-Time PCR standard curves were generated by cloning of the PCR-amplified +208 to +706nt fragment of the IFNβ transcript (NM_002176), the +335 to +1319nt fragment of the IFIT1 transcript (NM_001548), the +748 to +980nt fragment of the β-actin transcript (NM_001101) and the +57 to +652nt fragment of the ribosomal S9 transcript (NM_001013) into the pCR2.1-TOPO using EcoRI. The pCS2-myc-NOX2 construct encoding myc-tagged human NOX2 was a kind gift from Dr. Shah, King's College London [60]. The pcDNA3.1-myc-MAVS construct was previously described [61].
A549 cells (ATCC) were cultured in F12 Nutrient Mixture (Ham) medium (Gibco) supplemented with 10% heat-inactivated Fetal Bovine Serum (HI-FBS, Gibco) and 1% L-Glutamine (Gibco). Normal human bronchial epithelial cells (NHBE) were obtained from Lonza, cultured in BEGM medium (Lonza) and used between passage 2 and 4. Sendai virus Cantell strain was obtained from Charles River Laboratories. Infection of subconfluent cells was performed at 10–80HAU/106 cells depending on the experiments for the indicated times. Where indicated, DPI, Apocynin or Tempol (or the corresponding vehicle) were added 1h before infection in serum free medium (SFM). Infection was pursued for 2h in SFM before addition of HI-FBS.
The synthetic analog of dsRNA, polyinosine-polycytidylic acid (poly I:C) (InvivoGen, San Diego CA) was resuspended at 10 mg/ml in sterile PBS and annealed by heating at 56°C for 30 min before cooling down at RT until it reached 20°C. Before use, poly I:C was diluted to 1mg/mL with ice-cold PBS and sheared using a 26G syringe. Cells were then transfected using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions.
RNAi oligonucleotides and plasmid transfections were performed using the oligofectamine reagent (Invitrogen) and the TransIT-LT1 Transfection Reagent (Mirus), respectively, and luciferase assays were performed using the Dual Luciferase reporter assay (Promega) as previously described in [15].
Whole cell extracts (WCE) were prepared on ice in Nonidet P-40 (Igepal, SIGMA) lysis buffer [15] and quantified using the Bio-Rad Protein Assay (Biorad, Hercules CA). 30µg were subjected to SDS-PAGE electrophoresis followed by immunoblot analysis using anti-IRF-3-phosphoSer396 [62], anti-IRF-3-phosphoSer398 [36], anti-IRF-3 (Active Motif), anti-TBK1 (Imgenex), anti-IKKε (eBioscience), anti-MAVS (Alexis Biochemicals), anti-RIG-I (Alexis Biochemicals), anti-TRAF3 (santa-cruz), anti-TRAF6 (santa-cruz), anti-TRIM25 (BD transduction laboratories), anti-actin (Chemicon International) and anti-SeV (obtained from Dr. J. Hiscott, McGill University, Montreal, Canada) antibodies diluted in PBS containing 0.5% Tween and either 5% nonfat dry milk or 5% BSA.
For NOX2 detection, A549 were scraped directly in 125mM Tris-HCl (pH 6.8), 10% glycerol, 2% SDS, 0.1 M DTT buffer containing 10µg/ml leupeptin, 20µg/ml aprotinin and 1µM pepstatin. Lysis was pursued at RT for 15 min before sonication. After 10 min at 70°C, samples were quantified using RC-DC protein quantification assay (BioRad). 150µg of lysate proteins were resolved by SDS-PAGE and analyzed by immunoblot with anti-gp91phox-Cter (obtained from Dr. Dagher and Dr. Brandolin, CEA-Grenoble, Grenoble, France) and anti-tubulin (Santa-Cruz) antibodies.
Immunoreactive bands were visualized by enhanced chemiluminescence using the Western Lightning Chemiluminescence Reagent Plus (Perkin Elmer Life Sciences). In between phosphospecific- and anti-IRF-3-antibodies, the membrane was stripped in 0.2% SDS, 62.5 mM Tris-HCl pH 6.8, 0.1 mM β-mercaptoethanol for 20 minutes at 50°C.
Native-PAGE was conducted as described previously [63] using 8µg WCE prepared as described above. Immunoblot detection of IRF-3 was performed using anti-IRF-3-phospho-386 (1/200, IBL) or anti-IRF-3 (Active Motif) antibodies.
In vitro kinase assays were conducted as described previously [42], using 80µg of WCE immunoprecipitated using 1µg of TBK1 antibodies (obtained from Dr. T. Maniatis, Harvard, USA) and 1µg of recombinant GST-IRF-3 (aa387-427) protein produced in BL21(DE3)plysS following IPTG stimulation as previously described [64]. After resolution by SDS-PAGE, IRF-3-GST was detected by coomassie blue staining of the lower part of the gel and radioactivity incorporation (32P) was quantified using a Typhoon Trio apparatus (Amersham Biosciences). The upper part of the gel was transferred to nitrocellulose membrane and TBK1 was detected using anti-TBK1 antibodies (Imgenex).
Total RNA was prepared using the RNAqueous-96 Isolation Kit (Ambion) following the manufacturer's instructions without the included DNase1 treatment step. Total RNA (1µg) was subjected to reverse transcription using the QuantiTect Reverse Transcription Kit (Qiagen), which includes a genomic DNA removal step. PCR amplifications were performed using the QuantiTect SYBR Green Kit (Qiagen) or the Fast start SYBR Green Kit (Roche) in the presence of 0.4µM of ISG56-, IFNβ-, β-actin- or –S9 specific primers. Absence of genomic DNA contamination was analyzed using a reaction without reverse transcriptase. Sequences of primer used are as follows: - ISG56, S: gcccagacttacctggacaa, AS: ggttttcagggtccacttca- IFNß, S: gaactttgacatccctgaggagattaagcagc, AS: gttccttaggatttccactctgactatggtcc- SeV N, S: agtatgggaggaccacagaatgg, AS: ccttcaccaacacaatccagacc- MAVS, S: ggtgccatccaaagtgcctacta, AS: cagcacgccaggcttactca- S9, S: cgtctcgaccaagagctga, AS: ggtccttctcatcaagcgtc- Actin, S: acaatgagctgctggtggct; AS: gatggccacagtgtgggtga. Detection was performed on a Rotor-Gene 3000 Real Time Thermal Cycler (Corbett Research). For ISG56, IFNβ, actin and S9, standard curves were obtained using amplification of serial dilutions of pCR2.1-TOPO-ISG56, -IFNβ, -β-actin and –S9 plasmids. ISG56 and IFNβ data represent absolute mRNA copy numbers normalized to β-actin or S9 used as a reference gene. For SeV N expression, standard curves and PCR efficiencies were obtained using serial dilutions of cDNA prepared from positive control infected cells and data are presented as relative fold expression versus uninfected sample after normalization to S9. Relative fold expression values were determined applying the ΔΔCt method [65].
A549 cells were grown and transfected with RNAi oligonucleotides on glass coverslips. At 48h post-transfection, mitochondria were visualized with 100nM Mitotracker Red CMX ROS (Invitrogen) applied to cells for 30 minutes at 37C. Cells were subsequently washed, fixed with 3.7% formaldehyde for 15 minutes at 37C and permeabilized with 0.2% triton X-100. Cells were then blocked in 10% goat serum before incubation for 3h with anti-MAVS (Alexis Biochemicals) diluted in PBS containing 3% BSA. After washing, cells were incubated for 1h with anti-rabbit Alexa488 secondary antibody (Invitrogen) diluted in PBS containing 3% BSA. Cells were then washed and mounted with ProLong Antifade reagent (Invitrogen). Single plane images were acquired with a Leica SP5 confocal microscope equipped with 63× (1.7 NA) oil objective with a digital zoom of 2×.
Data are presented as the mean ± standard error of the mean (SEM). Statistical significance for comparison of two means was assessed by an unpaired Student's t test. For dose-dependent experiments or multiple inhibitor studies, a one-way analysis of variance (ANOVA) test was used followed by a Dunnett post-test. Analyses were performed using the Prism 5 software (GraphPad). Statistical relevance was evaluated using the following p values: p<0.05 (*), p<0,01 (**) or p<0,001 (***).
Please see Table 1 for accession numbers.
|
10.1371/journal.ppat.1004910 | The EBNA-2 N-Terminal Transactivation Domain Folds into a Dimeric Structure Required for Target Gene Activation | Epstein-Barr virus (EBV) is a γ-herpesvirus that may cause infectious mononucleosis in young adults. In addition, epidemiological and molecular evidence links EBV to the pathogenesis of lymphoid and epithelial malignancies. EBV has the unique ability to transform resting B cells into permanently proliferating, latently infected lymphoblastoid cell lines. Epstein-Barr virus nuclear antigen 2 (EBNA-2) is a key regulator of viral and cellular gene expression for this transformation process. The N-terminal region of EBNA-2 comprising residues 1-58 appears to mediate multiple molecular functions including self-association and transactivation. However, it remains to be determined if the N-terminus of EBNA-2 directly provides these functions or if these activities merely depend on the dimerization involving the N-terminal domain. To address this issue, we determined the three-dimensional structure of the EBNA-2 N-terminal dimerization (END) domain by heteronuclear NMR-spectroscopy. The END domain monomer comprises a small fold of four β-strands and an α-helix which form a parallel dimer by interaction of two β-strands from each protomer. A structure-guided mutational analysis showed that hydrophobic residues in the dimer interface are required for self-association in vitro. Importantly, these interface mutants also displayed severely impaired self-association and transactivation in vivo. Moreover, mutations of solvent-exposed residues or deletion of the α-helix do not impair dimerization but strongly affect the functional activity, suggesting that the EBNA-2 dimer presents a surface that mediates functionally important intra- and/or intermolecular interactions. Our study shows that the END domain is a novel dimerization fold that is essential for functional activity. Since this specific fold is a unique feature of EBNA-2 it might provide a novel target for anti-viral therapeutics.
| Epstein-Barr virus is an oncogenic γ-herpesvirus that may cause infectious mononucleosis in young adults and fatal lymphoproliferative disorders in immunocompromised patients and is associated with the pathogenesis of Burkitt's lymphoma, nasopharyngeal and gastric carcinoma. Epstein-Barr virus nuclear antigen 2 (EBNA-2) is a key regulator of viral and cellular gene expression which initiates and maintains a specific transcription program that promotes proliferation and differentiation of the infected B cell. EBNA-2 is a transcriptional activator that is recruited to DNA by cellular adaptor proteins, carries two transactivation domains, and has the capacity to form dimers or multimers. This study provides the first three-dimensional structure of the EBNA-2 N-terminal Dimerization (END) domain. Two END domain monomers, each consisting of four β-strands and a single α-helix, assemble into a dimer by interaction of two β-strands from each monomer in a parallel fashion. The dimer surface exposes residues that are critical for transactivation of target genes by EBNA-2. The dimeric fold of the EBNA-2 END domain has not been observed for any cellular protein and thus could provide a novel target for anti-viral therapeutics.
| Epstein-Barr virus (EBV) is a γ-herpesvirus that establishes a lifelong asymptomatic infection in the majority of human adults. EBV infection or reactivation can cause significant morbidity and mortality in immunocompromised transplant recipients of allogeneic hematopoietic stem cells or solid organs [1, 2]. EBV has the unique ability to transform resting human B cells into permanently proliferating latently infected lymphoblastoid cell lines. This process is controlled by the concerted action of six latent EBV nuclear antigens (EBNAs) and three latent membrane proteins (LMPs), which mimic cellular functions required for B cell proliferation and differentiation. EBNA-2 is a key viral factor in the initiation of the transformation process. The protein controls a specific transcription program that is associated with proliferation of the infected B cells and that closely resembles transcript patterns of EBV infected B cells described in post-transplant lymphoproliferative disorders (PTLD) of immunosuppressed patients [3]. Thus, EBNA2 could potentially serve as a target to develop therapeutic strategies which interfere with the proliferation of EBV positive PTLD originating from B cells. Structural information on EBNA2 could guide the development of new antivirals in the future.
EBV belongs to the genus of lymphocryptoviruses (LCV) and is the only LCV species that infects humans. Mainly based on the sequence diversity of the EBNA-2 alleles EBV can be categorized in two individual strains called type 1 and 2. Type 1 and 2 EBV strains differ in their capacity to immortalize primary B cells [4, 5] which is predominantly determined by sequence variation in the C-terminus of EBNA-2 [6, 7]. Most researchers in the field use the laboratory EBV strain B95-8 (type 1) which encodes a 487 amino acid EBNA-2 protein [8, 9]. Lymphocryptoviruses have also been isolated from baboon and macaque. While the EBNA-2 orthologs of baboon and macaque LCV show significant amino acid similarity with EBNA-2 encoded by the B95-8 strain [10, 11], similarity with the positional EBNA-2 homolog of marmoset LCV is below 20% (reviewed in [12]).
The transactivator EBNA-2 does not bind to DNA directly but uses cellular DNA binding proteins like CBF1/CSL as adapters to gain access to enhancer and promoter sites in the viral and cellular genome (reviewed in [13]). Two transactivation domains have been mapped within the primary structure of the EBNA-2 protein by tethering EBNA-2 fragments fused to the yeast GAL4 DNA binding domain to GAL4 dependent reporter genes (Fig 1A). The C-terminal acidic transactivation (C-TAD, aa 448–471) domain can recruit components of the basic transcriptional machinery like TFIIE via p100, TFIIB, TAF40, to the TFB1/p62 subunit of the TFIIH complex, RBP70 [14–18] and chromatin modifiers like p300/ CBP and PCAF [19] and might directly bind to the viral co-activator EBNA-LP [20]. The EBNA-2 C-TAD is intrinsically unstructured as shown by NMR. However, the C-TAD forms a 9-residue amphipathic α-helix when bound to the pleckstrin homology (PH) domain of the yeast homolog of fragments of the TFB1/p62 subunit of the TFIIH complex. Three hydrophobic residues (Trp458, Ile461, and Phe462) of this α-helix directly contact the TFB1 PH domain. The same EBNA-2 residues are critical for the interaction with CBP/p300 [21].
A second transactivation domain has been mapped to the N-terminus (N-TAD, aa 1–58) of the EBNA-2 protein [22]. The molecular mechanism by which this second EBNA-2 transactivation domain acts has not yet been elucidated. Like the C-TAD its activity can be enhanced by EBNA-LP although it does not bind directly to EBNA-LP [22–24]. When GAL4 DNA binding domain fusion proteins of the N- or C-TAD are compared directly, they score equally well in transient transactivation assays [22]. Deletion of the N-terminus causes a severe loss of activity, while deletion of the C-TAD completely abolishes transactivation of target genes indicating that the function of the two transactivation domains are neither equivalent nor redundant [15, 25]. The relevance of the N-terminus of EBNA-2 for the growth transformation process has been studied in two independent cellular systems. The results of both studies suggested that the N-terminus of EBNA-2 is of major importance for the transformation efficiency of the virus and the survival of EBV infected B cells [24, 26].
Two N-terminal regions separated by a poly-proline stretch have been proposed to mediate homotypic self-association of EBNA-2. The first, consisting of amino acid 1–58 coincides with the N-terminal transactivation domain [22, 23]. A second self-associating region is composed of amino acid 97–121 [23]. An additional self-associating domain has been mapped to a non-conserved region which is flanked by the second dimerization and the adapter region [27].
The N-terminal region of EBNA-2 comprising residues 1–58 appears to mediate multiple molecular functions including self-association, transactivation and functional cooperation with EBNA-LP. Similar functions have also been assigned to other parts of the protein. So far it is unknown if the N-terminus of EBNA-2 directly provides all these functions or if these activities merely depend on the dimerization involving the N-terminal domain. Thus, the molecular basis and functional importance of the dimerization regions are poorly understood since three-dimensional structural data for the entire EBNA-2 protein have not been reported.
Here, we present the three-dimensional structure of the EBNA-2 N-terminus which forms a compact parallel homodimer that is stabilized by a hydrophobic interface between the two monomers. The dimer interface involves two β-strands of each protomer that pack against each other in an anti-parallel manner. Based on this structural information we generated site-directed mutants which target either the hydrophobic dimer interface or solvent-exposed residues. We show that interface mutations abolish self-association of EBNA-2 and severely impair its transactivation function. Notably, surface mutants do not impair self-association. However, specific point mutations or deletion of a protruding α-helix on the surface of the END domain cause a major loss of biological activity. These data suggest that the EBNA-2 dimer provides a surface that is critical for its transactivation function.
Structure predictions for the full-length EBNA-2 amino acid sequence suggest that this viral protein does not form a globular three-dimensional fold, consistent with the presence of extended poly-proline or poly-glycine-arginine regions, and with a total proline content of 28%. The EBNA-2 protein thus appears to comprise intrinsically unstructured regions, which require interaction partners for proper folding. However, in silico analysis of the primary structure using PSIPRED [28], predicts that the N-terminal region comprises β-strands and thus might represent a small globular domain (Fig A in S1 Text).
To characterize biochemical and structural details of this region of EBNA-2, an N-terminal fragment comprising residues 1–58 was expressed in E.coli and purified with or without Z-tag under native conditions. The oligomerization status of the recombinant proteins was analyzed by analytical size exclusion chromatography (SEC) and static light scattering (SLS) (Table 1).The EBNA-2 N-terminal fragment lacking a Z-tag forms a single molecular species with a molecular mass of 13.1 kDa as expected for a dimer (2x6.7 kDa). Similarly, the EBNA-2 Z-tag fusion protein eluted as a single peak with a molecular mass of 46.3 kDa close to the theoretical molecular mass of a dimer (2x23.4 kDa).
We next determined the three-dimensional structure of this N-terminal fragment by heteronuclear nuclear magnetic resonance (NMR) spectroscopy. The solution structure of the N-terminal domain is well-defined by the NMR data and based on more than 1250 nuclear Overhauser effect (NOE)-derived distance restraints per monomer and 205 inter-monomer NOEs (Table 2). The structure reveals a parallel homodimeric arrangement of monomers each comprising four β-strands (β1-β4) and a short exposed α-helix (α1) remote from the dimer interface (Fig 1B and 1C). The central portion of the dimer is assembled by two curved anti-parallel β-sheets with an anti-parallel arrangement of β1-β4 with β4’-β1’ and β3-β2 with β2’-β3’ (un/primed secondary structures refer to the individual monomers). The dimer interface is constituted by anti-parallel interactions of β4-β4’ and β2-β2’, respectively (Fig 1B and 1C, right panel). The secondary structure observed in the structure is consistent with NMR secondary chemical shifts (Fig 2A). Structural similarity searches in the Protein Data Bank (PDB) using DALI and PDBeFold did not identify any structures with a similar fold (see Experimental Methods for details). Thus, the N-terminal domain of EBNA-2 represents a novel dimerization fold, which we propose to name “END” (EBNA-2 N-terminal Dimerization) domain.
The END domain is highly stable with a melting point of approximately 70°C (determined by thermal denaturation [29]). A strong interaction between the monomers is also consistent with a large buried surface area (1165 Å2, corresponding to one quarter of the total surface area per monomer) [30]. NMR relaxation data show that the folded region of the END domain between β1-β4 is highly rigid, while C-terminal residues (beyond Asn55) are flexible and exhibit internal dynamics at sub-nanosecond timescales (Fig A in S1 Text).
The END homodimer is stabilized by the formation of a hydrophobic core involving numerous residues from each monomer (Fig 1C). While some of these residues mainly stabilize interactions within each monomer, the dimer interface is formed by hydrophobic interactions of the side chains of Leu8, Tyr14, Leu16, Val18, Ile46, Leu48, Ile50, and Val52. Also, stacking of the solvent exposed side chains of His15 and Phe51 from both monomers contributes to the dimer interface. In addition to the hydrophobic interactions, hydrogen bonds between the peptide backbone of β2 and β2’, as well as β4 and β4’ are formed. These backbone interactions are supported by NMR-detected hydrogen-to-deuterium (H/D) exchange measurements, which indicate that most of the backbone amide protons that participate in intra-monomer or inter-monomer hydrogen bonds are protected against solvent exchange (Fig 2B and Fig B in S1 Text).
Taken together our structural and biophysical data shows that the recombinant wild-type END domain folds independently into a very stable dimer. Thus, we expect that the determined protein structure indicates a native assembly of the EBNA-2 protein and decided to further characterize and validate the dimer structure and its function using site-directed mutational analysis in vitro and in vivo.
The primary sequences of the END domain from type 1 EBV strains are highly homologous (>96% identity to B95-8). AG876, a type 2 strain, exhibits slightly lower sequence identity (79%), while the sequence identity of baboon and macaque LCV is significantly lower (41–50%). Interestingly, all hydrophobic amino acids which are an integral part of the dimeric interface are highly conserved between man and monkey viruses. Out of the eight residues, six are identical and two are highly similar (Fig 2B and 2C). This suggests that the dimer interface of the END domain is conserved in the EBNA-2 proteins of EBV and baboon and macaque LCV, and thus may play an important functional role. In addition, the program PSIPRED predicts 4 β-strands in similar positions for EBV and LCV END domains proposing that the dimer fold might be a conserved motif across species (Fig A in S1 Text).
To determine the contribution of particular residues for END domain dimer formation, we designed mutations to disrupt specific interactions in the dimerization interface (interface mutants, Fig 3A). We replaced Leu16 and Ile50 by either alanine or aspartate as both residues are positioned directly at the interface and interact with the same residue in the other monomer. Replacement by aspartate was considered to introduce charge repulsion in the dimer interface and thus expected to strongly impair dimerization. Leu16 and Ile50 mediate important hydrophobic interactions and are completely conserved in all human and monkey sequences (Fig 2C). In a second set of mutations, we altered solvent-exposed residues at the surface of the END structure (surface mutants, Fig 3A), such as His15 and Phe51. We also studied an END domain variant where helix α1, residues 35–39, had been deleted (Δα1). These surface residues and helix α1 are not expected to be essential for dimerization but could mediate molecular interactions that might be required for functional activity.
The dimerization properties and structural integrity of the mutant END domains were characterized by SEC/SLS and NMR spectroscopy (Table 1 and Figs C and D in S1 Text). The interface mutants were more difficult to purify than the wild-type protein and are prone to aggregation as judged by SLS analysis. Due to the low solubility of mutant END domains, SLS was also performed on Z-tag fusion proteins to enhance solubility of the fusion proteins. The L16A mutant exists in equilibrium between an unfolded monomeric and folded dimeric state. The L16D, I50A, and I50D mutants are greatly destabilized leading to high molecular weight aggregates (SLS, Fig D in S1 Text) and could not be analyzed by NMR. The data suggest that interface mutations destabilize the dimerization interface and thus promote aggregation of monomeric END domains, as monomers would expose hydrophobic residues.
H15A yields homogeneous protein samples and is a dimer as indicated by SLS analysis (Fig D in S1 Text) and a well-dispersed NMR spectrum (Fig D in S1 Text). SLS data for F51A and Δα1 mutant END domains indicate the presence of dimer populations but also some aggregated species. This is further confirmed by NMR spectra, which are recorded at higher concentration and show the presence of dimeric and aggregated species in solution for these mutants (Fig D in S1 Text). Residue F51 is located at the surface of the END domain but contributes to the dimerization interface. Mutation to alanine may thus destabilize the dimer and lead to aggregation due to solvent exposure of the hydrophobic dimerization interface. Similarly, although removal of helix α1 does not globally disturb the fold and dimerization it may enhance aggregation at the concentrations used in NMR and SLS. NMR spectra clearly indicate the presence of folded dimer species for all surface mutants, i.e. H15A, F51A and Δα1. To further characterize these mutations, we analyzed their effect on dimerization of the full-length protein in cells (see below).
As EBNA-2 has been reported to carry at least two domains implicated in dimerization (residues 1–58, i.e. the END domain, and residues 96–210), we tested whether mutants that abolish self-association of the END domain in vitro would also impair self-association of the full-length EBNA-2 protein [23, 24]. We expressed wild-type, deletion, surface and interface mutants as full-length EBNA-2 HA-tagged proteins and performed co-immunoprecipitation experiments in EBV negative DG75 cells [31] (Fig 3B–3E). For comparison we included HA-tagged mutants of EBNA-2 lacking amino acids 3–30 or 3–52 in our analyses (Δ3–30 and Δ3–52, respectively). All EBNA-2 mutants were expressed well and could be co-expressed with a FLAG-tagged EBNA-2 fragment encompassing amino acid 1–199 (F199). Co-immunoprecipitation studies using HA-specific antibodies indicated that all EBNA-2 mutants efficiently bound to endogenous CBF1. Both EBNA-2N-terminal deletion mutants were significantly impaired for self-association as has been reported previously (Fig 3D) [24]. The residual binding of Δ3–30 and Δ3–52 to F199 might be supported by the second self-association domain, comprising residues 96–210, which is still present in the F199 protein [23]. The self-association domain of a non-conserved region [27] is not present in F199 and thus cannot account for residual dimerization.
Next, we tested whether the interface mutants L16A, L16D, I50A and I50D can still mediate self-association with the EBNA-2 F199 fragment, which also harbors the END domain (Fig 3C, middle and right panel). While substitution of the Leu16 or Ile50 by alanine did not significantly affect F199 association, introduction of a negative charge by aspartic acid prevented self-association. These results confirmed the structural data indicating that hydrophobic residues facing each other across the dimer interface of the END domain are essential for EBNA-2 self-association. Surprisingly, Δ3–30 and Δ3–52 appeared to be less impaired than L16D and I50D.
In order to further validate the structural integrity of the END domain in the context of the complete EBNA-2 protein we tested the surface mutants H15A, Δα1 and F51A for association with F199 (Fig 3E). Consistent with the structural and biophysical data all surface mutants retained the capacity to self-associate, confirming that these residues are not essential for the dimerization of EBNA-2.
Nuclear localization and formation of nuclear speckles is a typical feature of EBNA-2 [32]. In order to analyze whether the END domain mutants had retained these features all EBNA-2 mutants were expressed in HeLa cells and the subcellular distribution of the EBNA-2 proteins was analyzed by confocal microscopy (Fig E in S1 Text). All mutants still showed strict nuclear localization, which typically excludes the nucleoli. Moreover, all mutants formed granular speckles, which are characteristic of wild-type EBNA-2 protein.
Based on previous work, EBNA-2 mutants impaired for dimerization were also severely impaired for activation of the viral target gene LMP1 [24]. In order to analyze the capacity of the EBNA-2 surface and interface mutants to activate the viral LMP genes we expressed EBNA-2 mutants in the EBV positive Burkitt's lymphoma cell line Eli-BL [33]. This B cell line exhibits a specific viral gene expression program where neither EBNA-2, nor EBNA-LP nor LMP proteins are expressed. By transient transfection of EBNA-2 expression constructs, endogenous LMP1 protein expression can be induced. We used this cellular system to measure the biological activity of EBNA-2 amino acid substitution mutants compared to N-terminal deletion mutants. The N-terminal deletion mutants, Δ3–30 and Δ3–52, which are expected to disrupt the END domain fold or delete it, are severely impaired for LMP1 activation, while the biological activity of the interface mutants L16A and I50A, which still self-associate, is comparable to wild-type EBNA-2 (Fig 4A and 4B). However, the functionality of the interface mutants L16D and I50D, which do not retain dimerization, is strongly attenuated. Notably, activation of LMP1 by the surface mutants H15A, F51A, and Δα1 is differentially affected. While F51A is unaffected, the activity of H15A and Δα1mutants is severely reduced (Fig 4C). EBNA-2 and all EBNA-2 mutants were expressed well in Eli-BL. Thus, the distinct biological activity of the mutant EBNA-2 proteins is not due to differential expression levels (Fig 4D).
In order to analyze the capacity of all EBNA-2 mutants to induce endogenous transcripts we selected two viral, LMP1 and LMP2A, and two cellular target genes, CCL3 and CD23, for quantitative RT-PCR analyses in Eli-BL (Fig 5). These four genes all carry functional CBF1/CSL binding sites in their promoter region within less than 500 base pairs upstream of the transcription start site [34, 35]. The LMP1 promoter is controlled by a complex network of transcription factors that includes CBF1/CSL. However, although CBF1/CSL enhances transactivation by EBNA-2, the LMP1 promoter is unique since it can still be activated by EBNA-2 to up to 50% in the absence of CBF1/CSL [36] (and our unpublished data). In contrast, the LMP2A promoter carries two adjacent CBF1/CSL sites which are essential for EBNA-2 transactivation. Activation of the two cellular genes CCL3 and CD23 is strictly CBF1/CSL dependent [37–39]. Compared to wild-type EBNA-2, all END domain mutants, even those that still dimerize in cells, showed some loss of activity indicating that the integrity of this domain is critical for EBNA-2 function. The surface mutant F51A appears to be affected the least. Neither Δ3–30 nor Δ3–52 could efficiently activate any of the four genes. LMP1 induction was impaired the most, while activation of LMP2A is the least sensitive.
In parallel we studied the activity of the viral C promoter and the endogenous EBNA-2 transcript levels after transfection. C promoter transcript levels were close to detection limits and were not modulated by either EBNA-2 or EBNA-2 mutants. Endogenous EBNA-2 transcript levels were undetectable and could also not be induced. Thus, we can exclude that endogenous EBNA-2 in Eli-BL interferes with our assay in Eli-BL (Fig F in S1 Text). It appears that the END domain is critical not only for LMP1 transactivation but rather is required in a universal manner for transactivation of unrelated genes although to different extent.
In order to prove that the END domain surface has a general impact on the transactivation capacity of the EBNA-2 protein we performed promoter reporter luciferase assays using Gal4 DNA-binding domain fusion proteins and two distinct promoter reporter constructs which either carried 10 GAL4 binding sites or 12 CBF1 binding sites to recruit GAL4 EBNA-2 (Fig 6). GAL4 EBNA-2 was efficiently recruited to both promoters and activated luciferase expression. The GAL4 EBNA-2 H15A mutant had lost more than 50% of its transactivation capacity on both luciferase constructs. The biological activity of GAL4 EBNA-2 Δα1 was almost completely abolished. Again the surface F51A mutant was affected the least.
In summary, EBNA-2 END domain mutations that do not affect dimerization are severely impaired for transactivation of endogenous target genes as well as artificial promoter reporter constructs. Loss of function was most pronounced for END domain deletion mutants and was almost as strongly observed with the surface mutants H15A and Δα1. The dramatic loss of function seen in mutants – that still dimerize, properly localize to the nucleus, and bind to CBF1 – suggests that the END domain not only promotes dimerization of EBNA-2 but conveys additional critical functions.
Here, we report the first three-dimensional structure information for the EBNA-2 protein. The N-terminal region of EBNA-2 represents a specific dimerization domain designated END (EBNA-2 N-terminal Dimerization) domain. The dimer is stabilized by anti-parallel interactions of β4-β4’ and β2-β2’, which generate a strong hydrophobic interface which stabilizes the dimer. In fact, dimerization via hydrophobic interfaces of diverse structures is a frequent feature of small dimers (<100 aa per monomer) [40]. However, to our knowledge the specific fold of the END domain dimer is novel. Notably, the hydrophobic residues which form the dimerization interface are completely conserved in EBV and rhesus LCV sequences. We thus expect that the dimerization by the END domain is conserved in all EBV sequences and most likely also in macaque and baboon EBNA-2 orthologs.
To probe the dimerization interface we generated END domain mutants which affect residues in the dimer interface. Mutation of these interface residues were indeed found to disrupt the fold of the END domain and/or lead to aggregation of recombinant protein. For further analysis, all END domain mutants were expressed as full length EBNA-2 protein in human B cells and tested for self-association and transactivation of endogenous target genes. While self-association of the EBNA-2 L16A and I50A interface mutants was marginally impaired, self-association of L16D and I50D was close to or below detection levels. Surprisingly, even the N-terminal deletion mutants (Δ3–30 and Δ3–52) exhibited residual binding activity stronger than L16D and I50D. Potentially the second dimerization domain (Dim2, Fig 1A) could be unmasked in the absence of the END domain. Or, single amino acid substitutions in the hydrophobic core may cause non-physiological aggregation-states of EBNA-2 and impair protein function even stronger than loss of the END domain.
Our data provide convincing evidence that the END domain is a conserved dimerization motif for the full-length EBNA-2 protein. As the END domain is separated from the rest of the EBNA-2 protein by an extended poly-proline hinge region, we suggest that the END domain acts as an independent module that mediates self-association of the entire protein.
EBNA-2 is recruited to DNA by adapters like CBF1/CSL but might require at least two factors to which it binds simultaneously to activate viral target genes. So, the viral LMP2A promoter carries two functional CBF1/CSL binding sites, while the LMP1 promoter requires PU.1 and CBF1/CSL for efficient activation by EBNA-2 [36, 41, 42]. By using CBF1/CSL as a DNA adapter, EBNA-2 mimics the activated Notch receptor which also is recruited to DNA by CBF1/CSL. Interestingly, Notch dimers frequently use paired CBF1/CSL1 binding sites in the cellular genome [43] which might also be used by EBNA-2 dimers. In the cellular genome, EBNA-2 binds preferentially to enhancers which can be located remote from the promoter of the regulated genes [44]. Thus, it may be proposed that dimerization promotes higher order protein complex assembly that bridges promoter and enhancer regions.
According to the NMR and SEC/SLS analyses all surface mutants of the END domain are folded and comprise dimeric species, although F51A and Δα1 have a tendency to aggregate. In B cells, the full-length surface mutants EBNA-2 H15A, Δα1, and F51A mutants self-associate, further corroborating the in vitro data. Notably, transactivation of target genes by the surface mutants H15A and Δα1 was severely reduced to similar levels observed for aspartic acid interface mutants, which abolish self-association. This indicates that the effects onto the functional activity are not due to impaired dimerization but suggest that these residues may be involved in additional intra- or intermolecular molecular interactions.
We directly compared the different END domain EBNA-2 mutants for their capacity to induce either LMP1 protein expression or endogenous LMP1, LMP2A, CCL3, or CD23 transcript levels in Eli-BL cells. These four genes share functional CBF1 binding motifs but rely on these motifs to varying degrees. Importantly, all END domain mutants retain the capacity to bind to CBF1. We find that the residual self-association of the two N-terminal deletion mutants (Fig 3D) is not sufficient to restore the biological activity of the mutants to wild-type levels (Figs 4 and 5). Although LMP1 and CCL3 induction are affected the most, all mutants produce similar patterns of loss of activity for all genes we have tested. Since we did not observe a gene specific phenotype for any of the mutants, a single so far unknown factor could interact with the END domain of EBNA-2 and be required for the activation of each of the four target genes. In EBV infected B cells, the EBNA-LP co-activator of EBNA-2 could be a candidate factor to play this role. However, since EBNA-LP is not expressed in EBV negative DG75 cells and neither expressed nor induced by EBNA-2 in Eli-BL cells [45], EBNA-LP can be excluded in our setting. At this point of our studies we speculate that basic mechanisms of transcriptional activation by EBNA-2 are impaired in the surface mutants H15A and Δα1.
In the past, multiple transactivation domains (TADs) have been defined by generating chimeras of protein fragments of interest and an unrelated DNA binding domain. These chimeras were tested for their activity to induce artificial promoters recruited by the DNA binding domain [46]. Most of the TADs, which scored positive in these assays, were enriched for hydrophobic or acidic amino acids or a 9aa TAD sequence motif [47]. In retrospect it was found that TADs not only bind to general factors of the transcription machinery, but also confer contact to components of the mediator, the SAGA complex or the chromatin remodeling machinery. Most TADs appear to be intrinsically unstructured. However, in complex with their cognate binding partners they may fold into specific structures which mediate protein-protein interactions (reviewed in [48]). In contrast to the acidic C-TAD of EBNA-2, which is intrinsically unstructured and attains a stable secondary structure only upon complex formation with cellular proteins [21], the END domain appears to be a non-typical TAD. In the absence of any cognate cellular binding partner the END domain folds into a well-defined rigid dimeric globular structure.
Taken together our structural and mutational analysis suggests that the dimerization by the END domain provides a surface that is critical for transactivation of target genes, for example, by exposing His15 and the α1-helix. Since all loss-of-function mutants interfere with activation of all genes that were tested, the END domain is likely to interact with candidate proteins which could be critical for transactivation at multiple steps.
EBNA-2 expression is a hallmark of B cell lymphomas arising in immunocompromised patients and considered to drive the proliferation of these cells. The END domain has a strong impact on the biological activity of EBNA-2 and thus it should be considered as a potential drug target for small molecules [24, 26]. The END domain forms a novel, highly stable parallel dimeric fold, which is stabilized by conserved hydrophobic interactions. Importantly, our in silico searches for cellular protein sequences or related folds similar to the END domain did not reveal any homologous cellular domains suggesting that the END domain is a unique structure that evolved in lymphocryptoviruses and thus is virus specific. Our future studies will focus on the identification of potential proteins which bind to the END domain and require His15 or the α-helix for protein interactions. The dimerization or the suggested binding surface of the END domain might be targeted by small molecules to impair EBNA-2 activity for potential therapeutic intervention.
The design of constructs for structural and biochemical studies was guided by secondary structure prediction (PSIPRED) [28]. Residues 1–58 of EBNA-2 (Strain B95-8; Uniprot: P12978) were cloned into a modified pET-24d expression plasmid following standard restriction digest procedures. The vector contained a Z-tag, as well as a 6xHis-tag to facilitate purification. The Z-tag is a 125 amino acid protein tag based on protein A from Staphylococcus aureus and is known to enhance the solubility of fusion proteins [49]. Both of these N-terminal tags could be removed by proteolytic cleavage using tobacco etch virus (TEV) protease. For cloning purposes and efficient TEV protease cleavage the final protein construct contained four additional residues at the N-terminus (Gly-Ala-Met-Glu). Mutations to study the functional importance of the END domain were introduced by overlap extension (also known as two-step) PCR. In brief, mutation primers were used in combination with the original forward or reverse primers in a first round of separate PCR experiments. The purified products were then combined and used as the template for a second round of PCR using only the original forward and reverse primers. Restriction digestion and ligation of the final product yielded expression plasmids in a similar way to the original construct. Mutant END domains were expressed and purified in similar fashion as the wild-type protein. For expression studies in mammalian cells all END domain mutant gene fragments were sub-cloned into pAG155, to generate EBNA-2 carrying an HA tag at the C-terminus of full-length proteins by conventional cloning techniques [24]. In order to express GAL4 EBNA-2 fusion proteins the GAL4 DNA binding domain (DBD) gene fragment was added to the 5’ end of the EBNA-2-HA ORF. Luciferase promoter reporter gene assays were performed using the Promega dual luciferase assay system. The CBF1 reporter (pGa981-6) carries 12 CBF1 binding sites [50] and the GAL4 (Gal4 tk-Luc) responsive reporter construct carries 10 GAL4 binding sites. For normalization the pRL-PGK Renilla Luciferase construct was used. The integrity of all expression plasmids was confirmed by sequencing.
Recombinant proteins were expressed in Escherichia coli BL21 (DE3). Using kanamycin for selection, one colony was picked from a fresh transformation plate to inoculate a 5 mL pre-culture in lysogeny brothmedium. The pre-culture was used to start larger culture volumes of unlabeled LB, or minimal M9 media for expression of isotope-labeled proteins. For production of 13C and 15N-labeled protein samples [U-13C]-D-glucose and 15NH4Cl were included as the sole carbon and nitrogen sources, respectively. Cultures were grown at 37°C until the optical density reached 0.8 and then, after cooling to 20°C, induced overnight (16 h) by addition of 0.5mM isopropyl β-D-1-thiogalactopyranoside. Cells were harvested by centrifugation (8000 g, 20min) and disrupted by pulsed sonication (6 min, 30% power, large probe, Fisher Scientific model 550) in lysis buffer (20 mM TRIS pH 7.5, 300 mM NaCl, 10 mM imidazole, and 0.02% NaN3), containing protease inhibitors, DNase, lysozyme, and 0.2% IGEPAL. After centrifugation and filtering the lysate was passed three times over Ni-NTA agarose resin (Qiagen) in gravity-flow columns (Bio-Rad). Bound protein was washed extensively with the lysis buffer, the lysis buffer containing no IGEPAL, and lysis buffer with high salt NaCl (1 M) or imidazole (30 mM) concentrations. The protein was eluted with the elution buffer (20 mM TRIS pH 7.5, 300 mM NaCl, 300 mM imidazole, and 0.02% NaN3). The eluted protein was buffer exchanged into TEV cleavage buffer (10 mM NaP pH 7.5, 150 mM NaCl, 1 mM DTT, and 0.02% NaN3). TEV protease was added to a molar ratio of 1:10, protease to recombinant protein, and incubated overnight at 4°C. To efficiently remove TEV protease and the cleaved off solubility tag, the sample was passed over an ion-exchange column (Resource Q, GE Healthcare) which was equilibrated with the buffer (20 mM sodium phosphate, pH 6.9, 20 mM NaCl, and 0.02% NaN3). The protein was eluted from Resource Q column with a NaCl gradient (0–0.5M over 60 ml). Additionally, a last purification step was implemented and included size-exclusion chromatography (HiLoad16/60, Superdex 75, GE Healthcare). The size-exclusion column was equilibrated and run in a buffer appropriate to subsequent studies.
NMR experiments were performed on Bruker instruments operating at a field-strength corresponding to a proton resonance frequency of 500, 600, 750, 800, and 900 MHz equipped with pulsed field gradients and cryogenic probes (except at 750 MHz). Spectra were generally recorded at 323K (50°C) on protein samples (1 mM) in20 mM sodium phosphate, pH 6.9, 20 mM NaCl, and 0.02% NaN3. Spectra were processed with NMRPipe [51] and analyzed in NMRView [52] and Sparky 3.
For assignment of backbone amides and side-chain signals the following multidimensional heteronuclear experiments were acquired [53]: 1H,15N-HSQC, 1H,13C-HSQC, HNCA, HNCACB, CBCA(CO)NH, (H)CC(CO)NH-TOCSY, H(C)CH-TOCSY, and HCC(H)-TOCSY. Assignment of aromatic protons was accomplished by two-dimensional (HB)CB(CG,CD)HD and (HB)CB(CG,CD,CE)HE spectra. Stereospecific assignment of the methyl groups in leucine and valine residues was achieved by partial 13C-labeling and by observing the presence or absence of a hydrogen-carbon J-coupling in a 2D 1H-13C HSQC [54]. Distance restraints were derived from three-dimensional NOESY experiments: 1H,15N-HSQC-NOESY,1H,13C-HMQC-NOESY (for both the aliphatic and the aromatic region), and 13C-edited-15N/13C-filtered NOESY (aliphatic region). Denaturation and refolding of the END dimer was required for measurement of the intermolecular NOEs. This was accomplished by taking equimolar amounts of unlabeled and double labeled (15N, 13C) protein and adding 8M urea. The mixture was heated to 80°C for 10 min and then dialyzed twice against NMR buffer at 4°C. Importantly, appropriate samples were lyophilized and dissolved in pure D2O to increase sensitivity of several experiments, and to simplify spectral analysis.
Automated NOESY assignment and derivation of distance restraints was performed using CYANA v3.0 [55]. Dihedral restraints were obtained with TALOS+ [56], using assigned chemical shifts as input, and inspected manually to remove less reliable predictions. The final structure calculations in ARIA v2.2 [57] included refinement in explicit water and activation of a non-crystallographic two-fold symmetry constraint. Out of one hundred calculated structures, ten models were selected as a representative ensemble based on low energy and restraint violations. Analysis of structure quality and restraint violations was performed with iCing including PROCHECK [58] and WHATCHECK [59]. Figures and structure ensemble alignment were prepared in Pymol v1.5 [60].
1H,-15N heteronuclear NOEs were measured at 318K on a 500 μM 15N-labeled sample (750 MHz proton Larmor frequency) as described previously [61], and analyzed in NMRView. The secondary chemical shift analysis was also done in NMRView. Hydrogen-deuterium exchange experiments were performed by NMR to detect solvent protected backbone amide protons. A 1H,15N-HSQC was recorded on a lyophilized protein sample 10 min after dissolving it in D2O, and compared to a reference spectra in H2O. Both spectra were recorded at 313K to reduce the amide proton exchange rates with the solvent. Any residual signals observed above noise were considered indicative of solvent protected amide protons.
A BLAST sequence search of the Protein Data Bank (PDB) generated no hits with reasonable E-values (< 1) or domains with structural similarities to the END domain. The fold of the END domain was further compared to previously determined protein structures deposited in the PDB using the DALI server as well as PDBeFold, available from EMBL/EBI. Interestingly, the DALI server only returned low-scoring hits for the complete dimer with relatively high RMSD values and low sequence identity. The structural superpositions of the END domain with the top twenty hits were manually examined, without the discovery of any similar folds. The most commonly matched structural feature of the END domain was the large anti-parallel beta-sheets (β1-β4-β4’-β1’), while the rest of the dimer and the ordering of the beta-strands, never exhibited an adequate fit. Likewise, PDBeFold produced no hits with reliable scores for the END monomer. Top hits only matched two out of the five secondary structure elements, and visual inspection confirmed lack of conserved structures. In conclusion this lack of similar structures strongly suggests that the END domain is of a novel fold and that this is the first structural determination of this viral dimerization motif.
SLS was measured with a Malvern-Viscotekinstrument (TDA 305) connected downstream to an Äkta Purifier equipped with an analytical size-exclusion column (Superdex 75 10/300 GL, GE Healthcare). Samples were run at a concentration between 150 and 400 μM in a running buffer containing 20 mM NaP pH 6.9, 20 mM NaCl, and 0.02% NaN3. Elution profiles were collected for 30 min with a flow rate of 1 mL/min. Data were collected using absorbance UV detection at 280 nm, right angle light scattering (RALS) and refractive index (RI). The molar masses of separated elution peaks were calculated using OmniSEC software (Malvern). As standard for calibration, 4 mg/mL Bovine Serum Albumin (BSA) was used prior to all experiments and the change in refractive index with respect to concentration (dn/dc) was set to 0.186 mL/g [62].
DG75 [31], Eli-BL [33], and 721 [63] cells were maintained in RPMI 1640 medium supplemented with 10% fetal calf serum, 100 U/mL penicillin, 100 μg/mL streptomycin and 4 mM glutamine at 37°C in a 6% CO2atmosphere. For transfection, 5x106DG75 or 2x107 Eli-BL cells were electroporated in 250 μL Optimem medium at 240 V and 975 μF using the Genepulser II (Bio-Rad) and allowed to recover in 10 mL of cell culture medium for 24 h. Luciferase promoter reporter gene assays were performed using the dual luciferase assay system (Promega) according to the manufacturer's instructions. Results obtained for firefly luciferase activity were normalized to Renilla luciferase activity.
HeLa [64] cells were cultivated in DMEM supplemented with 10% fetal calf serum, 100 U/mL penicillin, 100 μg/mL streptomycin and 4 mM glutamine at 37°C in a 6% CO2 atmosphere. Cells were transfected with a mixture 1.5 μg of EBNA2 expression plasmids and 4 μg polyethylenimine (Sigma) in the presence of Optimem (Gibco). After 4 h, the medium was replaced with cell culture medium and cells were allowed to recover for 24 h and subsequently cultured for 24 h on cover slips. The cells were fixed with 2% paraformaldehyde (PFA) at RT for 15 min and subsequently permeabilized with PBS/0.15% TritonX-100 3 for5 min at RT. All samples were blocked with 1% BSA/0.15% glycine 3x for 10 min and incubated with the EBNA-2 specific antibody (R3) over night at 4°C. Cells were washed with PBS for 5 min, with PBS/0.15% TritonX-10 for 5 min, with PBS 5 min, blocked with PBS/1% BSA/0.15% glycine for 7 min and incubated with Cy3-conjugated goat anti-rat immunoglobulin (Jackson Immuno Research) in the dark for 45 min at RT. Cells were washed again with PBS/0.15% TritonX-100, and with PBS and stained with 0.1μg/ml 4',6-diamidino-2-phenylindole (DAPI) (Sigma) for 90sec and washed with PBS. Samples were embedded in fluorescent mounting medium (DakoCytomation). Confocal microscopy was performed on a Leica LSCM SP5 microscope equipped with 405 nm, 488 nm, 561 nm and 633 nm lasers. Images were taken with an objective HCX PL APO 63/1.4 objective and an electronic zoom of 3.6. Laser line 405 nm (DAPI) and 561 nm (Cy3) were used for image acquisition. Detection settings were carefully chosen to exclude spill-over of DAPI and Cy3 fluorescence.
For immunoprecipitation studies DG75 cells were lysed in 1% NP-40 buffer (10 mM TRIS pH7.4, 1 mM EDTA, 150 mM NaCl, 3% Glycerol, 1x complete protease inhibitor tablets (Roche)). The lysates were submitted to immunoprecipitation and total cell lysates and immunoprecipitates were analyzed by immunoblotting. For direct immunoblotting of Eli-BL cells they were lysed in RIPA buffer (50mM TRIS pH7.5, 150mM NaCl, 1% Igepal, 0.1% SDS, 0.5% Na-deoxycholate, 1x complete protease inhibitor tablets (Roche)) for 1 h and sonicated for 10 min (30s on, 30s off) at 230 V using a Bioruptor (Diagenode). Immunoblot assays were performed as described previously [38]. HA (3F10, Roche) and Flag (M2, Sigma) specific antibodies were obtained from commercial sources. The EBNA-2 (R3) [65], the EBNA-1(1H4) [66] and the LMP1 specific monoclonal antibodies (S12) [67] are published. Chemilumiscence signals of immunoblots were quantified by digital imaging using the Fusion Fx7.
Total RNA was extracted from 1x107 transfected Eli-BL cells 24 h post-transfection using the Qiagen RNeasy Mini Kit and cDNA was synthesized from 2 μg of RNA using the High-Capacity cDNA Reverse Transcription kit (Applied Biosystems) according the manufacturer´s protocol. qPCR of the transcripts was performed on a LightCycler 480 SYBR Green I Master (Roche) and the data were processed with the LightCycler 480 software (version 1.5.0.39, Roche). A total of 1/80 of cDNA product was used for amplification of actin and 1/40 of cDNA for all other genes. Cycling conditions were 10 min at 95°C and 45 cycles of 3 s at 95°C, 10 s at 60 or 63°C, and 20 s at 72°C on a 96-well thermal block. PCR products were validated by melting curve analysis and agarose gel electrophoresis. Quantification was based on standard samples of known concentration and standard curves for each primer pair. Primer pairs for RT-PCR were selected by Primer3 software All pairs were chosen to support amplification across intron borders. Primers were GGTGTTCATCACTGTGTCGTTGTC and GCTACTGTTTTGGCTGTACATCGT for LMP1 [68], ATGACTCATCTCAACACATA and CATGTTAGGCAAATTGCAAA for LMP2A [69], CTGGGACACCACACAGAGTC and GACACCTGCAACTCCATCCT for CD23, ATGCAGGTCTCCACTGCTG and TTTCTGGACCCACTCCTCAC for CCL3, AGATCAGATGGCATAGAGAC and GACCGGTGCCTTCTTAGGAG for C promoter usage, GCTGCTACGCATTAGAGACC and TCCTGGTAGGGATTCGAGGG for EBNA-2 [70], and GGCATCCTCACCCTGAAGTA and GGGGTGTTGAAGGTCTCAAA for actin.
Atomic coordinates of the END domain have been deposited at the Protein Data Bank (PDB) with accession code 2N2J. Experimental NMR distance restraints have been deposited at the Biological Magnetic Resonance Bank (BMRB) with accession number 19390.
|
10.1371/journal.pntd.0001552 | The Genome of Mycobacterium Africanum West African 2 Reveals a Lineage-Specific Locus and Genome Erosion Common to the M. tuberculosis Complex | M. africanum West African 2 constitutes an ancient lineage of the M. tuberculosis complex that commonly causes human tuberculosis in West Africa and has an attenuated phenotype relative to M. tuberculosis.
In search of candidate genes underlying these differences, the genome of M. africanum West African 2 was sequenced using classical capillary sequencing techniques. Our findings reveal a unique sequence, RD900, that was independently lost during the evolution of two important lineages within the complex: the “modern” M. tuberculosis group and the lineage leading to M. bovis. Closely related to M. bovis and other animal strains within the M. tuberculosis complex, M. africanum West African 2 shares an abundance of pseudogenes with M. bovis but also with M. africanum West African clade 1. Comparison with other strains of the M. tuberculosis complex revealed pseudogenes events in all the known lineages pointing toward ongoing genome erosion likely due to increased genetic drift and relaxed selection linked to serial transmission-bottlenecks and an intracellular lifestyle.
The genomic differences identified between M. africanum West African 2 and the other strains of the Mycobacterium tuberculosis complex may explain its attenuated phenotype, and pave the way for targeted experiments to elucidate the phenotypic characteristic of M. africanum. Moreover, availability of the whole genome data allows for verification of conservation of targets used for the next generation of diagnostics and vaccines, in order to ensure similar efficacy in West Africa.
| Mycobacterium africanum, a close relative of M. tuberculosis, is studied for the following reasons: M. africanum is commonly isolated from West African patients with tuberculosis yet has not spread beyond this region, it is more common in HIV infected patients, and it is less likely to lead to tuberculosis after one is exposed to an infectious case. Understanding this organism's unique biology gets a boost from the decoding of its genome, reported in this issue. For example, genome analysis reveals that M. africanum contains a region shared with “ancient” lineages in the M. tuberculosis complex and other mycobacterial species, which was lost independently from both M. tuberculosis and M. bovis. This region encodes a protein involved in transmembrane transport. Furthermore, M. africanum has lost genes, including a known virulence gene and genes for vitamin synthesis, in addition to an intact copy of a gene that may increase its susceptibility to antibiotics that are insufficiently active against M. tuberculosis. Finally, the genome sequence and analysis reported here will aid in the development of new diagnostics and vaccines against tuberculosis, which need to take into account the differences between M. africanum and other species in order to be effective worldwide.
| Mycobacterium africanum causes up to half of human TB in West Africa and displays differences in patient characteristics and immunoepidemiological features with M. tuberculosis, as reviewed earlier in this journal [1]. First described in 1968 in Dakar, Senegal [2], M. africanum used to be classified using biochemical methods, until unambiguous classification became possible using molecular methods and two different lineages were identified, M. africanum West African type 1, common to Eastern West Africa, and M. africanum West African type 2 , common to Western West Africa [3]. Additionally, it became clear that the former “East African M. africanum” is genetically part of M. tuberculosis sensu stricto [4]. The prevalence of M. africanum varies within West Africa, with the highest prevalence of M. africanum West African 2 identified in Guinea Bissau (51%, [5]) and the highest prevalence of – West African 1 recorded in Benin (around 28%, [6]). While comparisons between prior prevalence estimates based on biochemical speciation and current estimates based on molecular speciation deserve caution, the prevalence of M. africanum appears to be decreasing in Cameroon [7] and Senegal (unpublished results). Comparisons between patients infected with M. africanum West African 2 and M. tuberculosis suggest that M. africanum is somewhat attenuated in its ability to cause disease in immunocompetent hosts [8] and is more common in HIV co-infected patients relative to M. tuberculosis in The Gambia [9], yet not in Ghana [10]. Moreover, patients infected with M. africanum West African 2, as well as their household contacts, are less likely to mount an IFNg response to ESAT-6 than those infected with M. tuberculosis [11]. These two types of M. africanum, West African 1 and West African 2, are distinct sub-species within the M. tuberculosis complex although it has been suggested that these clades are better described as ecotypes of the M. tuberculosis complex rather than sub-species [12]. M. africanum West African type 2 is phylogenetically closer to the animal strains like M. bovis , with which it shares deletions RD7, 8 and 10 [13], [14], [15], although an animal reservoir for M. africanum West African type 2 has not been detected [16]. Subtractive hybridization of M. africanum West African type 1 and type 2 from M. tuberculosis H37Rv revealed shared and unique genomic differences [3], yet these experiments were not designed to identify regions present in M. africanum but absent from M. tuberculosis. Here we take advantage of available genomic information for strains of the Mycobacterium tuberculosis complex from different sequencing platforms to present and analyze the first complete M. africanum West African type 2 genome, that of clinical isolate GM041182 (here designated M. africanum GM041182 in the remainder of this manuscript), detailing a novel lineage-defining deletion and an array of species-specific pseudogenes.
Mycobacterium africanum GM041182 was isolated in The Gambia in 2004 from a 27 year old HIV uninfected male patient with 3+ smear positive pulmonary tuberculosis. This patient provided written informed consent for participation in the TB Case Contact cohort study, which had been approved by the joint Gambia Government/MRC ethics committee. Moreover, the same ethics committee approved genotyping of bacteria isolated from the patients enrolled in this cohort, and the data were analyzed anonymously. Primary isolation was done in an automated liquid culture system (Bactec MGIT 960, BD) and drug susceptibility testing for first line drugs on solid medium identified no resistance. Genomic DNA was extracted from a single colony sub-culture using the CTAB method [17] and genotyped using spoligotype analysis [18] and PCR for Large Sequence Polymorphism RD702 [3].
The genome of Mycobacterium africanum GM041182 was sequenced to approximately 10-fold shotgun coverage, comprising 92612 end sequences, from pOTW12 (with insert sizes 3–4 kb) and pMAQ1Sac_BstXI (with insert sizes of 4–5 kb and 5–6 kb) genomic shotgun libraries using big-dye terminator chemistry on ABI3730 automated sequencers. End sequences from large insert Fosmid libraries in pCC1Fos with an average insert size of 38–42 kb provided scaffold information with approximately 0.2-fold coverage from 2077 end sequences. A 454 FLX sequencing run provided approximately 10-fold single-end shotgun coverage, comprising 224,378 end sequences from 3kb DNA fragments. In addition an Illumina GAII sequencing lane provided approximately 50-fold single-end shotgun sequence, comprising 6083237 end sequences from 200–300 bp fragments and 37 cycles of sequencing. All repeat regions and gaps were bridged by read-pairs or end-sequenced polymerase chain reaction (PCR) products again sequenced with big dye terminator chemistry on ABI3700 capillary sequencers. The sequence was manipulated to the ‘Finished’ standard [19] and is deposited in EMBL/Genbank/DDBJ under accession number FR878060.
Coding sequences were initially identified by using Glimmer3 [20] and then manually curated using Frameplot [21] and Artemis [22]. All genes were annotated in Artemis using standard criteria [23]. Genome comparisons were visualized in the Artemis comparison tool [24]. Sequence clustering and analysis was performed by using ClustalX 2.0 [25] and MEGA4 [26].
To corroborate the phylogenetic position of the GM041182 isolate within the MTBC we took advantage of the availability of Illumina GAIIx runs for different clinical strains representative of the MTBC [27]. We mapped reads for each strain to the genome of GM041182 using MAQ [28] and single nucleotide polymorphisms were called as described in Comas et al. 2010 [27]. A total of 9,699 positions were identified to vary in at least one strain after exclusion of positions with heterozygous calls or deletions (no coverage positions). A phylogeny was inferred using the number of nucleotide differences between strains as the distance measure and Neighbour-joining as the reconstruction method, and 1,000 bootstrap pseudo-replicates were performed to assess the reliability of the clades. Alternative molecular evolution models and phylogenetic methods were not carried out, as a similar set of strains was extensively analyzed before and no difference in topology was observed between the different approaches [27]. All the phylogenetic analyses were carried out using MEGA5 package [29].
A two step process was carried out to identify mutations that either led to the pseudogenization of previously described genes or generated new potential CDS in lineages of the MTBC. Because the genomes of M. tuberculosis H37Rv and M. bovis AF2122/97 were completed by shotgun sequencing and their annotation manually curated we used them to infer a first list of candidate pseudogenes when compared to M. africanum GM041182. CDS were designated as pseudogenes if they contain in the alignment of homologous positions between the three strains either a frameshift or nonsense mutation, were truncated by a deletion event, or interrupted by a large insertion event. As a second step we focus on the microevolution of the MTBC by assessing whether the events leading to truncated or novel CDS were shared among strains of the different lineages of the complex. We took advantage of the availability of draft shotgun sequences of strains belonging to the different lineages (see Supplementary Table S1 for a list of strains and sources). The polymorphisms were corroborated in other strains by blast searches and manual inspection of the alignments. To assign evolutionary directionality to the changes we used as an outgroup the M. canettii genome (accession number HE572590).
General features of the M. africanum GM041182 genome are unremarkable relative to other members of the M. tuberculosis complex with a typical %G+C content (65.6%) and a genome size (4,389,314 bp) between the usual values for M. bovis (4.34–4.37 Mbp) and M. tuberculosis (4.40–4.42 Mbp). The M. africanum GM041182 genome is also collinear with those of M. bovis and M. tuberculosis and shares the majority of coding sequences (CDSs). Identification of CDSs present in M. bovis and M. tuberculosis but absent from strains of M. africanum has been presented in several publications to date so will not be further detailed here [3], [30], [31]. However, the availability of the M. africanum GM041182 genome sequence has enabled the search for M. africanum-specific sequences and the identification of M. africanum-specific pseudogenes.
We took advantage of the publicly available Illumina sequencing data for 23 strains representative of the MTBC including the sequences of two lab-adapted strains, M. tuberculosis H37Rv and M. bovis Ravenel, as well as the sequence of a strain classified as M. canetti which we used as an outgroup. We mapped the Illumina short-reads to the newly generated M. africanum GM041182 and called for high-confidence polymorphisms. After exclusion of those SNP calls falling in PE/PPE genes and in phage-related regions of the genomes we used an alignment of 9,699 ‘core’ SNP calls (positions in the genome of M. africanum GM041182 where at least one strain has a SNP and no strain has a putative deletion or heterozygous call). The resulting phylogeny (Figure 1) placed M. africanum GM041182 as part of the M. africanum West-African 2 clade (also known as Lineage 6), clustered closely to another strain originally isolated in The Gambia (GM0981). The phylogeny also reflects the great diversity of human M. tuberculosis complex strains found in West African countries with circulating strains from at least three different lineages; the two M. africanum clades and different sub-lineages belonging to the Euro-American lineage (Lineage 4) which are thought to be recently re-introduced in Africa, as the Euro-American lineage is supposed to have originated in the European region [14].
It has been proposed that due to historical migrations and the low-infectious dose during aerosol transmission of human tuberculosis the effective population size of the bacilli could be reduced. This phenomenon could lead to increased genetic drift, limiting the removal of detrimental mutations through natural selection. Relaxed selection can also act during adaptation to a new niche on those genes for which a selective advantage for maintenance is lost; alternatively, the gene function has become disadvantageous in the new niche. Through base-level inspection of the genome sequences we identified pseudogenes in M. africanum GM041182 and verified pseudogene annotation in M. tuberculosis H37Rv and M. bovis AF2122/97. We identified 120 pseudogenes across the three genomes (M. africanum GM041182, M. tuberculosis H37Rv, M. bovis AF2122/97); 20 were in PE-PGRS/PPE family CDSs and in insertion sequence element transposase genes. Both PE-PGRS/PPE family CDSs and insertion sequence elements are known to be associated with intra-genomic recombination and are susceptible to gene disruption [32], [33], [34].
We compared the remaining candidate pseudogenes with available draft genomes from different strains belonging to the MTBC lineages and the genome sequence of a M. canetti strain as outgroup. By using an outgroup we could determine the genotype of the most likely common ancestor of the MTBC for the different candidate pseudogenes and determine which ones were shared by other strains apart from M. africanum GM041182, M. tuberculosis H37Rv, M. bovis AF2122/97 (Figure 2, Supplementary Table S2). We found that some of the pseudogenes identified were strain-specific occurring only in one of these three strains (20 in GM041182, 7 in H37Rv and 9 in M. bovis). More importantly, some of the pseudogene mutations were shared by a large group of strains. For example, 12 were common to the M. africanum West-African clade 2 and 13 common to both M. africanum clades.
In terms of function, the majority of pseudogenes are hypothetical proteins (N = 39), PE-PGRS family proteins and phage-related (20), metabolic enzymes (13) and transcriptional regulators (5) (Figure 3). Many seem to affect systems which are likely to have functional redundancy due to the presence of paralogous or analogous genes or pathways in the genome. For example, there are three pathways for trehalose biosynthesis in mycobacteria [35]; M. africanum and M. bovis each have a trehalose biosynthesis pseudogene but are affected in different genes. M. bovis treY, encoding maltooligosyltrehalose synthase, has a frameshift due to an internal 806bp deletion while M. africanum has a nonsense mutation in a gene (MAF20180) which has been shown to encode a trehalose-phosphate phosphatase. As a component of cell-wall glycolipids, trehalose has been implicated in host tissue damage [35]. Another example is the P450 family of enzymes: there are 21 in the genome of M. africanum GM041182, two of them, MAF35300 and MAF31280, are disrupted in all M. africanum strains while another (MAF22860), which has been shown to be essential for viability in M. tuberculosis, was identified in M. africanum GM041182 [36]. Furthermore, of the 10 loci containing at least one polyketide synthase gene, one is disrupted in GM041182, another is disrupted in all West-African 2 clade strains, and another in West-African 2 and animal strains.
The mycobacterial MmpL-family of proteins have a function in lipid transport and have been shown to contribute to M. tuberculosis intracellular survival [37]. Both clades of M. africanum, as well as animal strains, carry the same nonsense mutation in the 3′ end of the mmpL12 gene (MAF15490) and in M. bovis the mmpL1 gene (MAF04040) has a central frameshift. These mutations may be predicted to impair lipid transport function, although the presence of 12 mmpL paralogues per genome implies some degree of redundancy. Another redundant system affected by mutation in M. bovis is the so-called mammalian cell entry (mce) operons. M. bovis has two adjacent pseudogenes (mce2D and mce2E) in one of the four mce (mammalian cell entry) operons. In M. tuberculosis, deletion of the mce2 operon attenuates the ability to infect mice [38], and deletion of more than one mce operon has a cumulative effect indicating non-redundant roles during infection [39].
The ability to metabolize nitrate to nitrite is thought to be important for M. tuberculosis to persist under anaerobic conditions during dormancy and also appears to have functional redundancy [40]. M. africanum GM041182 has a pseudogene relevant to nitrate metabolism (narX) and all M. africanum and animal strains harbor a narU pseudogene. Although the majority of nitrate reductase activity in vitro is due to narGHJI [41], narX, which encodes a fusion protein equivalent to parts of NarG, NarJ and NarI, has also been shown to have a role in dormancy [40]. NarU is thought to be involved in transport of nitrate into and nitrite out of the bacterial cell though again its function is thought to be secondary to that of the more active narK2 which coincidentally is adjacent to narX.
M. africanum West-African clade 2 strains have frameshift mutations in one of the 17 adenylate cyclase genes in the genome (MAF03880). In M. tuberculosis the MAF03880 orthologue (Rv0386) was recently found to produce a cyclic AMP burst within macrophages that influences cell signaling. Loss of Rv0386 resulted in lower TNF-a induction, decreased immunopathology in animal tissues, and diminished bacterial survival [42].
Three genes with a role in drug efflux have been disrupted; one in M. africanum GM041182 strain (MAF03440), one in M. bovis (orthologue of MAF18990) and one in all the so-called ‘modern’ MTBC strains (MAF23460). The isoniazid inducible gene, iniA (MAF03440), thought to be involved in an efflux pump for two of the 1st line TB drugs, isoniazid and ethambutol [43] has a 5′ nonsense mutation in M. africanum GM041182. In M. tuberculosis an iniA deletion mutant showed increased susceptibility to isoniazid [43], suggesting that M. africanum may be more susceptible to isoniazid than M. tuberculosis. This mutation is however not present in other M. africanum strains of which the genome is available, nor in clinical isolates of the same lineage originating from Burkina Faso and Cote d'Ivoire (data not shown), suggesting that this polymorphism is unique to strain GM041182 . Deletion of the M. smegmatis orthologue of MAF18990 has been shown to result in reduced resistance to ethidium bromide, acriflavine and erythromycin [44]. More interesting is the evolution of the MAF23460 gene. Its homologue in H37Rv is Rv2333c. By inspecting the alignment of both genes a single base pair deletion in the H37Rv leads to a longer product than that observed in M. africanum GM041182 strain (524 residues in M. africanum GM041182 versus 538 residues in M. tuberculosis H37Rv). By comparing with the rest of strains of the complex it becomes clear that the single base deletion occurred in the common ancestor of ‘modern’ lineages representing in this case a possible gain of function rather than a pseudogenetization per se of the ancestral genes. Rv2333c has been shown to be involved in export of spectinomycin and tetracycline and thus contributes to the intrinsic resistance of M. tuberculosis to these antibiotics [45], which may thus be more effective against M. africanum.
Further pseudogenes affect non-redundant systems such as biosynthesis of vitamins B12 (cobalamin) and B6; three genes in a cobalamin biosynthesis operon (MAF20880, MAF20850 and MAF20870) have the same pseudogene allele in both M. africanum and M. bovis while the pdxH (MAF26250) vitamin B6 biosynthesis gene has a central frameshift mutation in M. africanum GM041822 strain. Supplementation of these vitamins may support growth of M. africanum and reduce the growth delay of M. africanum relative to M. tuberculosis.
A notable non-redundant pseudogene is the previously identified orthologue of Rv3879c (MAF38940), part of the RD1 region [46], that was found to be essential for ESAT-6 secretion, but not CFP-10, in M. marinum but not in M. tuberculosis. In a recent study using immunoblots for ESAT-6 and control antigens, we found ESAT-6 secretion to be similar between M. africanum GM041182, M. africanum GM041182 complemented with Rv3879c, and M. tuberculosis H37Rv [47], which does not corroborate the attenuated ESAT-6 response in M. africanum infected people [11]. Given the ESX-1 homology throughout the MTBC it is currently not clear how equal amounts of secreted ESAT-6 between M. africanum GM041182 and M. tuberculosis H37Rv can correlate with the attenuated ESAT-6 response observed in M. africanum infected people. Ongoing immunoepidemiological analyses however suggest that the attenuated ESAT-6 phenotype may cluster with sub-lineages within the M. africanum West African 2 lineage. In addition, we identified a deletion in the upstream regulatory region of Rv3616c whose expression is related with ESAT-6 secretion [48]. This polymorphism in GM041182 is shared with animal strains, in which it is responsible for decreased expression of Rv3616c (Roger Buxton, personal communication). Interestingly, ESAT-6 is highly immunogenic in M. bovis infected cows [49], suggesting that the genetic basis for the attenuated ESAT-6 response observed in M. africanum infected persons is specific to M. africanum.
Finally, the M. tuberculosis orthologue of MAF29630 (Rv2958c) encodes a glycosyl transferase which has been shown, in a co-infection assay, to confer increased resistance to killing by human macrophages [50]. Both M. africanum clades and M. bovis have a single base pair insertion that shortens the gene product (367 residues in M. africanum GM041182 versus 429 residues in M. tuberculosis H37Rv).
On comparative genomics of M. bovis and M. tuberculosis H37Rv, a region unique to M. bovis was designated TB deleted 1 (TbD1) [51]. Subsequent work identified the TbD1 deletion to be shared by “modern” M. tuberculosis lineages, with an intact TbD1 region in other animal strains, M. africanum, and “ancient” M. tuberculosis [15]. Proteins from the TbD1 region were however not immunogenic in an ELISPOT assay that aimed to identify lineage specific immune responses [52].
Comparison of the M. africanum GM041182 genome with reference shotgun sequences for M. bovis (strains BCG Pasteur 1173P2, BCG Tokyo 172, AF2122/97) and M. tuberculosis (strains H37Rv, H37Ra, CDC1551, F11) genomes revealed a single region present in M. africanum GM041182 but deleted in M. bovis and “modern” M. tuberculosis strains. This M. africanum specific locus was designated RD900. The locus is 3,141 bp long and contains a single complete gene (designated maf1 (MAF12860)) and the 3′ end of another (MAF12870); maf1 encodes a putative ATP-binding cassette (ABC) transport protein that has a central ATP-binding domain and six possible membrane-spanning domains in the C-terminal portion. In addition the N-terminal region contains a putative Forkhead associated (FHA) domain that may confer the ability to bind DNA and thereby potentially act as a transcriptional regulator. The LpqY-SugA-SugB-SugC ABC transporter (Rv1235-Rv1238), one of four ABC transporters in M. tuberculosis, has recently been characterized as a recycling system mediating the retrograde transport of the sugar trehalose produced and released by the bacterium [53]. Other ABC transport proteins may mediate efflux of drugs and other compounds (Rv1218c, [54]), with implications for immune responses (Rv1280c-Rv1283c, [55]).
Of the 51 proteins in the Pfam database with the same domain architecture as RD900, only twelve are from outside the order Actinomycetales (seven from Cyanobacteria and five from Chloroflexi); none have been experimentally characterised.
Assuming complete absence of recombination, comparison of the RD900 region of M. africanum GM041182 with all available genomes from the Mycobacterium tuberculosis complex suggests that intact RD900 (M. africanum GM041182) represents the ancestral state of this region and the RD900 region was independently deleted in two lineages: the “modern” M. tuberculosis lineage and a sub-branch of the animal associated lineage leading to M. bovis (Figure 1). We checked this in two ways. First, we aligned complete genomes available for the MTBC with M. africanum GM041182. Secondly, we did local BLAST searches of the region including the two flanking genes (from MAF12850 to MAF12880) (see Figure 4). The very unusual occurrence of independent deletions generating the same functional gene, pknH, is explainable when the flanking regions are considered.
In M. africanum GM041182, maf1 is flanked by similar, co-directional genes, each encoding a protein kinase. The RD900 deletion appears to have been generated by recombination between these flanking genes to form pknH, a protein kinase-encoding gene found in M. bovis and M. tuberculosis (Figure 2a), thus pknH is apparently a composite gene made up by intra-genomic recombination between two homologous, physically close, genes. Accordingly we have designated the flanking genes in M. africanum GM041182 as pknH1 (downstream) and pknH2 (upstream). These flanking CDSs in M. africanum GM041182 have a high level of amino acid identity for the first two-thirds of their length (Figure 2b) followed by divergent sequences from codon 424 (PknH1) and 373 (PknH2), onwards. However, the homologous regions of PknH1 and PknH2 in M. africanum GM041182 have two significant differences. The first is a 21 amino acid sequence insertion/deletion region, present in PknH1 from codons 194 to 214 but absent from PknH2. The second is a substitution region of 53 amino acids in PknH1 (298–350) with low identity to a region of 23 amino acids in PknH2 (277–299). These two differences account for the 50 codon difference in the region of high identity between PknH1 and PknH2. More importantly, the substitution region can be used to demonstrate that the deletion of the pknH gene in M. tuberculosis (called RD900h, Figure 1) was independent of the RD900 deletion found in M. bovis (RD900a, Figure 1). Deletion RD900h generates a composite pknH gene in M. tuberculosis identical to M. africanum pknH1 in the substitution region while the RD900a deletion found in M. bovis generates a composite pknH gene with a substitution region identical to that of M. africanum pknH2. This implies that the RD900h and RD900a deletions had end points before and after the substitution region, respectively. The remaining 3′ portion of pknH for M. tuberculosis and M. bovis has a high degree of similarity to M. africanum pknH1, consistent with the architecture of the region in M. africanum.
The formation of an intact protein kinase in M. tuberculosis and M. bovis, where deletion could easily have resulted in two non-functional gene fragments, could be used to suggest a selective advantage for the reduction in the number of protein kinases caused by the RD900 deletion. However, this conclusion must be tempered by the ease with which this deletion can be generated. We assume that at least one functional pknH gene is required by strains of the M. tuberculosis complex so it is not unexpected that extant strains appear to have a functional gene at this locus.
Broadening the phylogenetic scope of the analysis to include complete genomes from other members of the genus Mycobacterium reveals further patterns of mycobacterial genome evolution associated with RD900. It seems plausible that the ancestral chromosome arrangement for the clade including the M. tuberculosis and M. avium complexes was similar to that seen in M. marinum where an extra 13,772 bp flanks the pknH2 side of RD900 (Figure 5). This flanking region ends with another protein kinase gene which we designate here as pknH3 and the M. tuberculosis complex ancestral genome may have undergone a deletion due to recombination between pknH2 and pknH3.
The M. ulcerans Agy99 genome appears to have undergone the deletion of MURD111 (which removes pknH2) and a long-range rearrangement separating pknH1 and maf1 on the left flank from the right hand flank which actually carries pknH3.
The genomes of M. avium and M. avium subspecies paratuberculosis have simple RD900 deletions akin to those seen in the modern M. tuberculosis and M. bovis lineages; M. leprae TN genome has a similar deletion pattern but with extra DNA loss equivalent to the region from 1412285 to 1423476 in M. africanum GM041182, which results in the loss of nine genes including pknH1, maf1 and pknH2. Curiously, M. smegmatis has a similar deletion pattern to M. leprae suggesting a convergent event in this distant “rapid growing” relative. Also interesting is the high degree of synteny between M. africanum GM041182 and M. kansasii ATCC 12478 with equivalent RD900 arrangement and colinearity extending for at least 6 kilobases on both flanks.
The M. africanum GM041182 genome is, as expected, highly homologous to those of other members of the M. tuberculosis complex, yet contains a unique sequence, RD900, that was independently lost during the evolution of two important lineages within the complex; the “modern” M. tuberculosis group and the lineage leading to M. bovis. In addition, RD900 is variably present in atypical mycobacteria, with evidence for repeated independent deletion events. We can expect to learn more about the phylogenetic position of this deletion as more mycobacterial genomes are sequenced, with this complete M. africanum West African 2 sequence serving as an alternative reference for the mapping of further M. africanum genomes generated using Next Generation Sequencing techniques. Determining the function of the deleted gene, maf1, and the phenotypic consequences of its deletion will require further study but nevertheless this occurrence may provide valuable insight into the evolution of the complex.
The similarity in pseudogene repertoire suggests that M. africanum has a similar evolutionary history to M. bovis and it is tempting to speculate that this may have involved adaptation to a non-human animal host, though it must be noted that for both lineages nearly half of the pseudogenes are unique, so subsequent adaptations may have occurred since their divergence reflecting contemporary niche differences. Thus far, no candidate animal reservoir has been detected for M. africanum. Extensive searches among cattle, sheep, pigs, and goats in the Gambia and neighbouring countries have not identified mycobacterial infection nor disease [16], [56], [57]. Phylogenetically, the Dassie bacillus [58] and the recently identified Mycobacterium mungi [59], are the closest relatives of M. africanum within the M. tuberculosis complex. The Dassie bacillus has been isolated from Dassies, or Rock Hyrax, in South Africa [60], and M. mungi causes disease in troops of banded mongoose in northern Botswana [59]. However, an extensive search for mycobacteria in terrestrial small mammals in Benin, West Africa, did not identify any members of the M. tuberculosis complex [61].
The lack of spread of M. africanum from West Africa to the Americas at the time of the slave trade remains enigmatic. Today, M. africanum is rarely isolated outside of West Africa, typically in first degree immigrants [62]. In a study in Ghana, host polymorphisms were identified with differential protection against M. tuberculosis versus M. africanum in both directions [63], [64], although the degree of selective advantage conferred by these polymorphisms is unclear.
The majority of the pseudogenes detected are only disrupted by a single base mutation, either by an insertion/deletion leading to a frameshift or by substitution leading to a nonsense mutation. As expected for a recently evolved pathogen no further disrupting mutations have been identified in the pseudogenes. Similarly, in a comparison of several MTBC genomes that included GM041182, no mutations were identified in the promoter region of the pseudogenes, supporting the notion that “pseudogenization” in the MTBC is recent [65]. A formal statistical testing of the rate of acquisition of pseudogenes cannot be carried out because of the bias in the discovery of the pseudogenes described to those observed in the comparison of M. africanum GM041182 with the two other strains leading to a phenomenon of pseudogene discovery bias.
However, the high number of pseudogenes in M. africanum and other strains of the M. tuberculosis complex (MTBC) suggest that genome erosion is ongoing. Most likely this reflects several different phenomena that have lead to the downsizing of the MTBC genomes as compared to other free-living or opportunistic Mycobacteria [66], [67]. This could be partly due to its recent evolution as an intracellular pathogen, making some functions that served a free-living lifestyle redundant to the MTBC, which was therefore prone to lose the function due to relaxed selection. At the same time natural selection can act to favour the loss of some genes. These “anti-virulence genes” can be lost because they can be detrimental for the pathogenic lifestyle as has been described for other species [68] and suggested for some known deletion events in the MTBC [69]. Finally, the increased genetic drift imposed by transmission bottlenecks and changes in population size of its host, lead to a weakened effect of natural selection and increased accumulation of functional mutations, many of them detrimental [14]. Further studies, such as complementing the virulence gene Rv0386 in M. africanum and assessing the effect in the appropriate animal model, can assess to which extent its lower progression to disease is explained by these pseudogenes. Moreover, the presence in M. africanum GM041182 of the original version of MAF23460 (Rv2333c) without gain of function suggests that M. africanum (and other ancestral M. tuberculosis complex lineages) lack this functional efflux pump and may be more susceptible to antibiotics, possibly including spectinomycin and tetracycline.
Differentiating these processes by comparative genomics within and outside the complex could provide clues about how the tight relationships between MTBC species and their respective hosts arose in the first place, and how the ongoing erosion described here generates different genetic backgrounds within the complex than can explain some of the differences associated with diversity in disease outcome [70].
|
10.1371/journal.pntd.0000306 | Chlamydia trachomatis ompA Variants in Trachoma: What Do They Tell Us? | Trachoma, caused by Chlamydia trachomatis (Ct), is the leading infectious cause of blindness. Sequence-based analysis of the multiple strains typically present in endemic communities may be informative for epidemiology, transmission, response to treatment, and understanding the host response.
Conjunctival and nasal samples from a Gambian community were evaluated before and 2 months after mass azithromycin treatment. Samples were tested for Ct by Amplicor, with infection load determined by quantitative PCR (qPCR). ompA sequences were determined and their diversity analysed using frequency-based tests of neutrality.
Ninety-five of 1,319 (7.2%) individuals from 14 villages were infected with Ct at baseline. Two genovars (A and B) and 10 distinct ompA genotypes were detected. Two genovar A variants (A1 and A2) accounted for most infections. There was an excess of rare ompA mutations, not sustained in the population. Post-treatment, 76 (5.7%) individuals had Ct infection with only three ompA genotypes present. In 12 of 14 villages, infection had cleared, while in two it increased, probably due to mass migration. Infection qPCR loads associated with infection were significantly greater for A1 than for A2. Seven individuals had concurrent ocular and nasal infection, with divergent genotypes in five.
The number of strains was substantially reduced after mass treatment. One common strain was associated with higher infection loads. Discordant genotypes in concurrent infection may indicate distinct infections at ocular and nasal sites. Population genetic analysis suggests the fleeting appearance of rare multiple ompA variants represents purifying selection rather than escape variants from immune pressure. Genotyping systems accessing extra-ompA variation may be more informative.
| Trachoma is an important cause of blindness resulting from transmission of the bacterium Chlamydia trachomatis. One way to understand better how this infection is transmitted and how the human immune system controls it is to study the strains of bacteria associated with infection. Comparing strains before and after treatment might help us learn if someone has a new infection or the same one as before. Identifying differences between disease-causing strains should help us understand how infection leads to disease and how the human host defences work. We chose to study variation in the chlamydial gene ompA because it determines the protein MOMP, one of the leading candidates for inclusion in a vaccine to prevent trachoma. If immunity to MOMP is important in natural trachoma infections, we would expect to find evidence of this in the way the strains varied. We did not find this, but instead found that two common strains seemed to cause different types of disease. Although their MOMPs were very slightly different, this did not really explain the differences. We conclude that methods of typing strains going beyond the ompA gene will be needed to help us understand the interaction between Chlamydia and its human host.
| Trachoma is the leading infectious cause of blindness worldwide [1]. Repeated infection by Chlamydia trachomatis provokes chronic follicular conjunctivitis (clinically active trachoma), which leads to conjunctival scarring, entropion, trichiasis and ultimately blinding corneal opacification. Trachoma is a major public health problem affecting some of the world's poorest regions. Current estimates indicate 84 million have active trachoma, with 7.6 million visually impaired from trachomatous corneal opacification [2]. The World Health Organization is leading a global effort to control blinding trachoma through the implementation of the SAFE Strategy: Surgery for trichiasis, Antibiotics to reduce the burden of chlamydial infection, and face washing and environmental improvements to limit transmission [3].
Endemic trachoma is caused by 4 of the 19 recognised serovars of C.trachomatis: A, B, Ba and C. Serovars are distinguished from each other on the basis of surface variations in the Major Outer Membrane Protein (MOMP). As the main antigenic target for strain specific humoral immunity to C.trachomatis, MOMP has been considered a vaccine candidate [4]. MOMP is encoded by the ompA gene, which contains four variable segments (VS) interspersed between five conserved segments (CS). Comparative genome sequence analysis has indicated considerable variation in ompA, possibly driven by host immune pressure, and the study of ompA variants may therefore be informative in disease settings [5],[6] Originally serovars were distinguished according to their recognition by panels of patient sera, however the ompA sequence motifs for each serovar have now been well characterised. Organisms assigned to a serovar group on the basis of their ompA sequence are referred to here as genovars.
OmpA genotyping has been used previously to investigate C.trachomatis infections in trachoma endemic populations [7]–[14], usually with the goal of better understanding C.trachomatis transmission. However the analysis of ompA sequence variation is also relevant to the utility of MOMP as a target for chlamydial vaccine development. In genital infections caused by C.trachomatis D-K genovars, evidence that genovar and strain variants associate with clinically important differences in the biology of infection is marginal [15], and has not been described in human ocular infection. Here we analyse ompA genotypic diversity before and two months after mass antibiotic treatment of trachoma in Gambian villages [16],[17].
The Gambian Government/Medical Research Council Joint Ethics Committee (SCC 856) and the London School of Hygiene and Tropical Medicine Ethics Committee approved the study. All subjects, or their guardians, gave written informed consent, or witnessed consent by thumbprint where appropriate.
This study was conducted in 14 trachoma endemic Gambian villages, located within a defined geographical area [16],[17],[18]. The villages were surveyed and a population census was conducted. Individuals normally resident in the study area for at least 6 months of the year were enrolled. At baseline the entire available population was examined for signs of trachoma and classified using the WHO Trachoma Grading System [19]. A swab sample was collected from the upper tarsal conjunctiva of each subject for DNA isolation and kept cool until frozen at −20°C later the same day. Swabs of fresh nasal discharge were collected.
Following baseline clinical assessment, all participants were offered antibiotic treatment. Adults and children over 6 months old were given a single oral dose of azithromycin (20mg/kg up to a maximum of 1g). Infants under 6 months were given tetracycline eye ointment (twice daily, 6 weeks). All villages were examined and treated within a 9 day period [17].
Two months after baseline assessment and antibiotic treatment, participants were re-examined, and conjunctival and nasal discharge samples again collected. Between these two time points, the census was updated weekly, together with records of destination and duration of travel and of the presence of any external visitors.
DNA was extracted from the swabs and tested using the Amplicor CT/NG kit (Roche) [16]. Amplicor extracts from specimens with detectable C.trachomatis DNA were further purified and concentrated using the QIAamp DNA Mini Kit (Qiagen) [16]. Infection load was estimated by quantitative real-time PCR for the chlamydial ompA gene using a previously described method [20].
Sequencing of ompA used primers spanning VS1-4 and sequences were comfirmed by a second sequencing pass. A 1076bp fragment was amplified using primers 87: 5′ - TGA ACC AAG CCT TAT GAT CGA CGG - 3′ and 1163: 5′ - CGG AAT TGT GCA TTT ACG TGA G - 3′. If no amplified product was visible on an agarose gel, nested PCR was performed, with primers 87 (above) and 1059: 5′ - GCA AGA TTT TCT AGA TTT CAT C - 3′ used to amplify a 972bp target sequence. PCR products were purified using the QIAquick PCR purification kit (Qiagen) and sequenced using BigDye Terminator Cycle Sequencing Ready Reaction kit V3.1(Applied Biosystems) with outer primers 97: 5′ - CTT ATG ATC GAC GGA ATT TTC TAT GGG - 3′ and 1047: 5′ - GAT TTT CAT GAT TTC ATC TTG TTC AAC TG - 3′. Sequencing with inner primers 608: 5′ - CTC TCT GGG AAT GTG GGT GT - 3′ and 627: 5′ - ACA CCC ACA TTC CCA GAG AG - 3′ was performed to close sequencing gaps. Sequences were edited and aligned using DNA*DNASTAR 5.07 (DNASTAR), with HAR 13 (NC_007429) as genovar A reference and M33636 for genovar B. Here, a genotype denotes an ompA sequence variant differing from the ompA reference sequence or from another variant by one or more single nucleotide substitutions, and is identified using the letter of its genovar and an arbitrary number.
Data were analysed in Stata 9.0, with differences in loads per genotype examined using a two tailed t-test on logtransformed loads. Sequence alignments were imported into DNAsp4.00 and Tajima's D value calculated [21],[22]. P-values for each D test were calculated using 10,000 coalescent simulations without the presence of recombination to calculate the proportion of D values generated which were greater than the observed D value. D* and F* indices were calculated as further tests of the neutrality of mutations [23].
1319 (83%) of 1595 people enumerated at baseline were examined, sampled and treated. At two-months 1344 (85%) were examined and sampled. The overall prevalence of active trachoma in children <10 years was 16% before and 12% two months after treatment, with marked variations in prevalence between villages [16].
The prevalence of C.trachomatis infection was 7.2% (95/1319) before treatment and 5.7% (76/1344) two months after treatment. Of individuals infected at baseline, 30% were still infected two months after treatment and of those infections detected at two months 36/66 (59%) occurred in subjects uninfected at baseline (Table 1). Most infections (74/76; 97%) detected two months after treatment were in two villages. Almost all residents of these two villages travelled en masse to a religious festival one month after the treatment. This travelling event was very strongly associated with infection at two months [16]. In contrast, in the other 12 study villages all cases of C.trachomatis infection found at baseline had resolved by two months and there were only 2 new cases of infection in previously uninfected individuals.
77/95 (81%) baseline and 64/76 (84%) two-month ocular C.trachomatis samples yielded sequence data. On both occasions sequence data were obtained from all 5 Amplicor-positive nasal specimens. 73 (95%) of the baseline ocular sequences were genovar A and 4 (5%) were genovar B. Overall, ten separate genotypes were identified; 8 genovar A and 2 genovar B. Sequence variation compared to reference strains is shown in Table 2. For most genotypes single nucleotide polymorphisms (SNPs) resulted in amino acid changes in the variable sequence domains of MOMP. Within genovar A baseline sequences, there were eight polymorphic sites, of which five contained singletons (SNPs found only in a single isolate). Tajima's D value for baseline genovar A sequences was −1.06, revealing trend towards an excess of rare mutations, (p = 0.16). This was supported by significantly negative D* and F* indices, indicating an excess of singleton mutations amongst genovar A sequences (−2.59; P = 0.02 and − 2.45; P = 0.02 respectively). Only four genovar B sequences were found, therefore frequency based analyses could not be performed. However, addition of these four sequences to the genovar A sequences for calculation of an overall Tajima's D value revealed a significant excess of rare mutations within the baseline dataset as a whole (D = −1.76 ; p = 0.018).
Genotype frequencies are presented in Table 3. The dominant strain, A2, accounted for 74% of baseline ocular isolates. All other strains, except A1, were detected in only a few individuals. The 14 villages contained 79 family compounds (fenced areas inhabited usually by the members of one extended family). 16 (20%) contained subjects infected at baseline. Seven compounds contained multiple strains; three of which had 3 strains and one 5 different strains. Obvious environmental risk factors which might explain this concentration of diversity were not identified: however the latter compound had an unusually high proportion of its children attending the local primary school (7/25; 28%) compared to (30/773: 4%) in the study area generally.
At two months post-treatment only three strains A1, A2 and A5 were found. The A2 proportion increased to 90%. Rare strains had mostly disappeared. In 23 individuals ocular samples yielded sequence data at both time points. 18 (78%) of these had the same strain at both timepoints: 3 A1, 14 A2 and 1 A5 (the only example of A5 at either timepoint). 5 (22%) showed a change in genotype: from A1 to A2 in three cases, from A3/A4 to A2 in one case each. 34/35 (97%) newly infected individuals at two-months had the A2 genotype.
Infection load data from this population has been previously described [16],[17]. Geometric mean infection loads for strains A1 and A2 were compared by unpaired, two-sided t-tests on logarithmically transformed data. Chlamydial load was significantly higher in A1 infections before mass treatment: geometric mean for A1 5809 copies (95% CI 374–90189) (n = 6) and for A2 92 copies (95% CI 59–144) (n = 14) (p<0.0001). Similarly, after mass treatment geometric mean for A1 was 343 copies (95% CI 42–277663) (n = 3) compared to 115 copies (95% CI 66–202) (n = 19) (p = 0.0021). At both baseline and two-months, subjects infected with A1 were more likely than those infected with A2 to have clinically active disease: baseline: 7/10 vs 6/57 (RR = 6.65, χ 2 = 15.63, p<0.0001); two-months 3/5 vs 7/58 (RR = 4.97 p = 0.025 2-tailed Fisher's Exact Test). We have previously found that infected individuals with clinical signs of trachoma have higher chlamydial loads than those without signs [16],[17]. These analyses are not adjusted for potential clustering by village: however A1 only occurred in one village (village 3).
C trachomatis was detected in nasal samples from 5/58 subjects at baseline, and from 5/54 at two months. In seven subjects ompA sequence was determined in both ocular and nasal samples at the same time point: 5/7 (71%) had different genotypes at the two sites: A1(ocular)/A2 (nasal) in three cases, with A1(ocular)/A3(nasal) and A2 (ocular)/A7(nasal) in one each. Differing genotypes were found in all four individuals in whom baseline ocular and two-month nasal ompA sequence were both determined, and in the two individuals in whom baseline nasal and two-month ocular ompA sequences were both determined.
In this study, 972 bp sequences comprising almost the entire C.trachomatis ompA gene were determined in samples from infected individuals in a trachoma endemic area. Previous trachoma studies have sequenced primarily VS regions: variation in the interspersing ‘conserved’ segments is recognised but not usually examined at the pathogen population level. All variants were confirmed with double pass sequencing methods: dubious calls on the chromatogram were all clarified by resequencing. We discuss the utility of ompA genotyping for determining the existence and nature of selection pressure on the locus, for examining whether variants affect the features of infection or disease, and for distinguishing causes of reemergent infection after treatment.
Ten C.trachomatis genotypes were identified at baseline. Excepting B2, these differed from strains previously sequenced from The Gambia and elsewhere [7]–[11],[13]. Before treatment most (87%) infections were one of two strains (A1 and A2). Six of the minority genovar A strains had SNPs resulting in amino acid changes within variable segment domains. A similar pattern of a few dominant strains with several other strains present at low frequency has been described previously [7],[10],[14]. The variety of strains in this limited geographical area might suggest that new strains are regularly introduced through mixing with other populations or alternatively that the emergence of new variants is promoted by pressure from the human immune response. To test this frequency based analyses of polymorphism were carried out.
Population genetic analysis of baseline genovar A ompA sequences showed negative Tajima's D, Fu and Li's D* and F* statistics, suggesting that in this environment novel genovar A ompA mutations are being eliminated from the population. Despite this, the location of some of the polymorphic sites is intriguing. In genotype A5 the neutralizing antibody epitope which defines serovar A (70DVAGLEK76) is significantly altered (70DEAGLQK76): previously we noted significant alteration in close proximity to this epitope (69(S→R)DVAGLEK76) in strains which subsequently failed to establish themselves in the community [10]. One would expect that novel mutations which allow immune evasion offer the pathogen a selective advantage (at least while these strains remain uncommon), and ought to spread through the pathogen population until they reach intermediary frequencies. The excess of rare mutations observed at baseline therefore does not support the hypothesis that ompA polymorphisms are maintained within this population by immune selection pressure. Instead it implicates either ongoing negative selection (where most mutations are deleterious and removed from the population by purifying selection) or a recent selective sweep (whereby a single haplotype has reached fixation within the population, driving out diversity at the locus). Few studies have applied population genetic methods to analyse selection of C.trachomatis genes, but they have similarly generated little evidence that ompA is under immune selection pressure: both cross sectional studies of genovar A ompA sequences from Tanzania and sequence analysis of genital Ct genovars have found similar evidence of purifying selection in ompA 11,24 These data and the existence of individuals within trachoma endemic communities who are often or repetitively infected with the same ompA genovar lead us to question whether the ompA locus is a target of selective pressure in trachoma populations, and consequently whether targeting MOMP will lead to an effective vaccine.
Strain-specific differences affecting infection or disease manifestations are described in genital chlamydial infection, but not previously in trachoma. On both occasions strain A1 was associated with clinical signs of active trachoma and with higher mean infection loads to a greater extent than A2, but it was less common in the community and so not necessarily a more successful pathogen. The sampling method used here has been shown elsewhere to give adequate yields of host RNA [18], but the infection loads were not standardised, for example against host DNA yield in the sample. In natural infections the number of cells sampled, the proportion of host cells which are infected and the state of the chlamydial developmental cycle within them will all affect the measured load, and the best way to standardise the measurements is not clear. A1 and A2 might amplify differently by PCR, although there was no support for this suggestion in the amplification of standards, and no variation affecting primer binding sites. Differences in sampling, in PCR amplification or in the infection/disease course within the sampled individuals might explain this observation, or alternatively it could result directly or indirectly from variation in ompA .
Three differences exist in the ompA sequence of A1 and A2, of which two cause non-synonymous amino acid substitutions. These might alter the conformation of MOMP or have direct effects on ‘fitness’, transmission or the host response. The G→A mutation at position 304 introduces a cleavage motif for cathepsin-L, which generates of peptide fragments for antigen presentation [25],[26]. Whether peptide fragments of A1 and A2 MOMP are therefore presented differently during the generation of adaptive cellular immunity is unknown. Alternatively, strain differences might be unrelated to ompA itself but reflect linkage between ompA genotype and polymorphism(s) elsewhere on the chlamydial chromosome leading to differences in fitness or metabolic advantage. Trachoma strains may differ in their laboratory properties, and a recent study found differences in in vitro growth rate, interferon-γ sensitivity and virulence in non-human primates [27], attributable to variation affecting 22 open reading frames(ORFs) in addition to ompA. Both clinical differences between strains, and the purifying selection at the ompA locus could result from variation or selection pressure at linked chlamydial ORFs.
Following mass antibiotic treatment there was a modest reduction in the prevalence of infection [17]. Only 3 of the original 10 genotypes were still present. Most (90.5%) of these infections were with A2, and almost all in two villages (1 and 3 in Table 4), in which the prevalence of infection actually increased [17], with strains A1 and A2 continuing to dominate. New infections, 97% with strain A2, were strongly associated with travel to a festival in Senegal, at which over a million people from the region congregated in basic conditions, where the opportunity to acquire ocular C.trachomatis infection was probably considerable. These data suggest that a remarkable re-infecting exposure to strain A2 occurred in the treated subjects during this event. The persistence of the common A1 or A2 strains in 17 individuals in these villages could be due to treatment failure or to reinfection facilitated by the same unusually effective environment for C.trachomatis transmission. Genotyping provides some evidence that antibiotic treatment was not 100% effective, as strain A5 was found twice, but in the same individual both before and after treatment, strongly suggesting primary treatment failure. Nevertheless antibiotic treatment cleared all baseline infections in the other 12 villages [17].
The surprising demonstration of discordant genotypes in concurrent ocular and nasal samples may imply that these two mucosal surfaces function as distinct sites of infection, despite direct communication via the nasolacrimal duct. This could result from differences in the time course of infection or in the route of inoculation. Autoreinfection of the conjunctiva from extraocular sites such as the nasal mucosa has been suggested, however, a study from Tanzania did not support this hypothesis [28]. Here, the limited nasal genotyping data does not provide support significant transmission between eye and nose.
Our study illustrates the use and limitations of ompA sequence data in the molecular epidemiology of C.trachomatis infection. The pattern of ompA sequence diversity remains intriguing and inconsistent with immune selection pressure. Typing systems including other polymorphic loci may lead to better elucidation of key events in ocular C.trachomatis infection. An ongoing extended longitudinal study of C.trachomatis genotypes might better define the population dynamics, and determine implications for the long-term success of mass treatment [14].
|
10.1371/journal.pntd.0003590 | Unprogrammed Deworming in the Kibera Slum, Nairobi: Implications for Control of Soil-Transmitted Helminthiases | Programs for control of soil-transmitted helminth (STH) infections are increasingly evaluating national mass drug administration (MDA) interventions. However, “unprogrammed deworming” (receipt of deworming drugs outside of nationally-run STH control programs) occurs frequently. Failure to account for these activities may compromise evaluations of MDA effectiveness.
We used a cross-sectional study design to evaluate STH infection and unprogrammed deworming among infants (aged 6–11 months), preschool-aged children (PSAC, aged 1–4 years), and school-aged children (SAC, aged 5–14 years) in Kibera, Kenya, an informal settlement not currently receiving nationally-run MDA for STH. STH infection was assessed by triplicate Kato-Katz. We asked heads of households with randomly-selected children about past-year receipt and source(s) of deworming drugs. Local non-governmental organizations (NGOs) and school staff participating in school-based deworming were interviewed to collect information on drug coverage.
Of 679 children (18 infants, 184 PSAC, and 477 SAC) evaluated, 377 (55%) reported receiving at least one unprogrammed deworming treatment during the past year. PSAC primarily received treatments from chemists (48.3%) or healthcare centers (37.7%); SAC most commonly received treatments at school (55.0%). Four NGOs reported past-year deworming activities at 47 of >150 schools attended by children in our study area. Past-year deworming was negatively associated with any-STH infection (34.8% vs 45.4%, p = 0.005). SAC whose most recent deworming medication was sourced from a chemist were more often infected with Trichuris (38.0%) than those who received their most recent treatment from a health center (17.3%) or school (23.1%) (p = 0.05).
Unprogrammed deworming was received by more than half of children in our study area, from multiple sources. Both individual-level treatment and unprogrammed preventive chemotherapy may serve an important public health function, particularly in the absence of programmed deworming; however, they may also lead to an overestimation of programmed MDA effectiveness. A standardized, validated tool is needed to assess unprogrammed deworming.
| In countries with endemic soil-transmitted helminth infections, deworming medications are widely available from multiple sources, including over the counter. However, in many countries, national programs already provide deworming medications in mass drug administrations to primary school students, as part of World Health Organization recommendations. Evaluations of the effectiveness of such medications at reducing worm burden in children is based solely on the national program’s distribution schedules, primarily because little is known about how frequently deworming medications are obtained from other sources. We investigated sources of deworming medications received by children in a Kenyan slum, finding that more than half of school-aged and preschool-aged children received deworming medications outside of a national school-based deworming program. These drugs were received from multiple sources, including chemists, healthcare centers, and at schools, via the efforts of non-governmental organizations. These data strongly indicate a need to collect data on all sources of deworming medications when evaluating the effectiveness of national school-based deworming programs.
| Soil-transmitted helminth (STH) infections affect approximately 2 billion persons worldwide [1], with school-aged children generally having the highest-intensity infections and highest prevalence of infection [2–6]. Improper disposal of human feces contaminated with helminth eggs exposes humans to infection following ingestion of eggs (Trichuris trichura, or whipworm, and Ascaris lumbricoides, or roundworm) or skin contact with larvae that hatch from eggs (Ancystoloma duodenale and Necator americanus, or hookworm). A wide array of physical effects have been attributed to intestinal STH infections, including anemia (primarily from hookworm infection) [7–9], Vitamin A deficiency [10], decreased physical fitness [11], decreased cognitive function [12, 13], decreased growth [12, 14–16], and intestinal obstruction [17]. Morbidity is directly related to infection intensity [18].
Without meaningful improvements in sanitation infrastructure in low-resource settings, elimination of STH infections is likely not feasible. Because of this, and because most morbidity is attributable to high- and moderate-intensity infections, the World Health Organization (WHO) recommends, rather than elimination, reduction of worm burden in individuals [19]. In 2001, WHO set the goal of providing regular preventive deworming chemotherapy to at least 75% of at-risk school-aged children by 2010, and urged endemic countries to develop programs to administer these drugs through schools and primary healthcare systems [20]. In 2012, pharmaceutical companies GlaxoSmithKline and Johnson & Johnson agreed to donate billions of doses of anthelminthic drugs to countries in need [21], enabling the expansion of existing government-sponsored deworming programs, most of which have not yet reached the 75% target.
Many STH-endemic countries, including Kenya, are now planning or actively implementing national school-based deworming programs, provided as mass drug administration (MDA) conducted by Ministries of Health, Ministries of Education, and nongovernmental organization (NGO) partners [22]. Based on STH prevalence mapping and spatial modeling in Kenya [23–25], a phased-in approach to school-based deworming was planned for selected districts in five of the country’s eight provinces, excluding Nairobi, Rift Valley, and North Eastern Provinces [22, 26]; this program is in the process of being implemented [26]. However, smaller-scale deworming programs are also frequently carried out by other in-country partners [27], who may use different regimens for deworming. In addition, deworming drugs are widely available from clinics, drugstores, and other sources. This ‘unprogrammed deworming’—deworming outside of the context of a nationally-administered STH control program—is frequently neither documented nor reported to health officials [27].
There is currently great interest in evaluating the effectiveness of national deworming MDA programs and progress towards the WHO treatment coverage targets for 2020 [28–31]. However, accurate reporting of deworming is required to monitor this progress. The unknown extent and patterns of unprogrammed deworming challenge both the monitoring and evaluation of STH control programs. We describe unprogrammed deworming in Kibera, an urban slum in Nairobi, Kenya. Nairobi is not included in the national deworming program, due to an overall low prevalence of STH infection [32].
This study was conducted in two villages of Kibera, Kenya, during April—June 2012 as part of a larger cross-sectional evaluation of STH infection and nutritional status in children enrolled in the Centers for Disease Control and Prevention’s (CDC) International Emerging Infections Program (IEIP)/Kenya Medical Research Institute (KEMRI) surveillance platform. Details of this study are provided elsewhere [33]. In brief, IEIP enrolls all adults and children with head-of-household consent who have been living in all households in the two-village study area continuously for at least four months. Among households in the IEIP participant registry, approximately 25% were designated as potential sources for enrollment of preschool-aged children (PSAC) (aged 12–59 months) and infants (aged 6–11 months). The other 75% of households were designated as potential sources for enrollment of school-aged children (SAC) (aged 5–14 years); this ensured that two children were not selected from the same household. Households were selected with probability proportional to size from each target group, and one child was chosen randomly from each selected household. Up to three stools were collected from all selected children and tested by Kato-Katz analysis as described previously [33]. Field workers visited households and, using a visual recall aid showing the locally-available deworming medicines, asked parents about whether or not their SAC or PSAC had been dewormed in the past year, and the source of their deworming medication the last time they were treated in the past year. In addition, parents provided information on the school attended by the child. Schools were defined as ‘public’ (under the purview of the Ministry of Education), ‘private’ (privately-owned and operated, following public curricula), or ‘informal’ (not under the purview of the Ministry of Education and with independent curricula).
Because responses suggested that many children were receiving deworming medication at school, administrators and teachers were interviewed at five schools attended by children in the survey. These semi-structured interviews included questions about previous deworming and other health interventions that took place at the school (frequency of interventions, medications administered, and deworming partners). Information from these interviews was used to identify organizations providing deworming treatments in schools in this area. These organizations were subsequently contacted and asked to share any available deworming records from 2010–2012, including school name, school type and location, date of deworming event(s), enrollment figures, and treatment figures. Only 2012 data are included in this report. Drug coverage was estimated by dividing the number of doses of deworming medication reported to have been distributed at each deworming event by the total school enrollment at the time of deworming. Proportions were compared using chi-squared or Fisher’s exact test, where appropriate; missing data were excluded. Data were analyzed in SAS 9.3 (English).
This study was approved by Institutional Review Boards at the Kenya Medical Research Institute (KEMRI) and the U.S. Centers for Disease Control and Prevention (CDC). Written informed parental or guardian consent was required for all participants. Written assent was additionally obtained from all participants aged 13 years or older.
Of the 692 children included in the initial study, 679 had data on past-year deworming, including 18 infants, 184 PSAC, and 477 SAC. Of the 679, 377 (55.5%) reported receiving deworming drug treatments during the previous year. Past-year deworming occurred approximately equally among PSAC (62.0%) and SAC (54.1%) (p = 0.07), a median of three months (range, 0–12 months) before the interview for both groups. Five (27.7%) infants received deworming medication during the previous year. We examined frequency of past-year deworming by one- and two-year age categories; when infants were excluded, there were no differences between these age groups (p = 0.12) (Fig. 1).
Among the 377 children dewormed during the previous year, the most common source of the most recent deworming medication was school (39.5%), a chemist (independently-owned commercial drug kiosks) (27.9%), or a clinic/hospital or health center (26.3%) (Table 1). Among PSAC who were dewormed, the chemist (48.3%) and clinic/hospital or health center (37.7%) were the most common source of deworming medications; most SAC who were dewormed (55.0%) received the drugs at school. The five infants who were dewormed received drugs from a clinic/hospital or health center.
Of the 477 SAC, 442 (92.7%) normally spent the day at school, while 85 (46.2%) of the 184 PSAC spent the day at a nursery school or early childhood learning center. At least 150 different schools were named by respondents as being attended by the SAC included in our study; school name was not recorded for eight SAC. Of the 71 schools for which data were available, 45 (63.4%) were informal, 16 (22.5%) were private, and 10 (14.1%) were public. Of the 393 school-aged children with data on schools attended, 194 (49.4%) attended a public school, 140 (35.6%) attended an informal school, and 59 (15.0%) attended a private school.
Among 383 SAC with data on both school type and deworming, those who attended a public school were more likely to have taken deworming drugs in the past year (from any source) (120/190, 63.2%) than children who attended a private school (29/58, 50.0%)(p = 0.07) or an informal school (61/135, 45.2%) (p = 0.001). Among all SAC who were dewormed, public-school children were non-statistically-significantly more likely to have received their most recent deworming medications at school (78/120, 65.0%) than private-school children (16/29, 55.2%) or children attending an informal school (30/61, 49.2%) (p = 0.11).
Of the 679 children in this analysis, 268 (39.5%) were infected with at least one STH, including Trichuris (n = 176; 25.9%), Ascaris (n = 156; 23.0%), and hookworm (n = 1; <1%). Past-year deworming was associated with reduced frequency of any STH infection (34.8% vs. 45.4%, p = 0.005). When data were limited to the three most common sources of deworming medications (school, chemist, and clinic/hospital/health center), SAC whose most recent deworming medication was sourced from a chemist were more frequently infected with Trichuris (19/50, 38.0%) than those whose most recent deworming medication was obtained from a clinic/hospital/health center (9/52, 17.3%) or school (33/143, 23.1%) (p = 0.05). The presence of other STH infections and any STH infection were not significantly different by treatment source among SAC or PSAC.
Four NGOs were identified as providing school-based deworming medications to children in our study area and were contacted for follow-up. Local government in Nairobi County had partnered with one of these NGOs to deworm children at selected schools (independent from the Kenyan national school-based deworming program); the other three NGOs worked independently from the government and directly with schools. The NGOs reported deworming at 47 schools in Kibera during 2012; 44 (93.6%) schools received deworming treatment once and three were dewormed twice (a total of 50 deworming events). Of the schools with two deworming events during 2012, two had both events administered by the same NGO, and one received deworming treatments from two different NGOs, during February and December 2012. Calculated deworming coverage among enrolled pupils at these events ranged from 72%–130% (median 98.7%). Of 31 schools with data on school type, 19 were informal, eight were public, and four were private.
Unprogrammed deworming, defined as treatment with deworming drugs outside the context of a nationally-administered STH program, is increasingly recognized in many areas that are endemic for STH infections. Our data indicate that more than half of all preschool- and school-aged children in two villages of the informal settlement of Kibera, Kenya received unprogrammed deworming treatments during 2012. These treatments were obtained from a wide variety of sources, which differed by age group: while school-aged children most often obtained treatments in school, frequently through the efforts of NGOs, preschool-aged children more often received treatments from independent suppliers, such as clinics and chemist shops. The median time since last deworming was three months, suggesting that children may be treated several times each year. Because our survey questions were designed to identify only the most recent source of deworming medication, these data likely underestimate the true frequency of unprogrammed deworming events.
Although unprogrammed deworming is rarely reported, it is likely widespread. A recent evaluation of MDA in Bangladesh described high levels of unprogrammed deworming among school-aged children living in an area already receiving programmed school-based deworming [34]. The relatively low cost of such drugs also enhances their accessibility [34, 35]: at the local medical clinic in Kibera, albendazole and mebendazole cost approximately $0.02 USD for a single mebendazole tablet to approximately $0.22 USD for albendazole suspension (O. Mogeni, personal communication). Although individual-level deworming is indicated in specific circumstances, such as for children with palmar pallor in Integrated Management of Childhood Illness (IMCI) programs [36, 37] or in certain maternal health settings, mass unprogrammed deworming, when carried out in an area that already has programmed deworming, may represent wasted treatment in areas with already-limited resources.
In addition to the potential for wasted treatments, unprogrammed deworming may complicate evaluations of effectiveness of nationally-implemented STH control programs. Nonreporting of deworming events to Ministries of Health or to WHO, as previously reported, [27], compromises the monitoring of progress towards WHO-recommended anthelmintic coverage targets [18,20]. Typically, changes in prevalence and intensity of STH infection are assumed to be due to the effectiveness of drugs delivered through programmed MDAs [38–42], occasionally in combination with improvements in sanitation or hygiene [43]. However, high levels of unprogrammed treatment may inflate the apparent effectiveness of programmed MDA. In our study, 55% of children received unprogrammed deworming drugs, a proportion not far below the 75% coverage target set by the WHO [20]. Unprogrammed deworming at this level would almost certainly have an impact on the evaluated effectiveness of a nationally-run MDA program. While the mutual influences of unprogrammed deworming and programmed MDA on each other’s administration frequencies are unknown, in the Bangladesh study, unprogrammed deworming continued at a high rate despite the presence of a national school-based deworming program: 38.7% (95% CI 51.9–64.4) of school-aged children living in a district already receiving two programmed school-based deworming events during 2009 [during which surveyed coverage was 52.3% (95% CI 43.6–61.1%) and 54.3% (95% CI 44.8–63.8%)] reported that they had additionally obtained deworming drugs during that year from other, non-school sources [34]. The lack of coordinated timing of these unprogrammed treatments, even where >1 treatment per year is appropriate (for example, in very high STH-prevalence settings) [18], could also influence their effectiveness.
The effectiveness of unprogrammed deworming on the control and transmission of STH infections is unclear. Nairobi is not considered a high-prevalence region for STH infections [32], and therefore Kibera is not included in the Kenyan national school-based deworming program. Receipt of unprogrammed deworming was associated with reduced STH infection in our study area, and heavy infections were very rare [33]; this may be due in part to the unprogrammed deworming events reported by participants. However, unprogrammed deworming may expose infected individuals to drugs of suboptimal quality [41]. Our data indicated that school-aged children who received their most recent deworming treatment from a chemist were more likely to be infected with Trichuris spp than those who obtained their treatment from a clinic or at school. While this may reflect the use of different (and variably effective) drug brands or drug qualities—for example, levamisole, widely available in this setting, is less effective against Trichuris spp than albendazole or mebendazole [44], and its use in individual but not school-based deworming may have led to these differences—it may also be reflective of the spectrum of reasons individuals take deworming medications. For example, drugs at school may have been administered as MDA, without regard to actual infection status, while drugs from chemists may have been purchased with the intention of treating a known infection. If this is indeed the case, children whose parents purchased drugs from the chemist may be at higher risk for infection. Information about how and why parents obtain deworming medications for their children outside of MDA would be useful in answering these questions.
Beyond the potential individual effects of suboptimal treatment, use of suboptimal drugs on a population level may promote the selection of worms resistant to anthelminthic treatments [41, 45]. Veterinary data demonstrate the rapid spread of benzimidazole resistance in animal populations treated with the drug (reviewed in [45]). While there are few data to suggest that anthelminthic resistance is a widespread problem in humans at present, our inability to accurately determine the true frequency of deworming makes it more difficult to monitor drug efficacy and assess the potential for anthelminthic resistance [46]. Surveys of medications available and doses recommended by chemists would provide information on the occurrence and frequency of suboptimal drug treatment, and allow for opportunities to correct common problems.
As indicated from our data, in addition to NGO- and other-entity-driven deworming, individual treatment may also be common. Although reporting of individual treatment is impractical, individual treatment may prove particularly important in the‘endgame’ of STH control, when STH infections may still be present, but at too low a prevalence to warrant MDA [18]. In such settings, continued suppression of STH transmission will likely require the availability of high-quality and low-cost deworming drugs, health-seeking behavior to access these drugs on an individual basis, and improved sanitation and hygiene. In addition, individual treatment will remain important for infected subpopulations who often are not targeted for routine STH treatment, such as adult men, and, currently, women of child-bearing age, who are at increased risk of hookworm-related anemia [47–49]. Understanding the factors that drive self-treatment can serve to prepare public health officials for the STH’endgame.’
Despite the challenges associated with unprogrammed deworming, it currently fills a need in places where programmed deworming is not occurring. Although Nairobi is not eligible for school-based MDA due to its overall low prevalence of STH infection in school-aged children, Kibera, an impoverished slum inside the city limits, clearly represents a pocket of high prevalence of STH infection, likely due to its very poor water quality and poor sanitation [50]. For school-aged children in this area, the unprogrammed deworming carried out by the local government in partnership with NGOs fills an otherwise unmet need. In addition, for preschool-aged children in Kibera, there is no programmed treatment, including MDA to treat lymphatic filariasis (although IMCI deworming guidelines are implemented in clinical settings) [37]. Unprogrammed treatment partially fills this gap, at no cost to national governments. It is important to encourage further understanding of how, how much, and why unprogrammed treatment occurs, to assist in evaluating its contribution to STH control both inside and outside the context of national STH control programs.
To this end, the STH community must develop validated, comprehensive, and flexible tools to evaluate the frequency, source, and impact of unprogrammed deworming received whenever MDA coverage or effectiveness surveys are implemented. These tools should include questions about the number of times deworming medication was received outside of the context of MDA during the past year, where the drug was sourced from (if known), the type of drug obtained, and, among persons who choose to obtain their own deworming medications, when and why they opted to do so. Sources of drugs provided by NGOs should also be investigated and a sample of locally-available deworming medications tested, to evaluate their effectiveness at delivering the promised results. Different approaches may be required in different settings to confirm the extent of unprogrammed deworming and the source of drugs for different target groups. The variety of drug sources and informants involved in providing and reporting unprogrammed deworming adds a layer of complexity to its accurate assessment. It was not possible in our study to ascertain the veracity of parent-reported deworming; it is possible that other locally-available (and perhaps locally specific) deworming medication sources exist, and setting-specific tools should be developed to assess these during formal evaluations of STH MDA programs, perhaps including components of record review from NGOs and schools as well as individual recall.
In addition to possible recall challenges, potential limitations to our analysis include drug misclassification by respondents: what respondents thought was a deworming drug may have been something else, and what they thought was a drug for a different purpose may have been an anthelminthic. Future studies should verify agreement between the answer to deworming questions without visual aids and the answer when mothers are shown the different deworming drugs and preparations available in the area and asked which one, if any, their child received. In addition, this study took place in an area where a national government-sponsored deworming program was not occurring; were such a program taking place, the frequency of unprogrammed deworming, particularly by NGOs partnering with local government, might have been lower. However, non-school-based receipt of deworming medications accounted for approximately 60% of deworming drugs in this study, suggesting that individual deworming might continue. Partially as a result of the findings in this study, there are now discussions about implementing school-based anthelminthic MDA as part of the national program in parts of Nairobi. Should this occur, a repeat of this evaluation at that time will help shed light on the frequency of unprogrammed treatment while national government-sponsored MDA is occurring. Finally, as an urban African slum, Kibera is unusual in that it serves as a setting for multiple studies and NGO-based interventions. Due to the abundance of NGOs in Kibera, children living there may be more likely to have received NGO-sponsored unprogrammed deworming compared with children living in other urban slums.
In summary, unprogrammed deworming is substantial in Kibera, Kenya. Anecdotal and limited published evidence suggests that unprogrammed deworming is both prevalent and widespread. STH control programs must develop ways to determine the extent, impact, and patterns of unprogrammed deworming to inform guidelines and rational approaches to STH control.
|
10.1371/journal.ppat.1007069 | Langerin+ DCs regulate innate IL-17 production in the oral mucosa during Candida albicans-mediated infection | The opportunistic fungal pathogen Candida albicans frequently causes diseases such as oropharyngeal candidiasis (OPC) in immunocompromised individuals. Although it is well appreciated that the cytokine IL-17 is crucial for protective immunity against OPC, the cellular source and the regulation of this cytokine during infection are still a matter of debate. Here, we directly visualized IL-17 production in the tongue of experimentally infected mice, thereby demonstrating that this key cytokine is expressed by three complementary subsets of CD90+ leukocytes: RAG-dependent αβ and γδ T cells, as well as RAG-independent ILCs. To determine the regulation of IL-17 production at the onset of OPC, we investigated in detail the myeloid compartment of the tongue and found a heterogeneous and dynamic mononuclear phagocyte (MNP) network in the infected tongue that consists of Zbtb46-Langerin- macrophages, Zbtb46+Langerin+ dendritic cells (DCs) and Ly6C+ inflammatory monocytes. Of those, the Langerin+ DC population stands out by its unique capacity to co-produce the cytokines IL-1β, IL-6 and IL-23, all of which promote IL-17 induction in response to C. albicans in the oral mucosa. The critical role of Langerin+ DCs for the innate IL-17 response was confirmed by depletion of this cellular subset in vivo, which compromised IL-17 induction during OPC. In conclusion, our work revealed key regulatory factors and their cellular sources of innate IL-17-dependent antifungal immunity in the oral mucosa.
| IL-17 is a key cytokine for immune homeostasis and host defense in barrier tissues, which can also drive inflammatory diseases and immunopathology under certain conditions. Most studies addressing IL-17-mediated processes focus on the lower gastrointestinal tract, while other barrier tissues such as the oral mucosa remain largely understudied, despite their important role for entry of hazardous microbes. The protective role of IL-17 is particularly relevant for host defense against the fungal pathogen Candida albicans, as evidenced by individuals with genetic defects in the IL-17 pathway. Experiments with mice demonstrated that rapid IL-17 expression is essential for fungal control during oropharyngeal candidiasis. How this is regulated remained largely unclear. In this study, we identified a tripartite population of innate lymphocytic cells as the bona fide cellular source of IL-17 in the oral cavity. At the molecular level, we found IL-6 and IL-1 to act synergistically and complementary to IL-23 for instruction of IL-17 in response to infection. In search for the cellular source(s) of these IL-17-inducing factors we identified Langerin+ DCs as key players in the coordination of C. albicans-induced innate IL-17 production. Together, this greatly advances our understanding of IL-17 regulatory mechanisms in antifungal immunity in the oral mucosa.
| As part of the upper gastrointestinal tract, the oral cavity is colonized by microbes and constitutes an important entry point for hazardous pathogens. However, despite the relevance of the oral mucosa as a first site of interaction between microbes and the host, it remains little studied and its cellular composition is not well characterized. Oropharyngeal candidiasis (OPC) is a common infection of the oral cavity mediated by the opportunistic fungal pathogen Candida albicans in immunocompromised individuals [1]. It frequently develops as a consequence of impaired immune function due to administration of steroids and other immunosuppressant agents, or because of underlying diseases such as AIDS or primary immunodeficiencies [1]. The recent study of hereditary factors predisposing to OPC and other forms of mucocutaneous candidiasis determined the relevance of the interleukin-17 (IL-17) pathway as a key mechanism for protective immunity against this disease. Genes directly associated with disease include those encoding the IL-17 receptor subunits IL-17RA [2] and IL-17RC [3], the signaling adaptor Act1 (also referred to as CIKS or TRAF3IP2) [4] and the cytokine family member IL-17F [2], but also genes encoding transcription factors involved in the regulation of IL-17 production, such as STAT1 [5,6], STAT3 [7–9] and RORγt [10]. Work on experimental mouse models further confirmed the important role of IL-17, in particular IL-17A and IL-17F, in antifungal defense [11–13].
IL-17 is thought to act by promoting the antimicrobial function and the epithelial integrity of barrier tissues [14,15]. Although there is little doubt about the relevance of IL-17 in host-defense against C. albicans, the tissue-specific regulation of IL-17 production during candidiasis remains not well understood. Th17 cells are the major source of IL-17A and IL-17F in response to C. albicans [16–19]. The rapid induction of IL-17A and IL-17F within 24 hours post-infection in experimentally infected mice suggested that innate cells also participate in cytokine production in the infected mucosa. Indeed, previous work from our group demonstrated that Rag1-/- animals, lacking T and B cells, control the fungus and recover from infection within a week [13]. Genetic deletion or antibody-mediated depletion of CD90+ or IL-2 receptor gamma chain-dependent cells in Rag1-/- mice however rendered the animals susceptible to OPC to a degree similar to IL-17RA-deficient mice [13]. Hence, our oberservations strongly argued for the involvement of a RAG-independent cellular source of IL-17. This result was challenged by a study using an IL-17A reporter mouse strain suggesting that T cells produce IL-17 within 24 hours of primary infection [20]. Here we employed a flow cytometry approach to directly visualize IL-17A protein expression in the tongue of infected mice and identified three separate populations of IL-17-producing CD90+ leukocytes that act in an at least partially redundant manner during acute OPC. Moreover, we dissected and tested the functional relevance of different mononuclear phagocytes and IL-17 instructive cytokines in the infected tissue and thereby revealed the key determinants that orchestrate the innate IL-17 response against C. albicans in the oral mucosa.
We set out to characterize cells with the potential to produce IL-17A and IL-17F in response to C. albicans at the site of infection. To do so, we analyzed tongue leukocytes from RorcCreR26ReYFP reporter mice for the expression of eYFP as a reporter for the lymphocyte-associated transcription factor RORγt. We detected a distinct population of eYFP+ cells that uniformly co-expressed CD90+ in both naïve and infected tongues and comprised αβ T cells, γδ T cells and TCR- ILCs (Fig 1A). Innate IL-17 production in the murine tongue was so far never demonstrated directly and at the single cell level. We therefore established a protocol to visualize IL-17A and IL-17F protein in the C. albicans-infected tongue by combining in vivo Brefeldin A administration and intracellular cytokine staining (S1A Fig). By doing so we detected a well-defined population of CD90+IL-17A+ cells in the tongue of infected but not naïve wild type (WT) mice (Fig 1B). Because Il17a and Il17f transcript levels are maximal on day 1 post-infection with C. albicans strain SC5314 [13], we focused our analysis on this time point. Reminiscent of our analysis of RorcCreR26ReYFP mice, the CD90+IL-17A+ population comprised three subsets: αβ T cells, γδ T cells and TCR- ILCs (Fig 1B). The specificity of our IL-17A staining in CD90+ cells was confirmed by applying the same experimental conditions to Il17af-/- mice, which served as a biological negative control (S1B Fig). In WT mice, the majority of IL-17A+ cells co-produced IL-17F (S1C Fig). NKT cells were not found to contribute significantly to IL-17A production in the infected tongue, as only very few IL-17A+ cells could be stained with CD1d tetramers (S1D Fig). Quantification of the CD90+ and CD90+IL-17A+ subpopulations revealed an expansion of all subsets within the first 24 hours of infection (Fig 1C). Consistent with this, all IL-17A+ cells stained positive for the marker Ki67 (Fig 1D and S2 Fig), suggesting in situ proliferation of the three IL-17A-producing cellular subsets. To characterize the IL-17A-producing cells in the tongue of OPC-infected mice in more detail, we analyzed phenotypic and activation markers on their surface. Consistent with their TCR expression, αβ and γδ T cells, but not TCR- ILCs, co-expressed CD3 (Fig 1D and S2 Fig). All three IL-17A+ subsets displayed an activated phenotype as evidenced by their high expression of CD44 and partial expression of CD69. None of the IL-17A-expressing cellular subsets expressed CD122, NCR1, CCR6 or MHCII (S2 Fig).
To further define the ILC compartment, we examined RAG1-deficient mice. RorcCreR26ReYFP fate reporter mice, crossed to a RAG1-deficient background, harbored a clear population of eYFP+ cells in the naïve tongue that were uniformly CD90+ (Fig 1E). Infection of (reporter-less) Rag1-/- mice revealed, as expected, an overall reduction of total CD90+ and CD90+IL-17A+ cells due to the absence of TCRβ+ and TCRγδ+ cells. Yet, the number of IL-17A-producing cells was still significantly increased in infected as compared to naïve WT mice, which was attributed to the ILC subset (Fig 1F and 1G). Overall, our data demonstrate that direct visualization of IL-17A protein production by intracellular staining ex vivo revealed the involvement of three distinct CD90+ leukocyte subsets during OPC.
IL-23 is a key regulator of IL-17 immunity. To investigate the impact of IL-23 on IL-17A production by the three distinct cellular subsets during OPC, we infected Il23a-/- animals in comparison to WT controls and assessed the changes in absolute numbers of IL-17A producing CD90+ cells and the corresponding TCRβ+, TCRγδ+ and ILC subsets in the tongue on day 1 post infection. While the CD90+ populations overall remained unchanged in Il23a-/- mice in comparison to WT controls, the number of IL-17A-positive cells was significantly reduced (Fig 2A and 2B). This is consistent with previous studies analyzing the dependence of overall IL-17A and IL-17F expression on IL-23 in the infected organ [13] and the impact of IL-23 on fungal control [11,13]. Of the three identfitied IL-17A-producing cellular subsets, the TCRβ+ subset was most strongly affected, but also the TCRγδ+ and ILC populations showed a clear trend towards reduced IL-17A production in IL23-deficient mice compared to WT controls.
Instruction of IL-17A production by the adaptive immune system is driven by IL-1β and IL-6 in addition to IL-23 [21–23]. The impact of these cytokines on the regulation of IL-17A production by innate immune cells during OPC was not investigated in detail so far. Therefore, we set out to analyze mice with a nonfunctional IL-1 and/or IL-6 pathway. While genetic deletion of the IL-1 receptor or antibody-mediated neutralization of IL-6 alone had no measurable effect on IL-17A induction (S3 Fig), the combination of both resulted in a drop in CD90+IL-17A+ cells in the tongue of infected mice in comparison to controls (Fig 2C and 2D). Again, the effect was most pronounced for the TCRβ+ subset. Together, these data demonstrate that IL-23, IL-1 and IL-6 play a critical role for rapid IL-17A induction at the onset of OPC, whereby IL-6 and IL-1 act in a redundant manner.
Mononuclear phagocytes (MNPs) are a prominent source of IL-1β, IL-6 and/or IL-23 in diverse infectious settings and they may thus represent important players in the regulation of innate IL-17A production during acute OPC. However, the CD11c+MHCII+ MNP network in the tongue of mice remains poorly characterized. We thus set out to dissect this cellular compartment in detail. This required adaptation of our protocol for mouse tongue preparation to assure that the tissue-resident cells were liberated from the dense epithelial network. Based on CD11b and Langerin (CD207) expression, we identified four distinct subsets within the CD11c+MHCII+ population in the naïve tongue (Fig 3A). Unexpectedly, these cells appeared to be clearly distinct from the established DC populations e.g. in the spleen that comprise CD11b-CD24+ conventional DCs group 1 (cDC1s), CD11b+CD24- cDC2s [24] and Langerin+ DCs [25] (Fig 3A). Further characterization of the CD11c+MHCII+ MNPs in the naïve tongue showed that all four MNP subsets expressed low levels of XCR1 and Ly6C and high levels of Sirpα, whereby the highest expression of Sirpα was found in the CD11bhi subsets (Fig 3B). The double-negative subset was strongly positive for F4/80, while the CD11bhiLangerin- subset was heterogenous for most markers analyzed. Both Langerin+ subsets displayed a similar phenotype with high expression of CD24, EpCam and CD64 (Fig 3B). Consistent with a previous study on Langerhans cells in the oral mucosa [26], we found tongue Langerin+ cells to be radiosensitive (S4A and S4B Fig). Moreover, immunofluorescent staining of tissue sections and epithelial sheets from infected animals revealed an intraepithelial localization of the Langerin+/MHCII+ cells in the tongue (S4C Fig).
As the phenotypical analysis did not allow categorizing the tongue CD11c+MHCII+ MNP populations into DCs and/or macrophages, we examined their expression of Zbtb46, a transcription factor selectively expressed by cDCs but no other myeloid and lymphoid cells [27]. The analysis of Zbtb46GFP reporter mice in combination with EpCam expression as a surrogate for Langerin on the cell surface demonstrated that EpCam+ but not EpCam- CD11c+MHCII+ cells were bona fide cDCs, whereas the EpCam- subsets rather represented macrophages in the naïve mouse tongue (Fig 3C).
We then assessed the dependence of the tongue CD11c+MHCII+ MNPs on the transcription factors Batf3 and Klf4, which in other organs are lineage defining for cDC1s and a subset of cDC2s, respectivey [28,29]. Hematopoietic deletion of Klf4 had no impact on the tongue CD11c+MHCII+ MNPs (Fig 3D and 3E). In contrast, Batf3 deficiency resulted in a specific loss of the CD11blowLangerin- subset in the tongue, which was surprising since those cells were Zbtb46-negative (Fig 3C). For comparison, spleen samples were analyzed in parallel confirming that CD11b-CD103+ cDC1s were clearly Batf3-dependent whereas CD11b+CD103- cDCs were partially Klf4-dependent (Fig 3D and 3E). Together, our data revealed that the CD11c+MHCII+ compartment in the naïve tongue is unique and heterogenous with its Zbtb46+CD11b+/lowLangerin+ DCs and Zbtb46-CD11b+/lowLangerin- macrophages that display an unprecedented onthological signature.
Finally, we investigated the dynamics of these newly defined populations of tongue MNPs during OPC. The presence of C. albicans led to a rapid relocalization and clustering of intraepithelial MNPs in proximity of fungal hyphae (S5 Fig). The population of Langerin+ DCs was reduced in size on day 1 post-infection when compared to the naïve state (Fig 4A and 4B). The lack of Annexin-V+ staining suggested that the Langerin+ DCs were rather emigrating from the tongue epithelium than undergoing apoptosis (data not shown). The loss of Langerin+ cells was accompanied by an increase in Langerin-CD11c+MHCII+ MNPs in the tongue, which co-expressed Ly6C and CCR2, indicating that they were derived from Ly6Chigh inflammatory monocytes (Fig 4C).
In conclusion, these data show that the tongue bears a complex network of MNPs with important tissue-specific pecularities and that acute OPC leads to dynamic changes within this network including the decline in Langerin+ DCs and the infiltration and differentiation of CD11b+Ly6Chigh inflammatory monocytes.
Our observation that IL-1β and IL-6 are both important for induction of IL-17 during OPC prompted us to identify the cellular source(s) of these cytokines. We first assessed cytokine production by the two major myeloid cell populations in the infected tongue, namely CD11b+CD11c+MHCII+ MNPs (P1) and CD11b+CD11c-MHCII- cells (P2), the latter comprise pre-dominantly neutrophils. Both populations contributed to the overall IL-1β production (S6A Fig), while—consistent with previous data [30]—no IL-1β was produced by the non-hematopoietic compartment (S6B Fig). Intracellular staining of pro-IL-1β served to identify IL-1β-producing cells. The specificity of the intracellular staining for pro-IL-1β was confirmed by applying the same staining panel to cells obtained from infected Il1ab-/- mice (S6C Fig). In addition to IL-1β, the IL-1 receptor is also engaged by IL-1α, which is released by oral keratinocytes during OPC [30]. IL-6 on the other hand was produced by both, the hematopoietic and non-hematopoietic compartment in response to infection. Among the CD45+ cells, CD11b+CD11c+MHCII+ MNPs provided the main source of IL-6 with very little contribution of CD11b+CD11c-MHCII- neutrophils (S6A, S6B and S6D Fig). Overall, our data defined the presence of multiple hematopoietic and non-hematopoietic cellular compartments providing IL-1β and IL-6 at the onset of OPC.
Next, we aimed at investigating the contribution of individual MNP subset(s) to the overall production of IL-1β and IL-6 during acute OPC. Therefore, we combined our staining panel for the four MNP subsets (based on Langerin and CD11b expression) with intracellular cytokine staining for pro-IL-1β and IL-6 (S7A Fig). This revealed that pro-IL-1β and IL-6 were produced by three out of the four MNP subsets, namely Langerin+CD11b+ and Langerin+CD11b- DCs as well as Langerin-CD11b+ MNPs (predominantly macrophages), while CD11b-Langerin- MNPs did not produce either of the two cytokines (Fig 5A, 5B, 5D and 5E). Of all MNP subsets, the Langerin+ subsets displayed the highest proportion of IL-1β+ and IL-6+ cells (Fig 5C and 5F). Analyzing cytokine production by Ly6C+ inflammatory monocytes showed that these cells contributed only little to pro-IL-1β and not to IL-6 secretion (S7B and S7C Fig). In summary, our data show that tissue-resident MNPs, especially Langerin+ DCs, provide the IL-17A-inducing factors IL-1β and IL-6 during the onset of OPC.
Besides IL-1β and IL-6, IL-23 also contributes critically to innate IL-17A induction during OPC (Fig 2A and 2B). Determining IL-23 production at a cellular level remains difficult due to the lack of a functional antibody for detection of the specific cytokine subunit IL-23p19 by flow cytometry. To overcome this limitation, we FACS-sorted four major cell populations from infected tongues and quantified the expression of IL23a transcripts (coding for IL-23p19) by these cells by RT qPCR. For technical reasons we used EpCam instead of Langerin in the flow cytometry panel in this experiment. Our approach led to the identification of EpCam+CD11c+MHCII+ MNPs (corresponding to Langerin+ DCs) and CD11b+Ly6C+CD11c-MHCII- cells (predominantly neutrophils) as the main sources of IL23p19, while EpCam-CD11c+MHCII+ MNPs (mostly macrophages) and CD45-EpCam+ epithelial cells contributed only marginally (Fig 6A and 6B). The specificity of the RT qPCR analysis for IL-23p19 was verified by analyzing IL-23p19 expression in the four cell populations sorted from infected Il23a-/- mice (S8A Fig). The adequacy of using EpCam instead of Langerin was confirmed by analyzing Cd207 transcript expression (coding for Langerin) in the sorted cell populations (Fig 6C). Overall our data demonstrate that EpCam+/Langerin+ DCs as well as neutrophils provide IL-23p19 at the onset of OPC.
We also examined expression of Il1b and Il6 transcripts by the four sorted cell populations and could thereby confirm the results obtained by flow cytometry (Fig 5, S6 Fig), namely that IL-1β was expressed predominantly by EpCam+ MNPs, EpCam- MNPs and neutrophils, and that IL-6 was expressed by EpCam+ DCs and to a lesser degree by EpCam- macrophages (S8B and S8C Fig). Together, tissue-resident Langerin+/EpCam+ DCs thus stand out by their efficient expression of all three cytokines involved in innate IL-17A induction.
After identification of multiple and partially overlapping myeloid cell subsets producing IL-17A-inducing cytokines, we aimed at evaluating their functional relevance for IL-17A production during infection. First, we targeted neutrophils, which represent the majority of the tongue-infiltrating leukocytes during actue OPC (S9A Fig) [15,30] and contribute to the overall IL-1β and IL-23p19 production in the infected tongue (Fig 6B, S6A Fig). We therefore treated mice with anti-Ly6G and anti-G-CSF antibodies to deplete neutrophils prior and during infection. This had only a limited impact on IL-17A production by the three CD90+ celluar subsets that did not reach statistical significance (S9B Fig). However, the actual contribution of neutrophils to the IL-17 response may likely be underestimated due to the difficulty to fully deplete these rapidly infiltrating cells from the infected tissue despite the administration of two distinct neutralizing/blocking antibodies [15,30]. Moreover, additional cellular sources, such as EpCam+ DCs, provide IL-17A-inducing cytokines that may compensate for the compromised neutrophil response.
We also assessed the contribution of inflammatory monocytes to innate IL-17A induction during OPC, as these cells have previously been shown to promote the adaptive Th17 response to C. albicans in the oral mucosa [19]. However, infected Ccr2-/- mice, in which CD11b+Ly6Chigh monocytes recruitment to the infected tongue was strongly impaired (S9C and S9D Fig), displayed no defect in IL-17A production on day 1 post-infection by any of the three CD90+ cell subsets (TCRβ+, TCRγδ+, ILCs) compared to WT control mice (S9E Fig). Albeit not fully conclusive due to the incomplete monocyte depletion in Ccr2-/- mice, these data are in line with our observation that Ly6C+ inflammatory monocytes contribute only weakly to the overall IL-1β production and do not produce IL-6 (S7C and S7E Fig), indicating that in contrast to the later phase of OPC, inflammatory monocytes are dispensible for the early IL-17A response during acute OPC. Finally, we examined the role of the newly identified tongue MNP subsets in the initiation of innate IL-17A production during OPC. For this, we analyzed different genetic and antibody-depletion models with selective MNP defects. The lack of the CD11blowLangerin- MNP population in Batf3-/- mice (Fig 3D and 3E) had no impact on the number of IL-17A-producing CD90+ leukocytes during acute OPC (data not shown), which is in line with a previous publication showing that Batf3-deficiency in mice is dispensable for IL-17-mediated antifungal defense during OPC [31]. The same was true for IL-17A production in VAVCreKLF4fl/fl animals (data not shown), which was not surprising given that these mice did not display any changes in the tongue MNP network (Fig 3D and 3E). Homeostasis of tissue-resident MNPs, including macrophages and Langerin+ DCs, depends on CSF1R signaling and consequently these cells can be depleted in vivo upon antibody-mediated blockade of the CSF1R [32,33]. We thus aimed at depleting tongue CD11c+MHCII+ MNPs by administering an anti-CSF1R blocking antibody prior to OPC onset. This resulted in the selective loss of Langerin+CD11c+MHCII+ DCs, while Langerin-CD11c+MHCII+ MNPs, Ly6C+CCR2+ monocytes or neutrophils were not significantly affected by the treatment (Fig 7A and 7B). Importantly, the loss of Langerin+ DCs in the tongue was accompanied by a significant reduction in IL-17A-production by all three CD90+ subsets if compared to non-treated WT control mice (Fig 7C and 7D). These results were in line with our findings of Langerin+ DCs being the only tissue-resident MNP population in the tongue producing all three IL-17A-inducing cytokines IL-1β, IL-6 and IL-23 (Figs 5 and 6) and demonstrate that Langerin+ DCs are key players in the IL-17 response during acute OPC.
The cytokine IL-17 has gained much attention due to its association with auto-inflammatory disorders such as psoriasis, psoriatic arthritis or ankylosing spondylitis [34,35] and the remarkable success in treating these diseases with IL-17 targeting reagents [36,37]. However, IL-17 is also crucial for mediating immune homeostasis in barrier tissues that are continuously exposed to microbes. Over the past decade, the lower gastrointestinal tract has been intensively studied in this context [38–40]. IL-17 production in the gut is determined by the microbiota [41–43] and the Gram positive segmented filamentous bacterium SFB has been identified as a major driver of the response [44,45]. The regulation of IL-17 in other mucosal tissues than the gut is less well studied. Only recently, it was shown that the homeostatic Th17 response in the gingiva is independent of the microbiota but rather a consequence of constant tissue damage elicited by mastication [46].
IL-17 immunity plays an essential role in host defense against opportunistic infections with the fungal pathogen C. albicans as evidenced by primary immunodeficiency patients with defects in genes of the IL-17 pathway that suffer from chronic mucocutaneous candidiasis. One of the tissues most frequently affected by C. albicans is the oral cavity. Experimental OPC in mice triggers prominent IL-17 induction and fungal clearance depends on the rapid induction of IL-17 in the infected mucosa during the onset of infection [13]. Here, we demonstrated that upon OPC in mice, IL-17 is produced by a tripartite population of CD90+ leukocytes in the tongue, comprising αβ T cells, γδ T cells and ILCs. Production of innate IL-17 is under the control of CSF1-dependent Langerin+ DCs, which are the major source of the IL-17-inducing cytokines IL-1β, IL-6 and IL-23 in the oral mucosa.
Earlier work from our group already proposed ILCs to provide IL-17A at the onset of OPC in experimentally infected mice [13]. This observation was based on the rapid kinetics of IL-17 induction and the fact that Rag1-/- but not Rag1-/-Il2rg-/- or anti-CD90-treated Rag1-/- mice were protected from infection due to their capacity of upregulating IL-17 in the infected mucosa. Furthermore, MHC-II-deficiency did not impair IL-17A expression in the infected mucosa indicating that conventional MHCII-mediated antigen presentation is dispensable for IL-17A production during OPC [13]. However, our findings were challenged by the work from Conti et al., who reported IL-17 expression by RAG-dependent lymphocytes, foremost αβ and γδ T cells, during OPC [20]. These seemingly contradictory results have arisen not least because of indirect assessments of IL-17A production during infection by either measuring Il17a and Il17f transcripts in crude tongue extracts [13] or monitoring IL-17A promoter activity in Il17aeYFP fate reporter mice [20]. Here, we reconcile the discrepancy and demonstrate the existence of three separate and complementary IL-17-producing cell types by direct visualization of IL-17A and IL-17F cytokines in the infected tongue. These three cellular subsets act in an at least partially redundant manner: selective lack of αβ or γδ T cells does not affect fungal control and only deletion of all three subsets phenocopies the high susceptibility of IL-17RA or IL-17RC-deficient mice to OPC [11,13,15], underlining the robustness of the IL-17 response to the fungus.
Based on their independence of RAG and their expression of RORγt, the TCR-negative IL-17 producers are part of the family of group 3 ILCs, although they lack expression of CCR6, NCR1 and MHCII, which are characteristic of at least some ILC3s [47,48]. Visualization of IL-17A and IL-17F protein expression by flow cytometry not only allowed us to define the sources of IL-17 but also offered the opportunity to investigate the regulatory mechanisms of IL-17 production during acute OPC. We confirmed the critical role of IL-23 for innate IL-17 induction, an observation that is in line with previous work demonstrating that Il23a-/- mice phenocopy Il17ra-/- and Il17rc-/- mice in their inability to clear C. albicans [11,13]. However, it also became evident that the defect in IL-17 production in response to OPC was not complete in Il23a-/- animals, indicating that IL-23 may have (an) additional IL-17-independent function(s) in antifungal defense. Moreover, IL-23 may share redundancy with other IL-17-inducing cytokines. The dependence on IL-23 was not equally pronounced for all IL-17-producing subsets, suggesting that the relative contribution of different cytokines to IL-17 induction may differ for different cellular sources. In addition, the technical limitations of the experimental system due to the small cell numbers recoverable from the tongue may also mask clearer associations.
Both, IL-1 and IL-6 have been implicated in the regulation of IL-17 immunity in the gut [49,50] and IL-6 was also implicated in Th17 polarization in the gingiva during steady state [46]. Here, we report that these cytokines also trigger innate IL-17 production in response to C. albicans in the oral mucosa. The impact of IL-1 and IL-6 on innate IL-17 production was somewhat overlooked before when each pathway was examined in isolation [13,20]. We now demonstrate that only concurrent blockade of both IL-1 and IL-6 pathways resulted in a significant drop of IL-17 induction in infected mice.
Tongue-resident MNP populations have not been characterized in detail in mice. Our phenotypic and transcription factor analysis of CD11c+MHCII+ MNPs revealed the presence of two heterogeneous populations of Zbtb46-Langerin- macrophages and Zbtb46+Langerin+ DCs in the naïve tongue. Whether the Langerin+ DCs are bone fide Langerhans cells or represent the mucosal analogue of Langerin+ dermal DCs remains to be determined [51]. In terms of their phenotype and in situ localization they closely resemble Langerhans cells in the gingiva and the buccal mucosa [26]. Moreover, their independence of Batf3 further supports that they are indeed Langerhans cells [52]. Langerhans cells in the oral mucosa have been shown to differ from their skin counterparts in terms of their ontogeny, as they are derived from circulating radiosensitive precursors instead of radio-resistant embryonic precursors [26] and we confirmed this to be the case in the tongue. That MNPs in the tongue are different from MNPs in other tissues is also exemplified by the Langerin- MNP populations. We found the CD11blowLangerin- subset to depend on Batf3, a lineage-determining transcription factor for cDC1s, but at the same time to lack Zbtb46 expression, thus calling into question whether it constitutes a Batf3-dependent subset of macrophages or a special population of Zbtb46-independet DCs. While future work will be needed to fully clarify the ontogeny of all four oral MNP subsets, our detailed dissection of the MNP network in the murine tongue sets the stage for interrogating the contribution of individual populations to immune homeostasis and defense.
Among the accessory cells supplying IL-1 and IL-6 during acute OPC, we identified non-hematopoietic cells, which have been shown before to release IL-6 and IL-1α in the oral mucosa of mice [30]. In addition, we found (several) complementary myeloid cell populations serving as hematopoietic sources of IL-1, IL-6 and/or IL-23 during infection. Neutrophils that rapidly infiltrate to the site of infection were found to produce IL-1β and IL-23 during OPC. While neutrophils can themselves serve as a source of IL-17A under certain circumstances [53,54], we have no evidence for IL-17 production by neutrophils during OPC. Conversely, neutrophils may support IL-17 production by CD90+ leukocytes as they secrete IL-17-promoting cytokines during infection in the oral cavity. Macrophages and monocytes (defined as Langerin- MNPs) contributed to the production of IL-1β and IL-6, but did not produce IL-23. Langerin+ MNPs however were the only cellular subset producing all three IL-17-inducing factors. Together with their strategic location in the outermost layer of the epithelium at the onset of the infection, prior to the arrival of infiltrating inflammatory cells, this unique property predisposes Langerin+ DCs as the primary IL-17-inducing cellular subset. Targeting MNPs via antibody-mediated blockade of CSF1R confirmed the crucial role of Langerin+ cells as the primary cellular determinant for IL-17 induction at the onset of OPC. Langerin+ DCs have been implicated in IL-17-mediated immunity against C. albicans previously: during experimental cutaneous candidiasis, the constitutive absence of Langerhans cells in huLangerin-DTA mice [55] resulted in a drastric reduction in Th17 differentiation in skin-draining lymph nodes [56]. Reminiscent of experimental OPC, infection of mice with C. albicans via the epicutaneous route also triggered an immediate local IL-17 response within one day of infection, which is dominated by γδ T cells [12]. The release of IL-17 by γδ T cells in the skin during epicutaneous infection was not dependent on Langerhans cells, but rather on CD301b+ dermal DCs [12]. In the oral mucosa however and in contrast to the skin, adaptive immunity against C. albicans does not rely on Langerin+ cells, but instead depends on CCR2-dependent inflammatory DCs and other Flt3-dependent migratory DCs [19]. Therefore, the contribution of specific MNP subsets to the regulation of IL-17 production in different epithelial tissues and in different phases during infection emphasizes the dynamic and tissue-specific regulation of IL-17 immunity to C. albicans. Here, we revealed a novel role of Langerin+ DCs in the tongue coordinating the acute IL-17 response during OPC.
All mouse experiments in this study were conducted in strict accordance with the guidelines of the Swiss Animals Protection Law and were performed under the protocols approved by the Veterinary office of the Canton Zurich, Switzerland (license number 201/2012 and 183/2015). All efforts were made to minimize suffering and ensure the highest ethical and humane standards.
WT C57BL/6j mice were purchased by Janvier Elevage. Il1r-/- [57], Rag1-/- [58,59], Ccr2-/- [60], RorcCre [61], Il23p19-/- [62], Il17af-/- [63] (a kind gift from Immo Prinz, MH Hannover, Germany), RorcCreR26ReYFP x Rag1-/- (a kind gift from Burkhard Becher, University of Zurich, Switzerland) and Rosa26reYFP animals [64] were bred at the Institute of Laboratory Animals Science (University of Zurich, Zurich, Switzerland). Klf4fl/fl, VavCreKlf4flfl [65], Batf3-/- [28] and Zbtb46GFP/+ [27] animals were bred at the Department of Biomedicine, University of Basel, Switzerland. Il1ab-/- mice were obtained from Wolf-Dietrich Hardt, ETH Zurich, Switzerland. All mice were on the C57BL/6 background except for Batf3-/-animals, which were on mixed Sv129/B6 background. The animals were kept in specific pathogen-free conditions and used at 6–12 weeks of age in sex- and age-matched groups.
The C. albicans strain SC5314 [66] was used for all experiments if not stated otherwise. CAF-yCherry was obtained from Robert Wheeler [67]. Mice were infected with 2.5x106 cfu of C. albicans sublingually as described [68] without immunosuppression. Mice were monitored for morbidity and euthanized in case they showed severe signs of pain or distress.
All analyses of infected animals in this study were carried out at 24 hours post infection. Mice were anaesthetized with a sublethal dose of Ketamin (100mg/kg), Xylazin (20mg/kg) and Acepromazin (2.9mg/kg) and perfused by injection of PBS into the right heart ventricle prior to removing the tongue and/or the spleen. For most experiments except for the analysis of tongue-resident MNPs, we isolated leukocytes as previously described in detail [69]. Briefly, tongues were cut into fine pieces and digested with DNase I (200μg/ml) and Collagenase IV (4.8 mg/ml, Invitrogen) in PBS for 50 minutes at 37°C. Single cell suspensions were obtained by passing the digested tissue through a 70μm strainer using ice-cold PBS supplemented with 1% FCS and 2mM EDTA. Tongue leukocytes were enriched over a 40% Percoll gradient before they were stained for flow cytometry. For the characterization of tongue-resident MNPs, tongues were cut in half and the underlying muscle tissue was carefully removed using a scalpel. The remaining tongue tissue was cut into fine pieces and digested with Trypsin (1mg/ml), DNase I (200mg/ml) and Collagenase IV (2.4mg/ml) in PBS for 45 minutes at 37°C. Single cell suspensions were obtained by passing the digested tissue through a 70μm strainer using ice-cold PBS supplemented with 1% FCS and 2mM EDTA and then stained for flow cytometry.
For tongue sections, the tissue was embedded in Tissue-Tek OCT compound (Sakura) and snap-frozen in liquid nitrogen. Sagittal cryo-sections were cut at a thickness of 9 μm with a HM525 Microtome Cryostat and were mounted to super frost glass slides (Thermo Scientific). The specimen were allowed to dry at room temperature for 30min prior to immunofluorescence staining. For epithelial sheets, the tongue was cut longitudinally and the muscle tissue was carefully removed with a scalpel. The tissue was placed with the epithelial layer upwards onto a Dispase II solution (2.85 mg/ml PBS, Roche) and incubated for 1 hour at 37°C. Epithelial sheets were obtained by separating the lamina propria from the epithelium using two watchman tweezers. For immunofluorescence staining the specimen were fixed either with methanol at 20°C for 20 minutes or with acetone a room temperature for 10 minutes depending on the antibody used for the staining. The following antibodies were used: anti-Langerin (clone 929F.3, hybridoma supernatant), anti-MHCII (clone M5/114.15.2, Biolegend) and anti-CD11c (clone HL3, BD Bioscience). The stained specimens were mounted with Mowiol and stored at 4°C. Images were acquired with a digital slide scanner (NanoZoomer 2.0-HT, Hamamatsu) and analyzed with NDP.view2.
To block cytokine secretion, infected mice were treated with Brefeldin A (Axon Lab AG, 250μg per mouse i.p.) three hours prior to euthanization. For IL-6 neutralization, animals were injected with anti-IL-6 (clone MP5-20F3, BioXCell, 60μg per mouse i.p.) directly after infection and again eight hours later. For neutrophil depletion, mice were treated with anti-Ly6G (clone 1A8, BioXCell, 150μg per mouse i.p.) on day -1 and with anti-G-CSF (clone 67604, R&D Systems, 10μg per mouse i. p.) on day -1 and day 1 post-infection. Anti-CSF1R (clone AFS98, BioXCell or produced and in-house and obtained from M. Greter) was injected on day -3 (2mg per mouse i. p.), on day -1 (0.5mg per mouse i. p.) and day 0 (0.5mg per mouse i. p.) of infection.
All antibodies were from BioLegend, if not stated otherwise. For Flow cytometric analysis, single cell suspensions of the tongue and the spleen were stained in PBS supplemented with 1% FSC, 5mM EDTA and 0.02% NaN3. LIVE/DEAD Near IR stain (Life Technologies) was used for exclusion of dead cells. The following antibodies were used for surface markers: anti-CD90.2 (30-H12), anti-CD45.2 (104), anti-TCRβ (H57-597), anti-TCRγδ (GL3), anti-CD11b (eBioscience, M1/70), anti-CD11c (N418), anti-MHCII (M5/114.15.2), anti-CD3 (145-2C11), anti-CCR6 (29-2L17), anti-CD24 (M1/69), anti-NCR1 (29A14), anti-CD122 (TM-β1), anti-CD44 (IM7), anti-CD69 (H1.2F3), anti-ki67 (16A8), anti-Ly6C (HK1.4), anti-CCR2 (SA203G11), anti-CD64 (X54-5/7.1), anti-Langerin (929F3), anti-EpCam (G8.8), anti-Ly6G (1A8), anti-XCR1 (ZET), anti-Sirpα (P84), anti-CD103 (2E7). For intracellular cytokine staining, tongue cells were fixed and permeabilized using BD Cytofix/Cytoperm reagent (BD Bioscience) and subsequently incubated in Perm/Wash buffer (BD Bioscience) containing the following cytokine-directed antibodies or the respective isotype controls: anti-IL-17A (TC11-18H10.1), anti-IL-17F (8F5.1A9), anti-pro-IL-1β (NJTEN3) and anti-IL6 (MP5-20F3). CD1d surface expression was stained with anti-CD1d tetramers (Proimmune). All extracellular and intracellular staining steps were carried out on ice. Cells were acquired on a FACS LSR II Fortessa (BD Biosciences) or on a FACS Gallios (Beckman Coulter) and the data were analyzed with FlowJo software (Tristar). In all the experiments, the cells were pre-gated on viable and single cells for analysis. Absolute cell numbers of CD90+ and CD90+IL-17A+ cells and their respective subpopulations were calculated based on a defined number of counting beads (BD Bioscence, Calibrite Beads), which were added to the samples before flow cytometric acquisition.
For sorting cells from the infected tissue, single cell suspensions of the tongue were stained in PBS, supplemented with 1% FSC and 5mM EDTA, using the same antibodies as described in the previous section. Using a FACS Aria III, 50–100 target cells per defined population were sorted per well of a 96-well plate (Eppendorf) containing RLT Plus RNeasy® lysis buffer (Qiagen). Lysates were snap-frozen and stored at -80°C until further processing. Whole-transcriptome amplification was performed following the Smart-Seq2 protocol [70]. Briefly, Agencourt RNAClean XP paramagnetic beads (Beckman Coulter) in combination with a DynaMag-96 side skirted magnet (Thermo Fisher) were applied to purify whole-genome RNA. Subsequently cDNA was generated using the SuperScript II Reverse Transcriptase Kit (Thermo Fisher), and further amplified with HiFi HotStart PCR Mix (KAPA Biosystems). For DNA clean-up, Agencourt AMPure XP beads (Beckman Coulter) were used as above. Optimal DNA concentration for real-time qPCR assays was determined by testing sample serial dilution for the expression of the control gene Actb. RT qPCR was performed using SYBR Green (Roche) and a QuantStudio 7 Flex (Life Technology) instrument. The primers were Actb fwd 5´-CCCTGAAGTACCCCATTGAAC-3´, Actb rev 5´-CTTTTCACGGTTGGCCTTAG-3´; Il1b fwd 5´-TACAGGCTCCGAGATGAACA-3´, Il1b rev 5´-AGGCCACAGGTATTTTGTCG-3´; Il6 fwd 5´-GAGGATACCACTCCCAACAGACC-3´, Il6 rev 5´-AAGTGCATCATCGTTGTTCATACA-3´, Il23a fwd 5´- CCAGCAGCTCTCTCGGAATC-3´, Il23a rev 5´-TCATATGTCCCGCTGGTGC-3´; Cd207 fwd 5´-ATGTTGAAAGGTCGTGTGGAC-3´, Cd207 rev 5´- GTGGTGTTCACTATCTGCATCT-3´; All qRT-PCR assays were performed in duplicates and the relative expression (rel. expr.) of each gene was determined after normalization to Actb transcript levels.
Cell numbers of CD90+ and CD90+IL-17A+ cells and their respective subpopulations were transformed using the formula Y = y(Log10+1) to plot absolute cell numbers including 0 values. Bar of each data set indicate arithmetic mean. Statistical significance was determined by unpaired Student’s-test with Holm-Sidak correction for multiple comparison, one-way or two-way ANOVA with Tukey’s multiple comparison test using GraphPad Prism software with *p< 0.05; **p<0.01; ***p<0.001; ****p<0.0001.
|
10.1371/journal.pntd.0007412 | Malaria vector species in Amazonian Peru co-occur in larval habitats but have distinct larval microbial communities | In Amazonian Peru, the primary malaria vector, Nyssorhynchus darlingi (formerly Anopheles darlingi), is difficult to target using standard vector control methods because it mainly feeds and rests outdoors. Larval source management could be a useful supplementary intervention, but to determine its feasibility, more detailed studies on the larval ecology of Ny. darlingi are essential. We conducted a multi-level study of the larval ecology of Anophelinae mosquitoes in the peri-Iquitos region of Amazonian Peru, examining the environmental characteristics of the larval habitats of four species, comparing the larval microbiota among species and habitats, and placing Ny. darlingi larval habitats in the context of spatial heterogeneity in human malaria transmission. We collected Ny. darlingi, Nyssorhynchus rangeli (formerly Anopheles rangeli), Nyssorhynchus triannulatus s.l. (formerly Anopheles triannulatus s.l.), and Nyssorhynchus sp. nr. konderi (formerly Anopheles sp. nr. konderi) from natural and artificial water bodies throughout the rainy and dry seasons. We found that, consistent with previous studies in this region and in Brazil, the presence of Ny. darlingi was significantly associated with water bodies in landscapes with more recent deforestation and lower light intensity. Nyssorhynchus darlingi presence was also significantly associated with a lower vegetation index, other Anophelinae species, and emergent vegetation. Though they were collected in the same water bodies, the microbial communities of Ny. darlingi larvae were distinct from those of Ny. rangeli and Ny. triannulatus s.l., providing evidence either for a species-specific larval microbiome or for segregation of these species in distinct microhabitats within each water body. We demonstrated that houses with more reported malaria cases were located closer to Ny. darlingi larval habitats; thus, targeted control of these sites could help ameliorate malaria risk. The co-occurrence of Ny. darlingi larvae in water bodies with other putative malaria vectors increases the potential impact of larval source management in this region.
| The standard methods used to combat mosquitoes that transmit malaria, long-lasting insecticide treated nets and indoor residual spraying, target mosquitoes that bite people indoors and rest indoors after biting. In Amazonian Peru, the major malaria vector, Nyssorhynchus darlingi (formerly Anopheles darlingi), is known to bite and rest mostly outdoors and to feed on animals as well as humans. Therefore, additional methods are needed to control this species, such as targeting immature mosquitoes in water bodies using chemical or biological larvicides or environmental modification. To determine whether this is feasible, more ecological information about Ny. darlingi larval habitats in this region is needed. In this study, we found that Ny. darlingi were more likely to be collected from water bodies in more deforested areas, and in the presence of other species of mosquitoes that transmit malaria. We characterized the bacteria detected in three species of mosquito larvae, and found that the species of mosquito, and not the water body, determined which bacteria were identified. As we found that houses with more malaria cases were located closer to Ny. darlingi larval habitats, management of larval habitats may be an effective method to reduce the risk of malaria in this region.
| Despite substantial progress in reducing the global burden of malaria over the last two decades, no progress was made in decreasing the total number of malaria cases worldwide between 2015 and 2017 [1]. This emphasizes the need not only for continued commitment to the two most effective malaria vector control methods, long-lasting insecticidal nets (LLINs) and indoor residual spraying (IRS), but also for the use of alternative vector control methods [2, 3]. LLINs and IRS target mainly mosquitoes that feed and rest indoors (endophagic and endophilic, respectively). To combat residual malaria transmission that occurs despite universal coverage with LLINs and/or IRS, vector interventions need to incorporate tools targeting mosquitoes that feed and rest outdoors (exophagic and exophilic, respectively) [4]. In this study we focus on the main Amazonian malaria vector, Nyssorhynchus darlingi Root (formerly Anopheles darlingi [5]), which displays a high degree of plasticity in its biting behavior [6–9], making it difficult to target with LLINs and IRS.
Larval source management, used frequently in high-income countries for vector control, targets immature mosquitoes, and therefore its efficacy does not depend on the biting behavior of adult mosquitoes [10, 11]. Larval source management is currently recommended as a supplementary malaria vector control method by the WHO only in areas where larval habitats are “few, fixed, and findable” [12]. To evaluate whether larval source management will be effective in a malaria endemic region in the Amazon, and to plan targeted larval source management strategies to maximize the use of limited resources, it is necessary to have at least a basic understanding of the larval ecology of local and regional malaria vectors, a topic that is understudied globally [10, 12–14].
In Peru, over 90% of malaria cases occur in the Loreto Department, where malaria is endemic and seasonal, peaking during the rainy season [15, 16]. Plasmodium vivax causes the majority of regional malaria cases, but the proportion of cases caused by Plasmodium falciparum has been increasing since 2011 [16–18]. The primary malaria vector in Loreto is Ny. darlingi, reintroduced into the region in the 1990s following elimination in the 1960s [19]. A comprehensive survey of Ny. darlingi larval habitats in the peri-Iquitos region of Loreto in 2000–2001 focused only on water bodies along the Iquitos-Nauta Highway [20]. Our group has found previously that adult Ny. darlingi in riverine and highway environments in peri-Iquitos differ in their biting behavior and exhibit genetic differentiation [21]. An unanswered question is whether there are unique characteristics of riverine or highway larval habitats that contribute to such differentiation. Vittor et al. found that Ny. darlingi larvae in peri-Iquitos were associated with large water bodies, filamentous algae, the presence of human populations, and lower forest cover [20]. Elsewhere in its distribution, Ny. darlingi larvae have similarly been associated with human-modified habitats, including ponds constructed for fish farming and water bodies near the fringes of forested areas [22–25]. In peri-Iquitos, fish farming is common, and increased density of fish ponds has been associated with an increased risk of malaria [26], though a clear link between Ny. darlingi larvae and fish ponds has not been demonstrated [20]. To investigate these associations, and to determine how they relate to the risk of malaria, more current research on the relationship between Ny. darlingi and human-modified larval habitats in the peri-Iquitos region, including deforested sites and fish ponds, is necessary.
The composition of the microbiota of Neotropical malaria vectors is another critical knowledge gap. Mosquitoes rely on midgut microbiota for development [27], and the composition of the Anopheles gambiae microbiome affects its vector competence for P. falciparum [28, 29]. In the Neotropics, a few studies have focused on the Nyssorhynchus albimanus Wiedemann (formerly Anopheles albimanus [5]) microbiota [30–32], and until recently, only limited studies of Ny. darlingi using Sanger sequencing or culture-based methods on small numbers of mosquitoes had been published [33–35]. A recent study using next-generation 16S sequencing characterized the microbial composition of a large sample of Ny. darlingi and Nyssorhynchus nuneztovari s.s. Gabaldón (formerly Anopheles nuneztovari s.s. [5]) larvae and adults collected in coastal Colombia [36]. This study found that mosquito developmental stage and geographic location, but not mosquito species, influenced the composition of the gut microbiota. In mosquitoes from Africa and the United States, bacterial communities have been shown to differ across larval habitats or localities [36–40], between seasons [40, 41] and between species [39, 42, 43]. Where mosquito species share larval habitats, the extent to which the larval environment vs. mosquito species influences microbiome composition has varied across studies and may be different across systems. No studies addressing this question have been conducted in malaria vectors in the Peruvian Amazon.
Several other species of Nyssorhynchus that may be acting as secondary malaria vectors are also found in the peri-Iquitos region. These include Nyssorhynchus triannulatus s.l. Neiva & Pinto (formerly Anopheles triannulatus s.l. [5]), Nyssorhynchus rangeli Gabaldón, Cova-García & López (formerly Anopheles rangeli [5]), and members of the Nyssorhynchus Oswaldoi-Konderi complex (formerly the Anopheles Oswaldoi-Konderi complex [5]). Nyssorhynchus triannulatus s.l. is a species complex distributed throughout Latin America and the Caribbean that has been incriminated as a human malaria vector [44, 45]. Nyssorhynchus rangeli is distributed throughout the Amazon Basin and is a local vector of Plasmodium in southern Colombia [46] and Junín Department, Peru [47]. The Oswaldoi-Konderi complex consists of five species broadly distributed throughout South America, several of which have been implicated as malaria vectors [48]: Nyssorhynchus oswaldoi s.s. Peryassú (formerly Anopheles oswaldoi s.s. [5]), Nyssorhynchus oswaldoi A Ruiz-Lopez (formerly Anopheles oswaldoi A [5]), Nyssorhynchus oswaldoi B Ruiz (formerly Anopheles oswaldoi B [5]), Nyssorhynchus konderi Galvão & Damasceno (formerly Anopheles konderi [5]), and Nyssorhynchus sp. nr. konderi Ruiz-Lopez (formerly Anopheles sp. nr. konderi [5]).
For the current study, we collected Anophelinae mosquito larvae longitudinally from water bodies in eight villages on four rivers and one highway in the peri-Iquitos region to investigate the environmental drivers of differences in the composition of larval communities within water bodies and of bacterial communities within larvae in the context of malaria risk. Our aims were: 1) to characterize the larval habitats of malaria vectors in peri-Iquitos and test the hypothesis that Ny. darlingi larvae are associated with human-modified habitats; 2) to determine whether the spatial distribution of malaria cases is associated with the spatial distribution of Ny. darlingi larval habitats; and 3) to describe the larval microbiota of three malaria vectors and test the hypothesis that its composition is habitat-specific.
Authorization for the fieldwork included in this study was given by the Dirección de Gestión Forestal y de Fauna Silvestre and the Dirección General Forestal y de Fauna Silvestre of the Ministerio de Agricultura de la República del Perú, permit N. 0424-2012-AG-DGFFS-DGEFFS.
Anophelinae larvae were collected from eight villages in the peri-Iquitos region of Loreto, Peru (Fig 1). Four villages to the south and west of Iquitos (San José de Lupuna (LUP) on the Nanay River, Santa Emilia (SEM) on the Nahuapa Stream, and Nuevo Horizonte (NHO) and El Triunfo (TRI) on the Iquitos-Nauta Highway) have been described previously [21]. The four villages north of Iquitos are located in the Mazán District, that consists of small communities largely supported by agriculture, fishing, and timber extraction, and has a high overall incidence of malaria [49, 50]. In 2017, 1000 cases of P. vivax and 349 cases of P. falciparum were reported in the Mazán District (Annual Parasite Index (API) of 96.1 per 1000 inhabitants [51]). The four villages in the current study were selected because they each had an API greater than 10 cases per 1000 inhabitants, and to represent two ecologically distinct river systems [52]. Salvador (SAL) and Urco Miraño (URC) are on the Napo River, a large white water river that originates in Ecuador and flows into the Amazon River; and Libertad (LIB) and Visto Bueno (VIB) are on the Mazán River, a black water river that is a tributary of the Napo River.
In each village, collections were conducted 5–6 times over the study period (Table 1). Satellite images were used to identify water bodies within a 1km radius of each village (to correspond to the approximate flight range of Ny. darlingi [7]), and additional water bodies in this radius were identified by ground-truthing. Mosquito larvae dippers (350mL capacity) were used to sample each water body once per collection. Sampling points were selected 10m apart along the perimeter of each water body, with at most 20 sampling points per water body. Ten dips were taken at each sampling point and examined for the presence of Anophelinae larvae.
Characteristics of each water body were recorded at the time of each collection, including type of water body (fish ponds were recorded as active or abandoned); depth; cloud cover; shade level; presence of vegetation, fish, and amphibians; water movement; type of bed material; and light intensity (Foot Candle/Lux meter, Extech, Nashua, NH, USA). Additionally, the alkalinity, hardness, and nitrate and nitrite levels (Eco-Check 5-in-1 Test Strips, Industrial Test Systems, Inc., Rock Hill, SC, USA); and pH, temperature, conductivity and salinity (ExStik pH/Conductivity Meter, Extech, Nashua, NH, USA) of each water body was recorded.
Larvae collected in LUP, NHO, TRI, and SEM were reared to adults for species identification using morphological keys [53–55]. If the reared larvae died before adulthood, were not able to be identified, or were identified as a species other than Ny. darlingi, they were preserved in 100% ethanol (for larvae) or on silica gel (for adults) for molecular identification. All larvae collected in LIB, SAL, URC, and VIB were killed and preserved in 100% ethanol immediately after collection.
For molecular identification, total genomic DNA was extracted from whole larvae and reared adults using the DNeasy Blood & Tissue kit (Qiagen, Hilden, Germany). Mosquitoes were identified by PCR-RFLP of the ribosomal internal transcribed spacer 2 (ITS2) region [56] or, if the ITS2 region did not amplify, did not digest, or did not have an identifiable RFLP pattern, by cytochrome c oxidase subunit I (COI) barcode sequencing [57, 58]. Sequences of primers used for the ITS2 PCR-RFLP and for COI sequencing are included in Table I in S1 File. The COI barcode region was sequenced in one direction using the forward primer at the Wadsworth Center Applied Genomic Technologies Core (New York State Department of Health). All unique COI sequences were deposited in GenBank (accession numbers MK172893 to MK173015; Nyssorhynchus dunhami Causey (formerly Anopheles dunhami [5]) COI sequences were previously deposited in GenBank under accession numbers MH723612 to MH723658). Identifications were done by querying the sequences against the BOLD Identification System [59] or GenBank (https://www.ncbi.nlm.nih.gov/genbank/). Only larvae identified as species in the genera Anopheles, Kerteszia, Lophopodomyia, Nyssorhynchus, or Stethomyia (all formerly subgenera in the genus Anopheles [5]) were included in the analysis (S1 Dataset).
COI sequencing was done for all samples identified by ITS2-PCR-RFLP as members of the Oswaldoi-Konderi complex. Sequences were edited and checked for stop codons and pseudogenes using Geneious v9.1.4 [60]. Oswaldoi-Konderi complex (Ny oswaldoi s.s., Ny. oswaldoi A, Ny. oswaldoi B, Ny. konderi, and Ny. sp. nr. konderi) COI sequences from Ruiz et al. [61] and Saraiva et al. [48] were retrieved from GenBank and aligned to Oswaldoi-Konderi complex larval sequences from this study, using default settings of MUSCLE [62] in MEGA 7v7.0.26 [63]. Sequences were trimmed to 578 bp and a haplotype file created with DAMBE6 [64] (S2 Dataset). These data were used to construct a median-joining haplotype network in POPART v1.7 [65], with epsilon set to 0. Based on this haplotype network (Fig I in S1 File), one individual from this study was identified as Ny. konderi, while the remaining individuals were identified as Ny. sp. nr. konderi.
Enhanced Vegetation Index (EVI), Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) were calculated based on Landsat8 collections in Google Earth Engine repositories (Landsat 8 8-day EVI, NDVI and NDWI composites). These Landsat 8 composites are constructed from Level L1T orthorectified scenes, using the computed top-of-atmosphere (TOA) reflectance [66]. The collections have a 30m spatial resolution, and 8-day temporal resolution; between 3 to 4 images per month. EVI was generated from the Near Infra-Red (NIR), Red and Blue bands of each scene [67]. NDVI was generated from the NIR and Red bands of each scene as (NIR—Red) / (NIR + Red). NDWI is sensitive to changes in liquid water content of vegetation canopies. It is derived from the NIR band and a second Infra-Red (IR) band, ≈1.24μm when available or the nearest available IR band [68]. All normalized indices range from -1.0 to 1.0. Mean EVI, NDVI and NDWI were calculated within four buffers of 50, 100, 250 and 500m radius constructed around each water body at each collection. A two-month window was used when images were not available due to dense cloud coverage.
The distance from each water body to the closest forest patch was calculated in QGIS 3.0. Forest fringes were delimited by manual digitalization based on visual inspection of Google Earth imagery; any tree patch containing greater than 10m2 canopy cover was considered a forest patch. The dates of available imagery for each village are shown in Table II in S1 File. The resulting polygons that delineate the forest fringes were imported into QGIS along with the coordinates of the water bodies. A proximity raster was generated based on the rasterization of the forested area polygons. Briefly, in each cell, the distance to a target point was calculated using the proximity algorithm. As result, a raster with the minimum distance to the forest areas was generated for the entire study area and the values for each water body in meters were extracted.
The Hansen Global Forest Change dataset, the result of a time-series analysis of Landsat images that characterizes global forest extent and change [69], was used to compute forest cover and forest loss in Google Earth Engine. Yearly forest cover and forest loss area were calculated around each water body at different buffer sizes (50, 100, 250 and 500m) from 2010 to 2016. Percent forest cover in 2016 (the forest cover area divided by the total area of the buffer) and percent forest loss between 2010 and 2016 (the difference between the forest cover area in 2010 and 2016 divided by the forest area in 2010) at each buffer size was calculated in R v.3.5.3.
To determine whether Nyssorhynchus species larvae were co-occurring more often than would be expected by chance, affinity indices between pairs of the five most common Nyssorhynchus species were calculated using the formula described by Fager and McGowan [70]: JNA*NB× 12 NB, where J is the number of collection points (water bodies sampled at a collection) at which the species are both present, NA is the total number of collection points at which species A is present, and NB is the total number of collection points at which species B is present, chosen so that NA ≤ NB. An affinity index ≥0.5 indicates affinity between the species.
The characteristics of Ny. dunhami larval habitats from this study have been described [71]. The water body characteristics associated with the presence of the four most abundant species excluding Ny. dunhami were evaluated separately for each species at the level of the collection point (S3 Dataset). Unless otherwise indicated, all data processing and analysis described in this section was done in R v. 3.5.2 [72]. Collection points at which a water body was dry at the time of collection were excluded from the analysis. If a water body was dry at one or more collections, it was considered temporal, otherwise, it was considered permanent. The distance from each water body to the nearest non-dry water body, and to the nearest water body positive for Ny. darlingi at least once during this study, were calculated using the R package ‘sp’ [73].
Censuses that determined the number of people living in each georeferenced house were completed in May 2015 in SEM, November 2015 in LUP, and November/December 2016 in all other villages. Census data were used to calculate the distance between each water body and the nearest inhabited house, and the number of people living within 50, 100, 250, and 500m radius buffers from each water body (also used to calculate presence/absence variables for whether any people lived within each buffer) using the R packages ‘sp’ [73] and ‘rgeos’ [74].
Independent variables evaluated for association with the presence of each Nyssorhynchus species are listed in Table 2. Nitrate and nitrite levels were nearly always zero (99.3% and 99.8% of collection points, respectively), so they were not included. Alkalinity and hardness values were recoded as zero/non-zero to account for rare levels.
To account for the low percentage of missing data in the final dataset (2% overall, 8% or less for each variable), multiple imputation was done using the cart method in the R package ‘mice’ [75]. Five imputations were calculated, with 20 iterations each. The imputed datasets were used to construct logistic mixed-effects models in the R package ‘mitml’ [76], with the presence of each Nyssorhynchus species as the outcome and the water body ID as a random intercept to account for multiple collections conducted at each water body. Village was included as a fixed effect and not a higher-level random intercept so that between-village differences in the presence of Nyssorhynchus species could be assessed. The relationship between each variable and the presence of each Nyssorhynchus species was evaluated using bivariate models. For the population density variables, vegetation and water indices, and forest cover variables, it was unknown which, if any, variable at which radius buffer would affect the presence of the Nyssorhynchus species. Bivariate models were built for each variable at each radius, and for each category (vegetation index: EVI and NDVI; population density: distance to nearest inhabited house, number of people in radius, presence/absence of any people in radius; NDWI; percent forest cover; percent forest loss 2010–2016), the variable with the highest bivariate log likelihood was selected to be evaluated for inclusion in the multivariate model. A forward stepwise process was used to build the multivariate model using only variables with bivariate p<0.2. Each variable was added in order of its log likelihood in the bivariate model, and a variable was retained if its p-value was <0.2 in the final model.
To explore the effect of environmental variables on the overall community assemblage of Nyssorhynchus larvae, we used redundancy analysis (RDA). The RDA was computed in the R package ‘vegan’ [77], using the Hellinger-transformed presence/absence matrix for all identified larval species as the community data matrix and the environmental variables listed in Table 2 (with the exception of the presence of other Anophelinae species larvae) as the constraining variables. Missing data in the environmental dataset was imputed using the imputeFAMD function in the R package ‘missMDA’ [78], using five components to predict the missing entries as suggested by the estim_ncpFAMD function. A single variable and buffer size was selected for the vegetation index, NDWI, forest cover, forest loss, and population density variables by computing RDAs using each variable separately and selecting the variable in each group that resulted in the RDA with the highest adjusted R2. The RDA biplot was visualized using the R package ‘ggords’ [79].
Malaria case data for all 8 villages from 2016 were obtained from the local health authority (Dirección Regional de Salud Loreto, DIRESA). Where possible, cases were matched to georeferenced houses to correspond with the census data described above. To reduce the likelihood of including duplicate cases, all repeat diagnoses of P. vivax within 60 days or P. falciparum within 30 days for the same person were excluded, as in [80]. Since water bodies were sampled only once every 2–4 months, the cases and the presence of Ny. darlingi larvae were aggregated by 6-month season (rainy season: January-June 2016, dry season: July-December 2016). Separately for each season, the distance from each house to the nearest water body positive for Ny. darlingi at least once during the season was calculated using the R package ‘sp’ [73] (S4 Dataset). A separate Poisson regression for each season was constructed in R v. 3.5.2 using the log number of people in each house as the offset, the number of malaria cases in each house during the season as the outcome, and the distance to the nearest Ny. darlingi-positive water body as the explanatory variable. The village of VIB was excluded from the dry season analysis because no Ny. darlingi-positive water bodies were identified in this village during this season.
Larvae for bacterial 16S rRNA sequencing were selected from among 3rd and 4th instar larvae collected in LIB, SAL, and URC. Only larvae collected in September and November 2016 (dry season) were included, to minimize any potential seasonal effects on the microbiome. DNA was extracted from whole larvae using the DNeasy Blood & Tissue kit following surface sterilization in 100% ethanol. Prior to bead-beating, larvae were suspended in the manufacturer-recommended enzymatic lysis buffer (containing lysozyme) to enhance DNA purification from gram-positive bacteria as suggested in [81]. All extractions and post-extraction manipulation of samples and 16S PCR reactions were performed in a biosafety cabinet with pre-sterilized materials where possible to avoid contamination, and all extractions were performed using the same extraction kit to avoid batch contamination effects [82]. A negative extraction control with reagents only was processed along with the samples from extraction through sequencing. The larvae were identified to species using ITS2 PCR-RFLP as described above. Following identification, 95 larvae and the negative extraction control were selected for 16S bacterial sequencing (S5 Dataset). These samples include larvae of three species (Ny. darlingi, Ny. rangeli, and Ny. triannulatus s.l.) from 12 water bodies, selected to maximize the number of water bodies with multiple species present.
Bacterial 16S rRNA gene V3-V4 variable regions were amplified using the Illumina adapter overhang-linked primers suggested in the Illumina 16S Metagenomic Sequencing Library Preparation guide ([83], Table I in S1 File). PCR reactions were performed in a 25μl reaction including 5μl extracted DNA, 5μl each 1μM forward and reverse primers, and 10μl 2X KAPA HiFi HotStart ReadyMix. Reaction conditions were as follows: 95°C for 3 minutes; 30 cycles of 95°C for 30s, 55°C for 30s, and 72°C for 30s; and 72°C for 5 minutes. Products were visualized on 1% agarose gels. For samples that did not amplify using the above reaction conditions (n = 18 larvae + extraction control), an identical protocol using a 35-cycle reaction was used. All PCR reactions were performed in triplicate, and pooled products of the triplicate reaction for each sample were sent to the Wadsworth Center Applied Genomic Technologies Core for PCR clean-up, a second PCR to attach dual indices and Illumina sequencing adapters, and sequencing on the Illumina MiSeq system. All reads were deposited in the NCBI Sequence Read Archive (SRA; BioProject ID PRJNA494695).
The Quantitative Insights Into Microbial Ecology (QIIME) 1.9.1 pipeline [84] multiple_join_paired_ends.py and multiple_split_libraries_fastq.py scripts were used to prepare the sequencing reads for analysis. The QIIME 1.9.1 pick_open_reference_otus.py script, which wraps uclust for clustering [85], PyNAST for alignment [86], RDP Classifier for assigning taxonomy [87], and FastTree for building a phylogenetic tree [88], was used to assign 16S rRNA reads to operational taxonomic units (OTUs). Reverse strand matching was enabled in uclust, OTUs were matched to the SILVA 128 rRNA database [89] at a 97% identity threshold, alignments were filtered using an allowed gap fraction of 0.8 and an entropy threshold of 0.1, and lane mask filtering was suppressed. A single OTU (an uncultured Delftia spp., GU731299) had a higher relative abundance in the negative control than in any larva (control:maximum larvae relative abundance ratio = 6.38 vs. <0.5 for all other OTUs). This OTU accounted for 38% of the 3704 identified reads from the negative control (the next most prevalent OTU accounted for only 7% of the reads) and 1.5% of larval reads and was excluded for all downstream analyses. In addition, non-bacterial OTUs (n = 2) and low-abundance OTUs accounting for <0.1% of reads were filtered from the final table. The final OTU table is included as S6 Dataset.
The QIIME 1.9.1 summarize_taxa_through_plots.py script was used to visualize differences in bacterial composition across samples and groups. To compare the beta diversity across groups of samples, the QIIME 1.9.1 beta_diversity_through_plots.py script was used to rarefy the OTU table to 13,000 sequences per individual (rarefaction curves indicated that the alpha diversity saturated at about 10,000 sequences (Fig IV in S1 File)), and to compute unweighted and weighted UniFrac [90] and Bray-Curtis distance matrices, as well as a principal coordinates analysis (PCoA) for each distance matrix. Taxonomic composition and principal coordinates analysis plots were created in R using the ‘ggplot2’ package [91]. QIIME 2 v. 2017.12 was used to compute pairwise analyses of similarities (ANOSIMs) of the beta diversity distance matrices (beta-group-significance), alpha rarefaction curves (alpha-rarefaction), and an analysis of composition of microbes (ANCOM; qiime composition ancom) [92].
A total of 1579 larvae identified as Nyssorhynchus, Anopheles, or Stethomyia species were collected in 88 water bodies in the 8 villages between January 2016 and March 2017 (S1 Dataset). This excludes 102 larvae lost in processing, 32 larvae for which the COI product did not amplify, and 24 larvae identified as non-Anopheles, Kerteszia, Lophopodomyia, Nyssorhynchus, or Stethomyia species. The most commonly identified species was Ny. darlingi (n = 751), followed by Ny. rangeli (n = 269), Ny. triannulatus s.l. (n = 239), Ny. sp. nr. konderi (n = 131), Ny. dunhami (n = 116), Anopheles mattogrossensis Lutz & Neiva (n = 35), Anopheles forattinii Wilkerson & Sallum/Anopheles costai Fonseca & Ramos/Anopheles mediopunctatus Lutz (n = 17) (these three species are closely related [93] and could not be differentiated by COI barcode sequences), Nyssorhynchus benarrochi B Ruiz (formerly Anopheles benarrochi B [5]) (n = 16), Stethomyia nimbus Theobald (formerly Anopheles nimbus [5]) (n = 4), and Ny. konderi (n = 1). Each of the six most common species was collected in all 8 villages (with the exception of Ny. sp. nr. konderi in URC). Ny. benarrochi B was only collected in the four villages to the south and west of Iquitos (LUP, SEM, NHO, and TRI), and St. nimbus and Ny. konderi were only collected in SAL (Fig 2). The species composition varied by village (Fig 2) but was relatively consistent over time within each village (Fig II in S1 File).
Nyssorhynchus darlingi, Ny. rangeli, Ny. triannulatus s.l., and Ny. sp. nr. konderi were all collected from both artificial and natural water bodies (Table 3). All fish ponds that were active throughout the study period (n = 17) were positive for Ny. darlingi at least once.
The final analysis dataset included 403 collection points for 88 water bodies sampled across 5–6 collections, excluding 84 collection points at which the water body was dry. Anophelinae species were often found co-occurring at collection points; at least two species were present in 138 (70%) of 197 collection points that had any species present, and in one case, eight species were present at the same collection point. Affinity indices calculated between pairs of the five most common species indicated that two pairs of species showed an affinity (affinity index>0.5): Ny. darlingi and Ny. rangeli; and Ny. rangeli and Ny. triannulatus s.l. (Table 4).
Of the 403 collection points, Ny. darlingi was present in 169 (42%), Ny. rangeli in 66 (16%), Ny. triannulatus s.l. in 65 (16%), and Ny. sp. nr. konderi in 61 (15%). In multivariate logistic mixed-effects models, the presence of all four species was most strongly associated with the presence of other Nyssorynchus, Anopheles, or Stethomyia spp. larvae (OR>14, p<0.001 for all four species; Fig 3; Tables III-VI in S1 File). In addition, the presence of Ny. darlingi was significantly positively associated with the presence of emergent vegetation (OR = 2.54, p = 0.036) and percent forest loss between 2010 and 2016 at a 500m radius (OR = 1.38, p = 0.016), and significantly negatively associated with EVI at a 500m radius (OR = 0.01, p = 0.014) and light intensity (OR = 0.14, p = 0.007). The odds of Ny. darlingi presence differed significantly among villages in the bivariate, but not the multivariate model (Table III in S1 File). Ny. darlingi presence was not significantly associated with highway vs. riverine habitat (riverine OR = 1.72 vs. highway, p = 0.335).
The presence of Ny. rangeli was significantly positively associated with the villages SAL and URC (OR = 6.37 vs. the other 6 villages, p = 0.002), the presence of bushes (OR = 2.92, p = 0.018), the distance to the nearest forest (OR = 1.01, p = 0.017), the number of people living in a 100m radius (OR = 1.02, p = 0.013), and the January-March 2017 quarter (OR = 8.17 vs. July-September 2016, p = 0.004) in the multivariate model (Table IV in S1 File).
The presence of Ny. triannulatus s.l. was significantly positively associated with the same villages as Ny. rangeli: SAL and URC (OR = 16.45 vs. the other 6 villages, p<0.001), and significantly negatively associated with the presence of water movement (OR = 0.07, p = 0.002), the presence of emergent vegetation (OR = 0.15, p = 0.004), and the January-March 2016 quarter (OR of all other quarters>8, p≤0.034) in the multivariate model (Table V in S1 File).
The presence of Ny. sp. nr. konderi was significantly positively associated with partial and total shade (OR = 10.02/12.57 vs. no shade and p = 0.016/0.010, respectively), the percent forest cover in a 250m radius (OR = 1.06, p = 0.001), and the January-March and April-June 2016 quarters (OR = 4.52/4.05 vs. January-March 2017 and p = 0.013/0.023, respectively), and significantly negatively associated with the presence of any people living in a 100m radius (OR = 0.32, p = 0.014) in the multivariate model (Table VI in S1 File).
The RDA results were consistent with the results of the multivariate logistic mixed-effects models, highlighting, for example, the association of Ny. darlingi-positive habitats with emergent vegetation and recent forest loss (Fig III in S1 File). The RDA also emphasizes the ecological similarity of Ny. triannulatus s.l. and Ny. rangeli habitats, and of Ny. sp. nr. konderi and Ny. dunhami habitats.
A total of 556 malaria cases (367 (66%) P. vivax, 189 (34%) P. falciparum) were reported in the 8 study villages during 2016, excluding 22 repeat diagnoses. The Annual Parasite Index (API) for each village in 2016 ranged from 22 (URC) to 659 (LUP) (Table VII in S1 File). The final analysis dataset consisted of 442 houses from the 8 villages with a combined 1951 inhabitants and 498 malaria cases, excluding 18 inhabitants and 53 malaria cases that were unable to be linked to a georeferenced house, and 5 cases reported from VIB in the dry season excluded because no Ny. darlingi-positive water body was identified in this village during this season (S4 Dataset). The number of malaria cases in each house was negatively associated with the distance to the nearest water body positive for Ny. darlingi in the rainy season (Poisson rate ratio per 100m distance = 0.98, 95% CI 0.96–0.996, p = 0.02) and in the dry season (Poisson rate ratio per 100m distance = 0.91, 95% CI 0.86–0.95, p = 0.0005) (Fig 4).
Sequencing of 16S rRNA amplicons resulted in a total of 18,427,247 paired-end reads from 95 larvae, with a median of 166,453 reads per larva (range: 55,335–719,042), and 5,865 reads from the negative extraction control. After OTU picking and filtering, the final OTU table for the 95 larvae consisted of 12,392,204 reads matched to 89 bacterial OTUs (S5 and S6 Datasets).
A principal coordinates analysis (PCoA) of the unweighted Unifrac distance matrix indicated that the samples clustered by larval species (Fig 5A), with the Ny. darlingi larvae clustering together, and apart from the Ny. rangeli and Ny. triannulatus s.l. larvae. PCoAs of the weighted Unifrac and Bray-Curtis distance matrices showed similar clustering (Fig V in S1 File). The samples did not cluster by water body (Fig VI in S1 File). An ANOSIM of the unweighted Unifrac distance matrix indicated that the bacterial composition was more similar among larvae from the same species (overall R = 0.55, p = 0.001; Ny. darlingi vs. Ny. rangeli R = 0.61, p = 0.001; Ny. darlingi vs. Ny. triannulatus s.l. R = 0.63, p = 0.001; Ny. rangeli vs. Ny. triannulatus s.l. R = 0.07, p = 0.096) than among larvae from the same water body (overall R = 0.12, p = 0.001) or village (overall R = 0.08, p = 0.03). Measured ecological variables of the water bodies, including presence of vegetation, water chemistry measurements, and vegetation and water indices (S5 Dataset) were also tested for their effect on the bacterial composition; none had as strong as an effect as larval species (next highest ANOSIM R = 0.25).
The most abundant bacterial families identified from Ny. darlingi larvae were Enterobacteriaceae, Cytophagaceae, and Moraxellaceae; from Ny. rangeli and Ny. triannulatus s.l. larvae, the most abundant were Alcaligenaceae and Enterobacteriaceae (Fig 5B, Fig VII in S1 File). By ANCOM, 33 of the 89 OTUs were differentially abundant comparing Ny. darlingi to Ny. rangeli and Ny. triannulatus s.l. These included three OTUs in the family Cytophagaceae (all in the genus Flectobacillus) that were more abundant in Ny. darlingi; four OTUs in the family Alcaligenaceae (in the genera Bordetella, Candidimonas, Castellaniella, and Pusillimonas) that were more abundant in Ny. rangeli and Ny. triannulatus s.l.; and six OTUs in the family Enterobacteriaceae (all in the genus Thorsellia), two of which were more abundant in Ny. darlingi and four of which were more abundant in Ny. rangeli and Ny. triannulatus s.l. (S1 Table).
In this study, we investigate the larval ecology of Ny. darlingi in the peri-Iquitos region of Amazonian Peru in the context of human landscape modification and malaria risk. Furthermore, we provide evidence that Nyssorhynchus species from the same larval habitats have distinct microbiomes. This study represents, to our knowledge, the first published characterization of Ny. darlingi larval habitats in the peri-Iquitos region since 2001 [20], and the first that has included riverine villages. The epidemiology of malaria in the Loreto Department has changed significantly since 2001; a comprehensive malaria control program resulted in a dramatic drop in malaria incidence between 2005 and 2011, followed by a steady increase in the number of overall malaria cases and the proportion caused by P. falciparum since then [16]. This has been accompanied by behavioral [9] and genetic [21] changes in the adult Ny. darlingi population in peri-Iquitos in the same timeframe. This study provides an updated understanding of the ecology of the primary malaria vector in the region, as well as possible secondary vectors, which will fundamentally inform integrated vector control methods, including targeted larval source management.
In this study, Ny. darlingi were more likely to be present in water bodies in areas with a higher amount of recent deforestation (a higher percent forest loss in a 500m radius between 2010 and 2016) and a lower vegetation index (a lower EVI at a 500m radius). These associations are consistent with the results of Vittor et al.’s study in the peri-Iquitos region [20], that found that increased forest cover in a 1x1km grid decreased the probability of Ny. darlingi larval presence. The relationship between deforestation and both Ny. darlingi habitat suitability and overall malaria transmission has not been clearly defined, mainly due to differences in the definition of deforestation in different studies [94]. In the Brazilian Amazon, the forest fringe hypothesis has been proposed, whereby malaria risk is highest at the edges of deforested areas [95], particularly in small deforested patches [96]. In this transition zone between forested and deforested areas, vectors have ample access to human blood meals, but also to shaded water bodies [22, 24]. In the Peruvian Amazon, the forest cover level is overall much higher than it is in the Brazilian Amazon [94]; for the current study, the forest cover in a 500m radius around the sampled water bodies ranged from 33% to 89%. It is possible that deforestation differentially affects vector populations and overall malaria risk at different forest cover levels [94]; perhaps the water bodies in deforested landscapes in the peri-Iquitos region act comparably to the forest fringes in Brazil. Clearly, more research on the dynamics between deforestation and malaria transmission across the Amazon is necessary.
Nyssorhynchus darlingi larval habitats have also been found in previous studies to be associated with human presence directly [20, 22, 23], and indirectly via aquaculture; fish ponds have been implicated as Ny. darlingi larval habitats [20, 23, 25] and have been hypothesized to increase the risk of malaria transmission in the Peruvian and Brazilian Amazon [26, 97]. While we did not see a significant association between the presence of Ny. darlingi and the number of people living in a 100m radius (OR = 1.02, p = 0.09), a significant increase in the odds of Ny. darlingi presence in active fish ponds (OR = 3.24, p = 0.03) was observed in bivariate models (Table III in S1 File). Neither variable was included in the final multivariate model. Active fish ponds were consistently positive for multiple Anophelinae species characterized in this study, including Ny. darlingi. However, the five most abundant species were also frequently collected from natural water bodies, particularly streams and rivers, highlighting the need for larval control strategies to target natural as well as artificial water bodies.
The current study identified emergent vegetation and lower light intensity as predictive for Ny. darlingi. Associations between Ny. darlingi larval habitats and various types of vegetation have been reported previously, including algae [20, 98], grassy and floating vegetation [54, 99], patches of detritus [22, 100–102], and submerged vegetation [103, 104]. Vegetation cover in and around larval habitats could provide food for larvae, shelter from predators, and favorable oviposition conditions for adults [14]. An association between lower light intensity and the presence of Ny. darlingi has been reported previously in Brazil [22], and Ny. darlingi have been consistently reported to oviposit in shaded or partially shaded habitats [54, 99, 102, 103]. This association could represent a direct effect of light exposure or temperature on development of the larvae, or an indirect effect on food sources or habitat stability [14].
This study was conducted in villages with very high rates of malaria transmission (Table VII in S1 File). We found that houses in the eight study villages that were located closer to Ny. darlingi-positive water bodies had more cases of malaria in both the rainy and dry seasons. Proximity to Ny. darlingi larval habitats has been identified as a malaria risk factor previously in the Amazon [24, 97, 105]. While it is clear that not all cases of malaria are acquired at home [50], this association indicates that larval source management targeting water bodies near villages could be employed as part of an integrated intervention strategy to reduce malaria risk in the peri-Iquitos region.
The most significant predictor for the presence of all four larval species characterized in this study was the presence of other Anophelinae species in the same water body. In addition, high affinity indices were seen between Ny. darlingi and Ny. rangeli, and between Ny. triannulatus s.l. and Ny. rangeli. This could indicate that the Anophelinae species in this study have similar requirements for larval habitats, or that the larvae interact synergistically in these water bodies. Similar affinities have been identified previously between Nyssorhynchus species [106–108], including between Ny. darlingi and Ny. triannulatus s.l. in Venezuela [109]. The co-occurrence of these putative vector species could simplify larval source management control strategies, as multiple species could be targeted in a single water body.
Though Ny. darlingi, Ny. rangeli, and Ny. triannulatus s.l. were found to co-occur often in larval habitats, we found differences in the larval microbiome between Ny. darlingi and the other two species collected from the same water bodies. This could reflect different niches occupied by each species within the same water body, leading to differences in the composition of bacterial species to which each mosquito species is exposed. Alternatively, there could be underlying biological differences in the microbiome among larval species [110]. The redundancy analysis results, which show that Ny. rangeli and Ny. triannulatus s.l. larvae were collected in ecologically similar habitats (Fig III in S1 File), provide support for the first possibility.
Our results contrast with Bascuñán et al.’s recent study characterizing the bacterial composition of Nyssorhynchus species, that found no differences between Ny. darlingi and Ny. nuneztovari s.s. adults collected from coastal Colombia [36]. It is possible that the microbiomes of Ny. darlingi and Ny. nuneztovari s.s. are more similar than those of the species in our study, or that species differences in the Nyssorhynchus microbiome occur in larvae, but not in adults. As it is clear that there is a shift in microbiome composition after adult emergence in African [27, 38, 111] and Neotropical [36] malaria vectors, future studies could determine whether the species difference that we identified is maintained in adult Nyssorhynchus mosquitoes. Previous research in other genera has found a greater impact of larval habitat than species on the mosquito larval microbiome [38–40]. However, these studies collected larvae from small water bodies, such as containers [39] and irrigation ditches/puddles [38, 40]. All larvae included for bacterial 16S rRNA sequencing in the present study were collected from fish ponds or streams; larvae could be exposed to a wider variety of bacteria in these larger water bodies, which could explain the lack of a larval habitat-specific bacterial signature.
Our study adds to an increasing literature characterizing the microbiome of Neotropical malaria vectors. There are overlaps between the bacterial composition of Ny. darlingi larvae in the present study and that reported in Ny. darlingi adults in past studies, including the presence of bacteria in the families Enterobacteriaceae [33, 35, 36], Moraxellaceae [35, 36], Aeromonadaceae [36], Rhodocyclaceae [36], and Comamonadaceae [36]. Furthermore, our identification of Thorsellia spp. bacteria in Ny. darlingi is consistent with the recent description of a new species of bacteria, closely related to Thorsellia, isolated from Ny. darlingi larvae [34].
Both Ny. rangeli [54, 108, 112] and Ny. triannulatus s.l. [113] have been described as habitat generalists; this is consistent with the lack of significant environmental predictors for the presence of each species found in the current study. We did find Ny. rangeli to be associated with water bodies with bushes nearby that were located farther from forests and closer to human habitations. Larvae of this species was previously collected from an eutrophized dam in Rio de Janeiro State, Brazil [114]; the current study further confirms the tendency of Ny. rangeli to oviposit in human-associated habitats. The association we detected between Ny. triannulatus s.l. and non-moving water is consistent with previous studies that found associations of this species complex with large, permanent water bodies such as lagoons and fish ponds, while the negative association we found between Ny. triannulatus s.l. and emergent vegetation contrasts with previous studies that associated this species with vegetation [104, 115–118].
From the Oswaldoi-Konderi complex, we report the presence of both Ny. sp. nr. konderi (which has previously been collected in Loreto [61]) and a single specimen of Ny. konderi in our study villages. The association we identified between Ny. sp. nr. konderi and shaded water bodies has been previously reported for members of the Oswaldoi-Konderi species complex in Suriname [99]; however, a recent report from the Brazilian Amazon found Ny. oswaldoi s.l. habitats associated with more sun exposure [104]. We also found an association between Ny. sp. nr. konderi presence and habitats with a higher forest cover and further from human habitations. This association has not been previously reported for Oswaldoi-Konderi species complex members, though previous studies have found Ny. oswaldoi s.l. associated with roads [54, 119]. It is possible that different species in this species complex have different larval habitat preferences. Incorporating molecular identification methods into future Anophelinae larval studies in the Neotropics could help to resolve these inconsistencies.
This study had several limitations. First, it was a relatively short-term exploratory study conducted in the absence of recent baseline data of the larval ecology of malaria vectors in the peri-Iquitos region. There is a need for more longitudinal sampling in this region so that larval habitats can be more thoroughly characterized and seasonal differences explored. Second, we sampled only the water bodies we were able to identify by satellite imagery and ground-truthing within a 1km radius of each village. Future studies using more sophisticated technology such as drones [120] could help identify additional water bodies. Third, our larval dipping methods, though standard, may not have captured the entire species diversity in each water body in each collection. Neotropical Anophelinae larvae have been successfully collected using methods that sample the interior of large water bodies in addition to the perimeter [104], perhaps providing a more complete sampling of the larvae present. However, of the larvae that we collected, we were able to identify over 90% using a combination of morphological and molecular methods. Fourth, malaria cases included in this study were obtained from passive case reporting using health post data, which likely underestimates the overall malaria burden in these villages. Fifth, our larval microbiome analysis is limited by the small sample size and number of water bodies and species represented. Future studies including more comprehensive sampling could help to confirm the species differences we identified. Furthermore, the use of alternative analysis techniques, such as the use of exact sequence variants rather than OTUs [121], should be explored.
In this longitudinal study, we described ecological characteristics of the larval habitats of Anophelinae malaria vectors in eight villages on four river systems and a highway in the peri-Iquitos region. Nyssorhynchus darlingi, the primary regional malaria vector, was collected in both natural and artificial water bodies in all eight villages throughout the fifteen-month study period. Nyssorhynchus darlingi-positive water bodies were associated with more recent deforestation, a lower vegetation index, lower light intensity, and emergent vegetation, as well as the presence of other Anophelinae species. Despite the high co-occurrence of Anophelinae species in water bodies, we found that Ny. darlingi larvae had a distinct microbiome compared with Ny. rangeli and Ny. triannulatus s.l. larvae. Houses in the study area with more malaria cases were located closer to identified Ny. darlingi larval habitats. Our findings highlight the potential for larval source management to be a successful control measure in the peri-Iquitos region, as well as the continuing need to better understand the larval ecology of malaria vectors in the heterogeneous Amazon basin landscape so that these control efforts can be more efficiently targeted to reduce the risk of malaria.
|
10.1371/journal.pntd.0004625 | First Chikungunya Outbreak in Suriname; Clinical and Epidemiological Features | In June 2014, Suriname faced the first Chikungunya outbreak. Since international reports mostly focus on hospitalized patients, the least affected group, a study was conducted to describe clinical characteristics of mainly outpatients including children. In addition, the cumulative incidence of this first epidemic was investigated.
During August and September 2014, clinically suspected Chikungunya cases were included in a prospective follow-up study. Blood specimens were collected and tested for viral RNA presence. Detailed clinical information was gathered through multiple telephone surveys until day 180. In addition, a three stage household-based cluster with a cross-sectional design was conducted in October, December 2014 and March 2015 to assess the cumulative incidence.
Sixty-eight percent of symptomatic patients tested positive for Chikungunya virus (CHIKV). Arthralgia and pain in the fingers were distinctive for viremic CHIKV infected patients. Viremic CHIKV infected children (≤12 years) characteristically displayed headache and vomiting, while arthralgia was less common at onset. The disease was cleared within seven days by 20% of the patients, while 22% of the viremic CHIKV infected patients, mostly women and elderly reported persistent arthralgia at day 180. The extrapolated cumulative CHIKV incidence in Paramaribo was 249 cases per 1000 persons, based on CHIKV self-reported cases in 53.1% of the households and 90.4% IgG detected in a subset of self-reported CHIKV+ persons. CHIKV peaked in the dry season and a drastic decrease in CHIKV patients coincided with a governmental campaign to reduce mosquito breeding sites.
This study revealed that persistent arthralgia was a concern, but occurred less frequently in an outpatient setting. The data support a less severe pathological outcome for Caribbean CHIKV infections. This study augments incidence data available for first outbreaks in the region and showed that actions undertaken at the national level to mount responses may have positively impacted containment of this CHIKV outbreak.
| Chikungunya virus is transmitted to humans by mosquito bites and causes fever and joint pain. Chikungunya was first detected in Africa, but recently became a worldwide concern with outbreaks in many (sub)-tropical countries. We report the characteristics of the first outbreak in Suriname (2014–2015). Mainly non-hospitalized patients were followed-up to study the clinical manifestations and course of the disease, after presentation in the respective clinics with the standard Chikungunya symptoms (fever and arthralgia). Twenty percent of follow-up patients could clear the disease within one week and 22% (mostly women and elderly) still had complaints about arthralgia up to 6 months after infection. This is consistent with the assumption that Caribbean Chikungunya viral infection has a less severe pathological outcome. Furthermore, more insight was gained into the symptomatology of children (≤12 years). In addition, house-to-house surveys in Paramaribo were carried out to identify suspected cases to assess the incidence. Almost 25% of the survey participants experienced symptoms consistent with Chikungunya during the nine months spanning the investigation. The launch of a governmental campaign to eliminate mosquito breeding sites coincided with a sharp decline of Chikungunya cases, suggesting that such measures may be important in the containment of future CHIKV outbreaks.
| Chikungunya fever is caused by a classical arbovirus (genus Alphavirus, family Togaviridae), which is transmitted to humans primarily through Aedes aegypti and Aedes albopictus mosquitoes [1]. Acute onset of fever and polyarthralgia, mainly affecting the extremities (wrists, ankles, phalanges), are the primary reported clinical characteristics [2, 3]. Joint pain is often severe [4] and arthralgia may persist for weeks to years [5, 6]. Other reported symptoms include rash, headache and back pain [1, 7]. Despite the low hospitalization rate of Chikungunya patients (0.3% during the outbreak in La Reunion in 2005–2006 [8]), at present more is known about the clinical presentation and outcome (i.e. recovery or persistent pain) of hospitalized patients during outbreaks.
Chikungunya is endemic in tropical Africa, South-East Asia and on the Indian subcontinent [1, 3].
In recent years, outbreaks have been appearing outside the endemic zone probably due to increased global air travel and seaborne trade [9–12] and Chikungunya has recently emerged as a major public health concern in the Caribbean Region [13]. In December 2013, the first Chikungunya virus (CHIKV) infections were reported among non-travelers on the Caribbean island of Saint Martin [14]. Since then, the virus has spread rapidly into the Caribbean region and neighboring countries [15]. The Chikungunya viruses studied from this region belonged to the Asian genotype [16]. Recently, it was demonstrated that this strain causes a less severe pathological outcome compared to the East Central South African (ECSA) genotype [17].
In June 2014, the first locally acquired case of Chikungunya was reported in Suriname [18]. In this period, no other outbreaks of vector-borne diseases (e.g. yellow fever, malaria, leptospirosis) were recorded by the Bureau of Public Health, BOG (BOG Annual Report). A study was initiated with two objectives, firstly to assess the manifestation and course of Chikungunya infection following acute illness in a naïve population, and secondly to determine the cumulative incidence of the Chikungunya outbreak in Suriname. To assess the extent and severity of the symptoms during this first outbreak, patients with suspected CHIKV diagnosis were enrolled in the period between August and September 2014 in three different outpatient settings. CHIKV presence was determined with Real-Time reverse transcription-polymerase chain reaction (RT-rPCR). Detailed clinical information from Chikungunya infected patients was collected and the symptom development was recorded through multiple telephone interviews until 6 months after onset.
To determine the cumulative incidence of this first Chikungunya outbreak in Suriname, a community surveillance consisting of three successive household-based cluster investigations was conducted in the capital in October 2014, December 2014 and March 2015, gathering information on 1169 households with 4842 participants. The validity of self-reported CHIKV infections was crosschecked with serological analysis.
To our knowledge, the study is unique in describing in detail the clinical evolution of autochthonous CHIKV outpatients in a Caribbean country. Furthermore, clinical follow-up was conducted for six months for almost one hundred viremic CHIKV infected patients including 10 children (2–12 years). We therefore also add to the records internationally available for clinical manifestation of Chikungunya in children. The prospective follow-up study was complemented with cluster-based cross-sectional data, adding valuable data to the international Chikungunya outbreak information.
Suriname is a tropical country, located along the North Coast of South-America, bordering Brazil to the south, Guyana to the west and French Guiana to the east. Approximately 80% of Surname is covered by tropical rainforest. Suriname has a highly multiethnic population of nearly 550,000, most of whom live in the coastal area in and around the capital Paramaribo [19]. This study took place in Paramaribo and in Commewijne, a rural district adjacent to Paramaribo.
All participants provided oral informed consent. Acquired informed consent was registered prior to inclusion either by the attending physician on the short questionnaire in the clinical setting or by the trained pollster in the community surveillance. Participants, whose blood specimens were shipped abroad for testing, provided written informed consent. The national ethics committee within the Ministry of Health approved both studies (VG018-14 and VG008-15).
All reported signs and symptoms are presented descriptively, namely the presence/absence of arthritis, periarticular edema, arthralgia, fever, myalgia, skin rash, itching, headache, back pain, eye pain, vomiting, nausea, fatigue, asthenia and joint pain intensity and location. The Chi-square test was used to compare characteristics and symptoms of the participants. To evaluate differences in age distribution between the viremic CHIKV infected and the CHIKV- group, the Mann-Whitney U test was used.
The Statistical Packages for Social Sciences (SPSS 21.0) were used for analysis excluding observations with missing data. Statistical significance was set at p = 0.05.
Between October 2014 and March 2015, a total of 4842 participants (1637, 1583 and 1622 in the three surveys respectively) from 1169 households (385, 392, and 392, respectively) in all 12 regions in Paramaribo were included. The random distribution of households and the cross-sectional study design allowed inference and valid analysis.
The data presented encompass the whole study period. The gender distribution of this survey followed the general Surinamese population distribution (m/f ratio: 1.1 vs 1.0). The limited number of participants in each of the twelve regions in Paramaribo did not allow for comparisons per region.
After the emergence of CHIKV infections in the Caribbean region in 2013–2014 [15], autochthonous infections were continuously reported in Suriname since June 2014, as illustrated in this study. The prospective follow-up study included 180 symptomatic patients with 68% testing positive for viral RNA (viremic CHIKV infected patients), of which 70.8% had severe joint pain. At disease onset 73.5% viremic CHIKV infected patients had fever and 84.4% had arthralgia which was still reported at D180 for 22.2%.
The combination of fever and severe arthralgia/arthritis in the clinically suspected sample population (68% RT-rPCR positive), demonstrated that with these cardinal symptoms, general practitioners in Suriname were able to correctly diagnose the majority of the patients during this CHIKV outbreak. The fact that only RT-rPCR was used to define CHIKV infected patients may have induced a classification bias, especially in the context of the Asian lineage strain, circulating in the Caribbean Region [17]. However, we still observed clear differences between viremic CHIKV infected and CHIKV- patients. The manifestation of arthralgia in the fingers was characteristic for viremic CHIKV infected patients, while CHIKV- patients had more frequent pain in the eyeball compared to viremic CHIKV infected patients, and less frequent arthralgia. Moreover, rash in CHIKV- patients was mostly observed at D5 and D6, in contrast to the viremic CHIKV infected patients with rash mostly at D3. These symptoms may point towards Dengue virus infection, a common infection in South-America. Furthermore, the CHIKV outbreak substantiates the vector presence and Dengue is clinically difficult to differentiate from Chikungunya [23]. In the clinical differentiation, the symptoms arthralgia and onset of rash may therefore be good markers to differentiate Chikungunya from other exanthematous diseases. Besides, additional laboratory tools as leukocyte and thrombocyte count could be utilized, since patients with Chikungunya often have lymphocytopenia which is seldom seen in Dengue patients [24], whereas thrombocytopenia as seen in patients with Dengue hemorrhagic fever is not common in Chikungunya.
The gathered detailed information about daily clinical symptoms of acute disease and chronic illness of viremic CHIKV infected outpatients in Suriname revealed that the most common reported symptoms were abrupt onset of fever, arthralgia, asthenia and myalgia. These findings, in particular the very high frequency of fever and the incapacitating peripheral pain in multiple smaller joints (i.e. ankles, hands, feet, knees) matched the main reported features of CHIKV outpatients and hospitalized patients in other regions [3, 7, 25, 26]. However, the frequency of pain by location (ankles 54.3%, hands 48.9%, feet 47.8%, knees 40.2%) during acute CHIKV infection was lower than those reported by the TELECHIK cohort study [27] (ankles 74.9%, hands 75.7%, feet 73.1%, knees 67.6%) and the French soldiers cohort (ankles 68%, fingers and palms 76%, feet 68%, knees 58%) [28]. This finding is consistent with the presumed circulation of the Asian lineage in Suriname, which is less virulent than the ECSA strain circulating in the latter population-based studies in La Reunion [17].
The overall assigned score of pain intensity in other studies was generally high (i.e. NRS score ≥7) [25] as was corroborated by our findings (score 8). The intense joint pain caused walking difficulties in almost all our viremic CHIKV infected patients (92.1%), which is even higher than during the outbreak in La Reunion (2005–2006), where 46.4% to 75.0% [21, 25] of the viremic and/or serologically-confirmed patients reported discomfort in performing daily activities such as walking.
The presence of periarticular edema which was described earlier [29], was also observed at the day of consultation in 25.5% of individuals positive for CHIKV-RNA in our cohort. This corresponds with a hospital-based study (periarticular edema in 25.6% RT-rPCR CHIKV+ patients) [30] and another study from La Reunion where 30% of patients presented with soft-tissue swelling [1, 7].
The presence of acute arthritis in 35.8% of our viremic CHIKV infected cohort is lower in comparison to the French soldiers cohort study from La Reunion with 44.8% reporting polyarthritis [28], as could be expected from the Asian lineage. Moreover, a study from India (Maharashtra State) even observed arthritis in 68.8% [31].
Symptoms improved gradually during the acute phase, except for skin rash and itching, consistent with previous reports [4, 32]. The presence of skin rash, in 37.9% of our patients, falls within the broad range that is reported worldwide (10% to 81%) [33, 34]. The occurrence of skin rash towards the end of the febrile phase [35] is substantiated by the peak presence on day 3 in our study. The general observations on maculopapular rash [35], mostly reported on limbs and trunk, rarely affecting the face and occasionally spreading over the entire body was corroborated by our findings.
At day 7, fever receded for most patients, but asthenia and arthralgia were still reported by more than 40% of the patients, in coherence with the study of Thiberville et al. [4]. This study also reported a considerable percentage of patients with headache and myalgia at day 7, in contrast to our findings.
Twenty percent of our cohort was symptom-free within 7 days after disease onset, which seems slightly more favorable than in La Reunion with a duration of symptoms <15 days for 23% of the cases [25].
The manifestation and the course of CHIKV symptoms are variable and depend on several factors such as virus strain, age, gender, immune status [32, 36, 37] and possibly also on the genetic predisposition [38]. The latter is supported by the finding that Maroons were underrepresented in our study. This corresponds with a recent study where hospitalizations due to Dengue, which is transmitted by the same vector as Chikungunya, occurred least in the Maroons [39]. However, the underrepresentation of Maroons in our study could also be due to either the low population density of this ethnic group in at least one of our study sites (Commewijne) or other factors as differences in health care access.
Earlier studies report that women are more prone to CHIKV infection [31, 33, 40, 41], consistent with our results, probably because of greater home-based activities and different clothing behaviors enabling an increased accessibility for mosquitoes. This is further supported by a report in the Midwest region of Brazil, where women were significantly more frequent victims of Dengue [42]. On the other hand, a cross-sectional study in Mayotte (Indian Ocean), found that CHIKV seroprevalence was higher in men than women [43] and during an outbreak in Italy no difference in sex incidence was noticed [12]. The inconsistency of sex as factor in exposure to viral infection across countries/communities may be related to different lifestyles and behaviors.
The occurrence of Chikungunya viral infections in Suriname in all age groups substantiates the general finding. Moreover, the observation that children were the least affected group during first epidemic periods in La Reunion (18% of the cases <10 years) and India (Maharashtra State: 5% of the cases <15 years) [31, 41] was also corroborated with our findings (10.7% ≤12 years). In line with recent studies [44], arthralgia (50% at onset) had a milder course in the children younger than 13 years and back pain was less common (14.3% at D3). Most studies report a low frequency of headache in children (15% to 35.3%) [44, 45], whereas a higher frequency of headache for children was reported by our cohort (85.7% at onset). The frequency of vomiting (28.6% at D0) falls within the reported range from studies in India (12.2% to 47%) [33, 45]. However, these studies did not report our observation of a significant higher frequency of vomiting for children compared to adults at onset. This discrepancy could be due to different study settings. The presence of maculopapular rash in 44.4% of the children at D4 was in line with earlier reports (33–60%) [44]. However, the earlier reported more prominent presence of skin rash in children versus adults during a CHIKV outbreak in southern Thailand [24], could not be substantiated in this study. The standard triad of fever, joint pains and rashes for the diagnosis of Chikungunya in children may not be as effective in Suriname. The low number of children included in our study however could obscure this presumption.
Persistent arthralgia was described in La Reunion and Italy in more than half of the Chikungunya patients in outpatients and hospitalized patients [21, 25, 46], while in our setting, arthralgia was observed in 22% of the patients six months after disease onset. In contrast to our study, patients in the aforementioned studies reported underlying osteoarthritis or pre-existing rheumatic diseases, which are independent markers for persistent pain [21, 46]. A recent retrospective study from Columbia also reported a higher percentage of persistent polyarthralgia (44.3%) [47]. Our results are more in line with another study in Reunion Island including only outpatients, where the residual arthralgia was significantly lower (23%) after 300 days [4]. Our results are also consistent with a study done in Indonesia, where CHIKV infections by the Asian genotype caused mild and short lasting clinical symptoms [48]. An even lower percentage of only 13.3% of residual arthralgia, 2 to 3 months after illness, was observed in a study in Singapore, with patients without underlying medical conditions [32].
Other variables associated with persistent joint pain are age ≥ 45 years and gender (women are more likely to have persistent arthralgia) [4, 25, 26, 49], as was corroborated by our results.
The frequency of patients with continuous pain (25.0%) was lower than reported by the French soldiers cohort (41.3%) [28]. Moreover, in this cohort 93.7% of the symptomatic patients still complained about chronic pains 6 months after infection (our study: 22%). This difference could be caused by the less pathological outcome of the Asian lineage presumably present in Suriname compared to the Indian Ocean strain [17].
The reported relapse of arthralgia in some Chikungunya patients in the months after acute infection [21, 23], was also observed in our study (8.9%). However, relapse was less common in Suriname than in the hospitalised-based study from Borgherini et al. (21%) [21]. Moreover, in the population-based French soldier cohort, the more virulent ECSA strain caused a relapse in 58.7% of the patients [28].
Our investigation of the clinical manifestations had some limitations. Firstly, we only performed qualitative PCR analysis and did not determine the average viral load at inclusion of the study. We could therefore not relate viral load to symptomatology and severity of symptoms. Secondly, clinical symptoms were only described for Chikungunya infections in patients consulting a physician in the initial stage of disease. The described symptoms are therefore expected to be more severe than the general symptoms, since less severe or asymptomatic cases were not included. Thirdly, symptomatic differences between viremic CHIKV infected and CHIKV- patients could be obscured, since symptoms of CHIKV- patients were characterized only in the acute phase. Furthermore, no serology was performed to differentiate between real negative CHIKV cases and potential CHIKV+ cases in a late stage with no detectable viremia. These reasons may have skewed the estimation of some symptoms in this cohort. However, because of the robust size of the viremic CHIKV infected sample (2 fold higher than the CHIKV- group), we feel that our conclusions about the viremic CHIKV infected outcome and symptoms are well supported.
The prospective follow-up study was complemented with cluster-based cross-sectional data and the cumulative incidence in Paramaribo from July 2014 to March 2015 was 249 cases per 1000, which by inference is similar to the cumulative incidence of La Reunion in 2005/2006 (350 cases per 1000) over a 12-month period [41]. Caution is warranted in the country comparison for epidemic impact, since several factors as virus strain, influence of the weather, the density of the population and the effectiveness of vector control measures are involved.
The high value of serologically confirmed CHIKV infections (90.4% IgG) in self-reported cases clearly demonstrated the acquired skill of the population to recognize and diagnose CHIKV infection.
Internationally, not much data is available about the health care seeking behavior during Chikungunya outbreaks. In Mayotte, 52% of participants with confirmed CHIKV sought medical advice [50], while 79.5% of the CHIKV suspected persons in our study population sought medical care. The good access to health care in Paramaribo may account for this high percentage.
In the community surveillance, more women self-reported to be Chikungunya positive, underlining the observation among the laboratory confirmed CHIKV cases that women are more prone to CHIKV infection. Moreover, also the health care seeking behavior was higher for women than men, corroborating the results of a general study on the utilization of health care services in the USA [51] and underlining the generally poor recovery observed in women.
The use of insect repellent (47.4%) in Paramaribo is lower than in outbreaks in La Reunion (67.9%) [52] and India (Chennai) (88.4%) [53], despite the fact that in contrast to Reunion Island, Suriname was warned for an upcoming Chikungunya outbreak. However, the latter two studies reported individual use of insect repellent while in our study household data were gathered. Moreover, country comparisons may not be very useful without data of the mosquito index and data on actual repellent use.
In Suriname, the highest CHIKV incidence was noted in the long dry season, which may have been associated with increased storage of water filled objects in dry periods, positively influencing mosquito breeding and thus CHIKV transmission. This premise is further supported by the observation that the sharp increase in CHIKV cases occurred five weeks after the dry season had set in, similar to the largest Dengue epidemic in Suriname in 2005 (personal communication) and in coherence with findings in Thailand, where Chikungunya peaked in several provinces 6 weeks after the start of the dry period [54].
The sharp decline of CHIKV after week 40, coincided with the start of the governmental campaign to eliminate mosquito breeding sites, which may have positively impacted the containment of this outbreak. These results thus support country actions to increase effectiveness of vector control and to raise population awareness.
In conclusion, this is the first report about the emergence of CHIKV in Suriname. The clinical course of most symptoms in our naïve cohort was similar to those of other countries that faced a Chikungunya outbreak. The earlier finding that women and elderly persons are more at risk for persistent arthralgia was substantiated, although our study highlighted that persistent arthralgia, mostly intermittent, is a less frequent concern among viremic CHIKV infected patients in an outpatient setting. Furthermore, our data support the presumption of the circulation of the Asian lineage of CHIKV in Suriname. Our findings also provided more insight into the manifestation and course of this arboviral disease in children. Still, data from larger children cohorts in CHIKV affected areas are required to further enhance patient management. We also observed some clinical differences between viremic CHIKV infected and CHIKV- patients, which could be useful to differentiate Chikungunya from other diseases (such as exanthematous diseases as Dengue) in resource-limited countries lacking testing facilities. The finding that 20% of the patients cleared the disease within one week, while 22% still displayed persistent arthralgia after six months offers valuable indicators for countries facing their first outbreak. Furthermore, our study adds to the international data on cumulative Chikungunya incidence during first outbreaks, which is particularly important for the Caribbean region.
|
10.1371/journal.pgen.1002597 | Cis-by-Trans Regulatory Divergence Causes the Asymmetric Lethal Effects of an Ancestral Hybrid Incompatibility Gene | The Dobzhansky and Muller (D-M) model explains the evolution of hybrid incompatibility (HI) through the interaction between lineage-specific derived alleles at two or more loci. In agreement with the expectation that HI results from functional divergence, many protein-coding genes that contribute to incompatibilities between species show signatures of adaptive evolution, including Lhr, which encodes a heterochromatin protein whose amino acid sequence has diverged extensively between Drosophila melanogaster and D. simulans by natural selection. The lethality of D. melanogaster/D. simulans F1 hybrid sons is rescued by removing D. simulans Lhr, but not D. melanogaster Lhr, suggesting that the lethal effect results from adaptive evolution in the D. simulans lineage. It has been proposed that adaptive protein divergence in Lhr reflects antagonistic coevolution with species-specific heterochromatin sequences and that defects in LHR protein localization cause hybrid lethality. Here we present surprising results that are inconsistent with this coding-sequence-based model. Using Lhr transgenes expressed under native conditions, we find no evidence that LHR localization differs between D. melanogaster and D. simulans, nor do we find evidence that it mislocalizes in their interspecific hybrids. Rather, we demonstrate that Lhr orthologs are differentially expressed in the hybrid background, with the levels of D. simulans Lhr double that of D. melanogaster Lhr. We further show that this asymmetric expression is caused by cis-by-trans regulatory divergence of Lhr. Therefore, the non-equivalent hybrid lethal effects of Lhr orthologs can be explained by asymmetric expression of a molecular function that is shared by both orthologs and thus was presumably inherited from the ancestral allele of Lhr. We present a model whereby hybrid lethality occurs by the interaction between evolutionarily ancestral and derived alleles.
| When two different species mate, the hybrid progeny are often sterile or lethal. Such hybrid incompatibilities cause reproductive isolation between species and are an important mechanism for maintaining species as separate units. A gene called Lethal hybrid rescue (Lhr) is part of the cause of hybrid lethality between Drosophila species. Like many other hybrid incompatibility genes, Lhr protein sequences in the hybridizing species have diverged from one another by natural selection. This and other findings led to the hypotheses that the function of Lhr has changed between the two species, and this is what makes Lhr a hybrid lethality gene. Using a series of genetic, molecular, and cytological assays, we report evidence contrary to these hypotheses, that hybrid lethal activity is instead a function shared by both species and inherited from their common ancestor. This result is particularly surprising because the Lhr genes from the two species have different effects on hybrid viability. We discovered that these differential effects are caused by differences in expression levels of Lhr in hybrids rather than by changes in its protein-coding sequence. Our results demonstrate that, while natural selection may be important in evolving hybrid incompatibilities, how it does so in this case remains mysterious.
| Species can be isolated from one another by a variety of reproductive barriers. One widely observed barrier is hybrid incompatibility (HI), the inviability or sterility of interspecies offspring. The key premise of the Dobzhansky-Muller (D-M) model explaining the evolution of HI is that genetic changes fixed in one population need not be compatible with changes fixed in a different population [1], [2]. This is most commonly illustrated as two independently evolving populations that each diverge from the ancestral state and fix new alleles. Hybridization between the two populations brings together the independently derived alleles, thereby generating a genotype unscreened by natural selection. This genotype may suffer from an incompatible interaction between the derived alleles, resulting in developmental breakdown of the hybrid progeny. A key feature of this model is that HI alleles have diverged in sequence and function (perhaps extensively) from their ancestral states. A second important prediction of the model is asymmetry: Gene “A” from species one may interact with gene “B” from species two to cause HI, but not vice-versa [3]. Questions fundamental to understanding speciation then are: What molecular divergence between the ancestral and derived alleles is causing HI? Is this divergence at the level of regulatory or structural changes? What are the evolutionary forces causing this divergence?
One unifying emerging trend is that HI loci often show high levels of divergence caused by natural selection [3], [4]. These findings are exciting, because if molecular divergence created by selection is causing HI, then the phenotypic target of selection is, at least in part, the evolutionary basis of speciation. A major goal then is to understand the role of selection in the evolution of incompatible divergence. Interestingly, studies on several recently characterized HI genes implicate divergence of heterochromatin and heterochromatin-binding proteins as the cause of incompatibility [5], [6]. As heterochromatin is the graveyard of selfish genetic elements, this functional divergence could be the legacy of genetic conflicts between the host species and the invasion of selfish DNAs such as transposable elements and satellite DNAs [7], [8].
A variation of the D-M model suggests that HI can also be caused by interactions between alleles that have not diverged from the ancestral state and derived alleles that have diverged in only one lineage [9]. If an HI allele has not diverged from its ancestral state, then this model predicts that its HI effects will be symmetrical, with orthologs from both species contributing to HI. Several examples of ancestral-derived incompatibilities have been discovered, and consistent with expectations the HI genes, when known, have experienced limited sequence divergence [10]–[12].
On the other hand, the expectation of a strict dichotomy between ancestral and derived HI alleles may reflect an over-simplified view of HI. Hybrids are the sum of two independently evolving genomes and thus suffer from multiple suboptimal interactions [3]. For example, species-specific divergence at cis and trans-regulatory elements is associated with widespread transcriptional dysregulation in hybrids [13], [14]. This creates a genetic background distinct from either parental species, and several well-studied HI genes have genetic properties in hybrids that are significantly different from or even opposite to their intraspecific roles [3].
Crosses between D. melanogaster females and D. simulans males produce inviable hybrid sons and sterile hybrid daughters [15]. The incompatible D-M interaction in hybrid males can in part be explained by the interaction between two genes, Hybrid male rescue (Hmr) on the D. melanogaster X-chromosome and Lhr on the D. simulans 2nd chromosome [16]. A loss of function mutation in either Hmr or Lhr alone is sufficient to suppress the lethality of hybrid sons [16]–[19]. Thus it is the activity of these genes that causes hybrid breakdown.
Lhr (also known as HP3) encodes a protein that localizes to heterochromatin by directly binding to Heterochromatin Protein 1 (HP1) [16], [20], [21]. Population genetic analyses demonstrated that the Lhr protein coding sequence (CDS) has diverged extensively between D. melanogaster and D. simulans under positive selection, leading to the suggestion that Lhr has co-evolved with species-specific heterochromatin sequences [16]. If this co-evolution reflects a history of genetic conflict then one might predict that hybrid lethality is caused by defects in heterochromatin structure or maintenance, and that Lhr orthologs have functionally diverged in their heterochromatin localization properties such that they would mislocalize in the presence of heterochromatin from different species.
The hybrid lethality gene Lhr appeared initially to be a clear example of a derived D-M hybrid incompatibility locus. Consistent with the expectation of functional divergence, we previously found that the rescue of hybrid lethality via Lhr is asymmetric; removal of D. simulans Lhr (sim-Lhr) rescues lethal hybrid sons but removal of D. melanogaster Lhr (mel-Lhr) does not [16]. Surprisingly, however, Lhr orthologs from D. melanogaster, D. simulans and the outgroup species D. yakuba all have hybrid lethal activity when overexpressed in hybrids [21]. LHR proteins from these species also retain heterochromatic localization when expressed in polytenized salivary-gland cells, demonstrating that natural selection has not caused a wholesale change in Lhr function. This set of results suggests either that functional divergence is not an all-or-none property, or that Lhr is an ancestral HI locus, rather than a derived one.
To distinguish between these two possibilities, and to uncover the functional divergence underlying the asymmetric rescue properties of Lhr orthologs, we developed a native-promoter driven transgenic system that allows a sensitive comparison of the functions and localization properties of D. simulans and D. melanogaster Lhr orthologs. Using this system, we have compared Lhr function in both pure species and hybrids using three sets of experiments: (1) genetic tests for hybrid lethal activity and interaction with its D-M partner, Hmr; (2) detailed cytological mapping of the heterochromatic localization of LHR and its association with hybrid lethality, and (3) expression analysis comparing transcriptional levels of the Lhr orthologs.
We generated parallel strains of D. melanogaster containing either D. simulans Lhr (sim-Lhr) or D. melanogaster Lhr (mel-Lhr) transgenes using the φC31 site-specific integration system [22]. Each Lhr ortholog was C-terminally tagged with an HA epitope and was expressed under the control of its native regulatory sequences (Figure 1). The transgenic constructs contained the eye-color marker white+ and were each integrated into the attP2 site on the third chromosome. We tested the transgenes for wild type activity by assaying for complementation of the D. simulans Lhr1 hybrid rescue mutation. D. simulans Lhr1 is a loss-of-function mutation that acts as a dominant suppressor of hybrid lethality [16], [18]. Complementation here means that the transgene provides sufficient wild type Lhr activity to suppress rescue by the Lhr1 mutation, thus causing hybrid male inviability.
Complementation tests were performed by crossing D. melanogaster mothers heterozygous for an Lhr-HA transgene to D. simulans Lhr1 fathers. This cross generates two classes of hybrid sons: the control class that lacks the transgene and has white eyes, and the experimental class that inherits the transgene and has orange eyes. Complementation is detected as the lethality of orange-eyed sons. If hybrid lethal activity partitions discretely between Lhr orthologs, as expected from the functional divergence interpretation of genetic asymmetry, sons inheriting the φ{Dsim\Lhr-HA} transgene should be lethal, while those inheriting φ{Dmel\Lhr-HA} should be viable.
Unexpectedly, both transgenes fully complemented the D. simulans Lhr1 mutation (Table 1, crosses 1 thru 4), suggesting that both D. simulans and D. melanogaster Lhr orthologs have hybrid lethal activity. As this result was contrary to expectation we tested several possible causes of artifacts. First, the C-terminal HA-tag does not affect Lhr function because untagged versions of both mel-Lhr and sim-Lhr also complement Lhr1 (Table 1, crosses 5 and 6). Second, the adjacent gene Bap55 present in these constructs is not responsible for complementation because a modified mel-Lhr-HA transgene, φ{ΔBap55 Dmel\ Lhr-HA}, in which the Bap55 CDS is interrupted by two stop codons and a frameshift mutation, also complements Lhr1 (Table 1, cross 7). Third, the results are not caused by other unknown aspects of the strain background or by the attP2 site because the attP2 site itself without an integrated transgene does not complement Lhr1 (Table 1, cross 8). Furthermore mel-Lhr-HA integrated into a different site (attP86Fb) also complements Lhr1 (Table 1, cross 4). Fourth, these results are not due to an over-expression artifact because data presented below demonstrate that the mel-Lhr-HA transgene expresses Lhr at a level similar to the endogenous wild type locus (see section “cis-by-trans regulatory divergence causes functional divergence of D. melanogaster and D. simulans Lhr” below). These results clearly show that D. melanogaster Lhr has hybrid lethal activity even when expressed at its wild type level.
How can these results be reconciled with the original observation that only a mutation in D. simulans Lhr, and not the D. melanogaster ortholog, rescues hybrid sons? Those experiments were done in hybrid genotypes that had only a single dose of either mel-Lhr or sim-Lhr [16]. In contrast, the experiments here were performed by adding a transgenic copy of either mel-Lhr or sim-Lhr to hybrids that also carried the endogenous chromosomal copy of mel-Lhr. Increased dosage of mel-Lhr in the current experiments may therefore explain why we have not observed a difference between the mel-Lhr and sim-Lhr transgenes. This hypothesis raises the question of whether the hybrid lethal activity of the mel-Lhr-HA transgene would be eliminated in a background lacking the chromosomal copy of mel-Lhr. To test this we crossed D. melanogaster mothers that were doubly heterozygous for the mel-Lhr-HA transgene and an Lhr− deficiency to D. simulans Lhr1 fathers. If transgenic mel-Lhr behaves identically to the endogenous locus, then hybrid sons inheriting the Lhr− deficiency along with the mel-Lhr transgene should be equivalent in Lhr dosage to rescued +/Lhr1 hybrid males and thus be viable. However, hybrid sons from this cross were also inviable (Table S3). This result indicates that the mel-Lhr-HA transgene does not precisely phenocopy the native chromosomal mel-Lhr locus. In the Discussion we consider possible causes of this difference.
Because the complementation tests did not reveal a difference in the hybrid lethal effects of Lhr orthologs we used a more sensitive genetic assay to test for functional divergence between mel-Lhr and sim-Lhr. We previously demonstrated that Lhr-dependent hybrid lethality requires the presence of its D-M partner, the D. melanogaster gene Hmr [16].
We reasoned that the hypomorphic allele Hmr1 might exhibit different sensitivities to the HI effects of the different Lhr alleles, but that the null allele Df(1)Hmr− would not. We therefore introduced each of our Lhr transgenes into these Hmr mutant backgrounds and tested the effect of the transgenes on hybrid male viability in crosses to D. mauritiana and D. simulans. Crosses with the sim-Lhr-HA transgene recapitulated our previous experiments: Hmr1 hybrid males carrying sim-Lhr-HA were essentially inviable at room temperature and showed strongly reduced viability at 18°C, while Df(1)Hmr− hybrid males were equally viable with and without the transgene (Table 2). We then performed similar crosses with mel-Lhr-HA. This transgene had little effect on viability of males with the null mutation Df(1)Hmr− and the results were in general not significantly different compared to the crosses with sim-Lhr-HA (Table 2, sets 1 & 2). In crosses with the hypomorphic mutation Hmr1, hybrids carrying mel-Lhr-HA had reduced viability compared to their non-transgene carrying siblings, particularly at room temperature. Strikingly, we found that in all four cross conditions the magnitude of the viability reduction was significantly less for mel-Lhr-HA compared to sim-Lhr-HA (Table 2, sets 3 & 4). These data demonstrate that sim-Lhr is more potent than mel-Lhr in creating the hybrid lethal interaction with Hmr, and that our Lhr transgenes thus do in fact reveal a significant degree of functional divergence.
Having demonstrated that wild type mel-Lhr has hybrid lethal activity, we reinvestigated whether removal of mel-Lhr has any detectable hybrid rescue activity. We previously showed that deletion of mel-Lhr does not rescue hybrids with D. simulans [16]. We therefore looked for rescue in hybrids with D. mauritiana at 18°C, conditions that are maximally conducive for hybrid viability [17]. Unrescued hybrid males die as larvae [23]. We found that two D. melanogaster Lhr− deletions rescued 7–21% of males to the pharate adult stage (Table 3). This is clearly a modest rescuing effect and did not occur in one of the genetic backgrounds tested (Df(2R)BSC49 crossed to D. mauritiana W139), but it is significant because crosses with 45 other deletions across chromosome 2R gave no rescue. A third Lhr− deletion, Df(2R)BSC44, did not rescue hybrids, demonstrating that hybrid viability is sensitive to genetic background effects. The difference in magnitude of rescue for deletion of mel-Lhr versus sim-Lhr further supports our conclusion using transgenes that sim-Lhr has greater hybrid lethality activity than mel-Lhr.
We next set out to determine why sim-Lhr is more potent than mel-Lhr in causing hybrid lethality. Coding sequence evolution leading to different protein localization patterns is one possible cause of Lhr functional divergence. In order to test this hypothesis we examined the cellular localization of LHR orthologs in their wild type backgrounds using our Lhr transgenes. In D. melanogaster LHR protein is most abundant during embryogenesis (Figure S1). We therefore analyzed the distribution of LHR during early embryogenesis and found a cyclical on-off pattern through the cell cycle, with localization to chromatin mainly during interphase (Figure S2). This pattern is identical to its interaction partner, Heterochromatin Protein 1 (HP1) [24]. Thus, we focused on interphase nuclei, and unless otherwise specified all images were taken at embryonic nuclear cycles 12–14, when heterochromatin is first observed. Consistent with previous results, LHR-HA colocalized with HP1 at DAPI-rich heterochromatic foci on the apical surface of the nuclei (Figure 2A). Unlike HP1, however, which is found throughout the nuclear compartment including euchromatin, LHR is restricted to heterochromatin. Consistent with being localized to a sub-domain of HP1, LHR strongly overlapped with Histone-3 lysine 9 dimethylation (H3K9me2), a histone modification specific to pericentric heterochromatin [25], but not with Cid, a histone variant specific to the centromere.
LHR was also observed in the embryonic germline precursors, the pole cells, and in the somatic and germline cells of the ovary (Figure S3A), where it again colocalized with H3K9me2 (Figure S3B). However, LHR was excluded from the nucleolus, a sub-compartment within heterochromatin consisting of rDNA repeats (Figure S3B). This observation suggested that LHR has a specific distribution within heterochromatin. We therefore used immuno-FISH to investigate the localization pattern of LHR relative to various pericentric satellites in D. melanogaster. We observed no overlap between LHR and the 359 bp satellite, a 4–5 Mb block on the X-chromosome [26], [27], nor between LHR and the highly abundant AATAT satellite, which is distributed across multiple chromosomes [28] (Figure 2C). In contrast, LHR consistently overlapped with dodeca, a G/C-rich pericentric satellite on the third chromosome [29], although a substantial amount of LHR is also found in other heterochromatic regions that we have not mapped. During metaphase, however, four discrete foci of LHR were visible along the metaphase plate. Noticeably, each LHR focus corresponded to the pericentric region of the third chromosome, as identified by overlapping dodeca signal (Figure 2D).
We next tested whether LHR localization is conserved in D. simulans. We constructed transgenic lines of D. simulans using the sim-Lhr-HA construct described above. Like mel-LHR in D. melanogaster, sim-LHR in D. simulans also localized to apical heterochromatic foci, as marked by DAPI (Figure 3C). We were particularly interested to determine whether sim-LHR associated with the dodeca satellite, because the distribution of dodeca varies among melanogaster subgroup species [30]. In particular, dodeca satellite is present only in the pericentric region of the third chromosome in D. melanogaster, but is present in the pericentric heterochromatin of both the second and the third chromosomes in D. simulans [30]. We confirmed this difference and found that the dominant dodeca signal is on the D. simulans second chromosome in mitotic brain squashes (Figure 3A). We also noted significant differences in the interphase organization of dodeca between species. We quantified the number of dodeca foci per nucleus and the fraction of nuclear space occupied in interphase nuclei from wild type brains. The dodeca signal in D. simulans appeared fragmented into more foci and occupied a greater nuclear volume, indicating that dodeca-containing heterochromatin has evolved species-specific nuclear organization properties (Figure 3B).
Despite this divergence in both chromosomal location and structure of dodeca, immuno-FISH mapping in D. simulans showed that sim-LHR partially colocalized with dodeca in interphase nuclei (Figure 3C). As with mel-LHR, a substantial amount of sim-LHR localizes to other regions of heterochromatin which we have not mapped. However, our results show that its association with dodeca is conserved between species.
We were unable to detect sim-LHR on chromosomes during metaphase (data not shown). We note, however, that only a small fraction of mel-LHR appears to be on metaphase chromosomes in D. melanogaster (see Figure S2) and we have found that challenging to image. We are thus unable to determine whether the apparent absence of sim-LHR from metaphase chromosomes reflects a true difference between species or instead is due to technical limitations.
It is unclear how LHR localizes to specific domains within heterochromatin, but it might require associations with other heterochromatin proteins, some of which are also rapidly evolving [21]. If LHR is co-evolving with other rapidly evolving proteins, then its heterochromatic localization might be altered when expressed in a foreign species.
To test this possibility we examined the localization of sim-LHR-HA in D. melanogaster. We found that sim-LHR-HA localized to the H3K9me2-enriched heterochromatic regions (Figure 3D), and colocalized with the dodeca satellite in a pattern identical to that seen for mel-LHR above (see Figure 2C). In order to directly compare the localization of LHR orthologs within the same nucleus, we generated a recombinant transgenic line that expressed both YFP-tagged mel-LHR and HA-tagged sim-LHR. The two LHR orthologs showed complete overlap, demonstrating that the heterochromatic localization properties of LHR orthologs are conserved (Figure 3D).
To determine whether heterochromatin states are perturbed in hybrids we examined HP1 and H3K9me2 localization. Although hybrid embryos were not sexed in this experiment, the staining appeared uniformly wild type in all embryos (Figure 4A). In order to specifically compare LHR and/or dodeca localization in hybrid males versus females, we developed a FISH probe that hybridized to the D. simulans Y-chromosome (Figure S4). We found that mel-LHR staining was enriched within apical heterochromatin in both sexes, and that it overlapped partially with dodeca (Figure 4C). Importantly, we detected no difference in dodeca organization and LHR localization between lethal hybrid males and viable hybrid females. Since heterochromatin defects might become more apparent later in development we then looked at heterochromatin states in hybrid larval neuroblasts. Consistent with the embryo staining, we saw no defects in the organization of either dodeca or the 2L3L satellite in either inviable male or viable female larvae (Figure 4D). Furthermore, despite differences in the pericentric heterochromatic sequences between homologous chromosomes, somatic pairing during interphase appeared unaffected in hybrid nuclei.
In spite of the adaptive protein sequence divergence between D. melanogaster and D. simulans orthologs of Lhr, our results surprisingly suggest only a limited degree of functional divergence of Lhr, with both orthologs having significant hybrid lethal activity and similar patterns of protein localization within heterochromatin. We therefore asked if gene regulatory divergence of Lhr between D. melanogaster and D. simulans might instead be responsible for the asymmetry of the lethal effects of Lhr in hybrids. We first surveyed Lhr transcript levels using qRT-PCR in three strains from each of the two species, and found no significant difference between the two species (Figure 5A). Consistent with this, we detected similar levels of LHR protein between the species (Figure 5B). Expression levels of mel-Lhr-HA and sim-Lhr-HA transgenes were each at a wild type level in their own species background, as total Lhr transcript level was approximately double in strains homozygous for the transgenes compared to wild type controls (Figure 5C). However, sim-Lhr was significantly overexpressed in D. melanogaster. The different expression levels of the sim-Lhr-HA and mel-Lhr-HA transgenes in the same D. melanogaster background indicate that cis-regulatory divergence has occurred at Lhr (Figure 5C). Furthermore, the fact that wild type levels of Lhr are not significantly different between the species (Figure 5A) despite these cis-regulatory differences suggests that trans acting factors that regulate Lhr have diverged. Taken together these data demonstrate that Lhr has undergone cis-by-trans compensatory regulation, such that cis-regulatory regions and trans-factors have co-evolved within each species to maintain a constant level of gene expression [31]. The uncoupling of such species-specific compensatory changes in a foreign genetic background would explain why sim-Lhr is hyper-expressed in D. melanogaster.
Given these results, we hypothesized that such a mechanism might cause asymmetric expression of Lhr orthologs in hybrids and by extension underlie the asymmetric rescue properties of Lhr orthologs. To test this hypothesis, we did allele-specific pyrosequencing to estimate the relative expression levels of the two Lhr orthologs in hybrids (Figure 6). We examined 3–5 day-old larvae because temperature shift experiments have shown that the L2/L3 stage is the critical phase of the lethality [17]. As expected Lhr transcript from the pure species parents was essentially 100% for their respective species-specific SNP. However, there was a significant overrepresentation of the D. simulans-specific SNP in both hybrid males and females, with ∼65% of Lhr transcripts deriving from the D. simulans ortholog in hybrid males and ∼60% in hybrid females. These data confirm our expectation that cis-by-trans divergence of Lhr regulation causes asymmetric expression in hybrids, and strongly suggests that a D. simulans mutation rescues hybrid sons because it removes a greater fraction of the total pool of Lhr, compared to a mutation in the D. melanogaster ortholog. We emphasize that this regulatory evolution leads to asymmetric expression of Lhr in hybrids but does not appear to cause an increase in total levels. The abundance of transgenic mel-LHR protein is not elevated in hybrids compared to pure species, as determined by Western blots (Figure 5D). Moreover, because protein levels of LHR orthologs appear equivalent in hybrids, we infer that levels of D. simulans LHR are also not visibly elevated in hybrids (Figure 5D). We therefore conclude that hybrid male lethality is not caused by Lhr over-expression. As we discuss below, lethality instead appears to result from hybrids becoming sensitive to Lhr activity due to its interaction with additional genes including Hmr.
Lhr and Hmr are D-M interaction partners that cause hybrid lethality [16]. Population genetic analyses of Lhr, Hmr and other HI genes found their coding sequences to be evolving rapidly under positive selection [3], [4]. These results imply that selection-driven protein divergence is the molecular basis of incompatibility in hybrids. An experimental prediction then is that independently evolving orthologs of a D-M gene should be non-equivalent with respect to the HI phenotype. Our initial genetic data supported this expectation for Lhr, because a loss of function mutation in D. simulans Lhr rescues lethal hybrid sons, while a loss-of-function mutation in D. melanogaster Lhr does not [16]. These findings led to several hypotheses: 1) HI is due to divergence specific to the D. simulans lineage; 2) this divergence has caused significant changes in the heterochromatin association properties of LHR proteins from D. melanogaster and D. simulans; and 3) defects in heterochromatin states directly cause hybrid lethality.
Contrary to some of these expectations, a subsequent study found that both D. melanogaster and D. simulans Lhr could cause HI when overexpressed in hybrids and that both proteins localized to heterochromatin when ectopically expressed in salivary gland cells [21]. In order to further explore functional differences between sim-Lhr and mel-Lhr we developed a native-promoter-driven transgenic system and performed higher resolution mapping of LHR protein localization. We found that both Lhr orthologs suppress hybrid rescue by D. simulans Lhr1, supporting the inference that hybrid lethal activity is a shared ancestral function. However, using a more sensitive interaction assay with Hmr, we detected that the lethal interaction was greater with sim-Lhr (Table 2). This finding is consistent with the pattern of genetic asymmetry where a mutation in D. simulans Lhr rescues hybrid lethality, while a deficiency removing D. melanogaster Lhr does not [16]. Our further investigation here reveals that removing mel-Lhr does in fact provide a modest level of hybrid rescue (Table 3). The fact that this rescue only occurs to the pharate adult stage in a minority of male hybrids underscores our conclusion that mel-Lhr has weaker hybrid lethal activity than sim-Lhr. A major focus of this study then became to understand the cause of this difference.
We attempted to create transgenic constructs of Lhr that were functionally identical to the wild type locus. To achieve this we generated Lhr transgenes that were driven by their native cis-regulatory sequences (Figure 1). Although the boundary of the regulatory regions included in the constructs was arbitrary we did quantitative RT-PCR assays on the transgenes to confirm that they expressed at wild type levels in both D. melanogaster and D. simulans (Figure 5C). Additionally, we infer from western blots that the abundance of transgenic LHR protein is similar in hybrids and pure species (Figure 5D), suggesting comparable expression levels in both backgrounds.
Nevertheless, we found that our mel-Lhr-HA transgene has greater activity than wild type Lhr when directly tested against an Lhr− deletion (Table S3). We consider two explanations: One possibility is that the construct has aberrant expression in a limited number of tissues or developmental stages that is beyond the resolution of detection in qRT-PCR assays of whole embryos or animals. Two, genetic assays for Lhr rescue are highly sensitive to genetic background effects; for example a large screen for suppression of Lhr rescue found a wide range of rescue even in the control balancer-chromosome classes [32]. We also observed here variable effects of D. melanogaster Lhr− deletions on hybrid viability (Table 3). Thus it is possible that this anomalous result results from an interaction with the multi-locus deficiency used and/or its genetic background.
While the result in Table S3 remains unexplained, we emphasize that the major conclusions of this study are not affected. The inference that mel-Lhr has hybrid lethal activity is independently shown by the rescue activity of the mel-Lhr deletion (Table 3). That result also demonstrates the asymmetric lethal activity of mel-Lhr and sim-Lhr, as does pyrosequencing of cDNA from hybrids (Figure 6). Likewise, the inference from transgenic assays that Lhr has undergone cis-by-trans compensatory evolution (Figure 5C) is fully consistent with the quantification of Lhr transcription by qRT-PCR in pure species (Figure 5A) coupled with the pyrosequencing result in hybrids.
Our first hypothesis to explain the differential effects of mel-Lhr versus sim-Lhr on hybrid viability was that their respective proteins might have different localization patterns. Previous studies found the LHR localizes to heterochromatin in D. melanogaster, but did not determine whether it is a general heterochromatin factor or instead has a specific localization within heterochromatin [16], [20], [21]. The heterochromatic landscape is dramatically different in closely related species [33], which raises the question of whether rapid evolution of Lhr orthologs reflects functional divergence necessitated by its association with fast-evolving heterochromatic sequences.
We addressed this question by (1) mapping LHR localization within D. melanogaster pericentric heterochromatin, (2) comparing its localization in D. simulans, and (3) examining sim-LHR localization in a D. melanogaster background. Within both species LHR localized to heterochromatic foci but was not ubiquitous (Figure 2A). For example, mel-LHR does not overlap with the AATAT or the 359 bp satellites, two major components of D. melanogaster pericentric heterochromatin [28]. In contrast, a portion of LHR consistently colocalized with the dodeca satellite in both species during interphase. The conservation of this colocalization pattern was particularly striking, given that dodeca repeats are found only on chromosome III in D. melanogaster but on both chromosomes II and III in D. simulans (see Figure 3A and reference [30]). Thus, the chromosomal distribution of LHR between the two species is different.
However, despite this divergence in the genomic location of dodeca, sim-LHR when expressed in D. melanogaster colocalized perfectly with mel-LHR (Figure 3D), demonstrating full conservation of LHR's heterochromatic localization properties. For three reasons, it is highly unlikely that this conserved pattern is because LHR orthologs share a DNA-binding activity specific to the dodeca sequence. First, LHR contains no recognizable DNA-binding domain. Second, LHR localization to heterochromatin is dependent on HP1 binding [20], [21]. Finally, LHR signal is neither restricted to dodeca nor perfectly overlapping with it (Figure 2C and Figure 3C). Thus, it is unclear what features of DNA or chromatin are configuring this localization pattern of LHR.
Neither the structure of the dodeca satellite nor LHR localization differed between pure species and hybrids, nor between lethal male and viable female hybrids (Figure 4C and 4D). These results set Lhr apart from two other well-characterized heterochromatin-associated HI genes. OdsH is a fast-evolving homeodomain protein that mislocalizes to the heterochromatic Y-chromosome in hybrids [5]. Zhr is a species-specific satellite DNA that causes hybrid lethality by improperly segregating during mitosis [6]. Such defects have been interpreted as support for the hypothesis that internal conflict with selfish heterochromatic elements is driving HI [3], [4], [7], [8]. We cannot rule out the possibility that there are defects in heterochromatin undetectable by our cytological analyses, or that Lhr may have other functions related to telomeric [16] or euchromatic [20] localization that have been affected by genetic conflicts. Nevertheless, the observations that heterochromatin appears normal in hybrids and that LHR localizes normally in both hybrids and when expressed in foreign species are not consistent with straightforward expectations of genetic conflict theories involving satellite DNAs [3]. Further work will be required to understand how Lhr causes lethal hybrids to have defects in cell proliferation and abnormally few larval cells entering mitosis [34], [35].
Having found that LHR orthologs have not diverged in their heterochromatin localization, we tested whether the asymmetric effects of mutations in mel-Lhr versus sim-Lhr on hybrid lethality reflect a history of regulatory sequence divergence rather than protein sequence divergence. In particular, we hypothesized that asymmetric expression of Lhr orthologs in hybrids could explain the aforementioned genetic asymmetry. We tested this hypothesis by measuring allele-specific expression of Lhr orthologs in hybrid larvae. Our results strongly support this hypothesis: we found that approximately 66% of the total Lhr transcripts in lethal hybrid male larvae originates from the D. simulans allele (Figure 6). Thus a mutation in D. simulans Lhr creates hybrid sons with only 1/3rd the wild type level of Lhr transcript, while hybrid sons with a mutation in the D. melanogaster ortholog have twice that amount. We conclude that only a loss-of-function mutation in D. simulans Lhr produces viable hybrids because it removes a greater proportion of the total Lhr gene product.
The divergence leading to asymmetric expression does not, however, reflect species-specific divergence in expression levels, because Lhr expression is not significantly different between D. melanogaster and D. simulans (Figure 5A). Instead asymmetric Lhr expression in hybrids is likely caused by the uncoupling of species-specific compensatory changes between cis-regulatory sequences and trans-factors. Interestingly, studies comparing the evolution of transcriptional networks between species have found that this type of regulatory divergence is frequently associated with gene mis-expression in interspecific hybrids [14], [31]. Furthermore, Takahasi et al. recently found evidence that stabilization of expression levels within a species involves widespread cis- and trans-compensatory mutations that can be detected as incompatibilities between heterospecific regulatory elements in interspecific hybrids [36]. The authors also suggest that signatures of adaptive evolution might result from the rapid accumulation of compensatory changes, and thus reflect the maintenance of an existing function rather than the evolution of a novel one. To our knowledge Lhr is the first example of cis-by-trans compensatory evolution occurring at an adaptively evolving hybrid incompatibility gene. An intriguing possibility is that the rapid evolution of the protein coding region reflects compensatory changes required to maintain an existing regulatory function of Lhr, rather than to alter its protein function.
We emphasize that cis-by-trans regulatory divergence explains the asymmetric effect of Lhr mutations on hybrid viability, but is not the direct cause of Lhr having hybrid lethal activity. Instead our data argue that the hybrid male genotype has evolved an acute sensitivity to Lhr dosage. Our genetic assays further suggest that the activity of Lhr that causes hybrid lethality was likely present in the ancestral state because it is shared by both mel-Lhr and sim-Lhr. This hypothesis is further supported by the observation that GAL4-UAS driven expression of Lhr from D. yakuba, an outgroup species, also kills hybrid sons [21]. Unlike Lhr, however, transgenic assays with its D-M partner, Hmr, showed that only the D. melanogaster ortholog but not the D. simulans ortholog is capable of causing hybrid lethality [37]. That result is consistent with the HI effect of Hmr being derived during evolution in the D. melanogaster lineage.
HI involving ancestral gene function is compatible with the D-M model, and was first considered by Muller [9]. One model he proposed involves incompatibility between an ancestral and a derived allele, with loss of a suppressor allele being required to ‘release’ the incompatibility. Here, this would require a suppressor to evolve first and become fixed in the D. melanogaster lineage, before the incompatibility-causing substitutions evolved in Hmr (Figure 7A). In the hybrid background, the suppressor is diluted or inactivated, exposing the lethal interaction. Alternatively, incompatibility could result from a complex epistatic interaction involving three or more loci. In the simplest case, changes at a single D. simulans locus, Sen*, cause the hybrid background to become sensitive to the dosage of Lhr in the presence of Hmr from the D. melanogaster lineage (Figure 7B). We favor the latter model because in the first model over-expression of sim-Lhr in D. melanogaster might be expected to at least partially overcome the suppressor and create the incompatible interaction. However, GAL4-UAS over-expression of sim-Lhr has no effect in a D. melanogaster pure species background [16], [21].
Although we diagram only a single sensitizing locus, a polygenic model involving multiple genes is equally possible, because available data only establish that Hmr and Lhr are insufficient to cause hybrid lethality [16]. If many additional genes are involved, then the distinction between ancestral and derived alleles may become blurred. For example, interacting genes may co-evolve, and have high evolutionary rates that maintain interactions rather than alter molecular functions.
Other examples of ancestral-derived incompatibilities have been discovered, such as the inter-allelic incompatibility at the S5 locus in rice, and the bi-locus incompatibility between the derived S. cerevisiae splicing factor MRS1 and the ancestral COX1 mRNA [11], [12]. However, unlike the incompatible S5 alleles which differ by only two amino acid substitutions, and COX1 which retains the ancestral intron that causes HI, Lhr orthologs have diverged rapidly under selection [16]. It is therefore remarkable that despite extensive protein sequence divergence between the hybridizing species, hybrid lethality has evolved as sensitivity to the dosage of an ancestral function. The key mechanistic implication is that instead of searching for a process or function that differentiates Lhr orthologs as the source of hybrid lethality, we now know that the sensitivity to Lhr in hybrids is based on a function and/or interaction that is common to both orthologs.
There are least 6 HI genes known that are rapidly diverging under selection [3]. With the exception of OdsH and Prdm9, where the signature of selection is restricted to a single functional domain [38], [39], in the other HI genes peaks of nonsynonymous substitutions do not coincide with a specific functional domain within the protein coding sequence. In these cases, it has been assumed that changes derived under selection have led to functional divergence, in turn causing incompatibility. However, it remains to be tested if that is truly the case.
We have assayed the hybrid lethal activity of both Lhr orthologs and found that despite extensive selection-driven divergence of the protein sequence, hybrid lethal activity is a shared ancestral function. We do not rule out the possibility that protein divergence makes some minor difference in hybrid lethal activity. However, our results suggest that the asymmetric effect of Lhr in causing hybrid lethality is explained by regulatory divergence. This finding demonstrates the need to consider regulatory divergence when interpreting interspecies experiments. Our results also highlight the complexity of the interspecific background and emphasize that hybrids are far from being the stoichiometric sum of two parental genomes. We suggest that while positive selection of protein-coding sequences remains a characteristic of HI genes, the phenotypic target of selection and its connection to HI are in some cases much less direct than expected.
All crosses were done at room temperature, or at 18°C where explicitly stated. At least 2 replicates were done for each cross. Each interspecific cross was initiated with ∼15–20 1-day-old D. melanogaster virgin females and ∼30–40 3–4-day-old sibling-species males. The nomenclature used for the transgenic lines and a complete description of the constructs used to generate them are included in Table S1. Genetic markers, deficiencies, and balancer chromosomes are described on FlyBase [40]. We previously showed that the D. melanogaster stock y1,w67c23; P{w+mC = lacW}l(2)k01209[k08901a]/CyO, used here in Table 3 and Table S3, is deleted for Lhr (see Fig. S4 in ref. [16]).
To make a modified pCasper4 containing the attB site, we PCR amplified a 280 bp fragment using the pTA plasmid (gift from Michele Calos) as the template [22]. This PCR product, along with flanking SalI sites was cloned into the compatible XhoI site of pCasper4 to create the plasmid pCasper4\attB. In order to construct Lhr transgenes with Lhr under the control of its native regulatory sequences, we used a 4.8 kb genomic fragment that spans 2.7 kb upstream and 1 kb downstream of the Lhr CDS. This fragment includes the complete CDS of the adjacent gene Bap55 (Figure 1).
To generate the p{sim-Lhr} construct we amplified this fragment from D. simulans w501 genomic DNA, using primer pairs 691/664 (see Table S2 for primer sequences). This PCR product was gel purified and cloned into the pCR-BluntII TOPO vector (Invitrogen), according to manufacturer's directions. The insert was sequenced completely and subcloned into pCasper4\attB using NotI and KpnI restriction enzymes. Note that this transgene contains more upstream DNA than the sim-Lhr transgene used by Prigent et. al. [41], which was also functional.
The p{mel-Lhr} construct was generated similarly, a 4.8 kb fragment was PCR amplified from wild type D. melanogaster (strain Canton-S) genomic DNA using primer pairs 597/598, and TOPO cloned into pCR-BluntII vector. The forward primer contains a NotI site, allowing the insert to be released as a NotI fragment and cloned into the NotI site of pCasper4\attB. A clone was chosen with the same orientation as in p{sim-Lhr}.
To construct p{sim-Lhr-HA} a triple-HA tag was added in-frame to the C-terminus of the Lhr CDS using a two-piece fusion PCR strategy. The two overlapping PCR products were amplified using p{sim-Lhr} as the template, with primer pairs 691/728 and 729/664. These fragments were used as templates for the fusion PCR, and the gel-purified product was TOPO cloned into the pCRBluntII vector and sequenced completely. The insert was then subcloned into pCasper4\attB exactly as in p{sim-Lhr}. The construction of p{mel-Lhr-HA} followed the same logic, using the primer pairs 597/728 and 729/598. To synthesize the p{mel-Lhr-YFP} construct a three-piece fusion PCR strategy was used, the first and last PCR products, containing upstream and downstream genomic regions respectively, were amplified using p{mel-Lhr} as the template, with primer pairs 597/730 and 733/598. The central PCR product containing the YFP-tag was amplified from p{w+mC UAS-Lhr::Venus = UAS-Lhr::YFP} [16], with primer pair 731/732. The 3 overlapping PCR products were used as templates for the fusion PCR, and cloned into the pCR-BluntII vector and sequenced completely. The insert was subcloned into pCasper4\attB exactly as in p{mel-Lhr}.
The p{ΔBap55 mel-Lhr-HA} construct is identical to p{mel-Lhr-HA} except that the Bap55 CDS is interrupted by the insertion of “TAA TGA C”, i.e. two stop codons and a frame shift mutation after the second methionine at position 6. Two overlapping PCR products were amplified using p{mel-Lhr-HA} as template, with primer pairs 597/1171 and 1172/598. The products were stitched together using fusion PCR and cloned into pCasper4\attB exactly as done in p{mel-Lhr}.
φC31-mediated transformation of D. melanogaster was performed by Genetic Services Inc. The integration sites used were: i) P{CaryP}attP2 and ii) M{3xP3-RFP.attP}ZH-86Fb at cytological positions 68A4 and 86Fb, respectively [22], [42]. P{CaryP}attP2 carries the body color marker yellow+ (y+). Site specificity of integration was tested using the PCR assays of ref. [43]. We also developed attP docking-site specific PCR assays, primer pairs1086/1087 for attP2, and 949/1177 for ZH-86Fb. All D. melanogaster transformants were crossed into the strain w1118. P-element mediated integration was used to transform the D. simulans w501 strain with P{sim-Lhr-HA}.
Total RNA was isolated using the Trizol Reagent (Invitrogen), followed by DNaseI (Roche) treatment and purification using RNeasy columns (Qiagen). First strand cDNA was synthesized from 4 µg of total RNA using the SuperScriptIII first-strand synthesis system (Invitrogen) with the oligo(dT)20 primer in a 20 µl reaction according to the manufacturer's instructions. Quantitative real time PCR (qRT-PCR) was performed on a Biorad MyiQ cycler with SYBR detection using the 2× supermix from Biorad. Relative concentrations of Lhr transcripts were calculated against rpl32 as the reference gene with rpl32 primers from reference [44]. The rpl32 gene sequence is 99% identical between the species. For Lhr primer pair 1147/1148 was developed to recognize conserved sequences and to amplify both D. melanogaster and D. simulans Lhr with equal and high efficiency. For each sample real-time PCR on test and reference genes was done in technical triplicates, and the standard curve method was used to estimate transcript abundance. For each genotype RNA was isolated from between 3 and 4 independent 6–10 hr-old embryo collections. For all genotypes except D. simulans P{sim-Lhr-HA} cDNA was synthesized twice from each RNA isolate.
RNA was extracted from 3–5 day-old larvae collected from non-crowded vials. In hybrid crosses the D. melanogaster mothers carried the X-linked mutation y− allowing the sex of larvae to be determined by using mouth hook coloration (daughters are y+ and sons y−). Total RNA and genomic DNA were simultaneously extracted from the same biological samples using the SV RNA system (Promega). For the pure species control, RNA and genomic DNA were extracted once from a single biological collection, followed by a single round of cDNA synthesis. For the hybrid samples, RNA and genomic DNA were extracted from four independent biological samples. cDNA was synthesized twice from each independent RNA isolate. Pyrosequencing measurements were performed in triplicate on each cDNA and in duplicate on each genomic DNA.
Whole cell extracts were obtained by grinding samples in ∼3 volumes of lysis buffer (50 mM Tris-HCl pH 7.5, 10 mM EDTA, 1.25% TritonX-100, 1× Roche protease inhibitor tablet). Extracts were cleared by centrifugation at 14,000 rpm for 10 min at 4°C. Total protein concentration of the cleared extracts was measured using Bradford assay (Biorad) and the samples were boiled in 0.5× volume of 4× SDS-Sample buffer. For most westerns 40 µg of total protein was loaded in each lane. Primary antibodies used were: rat anti-HA 3F10 (Roche; 1∶1000) and mouse anti-tubulin T5168 (Sigma; 1∶10,000). HRP conjugated goat anti-rat and goat anti-mouse secondary antibodies (Jackson; 1∶5,000) were used and detected with ECL Western blotting substrate (Pierce).
Embryo FISH and immuno-FISH were performed as in reference [6] and immunostaining of ovarioles was performed as in reference [45] with the following antibodies: Rat anti-HA 3F10 (Roche; 1∶100), mouse anti-HP1 C1A9 (DSHB; 1∶100), rabbit anti-histone H3 lysine 9 dimethylation (Upstate 07-441; 1∶100), rabbit anti-Cid (a gift from S. Henikoff; 1∶1000), rabbit anti-GFP (Abcam ab6556; 1∶300), mouse anti-Fibrillarin (Cytoskeleton Inc. AFb01; 1∶400) and mouse anti-Hts 1B1 (DSHB; 1∶4). FISH probes are described in reference [6]. DNA was stained using TOPRO-3 iodide (Molecular Probes) or Vectashield containing DAPI (Vector Laboratories). All imaging was conducted at the Cornell University Core Life Sciences Microscopy and Imaging Facility, using either a Leica DM IRB confocal microscope or an Olympus BX50 epifluorescent microscope, except for embryo images with a DAPI channel which were taken in the Plant Cell Imaging Center at the Boyce Thompson Institute, with a Leica TCS SP5 confocal microscope. Images were processed using Photoshop (Adobe, version 7.0). Contrast and brightness changes, when used, were applied globally across images.
Quantification of dodeca signal in interphase larval brain tissue was done using ImageJ [46]. Watershed segmentation was applied on the DAPI-channel to generate a mask of nuclear territories. The Analyze Particle function was then used to identify individual nuclei as ROIs (regions of interest) and screened to exclude aberrant nuclear segmentations and non-nuclear entities. Each ROI was individually selected on the dodeca FISH channel of the same image and the FociPicker3D plug-in was used to identify regions of local maxima. We then calculated two measures to estimate the nuclear dispersion of dodeca satellite: (1) the total number of foci per nucleus and (2) the fraction of total nuclear area occupied by the dodeca signal.
|
10.1371/journal.ppat.1000474 | MARCO, TLR2, and CD14 Are Required for Macrophage Cytokine Responses to Mycobacterial Trehalose Dimycolate and Mycobacterium tuberculosis | Virtually all of the elements of Mycobacterium tuberculosis (Mtb) pathogenesis, including pro-inflammatory cytokine production, granuloma formation, cachexia, and mortality, can be induced by its predominant cell wall glycolipid, trehalose 6,6′-dimycolate (TDM/cord factor). TDM mediates these potent inflammatory responses via interactions with macrophages both in vitro and in vivo in a myeloid differentiation factor 88 (MyD88)-dependent manner via phosphorylation of the mitogen activated protein kinases (MAPKs), implying involvement of toll-like receptors (TLRs). However, specific TLRs or binding receptors for TDM have yet to be identified. Herein, we demonstrate that the macrophage receptor with collagenous structure (MARCO), a class A scavenger receptor, is utilized preferentially to “tether” TDM to the macrophage and to activate the TLR2 signaling pathway. TDM-induced signaling, as measured by a nuclear factor-kappa B (NF-κB)-luciferase reporter assay, required MARCO in addition to TLR2 and CD14. MARCO was used preferentially over the highly homologous scavenger receptor class A (SRA), which required TLR2 and TLR4, as well as their respective accessory molecules, in order for a slight increase in NF-κB signaling to occur. Consistent with these observations, macrophages from MARCO−/− or MARCO−/−SRA−/− mice are defective in activation of extracellular signal-related kinase 1/2 (ERK1/2) and subsequent pro-inflammatory cytokine production in response to TDM. These results show that MARCO-expressing macrophages secrete pro-inflammatory cytokines in response to TDM by cooperation between MARCO and TLR2/CD14, whereas other macrophage subtypes (e.g. bone marrow–derived) may rely somewhat less effectively on SRA, TLR2/CD14, and TLR4/MD2. Macrophages from MARCO−/− mice also produce markedly lower levels of pro-inflammatory cytokines in response to infection with virulent Mtb. These observations identify the scavenger receptors as essential binding receptors for TDM, explain the differential response to TDM of various macrophage populations, which differ in their expression of the scavenger receptors, and identify MARCO as a novel component required for TLR signaling.
| The causative agent of tuberculosis, Mycobacterium tuberculosis, has a lipid-rich cell wall that contains a high percentage of mycolic acids. These mycolic acids contribute to both the impermeable nature of the cell wall and to the immunostimulatory properties of the bacterium. Indeed, it has been known for over 50 years that trehalose 6,6′-dimycolate (TDM/cord factor) is the major immunogenic lipid of M. tuberculosis, which induces potent pro-inflammatory responses from macrophages, although the receptor has not been identified. We have demonstrated that the toll-like receptor (TLR) pathway is required for pro-inflammatory cytokine production in response to TDM; however, the TLRs alone, or in conjunction with known co-receptors, are not sufficient to induce a response. We demonstrate that the macrophage receptor MARCO, a scavenger receptor, is utilized preferentially to “tether” TDM to the macrophage and activate the TLR2 signaling pathway, and is used preferentially over the related SRA. Macrophages from MARCO−/− mice are defective in activation of TDM-induced signaling and subsequent pro-inflammatory cytokine production in response to both TDM-coated beads and virulent M. tuberculosis. By identifying the macrophage receptors involved in initial recognition we can now explain variable responses to TDM between different macrophage populations (which differ in scavenger receptor expression), and have identified a novel co-receptor that may be involved in lipid presentation to TLRs.
| Mycobacterium tuberculosis (Mtb), a causative agent of human tuberculosis, is responsible for 8 million new infections and 2 million deaths yearly. One third of the world population is currently estimated to be infected with M. tuberculosis, although less than 10% of those infected show clinical signs of infection [1]. This is mainly due to the robust granulomatous response that is initiated by the bacterium, which effectively contains the infection and allows the host to exist in equilibrium with a subclinical infection. The granulomatous response has been shown to be triggered by multiple components of the mycobacterial cell wall, such as phosphatidylinositol dimannoside, phosphatidylinositol hexamannoside, and trehalose 6,6′-dimycolate (TDM) [2]–[5].
The chemical structure of TDM (also known as cord factor) was solved in 1956 [6] and was identified as the predominant immunogenic mycobacterial cell wall glycolipid [7]. TDM can elicit pro-inflammatory cytokine production in vitro, and granulomatous responses in vivo, when administered as a monolayer or part of an oil-water emulsion [7]–[10] . Although it has been known for decades that this pro-inflammatory response is mediated primarily by macrophages, binding or signaling receptors for TDM on macrophages have yet to be identified. Biochemical approaches used in our laboratory to identify TDM receptors, such as affinity isolation using TDM columns, TDM bead phagosome isolation from labeled macrophages, and mass spectrometric analysis of phagocytosed TDM-coated beads after photoactivatable cross-linking, have been unsuccessful. These observations, as well as the finding that the stimulatory activity of TDM requires presentation over a larger surface area, such as emulsions, monolayers, or large diameter particles, suggests that the TDM-receptor interaction is of low avidity and requires the aid of co-receptors or other accessory molecules [9],[10].
One class of receptors implicated in TDM recognition is the TLRs. TLRs have not been shown to bind to TDM directly, but bone marrow-derived macrophages (BMMΦ) from MyD88−/− mice do not produce pro-inflammatory cytokines in response to TDM-coated polystyrene microspheres [9]. Although TLRs have been shown to sense and signal from within the phagosome, they are not phagocytic receptors and usually require the presence of a co-receptor (e.g. CD14) to present their ligands [11].
Class A scavenger receptors (SR), SRA and MARCO, are a class of phagocytic receptors that we have demonstrated mediate recognition and presentation of TDM. SRs bind a range of ligands of endogenous and exogenous origin with relatively low affinity. Ligands for the SRs include proteins [12] and lipids. These lipids can be derived from either the host (e.g. oxidized lipids) [13], or from exogenous sources (e.g. lipopolysaccharide) [14]. Of the two class A SRs, SRA has been clearly demonstrated to be involved in host defense by suppressing excessive pro-inflammatory cytokine production in mouse models of infection and septic shock [14],[15]. MARCO has also been implicated in host defense against bacterial pathogens, but it is not clear whether it is a positive or negative regulator of pro-inflammatory cytokine production [16],[17].
It has been proposed that the class A SRs may be involved in host defense against mycobacterial infection. SRA expression is increased after interferon-gamma (IFN-γ) treatment or exposure to M. tuberculosis, and is highly expressed on macrophages associated with M. bovis Bacille Calmette-Guérin (BCG)-induced granulomas [14],[18]. There are conflicting reports as to whether expression of SRA increases uptake of M. tuberculosis or BCG; however, its presence does not appear to affect the rate of replication of BCG, despite being protective against BCG-primed endotoxic shock [14],[18]. In mouse models, MARCO expression has been shown to be transiently up-regulated on macrophages in response to BCG infection and to be expressed on macrophages within, and adjacent to, BCG-containing granulomas [19]. MARCO-expressing macrophages in the splenic marginal zone appear to phagocytose more BCG than neighboring macrophages that do not express MARCO [19]. The mycobacterial ligands that mediate this recognition have not yet been identified.
Herein, we identify that TDM recognition and signaling is mediated, at least in part, by MARCO, TLR2, and CD14. Although SRA and MARCO have many common ligands, our results show that MARCO binds more TDM-coated beads than either isoform of SRA. MARCO is required for TDM-induced signaling via TLR2 and CD14 in a transfection system, whereas SRAI and SRAII require co-transfection of TLRs 2 and 4, and their accessory molecules, to permit even a minor response to TDM stimulation. Consistent with these data, both resident peritoneal macrophages (RPMφ) and BMMφ from TLR2/4 double-deficient mice (but not the individual mutants) have a markedly reduced response to TDM. This suggests that TDM engages TLR2 and TLR4 in a redundant fashion and that these predominantly MyD88-dependent pathways are required for the stimulatory effects of TDM [9]. When stimulated with TDM-coated microspheres, macrophages from MARCO−/− and MARCO−/− SRA−/− double-deficient (DKO) mice also show reduced activation of ERK1/2 compared to wildtype mice and are defective in subsequent pro-inflammatory cytokine production. These macrophages also produce fewer pro-inflammatory cytokines in response to infection with M. tuberculosis, indicating that SR-mediated detection of TDM may be an important component of the response to infection. On the basis of these data we propose a model in which MARCO, and to a lesser extent SRA, cooperate with TLR2 and CD14 for TDM recognition and signaling.
BMMφ secrete proinflammatory cytokines in response to TDM coated onto microspheres that are too large to be phagocytosed [9], whereas RPMφ readily produce TNF-α in response to phagocytosable TDM-coated microspheres and secrete much higher amounts than BMMφ in response to larger microspheres (Fig. 1A). RPMφ express high levels of MARCO and SRA, whereas BMMφ express high levels of SRA, but do not express MARCO at levels detectable by immunoblot (Fig. 1B). RPMφ upregulate MARCO at the RNA and protein level in response to TDM-coated 90 µm microspheres, however we have not observed any significant increase in MARCO expression on BMMφ by immunoblot after as many as 72 hours of exposure to LPS or TDM-coated 90 µm microspheres (data not shown). While there are most likely many differences between RPMφ and BMMφ, the absence of MARCO expression on BMMφ may account for the differences in the response to TDM between these two cell types.
We have previously shown that TDM-induced cytokine production is MyD88-dependent but had not identified which TLRs are required (9). Herein we demonstrate that both TLRs 2 and 4 are partially required for TDM-induced cytokine production. BMMφ from C3H/HeJ mice (defective in TLR4 signaling) and TLR2−/− mice showed no reduction in TNF-α production (Fig. 2A&B), while macrophages from TLR2/4−/− mice had greatly reduced TNF-α responses to TDM (Fig. 2C). This suggests that TLR2 and 4 can function at least in a partially redundant manner with respect to TDM responses.
In order to determine whether TDM-coated beads were a ligand for MARCO, CHO-K1 cells, which do not express SRs, were transfected with plasmids encoding either MARCO or SRA, and non-opsonic bead binding was assessed. MARCO- and SRA-transfected cells bound significantly more beads than mock-transfected cells, and binding was inhibited by the SR inhibitor dextran sulfate (DxSO4), but not chondroitin sulfate (ChSO4), which does not inhibit the SRs (Fig. 3A). Phosphatidylglyercol (PG) was used as a negative control lipid because we have previously shown that it can be coated onto microspheres in a similar manner and induces a minimal cytokine response from macrophages. PG-coated beads did not bind cells transfected with either MARCO or SRA to a greater extent over empty vector-transfected cells. Furthermore, PG bead binding to cells could not be inhibited by SR inhibitors, indicating that binding was not MARCO- or SRA-specific (data not shown). SRA-transfected cells bound fewer TDM-coated beads despite having higher transfection efficiency (Fig. 3B) indicating that SRA may have a lower binding affinity for TDM-coated beads.
HEK293 cells do not express most TLRs, except for TLRs 1, 5, and 6 (C. Leifer, personal communication) and are therefore often used for reconstitution assays. By co-transfecting these cells with a luciferase reporter driven by an NF-κB promoter and various combinations of TLRs and SRs, we determined which combinations of receptors allow TDM-induced NF-κB signaling. HEK293 cells transfected with only TLR2 or TLR4, with accessory molecules CD14 or MD2, respectively, did not respond to TDM-coated 90 µm microspheres more than PG-coated microspheres or medium alone (Fig. 4A). The addition of human MARCO (hMARCO) to TLR2/CD14 or TLR2/CD14/TLR4/MD2, however, allowed statistically significant (p<0.05) stimulation of luciferase activity in response to TDM (Fig. 4A). Interestingly, transfection of human SRAI (Fig. 4B) or SRAII (data not shown) required both TLRs 2 and 4, with their respective accessory molecules, in order for TDM-induced stimulation to be observed. These results suggest that MARCO is required for TDM-induced signaling via TLR2/CD14 and that SRA is less efficient in mediating responses to TDM, requiring both TLRs 2 and 4, as well as their accessory molecules.
The scavenger receptors have been implicated in modulating TLR signaling by a variety of TLR agonists, but it is not clear if they alter these responses by direct involvement in the signaling pathway. In order to determine whether the cytoplasmic region of MARCO (which would, in theory, transduce signaling events) could contribute to TDM-induced NF-κB activation, two constructs of MARCO were made. The first (Myr MARCO) lacks amino acids 1–40 but contains a putative myristoylation site; the second (N-tail) lacks the entire cytoplasmic domain (1–49) (Fig. 5A). After confirming that the constructs were expressed on the surface of transfected cells (Fig. 5C), their ability to restore TDM-induced NF-κB activation was tested. Cells transfected with either the wild-type MARCO (hMARCO), or the two deletion mutant constructs (Myr MARCO and N-tail MARCO), in addition to TLR2 and CD14, retained their ability to induce NF-κB activation (Fig. 5B). These data suggest that the cytoplasmic region of MARCO is not required for binding or signaling responses to TDM.
Confocal immunofluorescence microscopy was performed to determine whether MARCO or SRA localizes to the membrane of phagosomes containing TDM-coated or control beads. At early time points (5–10 minutes), MARCO recruitment to TDM bead phagosomes was visible in both transfected CHO-K1 cells (Fig. 6C) and RPMφ (Fig. 6A). Recruitment was not clearly evident in MARCO-transfected CHO-K1 cells or in RPMφ that had phagocytosed bovine serum albumin (BSA)- or PG-coated microspheres (Fig. 6A and C, and data not shown). SRA staining of RPMφ did not show strong co-localization with TDM- or BSA-bead phagosomes at this early time point (Fig. 5B), and SRA-transfected CHO-K1 cells showed no appreciable accumulation of SRA at the phagocytic cup in response to either TDM or BSA-coated beads (Fig. 6B and D). At later time points (15–30 minutes), both MARCO and SRA staining occurred at the site of binding to the TDM-, PG-, and BSA-coated microspheres (data not shown), consistent with previous observations that the scavenger receptors are involved in the binding and uptake of polystyrene beads [20]. From these results, we conclude that MARCO is specifically recruited to the phagosomal membrane surrounding TDM-coated beads, whereas SRA recruitment is non-specific.
RPMφ from wildtype, MARCO−/−, SRA−/−, MARCO−/−SRA−/− (DKO), TLR2−/−/TLR4−/− and CD14−/− mice were stimulated with lipopolysaccharide (LPS), Pam3Csk4, microspheres coated with either TDM or PG, or medium only, and lysates were immunoblotted for phosphorylated ERK1/2. ERK1/2 was phosphorylated in response to TDM-coated microspheres in wild-type macrophages (Fig. 7). There was no detectable decrease in the TDM response by SRA−/− macrophages (data not shown); however, MARCO−/− macrophages had a reduced response to TDM, and ERK1/2 phosphorylation was further reduced in macrophages from DKO mice (Fig. 7). The levels of ERK1/2 phosphorylation in response to LPS and Pam3Csk4 were similar between wildtype and SR-deficient macrophages indicating that the SR knockouts do not have a global defect in their ability to phosphorylate ERK1/2 in response to TLR agonists but are specifically defective in the TDM response (Fig. 7). Equivalent levels of ERK1/2 activation by LPS in CD14-deficient RPMφ may be due to soluble CD14 provided by fetal calf serum in the medium and the high dose of LPS used [21].
Because the SRs are required for TDM-induced ERK1/2 activation (Fig. 6), we hypothesized that SR-deficient macrophages might also be defective in downstream pro-inflammatory cytokine production. RPMφ from wildtype, MARCO−/−, SRA−/− or DKO mice were stimulated with 3 µm diameter TDM-coated beads for 24 hours. Both the MARCO−/− and SRA−/− macrophages had a statistically significant reduction in TNF-α production (P<0.05). TNF-α production from DKO macrophages was completely abrogated (P<0.025) (Fig. 8C). This is not due to decreased levels of TLR2 (Fig. 8A) or TLR4 (Fig. 8B) expression on the SR-deficient macrophages, as these are both expressed at equivalent levels between wildtype and SR−/− RPMΦ.
The mouse macrophage cell line, RAW264.7, is similar to BMMφ in expressing high levels of SRA (data not shown) but lacking MARCO at both the protein (Fig. 9B) and RNA levels (data not shown), and in producing lower levels of cytokines in response to TDM as compared to RPMφ (Fig. 9A). We predicted that transfection of RAW264.7 with MARCO could elevate responsiveness to TDM. Stable cell lines expressing either human MARCO (hMARCO-RAW) or an empty vector (vector) were created and stimulated with TDM- or PG-coated microspheres for 24 hours, after which the levels of TNF-α in the medium were assessed by ELISA. Only the hMARCO-expressing cells produced TNF-α in response to the TDM-coated microspheres (P<0.05) (Fig. 9A). This result further supports the important role of MARCO in TDM-induced cytokine production.
TDM is an essential virulence factor for M. tuberculosis pathogenesis and thus we hypothesized that pro-inflammatory cytokine production resulting from M. tuberculosis infection would also be impaired in MARCO-deficient macrophages. RPMφ were infected with an MOI of 5 for 24 hours and cytokine production in the supernatants was assessed by ELISA. Consistent with our hypothesis, MARCO−/− and DKO macrophages produced significantly less TNF-α, IL-6, and IL-1β than wildtype macrophages (Fig. 10).
The pathogenesis and establishment of M. tuberculosis infection requires phagocytosis of the bacterium by macrophages and initiation of the pro-inflammatory response. These two events are at least partially independent. Phagocytosis is mediated by a number of receptors including the mannose receptor and DC-SIGN which recognize mannose-capped lipoarabinomannan (ManLAM) [22],[23], and complement receptor which mediates the phagocytosis of both opsonized and non-opsonized bacteria [24]. The initiation of a pro-inflammatory response appears to be mediated primarily via TLRs [25] and possibly other signaling receptors such as dectin-1 [26]. Of the mycobacterial cell wall lipids that initiate a TLR-mediated inflammatory response, TDM appears to be one of the more potent [9].
Although we have demonstrated that the macrophage response to TDM is partially TLR2/4-dependent (Fig. 2), our initial attempts to reconstitute NF-κB signaling in a TLR2/4 stably-transfected cell line were not successful. It seemed likely that this was due to the requirement of an additional co-receptor, because many TLR ligands, and especially lipid-based ligands, require presentation via a co-receptor. The co-receptor CD14 has been implicated in facilitating TLR1/2-mediated responses to bacterial lipopeptides by enhancing the physical proximity of the ligand to the TLR1/2 heterodimers, without binding directly to the receptor complex [25],[27],[28]. However, CD14 expression in conjunction with TLR2 or TLR4 did not restore responsiveness to TDM (Fig. 4) suggesting that CD14 was not the only co-receptor for TDM. Because MARCO and SRA bind to TDM-coated beads (Fig. 3A), we hypothesized that they might be the additional co-receptors required for TLR2 signaling. Both MARCO and CD14 in conjunction with a TLR2 homo- or heterodimer appear to be required to initiate TDM signaling, and MARCO appears to be preferred over the closely related SRA (Fig. 4A and B, Fig. 7). We therefore propose that the scavenger receptors function as co-receptors that, in conjunction with CD14, present TDM to TLR2. Further work is warranted to determine whether TDM signaling requires an additional receptor such as TLR1 or TLR6. The structure of TDM, however, does not include di-acylated lipids that signal via TLR2/6 heterodimers or tri-acylated lipids that signal via TLR1/2 heterodimers [29].
SRA has been demonstrated to modulate TLR signaling [30],[31] and macrophages from both SRA- and MARCO-deficient mice have skewed cytokine responses in response to TLR agonists [16]. It is not entirely clear, however, if the SRs are able to signal directly or if they have an indirect function in the signaling pathway, for example, by phagocytosing and clearing TLR agonists. In order to test whether MARCO might be signaling directly, we created mutants that lacked the cytoplasmic domain, and thus any putative signaling motifs, of MARCO. The cytoplasmic region of MARCO has not been experimentally demonstrated to have any residues that are essential for cell signaling and indeed, apart from the putative myristoylation site at residues 41–46, does not contain any potential signaling motifs identifiable by scanning various protein motif databases (D.M.E.Bowdish, unpublished results). After confirming that these constructs were expressed on the surface of transfected CHO-K1 and HEK293 cells, we determined that they could indeed bind to TDM-coated beads and were in fact able to reconstitute TDM-induced NF-κB signaling (Fig. 4). Our receptor-ligand interaction assay results, however, show that MARCO alone is not sufficient for TDM binding, and that other receptors present on the surface of RPMφ must cooperate with MARCO for effective binding to occur (Fig. S1, Text S1). Signaling induced by TDM may be independent of the phagocytic function of the scavenger receptors, as macrophages from the SRA−/−, MARCO−/− and DKO mice did not have any detectable defect in phagocytosis of TDM-coated beads (data not shown). These data are consistent with a role for MARCO as a “tethering” receptor for TDM, perhaps extracting individual lipids and presenting them to the CD14/TLR2 complex (Fig. 11), but MARCO itself appears to lack a direct signaling function.
Our observation that MARCO is the preferred receptor for TDM may explain why some macrophage populations respond robustly to TDM, while others do not. For example, RPMφ, which express high levels of MARCO, respond strongly to TDM, whereas BMMφ and RAW264.7 cells, which express SRA but not MARCO, produce a minimal amount of pro-inflammatory cytokines in response to TDM. We therefore propose that MARCO is the preferred receptor for TDM, although the less avid interaction between SRA and TDM can also facilitate signaling through TLRs. This hypothesis is consistent with the work of Ozeki et al. in which it was shown that SRA can bind to TDM in vitro and plays a role in suppressing TNF-α production by alveolar macrophages or Kupffer cells in response to TDM-coated wells [32].
Because TDM is highly immunogenic, it is being studied as an adjuvant that boosts both humoral and cellular immune responses [33],[34], is a novel candidate for vaccine development [35], and is used to mimic the pathogenesis of M. tuberculosis infection [36]. Indeed, our observation that MARCO-deficient macrophages are defective in pro-inflammatory cytokine production in response to either TDM or virulent M. tuberculosis is consistent with our hypothesis that TDM is the major immunogenic lipid associated with pro-inflammatory responses. It is likely that TDM is not the only ligand for the scavenger receptors on M. tuberculosis and further study is warranted to elucidate what these interactions might be; however, our data suggests that this interaction is a major component of the macrophage response to infection. We propose that TDM is a novel SR ligand that binds to MARCO with a higher affinity than SRA and that the TDM-induced pro-inflammatory response is mediated in large part via SR/TLR2/CD14 receptors. . These observations may explain why some macrophage populations respond more strongly than others to TDM and thus will provide novel insight into the role of the scavenger receptors in the pathogenesis of tuberculosis.
C57BL/6, SRA−/−, MARCO−/−, MARCO−/− SRA−/− (DKO), CD14−/−, C3H/HeN, C3H/HeJ and 129sv/ev mice were bred and housed at the University of Oxford, Cornell University, or University of Georgia. All SR knockout mice were created on the C57BL/6 background strain [15],[17],[37]. The SRA−/− and DKO mice are deficient in SRAI and SRAII. Unless otherwise stated, C57BL/6 mice were used as wild-type and were purchased from either Taconic or Harlan. TLR2/4 double-deficient mice were created by Shizuo Akira (Osaka University), generously supplied by Lynn Hajjar (University of Washington), and were bred at the Cornell University Transgenic Mouse Core Facility. All mice were housed in specific pathogen-free conditions and experiments were designed to use age- and sex-matched mice, between 5 weeks and 3 months of age. All animal experiments were approved by the ethics board of the university at which the experiments were performed (i.e. University of Oxford or Cornell University).
CHO-K1 cell line (ATCC#CCL-61) was maintained in Ham's F12K medium (Gibco) supplemented with 10% heat-inactivated fetal calf serum (HI-FCS, Hyclone), 2 mM L-glutamine, 1.5 g/L sodium bicarbonate, 100 U/ml penicillin, and 100 µg/ml streptomycin (Gibco). CHO-K1 cells were transfected using Lipofectamine as per the manufacturer's instructions (Invitrogen). The HEK293 cell line (ATCC# CRL-1573) was provided by Cynthia Leifer, cultured in Dulbecco's Modified Eagle Medium (DMEM; Gibco) supplemented with 10% HI-FCS, 2 mM L-glutamine, 1 mM sodium pyruvate, 10 mM HEPES (Gibco), 100 U/ml penicillin, and 100 µg/ml streptomycin. Transfected RAW 264.7 cells were maintained in RPMI, 10% HI-FCS, 2 mM L-glutamine, and 100 µg/ml ascorbic acid (Sigma). BMMφ were cultured as described previously [9] and maintained in DMEM supplemented with 20% L-929 cell-conditioned media, 10% HI-FCS, 2 mM L-glutamine, 1 mM sodium pyruvate, 100 U/ml penicillin, and 100 µg/ml streptomycin. RPMφ were obtained by lavaging the peritoneal cavity with 10 ml cold phosphate-buffered saline (PBS), re-suspending the peritoneal cells in complete growth medium (DMEM supplemented with 10% HI-FCS, 2 mM L-glutamine, 1 mM sodium pyruvate, 100 U/ml penicillin, and 100 µg/ml streptomycin) and allowing the macrophages to adhere to Petri dishes overnight. Non-adherent cells were then rinsed off before RPMφ were used for experiments. All media components were routinely tested for endotoxin by Limulus Amoebocyte Assay (Cambrex). All cells were maintained at 37°C and 5% CO2. .
The monoclonal anti-MARCO clones ED31 (anti-mouse MARCO) and PLK-1 (anti-human MARCO), and anti-SRA clone 2F8 were grown and maintained at the Gordon laboratory. Polyclonal anti-hMARCO antibodies were a generous gift from Dr. Timo Pikkarainen. Anti- mouse TLR2 and TLR4 antibodies were purchased from eBioscience. Rabbit antibodies for phosphorylated and total ERK1/2 (Cell Signaling Technologies) were used according to manufacturer's protocols. Peroxidase-conjugated goat anti-mouse or anti-rabbit antibodies (Jackson Labs) were used at 1∶200 for immunoblotting. AlexaFluor 488 or 594 goat anti-mouse IgG antibodies (Molecular Probes) were used as secondary antibodies for immunofluorescence.
TDM from M. tuberculosis H37Rv strain and bovine-derived PG (Sigma-Aldrich) were resuspended in chloroform/methanol (2∶1 v/v) at 10 mg/ml and stored at −20°C under nitrogen. Sterile 3 µm and 90 µm diameter polystyrene microspheres or 2.5 µm fluorescent polystyrene microspheres (Polysciences) were coated with TDM or PG as described previously [9]. Coated microspheres were washed and re-suspended in endotoxin-free PBS (Invitrogen) at 2% solids. It should be noted that the manufacturing protocol for these microspheres was changed by Polysciences in 2007, resulting in less efficient coating of the polystyrene particles by lipids and therefore reduced immunostimulatory ability. Original results could be reproduced using 80 µm diameter polystyrene microspheres from Duke Scientific.
Human TLR4, MD2, TLR2, and CD14 plasmids were generously provided by Dr. Cynthia Leifer (Cornell University). All plasmids were amplified and purified using Endo-free Maxi Prep columns (Qiagen). The plasmids containing human MARCO (hMARCO), mouse MARCO (mMARCO), human SRAI and SRAII have been previously described [38],[39]. Constructs of human MARCO that were missing the cytoplasmic tail (N-tail hMARCO or Myr hMARCO) were created by designing primers that amplified the transmembrane region of hMARCO and contained restriction enzyme sites. After amplification and restriction digest, the amplified fragment was sub-cloned into pcDNA3.1 (Invitrogen).
Stable cell lines expressing either hMARCO or the empty vector (pcDNA3) were created in RAW264.7 cells by transfecting the cells per the manufacturer's directions (GeneJuice, EMDbiosciences) and selecting under G418. Surviving cells were tested for MARCO expression by flow cytometry and immunoblot, and were cultured in 1 mg/ml G418 until use.
CHO-K1 cells were transfected with plasmids encoding either MARCO or SRA as described above. Medium was removed and replaced with Opti-MEM (Invitrogen). Non-opsonic bead binding was assessed by incubating the cells on ice for 30 min with or without inhibitors, then adding TDM or PG-coated, fluorescent, 3 µm diameter microspheres. After 30 min on ice, the cells were washed, fixed, and bead binding was assessed by microscopy. Dextran sulfate (DxSO4) or chondroitin sulfate (ChSO4) were used at concentrations of 100 µg/ml.
HEK293 cells were seeded at 5×105 cells per well in 2 ml of medium per well in a 6 well plate overnight. HEK293 cells were transfected according to manufacturer's protocol using TransIT transfection reagent (Mirus) with 144 ng each of NF-κB-luciferase and β-galactosidase reporter plasmids, 30 ng each of TLR2, CD14, or TLR4, 90 ng MD2, and 300 ng MARCO or SRAI/II per well depending on the experiment. Total DNA was brought to 2 µg using empty vector (pcDNA3.1). Transfected cells were incubated for 24 hours before trypsinization and reseeding in 96 well plates (one row of a 96 well plate from each well of the 6 well plate) in 100 µl media/well. After another 24 hours, cells were stimulated with either 1.25×103 TDM- or PG-coated 90 µm polystyrene microspheres per well, or positive control ligands 1 µg/ml Pam3Csk4 (Calbiochem) or 100 ng/ml lipopolysaccharide (Sigma) as positive controls for TLR2 and TLR4, respectively. Only data from experiments, in which positive responses to LPS and Pam3Csk4 were consistent between relevant transfectants, were used in order to assure functional TLR4 and TLR2 complexes were expressed at equivalent levels. In addition, only experiments in which equivalent levels of MARCO and SRA were expressed, as determined by FACS, were used. After 18 hours, transfected cells were lysed using Reporter Lysis Buffer (Promega) and lysates were analyzed for luciferase (Promega) and β-galactosidase (Tropix) activity using a Veritas luminometer (Turner Biosystems). NF-κB activity (relative light units) was measured by dividing luciferase activity by β-galactosidase activity, and then fold activity was calculated by dividing TDM and PG results by medium only results. Statistical significance was determined using Student's t test.
RPMφ, or CHO-K1 cells transfected with hMARCO for 24 hours as described above, were seeded onto cover slips at 1×105 cells per well in 500 µl of media. Washed, 3 µm diameter, carboxylated silica beads (Kisker Biotech) were re-suspended in 25 mg/ml of the heterobifunctional crosslinker cyanimide and incubated with agitation for 15 minutes. Excess cyanimide was removed by washing twice in 0.1 M sodium borate pH 8.0 (Sigma-Aldrich). The beads were then cross linked to defatted BSA in a 10 mg/ml solution for 2 hours with agitation, followed by labeling with 2 µg/ml carboxyfluoresceinthiosemicarbazide (Molecular Probes) for 30 minutes with agitation. After washing with PBS, the beads were then passively coated with TDM as described above. Beads were added to cells at a ratio of 5∶1, maintained at room temperature for five minutes, then incubated at 37°C for five minutes, before rinsing with PBS and fixing in 4% paraformaldehyde in PBS. Cells were blocked overnight at 4°C in staining buffer (SB; 1% BSA, 1% heat-inactivated goat serum, 0.25% saponin in PBS), then incubated with 10 µg/ml primary antibody in SB for 1 hour at room temperature, washed three times with PBS, and then incubated with secondary goat anti-mouse antibody also for 1 hour at room temperature. Cover slips were rinsed three times with PBS, then quickly in double-distilled water before mounting using Prolong Gold Antifade medium (Molecular Probes). Confocal images were taken using an Olympus Fluoview 500 confocal laser scanning imaging system equipped with argon, krypton, and He-Ne lasers on an Olympus IX70 inverted microscope with a PLAPO 60× objective (Olympus America, Inc.). Confocal images were processed using Adobe Photoshop 6.0 (Adobe Systems, Inc.,).
For FACS analysis, cells were transfected with plasmids as described above and stained with either the appropriate MARCO- or SRA-specific antibodies or corresponding isotype controls as per standard protocols.
To assess SR expression, BMMφ, RPMφ, and RAW264.7 cells were seeded in 6 well plates overnight in the appropriate media (discussed above). The cells were lysed in RIPA buffer [50 mM Tris-HCl (pH 7.4), NP-40 1%, sodium deoxycholate 0.25%, NaCl 150 mM, EDTA 1 mM, PMSF 1 mM, and protease inhibitors (Roche)], then assessed for protein concentration by Bradford Assay (Bio-Rad). Equivalent amounts of protein per sample were boiled in 2× non-reducing sample buffer for five minutes, centrifuged, and separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). Samples were transferred to nitrocellulose membranes, and then blocked in 5% nonfat dry milk in 0.1% Tween-20 in PBS (PBST) overnight with gentle rocking. Both primary and secondary antibodies were diluted in blocking buffer, and applied to membranes for one hour at room temperature with gentle rocking, and with three 5 minute washes with PBST between incubations.
For MAPK immunoblots, RPMφ were seeded at approximately 5×105 cells per well in a 24 well plate in R10 medium and allowed to adhere overnight. After 16 hours, the cells were washed once with PBS and the medium was replaced with 450 µl of warm R10. RPMφ were stimulated with either 100 ng/ml of LPS, 1 µg/ml Pam3Csk4, or 1.9×103 TDM- or PG-coated 90 µm diameter beads, or PBS as a vehicle control. After stimulation, the medium was removed and the cells were lysed using 200 µl of hot 2× reducing sample buffer. Lysates were placed on ice, syringed to shred DNA, then boiled for five minutes at 100°C, centrifuged and stored at −20°C until use. Equal volumes of sample were loaded per lane for SDS-PAGE. Proteins were transferred to nitrocellulose membranes and blocked with 5% nonfat dry milk in TBST [10 mM Tris–HCl (pH 8), 150 mM NaCl, 0.1% Tween-20]. The filters were then incubated overnight at 4°C with anti-ERK1/2-P antibodies (Cell Signaling Technology). Anti-total ERK1/2 antibody was used to show equivalent lane loading. Immunoreactive bands were detected using horseradish peroxidase-conjugated goat anti-rabbit IgG antibodies (Jackson Immunoresearch) and chemiluminescence (Pierce).
BMMφ or RPMφ were counted and seeded at approximately 1×105 cells/well in 100 µl media/well in 96 well plates, or 4×105 cells/well in 500 µl media/well in 24 well plates. RAW264.7 constructs were counted and seeded at approximately 1×104 cells/well in 96 well plates. RPMφ were washed once with PBS to remove non-adherent cells and the media replaced with Opti-MEM. Macrophages were stimulated with either TDM or PG-coated microspheres, with dose normalization for bead surface area (total bead surface area of 5×107 µm2/well), 100 ng/ml of LPS (E. coli, Sigma-Aldrich), or 100 ng/ml Pam3Csk4 (Calbiochem); PBS was added as a vehicle control. After 24 h, the media were collected and the concentration of TNF-α was determined by ELISA as per the manufacturer's directions (BD Bioscience or eBioscience). In experiments comparing RPMφ from different mouse genotypes, cells were stained with crystal violet to normalize for cell numbers. The crystal violet was solubilized with 1% SDS and the resulting supernatant was read on a plate reader (OD 550 nm). TNF-α release is expressed as pg/ml/OD550 to normalize for cell numbers.
RPMφ were collected from MARCO−/−, SRA−/−, and DKO mice as described above and seeded in equivalent numbers in 24 well plates, depending on the genotype with the lowest yield (approximately 1×105 cells per well). Mid-log phase cultures of Mtb H37Rv strain were washed then added to the cells at an MOI of 5 in 250 µl of medium. After 24 hours the medium was filtered and analyzed for cytokines by ELISA.
Differences between the means of experimental groups were analyzed using Student's t test. Differences were considered significant when P≤0.05.
|
10.1371/journal.pbio.1001429 | Inhibition of the Prokaryotic Pentameric Ligand-Gated Ion Channel ELIC by Divalent Cations | The modulation of pentameric ligand-gated ion channels (pLGICs) by divalent cations is believed to play an important role in their regulation in a physiological context. Ions such as calcium or zinc influence the activity of pLGIC neurotransmitter receptors by binding to their extracellular domain and either potentiate or inhibit channel activation. Here we have investigated by electrophysiology and X-ray crystallography the effect of divalent ions on ELIC, a close prokaryotic pLGIC homologue of known structure. We found that divalent cations inhibit the activation of ELIC by the agonist cysteamine, reducing both its potency and, at higher concentrations, its maximum response. Crystal structures of the channel in complex with barium reveal the presence of several distinct binding sites. By mutagenesis we confirmed that the site responsible for divalent inhibition is located at the outer rim of the extracellular domain, at the interface between adjacent subunits but at some distance from the agonist binding region. Here, divalent cations interact with the protein via carboxylate side-chains, and the site is similar in structure to calcium binding sites described in other proteins. There is evidence that other pLGICs may be regulated by divalent ions binding to a similar region, even though the interacting residues are not conserved within the family. Our study provides structural and functional insight into the allosteric regulation of ELIC and is of potential relevance for the entire family.
| Pentameric ligand-gated ion channels (pLGICs) are ionotropic neurotransmitter receptors that mediate electrical signaling at chemical synapses. The pLGIC family includes receptors for acetylcholine, serotonin, GABA and glycine, which share a similar structural organization and activation mechanism: the channels are closed in the absence of ligands and open when neurotransmitters bind to a conserved site in the extracellular domain. In many family members, activation by the neurotransmitter can be affected by modulators (including several drugs in therapeutic use), which bind to different sites on the channel. Channel function can be modulated also by divalent cations, which either potentiate or inhibit pLGICs at physiological concentrations. Here, we analyze this mechanism in the pLGIC ELIC, a prokaryotic family member of known structure. We show that divalent cations such as calcium or zinc inhibit ELIC by occupying an extracellular site remote from the ligand-binding region thereby interfering with gating. Although the site of interaction is not conserved between different family members, we present evidence that regulation of other pLGICs involves the same region. Our study has thus provided insights into a regulatory process that appears to be general for the pLGIC family in both eukaryotes and prokaryotes.
| The pentameric ligand-gated ion channels (pLGICs) are ionotropic neurotransmitter receptors, which are activated by the binding of ligands to specific sites of the protein. The family includes both cation-selective channels, such as nicotinic Acetylcholine- (nAChRs) and Serotonin receptors (5HT3Rs), and anion-selective channels, such as GABA- (GABARs) and Glycine receptors (GlyRs) [1]. Despite these differences in ion selectivity, the overall molecular architecture and the mechanism by which ligands open the ion conduction path are conserved [2]–[8]. pLGIC subunits form either homo- or hetero-pentamers that consist of at least two functional units, an extracellular ligand-binding region and a transmembrane pore [9],[10]. Agonists open the channel by binding to a conserved site in the extracellular domain, at the interface between two subunits [11],[12]. A homomeric receptor contains five equivalent agonist binding sites, several of which need to be occupied for maximum channel activation and this makes the process highly cooperative [5],[13]–[16]. Agonist binding is accompanied by conformational rearrangements that are transmitted over a distance of tens of angstroms from the extracellular domain, via the domain interface to the pore [17]. These receptors have thus become important model systems for the study of allosteric mechanisms [18]. Many pLGICs are important drug targets and all aspects of their function can be influenced by pharmacological agents. These are a diverse set of molecules that include agonists and competitive antagonists (which act on the agonist binding site itself), pore blockers that inhibit ion conduction, and allosteric modulators that interact with regions distinct from the agonist-binding site. Modulators such as benzodiazepines [19], general anesthetics [20], alcohol [21], and the antiparasite ivermectin [22] can either enhance or inhibit pLGIC activation. pLGIC function is affected also by divalent cations (such as calcium and zinc) in two distinct ways. Cation-selective pLGICs are somewhat permeable to divalents, but the strong interaction between these ions and the pore decreases or blocks conduction in a voltage-dependent manner [23],[24]. In addition to that, divalent cations can also modulate channel gating. For instance, calcium potentiates the agonist responses of nAChRs [25]–[27] and inhibits those of 5HT3Rs [28],[29], and zinc can either potentiate or inhibit channel activation, depending on the type of pLGIC and the ion concentration [30]–[35].
Here we show that both the modulatory and the channel block effects of divalent cations are present also in ELIC, a prokaryotic pLGIC channel whose structure was determined in a nonconducting conformation [36]. Agonists of ELIC include primary amines such as cysteamine, propylamine, and the vertebrate neurotransmitter GABA. In ELIC, these agonists occupy the canonical ligand-binding site of the family and open a cation-selective pore with permeation properties similar to those of eukaryotic channels [37]. Here we describe how divalent cations permeate and block the ELIC pore, and how they also inhibit ELIC gating, by binding in the extracellular domain, to a site remote from the ligand-binding region.
We have investigated the effects of divalent cations on ELIC by electrophysiology and X-ray crystallography. Divalent cations can influence ELIC function in several different ways depending on concentration (Figure 1). The traces in Figure 1A show that low mM concentrations of the alkaline earth metal ion Ca2+ decrease the single channel conductance of ELIC when added to the extracellular medium at negative holding potentials. ELIC single channel currents are progressively reduced by increasing Ca2+ concentrations and decrease by approximately 25% of their control amplitude at 5 mM Ca2+ (Figure 1C) and by a maximum of about 50% at high Ca2+ concentration [37]. This effect is due to tight interactions of divalent ions with the channel pore and has been thoroughly characterized for different pLGIC family members [23],[24] including the homologous channel GLIC [38], whose structure was determined by X-ray crystallography in a conducting conformation [39],[40].
Low extracellular calcium (greater than 100 µM) produces also a voltage-independent decrease in agonist potency. This effect is detectable at Ca2+ concentrations too low to decrease channel conductance and is manifested as a parallel rightward shift in the agonist dose–response curve (Figure 1D, Table S1). A similar effect on agonist binding in the presence of calcium is observed in isothermal titration calorimetry experiments (Figure 1E). Up to 1 mM calcium, the shift in the agonist dose–response curves is truly parallel, as the maximum agonist current does not decrease more than the single channel conductance does (Figure 1B and 1C). This pattern appears to reproduce the effects of competitive antagonists, which bind to the ligand-binding site and reduce its occupancy by the agonist in a surmountable way (e.g., their effect can be overcome by increasing agonist concentration). This resemblance is obvious if the effects of Ca2+ are compared with those of the competitive antagonist acetylcholine, which is known to bind to the agonist-binding site of ELIC (Figure 1F) [41]. The Schild plot for acetylcholine [42],[43] is linear with a slope of unity and a binding affinity of 1.6 mM (Figure 1G, Table 1). The Schild plot for Ca2+ is also linear, with a potency of 260 µM, but a shallower slope of 0.8 (Figure 1G, Table 1).
The similarity between the effect of calcium and that of a competitive antagonist disappears as Ca2+ concentrations are increased above 1 mM. The current traces in Figure 1B show that the reduction in agonist potency is now associated with a decrease in the maximum agonist response. This decrease is too big to be explained by the effect of Ca2+ on conductance: at 5 mM Ca2+ the single channel conductance is reduced by 25% and the maximum agonist response by 55% (Figure 1C). At progressively higher concentrations of the divalent cation, the maximum current response continues to decline and this decrease can be described by a fit to a Langmuir equation with an IC50 of 6 mM (Figure 1H). Despite the strong reduction in the maximum currents, the shift in EC50 remains linear over a wide concentration range (Figure S1). The pronounced drop in maximum current strongly suggests that at higher concentrations calcium impairs the opening of the channel and reduces agonist efficacy.
Next, we tried to establish whether calcium impairs the maximum rate of ELIC gating (e.g., when the channel is fully bound to the agonist) by measuring the on-relaxation of currents elicited by rapid propylamine applications to outside-out patches from HEK293 cells. Figure 1I shows that increasing Ca2+ from 50 to 200 µM does slow the onset of the current elicited by a saturating agonist concentration (20 mM propylamine, red trace) but that this effect is overcome by increasing agonist concentration to 50 mM (green trace). Only minor changes in the time course of deactivation were detected (Table 2). Thus the maximum rate with which the agonist-bound channel opens is unchanged, which is unexpected given the observed change in agonist efficacy. This could be because we could test only low calcium (in high calcium the concentrations of agonist required to saturate channel gating are too high to be experimentally feasible). Alternatively, calcium impairs gating by affecting a step in the channel activation that controls the size of the maximum agonist response, but not the speed of overall gating (see Discussion).
Finally, we found that divalent cations other than Ca2+ also affect ELIC responses. In particular, other alkaline earth metal ions, such as Mg2+, Sr2+, and Ba2+, are slightly weaker than Ca2+ in inhibiting ELIC (Figure 2A–C and E, Table 1, Table S1), whereas the transition metal ion Zn2+ is considerably more potent (i.e., Schild plot x-intercept 8 µM, Figure 2D and E, Table 1, Table S1).
In order to understand the structural basis of the effects of divalent ions we aimed at identifying the region of interaction by X-ray crystallography. Since the crystal form that was used for the structure determination of ELIC contains high concentrations of sulfate, which forms insoluble salts in the presence of most alkaline earth metal ions, we had to identify novel crystallization conditions compatible with divalent ions. In a broad screen we observed crystals growing in Ba2+-acetate. Ba2+ can be readily located in the electron density by its strong anomalous scattering properties, and since it has comparable effects on channel function as Ca2+ (Figure 2A, Figure S2A), it is reasonable to assume that it will occupy the same sites in the protein. Crystals of the ELIC/Ba2+ complex belong to two different, yet related crystal forms, one similar to the original barium-free form of ELIC that was used for structure determination (space group P21) and another growing in a higher symmetry space group (P43) (Table 3). Datasets for both crystal forms were collected to 3.8 Å (P21) and 3.3 Å (P43) resolution and provide equivalent views of the channel and its interaction with divalent cations.
The structures show a conformation of the channel that is overall very similar to the structure of ELIC already described. Strong peaks in the anomalous difference density allow us to detect the presence of Ba2+ ions bound to three distinct sites of the protein (Figure 3).
Firstly, a single Ba2+ ion per channel is located on the 5-fold axis of symmetry at the extracellular end of the pore and is coordinated by the side-chains of Asn251 (position 20′ of the second transmembrane domain in the numbering system developed for the nAChR, Figure 3A–C). Throughout the article we will refer to this site as Spore.
There are two additional sets of binding sites for Ba2+ in the structure shown in Figure 3B. Both are found at the interface between subunits in the extracellular domain in five symmetry-related locations. One set of sites faces the channel vestibule and will be referred to as Sin. The barium ion in Sin is coordinated by Ser84 of the principal subunit and Asp86 of the complementary subunit (Figure 3D). Barium ions are bound also to a set of five equivalent sites on the outer rim of the extracellular domain (Figure 3B and E). These sites, which we will call Sout, are about 15 Å below the ligand-binding pocket, towards the membrane plane and are formed by the side-chains of acidic amino acids contributed by both subunits. These residues include Asp113 at the end of β6 on the principal side and Glu150 and Asp158 on the loop connecting β8 and β9 on the complementary side (Figure 3A and E). The refined 2Fo-Fc electron density map of this region indicates a direct interaction of the respective carboxylate groups with the bound ions resembling Ca2+-binding sites observed in other proteins (Figure 3E, Figure S2B and C). Remarkably, in none of the collected datasets did we find any evidence for Ba2+ in the ligand-binding pocket itself.
The structure of ELIC in complex with Ba2+ has revealed the location of three distinct sites for the interaction with divalent cations. If binding to any of these sites is relevant for the inhibition of the channel, we would expect that mutating the interacting residues should affect the functional modulation by divalent ions. Thus we mutated the residues that contact Ba2+ in the structure and measured again the effects of Ca2+ by two-electrode voltage-clamp electrophysiology (Figures 4 and 5, Table 1).
Given that the effects of low Ca2+ concentrations resemble those of competitive antagonists, we tested also whether the agonist binding site can play a role (even though we have no structural evidence that divalents bind there). Our functional data show that the agonist binding site is unlikely to be involved, because Ca2+ inhibition is not changed by a mutation here (R91A) that increases agonist potency by 3–4-fold ([37], Figure 4A and F).
We then proceeded to investigate whether the inhibitory effects of Ca2+ are produced via binding to the Spore site by truncating the side-chain of the Asn residue in contact with the divalent ion. Our X-ray crystallography data show that the structure of this N251A mutant is on the whole similar to WT but lacks the anomalous difference density in Spore. The structure of this mutant still shows strong density of ions bound to Sout (and weaker density for Sin), thus suggesting that effects of the mutation are local (Figure 4B). Electrophysiological recording shows that agonists activate WT and mutant N251A channels with similar potency and that the inhibition by Ca2+ of these responses is only modestly decreased in N251A (Figure 4C and Schild plots in 4F). This suggests that Spore is not the major site responsible for the Ca2+ inhibition.
Figure 4F shows also that mutating the binding residues in another set of divalent ion sites, Sin (which face the extracellular vestibule), has little effect on Ca2+ inhibition. Mutation S84A (on the principal side) changes neither the potency of the agonist nor the inhibition by Ca2+ (Figure 4D). Similarly, in the mutant D86A there is only a modest decrease in agonist potency, and the inhibitory effect of Ca2+ is virtually unchanged (Figure 4E and F). Thus we have shown that neither Spore nor Sin mediate the functional effects of calcium on channel activation.
In contrast to that, we found that Ca2+ modulation is greatly decreased when we change any of the residues that coordinate divalent cations in Sout. This is seen both when the residues with acidic side chains (Asp 113, Glu150, and Asp158) are individually replaced with their uncharged isosteric counterparts (Asn or Gln) and when the acidic side-chains are truncated to Ala (Figure 5, Figure S3). All of these mutations cause a variable but strong decrease in the potency of Ca2+, which suggests that they weaken the interaction with the ion and thus its inhibitory effects (Figure 5E and F). The strongest effect among single mutants is observed for residues Asp113 and Asp158 (Figure 5A, B, and E). Combining these two mutations in the double mutant D113A/D158A virtually abolishes the effects of both calcium and barium on the agonist dose–response curves (Figure 5D, Figure S3E and S3F). Remarkably, and in contrast to our observations in WT, in this double mutant the decrease in Imax at high Ca2+ concentration appears entirely due to the reduction in single channel conductance (Figure 5G). The binding of Ca2+ to Sout is thus responsible for both functional effects on the shift of the EC50 and the decrease of Imax. Figure 5 also shows that mutations in Sout shift the EC50 towards higher agonist concentrations, an effect that is not surprising given that this region is thought to be important in transducing agonist binding into channel activation (Figure 5A–D, Figure S3, Table S1).
The X-ray structure of the double mutant D113A/D158A in complex with Ba2+ is on the whole unperturbed. The double mutation has removed the density of ions bound to Sout, while leaving the strong anomalous difference density in Spore unchanged. This confirms that in this mutant divalents fail to modulate channel activation because they cannot bind to the Sout site (Figure 5H).
Given that the same mutations abolish also the modulation by Zn2+ (Figure 5I), it is very likely that Zn2+ inhibits ELIC by binding to the same site. This finding is somewhat unexpected as Zn2+ usually interacts with histidine or cysteine residues. However, since the ligand binding domain of ELIC does not contain any cysteines and since mutations of the two histidines, which are both located on β10, did not affect the inhibition by Zn2+ (Figure S4), it is likely that the interaction of this transition metal ion with ELIC occurs at this site and therefore deviates from common binding modes.
The results of our mutational analysis strongly suggest that the observed inhibition of ELIC by divalent cations is mediated by the specific interaction with a site that is located at the outer rim of the extracellular domain, at the interface between neighboring subunits. Since this site is distant from the agonist-binding region, we wanted to explore whether there is any direct competition between the effect of divalent ions and that of competitive antagonists binding to the ligand-binding site. Such competitive antagonists include quaternary ammonium compounds such as tetramethylammonium, a weak antagonist (Figure S5), or acetylcholine, which inhibits the channel with higher affinity. The X-ray structure of ELIC in complex with the heavy atom analogue tetramethylarsonium (Figure 6A) and the recently determined structure of ELIC in complex with acetylcholine [41] show that both antagonists bind to the ligand-binding pocket and prevent the binding of the agonist to the same site. The overlap of agonist- and antagonist-binding sites is also reflected in the 10-fold increase in the Schild affinity of acetylcholine in the mutant R91A. This is similar to the increase in agonist potency in the same mutant (Figure 6B and D). In contrast to the mutation in the binding site, the Sout double mutant D113A/D158A abolishes the modulatory effect of Ca2+ but does not alter the affinity of acetylcholine (WT 1.6 mM, D113A/D158A 2.1 mM), confirming that calcium and acetylcholine act via distinct sites (Figure 6C and F, Table 1).
Finally, in order to probe whether the presence of one antagonist would alter the effect of the other, we have studied the inhibition of ELIC by acetylcholine in the presence of different concentrations of Ca2+ and vice versa. In no case did we find any significant change in the potency of either antagonist, which suggests that the inhibitory effects are additive and the two compounds thus act independently (Figure 6E–H).
In the present study we have investigated how divalent cations modulate the function of ELIC, a bacterial member of the pLGIC family. ELIC is inhibited by alkaline earth metal ions and by the transition metal ion zinc. The modulation reported here resembles similar effects observed in other family members, where divalent cations act as either positive or negative modulators of gating. Ca2+ potentiates channel activity in a subset of nAChRs [26],[27], whereas it has an inhibitory effect on 5HT3Rs [29]. Like in ELIC, in 5HT3Rs calcium shifts the EC50 of activation towards higher ligand concentrations [29]. The action of Zn2+ appears to be more complex. In some subtypes of GABARs, Zn2+ inhibits channel activity [31], whereas in GlyRs, nAChRs, and 5HT3Rs, it acts as a potentiator at low concentrations and as an inhibitor at higher concentrations [30],[32],[33]. These opposing effects are believed to be mediated by the successive occupation of binding sites of different affinity.
X-ray structures of ELIC crystals grown in the presence of barium have allowed us to identify five structurally equivalent binding sites (Sout) located at subunit interfaces on the extracellular domain about 15 Å from the agonist-binding region. These are likely to be responsible for the observed inhibition, as mutations at this site have a strong effect on the potency of both Ca2+ and Zn2+. The sites resemble regulatory calcium-binding pockets found in other ion channel proteins, where the divalent ions interact with the side chains of acidic residues that are often organized in clusters on the protein sequence (Figure 7A) [44]–[46]. The interaction found in ELIC is, however, not typical for zinc-binding sites, as these usually contain either histidines or cysteines for ion coordination [47]–[49], residues that are unlikely to play this role in ELIC (Figure 7A, Figure S4).
While the residues that interact with divalent cations in ELIC are not conserved across pLGICs, there is evidence that equivalent modulatory effects in other pLGICs involve the same (Sout) region. In the α7-nAChR, the residue Glu 172, which has been identified as a key residue in the interaction with calcium [50]–[52], resides on the same loop as Glu 150 and Asp 158 (loop 8) in ELIC. Similarly, histidine and glutamate residues contributing to the interaction with Zn2+ in GABAARs were mapped to the same location, at the interface between two subunits [53], thus indicating that the Zn2+-dependent inhibition of GABAARs may follow a similar mechanism. Residues in the same loop of 5HT3Rs have also been proposed to participate to calcium regulation of this receptor [54].
Interestingly, a study on the 5HT3R has identified an aspartate residue in the pore domain as an important determinant for calcium-dependent inhibition. The equivalent Asn residue in ELIC coordinates the barium ion in the site Spore [55]. We investigated this site by mutagenesis but did not find any indication for a similar role in the calcium regulation of ELIC. The phenotypic difference may be due to a stronger interaction with a divalent ion in the 5HT3R where the respective residue is an aspartate and thus carries a negative charge (cf., an uncharged asparagine in ELIC).
The effect of calcium and other divalent cations on gating of ELIC results in a complex functional phenotype. At low extracellular calcium concentrations, we see a reduction in agonist potency that resembles competitive inhibition (with a linear Schild plot with a slope near unity). Despite this resemblance, the agonist binding site is not involved in this process and the presence of the antagonist acetylcholine (which binds in the canonical agonist site) has no effect on the action of calcium. Finally, higher calcium concentrations reduce the maximum agonist response (to a greater extent than can be accounted for by a conductance decrease). At first sight, these effects appear to be too complex to be explained by a single microscopic action of divalents (i.e., the binding of Ca2+ to the site Sout). However, they can all be accounted for, if calcium impairs a single step of ELIC activation, for example channel opening, provided gating is efficient in wild-type ELIC (i.e., the agonist efficacy E is high to start with, Figure 7B). This is a plausible hypothesis, given the high open probability of the single channel activity in Figure 1A.
In first approximation, the relation between maximum open probability Pmax and efficacy E is:and our observations of an ELIC Pmax greater than 95% are compatible with values of E that are greater than 20 (as reported for other pLGIC such as nicotinic and glycine receptors). If the value of E is high to start with, the reduction in efficacy produced by divalents must be substantial before a decrease in maximum response becomes apparent. That is why it is seen only at high calcium concentrations. More modest decreases in efficacy, at low calcium concentrations, will cause only a decrease in agonist potency. This is because agonist EC50 is directly affected by the value of E. In the simplest del Castillo-Katz model, EC50 is given by:where Kd is the microscopic dissociation constant of the agonist (Figure 7B) [56].
It can also be shown (Text S1) that the effects of calcium and those of a competitive antagonist are expected to be independent, if we model equilibrium channel activation with a simple scheme, where calcium binding impairs gating (by affecting E) and the antagonist binds to the resting form of the channel. This model not only predicts Schild-like behavior for the effect of calcium but suggests also that the Schild intercept is a reasonable estimate for the microscopic affinity of divalents (Text S1).
These conclusions are unchanged if we model channel activation by a more detailed and realistic activation scheme, incorporating an intermediate state between agonist binding and channel opening. The existence of one or more gating intermediate states for channels in the nicotinic superfamily is supported by several lines of evidence. For instance, φ analysis in muscle nicotinic AChRs [57] indicates that blocks of residues move asynchronously in the gating conformational change. In addition to that, mechanisms with reaction intermediates (referred to as flip, primed, or catch-and-hold [13],[58]–[61]) are needed to explain several aspects of the function of the GlyR and the muscle nicotinic AChR, such as agonist efficacy (Figure 7B). In our experiments, the presence of an additional intermediate step that limits the maximum rate of current onset in agonist-bound ELIC channels is required to explain the results of our agonist concentration jumps. This is because we observed that low calcium increased the agonist concentration needed to achieve the maximum rate of current onset, but did not change the limiting rate of channel gating. If activation went through a single conformational step as the channel gates (as in a simple del Castillo-Katz mechanism), this single step would control both the rate of current onset for the agonist-bound channel and the maximum response, and any changes in this would be experimentally detectable (see Text S1).
In our study we have shown how the binding of calcium to a single site remote from the ligand binding pocket modulates the activation of the pLGIC ELIC. Given that divalent ions impair ELIC gating, they are expected to bind more tightly to the resting state of the channel and stabilize it. The location of the divalent binding site at the interface between adjacent subunits is an intriguing mechanism to stabilize distinct states in an allosteric protein, given that these regions are involved in conformational changes (Figure 7C). Thus, occupancy by divalent ions of sites at a similar location in the different pLGICs will result in potentiation or inhibition, depending on whether the equilibrium is shifted towards conducting or nonconducting conformations. Allosteric modulation is important for the pharmacology of pLGICs, as many of pLGIC drugs in therapeutic use act by this mechanism, although by binding to sites distinct from those of divalent ions.
Modulation by divalent ions of pLGICs occurs at concentrations that are physiologically relevant in vertebrates and may regulate the activity of channels in their natural environment [31],[62]. It is not known whether such regulation is important for ELIC activity in its natural host Erwinia chrysanthemi, but it is remarkable that the observed mechanism has been conserved during evolution.
ELIC WT and point mutants were expressed and purified as described [36],[37]. E. coli BL21DE3 containing a vector encoding for a fusion protein consisting of the pelB signal sequence, a His10 tag, maltose binding protein, a HRV 3C protease site, and ELIC were grown in M9 minimal medium at 37°C to an OD of 1.0 and subsequently cooled to 20°C. Expression was induced by addition of 0.3 mM IPTG and carried out overnight. All the following steps were performed at 4°C. The protein was extracted from isolated membranes in a buffer containing 1% n-Undecyl-β-D-Maltoside (UDM, Anatrace, Inc.) and purified by Ni-NTA chromatography (Qiagen). The purified MBP-ELIC-fusion protein was digested with HRV 3C protease to cleave the His10-MBP protein. His10-MBP and 3C protease were subsequently removed from solution by binding to Ni-NTA resin. ELIC was concentrated and subjected to gel-filtration on a Superdex 200 column (GE Healthcare). The protein peak corresponding to the ELIC pentamer was pooled and concentrated to 10 mg/ml and used for crystallization.
The purified protein was crystallized in sitting drops at 4°C. Protein containing additional 0.5 mg/ml E. coli polar lipids (Avanti Polar Lipids, Inc.) was mixed in a 1∶1 ratio with reservoir solution (50 mM ADA pH 6.5, 50 mM BaAc2, and 10% (w/v) PEG4000). The crystals were cryoprotected by transfer into solutions containing 30% ethyleneglycol. All datasets were collected on frozen crystals on the X06SA beamline at the Swiss Light Source (SLS) of the Paul Scherrer Institut (PSI) on a PILATUS detector (Dectris). The data were indexed, integrated, and scaled with XDS [63] and further processed with CCP4 programs [64]. The structure of WT and mutants in space groups P43 and P21 were determined by molecular replacement in PHASER [65] using the ELIC pentamer (2VLO) as a search model. G164, which was not included in the original model (2VLO), was introduced according to the structure of the ELIC acetylcholine complex (3RQW). The absence of this amino acid had only a local effect and did not influence the location of neighboring residues. The model was rebuilt in Coot [66] and refined maintaining strong NCS constraints in PHENIX [67]. R and Rfree were monitored throughout. Rfree was calculated by selecting 5% of the reflection data in thin slices that were selected for the initial dataset of ELIC and that were omitted in refinement.
Binding of the agonist propylamine to ELIC in the presence and absence of calcium was measured by isothermal titration calorimetry (ITC) with a MicroCal ITC200 system (GE Healthcare). The syringe was loaded with agonist solution containing 30–37 mM propylamine dissolved in measurement buffer (containing 25 mM Tris-HCl pH8.5, 150 mM NaCl, and in certain experiments 0.6 mM CaCl2). The sample cell was loaded with 300 µl of purified ELIC in measurement buffer containing 0.9 mM UDM at a concentration between 80 and 110 µM. Agonist was applied by sequential injections of 2 µl aliquots followed by a 180 s equilibration period after each injection. The data were recorded at 4°C and analyzed by a fit to a single-site binding isotherm.
Constructs containing the gene of either the WT or mutant channels preceded by the signal sequence of the chicken α7nAchR were cloned into the pTLN vector for expression in X. laevis oocytes [68]. After linearization of the plasmid DNA by MluI, capped complementary RNA was transcribed with the mMessage mMachine kit (Ambion) and purified with the RNeasy kit (Qiagen). For expression, 1–50 ng of RNA was injected into defolliculated oocytes. Two-electrode voltage clamp measurements were performed 1 d after injection at 20°C (OC-725B, Warner Instrument Corp.). Currents were recorded in bath solutions containing 10 mM HEPES (pH 7), 130 mM NaCl, and the indicated concentrations of cysteamine and divalent cations. In case of solutions containing Zn2+, cysteamine was replaced by propylamine. The membrane potential in all dose–response measurements was set to −40 mV. As ELIC is permeable to divalent cations, we tested if endogenous calcium-activated chloride channels affected our measurements. To chelate intracellular calcium ions, the oocytes were incubated for 15 to 30 min in bath solutions lacking divalent ions but containing 10 µM BAPTA-AM. Dose–response curves in the presence of calcium obtained from BAPTA-AM-treated oocytes did not differ from the measurements of the untreated oocytes even at elevated Ca2+ concentration (Figure S6). The lack of a significant effect is likely due to the strong outward-rectification of calcium-activated chloride channels, which do not pass significant currents at negative voltages.
X. laevis oocytes were transferred to a hyperosmotic solution to manually remove the vitelline layer. Membrane patches were recorded in the excised outside-out configuration 3–5 d after injection of mRNA with an Axopatch 200B amplifier (Axon Instruments) at 20°C. Data were sampled at 100 µs, filtered with 1,000 Hz, and analyzed using Clampfit (Axon Instruments, Inc.). Bath solutions contained 10 mM HEPES (pH 7.0), 150 mM NaCl, and indicated concentrations of ligands and divalent cations. Electrodes had a resistance of 3–5 MΩ. Pipette solutions contained 150 mM NaCl, 10 mM EGTA, 5 mM MgCl2, and 10 mM HEPES at pH 7.0. Bath electrodes were placed in 1 M KCl solution connected to the bath solution by Agar bridges. The agonists were applied to the patch using a stepper motor (SF77B Perfusion fast step, Warner).
Human embryonic kidney 293 cells (American Type Culture Collection-CRL-1573;LGC Promochem) were maintained at 37°C in a 95% air/5% CO2 incubator in DMEM supplemented with 0.11 g/l sodium pyruvate, 10% (v/v) heat-inactivated fetal bovine serum, 100 U/ml penicillin G, 100 µg/ml streptomycin sulfate, and 2 mM L-glutamine (Invitrogen). Cells (passaged every 2 d, up to 30 times) were plated and transfected by calcium phosphate-DNA coprecipitation [69], with a total amount of DNA of 3 µg/dish (82% ELIC and 18% eGFP DNA, both subcloned in pcDNA3).
Cells were bathed in an extracellular solution containing (mM): 150 KCl, 0.05 or 0.2 CaCl2, and 10 HEPES, pH adjusted to 7.4 with KOH (osmolarity 310 mOsm). Patch pipettes were pulled from thick-walled borosilicate glass (GC150F; Harvard Apparatus) and fire polished to a resistance of 8–12 MΩ. Intracellular solution contained (mM): 150 KCl, 0.5 CaCl2, 5 EGTA, and 10 HEPES, pH adjusted to 7.4 with KOH. Agonist-evoked currents were recorded at 20°C with an Axopatch 200B amplifier (Molecular Devices) from outside-out patches held at −100 mV. Patches were stepped to this holding voltage 0.2 s before the agonist was applied and otherwise held at −40 mV. No correction for junction potential was applied (calculated value 0.2 mV). Currents were filtered at 5 kHz, digitized at 50 kHz with Digidata 1322A, and saved directly on computer with Clampex software (all MDS Analytical Technologies).
All concentration jumps were performed using a piezo stepper (Burleigh instruments) with an application tool made from theta tube glass (Hilgenberg; final tip diameter, 150 µm). Voltage commands for the piezo stepper were 200 ms square pulses conditioned by low-pass eight-pole Bessel filtering (−3 dB frequency 5 kHz) to smooth oscillations. Actual exchange time was estimated by recording the open-tip response to the application of diluted extracellular solution (70% water) after rupture of the patch. Only patches in which the 20%–80% exchange time was faster than 250 µs were included in the analysis.
Agonist solutions were freshly prepared every day from 1 M stock solutions. Propylamine was applied at a concentration known to elicit maximum response (20 mM and 50 mM, for 50 and 200 µM Ca2+, respectively). Traces shown are averages of 5 or 10 individual agonist currents, separated by at least 10 s. Responses were averaged, and the time course of activation and deactivation (between 95% and 5% of the peak current level) was fitted with one exponential component (program Clampfit 9.0).
The coordinates of the P43 crystal form of ELIC in complex with Ba2+ have been deposited with the Protein Data Bank under code 2yn6.
|
10.1371/journal.ppat.1005151 | Inhibition of Translation Initiation by Protein 169: A Vaccinia Virus Strategy to Suppress Innate and Adaptive Immunity and Alter Virus Virulence | Vaccinia virus (VACV) is the prototypic orthopoxvirus and the vaccine used to eradicate smallpox. Here we show that VACV strain Western Reserve protein 169 is a cytoplasmic polypeptide expressed early during infection that is excluded from virus factories and inhibits the initiation of cap-dependent and cap-independent translation. Ectopic expression of protein 169 causes the accumulation of 80S ribosomes, a reduction of polysomes, and inhibition of protein expression deriving from activation of multiple innate immune signaling pathways. A virus lacking 169 (vΔ169) replicates and spreads normally in cell culture but is more virulent than parental and revertant control viruses in intranasal and intradermal murine models of infection. Intranasal infection by vΔ169 caused increased pro-inflammatory cytokines and chemokines, infiltration of pulmonary leukocytes, and lung weight. These alterations in innate immunity resulted in a stronger CD8+ T-cell memory response and better protection against virus challenge. This work illustrates how inhibition of host protein synthesis can be a strategy for virus suppression of innate and adaptive immunity.
| Long after smallpox was eradicated by vaccination with vaccinia virus, the study of this virus continues to reveal novel aspects of the interactions between a virus and the host in which it replicates. In this work we investigated the function of a previously uncharacterized VACV protein, called 169. The results show that protein 169 inhibits the synthesis of host proteins in cells and thereby provides a broad inhibition of the host innate immune response to infection. Unlike several other virus inhibitors of host protein synthesis, protein 169 acts by inhibiting the initiation of protein synthesis by both cap-dependent and cap-independent pathways. Also unlike several other virus protein synthesis inhibitors, the loss of protein 169 does not affect virus replication or spread, but the virus virulence was increased. This more severe infection is, however, cleared more rapidly and results in a stronger immunological memory response that is mediated by T-cells and provides better protection against re-infection. This work illustrates how shutting down host protein synthesis can be a strategy to block the host immune response to infection rather than a means to manufacture more virus particles.
| The study of virus-host interactions continues to provide valuable information about the complex relationships between cells and pathogens. Large DNA viruses, in particular, encode many proteins that modify the intracellular environment to promote viral survival, replication and spread. Vaccinia virus (VACV) is the prototypic Orthopoxvirus of the Poxviridae and is the vaccine used to eradicate smallpox [1]. VACV replicates in the cytoplasm and encodes about 200 proteins that are required for viral transcription and replication [2, 3], alteration of cell metabolism [4–7], and immune evasion [8].
Between one-third and one-half of VACV proteins are devoted to evasion of innate immunity and these immunevasins may function inside or outside the infected cell. Intracellular immunevasins include those that inhibit innate immune signaling pathways leading to activation of nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB), interferon (IFN) regulatory factor (IRF)-3 and Janus kinase (JAK) / signal transducer and activation of transcription (STAT) signaling. Other intracellular immunevasins suppress apoptosis or the antiviral activity of IFN-stimulated gene products. Additional immunevasins are secreted from infected cells to bind complement factors, IFNs, cytokines or chemokines extracellularly and inhibit their activity. An interesting aspect of these immune evasion strategies is the apparent redundancy, with several proteins targeting the same activation pathway. For instance, there are at least 10 intracellular inhibitors of NF-κB encoded by VACV [9–18] and a VACV strain lacking all these factors still inhibits NF-κB [19].
VACV, like all viruses, relies on host ribosomes for virus protein synthesis. To ensure efficient translation of virus proteins, VACV shuts off host protein synthesis and re-directs the cellular translational machinery to the synthesis of viral proteins [20–27]. VACV mRNAs are translated by a cap-dependent mechanism facilitated by the eukaryotic initiation factor (eIF)4F complex that recognizes the 5’-methylated cap, and translation is initiated by interaction of the cap with eIF4E, a cap-binding protein [28]. VACV encodes capping [29] and methylating enzymes [30] that produce viral mRNAs that mimic cellular mRNAs and so evade detection by host pattern recognition receptors. VACV protein synthesis occurs in virus factories [21, 27, 31], and to ensure preferential translation of virus mRNAs, VACV expresses de-capping enzymes D9 and D10 that remove the cap from both cellular and viral mRNAs [25, 32, 33]. The abundance of viral transcripts ensures translation of viral mRNA continues despite this de-capping activity, which also promotes turnover of viral mRNAs and thereby aids the transition between the early, intermediate and late stages of viral gene expression. The importance of protein D10 for the virus replication cycle is illustrated by a D10 deletion mutant that has a smaller plaque phenotype and produces reduced yields of virus in cell culture [26]. Moreover, mutant viruses with a stop codon introduced into the D10 open reading frame (ORF) or with amino acid alterations in the D10 catalytic site have an attenuated phenotype in vivo [34]. D9 and D10 also reduce dsRNA accumulation and the consequential activation of host responses [35]. A similar outcome was observed after VACV infection of cells lacking the host exonuclease Xrn1 [36].
This report presents a functional characterization of VACV strain Western Reserve (WR) protein 169, a previously uncharacterized protein that is expressed by some, but not all VACV strains and orthopoxviruses. Protein 169 is an inhibitor of cap-dependent and cap-independent translational initiation. Protein 169 localizes in cytoplasmic puncta and is largely excluded from virus factories, enabling preferential inhibition of host mRNA translation. Consistent with this, protein 169 does not affect virus replication or spread in cell culture, but is a potent inhibitor of translation in cells in which it is expressed ectopically. Consequently, protein 169 blocks expression of host proteins that are induced following activation of diverse innate immune signaling pathways, and, in two in vivo models of VACV infection, a virus lacking 169 (vΔ169) induces a more severe primary infection than control viruses. The altered disease severity is not due to changes in viral replication, but instead is associated with increased production of pro-inflammatory cytokines and chemokines, and increased recruitment of immune cells at the site of infection. This altered response also affects the adaptive memory response and causes increased CD8+ T-cell memory and better protection against virus challenge.
Collectively, these results indicate that virus inhibition of host protein synthesis can be a strategy to suppress innate and adaptive immunity, rather than primarily a means to aid virus replication as considered hitherto.
VACV strain WR gene 169R encodes a small, charged protein of 78 amino acid residues. The protein lacks a nuclear localisation signal and a hydrophobic transmembrane sequence suggesting that protein 169 is likely to be cytosolic. The ORF is conserved in VACV strains modified vaccinia virus Ankara (MVA), Lister, Duke, Acambis 3000 and rabbitpox virus, and other orthopoxviruses such as camelpox virus, taterapox virus, cowpox virus and monkeypox virus (S1 Fig). However, the ORF is truncated in multiple variola virus strains (the cause of smallpox) after codon 38 and in ectromelia virus (ECTV) after codon 41. In cowpox virus and monkeypox virus there are minor changes in amino acid length and composition, but the protein is identical in the VACV strains shown and in taterapox virus (S1 Fig). The truncation of this ORF in VACV strain Copenhagen and in other orthopoxviruses indicates that the 78 amino acid protein is non-essential for orthopoxvirus replication.
The expression of protein 169 by several VACV strains was investigated by immunoblotting using a rabbit polyclonal antibody raised against VACV WR protein 169 that had been expressed in and purified from E. coli (Methods). This detected a 13-kDa polypeptide in cells infected with VACV strains WR, MVA, Lister, rabbitpox, International Health Department (IHD)-J and Tian Tan, and cowpox virus strain Brighton Red, but not VACV strain Copenhagen, or in mock-infected cells (Fig 1A). VACV infection was confirmed by immunoblotting with a mAb that recognizes the VACV structural protein D8 [37], although this mAb did not detect the D8 protein made by MVA (Fig 1A). Immunoblotting for α-tubulin demonstrated equal loading of samples.
The time of expression and localization of protein 169 during infection were investigated by immunoblotting (Fig 1B and 1C) and immunofluorescence microscopy (Fig 1D). HeLa cells were infected with v169 (a plaque purified, wild-type virus that expresses protein 169) in the presence or absence of cytosine arabinoside (AraC), a DNA replication inhibitor that blocks intermediate and late VACV gene expression. The anti-169 antiserum detected a 13-kDa protein from 2 h p.i. that was also present following addition of AraC, showing expression prior to DNA replication (Fig 1B). Similar expression kinetics were observed for early VACV protein C16 [38]. In contrast, the VACV late protein D8 [39] was expressed only in the absence of AraC.
The localization of protein 169 was investigated by biochemical fractionation of infected cells. Immunoblotting of lysates from cells infected with v169, vΔ169 (a deletion mutant lacking the 169R gene) and v169-rev (a revertant virus in which the 169R gene was reinserted at its natural locus into vΔ169) showed that protein 169 is expressed from v169 and v169-rev, but not vΔ169, and that it localizes predominantly in the cytoplasm. Satisfactory separation of cytoplasmic and nuclear fractions was confirmed by blotting for α-tubulin and lamin (Fig 1C). Analysis by immunofluorescence using purified anti-169 antibody (Methods) detected protein 169 from 4 h p.i. in cytoplasmic puncta (Fig 1D). VACV factories were also detected from 4 h p.i. by DAPI staining, but protein 169 was excluded from these structures.
To determine if protein 169 co-localized with specific cytoplasmic organelles, infected cells were stained with antibodies that detected the endoplasmic reticulum, mitochondria, Golgi apparatus, clathrin-containing vesicles and endosomes but no clear co-localization was observed (Fig 2A). Partial co-localization with 40S ribosomes was noted, although the abundance of 40S ribosomes makes a clear correlation uncertain. Staining with DAPI confirmed that protein 169 was excluded from virus factories (Fig 2B).
The contribution of protein 169 to virus replication and spread was investigated using recombinant VACVs v169, vΔ169, and v169-rev that were constructed by transient dominant selection [40] (Methods). These three viruses formed plaques of indistinguishable size in African monkey fibroblasts (BSC-1) and also in rabbit kidney (RK)-13 cells and human TK-143 cells (Fig 3A–3C). Similarly, the yields of intracellular and extracellular vΔ169 were unaltered compared to control viruses after high (10 PFU/cell) or low (0.05 PFU/cell) multiplicity of infection in BSC-1 cells (Fig 3D–3G). Therefore, the 169 protein is non-essential for virus replication and spread in cell culture.
The 169R gene is located in a terminal variable region of the VACV genome, is expressed early during infection and is non-essential for virus replication in cell culture. These properties are characteristic of VACV genes encoding immunevasins, such as the type I IFN binding protein [41, 42], the 3-β-hydroxysteroid dehydrogenase [43, 44] and the intracellular inhibitors of NF-κB activation [9, 10, 12–18]. Therefore, we hypothesized that protein 169 might be an immunevasin and this was tested by reporter gene assays. A plasmid in which firefly luciferase expression is driven by either an NF-κB, IRF-3 (ISG56.1), or interferon-stimulated response element (ISRE) responsive promoter was transfected separately into HEK 293T cells together with TK renilla luciferase (internal control), and plasmids expressing 169, FLAG-tagged 169 (FLAG-169) or other control proteins. The controls chosen were (i) VACV strain WR protein B14 that inhibits the NF-κB signaling by binding to IKKβ [15], (ii) VACV protein C6 that inhibits IRF-3 signaling by binding to TBK-1 adaptors [45], and (iii) paramyxovirus protein PiV5-V that inhibits type I IFN-induced signaling by degrading STAT1 [46]. Luciferase activity was measured by luminescence after stimulation with TNF-α (NF-κB Luc), IFN-α (ISRE Luc) or after transfection with poly (I:C) (IRF-3 Luc). Protein 169 and FLAG-169 inhibited NF-κB, IRF-3 and ISRE pathways as well as, or better than, known inhibitors of these pathways (Fig 4A–4C). The inhibition of all these pathways was surprising, and contrasted with the controls that generally inhibit specific pathways only. Interestingly, protein 169 also caused reduced expression of TK renilla, suggesting a general reduction in protein expression.
To investigate this further, the levels of chemokine CXCL10 were measured by ELISA. HEK 293T cells were transfected with plasmids expressing GFP, VACV B14, C6, 169 or Δ12A49 and then infected with Sendai virus (SeV). VACV protein A49 inhibits NF-κB signaling by binding to the E3 ubiquitin ligase β-TrCP but deletion of the first 12 amino acids abolishes this function [17] and so Δ12A49 served as a negative control. After 24 h, CXCL10 in the supernatant was measured by ELISA (Fig 4D). CXCL10 expression is induced by both NF-κB and IRF-3, and so levels of CXCL10 were lower in cells expressing either B14 or C6, but not in cells expressing Δ12A49, as expected. However, protein 169 also reduced CXCL10 levels, consistent with results of the reporter gene assays.
To test whether 169 mediates its inhibitory activity by blocking transcription, the levels of specific mRNAs were measured. A549 cells were transfected with plasmids expressing GFP, VACV B14, C6, or 169 and were stimulated 24 h later with TNF-α. mRNA levels of NF-κB-inducible genes such as intercellular adhesion molecule 1 (ICAM-1), IL-6, and NFκBia were measured by reverse transcription quantitative-PCR (RT-q-PCR) and normalized to the housekeeping gene hypoxanthine-guanine phosphoribosyltransferase (HPRT) (Fig 4E–4G). Levels of all three mRNAs were similar in cells expressing 169, GFP, or C6 following stimulation with TNF-α. Conversely, as expected, lower levels of these NF-κB-inducible mRNAs were detected in cells expressing the NF-κB inhibitor B14. No difference was detected in HPRT mRNA levels, confirming that the 169-mediated inhibition of multiple immune signaling pathways was not due to a general inhibition of transcription. Therefore, it was likely that protein 169 inhibited gene expression either by blocking mRNA transport to the cytoplasm, or by blocking protein synthesis. The former possibility was unlikely given that protein 169 is cytoplasmic, but was addressed by measuring the levels of cytoplasmic and nuclear mRNAs.
HEK 293T cells were co-transfected with plasmids expressing NEMO fused with renilla luciferase (NEMO-Luc) and protein 169. A plasmid expressing protein A49 and an empty vector were included as negative controls and cycloheximide was added as an inhibitor of translation. The levels of luciferase-tagged proteins were determined by luminescence (S2A Fig) and mRNA levels of NEMO-Luc were determined by RT-q-PCR (S2B Fig). In parallel, cytoplasmic and nuclear mRNAs were extracted and mRNAs levels of NEMO-Luc, HPRT and TATA box-binding protein were compared in these fractions (S2D–S2F Fig). As before, only low levels of NEMO-Luc was detected in cells expressing 169 or treated with cycloheximide. Slightly lower cytoplasmic mRNA levels of NEMO-Luc were found in cells expressing 169, but this slight decrease could not explain the profound (~10-fold) reduction of NEMO-Luc. There was also a slight reduction in NEMO-Luc mRNA in the cytoplasm in cycloheximide-treated cells, suggesting such reduction might derive from a general inhibition in protein synthesis. Lastly, no decrease in endogenous mRNAs was observed in the presence of protein 169.
Collectively these data indicate that mRNA transcription and export are not inhibited by protein 169 and therefore its inhibitory effect is downstream.
To investigate if protein 169 inhibits protein synthesis, HeLa cells were co-transfected with plasmids expressing GFP together with VACV N1, 169, FLAG-169 or empty vector. VACV N1 is another inhibitor of NF-κB signaling [47, 48] and served as a negative control. GFP levels were determined by immunoblotting and GFP mRNAs were measured by RT-q-PCR (Fig 5A and 5B). Cycloheximide, 169 and FLAG-169 reduced GFP levels greatly compared with N1 or empty vector. In contrast, GFP mRNA levels were similar in all cells and were higher in cells treated with cycloheximide. These experiments reveal that protein 169 inhibited protein synthesis and that this is generic rather than being specific to proteins functioning in innate immunity.
VACV inhibits cap-dependent translation of host mRNAs by the de-capping enzymes D9 and D10, but these do not affect cap-independent translation [25]. To determine if protein 169 has similar or different specificity, its ability to inhibit cap-dependent and internal ribosome entry site (IRES)-dependent translation was evaluated. A plasmid encoding a bicistronic gene in which firefly luciferase is translated in a cap-dependent manner and renilla luciferase is translated in a foot and mouth disease virus (FMDV) IRES-dependent manner was transfected into HEK 293T cells together with 169, FLAG-169 or 169-AAG. The latter plasmid has the 169 initiation codon and the fourth codon mutated from AUG to AAG to prevent translation and distinguish between inhibition mediated by 169 mRNA or 169 protein. Luciferase levels were determined by luminescence (Fig 5C and 5D), mRNA levels were determined by RT-q-PCR (Fig 5E and 5F) and protein expression was also measured by immunoblotting (Fig 5G). Low levels of both firefly and renilla luciferase were found in the presence of cycloheximide, 169 and FLAG-169, but not 169 AAG, confirming that the inhibitory effect of 169 on translation requires protein 169. In contrast, luciferase levels were unaffected by proteins N1 or A49. Similar mRNA levels of renilla luciferase were found in all samples. These data show that protein 169 inhibits both cap-dependent and FMDV IRES-dependent translation.
To evaluate the influence of protein 169 on protein synthesis in uninfected cells and during VACV infection, nascent proteins were analysed using surface sensing of translation (SUnSET) [49]. SUnSET is a non-radioactive method for monitoring protein synthesis that uses incorporation of puromycin into nascent polypeptide chains and causes termination of elongation. Puromycin-tagged polypeptides are then detected by immunoblotting with anti-puromycin antibody. In HEK 293 Trex cells expressing protein 169, protein synthesis was inhibited increasingly from 8 h post induction (Fig 6A). In contrast, in a control cell line expressing C6.TAP inducibly [50] no such inhibition was seen (Fig 6B).
The effect of 169 on protein synthesis during VACV infection was tested next. HeLa cells were infected with v169 or vΔ169, and puromycin was added at different times p.i. (Fig 6C). Host protein synthesis was inhibited by 6 h p.i. and more profoundly thereafter, but no difference was detected between v169 and vΔ169. This could be due to both viruses expressing the de-capping enzymes D9 and D10 that have profound effects on virus protein synthesis [25, 26, 32] and might mask effects of protein 169. This result is consistent with the observations that protein 169 is absent from virus factories (Figs 1D and 2), and does not affect virus replication and spread (Fig 3), suggesting that protein 169 might preferentially target host protein synthesis.
To determine at which stage of protein synthesis protein 169 might be acting, polysomes were profiled in HEK 293 Trex 169 cells with or without protein 169 expression (Fig 7A and 7B). Cytoplasmic extracts were prepared in the presence of cycloheximide to retain intact monosomes and polysomes and these were analyzed by sucrose density gradient centrifugation. The RNA and protein composition of the gradient was measured by absorbance (A254 nm) and immunoblotting, respectively. Protein 169 expression caused an increase in 80S ribosomes and decrease in polysomes (Fig 7B), indicating an inhibition of translational initiation. Immunoblotting of gradient fractions revealed that protein 169 co-purified partially with the 40S ribosomal fraction (Fig 7B), consistent with immunofluorescence data (Fig 2A). For comparison, HEK 293 Trex C6.TAP cells were analyzed in parallel and protein C6 expression caused no such alterations to polysomes or 80S monosomes (Fig 8A and 8B).
To investigate whether the 80S ribosomes accumulating in the presence of protein 169 contain mRNA, polysome profiling was repeated in a higher salt buffer (400 mM KCl), conditions in which 80S ribosomes lacking mRNA dissociate into constituent subunits. However, in the presence of protein 169, the 80S peak remained stable in high salt (Fig 9B), indicating that the 80S ribosomes are associated with mRNA. Increasing the concentration of salt in the sucrose density gradient reduced the sharpness of the peaks obtained. To confirm that this effect was due to the high salt concentration, the polysome profile of uninduced HEK 293 Trex 169 cytoplasmic cell lysates was examined on sucrose gradients (Fig 9C and 9D). Again, the high salt condition affects the overall sharpness of polysomal fractions independently of the expression of protein 169.
Since protein 169 inhibited cap-dependent and FMDV IRES-dependent translation, both of which require the concerted action of multiple eIFs, we tested whether protein 169 could affect translation from the cricket paralysis virus (CrPV) intergenic region (IGR) IRES. This IRES uses an unusual mechanism of translation initiation binding directly to the 40S subunit and initiating from the A-site without the requirement for initiation factors [51]. A plasmid encoding a bicistronic gene in which renilla luciferase is translated in a cap-dependent manner and firefly luciferase is translated in a CrPV IRES-dependent manner was transfected into HEK 293T cells together with A49.TAP, 169 or empty vector control. Luciferase levels were measured by luminescence (Fig 10A and 10B) and protein expression was determined by immunoblotting (Fig 10C). Low levels of both firefly and renilla luciferase were found in cycloheximide-treated cells as well as in cells expressing protein 169 indicating that protein 169 can inhibit translation from an IRES that does not require the activity of any initiation factors.
Taken together these data indicate that protein 169 inhibits the initiation of translation causing accumulation of 80S ribosomes and that this applies to both cap-dependent and IRES-dependent translation. This generic shut-down of host protein synthesis, while virus protein synthesis remains largely unaffected, affects the expression of many proteins induced by activation of innate immune sensing pathways and results in inhibition of innate immunity within infected cells. Such a strategy would be predicted to affect the outcome of infection in vivo and therefore this hypothesis was investigated.
The contribution of 169 to virus virulence was examined using two murine models of infection. The intranasal (i.n.) model represents a systemic infection, where the virus replicates in the lungs and spreads to other organs. Virus virulence is assessed by measuring weight loss, virus titers and signs of illness [52, 53]. In the intradermal (i.d.) model, mice are inoculated by intradermal injection into the ear pinna, which results in a localized infection, and virulence is determined by measuring lesion size and healing time [54, 55].
In the i.n. model, infection with vΔ169 resulted in significantly greater weight loss from day 5 onwards and more severe signs of illness than control viruses (Fig 11A). To investigate the basis for these differences, the levels of cytokines and chemokines in broncho-alveolar lavage (BAL) fluids were measured early (24 h) p.i. This showed that there were enhanced levels of IL-2, IL-6, TNF-α, CCL11, CXCL9 and CXCL10 following infection with vΔ169 compared to both control viruses, whereas the levels of CCL2, CCL9, IL-12 and IL-15 were unchanged (Fig 11B and 11C). Furthermore, infection with vΔ169 caused increased lung weights and the number of cells in BAL fluids on days 4 and 7 p.i. compared to control viruses (Fig 11D and 11E). Measurement of lung virus titers showed that all three viruses had replicated to the same extent on days 2 and 4 p.i., but by day 7 the titer of vΔ169 had decreased more than controls, indicating more rapid clearance (Fig 11F). These observations show that infection with vΔ169 caused a greater inflammatory response, with elevated synthesis of several cytokines and chemokines, enhanced recruitment of cells into BAL fluids and more rapid virus clearance.
To analyze the nature of cells recruited into BAL fluids, the cells were stained with monoclonal antibodies and quantified by flow cytometry. The majority of inflammatory cells recruited during infection were macrophages (Fig 12A) and lymphocytes (Fig 12C) including CD4+ and CD8+ T-cells (Fig 12F, 12G and 12H), with fewer numbers of neutrophils, (Fig 12B), NK cells (Fig 12D) and B cells (Fig 12E). Notably on days 4 and 7 p.i. the recruitment of macrophages, total lymphocytes, T-cells, and CD4+ and CD8+ T-cells was increased following infection with vΔ169 compared to controls, and these differences may explain the more rapid clearance of this virus. In contrast, neutrophils (Fig 12B), NK cells (Fig 12D), and B cells (Fig 12E) showed no difference between all viruses.
Changes in the inflammatory response to primary infection can alter the adaptive response and subsequent protection against virus challenge. This has been observed with VACV mutants that either have increased virulence, such the VACV WR strain lacking the soluble chemokine binding protein A41 [56–58], or decreased virulence, such as the inhibitor of IRF-3 activation C6 [45, 59, 60] and the inhibitor of apoptosis and NF-κB activation N1 [47, 61, 62]. A more severe primary infection can also lead to better protection [57], and to test whether enhanced immune response generated by vΔ169 is advantageous and would lead to better protection, the potency of vΔ169 as a vaccine was evaluated. Mice were immunized via the i.n. route with v169, vΔ169 or v169-rev and then were challenged with wild type virus i.n. at day 28 (Fig 13A). In this model, vΔ169 induced better protection against challenge as shown by reduced weight loss compared to controls (Fig 13A). To investigate the basis for this, the levels of VACV neutralizing antibodies were determined by plaque reduction neutralization assay (Fig 13B) and the cytotoxicity of NK cells on uninfected YAC-1 cells (Fig 13C) and CD8+ splenic T-lymphocytes (Fig 13D) on VACV-infected P815 cells was measured by chromium release assay. At day 28 p.i. all groups of immunized mice had high serum antibody titers that did not differ between the groups (Fig 13B). Similarly, the cytotoxicity of splenic NK cells on YAC-1 cell targets did not differ between the groups (Fig 13C). However, the lysis of target cells by splenic CD8+ T-cells within the total splenocyte population from mice infected with vΔ169 was significantly greater than lysis by cells from mice infected by control viruses (Fig 13D). Collectively, these data show that immunization with vΔ169 generates stronger CD8+ T-cell immunological memory and better protection against challenge.
The virulence and immunogenicity of vΔ169 was also assessed after intradermal (i.d.) infection (Fig 14). vΔ169 caused a statistically significant increase in lesion size and duration compared to control viruses (Fig 14A). Furthermore, as observed in the i.n. model, viral titers in the ears showed that all viruses replicated to a similar extent initially (day 3 and 6), but thereafter (days 10 and 14) viral titers were lower for vΔ169 compared to controls (Fig 14B). Additionally, mice immunized via the i.d. route with v169, vΔ169 or v169-rev were challenged with wild type virus i.n. at day 28 (Fig 14C). As observed for i.n. model, vΔ169 induced better protection against challenge as shown by reduced weight loss of mice immunized with vΔ169 compared to controls.
A functional study of VACV WR protein 169 is presented. This small, highly charged protein is expressed early during VACV infection, localizes in cytoplasmic puncta but is excluded from virus factories, and inhibits the initiation of cap-dependent and cap-independent protein synthesis. Thereby, protein 169 reduces production of host inflammatory mediators induced by activation of multiple innate immune signaling pathways. Protein 169 is conserved in many VACV strains and orthopoxviruses but nonetheless is non-essential for virus replication or spread in tissue culture. Instead, it affects the outcome of infection in vivo by decreasing the recruitment of inflammatory leukocytes, delaying clearance of virus, reducing the memory CD8+ T-cell response and diminishing protection against subsequent virus challenge.
The ability of protein 169 to inhibit the innate immune response, while not affecting virus replication in cell culture or in vivo, is characteristic of many VACV immunevasins [8]. However, a striking difference between many of the immunevasins characterized hitherto and protein 169 is that the former are often inhibitors of an individual innate immune signaling pathway (or sometimes two pathways) by binding to one or two specific host proteins. In contrast, protein 169 is a general inhibitor of protein synthesis and targets multiple pathways that require nascent protein synthesis. Thus, by blocking the translation of host mRNAs that are transcribed, for instance, following activation of NF-κB, IRF-3 or the JAK/STAT signaling pathways, there is a decreased production of many inflammatory mediators and consequential reduced recruitment of leukocytes to the site of infection.
The inhibition of host protein synthesis by viruses is widespread, but hitherto has been considered largely a strategy by which viruses subvert host metabolism to increase virus protein synthesis and production of virions. VACV protein 169 illustrates another purpose, namely, the decrease of host protein synthesis without a concomitant increase in production of virus proteins or infectious virus particles, but with the consequence of restricting the host innate immune response to infection, so aiding virus escape and diminishing immunological memory. Protein 169 is well adapted to this purpose for it is excluded from virus factories, the site of virus protein synthesis, and so targets host translation preferentially, and its loss does not affect virus replication in cell culture or in vivo.
Protein 169 is also unusual in that it targets both cap-dependent and cap-independent translation. Many RNA viruses exploit IRES-dependent translation to manufacture their proteins while disabling cap-dependent translation of host mRNAs by targeting the eIF4F complex. Popular strategies are (i) cleavage of eIF4G by viral proteases [63–65], (ii) cleavage of poly A-binding protein [66, 67], and (iii) decreasing phosphorylation of cap-binding protein eIF4E [68, 69]. In contrast, DNA viruses use mostly cap-dependent translation and stimulate eIF4F formation. Herpes simplex virus type 1 (HSV-1) protein ICP0 promotes phosphorylation of eIF4E and 4E-binding protein 1 (4E-BP1) that leads to degradation of 4E-BP1 and stimulation of formation of the eIF4F complex [70]. Also, HSV-1 protein ICP6 binds eIF4G to enhance eIF4F assembly [71]. VACV stimulates eIF4F complex formation through hyper-phosphorylation of 4E-BP1 enabling interaction between eIF4E and eIF4G [27]. Protein 169 acts differently, but has some similarity with the modulation of protein synthesis by hepatitis C virus (HCV) in that it leads to alterations in innate immunity. HCV relies mainly on IRES-dependent translation and causes stimulation of protein kinase R that leads to translation inhibition through phosphorylation of eIF2α to inhibit production of IFN stimulated genes (ISGs) [72]. However, the factors responsible for these changes and their mechanism of action remain unknown.
Protein 169 is the third VACV polypeptide shown to inhibit protein synthesis, the others being the de-capping enzymes D9 and D10 [25, 32, 33]. These enzymes are made either early or late during infection and de-cap both host and viral mRNAs, although some preferential affinity for different cap structures have been shown [33]. Since the viral mRNAs are synthesized in greater abundance, these soon become predominant and so virus proteins are made while host protein synthesis declines. Rapid mRNA turnover is also important for progression between early, intermediate and late stages of VACV gene expression. The importance of de-capping for virus replication is illustrated by the loss or mutation of protein D10 that results in a smaller plaque phenotype, accumulation of early transcripts, lower virus yield [26] and attenuation in vivo [34]. Recently a VACV strain expressing catalytically dead versions of D9 and D10 was shown to induce large amounts of dsRNA. This activates pathways leading to inhibition of protein synthesis and consequently reduces virus production and results in severe attenuation in vivo [35]. In contrast, loss of protein 169 has no effect on virus replication in vitro (Fig 3) or in vivo (Figs 11 and 14) and its loss causes an increase in virulence in both i.n. and i.d. models of infection (Figs 11 and 14). In the i.n. model, infection by vΔ169 caused enhanced production of several cytokines (IL-2, IL-6 and TNF-α) and chemokines (CCL11, CXCL9 and CXCL10) within 1 day p.i. (Fig 11) and subsequent greater recruitment of macrophages and CD4+ and CD8+ T cells (Fig 12) and increased lung weight (Fig 11D). Later, this greater recruitment of inflammatory cells leads to more rapid virus clearance and recovery (Fig 11). Similarly, in the i.d. model the greater inflammatory response is reflected in a greater lesion size, but again this is followed by more rapid virus clearance and recovery (Fig 14).
The early expression of protein 169 is consistent with prior RNA analysis of the VACV genome that showed early transcription of this ORF and 169 mRNAs were detected from 1 h p.i. [73, 74]. Sometimes viruses that induce exacerbated immune responses are more virulent and in this regard it is notable that orthopoxviruses lacking ORF 169 are generally of high virulence. For instance, all sequenced variola viruses and ectromelia virus lack ORF 169 and these viruses are highly virulent in man or mice, causing smallpox and mousepox, respectively. Similarly, VACV strain Copenhagen lacks ORF 169 and caused a higher frequency of post-vaccination complications in man than the more widely used VACV strains Lister and New York City Board of Heath (Wyeth) [1]. VACV strain Copenhagen also caused larger lesion sizes in the mouse intradermal model in comparison to other VACV strains used as smallpox vaccines in man [55]. However, VACV strain Copenhagen and all variola virus strains also lack another factor that diminishes virulence, namely the soluble IL-1β binding protein encoded by gene B15R of VACV strain WR [53, 75], and the causes of enhanced virulence are probably multi-factorial.
The increased virulence seen by loss of gene 169R has a few parallels in orthopoxvirus biology. In addition to deletion of the soluble IL-1β receptor encoded by VACV WR mentioned above [53], deletion of the chemokine binding protein A41 [56], and the B13 serine protease inhibitor [55] each caused an increase in virulence in either the i.n. or i.d. model, and in some cases also induced a stronger immunological memory response that resulted in better protection against virus challenge [57, 76]. Infection with vΔ169 generated a stronger innate response (Figs 11 and 12), that led to a stronger memory CD8+ T cell response and better protection to virus challenge (Figs 13 and 14). Increased immunological memory responses and better protection against challenge have also been observed with VACV mutants with diminished virulence, such as viruses lacking the C6 or N1 proteins [60, 77].
Protein 169 localizes mainly in the cytoplasm of infected cells throughout the course of infection. The punctate pattern observed might suggest co-localization of 169 with some specific organelles, but only some partial overlap with 40S ribosomes was observed. The precise mechanism by which protein 169 inhibits translation remains to be determined, but the polysome profiling experiments described (Figs 7–9) reveal that protein 169 expression leads to an accumulation of 80S monosomes and reduction of polysomes, particularly of heavier polysomes. This pattern is consistent with a reduced rate of translation initiation, and the stability of the 80S monosomes in high-salt indicates that the 80S ribosomes are mRNA-associated, rather than present in a free pool [78]. Reducing a pool of free ribosome is a strategy used by cardiovirus protein 2A that, in contrast to protein 169, causes accumulation of monosomes free of mRNA [79]. A direct interaction between protein 169 and either the mRNA cap or 40S subunit was not observed, nor was an effect of protein 169 on translation in vitro using rabbit reticulocyte lysate. However, we cannot be sure whether the prepared fraction of protein 169 is functional under the conditions tested. Nonetheless, the capacity of protein 169 to block FMDV IRES-directed translation initiation is consistent with an eIF4E-independent inhibitory mechanism. In addition, the inhibition of CrPV IRES-dependent translation by protein 169 suggests that its inhibitory activity is not mediated by interference with other eIFs.
In summary, 169 is an inhibitor of cap-dependent and cap-independent translation, it affects virus virulence and contributes to VACV immunogenicity by diminishing the innate and adaptive immune response. This study illustrates that viral inhibition of protein synthesis can be an immune evasion strategy rather than a mechanism to increase yields of virus from infected cells.
This work was carried out in accordance with regulations of The Animals (Scientific Procedures) Act 1986. All procedures were approved by the United Kingdom Home Office and carried out under the Home Office project licence PPL 70/7116.
BSC-1 (ATCC CCL-26), CV-1 (ATCC CCL70), HEK 293T (ATCC CRL-11268) and A549 (ATCC CCL-185) cells were maintained in Dulbecco’s modified minimal essential medium (DMEM) containing 10% fetal bovine serum (FBS) and penicillin/streptomycin (50 μg/ml). RK-13 (ATCC CCL-37) and TK-143 (ATCC CRL-8303) cells were grown in minimum essential medium (MEM) and supplemented as above. HeLa (ATCC CCL-2) cells were grown in MEM with addition of non-essential amino acids (1%) and supplemented as above. HEK 293 Trex (Invitrogen) cells were maintained in DMEM containing 10 μg/ml blasticidin, 100 μg/ml zeocin and supplemented as above.
The sequence of the VACV WR 169R gene was codon optimized by GENEART for expression in mammalian cells. 169R was then sub-cloned into mammalian expression vectors pcDNA 3.1 or pcDNA4 TO (Invitrogen) without a tag or with an N-terminal FLAG tag. E. coli expression plasmid pOPINE were engineered to express a 169R wild type sequence with a C-terminal His tag (169-His) and plasmid pGEX-6p-1 was engineered to express a 169R wild type sequence with an N-terminal glutathione S-transferase (GST) tag (GST-169). Plasmid Z11-Δ169 was used to construct the VACV mutant lacking gene 169R and contained flanking regions of the 169R gene locus cloned into plasmid Z11 that contains the E. coli guanine phosphoribosyltransferase (Ecogpt) fused with enhanced green fluorescent protein (EGFP) driven by an early/late VACV promoter as described [45]. Plasmid Z11-169-rev was used to construct the revertant virus v169-rev and contains the 169R gene and flanking sequences inserted into Z11 plasmid. A plasmid encoding a bicistronic gene expressing firefly luciferase in a cap-dependent manner and renilla luciferase in a FMDV IRES-dependant manner was a kind gift from Prof. Ian Goodfellow, Department of Pathology, University of Cambridge. A plasmid encoding a bicistronic reporter gene expressing firefly luciferase in a cricket paralysis virus (CrPV) IRES-dependent manner and renilla luciferase in a cap-dependant manner was a kind gift from Dr. Eric Jan, Department of Biochemistry and Molecular Biology, University of British Columbia, Canada. NF-κB-Luc, ISRE-Luc and TK renilla was obtained from Dr. Andrew Bowie (Trinity College, Dublin, Ireland), ISG56.1 Luc was from Ganeth Sen (Lerner Research Institute, Ohio), and M5P Luciferase–NEMO (Luc-NEMO) and M5P GFP-FLAG were obtained from Dr. Felix Randow (MRC Laboratory of Molecular Biology, Cambridge, United Kingdom). C6.TAP, N1.TAP, B14.FLAG and A49.TAP were described previously [15, 17, 45, 47]. V5-PiV5-V was provided by Jennifer H. Stuart (Department of Pathology, University of Cambridge, UK).
Rabbit polyclonal antiserum raised against recombinant 169 protein was used for immunoblotting (diluted 1:1000–2000) and purified anti-169 770 P antibody was used for immunofluorescence (diluted 1:50). Other antibodies used were mouse anti-FLAG (Sigma, F1804, diluted 1:1000), rabbit anti-FLAG (Sigma-Aldrich, F7425, diluted 1:5000), anti-D8 mouse mAb AB1.1 against VACV protein D8 [39] (diluted 1:500), anti-α-tubulin (Millipore, 05–829, diluted 1:5000), anti-actin (Sigma, A2066, diluted 1:1000), anti-lamins A+C (Abcam, ab898, diluted 1:1000), anti-VACV protein C16 [38] (diluted 1:1000), anti-protein disulphide isomerase (PDI, 1D3 clone, Enzo Life Sciences, diluted 1:50), anti-GM130 (Transduction laboratories, diluted 1:300), anti-clathrin (Abcam, diluted 1:50), anti-human transferrin receptor (Zymed, used at 2.5 μg/ml), anti-ribosomal protein S6 (Cell Signalling, diluted 1:25 for immunofluorescence and 1:1000 for immunoblotting), anti-eIF4AI (Santa Cruz Biotechnology, N-19, diluted 1:500 dilution, a kind gift from Prof. Ian Goodfellow), anti-eIF4E (Santa Cruz Biotechnology, A-10 diluted 1:500), anti-ribosomal protein L29 (Santa Cruz Biotechnology, P-14, dilution 1:200), anti-puromycin (Millipore, clone 12D10, diluted 1:15000–1:25000), Alexa Fluor 488 goat anti-rabbit IgG (H+L) (Invitrogen, A-11008, diluted 1:750), Alexa Fluor 549 donkey anti-mouse IgG (H+L) (Invitrogen, A-10036, diluted 1:750), and MitoTracker Red CM-H2XRos (Invitrogen, M7513, diluted 1:5000). Reagents used in this study were puromycin (InvivoGen), cycloheximide (Calbiochem), doxycycline (Melford), blasticidin (Gibco) and zeocin (Invitrogen).
VACV vΔ169 was constructed by transfecting plasmid Z11-Δ169 into VACV WR infected CV-1 cells using FuGENE 6 and a recombinant VACV was isolated by transient dominant selection [40] as described for other VACV deletion mutants [12, 80]. Plaque purified wild type 169 (v169) and deletion 169 (vΔ169) viruses were isolated from the same intermediate virus and were genotyped using PCR and primers amplifying the flanking regions of the 169R locus. The revertant 169 virus (v169-rev) was constructed by transfection of plasmid Z11-169-rev into vΔ169-infected CV-1 cells following the same procedure as described above. Genomic DNA isolated from recombinant VACVs (v169, vΔ169 and v169-rev) were compared to parental VACV WR virus using restriction endonuclease digestion with HindIII or SphI digestion and virus DNA was visualized after pulsed field gel electrophoresis.
E. coli BL21(DE3) R3 pRARE cells (kind gift from SGC Oxford), where R3 denotes a derivative of BL21(DE3) resistant to a strain of T1 bacteriophage (SGC Oxford) and the pRARE plasmid originates from the Rosetta strain (Novagen) and supplies tRNAs for rare codons, were transformed with the 169-His expression plasmid. The bacteria were grown in terrific broth and the expression of 169-His was induced by 1 mM IPTG at 37°C for 6 h. Bacteria were collected by centrifugation, lysed and disrupted by sonication. 169-His was purified from the soluble fraction by immobilized metal affinity chromatography (IMAC) using a His-Trap HP column followed by ion exchange chromatography (IEX) using a MonoQ GL column. Three and a half mg of 169-His was used to inoculate two rabbits (Eurogentec, Seraing, Belgium) to obtain polyclonal sera. Two rabbits (number 770, 771) were immunized at day 0, 14, 28 and 56 with Freund's complete adjuvant at day 0 and with incomplete Freund's adjuvant for the boosts with dose of 400 μg of 169-His. Sera prepared from venous blood drawn before immunization and at day 66 were tested for recognition of protein 169 expressed during VACV infection. Serum from rabbit 771 was sensitive enough to detect protein 169 from VACV-infected cells. This serum was used for immunoblotting analysis throughout this study (further referred as anti-169). Serum from rabbit 770 was further purified against GST-169 using AminoLink immobilization kit. GST-169 protein was produced in BL21(DE3) E. coli bacteria (Merck Millipore) transformed with pGEX-6p-1 GST-169 plasmid. Bacteria were grown in LB and expression of GST-169 was induced by 1 mM IPTG at 37°C for 6 h. Bacteria were collected by centrifugation, lysed and disrupted by sonication. GST-169 was purified from soluble fraction using glutathione-sepharose 4B and size exclusion chromatography (SEC) using Superdex 75 10/300 GL column. Two mg of GST-169 was used for polyclonal serum purification using AminoLink immobilization kit following the manufacturer’s instructions for the pH 7 protocol. Protein 169-specific purified IgG (further referred as an anti-169 purified antibody) were used for immunofluorescence studies.
For analysis of virus single step growth properties, BSC-1 cells were infected at 10 PFU/cell for 12 or and 24 h. Extracellular virus in the clarified growth medium (after centrifugation at 500 x g for 10 min) was titrated by plaque assay on BSC-1 cells. Cell associated virus was measured by scraping cells from the plastic flask, combining these with the debri from the supernatant and collection by centrifugation as above. Cells were then disrupted by three rounds of freeze-thawing and sonication and the virus was titrated by plaque assay on BSC-1 cells. For analysis of multiple step growth properties, BSC-1 cells were infected at 0.05 PFU/cell for 24 and 48 h. The extracellular and cell-associated viral titers were determined as described above.
BSC-1, RK-13 and TK-143 cells were infected with the indicated VACVs at 50 PFU/ well of a 6-well plate. The radius of plaques was measured after 72 h using Axiovision 4.8.2 software on an Axiovert.A1 microscope (Zeiss) with Axiocam MRc. In each condition 20 plaques per virus were measured in three independent experiments.
For intranasal (i.n.) model of infection, BALB/c mice (6–8 weeks old) were inoculated with VACVs, which had been purified by sedimentation twice through a sucrose cushion, (5 × 103 PFU into each nostril) and monitored daily for a weight loss and scored for signs of illness as follows hair ruffling, back arching, reduced mobility, pneumonia [52, 53]. For the intradermal (i.d.) model of infection, female C57BL/6 mice (6–8 weeks old) were inoculated with purified VACVs (104 PFU) in both ear pinna and the diameter of the lesion was measured daily using a micrometer [54]. The administered dose was confirmed by plaque assay. For challenge experiments, immunized animals were challenged i.n. 28 d p.i. with 5 × 106 PFU of v169 and weighed daily thereafter.
Bronchial alveolar lavage (BAL) fluids were prepared on the indicated days. These were centrifuged at 1500 g to obtain cells for flow cytometry and the clarified supernatant was used for ELISA. Live cells collected from BAL fluids were counted using a haemocytometer following staining with trypan blue.
For determination of lung and ear tissues viral titers, the lungs and ears tissues were homogenized and washed through a 70 μm nylon mesh using DMEM and 10% FBS. Cells were then frozen and thawed three times, and sonicated thoroughly to liberate intracellular virus. Infectious virus was titrated in duplicate by plaque assay on BSC-1 cell monolayers.
For chromium-release cytotoxicity assay, NK cell cytotoxicity and VACV-specific cytotoxic T lymphocyte (CTL) activity within total splenocyte populations was assayed with a standard 51Cr-release assay as described [77]. NK-mediated lysis was tested on uninfected YAC-1 cells, while VACV-infected P815 cells (H-2d, mastocytoma) were used as targets for VACV-specific CTL lysis. The percentage of specific 51Cr release was calculated as specific lysis = [(experimental release−spontaneous release)/(total detergent release−spontaneous release)]×100. The spontaneous release values were always < 10% of total lysis.
Anti-mouse CD3 (clone 145-2C11), CD4 (GK1.5), CD8 (5H10-1), CD45 (30-F11), CD45R (RA-6B2), NK1.1 (PK136), CD11b (M1/70), F4/80 (BM8), Ly-6G/Ly-6C (RB6-8C5), Ly6G (1A8) and CD16/32 (2.4G2) mAb were purchased from BD Biosciences or Biolegend. The mAbs were purified or conjugated with FITC, PerCP/cy5.5, APC, PE-Cy7, APC/Cy7, BV650 or PE. Isotype controls were used as negative controls. Flow cytometry was performed with a BD LSR Fortessa (BD Biosciences), and data were analyzed with FlowJo software (Tree Star Inc.). Events were gated for live lymphocytes on foward scatter × side scatter and dead cells were excluded on the basis of atypical fluorescence. Data were further analyzed using Prism (GraphPad, La Jolla, CA, USA).
Cytokines (IL-2, IL-6, IL-12, IL-15 and TNF-α) and chemokines (CCL2, CCL7, CCL11, CXCL9 and CXCL10) levels in the supernatants of BAL were determined following i.n. infection of 5 x 103 dose in 100 μl 24 h p.i. using DuoSet ELISA kits (R&D Systems Inc.) and were carried out according to the manufacturer's instructions.
HeLa cells were either mock-infected or infected at 10 PFU/cell for 7 h. The cells were fractionated using Cell fractionation kit (Thermo Scientific) according to the manufacturer’s instruction.
HeLa cells were seeded on glass coverslips and were either mock-infected of infected at 10 PFU/cell or 2 PFU/cell in case of 16 h time point. At the indicated times, the cells were washed twice with PBS and fixed with 4% paraformaldehyde in PBS containing 250 mM HEPES. The cells were permeabilized with 0.1% triton X-100 followed by blocking with 10% FBS in PBS (blocking buffer) for 0.5 h. Coverslips were incubated with primary antibodies for 1 h in a moist chamber followed by three 10 min washes with 10% FBS. Coverslips were incubated with secondary antibody (Alexa Fluor 488 Goat Anti-Rabbit IgG (H+L), Alexa Fluor 546 donkey Anti-Mouse IgG (H+L) 1:750 diluted in blocking buffer) for 30 min in a moist chamber followed by three 5 min washes with 10% FBS and PBS only. The coverslips were washed with water and mounted in Mowiol 4–88 containing DAPI. Coverslips were allowed to set and stored at 4°C. Cells were visualized by Axio observer Z1 confocal microscope (Zeiss) with a 63x oil objective.
Reporter gene assays was performed in HEK 293T cells in 96-well dishes as described [45]. Cells were transfected in triplicate with 60 ng of firefly reporter plasmid (NF-κB, ISG 56.1 or ISRE), 10 ng of TK renilla (as an internal control) and 100 ng of expression plasmid or empty vector control using TransIT-LT1 according to the manufacturer’s instruction. The following day cells were stimulated; (i) with 75 ng of TNF-α for 7 h (NF-κB Luc) or (ii) transfected with 200 ng/well of poly I:C for 24 h (ISG56.1 Luc) using lipofectamine, or (iii) with 100 U/ml of IFN-α for 7 h (ISRE Luc). Cells were lysed using passive lysis buffer (Promega) and firefly luciferase activity was normalized to the renilla luciferase activity, and these data were further normalized to the un-stimulated controls of each test plasmid.
A549 cells were transfected in triplicate with GFP.FLAG, B14.FLAG, C6.TAP and 169 using Lipofectamine LTX Plus (Life Technologies). The following day cells were stimulated with 50 ng/ml of TNF-α for 7 h. RNA was extracted using RNeasy Mini Kit (Qiagen) and converted to cDNA using SuperScript reverse transcriptase. ICAM-1, IL-6 and NF-κBia mRNA were quantified in comparison to hypoxanthine-guanine phosphoribosyltransferase (HPRT) using SYBR green master mix.
HeLa or HEK 293T cells were transfected with indicated plasmids for 24 h. RNA was extracted using RNeasy Mini Kit (Qiagen) and converted to cDNA using SuperScript reverse transcriptase. GFP, luciferase or 169 mRNA were quantified and compared to HPRT or glyceraldehyde 3-phosphate dehydrogenase (GAPDH).
For analysis of cytoplasmic and nuclear mRNA, HEK 293T cells were transfected with empty vector control, A49.TAP and 169 together with NEMO-Luc. After 4 h the cells were treated with CHX (1 μg/ml) for 16 h. Cells were lysed in RLN buffer (50 mM Tris HCl pH 8.0, 140 mM NaCl, 1.5 mM MgCl2, 0.5% (v/v) Nonidet P-40, 1 mM DTT, 500 U/ml RNAse out), scraped and incubated for 5 min on ice. Nuclei were sedimented by centrifugation at 1000 g for 3 min. Supernatant (cytoplasmic fraction) was taken and mRNA was extracted according to the manufacturer’s instruction (Qiagen). RLT buffer was added to the pellet (nuclear fraction) and forced through a 25G needle ten times. Further steps followed the manufacturer’s instructions (Qiagen). cDNA was prepared using SuperScript reverse transcriptase. Luc-NEMO, HPRT and TATA box binding protein mRNA were quantified in comparison to (GAPDH) using SYBR green master mix.
HEK 293T cells were transfected in triplicate with GFP.FLAG, B14.FLAG, C6.TAP, 169 and Δ12 A49.TAP. The following day cells were either mock-infected or infected with SeV for 24 h. The amount of CXCL10 in the supernatant was determined using human CXCL10 Quantikine ELISA Kit (R&D Systems). The results were analyzed using nonlinear standard curves for ELISA (GraphPad PRISM).
HEK 293 Trex (Invitrogen) empty cells were transfected with 169 (pcDNA4 TO) using TransIT-LT1. Transfected cells were selected in the presence of zeocin and were serially diluted to obtain individual clones. Expression of protein 169 within these clones were analyzed by immunoblotting and immunofluorescence. In the chosen clone at least 90% of cells were expressing protein 169.
HEK 293 Trex 169 or C6.TAP [50] cells were treated with 1 μg/ml of DOXY for the indicated times to express protein 169 or C6. Cells were treated with 5 μg/ml of puromycin for 25 min and harvested for analysis by immunoblotting [49]. HeLa cells or BSC-1 cells were mock-infected or infected with VACVs at 5 PFU/cell for the indicated times. Cells were treated with puromycin as described above.
HEK 293 Trex 169 or C6.TAP [50] cell were uninduced or induced with 1 μg/ml of DOXY for 16 h. Thirty min prior to harvesting, the cells were treated with CHX (1 μg/ml). Cells were washed and lysed in lysis buffer supplemented with protease inhibitors (cOmplete, Mini, EDTA-free, Roche, 1 tablet in 10 ml of lysis buffer) (20 mM Tris HCl pH 7.5, 100 mM KCl, 5 mM MgCl2, 1 mM CHX, 1 mM DTT, 0.1 mM EDTA) with DNAse I and NP-40 (0.03%) followed by trituration with a 25G needle. Cleared (19,000 g for 5 min at 4°C) cytoplasmic lysates were layered on top of sucrose density gradient (10–50% sucrose in lysis buffer) prepared by a Gradient Master (Biocomp) and resolved by centrifugation at 200,000 g for 90 min at 4°C. Absorbance (254 nm) composition within the gradient was measured during fractionation at 4°C using an Isco fractionator. Proteins from these fractions were extracted using methanol-chloroform extraction and subjected to immunoblotting analysis. Polysome profiling in higher salt condition was carried out with HEK 293 Trex 169 as described above except that the lysis buffer and sucrose density gradient contained 400 mM KCl.
Statistical analysis was performed using Student’s two tail t-test unless otherwise stated.
|
10.1371/journal.pcbi.1003438 | Robustness of DNA Repair through Collective Rate Control | DNA repair and other chromatin-associated processes are carried out by enzymatic macromolecular complexes that assemble at specific sites on the chromatin fiber. How the rate of these molecular machineries is regulated by their constituent parts is poorly understood. Here we quantify nucleotide-excision DNA repair in mammalian cells and find that, despite the pathways' molecular complexity, repair effectively obeys slow first-order kinetics. Theoretical analysis and data-based modeling indicate that these kinetics are not due to a singular rate-limiting step. Rather, first-order kinetics emerge from the interplay of rapidly and reversibly assembling repair proteins, stochastically distributing DNA lesion repair over a broad time period. Based on this mechanism, the model predicts that the repair proteins collectively control the repair rate. Exploiting natural cell-to-cell variability, we corroborate this prediction for the lesion-recognition factor XPC and the downstream factor XPA. Our findings provide a rationale for the emergence of slow time scales in chromatin-associated processes from fast molecular steps and suggest that collective rate control might be a widespread mode of robust regulation in DNA repair and transcription.
| The nucleotide-excision repair pathway removes mutagen-inflicted DNA lesions from the genome. Repair proteins recognize DNA lesions and form multi-protein complexes that catalyze the excision of the lesion and the re-synthesis of the excised part. Imaging the dynamics of fluorescently labeled repair proteins in living human cells has revealed that all factors continuously and rapidly exchange at repair sites. We asked how this dynamic mode of protein-complex assembly shapes the repair process. Measuring repair DNA synthesis in intact cells, we obtained a surprisingly simple result. Over the entire process, the rate is proportional to the amount of DNA lesions, where the proportionality factor is a single ‘slow’ rate constant. Such kinetic behavior is often regarded as evidence for a rate-limiting step, but we show here that it is an emergent property of the dynamic interplay of many repair proteins. As a consequence, the rate of DNA repair is a systems property that is controlled collectively by the expression levels of all repair factors. Given that transcription in living cells has similar dynamic features – rapidly exchanging components of the transcription machinery and slow bursts of mRNA synthesis – collective rate control might be a general property of chromatin-associated molecular machines.
| Chromatin-associated processes, including transcription, replication, DNA repair and epigenetic control, are executed by macromolecular complexes that assemble at specific sites on the chromatin fiber. The stepwise, cooperative ‘recruitment’ of components into stable complexes has, until recently, been the prevailing model for the assembly of these macromolecular machineries. In contrast to this notion, imaging of transcription and DNA repair in living cells has shown that the constituents of the respective complexes dissociate and rebind at the seconds-to-minutes time scale at specific sites of transcription initiation and DNA damage [1], [2] and that RNA polymerase clustering in transcription factories is dynamic on a similar time scale [3]. Similar findings have been made for replication, with the exception of stably bound Mcm proteins which mark licensed replication origins [4], [5]. Based on these data, an alternative model for the assembly of transcription-initiation and DNA repair complexes has been proposed: functional complexes form from randomly diffusing and rapidly exchanging components [6]–[8]. The kinetic and functional implications of this highly dynamic mode of macromolecular complex formation on DNA remain poorly understood.
Combining experiments on living cells with mathematical modeling, several studies have begun to unravel the relation between assembly dynamics and functionality of chromatin-associated machineries. Gorski et al. [9] demonstrated that moderate dwell-time changes of RNA polymerase components at transcription initiation regulate transcription rate. Voss et al. [10] showed that a rapidly exchanging transcription factor facilitates the subsequent binding of a second transcription factor by enzymatically increasing chromatin accessibility. In this “assisted-loading” mechanism, the first transcription factor enhances the binding of the second one via the modification of the chromatin template, thus obviating the need for simultaneous and cooperative binding. A comprehensive live-cell imaging study of nucleotide-excision repair (NER) in mammalian cells predicted that the rapid exchange of NER factors allows for high specificity of the pathway through kinetic proofreading [8].
A common theme linking these studies is the functional coupling between protein binding and modifications of chromatin by the bound proteins (including both DNA and protein modifications). This interplay suggests a cyclic principle (Fig. 1A): Appropriately assembled multi-protein complexes drive enzymatic reactions that modify the chromatin template, in turn changing protein affinities for the assembly of the next enzymatic complex. Multiple ‘recruitment-reaction’ cycles of this kind can integrate random and transient diffusion-dependent protein binding steps into an ordered sequence of regulatory events at a specific chromatin site [11]. Based on in vivo experiments, a detailed mathematical model of this type has been developed for NER in mammalian cells [8]. NER efficiently recognizes and removes a wide variety of DNA lesions from the damaged DNA strand and resynthesizes the excised part by using the intact strand as a template [2]. Live-cell imaging has shown that all core proteins for global-genome NER exchange rapidly at damaged DNA sites in the cell nucleus as repair takes place [8], [12]–[15]. The resulting model describes the NER pathway as the sequential conversion of DNA intermediates – damage recognition followed by DNA unwinding, lesion excision, re-synthesis and ligation of the excised strand – with each conversion step catalysed by a specific multi-protein complex that assembles reversibly at the site of repair when needed (Fig. 1B).
The NER model by Luijsterburg et al. [8] was able to integrate a large data set for all core NER proteins because a standard protocol was employed for each of them, consisting of inflicting DNA lesions locally in the cell nucleus by a pulse of UV light and subsequent monitoring of the repair factor dynamics by live-cell imaging [13], [16]. While the model presents a quantitative description of the in vivo data, the data themselves were not sufficient to uniquely identify all kinetic parameters (on-rate and off-rate constants as well as enzymatic rate constants). Therefore quantitative predictions with the model on the functioning of NER in intact cells were limited. This limitation of predictive power is due the fact that the model is based on measured assembly rates and dwell times of the repair factors whereas the repair of the DNA lesions has not been measured with comparable time resolution (similar limitations may apply to imaging-based models of transcription dynamics [9], [17], [18]).
Here we quantify the relation between the protein dynamics measured previously and the resulting DNA repair by measuring DNA repair synthesis in a time-resolved manner in individual living cells. We find that lesion repair by NER is effectively a slow first-order reaction and show this to be an emergent property of the fast dynamics of the repair proteins. The new data allow the development of an identifiable and hence predictive NER model. A conspicuous quantitative prediction of the model is that NER does not have a singular rate-limiting step but, instead, that the concentrations of all core repair factors have small or moderate effect on – and thus collectively control – the repair rate. Harnessing natural cell-to-cell variability of protein expression levels, we provide experimental support for this prediction. Our study introduces a quantitative framework for dissecting the dynamics of macromolecular machines that act on the genome.
To quantify the progression of DNA repair and expand our dataset on the dynamics of NER in intact mammalian cells, we determined the kinetics of repair DNA synthesis after locally inflicting UV-damage in cell nuclei [16]. Although DNA lesion removal is often employed as a measure to track repair, it only captures the systems behavior of the pre-incision steps and does not give information about the post-incision processes. To determine the repair kinetics accurately, we induced DNA damage by locally UV-irradiating cells that express an eGFP-tagged version of the lesion-recognition factor XPC in a cell line in which endogenous functional XPC is absent (XP–C [19]) and measured the incorporation of the uridine-analogue, 5-ethynyl-2′-deoxyuridine (EdU), into DNA that was fluorescently labeled after cell fixation [20]. Damaged nuclear areas accumulated XPC-eGFP, and EdU was exclusively present in these areas, consistent with EdU being incorporated through repair DNA synthesis (Fig. 2A). Cells in S-phase, which showed a strong EdU signal throughout the nucleus due to DNA replication, were excluded from our analyses.
To establish the time course of DNA repair synthesis we measured the extent of EdU incorporation at different time points after inflicting UV damage and averaged over multiple cells (Fig. 2B). Of the two main types of UV induced lesions, the 6-4 pyrimidine-pyrimidone photoproducts (6-4 PP) are repaired within a few hours, whereas repair of the cyclobutane pyrimidine dimers (CPD) takes much longer [8], [21], [22]. Indeed, we found that EdU incorporation essentially stops after ∼4 hours, which coincides with the disappearance of the 6-4PPs [8]. Labeling with EdU at different times after UV irradiation indeed showed a continuously declining rate of incorporation and hence of the rate of repair DNA synthesis (Supplementary Fig. S1). This finding is consistent with the progressive disappearance of DNA lesions and demonstrates that EdU availability was not limiting. Moreover, the incorporation of EdU followed the accumulation of PCNA as measured by Luijsterburg et al. [8] in living cells (Fig. 2C). PCNA is required for DNA polymerase function and remains bound to repaired DNA [8], [23], [24]. Taken together, these data establish EdU incorporation as a direct and quantitative measure of DNA repair synthesis in locally damaged cells.
The time course of repair synthesis measured through EdU incorporation is accurately fitted by a single exponential function:(1)where EdU(t) and EdUmax denote the amount of repaired DNA at time t and at saturation, respectively (Fig. 2D). The time constant is λ = 0.58 (±0.07) h−1. This result implies that, despite its molecular complexity, the repair of 6-4PP by NER is effectively a slow first-order reaction with a half-time of 1.2 hours.
The observed slow first-order kinetics of repair DNA synthesis could be due to a rate-limiting step with a long half-time of the order of an hour. However, there is no likely candidate among the molecular steps of NER for such a slow process [2]. Therefore, we asked whether the first-order characteristics of NER could arise by another mechanism. To gain analytical insight, we first considered a simplified model of repair, with N factors assembling into a repair complex and carrying out the repair reaction (Fig. 3A). Under appropriate assumptions, the average time after which a lesion is repaired can be computed as(2)where k denotes the pseudo-first order association rate constant of NER factors, l the dissociation rate constant of NER factors and ρ the rate of the repair reaction (Supplementary Text S1). The coefficients Ai and Bi depend on whether the repair complex is assembled in a sequential or in a random order. The assumption of common association and dissociation rate constants for all NER factors is motivated by our finding that the core NER factors bind reversibly to a DNA lesion with typical dwell times of the order of ∼1 min [8], taking k = l = 1 min−1 as reference values. Random and sequential assembly of the repair complex give rise to similar average repair times for ten or less assembling components (Fig. 3A); for comparison, global-genome NER in living cells involves 10 core components (cf. Fig. 1B). Remarkably, the repair time is in the experimentally observed hour range for a realistic number of factors, despite the fact that all individual processes (factor binding, dissociation and catalysis of the repair reaction) are much faster. Moreover, with these parameters the model yields a single-exponential time course for the repair of the DNA lesions (Fig. 3B). Further analysis shows that approximately single-exponential kinetics result when the repair factors bind reversibly, whereas high-affinity, near-irreversible binding would lead to a sigmoid time course (Fig. 3C; Supplementary Text S1).
Thus the simplified model of NER indicates that the assembly of a repair complex from rapidly exchanging components naturally generates slow, first-order kinetics of the repair process. This finding implies that the experimentally observed kinetics are an emergent phenomenon of the interplay of many fast components that stochastically distribute the repair of DNA lesions over a broad time period.
To further examine this hypothesis we derived a realistic model of NER based on a large set of quantitative experimental data. The model follows the concept developed by Luijsterburg et al. [8], assuming independent binding of the NER components to the DNA lesion that, transiently forming the appropriate complexes, catalyse the sequence of repair steps. The experimental data consist of the time course of repair DNA synthesis established here and the binding and dissociation rate measurements for all core NER factors (XPC, TFIIH, XPA, XPG, ERCC1-XPF and RPA), as well as PCNA involved in repair synthesis [8] (Fig. 4A). No live-cell imaging data are available for replicative DNA polymerases due to the lack of functional fluorescently tagged constructs; they are assumed to behave in a similar manner as PCNA (which loads the DNA polymerases). Using a standard maximum-likelihood approach for parameter fitting together with the profile-likelihood method to establish parameter bounds [25], we found that these data identify a realistic kinetic model of NER. The present model consists of the equations first derived in [8], with few simplifications to reduce the number of kinetic parameters (in particular, the incision of the DNA lesion has been assumed practically instantaneous once the pre-incision complex has been completely assembled; Supplementary Text S1). The catalytic rates are fast (except for the re-chromatinization rate δ). This is seen by the existence of lower bounds on the rate constants of the order of 1 s−1 (Fig. 4B). All protein binding and dissociation rate constants are identified within narrow bounds (Fig. 4B) as seen by computing the profile likelihoods (Supplementary Fig. S2). For the numerical values of the parameters see Supplementary Tables S1 and S2.
The model quantitatively simulates the net accumulation of NER factors and the incorporation of EdU (Fig. 4C, D), their behavior in FLIP-based protein dissociation measurements (Fig. 4E) and the time course of repair DNA synthesis (Fig. 4F). The concentrations of NER factors and DNA repair intermediates are given as local concentrations in the damaged area, based on our estimate that the volume associated with the damaged chromatin makes up on average 10% of the total nuclear chromatin per UV-damaged spot (note that in Luijsterburg et al. [8], we used a different normalization by dividing the number of bound proteins with the entire nuclear volume; this resulted in the average nuclear concentrations of bound factors that are tenfold smaller than the –more intuitive – concentrations in local damage given here).
The dissociation constants Kd, which account for the quantitative accumulation of the NER factors at local damage, fall into a rather uniform, biologically realistic range between ∼100 nM (for XPA) and ∼1 µM (for RPA) (Table 1). A notable exception is the initial XPC binding to damaged DNA, which shows a comparatively low affinity (Kd∼9 µM). These quantities are consistent with previous work [8]. However, due to the direct observation of the repair DNA synthesis kinetics, we have now been able to identify practically all model parameters. Importantly, the model implies that the excision of the lesion is followed by repair synthesis without much delay. Therefore, there is only little accumulation of the lesion-excised DNA repair intermediate (Fig. 4F).
The narrow confidence bounds on the parameters of the model allow us to make valid computational predictions. We asked whether the kinetics of repair DNA synthesis in this realistic model are indeed controlled by multiple NER factors, as the ‘cartoon’ model (Fig. 3A) suggests. To this end, we evaluated the response coefficients(3)giving the relative change of the repair rate v as a function of the relative change in the concentration of the ith repair factor Ci (i = XPC, TFIIH, …) [26]. All response coefficients have narrow confidence bounds as evaluated by the prediction profile likelihood method [25], implying a well-defined prediction (Supplementary Fig. S3A). The response coefficients are uniformly small (∼0.3 and below, Fig. 5A), and hence there is no singular rate-limiting component distinguished by a high response coefficient. A very similar result is obtained for the control of the dual incision rate (Supplementary Fig. S3B). We also computed the response of the repair rate to large variations in protein concentrations, finding that the response coefficients correctly predict the impact of the individual repair factors (Fig. 5B). The linear approximation (on which the response coefficients are based) yields a reasonable description for about two-fold concentration decreases or increases (corresponding to a knockdown or overexpression experiment), while for very large decreases the repair rate drops eventually to zero (corresponding essentially to a gene knockout).
In summary, these findings imply that the individual repair factors have similar, small control on the repair rate. Thus the model predicts that the repair rate is collectively controlled by all NER components and, therefore, robust against variations in the concentration of any individual factor.
To determine the effect of NER factors on repair synthesis, we focused on the lesion recognition factor XPC and the component of the pre-incision complex XPA. We observed considerable variability in the endogenous expression levels of both proteins in individual cells (Fig. 6A), with coefficients of variations (CV) of ∼0.3. However, the theoretical analysis of rate control predicted that across this range of natural variability (roughly from about one half of the average expression level to twice the average; cf. Fig. 5B), the repair rate of 6-4PP should be affected only moderately. We asked whether rate control could be quantified experimentally by exploiting the natural expression variability of the NER factors.
To this end, we first examined the accuracy of XPA and XPC antibody staining. In human XP–C and XP–A fibroblasts stably expressing XPC-eGFP and eGFP-XPA, respectively [19],[27], we observed a high correlation between the eGFP signal and immunofluorescent labeling (Fig. 6B). Fitting an error ellipse to these data (Supplementary Text S1), we estimated a relative measurement error of antibody labeling of 11%, showing that the technique is suitable for quantitation of nuclear XPC and XPA concentrations. We then examined whether the nuclear concentration of a repair factor is correlated with its accumulation at UV-induced DNA lesions and with the rate of repair DNA synthesis. Thirty minutes after local UV irradiation we observed a linear relationship between the nuclear concentration of XPC and the concentration of accumulated XPC in the damaged spot (Fig. 6C). This result demonstrates that the DNA lesions are not saturated with XPC, so that a higher free XPC concentration in the nucleus could accelerate the repair rate. The same holds true for XPA (Fig. 6D) and has previously been shown for the endonuclease XPG [8]. Therefore, we asked next whether increased accumulation of the repair factors accelerated repair. The measurement of the concentration of XPC or XPA in the nucleus simultaneously with EdU incorporation showed a significant positive correlation (Fig. 6E and F, respectively). However, the dependence of EdU incorporation on the concentration of either factor was weak, as seen by the small slopes of the regression lines. Remarkably, the slopes were in the range predicted by the response coefficients of XPC and XPA for the repair rate. These data suggest that the repair rate is largely robust against natural variability of XPC and XPA concentration, supporting the prediction of the sensitivity analysis of the model (cf. Fig. 5).
The finding of robustness of repair rate is further strengthened by the observation that in XP–C cells complemented with XPC-eGFP, the expression levels of the fluorescent protein expressed from the stably integrated transgene varied more strongly than the endogenous XPC protein levels in HeLa cells (CV∼1, with peak levels in a sizeable number of cells reaching 3–4 times the average), yet EdU incorporation after UV damage followed very similar dynamics in both cell types (Fig. 7).
The considerable scatter in Figures 6E and F indicates that there are further sources of cell-to-cell heterogeneity. We quantified the amount of inflicted DNA lesions (6-4PP) after UV damage and incorporated EdU after repair of 6-4PPs is completed. Both quantities show variability between individual cells (Fig. 8A,B). That the initial distribution of 6-4PP and the final distribution of EdU incorporation are of very similar width shows the consistency of the two measurements. These data indicate that the amount of inflicted DNA lesions contributes to the observed cell-to-cell variability in repair rate.
How do these two sources of cell-to-cell variability that we identified experimentally – concentration differences in NER factors and number of inflicted DNA lesions – control the rate of repair? To analyse this quantitatively we first simulated the simultaneous effect of both sources with the model. To simulate an individual cell we drew randomly an amount of 6-4PP from the measured distribution (cf. Fig. 8A) and equipped the cells with random amounts of all NER factors according to lognormal distributions with typical width (CVi = 0.25; cf. Fig. 6B) and the experimentally determined mean values [8]. Repeating this for several hundred cells (as in the experiments) we found that the simulated relation between nuclear XPC concentration and XPC accumulated in local damage agreed remarkably well with the measured data (Fig. 9A); the same held true for the relation between nuclear XPC and extent of DNA repair synthesis at 30 min after irradiation (Fig. 9B). Quantitative agreement was also found for the experimental data and the simulated relations between nuclear XPA concentration and XPA accumulated in local damage (Fig. 9C), as well as extent of DNA repair synthesis at one hour after irradiation (Fig. 9D).
These findings indicate that the variable concentrations of NER factors and inflicted DNA lesions explain the observed cell-to-cell variability in repair DNA synthesis. Therefore, we asked whether the individual contributions of the concentration distributions of the repair factors to the distribution of the repair rate ‘sum up’ according to the response coefficients predicted by the model. According to the law of propagation of uncertainty we have approximately(4)where CVv, CVi and CVL denote the coefficients of variation of the distributions of repair rate (v), repair factors (i = 1,2,…) and initial amount of lesion (index L) (Supplementary Text S1). Assuming that all CVi = 0.25, as measured for XPC and XPA, and CVL = 0.32, we find with the response coefficients of the model CVv = 0.34. This value compares well with the experimentally measured CV of the distribution of EdU incorporation at 30 min, 0.37, where the difference might be due to the fact that a few further factors (DNA polymerases and ligases) are involved in repair but have not been accounted for explicitly in the model.
To dissect the contribution of the individual factors to the overall variability, we computed the effect of heterogeneity in only one factor on repair synthesis. The initial distribution of 6-4PPs has the strongest impact but, clearly, does not account for the entire observed variability, particularly at earlier time points (Fig. 9E, observed: blue triangles). As expected from the size of the response coefficients, the cell-to-cell variability in the expression levels of the individual NER factors have all comparable and rather small effect (Fig. 9E, model individual factors: dots). Adding up the contributions from all factors closes the ‘variability gap’ left between the distribution of 6-4PP and repair synthesis (Fig. 9E, model all factors combined: red triangles). Indeed, the computed time evolution of the distribution of EdU incorporation largely matches the measured one (Fig. 9F).
To summarize, we uncovered two sources of cell-to-cell variability in the rate of NER: (i) the amount of inflicted DNA lesions and (ii) the expression of NER factors. The quantitative impact of these two parameters on the cell-to-cell variability of the repair rate is consistent with the robustness of the repair rate against concentration fluctuations in individual repair factors.
In this paper we have quantified the rate of DNA repair by the nucleotide-excision repair pathway in relation to the behavior of the individual NER components. The experimentally observed slow, apparent first-order kinetics of repair agree with the prediction from mathematical modeling of NER. Importantly, the model indicates that these kinetics are not due to a rate-limiting step in the pathway but, rather, are an emergent phenomenon of the rapid interplay of many transiently interacting components. An important functional consequence of this kinetic design is that the control of the repair rate is shared by all repair factors. This collective rate control implies that the rate is robust against concentration variations in individual repair proteins. Exploiting the natural variability of protein expression in mammalian cells, we have provided experimental support for this model.
Early measurements of repair rate by radionucleotide incorporation displayed two components of NER, one being completed in several hours and the other lasting much longer [28]. As indicated by the removal kinetics of the two main types of DNA lesions, 6-4PP and CPD, the first component represents the repair of 6-4PPs, which is practically complete within four hours, and the second component is CPD repair, which is still negligible after eight hours [8]. Here we monitored the first component, 6-4PP repair, via EdU incorporation and showed that it follows first order-kinetics with half-time of 1.2 hours. This long time scale (hours) is in sharp contrast with the results of imaging the dynamics of NER proteins in living cells. These latter studies showed that all protein-substrate interactions are transient and occur on a time scale of seconds to a few minutes [8], [12]–[15], [19], [27], [29]. The individual catalytic steps (DNA unwinding, lesion excision and repair synthesis) are also thought to be fast. Indeed, a recent direct measurement has shown that DNA repair synthesis and ligation in bacteria take seconds [30]. The present work proposes a resolution to the conundrum that the overall rate of lesion repair is nevertheless slow by showing how the repair kinetics emerge as a systems property from the interplay of the underlying molecular processes.
Our mathematical analysis shows that the reversibility in the binding of multiple proteins to the DNA lesion, and subsequent repair intermediates, is the key property that makes DNA repair an apparent first-order process. Intuitively, the broad distribution of repair times results from the fact that protein binding and dissociation events will be iterated a variable number of times before an active repair complex is formed. This is in contrast to a sequence of irreversible binding steps, which will create sigmoidal kinetics with an even sharper delay with increasing numbers of assembling components. When comparing with the degree of reversibility, the order of component assembly (random, sequential or combination of the two extremes) is less relevant for whether the repair kinetics will be first-order (exponential) or sigmoidal. Fast exchange of individual factors is a ubiquitous property of chromatin-interacting proteins in transcription, chromatin remodeling and repair [1], [31]. It is therefore tempting to speculate that the slow time scales of transcriptional bursting in mammalian cells (tens of minutes to hours; [32], [33]) also arise through the reversible assembly of large macromolecular complexes.
High molecular off-rates have been suggested to be a general feature of self-organizing systems in the cell as they allow for efficient exploration of an assembly landscape and selection of a functional steady state [34]. In the case of chromatin-associated machineries; specificity and regulatability will both benefit from reversible binding of repair proteins or transcription factors [8]–[10], [35].
Based on comprehensive experimental data, we have here developed a predictive mechanistic model of NER. Progressing beyond previous models [8], [36], [37], this appears to be the first model of a DNA-associated process for which the model parameters could be identified (i.e., parameter values uniquely assigned with narrow confidence bounds) from the experimental data. Achieving this required a number of simplifications, including the neglect of protein-protein interactions and the explicit consideration of most but not all core NER factors, selecting those for which live-cell imaging is currently possible. Judged by the rigorous statistical criterion of identifiability, the current model provides a concrete picture of what can be said quantitatively about the regulation of NER. Adding further mechanistic detail will require appropriate quantitative measurements to evaluate their impact on systems properties such as repair rate, robustness against protein-expression noise and fidelity of lesion recognition. We expect that the general dynamic behavior of the model will prove robust with respect to the addition of further molecular detail and components because the dynamics are already produced in a qualitative manner by a simplified ‘cartoon’ model of repair (see Fig. 3). Our work might also provide a useful reference for mathematical models of transcription, which so far have been parameterized on the basis of FRAP data without studying parameter identifiability [9], [17], [18].
The NER model makes two interesting predictions on the function of the repair pathway in living cells. First, lesion excision and repair synthesis are tightly coupled, preventing undue accumulation of single-stranded DNA after excision of the lesions. In this kinetic aspect, in vivo NER appears to differ from in vitro studies that have found a delay before repair synthesis [38], [39]. Preventing the accumulation of single-stranded repair intermediates in the cells that are vulnerable to being cleaved is a plausible functional objective and, as shown here, is consistent with data obtained on intact cells. Second, there is no singular rate-limiting step in the repair pathway. Instead, all repair factors control the repair rate collectively, with the contribution of each individual factor being rather small. Thus the control of the DNA repair rate has similarity to the control of flux in metabolic pathways which has also frequently found to be shared by several enzymes [26]. It is important to note that rate control is a kinetic property that is not linked to the order at which the factors bind. Thus the recognition factor XPC that binds first to a DNA lesion and XPA, which binds much later, have very similar rate control. In particular, our results show that the functioning of the NER pathway is robust against the natural fluctuations of protein concentrations in the cell.
The model can also be used to rationalize the effect of larger perturbations in the concentrations of NER factors. In particular, strong reduction of any factor participating in damage recognition and excision should compromise function (e.g., repair rate) in a similar manner. Indeed, reduction in XPA or XPC concentration leads to decreased cell survival and/or decreased lesion removal [40]–[42]. Increased expression levels of ERCC1 and XPA have been correlated with increased resistance of certain tumors to cisplatin, which induces DNA interstrand crosslinks specifically removed by NER [40]–[44]. It would be interesting to investigate whether other NER factors are also elevated under these conditions, as a collective rate control mechanism would suggest, or whether XPA or ERCC1 have particularly high control (e.g., through low endogenous expression level) in the cell types and conditions studied.
The mechanism that produces collective rate control in the NER model is not specific to this repair pathway, but should equally apply to the assembly of other chromatin-associated molecular machineries. Indeed, a quantitative study of the interferon-β gene transcription has shown that the overexpression of five out of six transcription factors needed to activate transcription had a significant effect on the probability of transcription [45]. Very similar results have recently been obtained for the regulation of expression of the cytokine IL-2 [46]. These experimental findings support a collective process of rate control also for transcription.
XPC-deficient XP4PASv cells (XP–C cells) stably expressing XPC-eGFP [19] and XPA-deficient XP20SSv cells (XP–A cells) stably expressing eGFP-XPA [27] were maintained in a 1∶1 ratio of F10∶DMEM supplemented with 5% FBS (Gibco) medium, glutamine and penicillin/streptomycin. HeLa cells were grown in DMEM medium supplemented with 5% FBS, glutamine and penicillin/streptomycin. All cells were kept in an incubator at 37°C incubator containing 5% CO2 for cultivation and in an incubator without CO2 after UV-irradiation.
Local irradiation was performed as described by Moné et al. [16]. Briefly, cells were grown to confluence on uncoated 24 mm glass coverslips. Before UV-irradiation cells on coverslips were briefly incubated with pre-warmed microscopy medium (137 mM NaCl, 5.4 mM KCl, 1.8 mM CaCl2, 0.8 mM MgSO4, 20 mM D-glucose and 20 mM HEPES, pH 7.0) at 37°C. Immediately upon removal from the incubator cells were overlaid with a mask, containing pores with 5 µm diameter, minimizing the distance between mask and cells. Local irradiation was performed with a dose of 100 J/m2 of UV-C with a fluency of 3.85 W/m2, as measured at 254 nm with an SHD 240/W detector connected to an IL 1700 radiometer (International Light), unless stated differently. Cells were incubated in microscopy medium in an incubator without CO2 at 37°C as indicated in the text.
Prior to UV-irradiation cells on coverslips were briefly incubated in microscopy medium supplemented with 10 µM EdU (5-ethynyl-2′-deoxyuridine, Life Technologies). After irradiation, cells were incubated in medium in the continuous presence of 10 µM EdU. After incubation for the desired time cells were fixed in PBS containing 4% PFA (pH 7.0), permeabilized by incubation with PBS/0.5% Triton X-100 and processed and fluorescently labeled following the manufacturer's instructions (Invitrogen). Briefly, coverslips were incubated with a proprietary Life Technologies reaction mix leading to copper-catalyzed covalent addition of an AlexaFluor-555 fluorophore to incorporated EdU. Subsequently, coverslips were washed twice with PBS and, if necessary, processed for immunolabeling as described below. After processing coverslips were rinsed with sterile ultrapure water, and mounted with Vectashield (Vectorlabs) mounting reagent for microscopy.
Locally irradiated cells were fixed in PBS/4% PFA for 10 minutes, washed 3× with PBS and permeabilized with PBS/0.5% Triton X-100 for 10 minutes. Immediately before 6-4PP immunolabelling cells were incubated with 0.1 M HCl for 10 minutes. Cells were washed 3 times with PBS/3% BSA and incubated with Rabbit anti-XPC (X1129, Sigma-Aldrich) and/or Mouse anti-64PP (64M-2, Cosmo Bio Co, Tokyo, Japan) dissolved in PBS/3% BSA for one hour at room temperature in the dark. Antibody was removed by washing 3 times with PBS/3%BSA and coverslips were incubated with donkey anti-rabbit Cy5 or Cy3 (711-175-152/711-165-152, Jackson Immunoresearch), and/or donkey anti-mouse Alexa488 (715-545-150, Jackson Immunoresearch) and DAPI for one hour at room temperature in the dark. Coverslips were washed 3 times with PBS/3% BSA, rinsed with water and mounted on slides with Vectashield (Vectorlabs) mounting medium before observation. For XPA detection a combination of rabbit anti-XPA (X1254, Sigma-Aldrich) and donkey anti-rabbit Alexa647 (711-605-152, Jackson Immunoresearch) was used, and nuclei were counterstained with Draq5 (Biostatus LTD).
All microscopic analyses were performed on a LSM510 (Zeiss Inc.) microscope with a fully open pinhole, equipped with the following lasers with the indicated wavelengths; Argon ion1 (364 nm), Argon ion2 (488 nm), Helium-Neon1 (543 nm) and Helium-Neon2 (633 nm). Slides were observed with either a 63× Plan-Apochromat NA 1.40 NA or a 63× Plan-NeoFluar NA 1.25 lens. Images were recorded by Zeiss software in multi-track mode. All images were recorded with 12-bit dynamic range (bits per pixel). The photomultiplier settings were corrected for each channel to ensure image detection within linear detection range.
All image analyses was done by using ImageJ software (NIH, Bethesda, MD). For indirect immunofluorescence experiments, values were normalized per experiment before comparison and analyses. Nuclear concentrations were measured by thresholding nuclei manually or, when possible, automatically. The mean background outside the nuclei was subtracted to correct for background. XPC and XPA intensities on local damage were determined by manually thresholding local damages; these intensities were background-corrected by subtracting the intensity in a similarly-sized undamaged portion of the nucleus. EdU incorporation was determined by manual thresholding of local damages and intensity measurements. Background correction was performed by subtracting the intensity of a similarly-sized area in the same nucleus. Raw values from different experiments were pooled and plotted directly (distributions) or scaled and plotted. The amount of inflicted DNA lesions was determined by measuring total intensity on local damage after antibody staining. Values were background-corrected by subtracting the mean background intensity outside local damage. Data was scaled per experiment (n = 5), pooled, and plotted. Different antibody concentrations were used to determine antibody linearity; similar CV values were obtained with all concentrations used. Figures and statistical analysis were done in MatLab (Mathworks, August 2010).
The mathematical models (analytical model and detailed, data-driven model) are based on balance equations for the abundance of protein complexes assembling on the different states of the DNA template (ordinary differential equations), as specified in the Supplementary Text S1. Parameter fitting and simulations were done with Matlab code supplied with this paper. Details on error estimation in Fig. 6B can be found in the Supplementary Text S1.
|
10.1371/journal.ppat.1000541 | Interplay in the Selection of Fluoroquinolone Resistance and Bacterial Fitness | Fluoroquinolones are antibacterial drugs that inhibit DNA Gyrase and Topoisomerase IV. These essential enzymes facilitate chromosome replication and RNA transcription by regulating chromosome supercoiling. High-level resistance to fluoroquinolones in E. coli requires the accumulation of multiple mutations, including those that alter target genes and genes regulating drug efflux. Previous studies have shown some drug-resistance mutations reduce bacterial fitness, leading to the selection of fitness-compensatory mutations. The impact of fluoroquinolone-resistance on bacterial fitness was analyzed in constructed isogenic strains carrying up to 5 resistance mutations. Some mutations significantly decreased bacterial fitness both in vitro and in vivo. We identified low-fitness triple-mutants where the acquisition of a fourth resistance mutation significantly increased fitness in vitro and in vivo while at the same time dramatically decreasing drug susceptibility. The largest effect occurred with the addition of a parC mutation (Topoisomerase IV) to a low-fitness strain carrying resistance mutations in gyrA (DNA Gyrase) and marR (drug efflux regulation). Increased fitness was accompanied by a significant change in the level of gyrA promoter activity as measured in an assay of DNA supercoiling. In selection and competition experiments made in the absence of drug, parC mutants that improved fitness and reduced susceptibility were selected. These data suggest that natural selection for improved growth in bacteria with low-level resistance to fluoroquinolones could in some cases select for further reductions in drug susceptibility. Thus, increased resistance to fluoroquinolones could be selected even in the absence of further exposure to the drug.
| The increasing frequency of human pathogens resistant to important classes of antibiotics poses a serious and growing challenge for medicine and society. We need improved strategies to reduce the rate of resistance development, for established and novel drugs, based on knowledge of the factors that drive the increase in resistance. Resistance to fluoroquinolones in most bacteria develops via a series of sequential genetic changes affecting several different genes. These are selected and enriched in bacterial populations by exposure to the drug. Relevant factors driving this increase include overuse, and inappropriate use, of these drugs. In this paper we show that mutant bacteria with low-level resistance (not itself a problem to treat with standard drug doses) can evolve by natural selection (for improved growth rate) to acquire mutations that dramatically increase their level of drug resistance. This means that we may need to consider how to reduce inappropriate drug use that can enrich for bacteria with low-levels of resistance, because at that stage some of the mutant bacteria in the population may continue to evolve higher level resistance even in the absence of any further drug exposure.
| Fluoroquinolones are potent antibacterial drugs [1] that bind to bacterial type II topoisomerases (DNA gyrase and topoisomerase IV) when they are in complex with DNA. The drugs inhibit chromosome re-ligation after enzyme-mediated cleavage [2]. Fluoroquinolones are effective against many bacteria including invasive E. coli, but resistance is increasing with 28 of 29 countries in Europe reporting a significant rise between 2001–2007 [3]. The rapid increase is surprising because clinically relevant levels of resistance in E. coli require multiple genetic changes, including mutations altering topoisomerases and up-regulating drug efflux [4], changes that are associated with reduced bacterial fitness in vitro and in vivo [5]–[7]. Inappropriate use of fluoroquinolones, or co-selection of resistant bacteria with the use of other antimicrobial drugs [8] may be factors driving the increase in resistance but these may not be the sole causes [9]. For example, resistant isolates of E. coli increased from ∼7–19% between 2001 and 2007 in the UK [3], while outpatient fluoroquinolone use remained unchanged from 1997–2003 [9]. To develop an effective strategy to restrict the increase in resistance frequency will require that we have a full understanding of the factors driving the increase. The aim of this paper was to investigate whether selection for improved fitness in bacteria might itself be a factor promoting increased resistance.
The fitness costs of drug resistance can be reduced by selection of low-cost mutations or by the accumulation of secondary fitness-compensating mutations that do not reduce resistance. During experimental evolution of clinical isolates of E. coli for decreased susceptibility to fluoroquinolones most lineages (16/18) suffered reduced growth competitiveness after only two or three selection steps [5]. However, lineages selected for further decreases in susceptibility were occasionally associated with a relative restoration of fitness [5]. A similar reversal was noted in some constructed strains of Streptococcus pneumoniae carrying one or two resistance mutations [10]. These data suggested that some resistance mutations might be selected because they decrease susceptibility to the drug and simultaneously reduce the fitness costs associated with existing resistance mutations. No cause for the phenomenon has been demonstrated, and it's relevance to bacterial fitness in vivo in not clear. To examine the phenomenon we constructed isogenic strains carrying various combinations of five resistance mutations found commonly in fluoroquinolone-resistant clinical E. coli, and measured their drug-susceptibility and fitness. The relationships between the number of resistance mutations and bacterial fitness were complex and the addition of a resistance mutation was shown in some cases to improve bacterial fitness. These findings have implications for the evolution of fluoroquinolone resistance in the absence of antibiotic exposure.
The E. coli urinary tract infection isolate C1186 [4] is highly-resistant to fluoroquinolones (MIC for ciprofloxacin ≥32 µg/ml) and carries resistance mutations altering topoisomerases (gyrA Ser83→Leu, Asp87→Asn; parC Ser80→Ile), and up-regulating drug efflux (marOR small deletion, and amino acid substitution; acrR IS1 insertion). These 5 mutations are typical of highly resistant clinical isolates [4]. C1186 has a growth rate similar to a laboratory wild-type. Thus, these resistance mutations may individually be low-cost, as found for some rifampicin-resistant patient isolates [11],[12] or the strain may carry additional fitness-compensatory mutations [7]. A third possibility is that some resistance mutations reduce existing fitness costs while simultaneously decreasing susceptibility to the drug. This last possibility is highly relevant to the multi-step nature of fluoroquinolone-resistance development. We constructed 28 isogenic derivatives of the wild-type MG1655, each mimicking in part the complex resistance genotype of C1186 (Table 1). Mutations were initially isolated spontaneously (gyrA, parC) or constructed by λ-red recombineering (ΔmarR, ΔacrR) and then separately introduced into MG1655 by P1 transduction. The gyrA and parC mutations were transduced by selection for a linked genetic marker (introduced using λ-red recombineering, see Materials and Methods) and at least 20 transductants of each cross were subsequently screened by phenotype (MIC) and DNA sequencing for the linked mutation. In every case only two phenotypic and genotypic classes were found, showing that the gyrA and parC mutations were not associated with other mutations.
MIC for ciprofloxacin was measured for each strain (Table 1). The margin of error of MIC values is ±1 half-doubling step. Accordingly, any change that is at least 2-fold is significant. Single mutations in gyrA, S83L and D87N, increased MIC 24-fold and 16-fold respectively. Knockout mutations in marR and acrR increased MIC only 2–3-fold, while the substitution S80I in parC had no effect on MIC. Double mutation combinations had MIC's that were 8–64-fold wild-type level, with the combination ΔmarR+ΔacrR having the smallest increase. Certain combinations with parC were not tested because in E. coli parC mutations only selected after the prior occurrence of a mutation in gyrA. Triple mutation combinations have MIC's that were 31–2000-fold wild-type level. Most strains with three resistance mutations (5/9), and all strains with 4 or 5 mutations had MIC's above the 1 µg/ml breakpoint that defines clinical resistance in Europe [13] equivalent to 64-fold wild-type MIC in these strains. On average there was a positive correlation between the number of resistance mutations carried by a strain and the MIC for ciprofloxacin (Figure 1).
The 28 mutant strains were tested in growth competitions against wild-type to measure their Malthusian fitness [14],[15] as a function of the resistance mutations they carried (Table 1). The fitness value associated with having either one or two resistance mutations ranged from ∼1 down to 0.82 per generation. Some single mutations (gyrA S83L, gyrA D87N, and parC S80I) were statistically neutral whereas others (ΔmarR and ΔacrR) caused a significant reduction in fitness (0.83 and 0.91 per generation, respectively). Strains carrying three resistance mutations had fitness values that ranged from ∼1 down to as low as 0.60, with 8/9 strains suffering a fitness deficit of ≥5% per generation. Interestingly, the addition of a fourth resistance mutation to these strains restricted the minimum fitness value to 0.66 per generation, higher than that measured with three mutations (Figure 1). When all five resistance mutations were present the fitness value was 0.68. Thus, some strains carrying 4 or 5 resistance mutations have a higher fitness than some strains carrying only 3 mutations. The major negative effect on fitness was associated with the presence of the marR and acrR mutations. Thus, fitness did not decrease as a simple function of the number of resistance mutations, but instead depended critically on the nature of those mutations.
In general the addition of a resistance mutation to a strain was either neutral with respect to MIC and fitness, or it caused an increased MIC and / or decreased fitness (Figure 1, Table 1). Across all the strains constructed decreased fitness was very strongly associated with the presence of one or both efflux mutations. In contrast, strain LM693 (gyrA S83L, D87N; parC S80I) has high level resistance (MIC 32) with no significant reduction in fitness relative to the wild-type (Table 1). LM693 shows that it is possible to evolve high level resistance with no, or minimal, fitness costs. However, mutations up-regulating drug efflux are highly relevant to resistance evolution because they arise at a rate hundreds of times higher than mutations in the structural genes for topoisomerases. This is because the genetic target for knockout mutations in efflux regulating genes is much larger than the target for the specific amino acid substitutions required in topoisomerase genes. Thus, even though, as shown here, efflux mutations are fitness-costly and contribute relatively small increases in resistance, they occur very frequently, and are found in many resistant clinical isolates [4]. For three low-fitness strains, each carrying three resistance mutations including a marR mutation, the addition of an extra resistance mutation increased both MIC and fitness: LM695→LM707; LM882→LM707; and LM871→LM707 (Figure 1 and Table 1). In these strains the added mutation affected gyrA or parC. The increased fitness was statistically significant (Students t-test, two-tailed, p<0.05) in two of the three cases, LM695>LM707 and LM882→LM707 (Table 2). The robustness of these results was verified by independently reconstructing each of these strains and re-measuring their MIC and fitness values. No significant differences were found from the original values. In addition, we tested these critical strains by de-construction experiments: replacing gyrA and/or parC mutations with equivalent wild-type genes and determining that the MIC and fitness values of the de-constructed strains were as expected. Based on these two experiments, re-constuction, and de-construction, we are confident that the isogenic strains do not carry any additional mutations affecting MIC or fitness. Thus, the addition of a single resistance mutation can increase growth fitness by 5–10% per generation and simultaneously increase MIC more than 40-fold (Table 2).
The data in the previous section showed that the addition of a fourth resistance mutation to either LM695 or LM882 could significantly increase competitive growth fitness measured in vitro. This result would be more interesting from a clinical viewpoint if the measured increase in fitness was not exclusively an in vitro phenomenon. To test this competition experiments were made in a mouse UTI infection model [16]. Each of the strains (LM695, LM883, and LM707) was competed against the isogenic wild-type in the mouse model and relative fitness expressed as a competitive index (C.I.).Three different measures of C.I. could be obtained in this model: from the urine; the bladder; and the kidneys. For each strain and tissue there was a clear positive correlation between relative fitness in vivo and in vitro (Figure 2). Thus, the relative order of fitness values of these strains, initially measured in vitro, was confirmed in the physiologically more complex in vivo environments.
The marR mutation makes the single largest contribution to loss of fitness in these 28 strains (Table 1). The marR mutation alone reduced fitness to 0.83±0.3 (LM202), and the average fitness of all strains carrying marR is 0.76 (SD 0.08). MarR protein regulates, directly and indirectly, transcription of many genes [17] and the reduction in fitness associated with loss of function mutations in mar is most likely because of disruption of gene regulation. This suggests a possible mechanism of fitness compensation associated with resistance mutations in topoisomerase genes. Thus, some topoisomerase mutations that alter chromosomal supercoiling levels may partially restore expression of growth-limiting gene(s) regulated by mar, reversing the fitness deficit. This hypothesis predicts that the improvement in fitness measured in LM695→LM707 should be associated with a change in global supercoiling level. This was tested by electrophoresis of pUC18 and pUC19 plasmids purified from MG1655, LM695 and LM707 in chloroquine-agarose gels [18]. With this method we were unable to detect differences in plasmid topoisomer patterns (data not shown). We also tested for differences in supercoiling using a reporter gene assay [19],[20]. This was done by introducing plasmids carrying a luciferase reporter gene fused to each of two different promoters (ptopA and pgyrA) whose expression is sensitive to, respectively, increased or decreased supercoiling levels [20]. In this assay the ratio (ptopA / pgyrA) of reporter gene expression from these two promoters, defined as the quotient of supercoiling (Qsc), is a measure of the relative level of negative supercoiling in isogenic strains [19],[20]. Qsc was 2.9 (SD 0.7) for MG1655 (n = 6), 2.6 (SD 0.7) for LM695 (n = 12), and 3.6 (SD 1) for LM707 (n = 13). In this assay neither of the mutant strains differed significantly from the wild-type, but LM695 and LM707 differed significantly from each other (P value 0.01, Students t-test, two tailed). From the expression data it was clear that the major cause of this difference in Qsc between LM695 and LM707 was due to a 25% decrease in expression from the gyrA promoter construct in LM707 relative to LM695. Thus the introduction of the parC S80I mutation to make strain LM707 caused a significant increase in the Qsc, coincident with the improvement in fitness. Although we cannot explain the absence of an effect in the chloroquine-agarose assay, the reporter gene assay suggests that some changes in chromosomal supercoiling associated with the acquisition of topoisomerase mutations may provide a mechanism linking bacterial fitness and decreased susceptibility to fluoroquinolones. However, an alternative explanation is that the combination of mutations in LM707 reduces expression of gyrA by a mechanism that does not change global supercoiling levels, and that it is a consequence of the reduced expression of gyrA that causes the increase in growth fitness. Additional experiments will be required to distinguish between these models.
In the sections above it was shown that the transfer of the parC S80I mutation into LM695 increased its MIC for ciprofloxacin and its competitive growth fitness versus the wild-type. This predicted that the resulting constructed strain, LM707, should outcompete LM695 in a head-to-head competition, and also raised the following questions: (i) could a mutant with an increased MIC for ciprofloxacin and a higher growth fitness be selected spontaneously from LM695 in the absence of drug; and (ii) would such a mutant be exclusively associated with of the acquisition of the parC S80I mutation. These predictions and questions were addressed experimentally. First, in head-to-head growth competition experiments in LB medium with no drug, LM707 (four resistance mutations) outcompeted LM695 (three resistance mutations), gaining ∼8% per generation, in good agreement with the relative differences in growth fitness of each strain versus the wild-type (Table 1). Second, 96 independent lineages of LM695 were grown in rich medium in the absence of drug for 4 growth cycles. Each growth cycle was inoculated with 2×106 cfu and grown to a total of 2×108 cfu (∼7 generations per growth cycle) in a volume of 200 µL. An aliquot of 5×106 cfu from each lineage was tested, after the completion of 2, 3, and 4 cycles of growth, for the presence of resistant mutants on solid medium (3 µg/mL ciprofloxacin, 4×MIC). Growth of LM695 is completely inhibited on 3 µg/mL ciprofloxacin and the spontaneous mutation rate to resistance on this media, measured in fluctuation tests, is 2×10−8. Accordingly, we expected that virtually none of the 96 independent lineages would contain a resistant mutant at the initiation of the experiment, but that a small number of resistant mutants would arise in each lineage during each growth cycle. If these mutants out-competed the parental LM695 they would be expected to increase relative to LM695, and thus have a greater probability of being transferred to the next growth cycle. The number of lineages from which resistant mutants were obtained was found to increase with successive growth cycles: from 2/96 (cycle 2)→8/96 (cycle 3)→15/96 (cycle 4). Because the spontaneous mutation rate to resistance (2×10−8) if much lower than the number of cells being tested from each lineage (5×106) it is very unlikely that these mutants arose on the selective media. Instead, the most reasonable conclusion is that the resistant mutants arose spontaneously and randomly during the growth of lineages in the absence of drug and were enriched because they out-competed the parent population. Three random drug-resistant mutants were chosen from independent lineages and tested by DNA sequencing for the presence of mutations in gyrA and B, in parC and E, and in marR and acrR. In each case the mutations originally present in LM695 were confirmed and in addition each of the mutants was found to have acquired a new mutation in parC (S80R in two cases, and E84K in one case). The MIC CIP of each of the mutants had increased from 0.75 to >32 µg/mL, and the exponential doubling time in rich medium had increased significantly, by ∼10% per generation (Table 3). Thus, the selection of a spontaneous parC mutation in LM695 decreased its susceptibility to ciprofloxacin and increased its growth rate. From these experiments we concluded that the phenotypes generated directly by strain construction (LM695→LM707; reduced drug susceptibility and increased growth fitness) could also be generated by a variety of spontaneous mutations, and that the growth advantage phenotype could be enriched by growth selection in the absence of drug.
A set of 28 isogenic E. coli strains was constructed and used to measure the relationship between the accumulation of fluoroquinolone resistance mutations, drug susceptibility, and growth fitness. The question was whether the accumulation of up to five resistance mutations, commonly found in resistant clinical isolates, would progressively reduce bacterial fitness. Most of the published data on the relationship between drug resistance and bacterial fitness would predict two possibilities: (i) that these mutations would cause little or no fitness cost, explaining their high frequency among resistant isolates; or (ii) that their accumulation would cause a progressive decrease in bacterial fitness, and require additional fitness-compensating mutations to restore fitness [6],[7]. The data from this study support, in part, each of these scenarios (Table 1, and Figure 1). However, they also revealed, for some combinations of resistance mutations, a positive relationship between reduced drug-susceptibility and increased bacterial fitness. This positive relationship could be another driving force in the development of increased resistance to these antibacterial drugs. Although this may be the first demonstration in bacteria that in the absence of an antimicrobial, selection can increase resistance to that antimicrobial, a similar phenomenon has been reported for HIV resistance to a protease inhibitor [21]. Thus, the phenomenon described here may have a broad biological significance.
The main conclusions from the data set were the following:
Molecular details of how particular parC and gyrA mutations together improve the fitness of a resistant strain with a marR mutation are beyond the scope of this study but we can suggest the outlines of a model. Growth rate depends on the rate of transcription regulated in accordance with physiological demands [22]. The MarR protein regulates, directly and indirectly, the transcription of many genes [17] and it is probable that the severe reduction in fitness associated with ΔmarR (Table 1) is because it causes inappropriate patterns of transcription regulation. We suggest that specific mutant forms of DNA gyrase and topoisomerase IV, possibly by acting in concert to influence the level of superhelicity in DNA, restore appropriate levels of gene expression at some loci where the loss of MarR regulation has a negative impact on growth rate [23],[24].
The particular gyrA and parC resistance mutations studied here are clinically relevant, being among the most common found in fluoroquinolone-resistant clinical isolates of E. coli [4]. Among 30 resistant UTI isolates analyzed: 30/30 had the gyrA S83L mutation; 18/30 had the double mutation gyrA S83L, D87N; 22/30 had the parC mutation S80I; and 15/30 had the triple combination gyrA S83L, D87N, parC S80I (23/30 had some form of triple mutation combination including other substitutions at position 87 of gyrA and/or position 80 of parC). Thus, the combination of target mutations studied here is typical of resistant clinical isolates. As measured by organic solvent tolerance, drug efflux was phenotypically up-regulated in 15/30 of these resistant isolates, associated in most cases with mutations in acrR and/or marOR [4]. The observed frequency of the efflux phenotype is very low relative to the frequency of specific gyrase and topoisomerase mutations in the same isolates, and given the much higher expected probability of mutations that knock out the function of efflux regulator genes it suggests selection against such mutations. This under-representation is consistent with our finding that mutations up-regulating efflux pumps carry significant fitness costs. The growth fitness of clinical isolates cannot be meaningfully compared with the data presented in this paper, in part because clinical isolates are not isogenic, and in part because clinical isolates will already have evolved to ameliorate the fitness costs, if any, associated with their resistance determinants. We believe that our data provide an insight into the likely initial effects on relative fitness and drug susceptibility of different pathways of resistance development. In particular, that one consequence of a following a high-probability but low-fitness evolutionary pathway may be that one or more steps may be driven by selection for increased fitness in the absence of drug exposure.
How fluoroquinolone resistance evolves in nature will depend on the genotype being selected and on the selective environment [25] but it is likely include mutational steps that reduce bacterial fitness. Bacteria that progress down an evolutionary path with reduced fitness relative to a competing population may be driven to extinction, or may, given the opportunity by mutation, acquire a change that increases their relative fitness thus improving their chances of survival. Evolutionary paths that could be taken on the road to extinction or antibiotic resistance are outlined in Figure 3. The low-fitness mutants LM695 and LM882 each have MICs that lie under the resistance breakpoint for ciprofloxacin [26],[27]. Such mutants are at a critical stage in resistance development: having low fitness they may be driven to extinction by natural selection; being under the resistance breakpoint they may avoid detection in a clinical setting; however, they are, as shown here, only one mutational step away from a high-level resistance phenotype with increased fitness, without additional exposure to the drug. The magnitude of these co-selected changes in fitness and susceptibility are significant (Table 2). These data argue in favor of testing anti-mutant dosing strategies, or other measures that could prevent the enrichment of low-level resistant mutants [28].
C1186 is a multiply mutant fluoroquinolone resistant UTI isolate previously described [4]. E. coli K12 MG1655 wild-type was the starting strain for all constructions (Table 1). Liquid growth medium was Luria broth (LB) while solid medium was Luria-Bertani agar (LA). Strains were grown at 37°C. Ciprofloxacin (Bayer HealthCare AG, Wuppertal, Germany) was dissolved in 0.1 M NaOH at 100 µg/mL then further diluted in LA in selective plates.
MIC was determined by Etest (AB BIODISK, Solna, Sweden) on Mueller-Hinton agar plates incubated for 16 to 18 h at 37°C with quality control reference strains [4] as recommended by the Clinical and Laboratory Standards Institute (www.clsi.org).
Spontaneous resistance mutations in gyrA and parC were selected sequentially in E. coli MG1655 in LB with ciprofloxacin at 2–8-fold MIC and mutations were identified by DNA sequencing. Individual mutations were always moved into a clean genetic background (MG1655) after initial selection. Deletion-replacement mutations in marR, acrR, yfaH, metC and araB were made by λ-red recombineering [29] in NC397 (a Lac+ Nad+ derivative of DY329) using PCR amplified linear DNA from pCP16 with an FRT-bounded TcR cassette [30]. The PCR reaction protocol was 95°C 5 min followed by 30 cycles of 95°C 15 sec, 55°C 20 sec, 72°C 240 sec. PCR primer sequences with details of the deletion-replacement boundaries are shown in Supporting Information (Table S1).
Isogenic derivatives of MG1655 were constructed by phage P1-mediated transduction. Transduction of the deletion-replacement mutations in marR and acrR was selected directly on LA+Tc. When transducing the gyrA and parC mutations selection was made for the linked markers, yfaH<>Frt::TcR::Frt and metC<>Frt::TcR::Frt, each ∼10 kb from gyrA and parC, respectively. Note that we are using the symbol <> to indicate a replacement generated by λ red homologous recombination technology. After transduction the TcR marker was removed by Flp-catalyzed excision expressed following transformation with pCP20 [30]. All strain constructs were confirmed by DNA sequencing.
LM347 (ΔaraB<>FRT) was used as the standard wild type strain in growth competitions against which each of the constructed mutant strains was competed. This strain was tested against its parent MG1655 showing that the ΔaraB mutation was neutral (relative growth rate 1.002±0.005). To support statistical analysis each competition was tested in at least 5 independent experiments (Table 1). To initiate growth competitions, each strain was grown in LB at 37°C 12 h, mixed in a 1∶1 ratio, diluted 10−3 into LB, then grown 23 h to complete a growth cycle. Each successive growth cycle was initiated by diluting the mixture 10−3 into LB. Each competition experiment (4 cycles) was made the number of times indicated in Table 1. For some low-fitness strains such as LM595 the mutant population became too low to detect after the second or third cycle of competition. After the initial mixing, and after each growth cycle, appropriate dilutions of the mixture were plated onto MacConkey agar plates containing 1% arabinose. Plates were incubated 37°C overnight. Red (ara+) and white (ΔaraB) colonies were scored. The change in the ratio of mutant/wild-type was used to estimate the selection coefficient per generation of each of the constructed strains according to the formula: S = ln2 (mutant/wild-type) / generation [31]. Relative fitness per generation with respect to the wild-type LM347 is defined as S+1. Note that the fitness defects could be due to defects at any stage of the growth cycle (lag, exponential, stationary phase).
Relative bacterial fitness in vivo was measured using an established urinary tract infection model [16]. Details, including mouse strain and ethical permission, are given in Text S1. Competitive index was calculated as the geometric mean of the ratio of mutant/wild-type bacteria isolated from each organ (urine, bladder, kidneys) of 8 mice per experiment, normalised to the ratio at the time of inoculation.
Exponential doubling times were calculated by measuring the increase in optical density at 600 nm (OD600) at 10-min intervals, using a BioscreenC machine (Oy Growth Curves Ab Ltd. Helsinki, Finland).
The mutation rate of LM695 to resistance on LA+3 µg/mL ciprofloxacin (CIP) is 2×10−8 measured by fluctuation test. Independent lineages were inoculated in 96×200 µL wells, and grown 24 h at 37°C to ∼109 CFU/mL. 2 µL was transferred to initiate a new growth cycle. 5 µL was plated on LA+3 µg/mL CIP to assay for resistant mutants.
The relative supercoiling degree was determined using a published assay. The quotient of supercoiling, Qsc [19] is defined as the ratio of luciferase activity from two different supercoiling sensitive promoters, ptopA-luc, and, pgyrA-luc, [20]. Details are given in Text S2.
|
10.1371/journal.pntd.0002017 | A Multiplex PCR for the Simultaneous Detection and Genotyping of the Echinococcus granulosus Complex | Echinococcus granulosus is characterized by high intra-specific variability (genotypes G1–G10) and according to the new molecular phylogeny of the genus Echinococcus, the E. granulosus complex has been divided into E. granulosus sensu stricto (G1–G3), E. equinus (G4), E. ortleppi (G5), and E. canadensis (G6–G10). The molecular characterization of E. granulosus isolates is fundamental to understand the spatio-temporal epidemiology of this complex in many endemic areas with the simultaneous occurrence of different Echinococcus species and genotypes. To simplify the genotyping of the E. granulosus complex we developed a single-tube multiplex PCR (mPCR) allowing three levels of discrimination: (i) Echinococcus genus, (ii) E. granulosus complex in common, and (iii) the specific genotype within the E. granulosus complex. The methodology was established with known DNA samples of the different strains/genotypes, confirmed on 42 already genotyped samples (Spain: 22 and Bulgaria: 20) and then successfully applied on 153 unknown samples (Tunisia: 114, Algeria: 26 and Argentina: 13). The sensitivity threshold of the mPCR was found to be 5 ng Echinoccoccus DNA in a mixture of up to 1 µg of foreign DNA and the specificity was 100% when template DNA from closely related members of the genus Taenia was used. Additionally to DNA samples, the mPCR can be carried out directly on boiled hydatid fluid or on alkaline-lysed frozen or fixed protoscoleces, thus avoiding classical DNA extractions. However, when using Echinococcus eggs obtained from fecal samples of infected dogs, the sensitivity of the mPCR was low (<40%). Thus, except for copro analysis, the mPCR described here has a high potential for a worldwide application in large-scale molecular epidemiological studies on the Echinococcus genus.
| The dog tapeworm Echinococcus granulosus (E. granulosus) is a cosmopolitan parasite. The adult worms reside in the small intestine of their definitive hosts (dogs). Infective eggs are shed with the feces into the environment and are orally ingested by intermediate hosts where they develop into the metacestode (larval) stage, causing cystic echinococcosis (CE) in humans and livestock. Ten intraspecific genotypes of E. granulosus (G1 to G10) have been reported from different intermediate host species. Based on the recently established molecular phylogeny, E. granulosus is now considered a complex consisting of four species: E. granulosus sensu stricto (G1/G2/G3), E. equinus (G4), E. ortleppi (G5) and E. canadensis (G6–G10). Simple and highly discriminative molecular epidemiological approaches are needed to explore dynamics, life cycle patterns, and the pathogenicity of the members of this complex. We here introduce a one-step multiplex PCR (mPCR) protocol for the genotyping and discrimination of the different members of the E. granulosus complex, allowing three levels of discrimination: (i) Echinococcus genus, (ii) E. granulosus complex, and (iii) genetic variants within the E. granulosus complex. The relatively complicated task of E. granulosus complex speciation and genotyping is clearly simplified by mPCR, and this technique therefore represents a useful tool for routine practice.
| Historically, four species have been recognized within the genus Echinococcus: E. multilocularis, E. oligarthrus, E. vogeli and E. granulosus [1]. E. shiquicus and E. felidis are two newly discovered additional species isolated from small Tibetan mammals and African lions, respectively [2], [3]. Extensive research on genetic variation, intermediate host affinities as well as morphological, biological and biochemical differences resulted in a more sophisticated classification of the dog tapeworm E. granulosus into ten genotypes/strains [4]–[6]: sheep strain (G1), Tasmanian sheep strain (G2), buffalo strain (G3), horse strain (G4), cattle strain (G5), camel strain (G6), pig strain (G7), cervid strain (G8), pig/human strain (G9) and Fenno-Scandian cervid strain (G10). The poorly characterized strain G9 is closely related to E. canadensis (G7) [7] and the existence of G9 as a separate genotype remains still controversial [8], [9].
More recently, new data obtained from phylogenetic analysis have shown an even more pronounced genetic divergence between these ten E. granulosus genotypes [5], [10]. Based on sequences of the complete mitochondrial genome [11] and several nuclear markers [8], [12], the phylogeny for E. granulosus was reconstructed. Data obtained from nuclear protein-coding genes resulting in two nuclear alternative phylogenies: (i) nuclear phylogeny [8] is supported by morphological data, whereas (ii) nuclear phylogeny [12] is in agreement with mitogenome phylogeny [13]. Thus, E. granulosus became considered as a complex consisting of four species: E. granulosus sensu stricto (G1/G2/G3), E. equinus (G4), E. ortleppi (G5) and E. canadensis (G6–G10). The phylogenetic relations within the latter group remain unresolved and are still under controversial discussion, since the E. canadensis cluster was proposed to be divided into the two species E. canadensis (G8/G10) and E. intermedius (G6/G7) [14], [15]. This proposal gained further support from nuclear phylogeny [8], but mitogenome phylogeny analyses contradicted this assumption by showing that E. canadensis (G6/G7/G10) form a subgroup and E. canadensis (G8) is a closely related sister taxon [16].
The adult worms of E. granulosus complex reside in the small intestine of their definitive hosts, principally wild or domestic canids. Infective eggs are shed with feces into the environment and are orally ingested by intermediate hosts where they develop into the metacestode (larval) stage, known as the aetiological agent of cystic echinococcosis (CE) in humans and predominantly ruminants, pigs and horses. Due to its success to undergo its life cycle in domesticated animals during both definitive and intermediate stages, E. granulosus constitutes an important worldwide public health problem with significant economic impact [17]–[19].
Human susceptibility to CE depends largely upon the infecting species or genotype of the E. granulosus complex. Worldwide molecular epidemiological studies revealed that E. granulosus s.s. (G1) is most commonly found in humans, but also a high prevalence of E. canadensis (G6) [20]–[25] and E. canadensis (G7) [26], [27] was reported. E. ortleppi (G5) has a very marginal impact on human health with only two reported cases [20], [28]. One major factor behind the worldwide spreading of many zoonoses can be the introduction of the parasite by host animals, as it happened in Australia, where E. granulosus was imported with domestic livestock about 200 years ago [29].
The worldwide distribution of CE reveals a geographic heterogeneity of E. granulosus species in many overlapping areas. Some examples are the co-existing genotypes E. granulosus s.s. (G1) and E. canadensis (G6) in North African countries [23], [30]–[32], E. granulosus s.s. (G1/G2), E. ortleppi (G5) and E. canadensis (G6/G7) in Argentina [20], [33], [34] or E. granulosus s.s. (G1), E. canadensis (G6) and E. equinus (G4) in Kyrgystan [35]. In these areas co-infections with more than one E. granulosus species/genotype might occur in the intermediate or definitive hosts. In addition, the not yet confirmed hypothesis of an eventual genetic exchange by sexual reproduction between E. granulosus species/genotypes is still discussed [36].
The knowledge about the distribution of the E. granulosus complex is important e.g. in the context of any control or eradication program. Thus, regular molecular epidemiological surveys provide key information on the spatio-temporal dynamics of parasite populations. Knowledge about the transmission and prevalence of E. granulosus in humans and animals, including dogs, is a basic step before and during control and/or surveillance strategies.
Different methods for genotyping genetic variants of the E. granulosus complex have been developed so far. Based on PCR amplified sequences of the mitochondrial cytochrome c oxidase subunit 1 (cox1) or the NADH dehydrogenase subunit 1 (nad1), genotyping can be performed in a relative time and/or cost intensive way by sequencing [37], RFLP (Restriction Fragment Length Polymorphism) [38], [39], fingerprinting [40] or SSCP (Single Strand Conformation Polymorphism) [41]. More recently, pure PCR based methods that simplify the genotyping have been designed. With a consecutive PCR approach a part of the E. granulosus complex (G1, G5, G6/G7) can be genotyped [42] and by applying four parallel PCRs the discrimination between E. multilocularis, E. granulosus s.s. (G1) and an E. ortleppi (G5)/E. canadensis (G6/G7) cluster is possible [43]. Parallel PCR approaches can be combined in a multiplex PCR setup and became rapidly and successfully applied worldwide in many aspects of DNA analyses, especially in the field of molecular diagnosis of infectious diseases such as bacterial [44], viral [45] and fungal [46] infections. For cestode infections, a 3-plex-PCR approach was already established to distinguish between E. multilocularis, E. granulosus complex and Taenia [47]. However, the potential of such an approach has not yet been evaluated for the specific detection and/or genotyping of different isolates within the E. granulosus complex.
Based on the identification of a number of discriminating polymorphism sites in nuclear and mitochondrial genes of the Echinococcus genus, we established a single-tube multiplex PCR (mPCR) approach that allows a rapid and simultaneous detection and discrimination among the following members of the E. granulosus complex: E. granulosus s.s. (G1/G2/G3), E. equinus (G4), E. ortleppi (G5), E. canadensis (G6/G7) and E. canadensis (G8/G10). We assessed the performance of the mPCR assay by re-identifying reference DNA panels (42 samples) and by genotyping 153 unknown DNAs from human and animal Echinococcus cyst samples isolated from infected intermediate hosts in Tunisia, Algeria and Argentina. Finally, we assessed the feasibility of applying mPCR for the detection and genotyping of E. granulosus complex in fecal egg samples, and directly in frozen or fixed parasite material (hydatid fluid or protoscoleces).
Based on known mitochondrial or nuclear DNA sequences, polymorphisms between Echinococcus strains/genotypes were identified and used for strain/genotype specific primer design. Each primer pair was first applied on its respective genotype-specific DNA, and if one clear PCR product was amplified, it was applied on DNA samples of all other genotypes/strains in order to exclude non-specific amplicons. Finally, 11 primer-pairs resulting in genotype/strain/genus specific targets were used for the mPCR.
The mPCR was set up with normalized known template DNAs in a sequential approach by starting with one specific primer pair in the PCR mix, followed by the incorporation of other primer pairs. The PCR was run with every additional new primer pair on all genotype/strain specific DNA samples to confirm specificity. Simultaneously the molar amount of primers was adjusted in order to achieve comparable amplicon intensities.
To reduce variable parameters and to allow comparison between experiments the basic mPCR conditions using GoTaq DNA polymerase from Promega were defined as followed: 94°C for 3 min, 25 cycles of 94°C for 30 sec, 56°C for 30 sec, 72°C for 30 sec and a final extension step for 5 min at 72°C. With this setup the sensitivity range was determined by adding different amounts of template DNA into the mPCR mix. The specificity of the mPCR was tested by (i) adding more PCR cycles, (ii) using mixed DNA templates derived from different Echinococcus genotypes/strains, (iii) using template DNAs of closely related genus Taenia or (iv) by the addition of foreign DNA derived from bovine thymus or dog feces.
To exclude lab-specific conditions, 13 samples were genotyped by mPCR in two different laboratories. To assess potential problems with materials derived from different suppliers, the system was tested with DNA polymerases from different companies. The mPCR performance was further validated by genotyping 42 E. granulosus complex samples derived from known origin and genotype, and subsequently 153 unknown DNA samples were genotyped. Furthermore, the mPCR was assessed on DNA derived from Echinococcus eggs isolated from feces of infected dogs. Finally, approaches were developed to perform the mPCR directly on fresh protoscoleces, either frozen or fixed, or on hydatid fluid.
Information on the complete mitochondrial genome sequences containing the genes cytochrome oxidase subunit I (cox1), cytochrome oxidase subunit 2 (cox2), ATP synthase subunit 6 (atp6) and NADH dehydrogenase subunit I (nad1) as well as mRNA sequences of the nuclear genes RNA polymerase II (rpb2), DNA polymerase delta (pold), ezrin-radixin-moesin-like protein (elp), elongation factor 1 alpha (el1a) and calreticulin (cal) were obtained from the databases of the National Center of Biotechnology Information (NCBI) for E. granulosus s.s. (G1/G2/G3), E. equinus (G4), E. ortleppi (G5), E. canadensis (G6/G7), E. canadensis (G8/G10), E. multilocularis, E. vogeli and E. oligarthrus. The respective sequences were retrieved via GenBank [http://www.ncbi.nlm.nih.gov/] and were aligned with BioEdit 7.0.9 to detect polymorphic sites. The accession numbers of the used Echinococcus sequences are listed at the end of the manuscript.
The primers were designed on the assumption that one specific 3′-base will be sufficient to result in genotype-specific amplification since Taq-polymerases lack a 3′–5′ proofreading activity. In consequence, primers were chosen such as to strain-specifically bind to the targets described above. If possible, primers were selected that contained more than one specific 3′-base, but five primers of the final set that were targeted to nuclear sequences matched this one base difference. Because genotyping based on single nucleotide polymorphisms (snips) is error-prone due to mutations [48], [49], we chose two genotype/strain-specific probes for all E. granulosus complex members. The exception was E. canadensis (G8/G10), where only one probe was selected due to its rare occurrence and close relationship to E. canadensis (G6/G7). Two additional probes were chosen: a common one for all E. granulosus complex members, and one for the overall detection of all known Echinococcus species: E. granulosus complex, E. multilocularis, E. vogeli, E. oligarthrus and E. shiquicus. Therefore, three levels of differentiation were obtained for each sample by determination of (i) the genus Echinococcus, (ii) the affiliation to the E. granulosus complex and (iii) the specific strain or genotype within the complex. For all primers, a Tm of approximately 55°C was selected, and for each primer-pair a PCR product of distinct size was anticipated, in order for the amplicons to be easily discriminated by 2% agarose gel-electrophoresis. Table 1 shows the complete list of the final 22 primers used in this study, including names, molar concentrations in the mPCR mix, the final product sizes, the specificities (genotypes), the primer sequences (including the polymorphic sites), the primer lengths, the target genes (gene marker), the accession numbers of the published DNA target sequences, and the corresponding positioning of the primer sequences within their targets.
The reaction mix for the final mPCR was composed of 100 µM dNTPs and 0.05 units µl−1 GoTaq DNA polymerase in 1× PCR Buffer (all Promega) and contained the 22 primers specific for 11 targets in the molarities shown in Table 1. For standard genotyping 5 ng template DNA were added into the PCR mix. Each reaction was performed in single tubes in a volume of 20 µl PCR mix. The cycling conditions were as follows: an initial denaturation step at 94°C for 3 min, 25 cycles (94°C–30 s, 56°C–30 s, 72°C–1 min) and a final extension step lasting 5 min at 72°C. 10 µl of the PCRs were separated by electrophoresis in a 2% agarose gel and visualized by ethidium bromide staining and subsequent UV excitation. The genotype specific amplicon profile is shown in Figure 1. The mPCR conditions were a result of pre-experiments described below, and these conditions were used throughout if not indicated otherwise.
Ethical statement: For the parasite samples of animal origin, these were taken from animals in abattoirs being processed as part of the normal work of the abattoirs, in the frame of conventional meat inspection. For the parasite samples of human origin, these were obtained for and thus part of the normal diagnostic investigation to determine the etiology of the biopsied tissue for clinical purpose. Thus the present investigation was part of the conventional diagnostic procedure used in clinical practice. Samples were all anonymized for carrying out data evaluation.
(A) For establishment of the mPCR and all evaluations concerning the sensitivity and the specificity of the method, a test panel of E. granulosus complex chromosomal DNAs was used. Genomic DNA specimens used for the test panel were: E. granulosus s.s. (G1), E. equinus (G4), E. canadensis (G6), E. canadensis (G7), and E. canadensis (G8). These were obtained from institutional DNA-collections in Berne/Switzerland, Zürich/Switzerland and Tartu/Estonia. Genomic DNA extracted from E. ortleppi (G5) was kindly provided by Dr. Karen Haag (Departamento de Genétic, Instituto de Biociências, Universidade Federal do Rio Grande do Sul/Brazil) and protoscoleces from E. canadensis (G10) were kindly provided by Prof. Thomas Romig (Institute of Parasitology, University of Hohenheim/Germany). All samples had been genotyped conventionally by sequencing cox1 and/or nad1. The genomic DNA of the E. canadensis (G10) protoscoleces was isolated using a standard phenol-chloroform protocol [50], using RNAse A (Sigma-Aldrich), Proteinase K (Sigma-Aldrich) and a subsequent isopropanol precipitation followed by multiple washes in 75% EtOH prior to drying and dissolving in ddH2O.
For most genotyped samples used in these parts of the study, the original extraction method for genomic DNA could not be retrospectively determined. A general problem in the usage of genomic DNA prepared by multiple methods (e.g. column based nucleic acid purification, phenol/chloroform extraction, presence or absence of RNAseA or proteinase K treatment) arises when quantifying the DNA concentration, e.g. by Nanodrop ND-1000 measurement. Therefore, an E. granulosus s.s. (G1) DNA amount (selected upon the most intense PCR amplification product when using the Echinococcus specific primers Echi-Rpb2 F and Echi-Rpb2 R, 1 µM, see Table 1), was defined as a reference measure point. The DNAs of all other species/genotypes were normalized to this sample by comparative PCR using the same primers. The PCRs were performed under the following conditions: 94°C for 3 min followed by 25 cycles of 94°C for 30 s, 56°C for 30 s and 72°C for 1 min and a final extension step of 5 min at 72°C.
(B) For the evaluation of specificity in the context of cross binding of the primers, DNA derived from Echinococcus species outside of the E. granulosus complex (E. multilocularis and E. vogeli) as well as DNA of the closely related Taenia saginata, T. solium, T. crassiceps, T. taeniaformis and T. pisiformis were obtained from the institutional DNA-collection at the University of Berne/Switzerland.
(C) For the evaluation of specificity in the context of contaminating DNA, bovine thymus DNA was obtained commercially from Serva, and dog feces DNA was isolated as described above by phenol/chloroform extraction from feces of a helminth-free dog that was obtained from the Small Animal Clinic of the Vetsuisse Faculty, University of Berne, Switzerland.
(D) For the assessment of the mPCR genotyping performance on DNA derived from metacestodes and/or protoscoleces, two panels of known (reference) and unknown Echinococcus metacestode DNAs were used. Known/genotyped materials were 20 reference DNA samples originating from Bulgaria [51] and 22 samples from Spain (unpublished) obtained from the institutional DNA-collection at the University of Berne/Switzerland. Unknown/non-genotyped materials were 13 DNA samples harvested from slaughterhouses in Buenos Aires/Argentina. Protoscoleces fixed in 95% (v/v) ethanol were obtained from 101 animal cysts harvested from slaughterhouses in Tunisia (75 samples) and Algeria (26 samples). Human isolates were collected after surgery from human patients in Tunisia (39 samples). Chromosomal DNA was prepared as described above. For more detailed information e.g. on host animal species, see Table 2. A part of these samples were used for the reliability and reproducibility tests. These 66 samples are marked with an asterisk in Table 2.
(E) For the assessment of the mPCR genotyping performance of feces, eggs were isolated according to Mathis et al. [52] from 28 dog fecal samples (Sample collection Zürich/Switzerland: 20 samples from a study in Kyrgyzstan [35] and 8 samples from a study in Lithuania [53]). DNA extraction was performed as previously described [54], and the DNA was characterized by a multiplex PCR for the simultaneous detection of E. granulosus (G1–G10), E. multilocularis and Taenia spp. [47]. Echinococcus was identified in all samples; 18 out of 28 with E. granulosus (10 from Kyrgyzstan, mainly sheep strain G1) and 8 from Lithuania where only E. canadensis G7 occurs and 10 with E. multilocularis (10 from Kyrgyzstan). These preselected samples were used to assess the potential of the mPCR as a molecular diagnosis tool for canine infection with adult Echinococcus.
(F) To evaluate the mPCR directly on parasite material, none genotyped Echinococcus samples obtained from the institutional sample collection of Berne/Switzerland were used: (i) frozen hydatid fluid, (stored at −20°C) and (ii) solid E. granulosus complex germinal layers and protoscoleces, used natively (frozen) or fixed in either 95% (v/v) ethanol or 4% PBS-buffered formaldehyde solution.
The mPCR conditions described above were a result of 3 preliminary sets of experiments. Used samples are described above in sample section A.
To specify the amount of template DNA which can be used in the mPCR, the sensitivity of the method was determined by varying the template concentrations of normalized test panel E. granulosus complex DNAs in the standard mPCR mix containing all 22 primers. Therefore 0.1, 0.5, 1, 2.5, 5, 10, 25, 50, 100, 250, 500 ng and 1 µg normalized DNA from E. granulosus s.s. (G1), E. equinus (G4), E. ortleppi (G5), E. canadensis (G6), E. canadensis (G7), E. canadensis (G8) or E. canadensis (G10) were tested individually by mPCR employing the conditions described above (sample origin is described in sample section A). For the readout of this experiment low amounts of the different template DNAs had to result in clearly visible bands and high amounts of template should not yield additional or smeary products. With these preconditions/definitions a usable template range resulting in clear genotyping patterns was determined.
To test the influence of additional PCR cycles (more than 25), the mPCR was performed individually with 5 and 250 ng template DNA of the different Echinococcus strains (see sample section A). The mPCRs were run with 25, 30 and 35 amplification cycles and after gel electrophoresis the amplicons were screened for smeary or unspecific products to detect the cycle number range which resulted in clear genotyping patterns.
To determine the detection limit of a specific E. granulosus complex strain in a dual-strain DNA mixture, normalized test panel DNA from E. granulosus s.s. (G1) and E. canadensis (G6) were mixed and applied in the standard mPCR in total amounts of 5 ng (ratios; 80∶20, 60∶40 and 50∶50), 50 ng (ratios; 97.5∶2.5, 95∶5, 90∶10 and 80∶20) and 250 ng (ratios; 99.37∶0.63, 98.75∶1.25, 97.5∶2.5, 95∶5 and 90∶10). Samples are described in sample section A. For the readout, clearly visible amplicons of the E. granulosus complex DNA applied in lower ratios indicated a successful detection. Depending on the applied template amount, different ratios were detected. Additionally a DNA cocktail containing 5 ng of normalized test panel DNA from each member of the E. granulosus complex was used as template for the mPCR to verify that all 11 targets could be amplified simultaneously in one tube.
To exclude unspecific cross binding of the primers on the closely related Taenia genus, 10 ng template DNA derived from different Taenia species were applied in individually performed standard mPCRs. The samples employed for assessment of cross-binding are described above in sample section B.
To assess the mPCR specificity in the presence of host-derived contaminations in individually performed mPCRs, 5 ng of normalized E. granulosus s.s. (G1) test panel DNA (sample section A) were mixed with different amounts (1∶1 up to 1∶200) of two types of foreign DNA (sample section C). Clearly visible specific amplicons combined with a lack of unspecific PCR products indicated successful genotyping.
To confirm the reliability of the mPCR, a set of 66 samples (sample section D) were genotyped first according to the PCR-sequencing technique described by Bowles et al. 1992 [37], using the cox1 primers JB3 (5′-TTTTTTGGGCATCCTGAGGTTTAT-3′) and JB4.5 (5′-TAAAGAAAGAACATAATGAAAATG-3′). The PCR products were purified with the High Pure PCR Product Purification kit (Roche Applied Science) according to the manufacturer's instructions and subsequently sequenced using an automated DNA sequencer (Applied Biosystems, ABI 3130× I Genetic Analyzer Sequencer). Sequence data were analyzed and compared with existing sequences derived from GenBank [http://www.ncbi.nlm.nih.gov/]. In a second step these 66 samples were used as templates in the standard mPCR setup using ∼20 ng DNA. Finally the results of both genotyping approaches were compared.
The reproducibility of the mPCR was assessed by performing the test in two different qualified laboratories and using the same mPCR protocol and test samples (see sample section D). Therefore, 13 samples from Argentina were genotyped in parallel by mPCR in the laboratories of Berne/Switzerland and Buenos Aires/Argentina. The mPCRs were performed with 20 ng template DNA as described above and the results were compared between the laboratories. Additionally, all 13 samples were genotyped by cox1 sequencing (see above).
Since the mPCR was set up with GoTaq DNA polymerase from Promega and the DNA polymerases from different suppliers can influence the mPCR performance, a panel of DNA polymerases was tested in a second reproducibility test by replacing the GoTaq polymerase and GoTaq PCR buffer by other products in the standard mPCR setup. For the mPCR, 5 ng of normalized E. granulosus s.s. (G1) template DNA was used (Sample section A). DNA polymerase systems, which clearly yielded the 4 expected products, were designated as “useful” and the others yielding unspecific products, smears or missing amplicons were designated as “needing optimization”. The tested DNA polymerases and the performance results are listed in Table S1.
In total 195 E. granulosus complex DNA samples were tested. The DNA concentrations in all metacestode derived samples were measured and 1 µl (∼20 ng) of the DNA samples was used as template. The mPCR was performed with the standard settings described above. Information on the samples tested is given above in sample section D.
In order to investigate whether the mPCR is suitable as a molecular diagnostic tool to detect Echinococcus eggs in canine fecal samples, a panel of positively preselected DNA samples prepared from Echinococcus eggs was investigated. Since contaminating DNA can be present, 2 µl of the DNA samples (150–350 ng DNA) were used for mPCR, which was first performed under standard conditions as described above, and subsequently with 35 instead of 25 cycles and with up to 1 µg of template DNA per reaction. Information on the samples is given above in sample section E.
To simplify the genotyping procedure, we elaborated protocols that allow omitting DNA extraction procedures for mPCR amplification by using frozen or fixed E. granulosus materials (Sample section F). Many Echinococcus samples contain high amounts of calcium corpuscles that could interfere with the mPCR. These calcium corpuscles form a relatively solid pellet at the lowest bottom of the tube after centrifugation and by using the upper cellular part of the pellet a carry-over can be avoided. Frozen hydatid fluid (HF) (stored at −20°C) was thawed at room temperature and 1 ml was heated to 100°C for 30 min, centrifuged at 13,000 rpm for 10 min, and different volumes (0.25, 0.5, 1, 1.5, 2, 2.5, 3, 10 µl) of the resulting supernatant were used as templates for mPCR. Additionally, 1 and 2 µl none heated HF were applied in the mPCR.
Solid E. granulosus complex germinal layers (cut into small pieces) and protoscoleces were used either natively (frozen) or fixed, either in 95% (v/v) ethanol or 4% PBS-buffered formaldehyde solution. The material was prepared either by boiling or by alkaline lysis. In both cases, frozen material was used directly, and fixed material was pre-washed twice with PBS. For the preparation of the material by boiling, 10 µl solid sedimented Echinococcus material was resuspended in 90 µl H2O and incubated in a shaking heater (1,200 rpm, 100°C) for 30 min. Shaking is important in this step and if no shaking heater is available, the samples have to be vortexed from time to time, or must be intensively resuspended by pipetting. After centrifugation at 13,000 rpm for 10 min, different volumes (0.25, 0.5, 1, 1.5, 2, 2.5, 3, 10 µl) of the supernatant were used for the mPCR. For alkaline lysis, 10 µl solid Echinococcus material was incubated in 50 µl of 0.4 M NaOH and 2 µl of 1 M dithiothreitol (Sigma) and the mixture was heated for 15 min at 65°C in a shaking heater (1200 rpm). The suspension was neutralized by adding 50 µl of 0.4 M HCl and 1 µl 1.5 M Tris-HCl pH 8, and centrifuged for 10 min at 13,000 g. Shaking is important at this step (see above). For the mPCRs, 2 µl of different supernatant dilutions (1∶1, 1∶2, 1∶4, 1∶6, 1∶8, 1∶10 and 1∶25) were used in 20 µl setups. Furthermore, 1 and 2 µl undiluted supernatant were applied in the mPCR.
The mitochondrial genome and different nuclear genes were aligned and analyzed for sequence differences appearing specifically within in the genes of the individual E. granulosus complex members: E. granulosus s.s. (G1/G2/G3), E. ortleppi (G4), E. equinus (G5), E. canadensis (G6/G7) and E. canadensis (G8/G10). Specific primer-pairs were designed and tested individually for sensitivity and specificity. In these preliminary experiments, primer concentrations were 0.5 µM, but template DNA amounts varied between 10 pg and 5 ng, and different numbers of amplification cycles (25, 30 or 40) were assessed. Primer pairs yielding specific and clear PCR products were combined to a set of 22 primers, which allowed the amplification of 11 different size-specific PCR products. This set of primers was used for the mPCR and the final concentrations of the primers in the mPCR mix were adjusted in order to achieve similar amplicon quantities. In this optimization step, 5 ng template and 25 amplification cycles were used, because by keeping the template DNA amount and amplification cycle numbers constant, the procedure for optimization of the final mPCR primer concentrations was simplified. In addition, keeping the numbers of cycles low reduced non-specific amplification and would speed up the procedure. The results of the single primer-pair tests that might be used for specific single primer-pair PCRs are depicted in Table S2 and all information about the chosen primers and their final concentrations used in the mPCR are shown in Table 1.
These pre-experiments resulted in a standard setup for the mPCR, which applies 22 primers at different concentrations. The mPCR was performed with GoTaq DNA polymerase in a final reaction volume of 20 µl and 25 amplification cycles. As template, 5 ng of normalized DNA of the different E. granulosus species were used. All reactions yielded a highly specific and clearly distinguishable banding pattern (Figure 1 A and B), allowing the discrimination among E. granulosus s.s. (G1), E. equinus (G4), E. ortleppi (G5), E. canadensis (G6/G7) and E. canadensis (G8/G10). The smallest band (110 bp) was designated to specifically indicate all members of the E. granulosus complex and was clearly present in all 5 species. The upper band (1232 bp) specifically identified the genus Echinococcus (Figure 1B) and detected the E. granulosus complex as well as E. multilocularis and E. vogeli.
The sensitivity of the mPCR was investigated by applying different concentrations of E. granulosus complex template DNA (0.1 ng–1 µg), and the results showed 5–250 ng template DNA are required for a successful detection of all members. When lower or higher amounts of DNA were employed, some PCR products were missing or non-specific amplification occurred. Out of the recommended amounts of template DNA, the detection limits depend largely on the species; E. granulosus s.s. (0.1 ng–1 µg), E. equinus (2.5 ng–250 ng), E. ortleppi (0.5 ng–250 ng), E. canadensis (G6/G7) (1 ng–500 ng) and E. canadensis (G8/G10) (5 ng–250 ng). Thus, in several experiments lower amounts of DNA (0.1–5 ng) were sufficient, but this occurred only when DNA of high quality was used (Figure 3A).
The specificity of the mPCR assay was investigated in four ways: (i) increasing the numbers of PCR cycles; (ii) employing mixed template DNA derived from different Echinococcus genotypes/strains; (iii) applying template DNAs of the closely related genus Taenia; (iv) addition of non-related DNA derived from bovine thymus or dog feces.
Increasing the cycle numbers had an influence on the specificity of the mPCR. In the case where up to 250 ng normalized template DNA was applied, a specific banding pattern was achieved at 25 amplification cycles, but as shown in Figure 2, increased numbers of cycles still allowed genotyping based on the most prominent bands. However, in some genotypes, application of 30 cycles or more resulted in smeary or unspecific amplicons. Thus, for mPCR 25 amplification cycles are recommended.
To test the specificity of the mPCR with mixed template DNA derived from different Echinococcus species, two experiments were performed. First, a DNA cocktail containing 5 ng of normalized DNA from each member of the E. granulosus complex was used as template for the mPCR, and this resulted in a clear and simultaneous expression of all specific amplicons. Additionally, this experiment showed that all specific PCR products could be amplified in parallel, without interference or non-specific amplification (Figure 3B). Since E. granulosus s.s. (G1/G2/G3) and E. canadensis (G6/G7) have been reported to co-exist in several areas, these two species were selected to determine the detection limit of a specific genotype in a dual-genotype DNA mixture. Thus, DNA from E. granulosus s.s. (G1) and E. canadensis (G6) were mixed in different ratios and analyzed by mPCR. When 5 ng of the mixed DNA was used as template, one genotype could be detected when it was present in a concentration of 20% (Fig. 3A, lanes 11 and 15). By using 50 ng template DNA one genotype was detectable in a concentration of 2.5% (Fig. 3A, lane 6) and if 250 ng template DNA were used, the detection of one genotype was possible at a concentration of 5% (Fig. 3A, lane 4). Both experiments showed that two or more genotypes can be detected in parallel by mPCR.
To test the cross-reactivity with closely related Taenia species, mPCRs were performed with 10 ng template DNA of T. saginata, T. solium, T. crassiceps, T. taeniaformis and T. pisiformis. As shown in Figure 1C (lanes 4–8) no products indicative for non-specific primer binding were amplified.
To mimic contaminations occurring during the isolation of DNA from metacestodes or E. multilocularis eggs, 5 ng normalized E. granulosus s.s. (G1) DNA and different amounts of DNA from bovine thymus or canine feces were mixed with at different rations (1∶1–1∶200). As shown in Figure 4, mPCR tolerated a 200-fold excess of foreign DNA (Figure 4).
To test the reliability of the mPCR, 66 unknown samples were genotyped by cox1-sequencing [37] and mPCR in parallel and both methods obtained identical results (Table 2, used samples marked with an asterisk).
The interlaboratory reproducibility of the mPCR was evaluated by genotyping 13 samples in parallel, namely in Berne/Switzerland and Buenos Aires/Argentina, respectively. Both laboratories employed GoTaq DNA polymerase, but otherwise worked independently from each other. Identical results were obtained; seven of the samples contained E. canadensis (G6/G7) and six contained E. granulosus s.s. (G1/G2/G3) isolates (data not shown).
In order to investigate whether the type of DNA polymerase used in mPCR could influence the results, a panel of DNA polymerases derived from different suppliers was tested. The GoTaq polymerase (Promega) originally used for the development of the mPCR yielded optimal results. However, similar results were obtained employing the 5× Multiplex PCR mix from New England Biolabs as well as AmpliTaq DNA Polymerase from Applied Biosystems. Other DNA polymerases failed to provide useful results, leading to non-specific amplicons, smeary products or missing amplification. A list showing the tested DNA polymerases is depicted in Table S1.
The newly established mPCR was applied on previously characterized metacestode DNA, and on metacestode DNA samples of unknown origin. A total of 195 hydatid cysts, 149 isolated from animals and 46 obtained in humans, and all originating from different regions and/or continents, were genotyped by mPCR (for details on the samples, see Table 2). The mPCR amplified the corresponding genotype-specific banding patterns, and in no case unspecific amplicons or mixed genotypes were detected (data not shown). All 46 human CE cases and 135 of the 149 animal CE cases clustered within E. granulosus s.s. (G1/G2/G3). Furthermore, the mPCR detected 7 European E. equinus (G4) cases isolated from 6 horses and 1 donkey from Spain, and 7 pig-derived E. canadensis (G7) cases from South American samples (Table 2).
In this experiment 28 preselected Echinococcus egg DNA samples extracted from dog feces were used: 10 E. granulosus s.s. (G1), 8 E. canadensis (G7) and 10 E. multilocularis samples. Employing mPCR and 150–350 ng template DNA, only 5 out of 10 E. granulosus s.s. (G1) samples, 0 out of 8 E. canadensis (G7) samples, and 4 out of 10 E. multilocularis samples could be positively identified. Increasing the number of amplification cycles up to 35 and/or employing increased amounts of template DNA (up to 1 µg) did not result in any improvement (data not shown).
In order to avoid time-consuming DNA extraction steps, the mPCR was performed directly on hydatid fluid (HF) and protoscoleces (Figure 5). The mPCR failed when these samples were used directly without any pre-treatment. However, heating HF followed by centrifugation and subsequent mPCR with 1–3 µl of the supernatant resulted in amplification of the entire E. equinus (G4) specific banding profile. Inclusion of lower or higher amounts of boiled HF supernatant, or inclusion of fresh, frozen or fixed Echinococcus tissue, did not result in mPCR amplification products (Figure 5A; data not shown). However, preparation of the material employing an alkaline lysis protocol resulted in effective genotyping with frozen and/or EtOH fixed samples, but not with protoscoleces fixed in 4% formaldehyde. When 2 µl of a 1∶8 or 1∶10 dilution of the alkaline lysed supernatant derived from frozen protoscoleces was used for mPCR the whole E. granulosus s.s. (G1) specific banding pattern was detected (Figure 5B). Application of 2 µl of a 1∶2 or a 1∶4 supernatant dilution of EtOH fixed protoscoleces resulted in the detection of a clearly amplified E. granulosus s.s. profile (Figure 5C). Conditions outside of these ranges yielded incomplete or lacking amplification of specific targets. It should be noticed at this point that calcium corpuscles interfere with the PCR. Best results were achieved when calcium corpuscles present at the bottom of the tube after centrifugation of the solid Echinococcus material were not included in the boiling or alkaline lysis steps.
The mPCR developed in this study represents an easy, rapid and inexpensive one-step detection method for the E. granulosus complex. This provides the unique opportunity to address directly speciation and genotyping within the framework of large-scale studies. However, as the E. granulosus complex at the genotypic level may considerably vary from region to region, we propose that routine control programs in a given area do not require the whole set of primers in the final mPCR mix as evaluated in our paper. Thus, a locally adapted primer combination may even render our approach strategy easier.
In the first step of the evaluation process, we defined a standard mPCR setup to minimize variable conditions. This setup enabled 100% specific amplification of targets for all E. granulosus complex members investigated. The setup was also successful when mixed genotypes or contaminating DNA from hosts (dog feces or cattle) were present. Additionally, the presence of closely related species such as other members of the genera Echinococcus or Taenia did not result in false positive amplification products. However, specificity diminished upon introduction of high amounts (>250 ng) of template DNA into the system or, conversely, when very low amounts (<5 ng) of template DNA were applied in combination with more than 25 rounds of PCR amplification. Highly reliable results were provided by using template DNA in the range of 5 to 250 ng. The method of DNA extraction could also have a substantial influence on the mPCR performance, since residual RNA, salt, ethanol or phenol could still be present, biasing DNA concentration measurement. Thus, for a standard genotyping experiment we finally recommend using 20–50 ng of template DNA per reaction, and for the simultaneous detection of different genotypes in one DNA sample up to 250 ng of template DNA should be applied. The same accounts for situations where a high contamination with foreign DNA is expected. In every case it should be taken into account that a minimal amount of specific Echinococcus DNA (approximately 5 ng) is necessary for the amplification. This is especially problematic when DNA extracted from eggs is used, as a single Echinococcus egg contains only approximately 8 pg of nuclear DNA [55], and therefore 600 Echinococcus eggs would be needed for reaching the minimal sensitivity threshold of one mPCR assay. As the worm burden of Echinococcus is highly dispersed in the dog population, the majority of animals are infected with low (<100) numbers of worms, which can result in relatively low egg numbers in the feces. In a study in Lithuania E. canadensis eggs were found in 9 of 240 dogs with egg numbers between 0.25 and 100 eggs per gram feces [52]. Therefore, for epidemiological investigations, the required amount of template DNA for the mPCR might be too high to reach the minimal amount of 5 ng Echinococcus DNA.
Nevertheless, the detection of canine echinococcosis is of essential interest since control programs are based mainly on the anthelmintic treatment of dogs, which interrupts the life cycle of the parasite. A highly sensitive PCR method to discriminate between the E. granulosus complex, E. multilocularis and other Taeniidae in fecal samples was established by Trachsel et al. [47]. This PCR is based on the amplification of mitochondrial genes employing a PCR setup with 40 amplification cycles, and thus low amounts of parasite DNA can be detected. However, genotyping of the E. granulosus complex is not possible with this approach. In comparison, the mPCR developed in this study could only detect 32% of those fecal samples that had tested positive by the PCR developed by Trachsel et al. [47]. Since those fecal samples had been pre-selected as Echinococcus PCR-positive from previous studies [35], [53], the real sensitivity of the mPCR might be even lower. Increasing template DNA concentration and increasing the numbers of PCR cycles did not result in improved sensitivity (data not shown). One possibility to apply the mPCR for the genotyping of canine derived samples would be the use of DNA extracted either from adult worms, isolated after necropsy or from purged dogs. With this approach 3 genotype groups (G1, G4 and G6/G7) were identified in dogs in Kyrgyzstan [35]. Another possibility would be the optimization of the mPCR protocol supporting the very low target amounts in a foreign DNA background, for example by using higher primer concentrations in a 30–40 amplification cycle setup.
So far, methods for genotyping E. granulosus complex members have been based on extracted DNA derived from protoscoleces (fertile cysts) or germinal layers (infertile cysts). In contrast to these methods, the mPCR protocol described here allows also genotyping without the need for DNA extraction steps, provided the material is frozen or fixed in ethanol. For solid materials such as protoscoleces, a direct testing, or testing upon pre-boiled treatment was not successful, but a pre-alkaline lysis step was sufficient for tissue dissolution and release of genomic DNA into the supernatant. Hydatid fluid of the metacestodes contains secretory parasite proteins and other metabolites derived from the germinal layer, and may also contain released and live germinal layer cells, and/or DNA derived from degraded cells. Thus, minimal amounts of boiled hydatid fluid can be used directly for mPCR-based genotyping. While this is possible with clear hydatid fluid, problems could occur in cases where the fluid is bacterially infected. In addition, application of excessive amounts of hydatid fluid or undiluted supernatants from boiled or alkaline-lysed E. granulosus might result in missing amplicons and therefore in imperfect genotyping. Nevertheless, compared to standard DNA extractions, both methods are fast, simple to perform and inexpensive. In our opinion, the most interesting finding was that the mPCR could be applied reliably with minimal amounts of boiled HF.
Another variable parameter of mPCR performance concerns the DNA polymerases. We optimized the protocol for GoTaq from Promega, but also Amplitaq (Applied Biosystems) or the Multiplex PCR 5× Mastermix (New England Biolabs) rendered good results, while the use of many other polymerases resulted in poor performance. In cases where other polymerases are used, the described protocols may have to be optimized.
In the explorative epidemiological application of our mPCR, a large amount of field samples obtained from different collaborating groups were investigated (Table 2). For all previously characterized isolates, the genotypes could be successfully confirmed by mPCR, including 20 samples from Bulgaria [51] and 22 samples from Spain (unpublished). All 176 samples derived from North African countries (Algeria and Tunisia; human and animal cases) were genotyped as E. granulosus s.s. by mPCR, and the 39 Tunisian human samples were additionally confirmed by cox1 sequencing. In experiments carried out independently in two distinct laboratories, 13 Argentinean samples were genotyped by mPCR in Buenos Aires/Argentina and in Berne/Switzerland, and all results were comparable. These samples were additionally confirmed by cox1 sequencing. Taken together 195 samples were genotyped, or the known genotype was confirmed by mPCR in this study. For all samples a clear genotype-specific banding pattern was observed, thus demonstrating the high accuracy of the E. granulosus complex mPCR. Compared to other genotyping methods (PCR-RFLP, sequencing or other approaches [56]–[59]) the mPCR resulted in similar findings, but results were obtained employing a rapid one-tube assay. Chromosomal DNA was used in this test-approach, but by applying hydatid fluid or cellular Echinococcus material as templates for the mPCR, the speed, price and hands-on-time for genotyping the E. granulosus complex can be further decreased. The relatively complicated task of E. granulosus complex speciation and genotyping is clearly simplified by mPCR, and therefore this method represents a useful tool for future routine practice.
In conclusion, the mPCR described herein represents a robust and reliable technique to characterize (i) any E. granulosus complex derived sample at the genus level, (ii) the membership within the E. granulosus complex and (iii) the species/genotype level, all in a single tube. Within the last two years, more than thirty studies addressed the question of genotyping of E. granulosus isolates around the world. This demonstrates the importance of the epidemiology of Echinococcosis, and the mPCR can contribute to a better understanding of the spatio-temporal circulation of this complex.
A) The primers Echi Rpb2 F and Echi Rpb2 R used for the detection of all Echinococcus species were designed using the Echinococcus gene RNA polymerase II (rpb2): E. granulosus s.s. (G1/G2/G3) - FN566850.1, E. equinus (G4) - FN566851.1, E. ortleppi (G5) - FN566852.1, E. canadensis (G6) - FN566853.1, E. canadensis (G7) - FN566854.1, E. canadensis (G8) - FN566855.1, E. oligarthrus - FN658827.1, E. vogeli - FN566847.1, E. multilocularis - FN566845.1. B) The complete mitochondrial genome sequence was used to design the E. granulosus complex specific primers E.g. complex F and E.g. complex R (gene marker: cox2), the E. ortleppi (G5) specific primers E. ortp ATP6 F and E. ortp ATP6 R (gene marker: atp-6) as well as E. ortp CoxI F and E. ortp CoxI R (gene marker: cox1) and the E. canadensis (G6/7) specific primers E.cnd G6/G7 NDI F and E.cnd G6/G7 NDI R (gene marker: nad1): E. granulosus s.s. (G1/G2/G3) - AF297617.1, E. equinus (G4) - AF346403.1, E. ortleppi (G5) - AF235846.1, E. canadensis (G6) - AB208063.1, E. canadensis (G7) - AB235847.1, E. canadensis (G8) - AB235848.1. C) The ezrin-radixin-moesin-like protein (elp1) was used to design the E. canadensis (G8/G10) specific primers E.cnd G8/G10 F and E.cnd G8/G10 R: E. granulosus s.s. (G1/G2/G3) - EU834886.1, E. equinus (G4) - EU834891.1, E. ortleppi (G5) - FN582298.1, E. canadensis (G6/G7) - EU834893.1, E. canadensis (G8) - EU834894.1, E. canadensis (G10) - EU834896.1. D) The DNA polymerase delta (pold) gene was used to design the E. canadensis (G6/7) specific primers E.cnd G6/G7 pold F and E.cnd G6/G7 pold R: E. granulosus s.s. (G1) - FN568361.1, E. equinus (G4) - FN568362.1, E. ortleppi (G5) - FN568363.1, E. canadensis (G6) - FN568364.1, E. canadensis (G7) - FN568365.1, E. canadensis (G8) - FN568366.1. E) The calreticulin (cal) gene was used to design the E. granulosus s.s. (G1/G2/G3) specific primers E.g ss cal F and E.g ss cal R as well as the E. equinus specific primers E.eq cal F and E.eq cal R: E. granulosus s.s. (G1) - EU834931.1, E. equinus (G4) - EU834936.1, E. canadensis (G6/G7) - EU834937.1, E. canadensis (G8) - EU834939.1, E. canadensis (G10) - EU834940.1. F) The elongation factor 1 alpha (ef1a) gene was used to design the E. granulosus s.s. (G1/G2/G3) specific primers E.g ss Ef1a F and E.g ss Ef1a R: E. granulosus s.s. (G1) - FN568380.1, E. equinus (G4) - FN568381.1, E. ortleppi (G5) - FN568382.1, E. canadensis (G6) - FN568384.1, E. canadensis (G7) - FN568383.1, E. canadensis (G8) - FN568385.1. G) The cytochrome oxidase subunit I (cox1) gene was used to design the E. equinus (G4) specific primers E.eq cox1 F and E.eq cox1 R: E. granulosus s.s. (G1/G2/G3) - M84661.1, E. equinus (G4) - M84664.1, E. ortleppi (G5) - M84665.1, E. canadensis (G6) - M84666.1, E. canadensis (G8) - DQ144021.1, E. canadensis (G10) - DQ144022.1.
|
10.1371/journal.ppat.1004375 | Helminth Infections Coincident with Active Pulmonary Tuberculosis Inhibit Mono- and Multifunctional CD4+ and CD8+ T Cell Responses in a Process Dependent on IL-10 | Tissue invasive helminth infections and tuberculosis (TB) are co-endemic in many parts of the world and can trigger immune responses that might antagonize each other. We have previously shown that helminth infections modulate the Th1 and Th17 responses to mycobacterial-antigens in latent TB. To determine whether helminth infections modulate antigen-specific and non-specific immune responses in active pulmonary TB, we examined CD4+ and CD8+ T cell responses as well as the systemic (plasma) cytokine levels in individuals with pulmonary TB with or without two distinct helminth infections—Wuchereria bancrofti and Strongyloides stercoralis infection. By analyzing the frequencies of Th1 and Th17 CD4+ and CD8+ T cells and their component subsets (including multifunctional cells), we report a significant diminution in the mycobacterial–specific frequencies of mono- and multi–functional CD4+ Th1 and (to a lesser extent) Th17 cells when concomitant filarial or Strongyloides infection occurs. The impairment in CD4+ and CD8+ T cell cytokine responses was antigen-specific as polyclonal activated T cell frequencies were equivalent irrespective of helminth infection status. This diminution in T cell responses was also reflected in diminished circulating levels of Th1 (IFN-γ, TNF-α and IL-2)- and Th17 (IL-17A and IL-17F)-associated cytokines. Finally, we demonstrate that for the filarial co-infections at least, this diminished frequency of multifunctional CD4+ T cell responses was partially dependent on IL-10 as IL-10 blockade significantly increased the frequencies of CD4+ Th1 cells. Thus, co-existent helminth infection is associated with an IL-10 mediated (for filarial infection) profound inhibition of antigen-specific CD4+ T cell responses as well as protective systemic cytokine responses in active pulmonary TB.
| While it has long been recognized that helminth infections alter the pathophysiology of allergic and autoimmune disease, data suggest that helminth infections also exert an important immunological effect on concomitant infections and vaccine responses. In particular, helminth coinfection can modulate the severity, pathogenesis and transmission of other infectious diseases. In this study, we examine the mechanism by which helminth infections modulate the immunological responses to tuberculosis antigens in individuals with active pulmonary tuberculosis. Our data suggest that two different helminth infections, with different life cycles, tissue localization and modes of transmission essentially exert very similar effects on the adaptive immune response to tuberculosis antigens in pulmonary tuberculosis. This includes a compromised induction of protective cytokine-expressing T cells as well as inhibitory effects on systemic cytokines that are potentially protective in tuberculosis. The strength of this study lies in the fact that this is the first study to demonstrate that two different helminth infections essentially impair cytokine responses in a similar manner in pulmonary tuberculosis.
| Helminth parasites are complex eukaryotic organisms, characterized by their ability to maintain long-standing infections in humans, sometimes lasting decades. Two of the most common persistent helminth infections are Wuchereria bancrofti, the major causative agent of lymphatic filariasis, and Strongyloides stercoralis, the causative agent of strongyloidiasis together infecting close to 250 million people worldwide [1], [2]. In addition, both these infections are often clinically asymptomatic due, in large part, to the parasites' ability to manipulate the host immune system, a feature that insures their survival largely because of their ability to restrict local inflammatory pathology [3], [4]. Modulation of the host immune response involves a variety of strategies including the induction of regulatory networks that leads to dysregulation of innate and adaptive immune responses [3], [4]. The immune down modulation associated with helminth infections is primarily parasite-antigen specific, but some bystander effects on parenterally administered vaccine responses, allergen skin test positivity, non-helminth pathogen-specific immune responses and autoimmune diseases have been noted [5], [6], [7]. In terms of interaction in human TB, filarial infections have been shown to alter the antigen - specific protective immune responses in latent TB by modulating the Th1 and Th17 responses to TB antigens [8]. In addition, Strongyloides has been shown to alter the protective Th17 cytokine responses in animal models of co-infection [9]. Finally, helminth infections are strongly associated with an IL-10 dominant regulatory environment that could potentially down modulate antigen - specific responses to third party antigens [10].
Active TB reflects the progression from latent TB to active symptomatic disease that is usually attributed to failure to contain Mtb within a granuloma. However, it is well established that the control of TB infection is dependent on Th1 (IL-12, IFN-γ and TNF-α) and, to a lesser extent, Th17 (IL-17 and IL-23) responses [11]. Both Th1 and Th17 responses have been shown to be important in the induction and maintenance of protective immune responses in mouse models of TB infection or for control of human TB infection (as seen in latent TB) [12], [13], [14]. The presence of multifunctional T cells (expressing more than one cytokine) has also been shown to be an important correlate of protective immunity to a wide variety of pathogens, including TB [15]. Multifunctional CD4+ Th1 cells, co-expressing IFN-γ/TNF-α/IL-2 or IFN-γ/IL-2 or IFN-γ/TNF-α have been shown to be associated with protection against active pulmonary disease in TB [16], [17], [18]. In addition, the absence or reduced frequency of multifunctional Th1 cells is thought to correlate with the severity of TB disease [19]. Th17 cells have also been shown to play a role in protection against TB infection as well as in the induction of memory responses in animal models [12]. However, the role of multifunctional Th17 cells, if any, in active human pulmonary TB remains unexplored. Finally, Type 1 and Type 17 cytokine production by CD8+ T cells is also thought to play an important role in protection against TB infection/disease [12].
Helminth infections commonly occur throughout the tropics and subtropics and in many regions of the world have an overlapping geographic distribution with Mycobacterium tuberculosis (Mtb) [6]. Moreover, age-specific prevalence studies have indicated that helminth infections usually precede the acquisition of pulmonary tuberculosis [20]. Finally, both filarial parasites (present in the circulation) and Stronglyoides (which is an intestinal helminth but has a lung migratory larval stage) could directly influence the outcome of TB infection. We therefore hypothesized that immune responses in active TB might be modulated by the regulatory immune networks often seen in chronic helminth infections that could have a negatively impact on the course of active TB. To this end, we examined CD4+ and CD8+ Th1 and Th17 responses in patients with active TB with or without concomitant filarial or Stronglyloides infection. Our data suggest that coincidental helminth infection has a profound inhibitory effect on multi - functional Th1 and Th17 responses as well as on systemic cytokine responses in active pulmonary TB. Our data also suggest that IL-10 is an important mediator of these inhibitory effects for filarial co-infections.
To determine the impact of helminth infection on the hematological and immunological parameters of active TB individuals at baseline (or steady state), we performed hematological and flow cytometry analysis on these individuals. As shown in Table 1, infection with W. bancrofti or S. stercoralis in the context of active pulmonary TB was not associated with significant alterations in the absolute numbers of CD4+ and CD8+ T cells nor in the frequency distribution of the various T cell subsets - naive, central memory, effector memory and regulatory T cells - when compared to helminth-uninfected individuals with active TB. Similarly, all other hematological and immunological parameters examined including total leukocyte and differential cell counts were similar between those helminth-infected and –uninfected individuals with active TB.
Since a decrease in multifunctional CD4+ Th1 cells is known to be associated with increased bacterial burdens in active TB [19] and since both mono - and multifunctional CD4+ Th1 cells are potential correlates of protective immunity in TB [11], we sought to determine the impact of helminth infection on both the mono-functional and multifunctional CD4+ Th1 responses in TB infected individuals. To this end, we cultured whole blood from FIL/TB, STR/TB and TB only individuals with media alone, CFP-10, ESAT-6 and anti-CD3 and measured the frequency of CD4+ T cells expressing each of the Th1-associated cytokines (Figure 1A). As shown in Figure 1B, co-incidental filarial infection was associated with significantly lower frequencies of CD4+ T cells expressing IL-2 alone or co – expressing TNF-α/IFN-γ or IL-2/IFN-γ at baseline. Similarly, Strongyloides co-infection was associated with decreased frequencies of CD4+ T cells co-expressing TNF-α/IFN-γ or IL-2/IFN-γ/TNF-α at baseline in comparison to individuals with active TB only. In addition, as shown in Figures 1C and D, co-incidental filarial infection was associated with significantly lower frequencies of CFP-10 and ESAT-6 induced net frequencies of CD4+ T cells expressing IL-2 or IFN-γ or TNF-α alone or co-expressing TNF-α/IFN-γ or IL-2/IFN-γ or IL-2/TNF-α or IL-2/IFN-γ/TNF-α in comparison to individuals with active TB only. Similarly, Stronglyloides co-infection was also associated with significantly decreased frequencies of almost all of the above-mentioned mono - and multifunctional CD4+ Th1 cell subsets in response to CFP-10 and ESAT-6 (Figures 1C and D). Finally, no significant differences in the net frequency of CD4+ Th1 cells was observed between the helminth-infected and -uninfected groups following anti-CD3 stimulation, with the exception of CD4+ T cells expressing IL-2 alone in FIL/TB individuals (Figure 1E). Thus, helminth infections are associated with a down modulation of spontaneous and/or antigen - specific mono - and multifunctional Th1 responses in active TB.
Since both mono - and multifunctional CD4+ Th17 cells have also been implicated as being important in the immune response in active TB [11], we sought to determine the impact of helminth infection on the CD4+ Th17 responses in TB infected individuals. To this end, we cultured whole blood from co-infected (FIL/TB or STR/TB) and TB only (TB) individuals with media alone, CFP-10, ESAT-6 and anti-CD3 and measured the frequency of CD4+ T cells expressing each of the Th17-associated cytokines (Figure 2A). As shown in Figure 2B, at baseline, the frequencies of CD4+ T cells expressing IL-22 or co – expressing IL-17A/IFN-γ or IL-17A/IL-17F or IL-17A/IL-22 was significantly reduced in FIL/TB compared to TB alone individuals. In addition, as shown in Figures 2C and D, upon Mtb-specific antigen stimulation, the frequencies of CD4+ T cells expressing IL-17A or IL-22 or co – expressing IL-17A/IFN-γ or IL-17A/IL-17F was significantly reduced in FIL/TB compared to TB alone individuals. Moreover, similar to the pattern observed in Th1 cells, the differential frequencies of Th17 cells was also mycobacterial - antigen specific since anti-CD3 stimulated frequencies of these cells did not exhibit any major significant differences (Fig. 2E). In contrast, STR/TB individuals did not exhibit any significant differences in the frequencies of mono - or multifunctional CD4+ Th17 cells in comparison to TB alone individuals ex vivo or following stimulation with TB antigens or anti-CD3. Thus, helminth infections, and more specifically filarial infections, are associated with a modulation of spontaneous or antigen - specific Th17 responses in active TB.
Since CD8+ T cells play an important role in protection against TB [21], we sought to determine the impact of helminth infection on the CD8+ Th1 and Th17 cytokine responses in TB infected individuals. To this end, we cultured whole blood from co-infected (FIL/TB or STR/TB) and TB only individuals with media alone, CFP-10, ESAT-6 and anti-CD3 and measured the frequency of CD8+ T cells expressing each of the Th1 and Th17-associated cytokines. As shown in Fig. 3A, FIL/TB individuals exhibited significantly lower frequencies of CD8+ T cells expressing IL-2 or TNF-α or IL-17A in comparison to TB alone individuals ex vivo. Similarly, STR/TB individuals also exhibited significantly decreased frequencies of CD8+ T cells expressing IL-17A or IL-17 or IL-22 ex vivo (Figure 3A). In addition, FIL/TB individuals exhibited significantly lower frequencies of CD8+ T cells expressing IL-2 or INF-γ or TNF-α or IL-17A or IL-17F in comparison to TB only individuals upon CFP-10 and ESAT-6 stimulation (Figures 3B and C). Similarly, as shown in Figures 3B and C, STR/TB individuals also exhibited significantly lower frequencies of CD8+ T cells expressing INF-γ or TNF-α or IL-17A or IL-17F in comparison to TB only individuals upon CFP-10 and ESAT-6 stimulation. In contrast, the frequencies of CD8+ T cells expressing Th1 and Th17 cytokines was not significantly different between the 3 groups upon anti-CD3 stimulation. Thus, helminth infections are also associated with a down modulation of CD8+ T cell responses in active TB.
Since Th1 and Th17 cytokines are cytokines important components of the immune response in active TB [11], we wanted to explore the effect of coincident helminth infection on systemic levels of these cytokines. To determine the impact of helminth infections on the circulating levels of the prototypical Th1 and Th17 cytokines as well as regulatory cytokines, we measured the levels of IFN-γ, TNF-α, IL-2, IL-17A, IL-17F,IL-22, IL-10 and TGFβ in the plasma of three groups of individuals with active TB - FIL/TB, STR/TB or TB alone. As shown in Figure 4, we observed significantly lower plasma levels of Th1 associated cytokines - IFN-γ (Geometric Mean of 936.6 pg/ml in TB alone vs. 59.2 pg/ml in Fi/TB and 60.5 pg/ml in STR/TB), IL-2 (GM of 27.6 pg/ml in TB alone vs. 11.7 pg/ml in FIL/TB and 13.7 pg/ml in STR/TB) and TNF-α (GM of 1017 pg/ml in TB alone vs. 493.1 pg/ml in FIL/TB and 202.7 pg/ml in STR/TB) as well as Th17 - associated cytokines - IL-17A (GM of 219.2 pg/ml in TB alone vs. 84.1 pg/ml in FIL/TB and 90.6 pg/ml in STR/TB) and IL-17F (GM of 110.2 pg/ml in TB alone vs. 59.9 pg/ml in FIL/TB and 73.6 pg/ml in STR/TB) but not IL-22 in co-infected individuals compared to TB infected individuals. In contrast, we observed significantly higher plasma levels of IL-10 (GM of 116.9 pg/ml in TB alone vs. 209.7 pg/ml in FIL/TB and 177.7 pg/ml in STR/TB) but not TGFβ (data not shown) in helminth co-infected individuals compared to TB-infected individuals. Thus, both helminth infections are associated with profound alterations systemic levels of Th1 and Th17 cytokines in co-infected individuals.
To determine the role of IL-10 and other known immunomodulatory cytokines (e.g.TGFβ in the modulation of CD4+ Th1 cells in active TB with concomitant helminth infection, we measured the frequency of cells following stimulation with the TB antigen -CFP-10 in the presence or absence of anti-IL-10 or anti-TGFβ neutralizing antibody in FIL/TB and TB alone individuals (n = 10). As shown in Figure 5A, IL-10 neutralization resulted in significantly increased frequencies of monofunctional (IL-2 or INF-γ or TNF-α expressing) and multifunctional (IL-2/IFN-γ or IFN-γ/TNF-α or IL-2/TNF-α co-expressing) Th1 cells in FIL/TB individuals. In marked contrast, as shown in Figure 5B, TGFβ neutralization had no significant effect on the frequencies of mono- or multi - functional Th1 cells. On the other hand, IL-10 neutralization resulted in significantly increased frequencies of monofunctional (IL-2 or INF-γ or TNF-α expressing) but not multifunctional (IL-2/IFN-γ or IFN-γ/TNF-α or IL-2/TNF-α co-expressing) Th1 cells in TB alone infected individuals (Figure 5C). Thus, IL-10 plays an important role in the modulation of CD4+ Th1 cells in FIL/TB co-infection.
Helminth infections afflict over 1.5 billion people worldwide, while Mtb infects one third of the world's population resulting in a million deaths per year [6]. The overlapping geographic distributions of the helminth infections and tuberculosis demonstrate very clearly that, on a population level, the potential for interaction among these various pathogens can occur. A wide variety of studies have been performed to examine the possible effect of helminth infection on the induction of a protective immune response against mycobacteria [22], [23]. Both intestinal and systemic helminths have been shown to modulate proliferation and IFN-γ production in response to PPD in helminth – latent TB coinfected individuals [22], [23]. Some of these effects have been shown to be reversible following anthelmintic chemotherapy [22]. Indeed, we have previously demonstrated that concurrent filarial infection could inhibit the generation of potentially protective Th1 and Th17 immune responses in latent TB infected individuals [8]. In addition, we have also shown that concomitant hookworm infection modulates the frequency of Th1 and Th17 cytokine-producing cells in latent TB [24]. The immunogenicity of BCG vaccination has been shown to be impaired in helminth-infected individuals, and this is associated with enhanced TGF-β production but not enhanced Th2 responses [25], while there exists an inverse association between BCG immunization and intestinal nematode infection [26]. Despite these studies on the interaction of helminth infection and latent TB or TB vaccination, the relationship of helminth infection on the development of active tuberculosis or outcome following treatment is not completely clear.
The two major subsets of CD4+ T cells that form an important component of adaptive immune responses to TB are Th1 and Th17 cells [11], [12]. Th1 responses are known to be important in resistance to TB, while Th17 responses are known to be important in inducing and maintaining memory and recall responses to TB [12]. Finally, multifunctional Th1 cells are also thought to play an important role in protection against TB disease [27]. Because immune-mediated protection against Mtb is characterized by strong mycobacterium-specific Th1 and Th17 responses [12], it has been postulated that coincident infections with helminth parasites could modulate these immune responses by driving Th2 and/or Tregs that induce anti-inflammatory responses [6]. Therefore, we have examined the effect of helminth infection on TB - antigen specific immune responses in individuals with active microbiologically confirmed pulmonary TB.
Our data reveal significant alterations in the baseline frequencies of mono - and multifunctional CD4+ and CD8+ Th1 and Th17 cells in TB-infected individuals with active helminth infection. This is associated with perturbations in the homeostatic or steady - state levels of Th1 and Th17 cytokines in pulmonary TB individuals in comparison to co-infected individuals as well. Our examination of plasma levels of these cytokines clearly reveals that a profound depression of both Th1 and Th17 cytokines is found in those with helminth infection and active TB. Amongst all the cytokines, IFN-γ and TNF-α are known to be critically responsible for protection against TB [11]. Therefore, the diminished circulating levels of these cytokines in helminth co-infected individuals, suggest an impairment in Th1 responses in pulmonary TB with coincident filarial infection. In addition, the diminished systemic production of IL-2, IL-17A and IL-17F also indicate a more extensive impairment in Th1 and Th17 responses in co-infection settings. Thus, helminth infection appears to be associated with homeostatic alterations in the Th1 and Th17 cellular responses in pulmonary TB.
Our study highlights the association of filarial co-infection with a profound impairment in TB - antigen specific CD4+ Th1 and Th17 responses. Our data on STR/TB co-infection also reveals remarkably similar yet more pronounced effects of helminth infection on CD4+ T cell responses in active TB. Thus, co-infected individuals exhibit a spontaneous deficiency in the frequencies of Th1 and Th17 cells and a much more potent deficiency in the expansion of mono - and multifunctional Th1 and Th17 cells in response to Mtb-specific antigens. In contrast, our data suggest that the intrinsic potential of CD4+ T cells to respond to polyclonal stimulation and induce Th1 and Th17 cytokine expression is unaltered in the presence of coincident helminth infection. CD4+ T cells expressing IL-2 alone or those co-expressing IL-2 and IFN-γ or TNF-α and IFN-γ have been show to be potential correlates of protective immunity to Mtb [18], [28]. Similarly, multifunctional CD4+ T cells co-expressing IFN-γ, TNF-α and IL-2 have also been shown to correlate with immunity to Mtb in a study comparing smear-positive TB to those with smear-negative TB or latent TB [16]. Thus, mono – and multifunctional Th1 cells clearly play an important role in susceptibility or resistance to infection and/or disease. In addition, Th17 cells, secreting one or more cytokines, are also known to play an important role in protective memory responses in TB infection [29]. Since multifunctional T cell responses are known to be better correlates of protective immunity and also to be more persistent [30], the impairment of multifunctional CD4+ Th1 and Th17 cells could potentially have an impact on the clinical course of TB disease in co-infected individuals.
Although a role for CD4+ T cells in protection against Mtb is well established, there is also a large body of evidence derived from both humans and animal models that suggest an essential role for CD8+ T cells [12], [21] as well. CD8+ T cells are known producers of Th1 and Th17 associated cytokines and also possess direct antimicrobial activity through granule-exocytosis dependent mechanisms [12]. Since helminth infections can also modulate CD8+ T cell responses, we examined the effect of co-incidental helminth infection on CD8+ T cell cytokine responses in active TB. Similar to the effect on CD4+ T cells, helminth infections appear to exhibit a profound inhibitory effect on the expression of Th1 and Th17 associated cytokines in the context of pulmonary TB. Alterations in cytokine producing CD4+ and CD8+ T cell subsets could be the result of altered T cell numbers at baseline. Our data suggest that helminth infections are not associated with any such alterations (see Table 1). Moreover, our data also clearly reveal that helminth infections have very little effect on the naive and memory compartmentalization of CD4+ and CD8+ T cells in active TB. In addition, while the induction of natural Tregs by filarial infections is a major mechanism by which these infections could dampen host immune responses [31], our data also clearly indicate no significant difference in the frequency of nTregs between helminth - infected and uninfected individuals, suggesting that nTreg expansion might not play an important role in modulation of the T cell subsets observed in the present study.
The other major mechanism by which helminth infections are known to alter immune responses to bystander antigens is by the production of immuno-modulatory cytokines - IL-10 and TGFβ [3]. Indeed, filarial infections are known to be associated with an IL-10 dominant cytokine milieu [31]. Moreover, helminth infections were associated with elevated circulating levels of IL-10 in the co-infected individuals, implicating a potential regulatory role for IL-10 in co-infections. Our data on the role of IL-10 and TGFβ in the helminth infection associated modulation of CD4+ Th1 responses implicate IL-10 as the major player in the down modulation of Th1 responses in active TB, at least in the context of filarial infections. Moreover, our data also reveal an important role for IL-10 in the down regulation of both mono - and multifunctional Th1 cells in this setting. Interestingly, TGFβ appeared to have a negligible effect on the modulation of the Th1 response to TB antigen, although an effect on Th17 responses or CD8+ T cell responses cannot be excluded. In addition, while IL-10 also appears to play an important role in down modulation of Th1 responses in active TB individuals without helminth infection, this effect appears to be selective to mono- functional Th1 cells only. In contrast, filarial infection modulated effector CD4+ T cell responses encompass both mono- and multi-functional Th1 cells. Our data, therefore, suggest a major role for IL-10 in the regulation of immune responses of active TB.
Our findings suggest that in the presence of coincident helminth infection, the ability to restore homeostatic CD4+ and CD8+ T cell responses in active disease could be worsened. Our study did not have the sample size required to assess the impact of helminth infection on severity of disease or bacterial burdens but the immunological correlates nevertheless highlight a potentially deleterious effect of filarial infection on active TB. In addition, the major strength of our study is the finding that two different helminth infections, with different modes of transmission as well as localization, are both associated with down modulation systemic and antigen - specific immune responses in active TB. Our findings, therefore, have significant implications for treatment and vaccine discovery in TB and suggest that treatment of concomitant helminth infections could have an impact on both the clinical course of TB as well as on vaccine studies in TB-endemic areas.
All individuals were examined as part of a clinical research protocol approved by Institutional Review Board of the National Institute for Research in Tuberculosis, and informed written consent was obtained from all participants.
We studied a group of 50 individuals with active pulmonary TB, 17 of whom were infected with W. bancrofti (hereafter FIL/TB) and 13 of whom had S. stercoralis (hereafter STR/TB) infection in Tamil Nadu, South India (Table 1). Another set of 10 individuals with active pulmonary TB and coincident filarial infection and 10 individuals with active TB alone were used for cytokine neutralization experiments. Active pulmonary TB was diagnosed microbiologically on the basis of being at least culture positive for Mtb by solid cultures in LJ medium (some were also sputum smear positive). Filarial infection was diagnosed by the presence of circulating filarial antigen by the TropBio Og4C3 enzyme-linked immunosorbent assay (ELISA) (Trop Bio Pty. Ltd, Townsville, Queensland, Australia). Strongyloides infection was diagnosed by the presence of IgG antibodies to the 31-kDa recombinant NIE antigen by the Luciferase Immunoprecipitation System Assay, as described previously [32]. All individuals were HIV negative and anti-tuberculous and anthelmintic treatment naive. The two groups of individuals did not differ significantly in the radiological extent of disease or bacillary burden (as estimated by smear grades).
Leukocyte counts and differentials were performed on all individuals using the Act-5 Diff hematology analyzer (Beckman Coulter).
Flow cytometry acquisition was done on BD FACS Canto II (BD Biosciences, San José, CA, USA). Analysis was done using FlowJo software v9.4.10 (TreeStar Inc., Ashland, OR, USA).
Absolute CD4+T cell counts were enumerated in whole blood using BD Multiset 6-Color TBNK cocktail (BD Biosciences). Naïve and memory T cell phenotyping was performed using FITC-CD45RA (BD Pharmingen, BD Biosciences) and APC-CCR7 (eBioscience, San Diego, CA, USA) staining in CD4+ and CD8+ T cells. Naïve cells were classified as CD45RA+CCR7+, effector memory cells as CD45RA−CCR7−, and central memory cells as CD45RA−CCR7+. Natural Tregs (nTregs) were classified as CD4+CD25+Foxp3+CD127dim (BD Pharmingen and eBioscience).
Mycobacterial antigens— recombinant early secreted antigen-6 (ESAT-6) and culture filtrate protein-10 (CFP-10) (Fitzgerald Industries Intl. Inc, Acton, MA)— were used as the antigenic stimuli. These antigens contain epitopes reactive to both CD4+ and CD8+ T cells [33]. Final concentrations were 10 µg/ml for ESAT-6 and CFP-10. Anti-CD3 at a concentration of 10 µg/ml was used as the positive control stimuli.
In vitro cultures and subsequent intracellular cytokine staining was performed. Whole blood cell cultures were performed to determine the intracellular levels of cytokines. Briefly, whole blood was diluted 1∶1 with RPMI-1640 medium supplemented with penicillin/streptomycin (100 U/100 mg/ml), L-glutamine (2 mM), and HEPES (10 mM) (all from Invitrogen, San Diego, CA) and distributed in 12-well tissue culture plates (Costar, Corning Inc., Corning, NY). The cultures were then stimulated with ESAT-6 or CFP-10 or anti-CD3 or media alone in the presence of the costimulatory molecules CD49d/CD28 at 37°C for 6 h. FastImmune Brefeldin A solution (10 µg/ml) was added after 2 h. After 6 h, centrifugation, washing, and red blood cell lysis were performed. Cells were fixed using cytofix/cytoperm buffer (BD Biosciences) and cryopreserved at −80°C. For cytokine neutralization experiments, whole blood from individuals with filariasis and active TB or active TB alone (n = 10) was cultured in the presence of anti-IL-10 (5 µg/ml) or anti-TGFβ (5 µg/ml) or isotype control antibody (5 µg/ml) (R& D Sytems) for 6 h following which CFP-10 and brefeldin A was added and cultured for a further 12 h.
The cryopreserved cells were thawed, washed, and then stained with surface antibodies for 30–60 min. Surface antibodies used were CD3 (Amcyan), CD4 (APC-H7), and CD8 (PE-Cy7). The cells were washed and permeabilized with BD Perm/Wash buffer (BD Biosciences) and stained with intracellular cytokines for an additional 30 min before washing and acquisition. Cytokine antibodies used were IFN-γ, TNF-α, IL-2, IL-17A, IL-17F, and IL-22. Eight-color flow cytometry was performed on a FACSCanto II flow cytometer with FACSDiva software v.6 (Becton Dickinson and Company, Cockeysville, MD). Lymphocyte gating was set by forward and side scatter, and 100,000 lymphocyte events were acquired. Gating for CD4+ T cells expressing cytokines was determined by FMO (fluorescence minus one). Data were collected and analyzed using Flow Jo software (TreeStar Inc., Ashland, OR). All data are depicted as frequency of CD4+ T cells expressing cytokine(s). Baseline values following media stimulation are depicted as baseline frequency, while frequencies following stimulation with antigens or PMA/ionomycin are depicted as net frequencies (with baseline values subtracted).
Plasma cytokines on all 50 individuals were measured using Bioplex multiplex cytokine assay system (Biorad). The cytokines analyzed were IL-2, IFN-γ, TNF-α, IL-10, IL-17A, IL-17F and IL-22. TGFβ levels were measured using a standard ELISA kit from R&D Systems.
Data analyses were performed using GraphPad PRISM (GraphPad Software, Inc., San Diego, CA, USA). Geometric means (GM) were used for measurements of central tendency. Comparisons were made using either the Kruskal-Wallis test with Dunn's multiple comparisons (unpaired comparisons) or the Wilcoxon signed rank test (paired comparisons).
|
10.1371/journal.pgen.1006111 | Genetic Drift, Purifying Selection and Vector Genotype Shape Dengue Virus Intra-host Genetic Diversity in Mosquitoes | Due to their error-prone replication, RNA viruses typically exist as a diverse population of closely related genomes, which is considered critical for their fitness and adaptive potential. Intra-host demographic fluctuations that stochastically reduce the effective size of viral populations are a challenge to maintaining genetic diversity during systemic host infection. Arthropod-borne viruses (arboviruses) traverse several anatomical barriers during infection of their arthropod vectors that are believed to impose population bottlenecks. These anatomical barriers have been associated with both maintenance of arboviral genetic diversity and alteration of the variant repertoire. Whether these patterns result from stochastic sampling (genetic drift) rather than natural selection, and/or from the influence of vector genetic heterogeneity has not been elucidated. Here, we used deep sequencing of full-length viral genomes to monitor the intra-host evolution of a wild-type dengue virus isolate during infection of several mosquito genetic backgrounds. We estimated a bottleneck size ranging from 5 to 42 founding viral genomes at initial midgut infection, irrespective of mosquito genotype, resulting in stochastic reshuffling of the variant repertoire. The observed level of genetic diversity increased following initial midgut infection but significantly differed between mosquito genetic backgrounds despite a similar initial bottleneck size. Natural selection was predominantly negative (purifying) during viral population expansion. Taken together, our results indicate that dengue virus intra-host genetic diversity in the mosquito vector is shaped by genetic drift and purifying selection, and point to a novel role for vector genetic factors in the genetic breadth of virus populations during infection. Identifying the evolutionary forces acting on arboviral populations within their arthropod vector provides novel insights into arbovirus evolution.
| During infection of their arthropod vectors, arthropod-borne viruses (arboviruses) such as dengue viruses traverse several anatomical barriers that are believed to cause dramatic reductions in population size. Such population bottlenecks challenge the maintenance of viral genetic diversity, which is considered critical for fitness and adaptability of arboviruses. Anatomical barriers in the vector were previously associated with both maintenance of arboviral genetic diversity and alteration of the variant repertoire. However, the relative role of random processes and natural selection, and the influence of vector genetic heterogeneity have not been elucidated. In this study, we used high-throughput sequencing to monitor dengue virus genetic diversity during infection of several genetic backgrounds of their mosquito vector. Our results show that initial infection of the vector is randomly founded by only a few tens of individual virus genomes. The overall level of viral genetic diversity generated during infection was predominantly under purifying selection but differed significantly between mosquito genetic backgrounds. Thus, in addition to random evolutionary forces and the purging of deleterious mutations that shape dengue virus genetic diversity during vector infection, our results also point to a novel role for vector genetic factors in the genetic breadth of virus populations.
| Due to the low fidelity of their RNA-dependent RNA polymerase, rapid replication kinetics and large population size, RNA viruses consist of a heterogeneous intra-host population of related mutants, sometimes referred to as a quasispecies [1]. This mutant swarm as a whole defines the properties of the viral population, and is considered critical for the fitness and adaptive potential of RNA viruses [1]. For example, high fidelity poliovirus mutants are attenuated in mice in vivo, demonstrating the functional importance of intra-host genetic diversity for pathogenesis [2].
Arthropod-borne viruses (arboviruses) are maintained by transmission between vertebrate hosts and blood-feeding arthropods such as mosquitoes that serve as vectors. Although arboviruses span a wide range of viral taxa in the Togaviridae, Flaviviridae, Bunyaviridae, Rhabdoviridae and Orthomyxoviridae families, the vast majority are RNA viruses, with the single known exception of a DNA arbovirus (African swine fever virus). The genetic plasticity of an RNA genome may confer arboviruses the remarkable ability to alternate between two fundamentally different hosts, and to quickly adapt to novel hosts [3]. Like other RNA viruses, high levels of intra-host genetic diversity are critical for arboviral fitness, as demonstrated in both host types for chikungunya virus [4,5] and West Nile virus [6–8].
Arboviruses usually rely on horizontal transmission between vertebrate hosts and arthropod vectors, although vertical transmission from an infected female arthropod to her offspring may occur [9,10]. After being ingested in a blood meal taken from a viremic vertebrate, arboviruses initially establish infection in the midgut epithelial cells of the arthropod vector. Transmission to another vertebrate host occurs after an extrinsic incubation period during which the arthropod develops a systemic infection that results in the release of viral particles in saliva. During infection of the arthropod vector, arboviruses are confronted with several anatomical barriers that are believed to impose severe population bottlenecks on viral populations [11]. Bottlenecks are dramatic reductions in population size, resulting in stochastic sampling of a small number of viral genomes from the mutant swarm. Population bottlenecks can significantly reduce the fitness of RNA viruses through accumulation of deleterious mutations that cannot be efficiently removed by purifying selection [12]. Initial infection and traversal of midgut cells, followed by virus dissemination and invasion of the salivary glands are expected to result in strong population drops that represents a challenge to maintaining arboviral genetic diversity during systemic vector infection [13].
Despite such population bottlenecks, arboviruses typically maintain high levels of genetic diversity during transmission by their arthropod vectors [11]. For example, analyses of West Nile virus populations in the midgut, hemolymph and saliva of Culex mosquitoes failed to document reductions in genetic diversity [14]. However, the authors of this study did not determine whether a large effective population size was maintained, or if viral genetic diversity was quickly replenished by mutation and demographic expansion following population bottlenecks. In a recent study of dengue virus (DENV), genetic diversity was maintained during human-to-mosquito transmission but the variant repertoire changed substantially between venous blood and different organs of Aedes mosquitoes that became infected by feeding on the person [15]. Over 90% of DENV genetic variants were lost upon transition from venous blood to mosquito abdomen, as well as from abdomen to salivary glands, which led the authors to estimate that about a hundred viral genomes initially established a productive midgut infection [15]. However, this number could have been underestimated because the calculation assumed that the observed change in variant frequency was due to chance alone (i.e., it did not account for the effect of natural selection). Genetic drift and purifying selection, for example, can result in a similar loss of genetic diversity.
The relative strength of natural selection and genetic drift is informed by the effective population size (Ne), defined as the size of an idealized population that would drift at the same rate as the observed population [16]. Ne indicates whether the evolution of a population is better described as a deterministic (selection) or stochastic (drift) process. When Ne is large, competition between variants occurs with little interference of random processes. When Ne is small, stochastic sampling of variants counteracts selection and hinders adaptation. For example, genetic drift plays a limited role during systemic infection of the plant host by cauliflower mosaic virus, as viral populations maintain an effective size of several hundreds of viral genomes [17]. Understanding the relative role of genetic drift and natural selection is critical to evaluate the risk of arboviral emergence through adaptive processes [3]. For example, limited epidemic potential of an Asian lineage of chikungunya virus was associated with fixation of a deleterious deletion likely due to a founder effect [18].
In the present study, we investigated the intra-host evolution of DENV in the main mosquito vector Aedes aegypti by deep sequencing the full genome of viral populations at different time points of infection. Importantly, we accounted for the potential role of mosquito genetic variation on DENV intra-host genetic diversity. DENV intra-host genetic diversity has attracted considerable interest since the confirmation of its quasispecies nature [19]. Until now, however, most of this research has focused on viral genetic diversity in humans [20–23]. A few studies examined DENV intra-host genetic diversity in the mosquito vector [15,24,25], but these studies did not account for vector genetic heterogeneity. There is substantial evidence for genetic variation in Ae. aegypti vector competence for DENV [26–32], as well as specific interactions between Ae. aegypti genotypes and DENV genetic variants [33–37].
Our objectives were three-fold: (i) measure the bottleneck size during initial midgut infection of Ae. aegypti mosquitoes by DENV; (ii) monitor DENV intra-host genetic diversity during population expansion and systemic infection; and (iii) determine the influence of the vector genotype on bottleneck size and intra-host DENV genetic diversity.
The Institut Pasteur animal facility has received accreditation from the French Ministry of Agriculture to perform experiments on live animals in compliance with the French and European regulations on care and protection of laboratory animals. This study was approved by the Institutional Animal Care and Use Committee at Institut Pasteur.
This study used a wild-type DENV-1 isolate (KDH0026A) that was originally recovered from the serum of a clinically ill dengue patient attending Kamphaeng Phet Provincial Hospital, Thailand as previously described [36]. Informed consent of the patient was not necessary because the virus was isolated in laboratory cell culture for diagnostic purposes (unrelated to this study) and, therefore, was no longer a human sample. The isolate was passaged three times in Aedes albopictus C6/36 cells prior to its use in this study. The full-length consensus genome sequence of the isolate is available from GenBank under accession number HG316481.
Aedes aegypti females used in this study belonged to the 16th generation of four isofemales lines (referred to as A, B, C, and D thereafter) derived from wild Ae. aegypti specimens collected in Kamphaeng Phet Province, Thailand. The lines were initiated by single mating pairs of field-caught males and females as previously described [36]. One male and one female from different collection sites (subdistricts) of the Muang district, Kamphaeng Phet Province, were randomly paired. The mothers of lines A and B, and the father of line C were collected in Thep Na Korn. The fathers of lines A, B and D were collected in Mae Na Ree. The mothers of lines C and D were collected in Nhong Ping Kai. They were maintained in the laboratory by mass sib-mating and collective oviposition at each subsequent generation. Quantification of genetic variation within and between the four isofemale lines was conducted as part of this study (see below).
To initiate the experiment, eggs were hatched in filtered tap water. Larvae were reared in 24×34×9 cm plastic trays at a density of about 200 larvae per tray. Adults were maintained in 30×30×30 cm screened cages under controlled insectary conditions (28±1°C, 75±5% relative humidity, 12:12 hour light-dark cycle). They were provided with cotton soaked in a 10% (m/v) sucrose solution ad libitum and allowed to mate for 6–7 days before the experimental infection.
Genetic characterization of the Ae. aegypti isofemale lines used single nucleotide polymorphism (SNP) markers identified and genotyped by Restriction-site Associated DNA (RAD) sequencing [38]. Ten females from the 16th generation of each isofemale line (i.e., from the same generation that was used in the experimental infection) and 10 females from the 1st generation of an outbred population collected in 2013 in Thep Na Korn, Kamphaeng Phet Province, Thailand (i.e., the region where the isofemale lines originated) were genotyped using RAD sequencing.
Mosquito genomic DNA was purified using the procedure developed by Pat Roman's laboratory at the University of Toronto [39]. DNA concentration was measured with Qubit fluorometer and Quant-iT dsDNA Assay kit (Life Technologies, Paisley, UK). A modified version of the original double-digest Restriction-site Associated DNA (ddRAD) sequencing protocol [40] was used as previously described [41] with minor additional modifications. Briefly, 350 ng of genomic DNA from each mosquito were digested in a 50-μl reaction containing 50 units each of NlaIII and MluCI restriction enzymes (New England Biolabs, Herts, UK), 1× CutSmart Buffer and water for 3 hours at 37°C, without a heat-kill step. Digestion products were cleaned with 1.5× volume of Ampure XP paramagnetic beads (Beckman Coulter, Brea, CA, USA) and ligated to the modified Illumina P1 and P2 adapters with overhangs complementary to NlaIII and MluCI cutting sites, respectively. Each mosquito was uniquely labeled with a combination of P1 and P2 barcodes of variable lengths to increase library diversity at 5’ and 3’ ends (S1 Table). Ligation reactions were set up in a 45-μl volume with 2 μl of 4 μM P1 and 12 μM P2 adapters, 1,000 units of T4 ligase and 1× T4 buffer (New England Biolabs) and were incubated at 16°C overnight. Ligations were heat-inactivated at 65°C for 10 minutes and cooled down to room temperature (20–25°C) in a thermocycler at a rate of 1.5°C per 2 minutes. Adapter-ligated DNA fragments from all individuals were pooled and cleaned with 1.5× bead solution. Fragments from 350 to 440 base pairs (bp) were selected using a Pippin-Prep 2% gel cassette (Sage Sciences, Beverly, MA, USA). Finally, 1 μl of the size-selected DNA was used as a template in a 10-μl PCR reaction with 5 μl of Phusion High Fidelity 2× Master mix (New England Biolabs) and 1 μl of 50 μM P1 and P2 primers (S1 Table). To reduce bias due to PCR duplicates, 8 PCR reactions were run in parallel, pooled, and cleaned with a 0.8× bead solution to make the final library. At this step, final libraries were quantified by quantitative PCR using the QPCR NGS Library Quantification Kit (Agilent Technologies, Palo Alto, CA, USA).
Libraries containing multiplexed DNA fragments from 50 mosquitoes were sequenced on an Illumina NextSeq platform using a NextSeq 500 High Output 300 cycles v2 kit (Illumina, San Diego, CA, USA) to obtain 150-bp paired-end reads. An optimized final library concentration of 1.1 pM, spiked with 15% PhiX, was loaded onto the flow cell. Raw sequences were deposited in the NCBI Sequence Read Archive under accession number SRP075401.
A previously developed bash script [41] was used to process raw sequencing reads with minor modifications. Briefly, the DDemux program was used for demultiplexing fastq files according to the P1 and P2 barcodes combinations. Sequence quality scores were automatically converted into Sanger format. Sequences were filtered with FASTX-Toolkit. The first 5 bp (i.e., the restriction enzyme cutting site) and last 11 bp of P1 and P2 reads were trimmed. All reads with Phred scores <25 were discarded. P1 and P2 reads were then matched and unpaired reads were sorted as orphans.
Paired reads were aligned to the reference Ae. aegypti genome (AaegL3, February 2016) [42] using Bowtie version 0.12.7 [43]. Parameters for the ungapped alignment included a maximum of three mismatches allowed in the seed, suppression of alignments if more than one reportable alignment existed, and a “try-hard” option to find valid alignments. Orphans were combined with all unaligned paired reads and single-end alignment was attempted. All aligned Bowtie output files were merged per individual and were imported into the Stacks pipeline. A catalog of RAD loci used for SNP discovery was created using the ref_map.pl pipeline in Stacks version 1.37 [44,45]. First, sequences aligned to the same genomic location were stacked together and merged to form loci using Pstacks. Only loci with a sequencing depth ≥3X per individual were retained. Cstacks was used to create a catalog of consensus loci, merging alleles together and Sstacks was used to match all identified loci. The Stacks pipeline identified a total of 899,892 RAD loci. The “populations” module was used to export markers with a sequencing depth ≥10X that were present in ≥98% of samples. The mosquito phylogenetic analysis was performed with the resulting subset of 2,321 SNPs, which belonged to 1,319 distinct RAD loci (0.15%).
Phylogenetic trees were constructed using a Bayesian Markov Chain Monte Carlo (MCMC) algorithm, implemented in the BEAST 1.8.3 package [46]. Inferences were produced under the coalescent model (constant size), and under the GTR+G (global time reversible with gamma distribution and no invariable sites) and the HKY+G (Hasegawa-Kishino-Yano) substitution models. Heterozygote positions were considered in calculations by enabling the use of IUPAC code and associated degeneracy within the substitution model. The length of MCMC was set at 3x107 states to obtain Effective Sampling Size (ESS) values >200.
Six- to seven-day-old Ae. aegypti females were deprived of water and sucrose for 24h prior to the infectious blood meal. The virus stock was diluted in cell culture medium (Leibovitz’s L-15 medium + 10% heat-inactivated fetal calf serum + non-essential amino-acids + 0.1% penicillin/streptomycin + 1% sodium bicarbonate) to reach an expected infectious titer of 3×106 focus-forming units (FFU) per mL. One volume of virus suspension was mixed with two volumes of freshly drawn rabbit erythrocytes washed in distilled phosphate-buffered saline (DPBS). After gentle mixing, 2.5 mL of the infectious blood meal was placed in each of several Hemotek membrane feeders (Hemotek Ltd, Blackburn, UK) maintained at 37°C and covered with a piece of desalted porcine intestine as a membrane. Sixty μL of 0.5 M ATP were added to each feeder as a phagostimulant. Each isofemale line was allowed to feed during two rounds of 15 min on different feeders to ensure randomization of a potential feeder effect. Actual virus titer in the blood meal was measured by standard focus-forming assay in C6/36 cells [33]. After feeding, mosquitoes were cold anesthetized on ice and fully engorged females were transferred to 1-pint cardboard cups. They were incubated under controlled conditions (28±1°C, 75±5% relative humidity, 12:12 hour light-dark cycle) in a climatic chamber.
At 4, 7 and 14 days post exposure (dpe), the midgut of 8–12 individuals from each isofemale line (i.e., biological replicates) were dissected. Midguts were homogenized individually in 140 μL of DPBS + 560 μL of QIAamp Viral RNA Mini Kit lysis buffer (Qiagen, Hilden, Germany) during two rounds of 30 sec at 5,000 rpm in a mixer mill (Precellys 24, Bertin Technologies, Montigny le Bretonneux, France). At 7 and 14 dpe, the legs of midgut-dissected mosquitoes were removed and homogenized as described above. At 14 dpe, the salivary glands of the midgut- and leg-less individuals were harvested and processed as above.
Total RNA was extracted from mosquito homogenates using QIAamp Viral RNA Mini Kit (Qiagen) and reverse transcribed using Transcriptor High Fidelity cDNA Synthesis Kit (Roche Applied Science, Penzberg, Germany) and a specific reverse primer located at the 3’ end of the viral genome (S1 Table). Presence and amount of viral cDNA was assessed by quantitative PCR using the LightCycler DNA Master SyberGreen I kit (Roche Applied Science) and custom primer pairs (S1 Table). Absolute quantification used a standard curve generated with serial dilutions of PCR amplicons of known concentration. Selected samples were amplified by 40 cycles of PCR in 10 overlapping amplicons with Q5 High Fidelity DNA polymerase (New England Biolabs) and custom primer pairs (S1 Table).
PCR products were purified with Agencourt AMPure XP magnetic beads (Beckman Coulter) and their concentration was measured by Quant-iT PicoGreen dsDNA fluorometric quantification (Invitrogen). Equal amounts of each amplicon were pooled by sample and brought to a final concentration of 0.2 ng/μL. Multiplexed libraries were prepared using Nextera XT DNA Library Preparation Kit (Illumina) and single-end sequenced on an Illumina NextSeq 500 platform using a high-output 75 cycles v1 kit (Illumina). Sequencing reads were demultiplexed using bcl2fastq v2.15.0 (Illumina). Raw sequences were deposited in the NCBI Sequence Read Archive under accession number SRP075335.
After demultiplexing, reads were trimmed to remove Illumina adaptor sequences using Trimmomatic v0.33 [47] and amplification primers if matching sequences were found on either the 5’ or 3’ end of the reads using Cutadapt v.1.8.3 [48]. Reads shorter than 32 bp were discarded and remaining reads were then mapped to the reference DENV genome sequence using Bowtie2 v2.1.0 [49]. The alignment file was converted, sorted and indexed using Samtools v0.1.19 [50]. Coverage and sequencing depth were assessed using bedtools v2.17.0 [51]. Single nucleotide variants (SNVs) and their proportion among all reads were called using LoFreq* v2.1.1 [52] and their effect at the amino-acid level assessed by SNPdat v.1.0.5 [53].
Two sets of SNV markers were used for analyses of genetic diversity and natural selection. The ‘full’ marker set excluded all nucleotide positions in a given sample that had (i) a sequencing depth <500X or (ii) where potential sequencing or library preparation artifacts [54] were detected. The ‘conservative’ marker set excluded all nucleotide positions in all samples that had (i) a sequencing depth <500X or (ii) where potential sequencing or library preparation artifacts [54] were detected in a least one sample. The conservative marker set minimized the potential bias owing to the unique mutational profile of each nucleotide position. However, because some of the overlapping fragments covering the viral genome could not be successfully amplified in a few samples (S1 Fig), the conservative marker set failed to cover large fractions of the viral genome (S2A Fig). The full marker set, conversely, minimized the potential bias owing to distinct evolutionary properties of the different regions of the viral genome.
Genetic complexity of the viral population was estimated using normalized Shannon entropy (Sn) for each nucleotide site [55]:
Sn = -(p ln(p))+((1-p)×ln(1-p))ln(4)
where p is the SNV minor allele frequency at the considered position, and ln(4) corresponds to maximum complexity (i.e., four possible nucleotides at each position). For individual SNVs, Sn values range from 0 to 1. For diallelic SNVs, Sn values range from 0 (no diversity) to 0.5 (maximum complexity, when the two alternative nucleotides are present at equal frequency). For each sample, Sn was averaged over all nucleotide sites included in either the full or the conservative set of SNV markers (i.e., total genome length minus number of excluded positions).
Genetic diversity of the viral population was also estimated using nucleotide diversity at each nucleotide site [56]:
π = DD-1 × (1-(p2+(p-1)2)
where D is the sequencing depth at the considered position and p is the SNV minor allele frequency. Like for Sn, π values for a diallelic SNV range from 0 (no polymorphism) to 0.5 (when the two alternative nucleotides are present at equal frequency). For each sample, π was averaged over all nucleotide sites included in either the full or the conservative set of markers. This index of genetic diversity is less sensitive to low-frequency variants than Sn, due to the lack of log-transformation of the frequencies.
The magnitude and direction of natural selection were assessed using the dN/dS ratio, which is the ratio between the number of non-synonymous substitutions per non-synonymous site (dN) over the number of synonymous substitutions per synonymous site (dS) of a coding sequence, assuming synonymous substitutions are selectively neutral:
dS = -3 ×ln(1-4 ×SdSs3)4 and dN = -3 ×ln(1-4 ×NdNs3)4
where Sd is the number of synonymous substitutions in the sequence, Ss is the number of synonymous sites, Nd is the number of non-synonymous substitutions in the sequence and Ns is the number of non-synonymous sites [57]. A dN/dS ratio >1 means that there is an excess of normalized number of non-synonymous substitutions relative to the normalized number of synonymous substitutions and is interpreted as evidence for positive selection (i.e., driving change). A dN/dS ratio <1 means that there is an excess of normalized number of synonymous substitutions relative to the normalized number of non-synonymous substitutions and is interpreted as evidence for negative selection (i.e., acting against change). A dN/dS ratio equal to 1 is interpreted as evidence for the absence of natural selection (i.e, neutral evolution).
The dN/dS ratio was computed using the Nei-Gojobori method [57] with suggested modifications for high-throughput sequencing data [58]. Briefly, Nd and Sd were calculated for each sample as the sum of SNV frequencies. Mean Nd and Sd were computed for each isofemale line at each time point and used for dN and dS calculation, respectively. Numbers of synonymous and non-synonymous sites from the initial population consensus sequence were estimated using MEGA v.7.0.16 [59] by computing the number of 0-, 2-, 3- and 4-fold degenerate sites following the Nei-Gojobori method [57]. The full marker set had a variable number of synonymous and non-synonymous sites depending of the number of nucleotide sites retained or excluded for each sample. The conservative marker set had 328.67 synonymous and 1,481.33 non-synonymous sites for all samples.
Statistical analyses were performed in the statistical environment R, version 3.2.0 (http://www.r-project.org/) using the packages car [60], plyr [61], ggplot2 [62], stringr [63], reshape2 [64], gridExtra [65], fitdistrplus [66] and boot [67]. In all analyses, the individual mosquito sample was considered a biological unit of replication.
Infection prevalence and cDNA copy numbers were compared among isofemale lines at each time point by pairwise Pearson χ2 tests and pairwise Wilcoxon signed-rank tests, respectively, followed by a Holm correction of p-values for multiple testing.
The proportion of SNVs per position, mean Sn and mean π estimates were compared between the input and later time points using pairwise Wilcoxon signed-rank tests and a Holm p-value adjustment. The proportion of SNVs per position, Sn, π and dN/dS estimates in midgut samples were analyzed between 4 and 7 dpe as a function of the combined effects of time point and mosquito genotype using a linear model with an identity link function and a normal error distribution. Model validity was verified with quantile-quantile (Q-Q) plots of residuals and by computing Cook’s distance to assess influence of observations. Statistically significant effects (p<0.05) of time point, mosquito genotype and their interactions were determined using type-II analysis of variance. Statistically insignificant interactions were removed from the model, subsequently repeating model validation. Statistical testing of pairwise differences between isofemale lines used the linear regression coefficients. Estimated regression coefficients were extracted and their 95% confidence intervals and p-values were calculated based on their standard errors compared to a reference level. Isofemale line A was arbitrarily chosen as the reference level.
Following a published method [17], bottleneck size at initial midgut infection was estimated by analyzing the change in frequency distribution of neutral markers between blood meal (initial) and midgut (final) samples. Under the assumption of neutrality (i.e., absence of natural selection), the idealized number of founding genomes (N) initiating the midgut infection can be computed as:
N = p(1-p)Var(p′)-Var(p)
where p is the marker allele frequency in the initial population and p′ is the marker allele frequency in the final population [17]. This method considers that changes in the genetic variance between sequential samples result exclusively from genetic drift and therefore requires neutral or quasi-neutral markers.
SNVs that were presumably neutral were selected based on the following set of criteria: (i) synonymous change at the third codon position, (ii) no significant change in mean frequency between sampling time points, (iii) SNV detected in ≥80% of the five viral input replicates (viral stock and blood meal samples), and (iv) mean frequency >0.02 in the input population. Confidence intervals of N estimates were computed using a bootstrapping procedure as described in [17]. Briefly, for each bootstrap all individuals were sampled with replacement to calculate N. This was repeated 1,000 times to generate a distribution of N values and derive 95% confidence intervals corresponding to the 2.5 and 97.5 percentiles of the distribution.
The effect of the initial midgut infection bottleneck on viral diversity indices was simulated in R based on 100 sampling events from an initial viral population containing 100 independent SNVs. SNV minor allele frequency was randomly drawn from an exponential distribution (λ = 100). Initial viral population size (equivalent to the infectious dose ingested in the blood meal) was drawn from a normal distribution (mean = 2,000; standard deviation = 200). Bottleneck size was drawn from a normal distribution (mean = 28; standard deviation = 5). Mean Sn and mean π were computed for all samples in the presence or the absence of a detection threshold arbitrarily set at an SNV minor allele frequency of 0.01.
A genome-wide set of 2,321 SNPs generated by RAD sequencing was used to genetically characterize the four Ae. aegypti isofemale lines (A, B, C, and D). These markers had a sequencing depth ≥10X per sample and were missing in <2% of samples. An outbred Ae. aegypti population from the same geographic location where the lines were created was also genotyped to provide a phylogenetic background. Phylogenetic relationships among individuals from the four isofemale lines and the outbred population were determined with a Bayesian method (Fig 1). As expected, the outbred mosquito population was paraphyletic, reflecting its genetic diversity. Mosquitoes from isofemale lines A and B clustered independently with strong statistical support, confirming their distinct genetic identity. Unexpectedly, mosquitoes from isofemale line C grouped with mosquitoes from isofemale line D within the same clade. This could be the result of relatedness between the parents randomly chosen to initiate the lines, as the mothers of lines C and D came from the same collection site and may have been siblings. Two different substitution models for the phylogenetic reconstruction gave similar clustering patterns. Similar results were also obtained when testing a variable number of markers (allowing from 0% to 30% of missing genotypes) with the same method. Because isofemale lines C and D were not unambiguously assigned to different monophyletic groups, they could not be considered as distinct genotypes and were thus combined in all subsequent analyses. They are jointly referred to as line CD hereafter.
Mosquitoes from the three different genotypes (A, B, and CD) were exposed to DENV through an artificial blood meal at a final titer of 1.52×106 focus-forming units (FFU)/mL. Assuming a blood meal size of approximately 2 μL, the infectious dose ingested by each mosquito was about 3,000 infectious viral particles. The proportion of mosquitoes that acquired a midgut infection ranged from 75 to 100% and did not differ significantly between time points or isofemale lines (Fig 2A). The proportion of mosquitoes with a DENV infection that disseminated to their legs increased from 10–40% at 7 days post exposure (dpe) to 60–100% at 14 dpe, but the rate of virus dissemination to the legs did not differ significantly between isofemale lines (Fig 2A). However, the proportion of mosquitoes with a disseminated infection in the salivary glands was significantly higher for line CD (87.5%) than for line A (37.5%) and line B (41.7%) at 14 dpe (line A vs. line CD, p = 0.037; line B vs. line CD, p = 0.037). Among infected mosquitoes, viral load ranged from 8.9×102 to 2.8×106 DENV genome copies per sample with no significant difference between isofemale lines at any of the time points, with the exception of lines B and CD that had significantly different viral loads (p = 0.037) in their salivary glands at 14 dpe (Fig 2B).
Deep sequencing of viral genomes was performed for a subset of 78 infected samples at selected time points (Fig 2B) that were processed individually and treated as biological replicates. Some samples were excluded because their low concentration of viral RNA resulted in unsuccessful RT-PCR amplification. A total of 4, 7 and 13 infected midguts at 4 dpe and 7, 11 and 21 infected midguts at 7 dpe were analyzed for lines A, B, and CD, respectively. Ten infected salivary glands at 14 dpe were analyzed in line CD. In addition, DENV genomes were deep sequenced in the initial viral stock and in four replicates of the infectious blood meal. On average, 3,615,466 sequencing reads per sample aligned to the reference DENV genome. Mean DENV genome coverage with a sequencing depth >500X was 10,594 nucleotides per sample, which represents 98.8% of the 10,718 nucleotides of the total genome length. Mean sequencing depth was 24,212X per sample (S1 Fig).
The full set of SNV markers retained for population genetic analyses included an average of 5,843 nucleotide sites across the DENV genome, whereas a more conservative set (see Materials and Methods) was restricted to 1,810 nucleotides (S2A Fig). SNVs of the full marker set were randomly distributed across the genome without obvious mutation hot or cold spot (S3 Fig). A new variant reached consensus level (frequency >0.5) in one midgut sample at 4 dpe and one midgut sample at 7 dpe, but the SNV was different in each case. In salivary glands collected at 14 dpe, new variants reached consensus level at 11 positions, none of which was shared among individuals within or between isofemale lines (S3 Fig). In the more restricted conservative set of markers, no variant reached consensus level at any time point (S2B Fig).
To determine the effect of initial midgut infection on DENV genetic diversity, a first series of analyses compared viral genetic diversity observed in the input samples with genetic diversity observed at any of the later time points. In the full marker set, initial infection of the midgut was associated with an increase in viral genetic diversity relative to the input (Fig 3) both when measured with normalized Shannon entropy Sn (0 dpe vs. 4 dpe, p = 0.0001; 0 dpe vs. 7 dpe, p = 0.002; 0 dpe vs. 14 dpe, p = 0.003) (Fig 3A) and when measured with nucleotide diversity π (0 dpe vs. 4 dpe, p = 0.0001; 0 dpe vs. 7 dpe, p = 0.0004; 0 dpe vs. 14 dpe, p = 0.003) (Fig 3B). Viral diversity was also significantly higher in the salivary glands at 14 dpe than in the midgut at 7 dpe (Sn: p = 0.012; π: p = 0.012). The proportion of variable sites detected also increased following initial midgut infection (S4 Fig) although differences were only statistically significant between 0 dpe and 4 dpe (p = 0.0029) and between 0 dpe and 14 dpe (p = 0.0067). Similarly, in the conservative set of markers, mosquito infection was associated with a relative increase in viral genetic diversity following initial midgut infection, albeit more modestly due to the smaller number of markers, both when measured with normalized Shannon entropy Sn (0 dpe vs. 4 dpe, p = 0.046; 0 dpe vs. 14 dpe, p = 0.008) (S5B Fig) and when measured with nucleotide diversity π (0 dpe vs. 14 dpe, p = 0.008) (S5C Fig). The proportion of variable sites detected, however, did not differ statistically between time points (S5A Fig).
To evaluate the dynamics of DENV genetic diversity during viral population expansion in the midgut, a second series of analyses compared viral genetic diversity between 4 and 7 dpe, accounting for potential differences between mosquito genotypes. In the full set of markers, both the time point (proportion of variable sites: p = 0.03; Sn: p = 0.04; π: p = 0.04) and the isofemale line (proportion of variable sites: p = 0.0035; Sn: p = 0.0002; π: p = 0.0005) significantly influenced viral genetic diversity. Overall, DENV genetic diversity slightly decreased between 4 dpe and 7 dpe. Isofemale line A displayed significantly higher viral genetic diversity than lines B and CD, for all three indices: proportion of variable sites (p = 0.012 and p = 0.0008, respectively), Sn (p = 0.006 and p = 0.0005 respectively) and π (p = 0.017 and p = 0.0001, respectively). Similar results were obtained with the conservative set of markers. Both the time point (proportion of variable sites: p = 0.01; Sn: p = 0.013; π: p = 0.015) and the isofemale line (proportion of variable sites: p = 0.0011; Sn: p = 0.0006; π: p = 0.0029) significantly influenced viral genetic diversity. Overall, DENV genetic diversity slightly decreased between 4 dpe and 7 dpe. Isofemale line A displayed significantly higher viral genetic diversity than lines B and CD, for all three indices: proportion of variable sites (p = 0.017 and p = 0.0002, respectively), Sn (p = 0.0047 and p = 0.0001, respectively) and π (p = 0.012 and p = 0.0007, respectively).
Based on the full set of SNV markers, dN/dS ratios were predominantly negative indicating purifying selection (Fig 3C). There was no statistically significant difference in dN/dS ratios between time points or mosquito isofemale lines. Computing dN/dS ratios was not possible with the conservative set of markers because the smaller number of SNVs resulted in a large proportion of samples with dS = 0. Analysis of dN/dS ratios calculated per isofemale line, however, provided results consistent with predominantly purifying selection using the conservative set of markers. Average dN/dS ratios were remarkably similar among lines and time points around 0.2218 (S1 Table).
Three SNVs that complied with criteria of quasi-neutral evolution were selected to estimate the idealized number of founding viral genomes (N) initiating the midgut infection based on changes in the variance of their frequency between input and midgut samples (Table 1). Based on the three markers, initial midgut infection was founded by 23–34 genomes when estimated at 4 dpe (Fig 4A) and 5–42 genomes when estimated at 7 dpe (Fig 4B). Collectively, 95% confidence intervals ranged from 2 to 161 founding genomes. N estimates and their confidence intervals were consistent across time points, especially for marker at position 1556. For this marker, 4 dpe and 7 dpe data were pooled to compute isofemale line-specific N estimates. There were no statistically significant differences among lines in the estimated bottleneck size (Fig 4C), ranging from 83 (95% confidence interval: 52–396) founding genomes for line A, to 23 (9–220) for line B and 33 (16–108) for line CD.
Simulations were performed to model the effect of population bottlenecks on DENV intra-host genetic diversity. The simulation randomly assigned SNV minor allele frequency, initial viral population size and bottleneck size to explore whether a minimum threshold for SNV detection would alter the observed genetic diversity following a population bottleneck compared to the true genetic diversity. When 100 SNVs were present in the input viral population and no minimum detection threshold was set, mean Sn and π estimated in 100 replicate samples decreased following the bottleneck (Fig 5A). However, when only SNVs with a minor allele frequency >1% were detected, mean Sn and π estimates increased after the bottleneck (Fig 5B).
We investigated the evolutionary forces acting on DENV populations within their arthropod vector. Specifically, we evaluated the relative effects of natural selection and genetic drift on DENV intra-host evolution in the midgut of Ae. aegypti. In addition, we assessed the influence of vector genetic heterogeneity on intra-host viral genetic diversity. Our results show that DENV intra-host genetic diversity in Ae. aegypti is shaped by the combined effects of genetic drift, purifying selection and vector genotype. Reshuffling of the variant repertoire during initial infection of the midgut was associated with a bottleneck size ranging from 5 to 42 founding viral genomes, irrespective of the mosquito genotype. DENV genetic diversity increased significantly following initial infection, but was restricted by strong purifying selection during DENV population expansion in the midgut. Observed levels of DENV genetic diversity in the midgut differed significantly between mosquito isofemale lines despite a similar bottleneck size at initial infection.
Arboviruses typically maintain high levels of genetic diversity during transmission by their arthropod vectors despite anatomical barriers that often result in severe population drops [11]. Such population bottlenecks have been documented for several arboviruses in their vectors using artificial mutant swarms [68], marked viral clones [13], viral replicons [69] or stochastic simulations based on observed changes in variant frequencies [15]. Although the overall level of arboviral genetic diversity is usually maintained during vector infection [14], the viral variant repertoire can be significantly altered [15,68,70]. Presumably, viral genetic diversity is quickly replenished by mutation and demographic expansion following population bottlenecks [11]. However, whether changes in the viral variant repertoire are due to stochastic sampling (i.e., genetic drift), purifying selection (i.e., removal of variants with lower fitness), or vector genetic heterogeneity combined with specific interactions between vector and virus genotypes [33–37] has remained largely unresolved. Our analysis used neutral or quasi-neutral genetic markers to estimate the effective DENV population size during initial infection of the Ae. aegypti midgut. This approach rules out natural selection and only measures the effect of genetic drift due to random sampling. It is worth noting, however, that true neutral mutation may not exist because even synonymous mutations can have a fitness effect, especially in RNA viruses [71]. Deviation from our assumption of neutrality or quasi-neutrality of the chosen markers may have overestimated the bottleneck size. Indeed, both positive selection and negative selection would likely act to decrease the variance of marker frequency and therefore result in a larger estimate of Ne with our method. Therefore, our conclusion that DENV populations undergo a strong population bottleneck during initial midgut infection should be robust to any undetected departure from neutrality. Moreover, we chose markers whose average frequency was similar before and after the bottleneck, supporting the assumption that they were not under directional selection. Our estimation that initial midgut infection is founded by only a few tens of DENV genomes is consistent with previous estimations for DENV based on stochastic simulations using empirical data [15]. We went one step further by demonstrating that genetic drift, rather than natural selection, is the main evolutionary force underlying this population bottleneck.
Although our estimated bottleneck size is larger than for other RNA viruses during host-to-host transmission [72], it is expected to substantially reduce the genetic breadth of the viral quasispecies [73]. This finding has important implications for DENV evolution in general. A small effective population size means that natural selection will be relatively inefficient during human-to-mosquito transmission. It will prevent adaptive evolution especially if beneficial SNVs are present at low frequencies in the mutant swarm transmitted from the human host [74]. On the other hand, the population bottleneck associated with initial midgut infection may not be small enough to prevent the long-term maintenance of defective viral genomes through complementation by co-infection of host cells with functional viruses. Such a phenomenon was previously documented in the case of a stop-codon mutation that became widespread in DENV populations sampled in Myanmar in 2001 [75]. The frequency of the stop-codon mutation was likely high enough to overcome the effect of population bottlenecks during multiple host-to-host transmission events.
During DENV population expansion following initial midgut infection, natural selection was predominantly negative (i.e., acting against change). Accordingly, the consensus sequence remained unchanged in most of the midgut samples. Only in the salivary glands did several SNVs reach consensus level (frequency >0.5), but with no evidence of evolutionary convergence. As was observed for West Nile virus [68], DENV intra-host genetic diversity in midguts slightly decreased between 4 and 7 dpe. Importantly, we found that overall levels of DENV intra-host genetic diversity differed significantly between distinct mosquito genetic backgrounds. Both the initial bottleneck size and the census size of the viral population did not significantly vary among mosquito genotypes, and thus are unlikely to explain this difference. The mechanistic basis of this finding remains to be determined, but we speculate that viral populations may undergo different selective constraints in different mosquito genotypes. Mosquito genotypes could vary in the intensity of purifying selection (i.e., variation in the efficiency of mechanisms that remove deleterious de novo mutations), but this was not supported by our data. Likewise, the overall lack of positive selection that we observed indicates that it is unlikely to be the underlying mechanism. Alternatively, mosquito genotypes may differ in the level of balancing selection (i.e., mechanisms that act to promote genetic diversity such as negative frequency-dependent selection or spatiotemporal fluctuations in the strength and direction of selection). The antiviral RNA interference (RNAi) pathway of mosquitoes was suggested to play a role in viral genetic ‘diversification’ [76,77], by promoting escape to complementary base-pairing required for RNAi-mediated cleavage. Variation in host factors could also result in differences in viral intra-host genetic diversity through subtle changes in viral RNA-dependent RNA polymerase fidelity [78]. Mutation rates of RNA viruses are not only determined by virus-encoded factors, by also by host-dependent processes. Replicase fidelity of a plant RNA virus was found to differ according to the host type [79]. Replication fidelity in retroviruses can be affected by intracellular dNTP imbalance [80,81]. Viral mutation rate can also be influenced by the expression of host genes, such as cellular deaminases that promote hypermutation in RNA viruses [82–84].
Interestingly, the isofemale line that displayed the lowest level of DENV genetic diversity in the midgut (i.e., line CD) was also associated with the highest prevalence and highest viral load in salivary glands. Because we did not examine viral populations in saliva samples, whether this translates in differences of virus transmission potential is unknown. It is tempting to speculate that the vector competence phenotype relates to the level of viral genetic diversity. Unfortunately, we could not compare DENV intra-host diversity in salivary glands between mosquito isofemale lines because DENV amplification was unsuccessful in two out of three lines due to low template concentration. A recent study found differences in the intra-host genetic diversity of West Nile virus among different species of Culex mosquitoes [85]. Here, we provided evidence that such differences exist at the intra-specific level. The potential relationship between viral intra-host genetic diversity and vector competence variation among mosquito genotypes deserves further investigation. It will be interesting to determine in future experiments whether the effect of the vector genotype varies according to the mosquito generation, the virus strain, and/or the specific combinations of mosquito lines and virus strains.
Finally, we introduced a non-exclusive, alternative scenario to the ‘diversification’ hypothesis that may contribute to explain why the level of arboviral genetic diversity increases despite a population bottleneck. Our proposed scenario is based on the counter-intuitive idea that a strong initial population bottleneck may actually result in a higher observed level of genetic diversity if low-frequency SNVs go undetected for methodological reasons. We used a model based on stochastic simulations to illustrate the effect of a minimum detection threshold of low-frequency SNVs on observed genetic diversity. When all SNVs present were detected regardless of their frequency (i.e., no detection threshold), mean viral genetic diversity indices decreased following a simulated population bottleneck. Conversely, mean genetic diversity indices increased when only SNVs present at a frequency >1% were successfully detected. In our empirical data, it was not possible to ascertain whether SNVs newly detected after the initial population bottleneck resulted from de novo mutations or were already present prior to the bottleneck at frequencies lower than the detection threshold. However, our model indicated that a change in the SNV frequency spectrum following the population bottleneck combined with a minimum detection threshold is a potential explanation to the observed increased genetic diversity following the initial bottleneck.
Taken together, our results show that DENV intra-host genetic diversity in the mosquito vector is shaped by stochastic events during initial midgut infection due to a sharp reduction in population size, followed by predominantly purifying selection during population expansion and diversification in the midgut. Differential diversification between mosquito isofemale lines indicates a genetic foundation, but the lack of convergent SNVs does not support the existence of mosquito genotype-specific directional selection. We conclude that the evolution of DENV intra-host genetic diversity in mosquitoes is not only driven by genetic drift and purifying selection, but is also modulated by vector genetic factors. Characterizing the evolutionary forces that govern arboviral genetic diversity contributes to understanding their unique biology and adaptive potential.
|
10.1371/journal.pcbi.1003316 | Conformational Changes in Talin on Binding to Anionic Phospholipid Membranes Facilitate Signaling by Integrin Transmembrane Helices | Integrins are heterodimeric (αβ) cell surface receptors that are activated to a high affinity state by the formation of a complex involving the α/β integrin transmembrane helix dimer, the head domain of talin (a cytoplasmic protein that links integrins to actin), and the membrane. The talin head domain contains four sub-domains (F0, F1, F2 and F3) with a long cationic loop inserted in the F1 domain. Here, we model the binding and interactions of the complete talin head domain with a phospholipid bilayer, using multiscale molecular dynamics simulations. The role of the inserted F1 loop, which is missing from the crystal structure of the talin head, PDB:3IVF, is explored. The results show that the talin head domain binds to the membrane predominantly via cationic regions on the F2 and F3 subdomains and the F1 loop. Upon binding, the intact talin head adopts a novel V-shaped conformation which optimizes its interactions with the membrane. Simulations of the complex of talin with the integrin α/β TM helix dimer in a membrane, show how this complex promotes a rearrangement, and eventual dissociation of, the integrin α and β transmembrane helices. A model for the talin-mediated integrin activation is proposed which describes how the mutual interplay of interactions between transmembrane helices, the cytoplasmic talin protein, and the lipid bilayer promotes integrin inside-out activation.
| Transmission of signals across the cell membrane is an essential process for all living organisms. Integrins are one example of cell surface receptors (αβ) which, uniquely, form a bidirectional signalling pathway across the membrane. Integrins are crucial for many cellular processes and play key roles in pathological defects such as cardiovascular diseases and cancer. They are activated to a high affinity state by the intracellular protein talin in a process known as ‘inside-out activation’. Despite their importance and the existence of functional and structural data, the mechanism by which talin activates integrin remains elusive. In this study we use a multi-scale computational approach, which combines coarse-grained and atomistic molecular dynamics simulations, to suggest how the formation of the complex between the talin head domain, the cell membrane and the integrin moves the integrin equilibrium towards an active state. Our results show that conformational changes within the talin head domains optimize its interactions with the cell membrane. Upon binding to the integrin, talin facilitates rearrangement of the integrin TM region thus promoting integrin activation. This study also provides a demonstration of the strengths of a computational multi-scale approach in studies of membrane interactions and receptor conformational changes and associated proteins that enable transmembrane signaling.
| Integrins are cell surface receptors involved in many essential cellular processes, such as cell migration, and in pathological defects, such as thrombosis and cancer [1]. Integrins are αβ heterodimers. Each subunit has a large ectodomain, a single transmembrane (TM) helix and a short flexible cytoplasmic tail [2]. Integrins are crucial for many signal transduction events [3]–[6]. Unusually, they can transmit signals in both directions across the cell membrane [2], [7]. In the inside-out activation pathway, formation of a complex between talin, the integrin β cytoplasmic tail, and the membrane is thought to shift the integrin conformational equilibrium towards an active state [8]–[15]. Activation is believed to proceed via changes in TM helix packing [2], [3] coupled to substantive conformational changes in the integrin ectodomain.
Talin consists of a head domain (∼50 kDa) and a large rod domain (∼220 kDa) [16], [17]. A crystal structure of the talin head domain revealed a novel linear arrangement of four subdomains, F0, F1, F2 and F3 (Fig. 1A). Although the head domain has sequence homology with other FERM (band four-point-one, ezrin, radixin, moesin) domains [18], the linear arrangement of the subdomains, an inserted loop in the F1 domain, and the extra F0 domain confer on talin significantly different features from canonical FERM domains [19]. Experimental studies of the talin head have yielded valuable insights into the role of the different subdomains [8], [12], [17], [20], [21]. A key step in integrin activation is binding of the F3 subdomain to the integrin β tail [9], [10], [12]–[14]. The other subdomains have also been shown to contribute to integrin activation, although they do not bind directly to the integrin β subunit [8]. The F2 subdomain has a positively charged patch which has been shown to enhance activation by interacting with negatively charged lipid headgroups in the membrane [11], [21]. The F0 and F1 subdomains have an ubiquitin-like fold [17], [20] along with a flexible, positively charged loop of ∼40 residues inserted in the F1 domain. This loop (which was removed to facilitate structure determination of the head domain [17]) has been proposed to form a transient helix stabilized by interactions with acidic phospholipids [20]. Structural studies of the talin head domain and fragments have also suggested that the F2–F3 and the F0–F1 domain pairs are relatively rigid but the pairs are connected by a flexible linker [17].
The inactive state of integrins is in part maintained by interactions between the α and β TM helices. Important interactions are defined by regions known as the outer (OMC) and inner (IMC) membrane clasps (Fig. 1B). In the OMC, a GxxxG motif in the α integrin TM region allows close packing with the β integrin helix whereas in the IMC there are hydrophobic interactions involving a cluster of phenylalanines and a salt bridge between the α and β chains [22]. Various models for TM helix rearrangements, including ‘scissor’ or ‘piston’ movements and increased helix separation have been proposed to explain activation and trans-membrane signaling [10], [23]–[29] but there remains a need to distinguish and refine these models.
Molecular dynamics (MD) simulations allow us to explore the conformational dynamics and lipid interactions of membrane proteins [30]. Multiscale approaches combine coarse-grained molecular dynamics (CG-MD) simulations [31], [32], which extend the time scales that can be studied, with subsequent all-atom simulations, which allow refinement of the system [33]. We previously used this approach to develop a model that explained how the F2–F3 fragment of the talin head domain associates with the plasma membrane in a way that led to a scissoring movement of the two integrin TM helices [34]. In the current study, we build upon these studies to elucidate the interactions between the complete talin head domain (i.e. domains F0–F3), a lipid bilayer, and the α/β integrin TM regions. Our results provide novel information about the orientation of the intact talin head in complex with the lipid bilayer, and about changes induced in the TM helical regions. Overall, the results reveal how binding of talin to the membrane and to integrins tails leads to integrin activation.
We have used a serial multiscale MD simulation [35] approach to explore the dynamics of the talin head (F0–F3) domain, its interaction with lipid bilayers, and the resultant conformational changes of a talin/integrin TM complex embedded in a phospholipid bilayer. This approach has previously been used to explore the interactions of a number of peripheral proteins with membrane surfaces and lipids [11], [36]–[39], of TM helices within a lipid bilayer [40], [41], and of more complex signaling and related assemblies within membranes [34]. It enables one to combine coarse-grained (CG) simulations of membrane association and related events, with more detailed atomistic simulations to refine the resultant models. It thus provides a complementary approach to extended atomistic simulations [42].
The principal simulations underlying the current study are summarized in Table 1 and in Fig. 2. The talin head domain crystal structure (PDB:3IVF [17]) lacks a long loop region in the F1 domain and therefore atomistic simulations of the talin head domain (tal-sol-AT) in solution were first used to explore potential internal flexibility between the four component sub-domains (F0–F3) along with possible conformations of the F1 domain insertion. For these simulations the insertion in the F1 domain (res: 134–172) was modeled in a random coil conformation, and was located away from the F0–F1 pair (see Fig. 3B and Fig. S1A) using Modeller 9v8 (http://salilab.org/modeller/) [43], [44]. This configuration of the loop allows exploration of all possible conformations/orientations and selection of a preferred conformation/orientation relative to the talin head domain.
The conformation of the talin head domain suggested by the above simulations was used to model the association of the talin head domain with a phospholipid bilayer. Since NMR studies suggested a helical propensity for the region involving residues 154–167 [20] this region was modeled as an α-helix (tal-h2F0-CG and Fig. S1). Subsequently, the same multiscale simulation approach was used to explore the dynamic behavior of a talin/TM integrin complex in a bilayer (αβ-talh2-CG) on a multi-microsecond timescale. The talin head domain/αβ complex was constructed as described in Kalli et al. [34] using the αΙΙbβ3 TM region NMR structure [22], the F2–F3/β1D complex crystal structure [12] and the talin head domain configuration obtained from the talin head domain simulations described in this study. An ‘open’ model generated by these CG simulations was subsequently explored via a microsecond duration atomistic simulation (αβ-talh2o-AT).
A number of control simulations were also performed to evaluate the robustness/sensitivity of the results and to explore the contributions of different regions and interactions (e.g. flexibility within the domain, electrostatic interactions and other helical conformations in the F1 loop) to the binding of the talin head to anionic lipid bilayers. Detailed descriptions of these simulations are provided in the Supporting Information (Tables S1 and S2). In total our study amounts to ca. 60 µs of CG-MD and ca. 2 µs of atomistic molecular dynamics simulations (AT-MD) simulation time.
To study the conformational dynamics of the talin head domain prior to the association with the bilayer atomistic (AT-MD) simulations of the talin head domain (i.e. subdomains F0 to F3) in aqueous solution in the absence of a bilayer were performed (tal-sol-AT in Table 1). During these simulations the flexible linker between the F2–F3 and F0–F1 pairs allowed transient displacement of the F0–F1 subdomain relative to the F2–F3 subdomain with the angle defined in Fig. 3A. This angle, equal to 0° for a linear arrangement of F0-F1-F2-F3 as seen in the crystal structure, ranged from 0° to 90° in the simulations. During these simulations the long loop in the F1 domain moved closer to the F0–F1 pair, and adopted an extended conformation on the same side of the protein as the positively charged patch on F2 (Fig. 3B). Calculation of the electrostatic field around the talin head conformation observed at the end of these simulations suggests that localization of this loop close to the F0–F1 pair creates an extensive positively charged surface on one side of protein; this could facilitate strong talin/bilayer interactions (Fig. S2). Despite experimental evidence for an α-helical propensity in the F1 loop [17], no helix formation was detected (this might be due to insufficient simulation time for a coil-to-helix conformational transition to occur). Although there was a relatively large change in the angle between the F0–F1 and F2–F3 domain pairs, no significant angle change was observed within either the F0–F1 or the F2–F3 domain pair, suggesting that each pair behaves approximately as a rigid body.
Having established in the tal-sol-AT simulations (see above) that the F1 loop interacts with F0–F1 to form a positively charged surface that extends the positive patch on F2–F3, CG simulations with the loop in this location were performed to explore the nature of the interactions of the complete talin head domain with an anionic lipid bilayer. Note that in this simulation system a small helical region (h2 helix; see Fig. S1) was included within the F1 loop as indicated by NMR data [20]. During this modeling of the loop the remainder of the structure, with the exception of the region modeled as helical (res: 154–167), was restrained to maintain the talin conformation derived from the above simulations. These restraints were removed during the simulations. In the tal-h2F0-CG simulation (Table 1; Fig. 4 and Fig. 5), talin was observed to associate with the bilayer in four out of five simulation and in all four of these simulations talin bound to the bilayer initially via a basic loop (res: 318–330) in the F3 domain, and subsequently via the positively charged patch in the F2 domain (res: 255–285) (Fig. S3A). Contacts were defined by using a distance cut-off of 7 Å between the protein residues and the lipids. These two regions have been identified previously [11] to promote productive binding of the isolated F2–F3 fragment to an anionic lipid bilayer. The additional surface created by the F1 loop also interacted with the bilayer (Fig. 4C). Interestingly, in all simulations which resulted in a talin/bilayer complex, the talin head domain with the F1 loop adopted a V-shaped conformation due to rotation of the F0–F1 pair relative to the F2–F3 pair in the bilayer plane (Fig. 4A). The reorientation of the F2–F3 and F0–F1 domain pairs prior to the binding to the bilayer was also observed in other simulations in which a different starting conformation of talin was used (e.g. the talin crystal structure; data not shown). In contrast to the more dynamic variations in the angle between the F0–F1 and F2–F3 observed when talin was in solution (see above), the V-shaped conformer was stabilized by association with the bilayer (Fig. 4B). This conformation optimizes talin/lipid interactions and induces a more compact arrangement of domains, although this new arrangement is still different from the linear arrangement in the X-ray structure [17] and the canonical FERM domain packing of F0 to F3 [18]. During the AT-MD simulations (tal-h2F0-AT; see Table 1) that started from the final snapshot of the tal-h2F0-CG simulation, the V-shaped conformation of talin was retained, with talin interacting preferentially with the headgroups of the anionic POPG lipids (Fig. S6B). No restrictions in the position/flexibility of the loop or the domains were imposed in the AT-MD simulations. Simulations of the talin head domain with a neutral bilayer (containing 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidyl-choline (POPC) lipids) resulted in no association of talin with the bilayer (Fig. S3B and Text S1). These results are in good agreement with the available experimental data [11], [12], [17] and augment our previous observation that electrostatic interactions are important in regulating the formation of a talin/membrane complex.
Control CG simulations starting with the talin head domain crystal structure (i.e. without the F1 loop) with the same POPC/POPG (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidyl-choline/1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidyl-glycerol) bilayer showed that the positively charged region of F0–F1 augmented by the F1 loop promotes direct F0–F1/bilayer interactions (data not shown). CG models of talin with restricted flexibility between the F0–F1 and F2–F3 domain pairs did not bind to the bilayer in a way that would facilitate binding of F2–F3 to the β-tail in any of these simulations (see Text S1). Disruption of the electrostatic interactions between talin and the membrane (simulating with experimentally tested mutations [10], [12]) also resulted in the ‘non-productive’ orientation of the talin head domain relative to the membrane (see Methods for description of mutations). An orientation is judged here to be ‘non-productive’ when the talin/bilayer complex formed is incompatible with binding to the β integrin cytoplasmic region in a manner similar to that observed by Anthis et al. [12] (see Text S1).
Overall, our simulations suggest that optimal association of talin with the membrane is enhanced by conformational flexibility within the head domain, especially between the F0–F1 and F2–F3 pairs. This flexibility facilitates optimal interactions of the V-shaped conformation of talin with the headgroups of anionic lipids. Reduction in the anionic lipid content significantly decreased the talin/bilayer association and, in the presence of mutations that disrupted the talin/lipid electrostatic interactions, talin bound to the bilayer in a perturbed orientation (Fig. S5).
Having established the nature of the interactions of the intact talin head domain with a bilayer, we set out to explore the impact of these interactions on the structure of the integrin TM region. To this end, the talin head domain with optimal bilayer interactions (i.e. from simulation tal-h2F0-CG) was modeled in a complex with the α/β integrin TM complex (see Methods for details of modeling this complex). This complex was inserted in a POPC/POPG bilayer (Fig. 6A), and five extended CG-MD simulations (αβ-talh2-CG; Table 1) were performed. In these simulations all restraints between the integrin α subunit TM domain and the F2–F3/β TM complex were removed to allow the α TM domain to move relative to the rest of the complex. Closer examination of the movement of the individual domains within the complex during the simulations revealed a reorientation of the talin head domain relative to the bilayer surface, comparable to the reorientation observed in our earlier studies of the F2–F3/αβ complex (Fig. 6B). This rotation, in turn, induced a ∼25° rotation of the β TM helix perpendicular to the bilayer normal, similar to that seen in our previous studies (Fig. 6B) [34]. This resulted in the disruption of the interactions in the OMC and IMC regions of the α/β TM segment which were shown to maintain the integrin inactive state [22], [27], [29], [45]–[52]. In particular, the β TM helix rotation perturbed the close packing in the OMC region (because of the 972GxxxG976 motif in the α TM helix) and disrupted the hydrophobic interactions in the IMC region formed by F992 and F993 residues and the αIIb995β3723 salt bridge. Calculation of the angle between the F2–F3 and the F0–F1 domain pairs during the simulations revealed a stable angle of ∼60° (the definition of the angle is the same as described above), indicating that the V-shaped conformation of the talin head was retained in all the simulations (Fig. S7A). We note that simulations of the α/β dimer and the isolated β integrin tails in the same bilayer were performed in our previous study [34].
To explore the talin/αβ complex in more detail, a structure which represented an “open” state of the integrin TM domain was selected from the αβ-talh2-CG simulation (using similar criteria to those described in our previous simulations of the α/β dimer and the isolated β integrin tails in a bilayer [34]) and converted to an AT representation. In this “open” integrin model, the interactions in both the IMC and OMC regions are disrupted. We have previously suggested that the scissoring motion of integrin TM helices may be “a trigger for inside-out activation” [34]. To explore the consequences of activating this trigger, we performed an extended (microsecond) AT-MD simulation of the complex (αβ-talh2o-AT; Table 1). At the end of this atomistic simulation the V-shaped conformation of the talin head domain was retained. Calculation of the inter-helical distances in the OMC and IMC regions revealed an increase in the helix-helix distance in both the IMC and OMC regions (Fig. 7A,B), corresponding to dissociation of the α and β TM helices. Dissociation was preceded by a ‘scissoring’ movement similar to that seen in our previous studies of the F2–F3/αβ complex but with more extensive movement of the TM helices and a large increase in the tilt angle of the β helix relative to the bilayer normal, to a final value of 40°, (Fig. S7B). On a similar simulation timescale, the F2–F3 fragment alone induced a much smaller separation of the α/β subunits (Fig. S8B) with a comparable increase in the β TM helix tilt angle to that observed in our previous studies [34]. This suggests that the F2–F3 domains are sufficient to modulate the β tail tilt angle but the entire head is more effective in producing TM helix separation.
The increase in tilt angle induced by the talin head increases the extent of membrane-embedding of the β tail TM region. In the final orientation seen in the simulation with the talin/αβ complex, residues from K716 up to K725 are embedded in the membrane (Fig. 8). A similar increase in the extent of the membrane-embedded region was observed in a recent experimental study [53]. Despite a large increase in β-TM tilt angle, the T715 sidechain remains oriented toward the lipid phosphate atoms, but the K716 sidechain is no longer in contact with the lipid headgroups. This is in good agreement with experimental data that identified interactions between the β3 K716 ε-amine group and the lipid phosphate in the integrin inactive state, suggesting that this interaction controls the tilt angle of the β3 integrin tail [54]. Mutation of K716 in these experimental studies shifted the integrin conformational equilibrium towards an active state, possibly by perturbing the β subunit tilt angle and the α/β TM region crossing angle. Thus, in the active state one would expect weaker K716/lipid headgroup interactions, as observed in our simulations of the integrin active state. The tendency of the β TM helix to adopt a tilted orientation in the membrane is also suggested by other experimental [55] and computational [21] data.
Our studies show that conformational changes in the talin head on binding to anionic phospholipid membranes mediate transmembrane signaling by the integrin TM helix dimer. Simulations have revealed how optimization of the interactions of the complete talin head domain (F0–F3) with an anionic phospholipid bilayer promotes a rearrangement of the subdomains, which in turn initiates a conformational rearrangement of the integrin TM region. In particular, on binding to the membrane the F0–F3 talin head domain undergoes a conformational change to a V-shape, rather than the linear arrangement of the F0-F1-F2-F3 subdomains observed in the crystal structure. Formation of a complex between the rearranged talin head and the integrin TM domain subsequently triggers separation of the TM helices i.e. disassembly of the TM helix dimer. The first stage of this disassembly is an initial scissoring motion of the helices, as seen in our previous studies [34].
The current study reveals how the flexible linker between the F0–F1 and F2–F3 pairs allows talin to adopt a conformation which optimizes its contacts with anionic headgroups of lipids (Fig. 4A). This correlates well with the studies by Bouaouina et al. [8] which have suggested that the activation responses of β3 and β1 integrins have different dependencies on talin head fragments, indicating that formation of an optimal configuration of the talin/membrane complex could also have mechanistic importance.
Our simulations reveal how electrostatic interactions between the protein and anionic lipid headgroups orient the talin head domain and optimize its interactions with the membrane. The cationic surface of the talin head, which binds to the anionic lipids, is formed by the F2 and F3 domains plus the F1 loop. In silico mutations (i.e. K324D and K256E, R277E, K272E, R274E) of this surface perturb binding of talin to the membrane, in agreement with experimental studies [12], [17], [20], [21]. This binding surface is consistent with NMR [10], crystallographic [12] and TIRF microscopy studies [21].
Experimental studies indicate that inserted loop in the F1 domain is dynamic in nature [17], [20]. Our simulations suggest that it provides a binding surface close to the cationic surface of F2–F3. In all simulations that yielded a ‘productive’ orientation of the talin head domain on the membrane, the F2–F3 pair always associated prior to F0–F1. This agrees with recent data that the association constant of the complete talin head domain with lipids is similar to that of the F2–F3 fragment [56]. We note that in kindlin, a homolog of talin that co-activates integrins, an even longer lysine-rich loop is inserted in the F1 domain. This lysine rich loop in kindlin is highly conserved and is believed to support binding to anionic phospholipid head groups [57].
Simulations that included the entire talin head/membrane/TM complex revealed that talin interactions with the integrin β tail and the membrane surface disrupt the interactions of both the OMC and IMC regions, resulting in eventual dissociation of the TM helices. Calculation of the hydrogen bonds between the lipids and the membrane (see Fig. S7C) after the disruption of interactions in the TM region show an increase in the number of lipid/talin hydrogen bonds. This suggests a stronger association between talin and the bilayer after formation of the talin/αβ complex. This could explain why the intact talin head domain has more dramatic effect than F2–F3 alone. The crystal structure of the integrin ectodomain and cysteine disulfide mapping of an intact integrin [58], [59] suggest that the α and β domains are in close proximity to one another in the inactive state. Thus a scissoring movement followed by dissociation of the two TM helices provides a plausible model for how the TM domain may trigger a conformational change in the ectodomain leading subsequent adoption of an extended active state.
From a more general perspective, this study reveals the interplay of membrane interactions and conformational changes involved in transmembrane signaling by receptors and associated proteins and/or domains. In particular, it may be compared with recent simulation studies, e.g. of the EGF receptor [42], which suggested that substantive repacking of the TM helix dimer and interactions between the intracellular kinase domain and anionic lipids play a key role in signaling across a membrane. The mechanisms in these two classes of membrane receptors (i.e. integrins and receptor tyrosine kinases respectively) may be compared with movements of TM helices thought to mediate signaling in GPCRs (as revealed by crystallographic, NMR and simulation studies [60]). One possible consequence of the extensive movements of TM helices is that signaling mechanisms are likely to be modulated by changes in (local) lipid bilayer properties [61]. This clearly merits further investigation by both computational and biophysical approaches.
The CG-MD simulations were performed using a local variant [31], [62] of the MARTINI forcefield [32]. A mapping of approximately 4∶1 heavy atoms to CG particles was used. Harmonic restraints (i.e. an elastic network model; ENM) between backbone particles within a cut-off distance of 7 Å was applied with a harmonic restraint force constant of 10 kJ/mol/Å2. In the tal-l25-CG and tal-l50-CG simulations (Table S1) the force constant for the ENM was set to 25 kJ/mol/Å2 and 50 kJ/mol/Å2 respectively and the cut-off distance was increased to 10 Å. The bilayer was constructed by self-assembly CG-MD simulations. In these simulations the lipids were placed randomly within a simulation box and solvated with CG water molecules and ions to neutralize the system. Subsequently, a production simulation was performed for 200 ns. After the first 10–15 ns of simulation the bilayer formed with an equal distribution of lipids in the two leaflets. For the simulation systems discussed here, two different bilayers were constructed. The first bilayer contained 832 zwitterionic POPC lipids and the second had 512 POPC and 320 POPG lipids (ratio of 3∶2). In the CG simulations the center of mass of the protein was placed 120 Å from the center of mass of the preformed bilayer (Fig. S3A). This starting distance between protein and bilayer was chosen to be much larger than the cut-off distance used for the electrostatic and the van der Waals terms in the CG forcefield. All systems were subsequently solvated with CG water molecules and neutralized with CG sodium particles, energy minimized for 250 steps and equilibrated for 5 ns with the protein Cα particles restrained (force constant 10 kJ/mol/Å2). Finally, CG-MD simulations were performed. The final snapshot of the CG-MD simulation was converted to an atomistic (AT) representation, using a fragment-based approach [33], for further refinement.
For the simulations with the talin/αβ complex, the same POPC/POPG bilayer as above was used. The TM region of the talin/αβ complex was inserted in the bilayer using GROMACS. The lipids that overlapped with the integrin TM region were removed. The same energy minimization and equilibration steps were performed as described above. A modified CG model was used where all the ENM restraints between the integrin α TM region and the rest of the complex were removed.
All CG-MD simulations were performed using GROMACS 4.5 (www.gromacs.org) [63], [64]. A Berendsen thermostat [65] was used for temperature coupling with a coupling constant of 1.0 ps and a reference temperature of 310 K. The Lennard-Jones and Coulombic interactions were shifted to zero between 9 Å and 12 Å, and 0 to 12 Å respectively. The time step was 20 fs. A Berendsen barostat was used for pressure coupling. The coupling constant was 1.0 ps, the compressibility was 5.0×10−6 bar−1 and the reference pressure was 1 bar.
The AT-MD simulations were performed using the GROMOS96 43a1 forcefield [66]. The Parrinello-Rahman barostat [67] and the Berendsen thermostat [65] were used for pressure and temperature coupling, respectively. The bond length was constrained using the LINCS algorithm [68] and the particle mesh Ewald (PME) algorithm [69] was used to model long-range electrostatic interactions. A cut-off distance of 10 Å was used for the van der Waals interactions. All the AT simulation systems were energy minimized using a steepest descent algorithm and equilibrated for 2.5 ns with the protein Cα atoms restrained (force constant 10 kJ/mol/Å2). Subsequently, unrestrained AT-MD simulations were performed. All the analyses were performed using GROMACS (www.gromacs.org) [63], [64], VMD [70] and locally written codes.
|
10.1371/journal.ppat.1004630 | Cell Cycle-Independent Phospho-Regulation of Fkh2 during Hyphal Growth Regulates Candida albicans Pathogenesis | The opportunistic human fungal pathogen, Candida albicans, undergoes morphological and transcriptional adaptation in the switch from commensalism to pathogenicity. Although previous gene-knockout studies have identified many factors involved in this transformation, it remains unclear how these factors are regulated to coordinate the switch. Investigating morphogenetic control by post-translational phosphorylation has generated important regulatory insights into this process, especially focusing on coordinated control by the cyclin-dependent kinase Cdc28. Here we have identified the Fkh2 transcription factor as a regulatory target of both Cdc28 and the cell wall biosynthesis kinase Cbk1, in a role distinct from its conserved function in cell cycle progression. In stationary phase yeast cells 2D gel electrophoresis shows that there is a diverse pool of Fkh2 phospho-isoforms. For a short window on hyphal induction, far before START in the cell cycle, the phosphorylation profile is transformed before reverting to the yeast profile. This transformation does not occur when stationary phase cells are reinoculated into fresh medium supporting yeast growth. Mass spectrometry and mutational analyses identified residues phosphorylated by Cdc28 and Cbk1. Substitution of these residues with non-phosphorylatable alanine altered the yeast phosphorylation profile and abrogated the characteristic transformation to the hyphal profile. Transcript profiling of the phosphorylation site mutant revealed that the hyphal phosphorylation profile is required for the expression of genes involved in pathogenesis, host interaction and biofilm formation. We confirmed that these changes in gene expression resulted in corresponding defects in pathogenic processes. Furthermore, we identified that Fkh2 interacts with the chromatin modifier Pob3 in a phosphorylation-dependent manner, thereby providing a possible mechanism by which the phosphorylation of Fkh2 regulates its specificity. Thus, we have discovered a novel cell cycle-independent phospho-regulatory event that subverts a key component of the cell cycle machinery to a role in the switch from commensalism to pathogenicity.
| The fungus Candida albicans is a commensal in the human microbiota, responsible for superficial infections such as oral and vaginal thrush. However, it can become highly virulent, causing life-threatening systemic candidemia in severely immunocompromised patients, including those taking immunosuppressive drugs for transplantation, sufferers of AIDS and neutropenia, and individuals undergoing chemotherapy or at extremes of age. With a rapidly increasing ageing population worldwide, C. albicans and other fungal pathogens will become more prevalent, demanding a greater understanding of their pathogenesis for the development of effective therapeutics. Fungal pathogenicity requires a coordinated change in the pattern of gene expression orchestrated by a set of transcription factors. Here we have discovered that a transcription factor, Fkh2, is modified by phosphorylation under the control of the kinases Cdc28 and Cbk1 in response to conditions that activate virulence factor expression. Fkh2 is involved in a wide variety of cellular processes including cell proliferation, but this phosphorylation endows it with a specialized function in promoting the expression of genes required for tissue invasion, biofilm formation, and pathogenesis in the host. This study highlights the role of protein phosphorylation in regulating pathogenesis and furthers our understanding of the pathogenic switch in this important opportunistic fungal pathogen.
| The fungus Candida albicans is commonly found as a harmless commensal on the skin and mucosal surfaces of the vaginal and gastrointestinal tracts of healthy people. However, it is also an opportunistic pathogen causing diseases that range from superficial infections, such as vaginal and oral thrush in otherwise healthy people, to life-threatening bloodstream infections that disseminate to internal organs in immunocompromised patients [1–3]. A key aspect of C. albicans pathogenicity is the capability to grow in both budding yeast and hyphal forms [4, 5]. When growing at low densities on mucosal surfaces C. albicans mostly exists as a commensal and is tolerated by the host immune system [6, 7]. Hyphal and pseudohyphal forms are found at sites of mucosal infections and are responsible for tissue invasion and damage [8, 9]. Hyphae preferentially invade epithelial cells, either by active penetration or host-mediated endocytosis [10–13]. Yeast cells in the bloodstream are engulfed by macrophages [14], but immediately switch to hyphal growth to escape and invade internal organs [15]. Hyphal forms are also a key part of the structure of biofilms [16]. Biofilm formation on the surfaces of implant medical devices has been recognized as a primary source of invading fungal cells, because biofilms provide protection against the host immune system and anti-fungal drugs [16].
Associated with the yeast-hyphal morphological switch, transcriptional changes occur resulting in the expression of proteins required for pathogenesis. This hyphal-specific gene set includes genes required for tissue damage, adhesion and invasion [17]. For example, they encode cell wall proteins such as Hyr1, secreted aspartyl proteases (SAPs) that cause tissue damage [18], and adhesins such as Als3 and Hwp1 that promote hyphal endocytosis by epithelial cells [19, 20]. Transcriptional responses on hyphal induction have been well studied, identifying many genes that are commonly up regulated during the yeast-hyphal switch [21–23].
Gene knockout studies have provided invaluable information on the molecular mechanisms underlying the morphological and transcriptional changes involved in C. albicans pathogenesis. This has led to the discovery that the cAMP-PKA-Efg1, MAPK-Cph1, and pH-responsive pathways play a key role in transcriptionally activating the hyphal program, along with the identification of several transcriptional repressors such as Nrg1, Tup1 and Sfl1 [5, 24]. Among the many hyphal-specific genes identified so far, only a few are required for hyphal formation. One example is HGC1, which encodes a cyclin homologous to the G1 cyclins Cln1 and Cln2 of the budding yeast Saccharomyces cerevisiae that partner the cyclin-dependent kinase (CDK) Cdc28 [25]. Cells lacking HGC1 are severely defective in hyphal morphogenesis under all conditions tested, and in causing infection in animals.
The discovery of the crucial role of Hgc1 and Cdc28 in C. albicans hyphal growth has uncovered multiple regulatory mechanisms involved in hyphal morphogenesis. Rga2 is a negative regulator of Cdc42, a Rho GTPase that orchestrates polarized growth processes at the hyphal tip [26]. Phosphorylation of Rga2 by Cdc28-Hgc1 inhibits its tip localization and keeps Cdc42 in the active state [27]. Cdc28-Cln3 regulates endocytic actin patch dynamics by phosphorylating Sla1, which leads to further phosphorylation by Prk1. Upon hyphal induction, Sla1 is rapidly dephosphorylated resulting in enhanced actin patch activity in hyphae [28]. Sec2 is a secretory vesicle-associated guanine-nucleotide-exchange factor (GEF) for the Rab GTPase Sec4. Phosphorylation of Sec2 by Cdc28-Hgc1 is necessary for its localization to the Spitzenkörper and correct hyphal growth [29]. Cdc28-Ccn1 acts in concert with the Gin4 kinase to phosphorylate a pair of serine residues of the septin Cdc11 within a few minutes of hyphal induction [30]. In the absence of this event, polarized growth is lost after the formation of the first septum.
Another kinase required for hyphal growth is the cell wall integrity kinase Cbk1 and its regulatory subunit Mob2. Cbk1 is a member of the evolutionary conserved Large Tumour Suppressor / Nuclear Dbf2 Related (LATS/NDR) superfamily of kinases that are involved in control of cell shape and growth [31]. In C. albicans loss of Cbk1 completely abrogates germ tube formation and polarized growth, disturbs cell separation in yeast cells and reduces expression of hyphal specific genes [32, 33]. Defects in polarised growth are seen when its homologue is lost in other fungi [34]. For example, orb6 mutants in Schizhosaccharomyces pombe [35], cot1 mutants in Neurospora crassa [36] and cbk1 mutants in S. cerevisiae [37, 38] all show profound defects in polarised growth. Cbk1 forms part of the regulation of Ace2 activity and cellular morphogenesis (RAM) network of physically interacting proteins including its activating subunit Mob2, the kinase Kic1, scaffolding proteins Tao1 and Hym1, and the RNA binding protein, Ssd1 [39]. As well as polarised growth the RAM network is also required for cell separation [37, 39]. It phosphorylates the transcription factor Ace2, which then translocates to daughter cell nuclei and transcribes genes that encode hydrolytic enzymes to degrade the primary septum. In C. albicans the mechanism of Cbk1 action in hyphal growth and the target proteins it phosphorylates remains largely unknown. One known target of Cbk1 is the transcription factor Bcr1, phosphorylation of which promotes biofilm formation [40].
Clearly, protein phosphorylation plays a key role in morphogenesis and pathogenesis in C. albicans. To further investigate the role of phospho-regulation in hyphal growth, we searched for proteins that showed a hyphal-specific pattern of phosphorylation. We observed that the fork-head family transcription factor Fkh2 changes its phosphorylation profile dramatically within five minutes of hyphal induction. In S. cerevisiae, two fork-head transcription factors, Fkh1 and Fkh2, control a G2 transcription program including the expression of the G2 cyclin Clb2 required for mitotic entry [41], and the transcription factors Swi5 and Ace2 required for the M to G1 phase gene expression program. We show here that upon hyphal induction in C. albicans Fkh2 undergoes a radical shift in its phosphorylation profile mediated by two kinases: Cdc28-Ccn1/Cln3 and Cbk1-Mob2. This shift specifically activates Fkh2 to promote the expression of genes required for pathogenic processes in addition to its normal general housekeeping function. Thus, phosphorylation not only plays a part in promoting polarized growth but also has a role in a novel regulatory circuit that activates hyphal-specific gene transcription necessary for pathogenesis.
To identify further targets of Cdc28 in hyphal development, we first identified C. albicans proteins which contain a cluster of the consensus Cdc28 target motifs, S/TPxK/R (x, any amino acid). We then used a band shift assay in one dimensional polyacrylamide gel electrophoresis (1D PAGE) to determine whether any of these proteins were differentially phosphorylated in hyphae compared to yeast. In addition to the proteins described in the introduction we identified changes in the phosphorylation profile of Orf19.3469 (S1 Fig.), a possible homolog of the S. cerevisiae Stb1 protein that regulates the MBF transcription at START [42], Orf19.1948 (S1 Fig.), a protein of unknown function, and Fkh2 which is known to play a key role in cell cycle progression (S1 Fig.). Here we report our analysis of Fkh2 phosphorylation and its cell-cycle independent role in promoting the expression of genes involved in pathogenesis. Fig. 1 presents an experiment where early G1 yeast cells expressing Fkh2-YFP were collected by elutriation and then reinoculated either into yeast growth conditions (YEPD and 30°C, pH 4.0) or hyphal growth conditions (YEPD plus 10% serum and 37°C, pH 7.0). In yeast growth conditions, the appearance of small buds, large buds and binucleate cells was recorded and plotted against time (Fig. 1A). In hyphal cells we plotted germ tube emergence, the appearance of a septin ring within the germ tube as visualized by Cdc12-mCherry fluorescence, nuclear migration as visualised by DAPI staining, and the appearance of binucleate cells (Fig. 1B). A change in the phosphorylation profile of Fkh2 was indicated by the appearance of a double band and the disappearance of the slower migrating band upon phosphatase treatment (Fig. 1C). (Note in Fig. 1A the septin Cdc11 was used as a loading control whereas in Fig. 1B the loading control was Cdc28/Pho85 identified by a monoclonal anti-PSTAIRE antibody). In yeast cells, the Fkh2-YFP band became double after the appearance of small buds (Fig. 1A), consistent with phosphorylation in S-phase as previously documented in S. cerevisiae [43]; this then collapsed to one band when the cells became bi-nucleate. In contrast, Fkh2 was present as a double band from 20–60 min after hyphal induction, well before the appearance of the septin ring (Fig. 1B), which marks the start of the cell cycle [44]. Cdc28, partnered by the cyclin Ccn1, and in conjunction with the Gin4 kinase, has been shown to phosphorylate the septin Cdc11 within 5 min of hyphal induction [30]. To determine if Fkh2 is similarly targeted at this early stage, we repeated the experiment collecting samples at 5-min intervals after hyphal induction. Fkh2 showed an additional retarded band after 5 min (Fig. 1D). Thus, whereas Fkh2 is phosphorylated in S-phase in yeast cells, it is rapidly phosphorylated upon hyphal induction in a cell cycle-independent fashion.
In order to generate a more detailed picture of Fkh2 phosphorylation, we carried out two-dimensional (2D) protein electrophoresis using an immobilised pH gradient (IPG) of 3–10 for isoelectric focussing. Quantitative intensity profiles were generated and are displayed above each of the resulting autoradiograms (Fig. 1E). On phosphatase treatment Fkh2 is only present as a single spot at the basic end of the IPG, representing Fkh2 without any negative charges added due to phosphorylation. In contrast to the single band observed in 1D gels in stationary phase, these data showed that there is a diverse pool of differentially charged Fkh2 phospho-isoforms. Five minutes after hyphal induction this profile begins to change and after 40 min of hyphal growth this profile is transformed, but after 80 min it resembles cells growing in the yeast form. A similar change is not observed in cells growing in the yeast morphology 40 min after reinoculation of the stationary phase culture. S2 Fig. shows an independent replicate of this experiment to demonstrate the reproducibility of this key observation. Other 2D gels described in this paper show a similar degree of reproducibility. Thus, shortly after hyphal induction there is a window in which the spectrum of Fkh2 phospho-isoforms is transformed before resuming the characteristic yeast profile.
To identify the phosphorylated residues on Fkh2 during the window where the shift in the phosphorylation pattern is observed, Fkh2-HA was immuno-precipitated from a hyphal culture 40 min after induction and subjected to phospho-site mapping by mass spectrometry (MS). Six residues were identified with high confidence at four full Cdc28 consensus sites in the C-terminal region and two minimal sites (S/TP), one of which was also in the C-terminal region (Fig. 2A). The full results of the phospho-site mapping are shown in S3 Fig. and S1 Dataset Fkh2-YFP immuno-precipitated from a hyphal culture was detected by an antibody that recognizes phosphorylated serine in the context of a full Cdc28 target sequence (SPxK/R) in a Western blot (Fig. 2B), thus providing further evidence that these Cdc28 target sites are phosphorylated. To test this conclusion more fully and to investigate the physiological role of the phosphorylation, we constructed strains expressing mutant versions of Fkh2 that had the six MS-identified and other potential phospho-acceptor residues replaced by either the non-phosphorylatable alanine (A) or the phosphomimetic glutamate/aspartate (E/D) residues in the following combinations: 1) Fkh2(6AMS) had all six MS-identified sites substituted with alanine; 2) Fkh2(6A) or Fkh2(6DE) carried the indicated substitutions at the six Cdc28 consensus sites including the four C-terminal sites identified by MS and additional two full Cdc28 sites not detected by MS (Fig. 2A); 3) Fkh2(10A) carried the indicated substitutions at all the full and minimal Cdc28 sites C-terminal to the DNA binding domain; 4) Fkh2(15A) or Fkh2(15DE) harboured the indicated substitutions at all the full and minimal Cdc28 sites across the whole protein; and 5) Fkh2(1–426) was a truncated version with the C-terminal domain, containing five full and five minimal Cdc28 sites, removed. In each case, the mutant protein was C-terminally fused to GFP. We confirmed that the Fkh2-GFP protein was functional, because the fkh2/FKH2-GFP strain did not show the fkh2ΔΔ phenotype previously described [45] and also shown below in this study. We also confirmed that the mutant proteins were present in similar levels to the wild-type protein (S3D,E Fig.), or in the case of Fkh2(1–426) the protein more abundant than the wild-type protein (Fig. 2D).
To test the hypothesis that Cdc28 directly phosphorylates Fkh2 at these sites, we used an in vitro kinase assay to demonstrate that immuno-purified Cdc28-HA can phosphorylate an E. coli-expressed recombinant C-terminal GST-Fkh2 fragment (Fig. 2C); however, the C-terminal fragment carrying the serine/threonine to alanine substitutions in the Cdc28 consensus sites was not a substrate as predicted by this hypothesis. We then examined whether these phospho-acceptor substitutions affected Fkh2 phosphorylation in vivo. The Fkh2(1–426)-GFP C-terminal truncation did not show the characteristic double band of the wild-type protein, suggesting that phosphorylation of the cluster of C-terminal Cdc28 target sites contribute to the band shift (Fig. 2D). 1D gels of the mutants showed that Fkh2(6AMS) was still phosphorylated, by the presence of a more retarded isoform (S3D Fig.). The Fkh2(6A), Fkh2(10A) and Fkh2(15A) proteins were only present as one phospho-isoform on 1D-PAGE; whereas the Fkh2(6DE) and Fkh2(15DE) proteins still showed a more retarded phospho-isoform (S3E,F Fig.). We used 2D gels to further examine the phosphorylation profile of the above phospho-site mutants in stationary phase, growing yeast cells, and hyphal cells early after induction. The results are shown in Fig. 2E. In each panel the profile of the wild-type Fkh2 protein in the corresponding culture condition is shown as a grey dashed line for comparison. The Fkh2(6A) mutant showed only three major spots in both stationary phase, growing yeast and early hyphal growth in contrast to the more complex pattern of the parental cells (Fig. 2E). Thus, Fkh2 is phosphorylated at these sites in both yeast and hyphae and requires this phosphorylation to show the characteristic early hyphal profile. The Fkh2(15A) mutant, which lacks all 15 possible Cdc28 target sites, shows little, if any evidence of phosphorylation in stationary phase and in yeast growth conditions. However, 40 min after hyphal induction additional peaks are present providing clear evidence of phosphorylation events that are specific to the early period of hyphal induction. Nevertheless, the profile is different from the characteristic early hyphal pattern showing that the Cdc28 sites are required for the transition to the early hyphal profile. These peaks could represent additional cryptic Cdc28 target sites that are specific to the early hyphal form or they could indicate the action of one or more additional kinases. Like the Fkh2(6A) mutant, the Fkh2(6DE) mutant is also present as three spots in stationary phase (Fig. 2E), but they are shifted to the more acidic end of the pH gradient due to the negative charge on the glutamate/aspartate residues. Interestingly, the Fkh2(6DE) mutant is able to undergo further phosphorylation on hyphal induction to generate the hyphal specific profile. This suggests that a second kinase acts on Fkh2 after hyphal induction in addition to phosphorylation at the Cdc28 consensus sites and is consistent with the evidence from the Fkh2(15A) mutant, which is still present as multiple phospho-isoforms even though all the potential Cdk1 target sites are blocked in this mutant.
We next examined the Fkh2 2D profile in the cdc28-as1 mutant, and in mutants in which one of the Cdc28 G1 cyclins was either deleted (hgc1ΔΔ and ccn1ΔΔ) or down regulated using the MET3 promoter (CLN3-sd) (Fig. 3). The results showed that both inhibition of Cdc28 or lack of either Ccn1 or Cln3 prevented the shift to the hyphal profile, but there was less of an effect on the early hyphal profile in cells that lack Hgc1. Thus, the data support the conclusion that Cdc28 acts on Fkh2 early after hyphal induction, but suggest both Cln3 and Ccn1 are required to partner Cdc28. Inhibition of Cdc28 or lack of Cln3 also had a major effect on the Fkh2 profile in growing yeast cells, similar to the effect of the Fkh2(6A) mutant in these cells. Thus, while Cdc28 mediated phosphorylation is necessary for the transition to the early hyphal pattern, Fkh2 is also phosphorylated by Cdc28 during yeast growth, as would be expected from previous studies in S. cerevisiae.
To investigate the physiological role of Fkh2 phosphorylation after hyphal induction, we used microarrays to compare the transcriptome of the fkh2(6A) and fkh2ΔΔ mutants with their respective parental strains (fkh2(6A)-GFP versus FKH2-YFP and fkh2ΔΔ versus FKH2/FKH2) (Figs. 4, 5 and S2 Dataset). Previous microarray analysis with the fkh2ΔΔ mutant used a limited array containing probes for only 319 C. albicans ORFs, thus likely under-representing the transcriptional role of Fkh2 [45]. To provide a more complete analysis of the role of Fkh2, we also carried out microarray analysis using the above strains in exponentially growing yeast cultures as well as hyphal cultures. We also examined the effect of over-expressing FKH2 driven by the GAL1 promoter in yeast cultures, since overexpression of FKH2 has been shown to induce hyphal-like growth [46]. Full microarray results can be found in S2 Dataset.
Many genes known to be up regulated during hyphal growth and pathogenesis were found to be down-regulated in both the fkh2ΔΔ and fkh2(6A) mutants (Fig. 4). These included HYR1 and ECE1, which have been previously shown to be down-regulated in the fkh2ΔΔ strain [45]. In addition, the secreted aspartyl protease genes SAP4 and SAP6, the chitinase gene CHT2, and the hyphal-specific kinesin-like protein gene KIP4 were also down-regulated in both the fkh2(6A) and fkh2ΔΔ mutants (Fig. 4). The overlap between the sets of genes down regulated in the two strains is summarised by the Venn diagram in Fig. 5A. However, comparing the gene sets in this way is potentially misleading for two reasons. First, while the degree of down-regulation may be greater than an arbitrary threshold in both mutant strains, the degree of down-regulation may actually be significantly different: e.g. 2.1-fold down in one strain compared to 10-fold down in the second strain. Second, the degree of down-regulation may actually be quite similar, but exceeds the arbitrary threshold in one strain but just fails to exceed the threshold in the second strain: e.g. 1.9-fold compared to 2.1-fold down-regulated. For this reason we compared the fold change for every gene in the fkh2(6A) and fkh2ΔΔ data sets to determine which genes showed a significantly greater degree of down regulation in each of the two mutants (S2 Dataset). The results are summarised in Fig. 5B together with the Gene Ontogeny (GO) processes co-ordinately affected in each mutant together with the GO processes of those genes that were significantly more affected in one mutant compared to the other.
Of the 237 genes down-regulated in fkh2ΔΔ, the change was significantly greater in fkh2ΔΔ than in fkh2(6A) cells in 201 cases (FDR threshold 0.05, empirical Bayes moderated t-statistic). Of the 67 genes down-regulated in the fkh2(6A) mutant, the change was significantly greater in this mutant than in fkh2ΔΔ in 37 cases (FDR threshold 0.05, empirical Bayes moderated t-statistic). GO analysis showed that the genes only affected in the fkh2ΔΔ strain during hyphal growth were associated with a wide range of metabolic processes including sterol and ergosterol biosynthesis, ion homeostasis and filamentous growth (Fig. 5B). In contrast, GO analysis showed that genes down-regulated in the fkh2(6A) mutant were involved in only biofilm formation and biological adhesion (Fig. 5B). Importantly, HGC1, the Cdc28 cyclin essential for hyphal development [25], and SUN41, which is critically required for hyphal and biofilm formation [47], are only down-regulated in the fkh2(6A) mutant (Fig. 4 and S2 Dataset). We confirmed the down-regulation of HGC1, SAP4, KIP4 and ECE1 in the fkh2(6A) mutant by qPCR (Fig. 5C). We also showed that in the fkh2(6DE) mutant the expression levels of these genes were near wild-type levels with the exception of HGC1, whose expression also appears reduced, but not to the extent as seen in the fkh2(6A) strain. Thus, correct phospho-regulation of Fkh2 at the six Cdc28 consensus sites is required to specifically activate a subset of genes that are associated with interaction with the host.
There was also a limited overlap between the sets of genes up-regulated during hyphal growth in the fkh2ΔΔ mutant compared to the fkh2(6A) mutant. However, the GO analysis indicated fewer processes being co-ordinately affected. In the fkh2ΔΔ mutant the genes up-regulated were involved in GO processes: oxidative reduction and cellular response to oxidative stress. There were no GO processes that were significantly over-represented in the set of genes up-regulated in the fkh2(6A) mutant during hyphal growth. However, PDE1, CLB4 and RAD6 were all up-regulated in both the fkh2(6A) and the fkh2ΔΔ mutants. PDE1 encodes a phosphodiesterase that hydrolyses cAMP, thus inhibiting the major signalling pathway for hyphal morphogenesis. Over-expression of CLB4 may promote the non-polar growth that is characteristic of G2 cyclin mutants. RAD6 is known to be a negative regulator of hyphal growth [48]. The effect of the fkh2ΔΔ mutation was again greater than the fkh2(6A) allele during yeast growth. In the fkh2ΔΔ cells 125 genes were down-regulated and 219 genes up-regulated. In contrast, in fkh2(6A) cells only 37 genes were down-regulated and 42 genes up-regulated. Only 14 genes were down-regulated and 13 genes up-regulated in both cell types (S2 Dataset). There were 24 genes that were down-regulated only in the fkh2(6A) mutant. However, there were no GO processes co-ordinately affected in this gene set.
Overexpression of FKH2 from the GAL1 promoter resulted in a filamentous phenotype as previously reported [46] (S4 Fig.). Close inspection suggested that these were not true hyphae, but rather resembled the phenotype of cells blocked in the cell cycle by treatments such as hydroxyurea. DAPI staining revealed that elongated daughter cells were often anucleate or contained fragmented nuclei. Microarray analysis revealed down-regulation of the CDC14 phosphatase and the Gin4 kinase both of which are required for cell cycle progression (S2 Dataset). Furthermore, DAM1 and ASK1 were also down-regulated. These genes encode a complex that is required for the coupling of kinetochores to microtubules. Thus, down-regulation of these genes required for progress through mitosis provides a plausible explanation for the filamentous phenotype observed upon FKH2 overexpression.
One explanation for the altered pattern of gene expression in the fkh2(6A) mutant is that this mutant is perturbed in cell cycle progression in hyphae. To address this possibility we quantified the nuclear distribution in fkh2(6A) and parental BWP17 cells at 30-min intervals after stationary phase yeast cells were inoculated into hyphal inducing conditions. To do this, we characterised the developing hyphae according to whether they contained a single nucleus in the mother cell, a single nucleus that was in the process of migrating into the developing germ tube, two nuclei thus having finished the first mitosis or three nuclei thus having completed the second mitosis (after the first mitosis the nucleus that returns to the mother cell does not re-enter the cell cycle [49]) (Fig. 6). Nuclear migration commenced 90 min after induction in both wild-type and mutant cells (Fig. 6A). fkh2(6A) cells did show a delay in completing the first mitosis (Fig. 6A). However, by 180 min, when the samples for microarray analysis were harvested, fkh2(6A) cells had completed the first mitosis and the nuclear distribution of these cells was essentially identical to parental fkh2/FKH2 cells; indeed approximately 20% of both parental and mutant cells had completed the second mitosis (Fig. 6A–C). Fig. 6B,C also shows the cell cycle distribution of fkh2(6DE), fkh2(1–426) and fkh2ΔΔ mutants 180 min after hyphal induction. The fkh2(6DE) and fkh2(1–426) mutants also showed a similar distribution to parental cells except that fewer of these cells had completed the second mitosis. In contrast, the fkh2ΔΔ mutant cells failed to form normal hyphae. The cells were swollen and the nuclear distribution was grossly abnormal with some cells containing two nuclei and other cells containing no nuclei (Fig. 6C). Thus, while Fkh2 is essential for normal cell cycle progression, mutations affecting the C-terminal domain have only a mild effect.
Microarray analysis revealed that the fkh2(6A) mutant is defective in the expression of genes that have been associated with pathogenic processes and host interaction. We therefore went on to characterise the phenotype of the fkh2(6A), fkh2(6DE) and fkh2(1–426) mutants, to investigate whether they are defective in the corresponding pathogenic processes. The fkh2(6A) mutant initially formed hyphae normally as shown by the images after 180 min (Fig. 6C); however, after 6 h of hyphal induction in liquid culture hyphae displayed an increased branching frequency and subtle morphological abnormalities such as swelling of the hyphal tip, not seen in the wild-type cells (Fig. 7A–C). These abnormalities were reduced in the phosphomimetic fkh2(6DE) mutant (Fig. 7A–C). However, the fkhΔΔ mutant showed pseudohyphal-like growth in yeast growth conditions and more severe hyphal defects than the fkh2(6A) mutant (Fig. 7A). On solid Spider medium, both the fkh2(6A) and fkh2(6DE) mutants failed to produce the wrinkled colony morphology normally seen in the wild-type cells, while the fkh2(1–426) mutant showed reduced wrinkling (Fig. 7D). Importantly, the fkh2(6A) mutant was defective in invasion of the agar substratum in a wash-off test, whereas the fkh2(6DE) and fkh2(1–426) strains showed a similar invasive capacity to the fkh2/FKH2 strain (Fig. 7D). The fkh2ΔΔ strain grew poorly on Spider medium and failed to invade the agar. Taken together, these observations suggest that the correct phospho-regulation of Fkh2 is required for long-term hyphal maintenance and invasive growth.
The microarray analysis of the fkh2(6A) mutant showed defects in the induction of genes involved in biofilm formation, interaction with host cells, and activation of host immune response. Fkh2ΔΔ cells were unable to form biofilms (Fig. 8A,B). The biofilm matrix was visibly reduced in the fkh2(6A) strain and the average biofilm mass was only half that of the fkh2/FKH2 and BWP17 parental strains (Fig. 8A,B). The fkh2(6DE) and fkh2(1–426) also showed biofilm formation defects, but these were not as severe as those observed in the fkh2(6A) strain (Fig. 8A,B). Furthermore, the fkh2(6A) mutant was markedly defective in causing tissue damage as measured by a reduction in lactate dehydrogenase release from damaged cells in a TR146 oral epithelial monolayer infection model (Fig. 8C). Cells lacking Fkh2 were completely unable to cause damage, as would be expected from the deleterious phenotype. There was also a significant reduction in the levels of the interleukins IL1-α and IL-1β released when fkh2(6A) was used in the infection model (Fig. 8D–E). This suggests that in the absence of Fkh2 phosphorylation C. albicans elicits a reduced immune response. The above data show that the changes in gene expression resulting from loss of Fkh2 phosphorylation do indeed have corresponding effects on pathogenic processes. Thus, the change in Fkh2 phosphorylation upon hyphal induction is required to positively regulate multiple virulence mechanisms.
To further investigate the physiological role of Fkh2 phosphorylation, we first examined whether it affected Fkh2 nuclear localisation and found that the Fkh2(6A), Fkh2(6DE) and Fkh2(1–426) mutant proteins all localised to the nucleus in the same way as the wild-type protein (Fig. 9A). Next we investigated which proteins interacted with Fkh2 early upon hyphal induction and whether any such interactions were altered by Fkh2 phosphorylation. To do this, Fkh2 was immuno-precipitated from a strain expressing Fkh2–6Myc and fractionated by SDS-PAGE; the proteins revealed by Coommassie Blue staining were then identified by MS. Most of the bands were found to be proteins already present in our database of common contaminants, but one band was identified as a mixture of the Candida orthologues of ScSrp1 and ScPob3. ScSrp1 is a karyopherin [50], while ScPob3 is a member of the facilitates chromatin transcription (FACT) nucleosome remodelling complex [51]. To verify this finding, we constructed strains co-expressing Pob3-HA or Srp1-HA with Fkh2-YFP and determined whether the proteins could co-immunoprecipitate. Western blot analysis showed that when Fkh2-YFP was immuno-precipitated, a more intense band corresponding to Pob3-HA can be detected in a strain co-expressing Fkh2-YFP and Pob3-HA, than in a strain only expressing Pob3-HA (Fig. 9B). A band corresponding to Srp1-HA could be detected at the same intensity in a strain only expressing Srp1-HA as well as in the co-expressing strain (Fig. 9B), suggesting the Fkh2-Srp1 interaction may be non-specific. Further investigation of the Fkh2-Pob3 interaction showed that when Fkh2(6A)-GFP was immuno-precipitated the amount of Pob3-HA co-precipitated was reduced by half in a co-expressing strain (Fig. 9C). However, the co-immunoprecipitation of Pob3 was present when Fkh2(6DE)-GFP was used as the bait, suggesting that the interaction of Pob3 with Fkh2 is phosphorylation dependent. Thus we have uncovered a possible mechanism through which the phosphorylation of Fkh2 on hyphal induction could bring about the changes in gene expression required for virulence, through mediating the interaction of Fkh2 with Pob3.
Residual phosphorylation was observed in the Fkh2(15A) and Fkh2(6DE) mutants during early hyphal growth and in cdc28-as1 cells upon the addition of inhibitor (Fig. 2E). This suggests that an additional kinase may be targeting Fkh2. We used 1D gels to profile Fkh2-GFP in a number of kinase mutants and found an altered profile in cells lacking the Cbk1 kinase (S5 Fig.). We therefore used 2D gels to profile Fkh2 in cbk1ΔΔ cells and found that Fkh2 failed to show the characteristic early hyphal profile (Fig. 10A). Interestingly, although the stationary phase profile was also altered, the profile of growing yeast was largely unaltered. S. cerevisiae Cbk1 has been shown to preferentially target serine or threonine with histidine at-5, arginine at-3 and serine at-2 [52]. Such a site is present in the C-terminal domain of CaFkh2 (528-HSRSTS-533). We constructed a strain expressing Fkh2 with a non-phosphorylatable substitution at this site, Fkh2(S533A)-GFP, and used 2D gels to characterise this protein in stationary phase yeast, growing yeast, and early hyphal cells (Fig. 10A). We found that the Fkh2(S533A)-GFP profile was different from the wild type profile at 40 minutes after hyphal induction, thus supporting the conclusion that Cbk1 targets Fkh2 at this site and that this phosphorylation is necessary for the transit to the early hyphal profile. To provide further evidence that Cbk1 targets Fkh2, and to test whether the action is direct, we carried out an in vitro kinase assay. To do this, we used Mob2-HA to immunoprecipitate the Cbk1-Mob2 kinase from an early hyphal lysate and recombinant Fkh2 C-terminal domain fused to GST (GST-Fkh2(CT)) as substrate. Fig. 10B shows that the immune-precipitated Mob2-HA can indeed phosphorylate GST-Fkh2(CT) in vitro. Only background levels of signal resulted from negative controls lacking either immuno-precipitated Mob2-HA or the immuno-precipitation products from a mock lysate. Thus, the reaction is specific to Mob2-Cbk1 and is not due to a co-purifying kinase. The fkh2(S533A) strain had a normal yeast morphology but showed morphological defects on hyphal induction, thus showing that phosphorylation of Fkh2 S533 by Cbk1-Mob2 is necessary for normal hyphal development (Fig. 10C).
The Fork-head transcription factors Fkh1 and Fkh2 have been extensively studied in S. cerevisiae where they regulate the G2 transcription program, including the expression of the G2 cyclin gene CLB2 required for mitotic entry, and the transcription factor genes SWI5 and ACE2 required for the M to G1 phase gene expression program [53–56]. In S. cerevisiae, Fkh2 is phosphorylated by Cdc28-Clb5 during S-phase to activate transcription at the promoters of CLB2 cluster genes [43]. Here we show that Fkh2, the single ScFkh1 and ScFkh2 homologue in C. albicans, undergoes a radical change in its phosphorylation profile within 5 min of hyphal induction. Previously, it has been reported that spindle pole body duplication, which marks the start of the cell cycle, is coincident with the appearance of the septin ring within the germ tube [44]. In the experiments described in Fig. 1, septin rings are only present in hyphae after 60 min, and yet the change in phosphorylation was apparent as early as 5 min, and is disappearing by 60 min. Clearly, Fkh2 phosphorylation triggered by hyphal induction of early G1 cells occurs far before START and is thus cell cycle-independent. In contrast, Fkh2 phosphorylation in yeast cells occurs after bud emergence, consistent with the timing of S-phase phosphorylation reported in S. cerevisiae [43].
We initially identified the change in Fkh2 phosphorylation status using 1D gels which suggested that Fkh2 was not phosphorylated in stationary phase but was rapidly phosphorylated upon hyphal induction before the start of the cell cycle. In contrast, 1D gels suggested that Fkh2 in yeast cells was phosphorylated at a later time and only after the start of the cell cycle. However, the increased detail provided by 2D protein gels showed a more complex pattern and revealed that both growing and stationary phase yeast cells contained a complex pool of Fkh2 phospho-isoforms. Importantly, this complex profile underwent a radical, but transient transformation upon hyphal induction.
Mass spectrometry identified that Fkh2 was phosphorylated as sites corresponding to the full Cdc28 consensus target sites. Non-phosphorylatable substitutions at the six full target sites reduced the number of 2D gel peaks in the profile of both yeast and hyphal cells and the shift to the characteristic hyphal profile was no longer evident. However, the transition to the early hyphal profile occurred in Fkh2 carrying phosphomimetic substitutions at these six sites. Thus, phosphorylation of these residues is necessary for the shift to the hyphal profile. Non-phosphorylatable substitutions at all 15 potential full and minimal Cdc28 target sites essentially eliminated all peaks in stationary phase and growing yeast cells, but clear peaks remained in early hyphal cells suggesting the action of a second kinase, the action of which is specific to early hyphal cells. We showed that a strong candidate for this second kinase is Cbk1-Mob2 acting on the S533 residue. An in vitro kinase assay and the altered Fkh2 phosphorylation profile in 2D gels when either Cdc28 or Cbk1 activity was inhibited both add support to the conclusion that both Cdc28 and Cbk1 kinases act on Fkh2. We investigated the likely cyclin partner for Cdc28 using mutants in which G1 cyclins were either deleted or depleted. The results suggested that both Cln3 and Ccn1 may be required for the switch to the hyphal-specific pattern. Interestingly, previous reports have shown that cells lacking either Cln3 or Ccn1 initiate hyphal formation but are unable to maintain hyphal growth long term, phenotypes that are reminiscent of the fkh2(6A) phenotype which is also unable to maintain the hyphal state [57–59]. In addition, Cdc28-Ccn1 in collaboration with Gin4 phosphorylates the septin Cdc11 on adjacent residues [30]. When this phosphorylation is prevented, hyphal germ tubes form normally but the tip swells and polarized growth ceases after the formation of the first septum.
Taken together these data show that Fkh2 exists as multiple phosphorylated isoforms in both yeast and hyphal growth modes and that Cdc28 is likely to target six key residues and possibly up to a total of 15 residues. However, immediately after hyphal induction there is a transient shift in the phosphorylation profile which is dependent on phosphorylation at the six full Cdc28 target sites. Thus while Fkh2 is phosphorylated in both yeast and early hyphal cells, the phosphorylation state of Fkh2 is qualitatively different at early times after hyphal induction compared to either stationary phase yeast or growing yeast cells. It is important to bear in mind that the complex 2D profiles show that there is a pool of differentially phosphorylated phospho-isoforms of Fkh2 in stationary phase cells, growing yeast and hyphae. With six possible sites that we have shown to be phosphorylated, there are 64 different possible phosphorylated states (26). With 15 possible sites the number of different phosphorylation states is much larger. Our data do not allow us to definitively resolve whether unique sites are phosphorylated in hyphae compared to yeast. However, it is clear that the spectrum of phosphorylated isoforms changes rapidly on hyphal induction in a way that is not observed in yeast, and that this rapid change contributes towards the induction of genes that are required for pathogenesis.
The characteristic early hyphal profile may reflect a change in the spectrum of phosphorylated states of Cdc28 target sites, but the change also depends on the action of Cbk1-Mob2. The change in the phosphorylation pattern of Fkh2(6DE), but not Fkh2(6A), upon hyphal induction suggests that phosphorylation at Cdk1 sites may prime Fkh2 for further phosphorylation by a second kinase. While it is attractive to speculate that Cbk1 phosphorylation is dependent on Cdk1 phosphorylation, we have not addressed this issue experimentally so there is no direct evidence that this is the case. Relatively little is known about the action of Cbk1 and its regulatory subunit, Mob2, despite it being absolutely required for hyphal growth [32]. Moreover, its action is necessary for the expression of hyphal-specific genes [33], consistent with the idea that phosphorylation of Fkh2 by Cbk1 acts to redirect the specificity of Fkh2 to promote hyphal gene expression. Interestingly, Cbk1 has recently been shown to phosphorylate Bcr1, a key transcription factor in biofilm formation [40]. Here we demonstrate another mechanism by which Cbk1 positively regulates biofilm formation, through Fkh2, suggesting that Cbk1 has a general role in promoting biofilm formation.
Identification of the sites targeted by Cdc28-Ccn1/Cln3 allowed us to investigate the physiological role of Fkh2 phosphorylation. Transcript profiling showed that the fkh2(6A) mutant is defective in the expression of genes involved in important aspects of C. albicans pathogenicity. In contrast, fkh2ΔΔ mutants are defective in a wide range of metabolic functions. Moreover, the gene set whose expression is reduced in fkh2ΔΔ shows only a partial overlap with the loss of functions that result from non-phosphorylatable alanine substitutions at the six Cdc28 target sites. Thus, while we confirmed a previous report that Fkh2 plays a role in the expression of hyphal-specific genes [45], we have also revealed that phosphorylation of Fkh2 is specifically required for the expression of genes that promote pathogenesis within the host. These include: SAP4,6 which are expressed during mucosal and systemic infections for nitrogen utilisation from host proteins; HYR1 which encodes a GPI anchored cell wall protein required for protection against killing of C. albicans by neutrophils [60]; HGC1, which is required for hyphal growth, biofilm formation, and virulence in a mouse model of systemic infection [25]; and SUN41 which is required for biofilm formation [47]. Expression of these genes has previously been shown to be regulated through the cAMP-PKA and MAPK pathways. Thus we have identified a new pathway, through Cdc28-Ccn1/Cln3 and Cbk1 that acts to positively regulate hyphal gene expression alongside the aforementioned pathways. Importantly, we have also shown that the change in gene expression observed in the fkh2(6A) mutant was associated with a corresponding reduction of C. albicans virulence functions such as: hyphal maintenance, host cell damage, biofilm formation and host immune response activation. Our work suggests that Cdc28-Ccn1/Cln3 positively regulates HGC1 expression on hyphal induction via Fkh2. This early activation of HGC1 expression through Fkh2 may be a key role of Ccn1 and Cln3 on hyphal induction, and thus could explain why Fkh2 is only phosphorylated for a short period on hyphal induction.
Here we have identified that Fkh2 physically associates with Pob3 in a manner that is dependent on Fkh2 phosphorylation. Pob3 in S. cerevisiae is a subunit of the heterodimeric FACT complex that reorganizes nucleosomes to allow access of RNA polymerase II and DNA polymerase to promoters, thus promoting transcription initiation and DNA replication [51]. Interestingly, a role for ScFkh2 in controlling origin firing has recently been reported [61]. The fkh2(6A) mutant initiates germ tube formation normally, but fails to maintain hyphal growth and is defective in agar invasion. Chromatin remodelling has been shown to play a key role in hyphal maintenance [62, 63]. Significantly, we observe a peak in Fkh2 phosphorylation after 40 min of hyphal induction, which is consistent with the timing of chromatin reorganisation that results in the ejection of the repressor Nrg1 from the promoter of hyphal-expressed genes, a requirement for the long-term maintenance of hyphal growth [62, 63]. This suggests that phosphorylation of Fkh2 on hyphal induction may be required to direct Pob3 to the promoters of hyphal specific genes, in order to regulate the remodelling events that are required to sustain hyphal and pathogenic gene expression.
In this study, we identified how the C. albicans Fkh2 transcription factor has evolved to acquire a new function in addition to its conserved housekeeping role. CaFkh2 switches from cell cycle-dependent phospho-regulation during yeast growth, to a specific early phospho-regulatory event during hyphal growth that is independent of the cell cycle. This early hyphal phosphorylation is necessary for the positive regulation of genes involved in invasive growth and pathogenesis. Therefore, we can propose a new mechanism, by which C. albicans specifically modifies a key component of its cell cycle transcription machinery by phosphorylation, in the switch from commensalism to pathogenicity.
C. albicans cells were routinely grown in 1% yeast extract, 2% peptone and either 2% glucose (YEPD) or galactose (YEPG), or in 0.67% yeast nitrogen base without amino acids and 2% glucose (GMM) with appropriate amino acids for auxotrophic mutants. For yeast growth, cells were diluted to OD600 = 0.6 in pH 4.0 YEPD media and incubated at 30°C. For hyphal induction, cells were washed in sterile water and diluted to OD600 = 0.6 in pH 7.0 YEPD media supplemented with 10–20% fetal calf serum (FCS) and incubated at 37°C. G1 cells were obtained by centrifugal elutriation from a log phase YEPD culture using a Beckman elutriating rotor JE 5.0. For MET3 promoter shutdown, methionine and cysteine were added to a final concentration of 2.5 mM and 0.5 mM respectively. Inhibition of the analogue sensitive cdc28-as1 allele was carried out by adding 1NM-PP1 (Merck-Calbiochem) to a final concentration of 30 μM.
C. albicans strains, plasmids and primers are listed in the S1 Table. Strains were generated either by PCR-generated fragment recombination or plasmid integration as previously described [64–66]. Site-directed mutagenesis was carried out using the QuickChange Multi Site-Directed mutagenesis kit (Agilent technologies, Edinburgh, UK) following the manufacturer’s instructions.
Cells were harvested by centrifugation, washed once with sterile water and then re-suspended in two volumes of lysis buffer (50 mM Tris-HCl pH 7.5. 100 mM NaCl, 0.1% Triton X-100 and 0.1% w/v sodium deoxycholate) containing a protease inhibitor cocktail (Roche) and phosphatase inhibitors (50 mM NaF and 100 mM β-glycerophosphate). One cell volume of 0.4 mm glass beads were then added before 3 rounds of 30 second homogenization in a mini-bead beater (Biospec products, Bartlesville, OK,USA), with 2 min on ice in between. Lysates were then cleared by centrifugation at 13,000 rpm for 10 min. For direct Western Blotting, 30–50 μg protein aliquots were resolved by 1D PAGE. For Immuno-precipitation (IP) 2 mg of total protein was incubated with 50 μl of protein-G Dynabeads (Life Technologies, Carlsbad CA USA), after pre-binding with GFP (Roche) or Myc/HA (Bioserv, Sheffield UK) mouse monoclonal antibodies, in a total IP volume of 500 μl for 90 min at 4°C. IP products were washed 2–4 times with 10 bead volumes of cold lysis buffer before being re-suspended and boiled for 5 min in 48 μl of 1x protein loading dye. αPSTAIR loading control is a monoclonal antibody raised against a synthetic peptide (P7962 Sigma). αPser CDK (Cell-Signalling 2324S) recognises phosphorylated serines in the Cdc28 consensus target motif SPxR/K.
Dephosphorylation of proteins was carried out using Lambda Phosphatase (NEB # P0753). 400 units of lambda phosphatase were used for 40 μl of 10 mg/ml lysate in a total reaction volume of 50 μl, including 1mM MnCl2 and 1x PMP buffer. The reaction was carried out at 30°C for 1 h on a shaking platform set at 200 rpm
Proteins were visualised on Western blots using ECL (GE Healthcare, Amersham, UK) and resulting chemifluorescence recorded using a GeneGnome (Syngene, Cambridge, UK)
C-terminal fragments of the Fkh2 wild-type sequence or substituted with alanine at the Cdc28 target site were expressed and purified from E. coli as previously described [28]. Cdc28 or Cbk1 was purified from 2 mg lysate in a strain where both alleles were fused to HA and then washed three times with lysis buffer containing 750 mM NaCl and once with standard lysis buffer. 25 μl of 2x kinase assay buffer (100 mM Tris-HCl pH 7.5, 2 mM EGTA, 0.02% v/v Tween-20 (Sigma-Aldrich) 2 mM DTT, 2 mM β-glycerophosphate and 20 mM MgCl2) was added to the beads along with 24 μl of 1 mg/ml GST fragment and 1 μl ATP from a 5 mM stock. Reactions were incubated at 37°C for 1 h with gentle agitation. The reaction was quenched by boiling in 1x loading dye for 5 min.15 μl of reaction was separated by 10% 1D PAGE before subsequent Western blotting to PVDF membrane. Kinase assay blots were initially probed with either αPSER(CDK) (Cell-Signalling 2324S) or αPSER Q5 (Qiagen 37430), and then stripped and re-probed with rabbit monoclonal αGST (Santa-Cruz Biotechnology)
Fkh2-HA or Fkh2-Myc was immuno-purified from approximately 1 g of total cell lysate using 60 μl of the appropriate Ez-view agarose slurry (Sigma-Aldrich). The protein-bound agarose was washed four times with cold lysis buffer before boiling for 10 min in 30 μl 1x protein loading dye. The denatured proteins were separated on a 4–20% gradient SDS-PAGE gel (Bio-Rad) and then stained using Instant Blue™ Coomassie stain (Expedeon). Bands of interest were excised for subsequent MS analysis as previously described [28].
2D electrophoresis was carried out as previously described [67] with modifications. Fkh2-GFP was purified from 2 mg of total protein by IP, and then removed from the beads by adding 100 μl of hydration buffer with 4 mM DTT and incubating for 1 h at room temperature before subsequent isoelectric focusing (IEF) using the IPGphor3 IEF system (GE Life-sciences) on a pH 3–10 range IPG strip. Western blots were imaged with a GeneGnome (Syngene, Cambridge UK). The images shown in Figs. 1E, 2E and 3B and 10A are 400×70 pixels aligned 100 pixels from the fixed marker in a 1.3 MP image. The intensities of the fluorescence in these images were profiled using the FIJI distribution of ImageJ (http://fiji.sc/Fiji). The profiles were derived from a line that was manually drawn through the signal and the resulting intensity values exported as csv files into Microsoft Excel (Microsoft Corporation, US) to generate the intensity plots shown above the autoradiograms of the 2D gels.
Log phase yeast and 3 h post-induction hyphal cells were harvested and the pellets were snap-frozen in liquid nitrogen. RNA was extracted using the RNeasy minikit (QIAGEN) following the manufacturer’s instructions for yeast mechanical disruption, using a TOMY Microsmash (TOMY Digital Biology, Tokyo JP). Total RNA was treated with DNase I (Roche) before RNA-Cleanup using the RNeasy kit. 10 µg of total RNA was reverse-transcribed with either Cy3 (control) or Cy5 (experimental) dyes and hybridised to full C. albicans arrays from Microarrays Inc. (Huntsville Alabama). Arrays were subsequently scanned with an Axon GenePix 4000B scanner. Duplicates of each array were carried out, with each array containing multiple replicate probes.
Microarray data was analysed using the limma (version 3.21.18) package for the R language (version 3.1.1) [68]. Briefly, after quality control the arrays were background normalised using the “normexp” method [69] and spots with an intensity of less than 50 over background in both channels were removed. Arrays were then normalized using the “print-tip loess” method [70]. Within array replicate spots were averaged. Differentially expressed genes between wild type, fkh2(6A) and fkh2ΔΔ yeast samples were assessed using empirical Bayes moderated t-tests [71]. For hyphal samples we fitted a linear model with indicator variables for genotype and used moderated t-statistics to test the significance of the coefficients for the phosphorylation mutant and deletion genotype, as well as the significance of the difference between these coefficients. False discovery rate was assessed across all tests simultaneously using the method of Benjamini and Hochberg [72]. Where multiple probes targeted the same gene, the least significant was selected. Genes were regarded as more down regulated in fkh2(6A) than fkh2ΔΔ if: a) The gene was down-regulated more than two fold at the 5% FDR threshold in the fkh2(6A) samples compared to wild type b) the fold change was less negative (or was positive) in the fkh2ΔΔ samples and c) the difference in fold changes was significant at the 5% FDR threshold. Similar criteria were used for genes more down regulated in fkh2ΔΔ samples. Data and detail analysis protocol are deposited in GEO with accession GSE64383. GO analysis was conducted using the GO Stats R/Bioconductor package [73], using conditional testing (account for GO graph structure) and calculating FDR using the Benjamini and Hochberg method [72].
Total RNA was prepared as above, and 0.5 μg was reverse-transcribed using Superscript III (Invitrogen) in a reaction volume of 30 μl following the manufacturer’s instructions. The mock reverse transcription reaction contained everything except the Superscript enzyme. Reaction mixes were then diluted two-fold for use in subsequent qPCR reactions. Triplicate qPCR reactions were carried out using 0.5 μM of the primer pairs listed in the supplementary materials in a total 10 μl volume with 2x Sensi-mix (Bioline). Samples were run on the Rotor-gene-6000 system (QIAGEN). Normalisation of expression levels was carried out using the ADE2 gene, before comparative ΔΔCT analysis of expression levels in wild-type and mutant strains.
Fluorescence microscopy was carried out as previously stated [67]
These assays were carried out as previously described [74]
Cell culture. TR146 oral (buccal) epithelial cells [75] (SkinEthic Laboratories, Lyon FR) were maintained in DMEM (Sigma) supplemented with 10% (v/v) heat-inactivated foetal bovine serum (Sigma), 1% (v/v) penicillin-streptomycin solution and cultured at 37°C, 5% CO2. All experiments were performed in serum-free DMEM (Sigma). C. albicans strains were cultured overnight in YEPD medium at 30°C, 200 rpm. Cells were collected by centrifugation and washed twice in sterile PBS prior to use in epithelial cell damage assays.
Confluent TR146 monolayers were infected with C. albicans strains (multiplicity of infection = 0.01) and cultured for 24 h (37°C, 5% CO2). Cell culture supernatants were collected and epithelial cell damage was determined by quantification of lactate dehydrogenase (LDH) activity using the Cytox 96 Non-Radioactive Cytotoxicity Assay kit (Promega) according to the manufacturer’s instructions. Recombinant porcine LDH (Sigma) was used to generate a standard curve.
TR146 monolayers were infected with C. albicans strains as described above. Cell culture supernatants were collected and cytokines quantified using the Fluorokine MAP cytokine multiplex kit (R&D Systems), coupled with a Bio-Plex™ 200 machine according to the manufacturer’s instructions. The trimmed median value was used to derive standard curves and calculate sample concentrations
|
10.1371/journal.ppat.1006093 | A Large Collection of Novel Nematode-Infecting Microsporidia and Their Diverse Interactions with Caenorhabditis elegans and Other Related Nematodes | Microsporidia are fungi-related intracellular pathogens that may infect virtually all animals, but are poorly understood. The nematode Caenorhabditis elegans has recently become a model host for studying microsporidia through the identification of its natural microsporidian pathogen Nematocida parisii. However, it was unclear how widespread and diverse microsporidia infections are in C. elegans or other related nematodes in the wild. Here we describe the isolation and culture of 47 nematodes with microsporidian infections. N. parisii is found to be the most common microsporidia infecting C. elegans in the wild. In addition, we further describe and name six new species in the Nematocida genus. Our sampling and phylogenetic analysis further identify two subclades that are genetically distinct from Nematocida, and we name them Enteropsectra and Pancytospora. Interestingly, unlike Nematocida, these two genera belong to the main clade of microsporidia that includes human pathogens. All of these microsporidia are horizontally transmitted and most specifically infect intestinal cells, except Pancytospora epiphaga that replicates mostly in the epidermis of its Caenorhabditis host. At the subcellular level in the infected host cell, spores of the novel genus Enteropsectra show a characteristic apical distribution and exit via budding off of the plasma membrane, instead of exiting via exocytosis as spores of Nematocida. Host specificity is broad for some microsporidia, narrow for others: indeed, some microsporidia can infect Oscheius tipulae but not its sister species Oscheius sp. 3, and conversely some microsporidia found infecting Oscheius sp. 3 do not infect O. tipulae. We also show that N. ausubeli fails to strongly induce in C. elegans the transcription of genes that are induced by other Nematocida species, suggesting it has evolved mechanisms to prevent induction of this host response. Altogether, these newly isolated species illustrate the diversity and ubiquity of microsporidian infections in nematodes, and provide a rich resource to investigate host-parasite coevolution in tractable nematode hosts.
| Microsporidia are microbial parasites that live inside their host cells and can cause disease in humans and many other animals. The small nematode worm Caenorhabditis elegans has recently become a convenient model host for studying microsporidian infections. In this work, we sample Caenorhabditis and other small nematodes and 47 associated microsporidian strains from the wild. We characterize the parasites for their position in the evolutionary tree of microsporidia and for their lifecycle and morphology. We find several new species and genera, especially some that are distantly related to the previously known Nematocida parisii and instead closely related to human pathogens. We find that some of these species have a narrow host range. We studied two species in detail using electron microscopy and uncover a new likely mode of exit from the host cell, by budding off the host cell plasma membrane rather than by fusion of a vesicle to the plasma membrane as in N. parisii. We also find a new species that infects the epidermis and muscles of Caenorhabditis rather than the host intestinal cells and is closely related to human pathogens. Finally, we find that one Nematocida species fails to elicit the same host response that other Nematocida species do. These new microsporidia open up many windows into microsporidia biology and opportunities to investigate host-parasite coevolution in the C. elegans system.
| Microsporidia are fungi-related obligate intracellular pathogens, with over 1400 described species [1,2]. Interest in these organisms started 150 years ago when researchers, especially Louis Pasteur, studied silkworm disease that was caused by a microsporidian species later named Nosema bombycis [3]. In the past decades, microsporidia have attracted more attention when they were revealed to be a cause of diarrhea in immunocompromised patients and were further demonstrated to have a high prevalence in some areas in immunocompetent patients and healthy individuals [4–6].
Microsporidia are transmitted between hosts through a spore stage. Inside the microsporidian spore is found a characteristic structure called the polar tube, which at the time of infection can pierce through host cell membranes and introduce the sporoplasm (the spore cytoplasm and nucleus) into host cells [1,7]. These obligate intracellular pathogens are known to infect a wide range of hosts among protists and animals, especially insects, fish and mammals [1]. Even though nematodes constitute a huge phylum with over 25,000 described species, very few studies on microsporidian infections in nematodes have been reported so far [1].
The free-living nematode Caenorhabditis elegans has been used as a major biological model species over the last 50 years [8]. However, until the past decade, little was known about its biology and ecology in its natural environment and no natural pathogens were isolated until C. elegans could be readily isolated from natural environments. C. elegans is now known to be found in compost heaps, rotting fruits (apples, figs, etc.) and herbaceous stems, as well as with diverse carrier invertebrates (snails, isopods, etc.) [9–11]. C. elegans coexists with a variety of prokaryotic and eukaryotic microbes, including both its food and pathogens, which likely have a large impact on its physiology and evolution [12–15].
With an improved understanding of the natural history of Caenorhabditis [16,17], dramatically increased number of various wild rhabditid nematode strains and species have been isolated and identified. C. elegans' close relatives such as Caenorhabditis briggsae or Caenorhabditis remanei are isolated from similar environments [18]. Oscheius tipulae, a very common bacteriovorous nematode species, also in family Rhabditidae, can be readily isolated from soil and rotting vegetal matter [10,19,20], as well as its closest known relative Oscheius sp. 3, with which it cannot interbreed [21]. Interest in these rhabditid nematodes concerns not only the evolution of genomic and phenotypic characters, but also their inter- and intraspecific interactions and co-evolution with other organisms, especially with various microbes in their natural habitats. While nematodes feed on bacteria and small eukaryotes, some microbes take nematodes as their food source [13,14,16]. Among them, microsporidia are obligate intracellular parasites and thus are in particularly tight association with their hosts.
The microsporidian Nematocida parisii was the first found natural intracellular pathogen of C. elegans, which we isolated from a wild C. elegans sampled near Paris, France [22]. Nematocida sp. 1 (described here as Nematocida ausubeli) was further isolated from a wild Caenorhabditis briggsae strain in India [22]. A microsporidian species isolated in C. elegans was found to infect the epidermis and muscles and was named Nematocida displodere [23]. Two microsporidia infecting marine nematodes have also been described, namely Sporanauta perivermis [24] and Nematocenator marisprofundi [25,26]. However, the extent and diversity of microsporidia infections in nematodes remained sparsely described.
Here we describe a collection of 47 terrestrial nematode strains that we isolated from the wild with a microsporidian infection. The microsporidia can be grown in the laboratory in their host using C. elegans culture conditions and stored frozen with their nematode host. They are all transmitted horizontally. In this set, we found that N. parisii and N. ausubeli (formerly called N. sp. 1) are in association with further host species and display a wider geographical distribution than so far reported [22]. N. parisii is the most common C. elegans-infecting species we found in the wild. We further discovered new nematode-infecting microsporidian species. From our phylogenetic analysis using small subunit (SSU) ribosomal DNA and β-tubulin sequences, five new microsporidia species were placed in the Nematocida genus, while the others defined two new genera in the microsporidian clade often designated as Clade IV, which includes human pathogenic microsporidia such as Enterocytozoon bieneusi and Vittaforma corneae. The similarities and differences in the morphological features of these microsporidia matched their groupings by sequence similarity. We therefore describe two new microsporidian genera, Enteropsectra and Pancytospora. and nine new species in these two genera and Nematocida. We further examined Nematocida ausubeli and Enteropsectra longa by electron microscopy, which allowed us to observe different mechanisms for their exit from host intestinal cells, through a vesicular pathway for N. ausubeli (as described for N. parisii; [27]), but surprisingly through membrane budding for E. longa. Concerning specificity of infection, we find cases of tight specificity between host and pathogen. We also find that N. ausubeli fails to strongly induce the transcription of genes that are induced in C. elegans by N. parisii infection. Overall, our study points to strong and diverse interactions between wild rhabditid nematodes and microsporidia, and provides a platform for further study of these infections.
Our worldwide sampling of bacteriovorous nematodes was primarily aimed at isolating Caenorhabditis species and, to a lesser degree, Oscheius species. From this sampling, we identified a subset of strains with a pale body color (Fig 1A), some of which, upon morphological examination using Nomarski microscopy [22], appeared infected with microsporidia. In total, we collected 47 nematode strains (S1 Table) displaying putative microsporidian infections, comprising 10 nematode species from different parts of the world (Tables 1 and 2; Fig 1B). The microsporidia strain JUm2807 was isolated during these sampling efforts and described elsewhere as Nematocida displodere, and is not considered here [23].
The unidentified microsporidian strains were characterized by sequencing of PCR fragments of the SSU rDNA and β-tubulin genes. We were able to amplify 45 SSU rDNA sequences (most 1390 bp long) and 32 β-tubulin sequences (most 1210 bp long) (S1 Table). We first blasted the sequences in GenBank for initial grouping, then built phylogenetic trees and calculated interspecific genetic distances, based on our sequences and the sequences of related species from GenBank. We present below the grouping and phylogenetic distribution of new microsporidia strains, starting with those closest to N. parisii.
Molecular sequences of microsporidia in ten wild C. elegans strains and four C. briggsae strains showed ≥ 99% SSU rDNA and ≥ 97% β-tubulin sequence identities to N. parisii sequences in GenBank. In the global phylogenetic analysis of microsporidia, these 14 sequences form a group with previously reported sequences of N. parisii strains ERTm1, ERTm3 and ERTm5 [26] (Fig 2). The N. parisii isolates were all found in Europe (note however that the sampling is highly biased towards Europe, especially France), with the exception of the previously reported ERTm5 (JUm2055), isolated from a C. briggsae strain sampled in Hawaii (Fig 1B; Table 1A) [29]. Note that a recent article assigns this strain to a new species based on genome divergence [28].
Eight other microsporidian strains showed ≥ 99% SSU rDNA and ≥ 95% β-tubulin sequence identities to the corresponding genes of the unnamed Nematocida sp. 1 in GenBank (Table 1B), previously reported in C. briggsae [22,30]. This N. sp. 1 group is most closely related to N. parisii in the microsporidian phylogeny and the sequences of both SSU and β-tubulin genes gave the same grouping (Figs 2 and 3; S2 Fig; Table 3). Because of these new samples of N. sp. 1 and their phylogenetic difference and genetic distance to the N. parisii group, here we describe N. sp. 1 as Nematocida ausubeli n. sp. (see Taxonomy section after the Discussion). Whereas N. ausubeli was so far only reported from C. briggsae (India, Cape Verde [30]), we also found it in C. elegans and C. remanei, in France, Portugal and Germany (Table 1B; Fig 1B), thus broadening its geographic and host range to several species of the Elegans group of Caenorhabditis from Europe.
The remaining 20 microsporidia strains that we identified are distributed among several other species, including some species in another clade (see below). Thus the dominant microsporidia species in our collection of Caenorhabditis and other nematodes are Nematocida parisii and Nematocida ausubeli n. sp., with 17 and 10 strains, respectively. They were found in several species of the Elegans group of Caenorhabditis nematodes.
Of the remaining 19 microsporidian strains, nine had a Nematocida species as their top blast hit in GenBank, with similarity between 81% ~ 86% of SSU rDNA and 74% ~ 84% of β-tubulin genes. In terms of host and geographical ranges, these microsporidia were found in two C. briggsae strains (Thailand and Guadeloupe), one C. tropicalis strain (Guadeloupe), one C. sp. 42 strain (NIC1041 from French Guiana), three Oscheius tipulae strains (France, Czech Republic, and Armenia), one Rhabditella typhae strain (Portugal) and one Procephalobus sp. strain (JU2895 from Spain). In the phylogenetic analysis of SSU rDNA, the corresponding sequences formed a single clade with N. parisii and N. ausubeli, with Ovavesicula popilliae as sister group within Clade II of the microsporidian phylum (see Fig 2) [34]. In addition, the JUm2807 strain that has been recently described as Nematocida displodere [23] is distinct from all of them.
From phylogenetic analysis and genetic distance of SSU rDNA genes, these Nematocida strains form four groups. These putative new Nematocida species have a mean genetic distance among them of at least 0.06 (Table 3), while their intra-specific genetic distances are all 0.00 (when several strains were isolated). This inter-group distance is also greater than the distance between N. parisii and N. ausubeli. Hence we describe them below as four new species: Nematocida minor, Nematocida major, Nematocida homosporus and Nematocida ciargi n. spp. (see Taxonomy section).
In terms of the phylogenetic relationships within the Nematocida genus in the SSU rDNA tree, the first outgroup clade to N. parisii + N. ausubeli was formed by JUm2751, JUm2747 and JUm2751, corresponding to N. major (Fig 2). The second branch out was formed by JUm1510 and JUm2772, described here as N. minor. N. ciargi JUm2895 was placed in a basal position to the clade formed by N. parisii, N. ausubeli, N. major and N. minor (Fig 2). At the base of the Nematocida genus, the most externally branching sequences appeared to be those of N. displodere JUm2807, and of N. homosporus JUm1504 and NICm516. All topologies were highly supported, except for the node defining the latter clade of N. homosporus and N. displodere (Fig 2). In the phylogenetic tree based on both genes (SSU rDNA and β-tubulin), N. ciargi was placed at the base of Nematocida genus, while N. displodere and N. homosporus still formed one clade (Fig 3). The phylogenetic tree only based on β-tubulin sequences supported the grouping of strains and overall their relative positions (S2 Fig), except that the relative placement of N. displodere and N. ciargi was exchanged. The β-tubulin phylogeny has one more branch formed by NICm1041, numbered provisionally N. sp. 7, for which we failed to amplify the SSU rDNA fragment. Whole-genome analysis could be performed in the future to refine these placements.
The Nematocida consensus phylogeny is shown in Fig 3 next to the phylogenetic relationships of the nematode hosts in which they were naturally found (see below for further specificity tests). Although the numbers of samples and species are too low for rigorous testing, the data are at least consistent with the intestinal microsporidia species branching through continuous co-evolution with their nematode host. For example, all intestinal Nematocida species found in Caenorhabditis species form one clade, with a first outgroup including Oscheius and Rhabditella pathogens and a distant outgroup infecting the distant outgroup Procephalobus (Fig 3). The exception is N. displodere that was found a single time, in C. elegans, and corresponds to a change in tissue tropism.
As with previously isolated Nematocida, the newly identified microsporidia appeared to be transmitted horizontally, because a bleaching treatment [35] of infected gravid adults eliminated the infection in the culture and reinfection could be obtained by exposure to spores in the environment. All Nematocida microsporidia stages described here were found exclusively in the intestinal cells and were not detected in the germ line.
As previously described for N. parisii [22], two main stages could be distinguished by Nomarski optics. First, the meront stage appeared as areas of infected intestinal cells devoid of storage granules. These areas were first small circular regions, then extending to longer grooves. Second, rod-shaped sporoblasts and spores appeared in the intestinal cell cytoplasm. In host cells that were heavily infected with N. parisii and some other species, groups of spores inside vesicles could be seen [22], possibly derived from spore re-endocytosis [36]. In this study, as described before [22], all N. parisii and N. ausubeli infections displayed two distinct classes of spore size (Table 1; Fig 4A and 4B; S3A Fig).
N. major and N. minor also displayed two spore size classes. N. major formed slightly longer but thinner spores than N. parisii. N. minor showed however much smaller spores, for each class taken separately (Tables 2 and 4; Fig 4C and 4D). In contrast, N. homosporus and N. ciargi only have a single class of spore size, with N. homosporus spores having an intermediate size (2.00 ± 0.22 μm long, 0.72 ± 0.12 μm wide) and N. ciargi spores having a smaller size (1.39 ± 0.20 μm long, 0.59 ± 0.13 μm wide). Spore vesicles were observed more frequently with N. homosporus or N. ciargi infections than with other Nematocida infections (Fig 4E and 4F).
N. ausubeli being the most commonly found parasite of C. elegans besides N. parisii, we further chose to study its lifecycle by electron microscopy. The ultrastructure by electron microscopy and the deduced lifecycle of N. ausubeli overall resembled those of N. parisii, with possible differences outlined below. High-pressure freezing/freezesubstitution allowed better to visualize lipid membranes compared to room temperature preparation methods. We observed meronts, which are separated from the host cell by a single membrane bilayer, likely pathogen-derived (Fig 5A and 5B; S1A and S1B Fig). Their cytoplasm appeared packed with ribosomes. Some meronts displayed an elongated shape and contained several nuclei (Fig 5B and 5L). The membrane enclosing the meronts appeared to darken progressively and intracellular membrane compartments developed, likely corresponding to the progressive transition to a sporont stage (Fig 5C). We further observed sporogony, whereby individual sporoblasts with a single nucleus are formed, each surrounded by a membrane (Fig 5D and 5E; S1A and S1D Fig). We did not observe any nuclear division at this stage (unlike in Enteropsectra longa, where they were easily found; see below). We observed progressive stages of sporogenesis, including formation of the anchoring disk, polaroplast membranes, polar tube, posterior vacuole and spore coat (Fig 5E–5G; S1C, S1D, S1F and S1G Fig).
In the final stages of sporogenesis and in mature spores that corresponded to the small size class observed in light microscopy, two polar tube coil cross-sections could usually be observed (Fig 5H and 5J; S1L Fig). A single large spore could be found, which displayed three polar tube coil sections on either side of the spore (six sections in total; Fig 5K). Thus, the tube coiled several times in large spores, instead of once in the small spores. In N. parisii, five polar tube sections were reported on one side of the large spores [22]; it is thus possible that large spores of N. ausubeli harbor fewer polar tube coils than those of N. parisii (because a single large spore was found in each species, it is however difficult to conclude). The anchoring disk defines the anterior pole of the spore. Below the anchoring disk, the polar tube is lined on either side by polaroplast membranes (visible in Fig 5F and 5G). A polar tube cross-section with several layers could be seen in Fig 5J and the posterior turn of the polar tube in S1I Fig. The mature spore was seen to contain a posterior vacuole on the side opposite to the anchoring disk (Fig 5K; S1H–S1J Fig). This vacuole seemed to develop from a dense membrane compartment of the sporoblast (S1C and S1D Fig). The spores displayed an external coat with several layers (Figs 5H, 5J, 5K and 6A; S1I–S1L Fig).
The spores in the host cytoplasm appeared either isolated, or clustered within a large vesicle. Some isolated spores were surrounded by an additional membrane outside the spore coat and the inner face of this membrane appeared coated (Fig 5H). Unlike in N. parisii [27], we could not see the additional membrane around all spores. Fig 6A shows a spore apparently exiting the host cell through exocytosis (although we cannot rule out that such images correspond to endocytotic events). Spores in the lumen were not surrounded by any additional membrane (Figs 5K, 5L and 6A).
When spores were clustered in a vesicle, two membranes could be observed around them (Fig 5J, and other instances).
Whereas the Nematocida genus is in Clade II of the microsporidia [22,34], the remaining nine microsporidia strains in our collection were placed in Clade IV, which, unlike Clade II, contains several human-infecting microsporidia (Fig 2). This clade assignment was based on SSU rDNA sequences, which had closest (88–89%) identities to the insect parasite Orthosomella operophterae (host: moth Operophtera brumata) (Table 2). Only four β-tubulin sequences could be obtained, and these were closest (75% ~ 76% identity) to Vittaforma corneae, a human-infecting microsporidia species and a close relative of Orthosomella operophterae (whose β-tubulin sequence is not available), consistent with rDNA analysis. We thus isolated nematode-infecting microsporidia that are in a distinct evolutionary branch compared to Nematocida and are closer relatives of the human-infecting microsporidia.
Eight out of the nine strains in this group have Oscheius species as their nematode host and infect their gut: seven of them from different locations in France were found in O. tipulae, while JUm408 was found in Oscheius sp. 3 [21] from Iceland. The ninth strain, JUm1396, was isolated from a C. brenneri strain and is the only one in this set to infect non-intestinal tissues.
In the phylogenetic analysis, these nine strains separated into two groups, corresponding to the two new genera described below, Enteropsectra and Pancytospora (see section on Taxonomy) (Fig 2; S4 Fig). The first group included four strains, JUm408, JUm1456, JUm2551 and JUm1483, which were phylogenetically placed as a sister group to Liebermannia species (with hosts such as grasshoppers) (Fig 2). In the β-tubulin phylogeny, Enteropsectra strains also showed a sister relationship with the group of V. corneae and Enterocytozoon bieneusi, a human intestinal parasite (S2 Fig). However, with β-tubulin, JUm408 and JUm1483 formed a branch, JUm1456 and JUm2551 another branch, which was different from their SSU rDNA phylogenetic position. Based on molecular sequences, spore morphology and host specificity (below), we describe two species in the Enteropsectra genus, E. longa (type strain JUm408) and E. breve (type strain JUm2551), and do not assign the two other strains to a species. E. longa and E. breve strains have a small mean SSU genetic distance of 0.005 (Table 3) but differ in spore size and host specificity (see below). While E. longa and E. breve form a sister group to Liebermannia species on the SSU rDNA phylogeny, they have a smaller mean genetic distance to O. operophterae (0.08) than to Liebermannia (0.11).
The second new clade of nematode-infecting microsporidia includes the five remaining strains and showed strong support as sister lineage to the clade formed by Enteropsectra and Liebermannia species, with O. operophterae as outgroup (Fig 2). Based on molecular sequences, host and tissue specificity, we describe two new species: Pancytospora philotis (JUm1505 as type strain, JUm1505, JUm1670, JUm2552), found in the Oscheius gut, and P. epiphaga (JUm1396) from a C. brenneri strain from Colombia that caused an epidermis and muscle infection (Fig 7; S5 Fig).
As with Nematocida, all of the infections by Clade IV microsporidian strains mentioned above appeared to be transmitted horizontally, as bleaching of the nematode culture eliminated the infection. The Enteropsectra strains and P. philotis were only observed to infect the intestine of Oscheius nematodes. By contrast, P. epiphaga (JUm1396) was found to infect epidermis and muscles of C. brenneri (Fig 7D; S5D, S5E and S5F Fig), thus sharing its tissue tropism with N. displodere, although on a different evolutionary branch. P. epiphaga could also infect C. elegans (N2 reference background) (S7F Fig) and C. briggsae (AF16).
A striking feature of Enteropsectra strains is their cellular localization within the nematode intestinal cells: Enteropsectra were all observed to form their spores on the apical side of the epithelial cell at first, while meront stages could be seen in a more basal position (Figs 7A, 7B, 7E; 8L). This polarization within the host intestinal cell was not observed in infections of P. philotis nor of any Nematocida species (Table 2; Figs 4 and 7).
The Enteropsectra and Pancytospora species displayed quite different sizes and shapes of spores from those of Nematocida species and we did not see any spore-containing vesicles in these microsporidian infections. They all show a single class of spore size. Though apart in the phylogenetic analysis, E. longa (JUm408) and P. philotis share similar dimensions of spores, which are particularly long and thin: E. longa (JUm408) spores measure 3.76 ± 0.38 μm by 0.49 ± 0.06 μm, while P. philotis spores measure 3.46 ± 0.48 μm long by 0.42 ± 0.06 μm. These spores are even longer than the largest spores and thinner than the smallest spores in Nematocida. In stark contrast, E. breve (JUm2551) form small rod-shaped and crescent-shaped spores (Fig 7B; Table 4).
Because of the striking difference in spore distribution, we further analyzed by electron microscopy the type species of the Enteropsectra genus, Enteropsectra longa (JUm408) in Oscheius sp. 3 JU408. The meront stage appeared overall similar to that of Nematocida species: the early stages displayed a cytoplasm packed with ribosomes and very few membranes (Fig 8A); elongated multinucleated meronts could also be observed (Fig 8B). The parasite membrane then progressively darkened, indicating the transition to the sporont stage (Fig 8C–8F). Figures of intranuclear mitosis could be seen at this stage, with intranuclear microtubules and spindle plaques at the nuclear membrane (Fig 8E; S6A Fig). Signs of sporogenesis then developed, with a nascent polar tube (Fig 8F–8H; S6B Fig). The spore membrane and nascent wall appeared wrinkled (Fig 8G) before becoming smooth in mature spores (Fig 8H–8J). The spore wall with its endospore and exospore layers could be clearly observed (Fig 8J). An anchoring disk formed (Fig 8I and 8K), but the polaroplast membranes were less developed than in Nematocida species. In most spores, the polar tube presented a single section (Fig 8G, 8H and 8J; S6D Fig). The polar tube however could be seen to turn on the posterior side of the spore (S6C Fig) and occasionally two polar tube sections could be counted in the same spore section (S6E and S6F Fig). The polar tube thus likely folds back anteriorly on the posterior side of the spore on a short part of its length.
By electron microscopy, we observed a potential key difference in the exit mode of the spores between Enteropsectra longa (JUm408) on one hand, and N. parisii and N. ausubeli on the other hand. First, the sporoblasts and mature spores of E. longa were never seen to be surrounded by an additional membrane outside the spore wall, precluding exocytosis as an exit route. Second, the spores were seen to protrude on the apical side of the host cell, pushing out the host cell membrane like a finger in a glove (Fig 6B; S6G–S6I Fig). We further focused on spore sections in the intestinal lumen and saw both spores with a surrounding membrane (S6I Fig) and spores without any membrane (Fig 8K).
On the host side, rough endoplasmic reticulum was often seen to wrap around sporoblasts, yet never encircling them fully (Fig 8G). The host cell nuclei presented a characteristic nucleolar structure, which became organized in long tubules (often appearing circular in cross-sections;. S6J and S6K Fig). On one occasion, microsporidian spores were observed within the host intestinal cell nucleus, whose nucleolus had apparently further degenerated (S6L Fig).
The pattern of natural association revealed an apparent specificity of a given microsporidian species for a nematode genus, mostly Caenorhabditis versus Oscheius in our collection. Strikingly, N. parisii, N. ausubeli and N. major infections were found in Caenorhabditis species, while N. minor, N. homosporus and Clade IV microsporidia species infections were all found in Oscheius species and not in Caenorhabditis (or, for N. homosporus, in Rhabditella, a closer relative of Oscheius compared to Caenorhabditis; Fig 3). The notable exception in Clade IV was the epidermal P. epiphaga JUm1396, found in C. brenneri. These results suggested a pattern of host-pathogen specificity between nematode and nematode-infecting microsporidia.
We further complemented these natural associations with infections performed in the laboratory. To test for the capacity of a given microsporidia strain to infect a given host, uninfected nematode cultures (cleaned by bleaching) were exposed to microsporidian spores. We used clean spore preparations from seven microsporidian species (see Materials and Methods), namely N. parisii, N. ausubeli, N. major, N. homosporus, E. longa, E. breve and P. philotis. On the host side, we focused on four nematode species of two genera: C. elegans, C. briggsae, O. tipulae and O. sp. 3, all of which reproduce through self-fertilizing hermaphrodites and facultative males [19,20]. We favored wild strains that had been found naturally infected with microsporidia and were thus not generally resistant to microsporidian infections (Table 5).
N. parisii (JUm2816) infected more than 50% of C. elegans (N2, JU2009) and C. briggsae (JU2747, JU2793) individual animals at 72 hpi. However, no microsporidian infection symptom was observed in O. tipulae (JU1483, JU170) nor O. sp. 3 (JU408, JU75) at 72 and 120 hpi (Table 5). O. tipulae strains JU1504, JU1510 and JU2552 were also exposed to N. parisii spores, and none of them became infected either. These infection results indicated that N. parisii was unable to infect O. tipulae nor O. sp. 3 (Table 5).
Specificity of N. ausubeli (JUm2009) slightly differed from that of N. parisii. By 72 hpi, half of all Caenorhabditis animals and about 30% of O. sp. 3 showed signs of infection. None of O. tipulae worms were infected even at 120 hpi (Table 5). However, when we made a new N. ausubeli (JUm2009) spore preparation and used it directly for infection tests, O. tipulae strains JU1510 and JU2552 could be infected, but the preparation lost its ability to infect O. tipulae over storage at -80°C (see Materials and Methods). We conclude that O. tipulae was far less susceptible to infection than C. elegans, C. briggsae and O. sp. 3 to N. ausubeli infection.
The host spectrum of N. major (JUm2747) was quite similar to that of N. parisii: it infected C. elegans (N2, JU2009) and C. briggsae (JU2507, JU2793), but not O. tipulae (JU1483, JU170) nor O. sp. 3 (JU408, JU75) (Table 5). N. homosporus, however, could infect both Caenorhabditis and both Oscheius species and thus appeared as the most generalist (Table 5). Yet O. tipulae seemed relatively less sensitive than C. elegans, C. briggsae and O. sp. 3 to N. homosporus infection.
Enteropsectra spp. and Pancytospora philotis showed different and even opposite specificities compared to the four tested Nematocida species. Indeed, none could successfully infect any tested Caenorhabditis strains at 120 hpi. Within the two Oscheius species, specific interactions were further observed. Enteropsectra longa (JUm408) only infected O. sp. 3 strains (JU408, JU75), but not O. tipulae (JU1483, JU2551), while E. breve (JUm2551) infected all four O. tipulae and O. sp. 3 strains (Table 5). Pancytospora philotis (JUm1505) only infected O. tipulae (JU1483, JU1505), but not O. sp. 3 strains (JU408, JU75). Since O. sp. 3 is the closest known species to Oscheius tipulae, E. longa and P. philotis are examples of narrow specialization in the host-parasite interaction. We also found that C. elegans N2 could be infected with P. epiphaga (JUm1396), showing epidermal and muscle infection (Table 5; S7F Fig).
The spore morphology of a given microsporidian species was maintained in different nematode species, indicating that host genotype does not affect this pathogen phenotype. For instance, O. tipulae (JU1510) infected with N. ausubeli (JUm2526) displayed two sizes of spores in its intestinal cells as upon Caenorhabditis infection by N. ausubeli (S7A Fig). Oscheius sp. 3 (JU408) infected with Enteropsectra breve (JUm2551) formed small rod-shaped or crescent-shaped spores along the apical side of the worms’ intestinal cells, as upon O. tipulae infections (S7D Fig). C. elegans N2 infected with P. epiphaga (JUm1396) formed long and thin spores in the epidermis and muscles, as upon C. brenneri infection (S7F Fig).
Given the capacity of all Nematocida species to infect C. elegans, we next sought to compare the C. elegans response to infection among our newly isolated microsporidia species. N. parisii infection in C. elegans has been shown to induce a broad transcriptional response [38]. Among genes that were highly upregulated at all infection timepoints were C17H1.6 and F26F2.1, two genes of unknown function. Two transgenic C. elegans strains, ERT54 and ERT72, were generated as transcriptional reporters for these two genes and have been previously shown to be strongly induced in early N. parisii and N. displodere infection [38]. We tested these reporter strains with our new microsporidia species by placing them onto plates with a culture of infected worms and microsporidian spores, then monitoring GFP expression at different timepoints in the reporter strains, as well as monitoring microsporidian meront and spore formation. As expected, N. parisii, N. ausubeli, N. major and N. homosporus could all infect these reporter strains, forming meronts and spores, and induce reporter GFP expression. By contrast, E. longa and Enteropsectra JUm1483 failed to show evidence of proliferative infection and did not robustly induce reporter expression (Fig 9A and 9B; S2 Table). Most interestingly, while N. parisii, N. major or N. homosporus consistently induced the GFP reporters, different strains of N. ausubeli (JUm2009, ERTm2, ERTm6; Fig 9; S2 Table) did not, although this species did robustly infect and proliferate within the C. elegans intestine.
To verify that this differential induction of the GFP reporters matched the transcripts of the endogenous genes, we conducted qRT-PCR after controlled N. parisii (ERTm1) and N. ausubeli (ERTm2) infections of N2 using purified spore preparations that were normalized for an equivalent level of invasion (see Materials and Methods). Indeed, we saw that both C17H1.6 and F26F2.1 transcripts, along with another gene highly induced by N. parisii infection W04B5.5 [39], were induced approximately 6-7-fold lower upon N. ausubeli infection compared to N. parisii infection (Fig 9C), while the levels of pathogen rRNA (indicative of pathogen load) remained similar. Thus, N. ausubeli infection caused a much reduced host response compared to other Nematocida species (Fig 9; S2 Table), despite causing an equivalent, or even more robust infection [40]. Considering the phylogenetic relationships of the Nematocida species (Figs 2 and 3; S2 Fig), this evolutionary change can be polarized: N. ausubeli seems to have lost the capacity to activate these transcriptional reporters as strongly as its closest relatives, or has acquired the capacity to inhibit their activation. Thus, although morphologically quite similar and both are able to infect C. elegans, N. parisii and N. ausubeli elicit distinct host responses.
Microsporidia are ubiquitous obligate intracellular pathogens that have agricultural and medical significance, but have been difficult to study in the laboratory. Our study provides a collection of microsporidia that can infect bacteriovorous nematodes and can easily be studied in the laboratory in their natural hosts and in related species. These rhabditid nematode-infecting microsporidia seem to have more than one origin within the Microsporidia phylum: at least one origin within Clade II and one or two within Clade IV. We thus here enlarge considerably the spectrum of microsporidia that can be cultured in nematodes, including some that are genetically close to human pathogens in Clade IV.
Environmental SSU rDNA microsporidian sequences have been reported from soil, sand and compost samples from North America [41]. (The corresponding species have not been named.) Some of them branch in the SSU phylogeny in the vicinity of the nematode-infecting microsporidia that we isolated (S4 Fig). Specifically, some branch close to Nematocida homosporus and some may be outgroups to Nematocida or further species of the genus. In Clade IV, one is closely related to the Pancytospora epiphaga JUm1396 sequence.
The clades of nematode-infecting microsporidia that we describe have close relatives that infect arthropods, especially insects. This relationship may be due to deep co-evolution (arthropods and nematodes being close relatives on the animal phylogeny), or to the fact that nematodes share their habitats and interact with insects by using them as hosts or carriers [16], which may have facilitated a host shift or a complex lifecycle with several hosts. The microsporidia described here can be cultured continuously in their nematode hosts, but we cannot rule out the possibility that some of them may use non-nematode hosts as well, including insects. Of note, all of them use a horizontal mode of transmission, despite the fact that many instances of vertical transmission of microsporidia in arthropods, molluscs and fish are known [42,43]. In addition, Nematocida species are diploid with evidence of recombination and thus possibly a sexual cycle [30,39], which might occur in another host.
Our results suggest that infections by N. parisii and N. ausubeli are quite common in wild Caenorhabditis strains, especially in C. elegans and C. briggsae. In our collection, N. parisii, N. ausubeli and N. major infections were found in 30 strains of four Caenorhabditis species. Though we have a sampling bias towards France, N. ausubeli was found in Asia, Europe and Africa, while N. parisii was found mostly in France and once (ERTm5) from Hawaii. N. major was only found from three Caenorhabditis strains of C. briggsae and C. tropicalis, all of which were sampled in tropical areas, despite the fact that that we have sampled many hundreds of C. elegans isolates and that N. major can easily infect C. elegans in our specificity infection tests (Table 5). A possibility is that N. major may be preferentially distributed in the tropics rather than temperate zones, where C. elegans are mostly found (Table 2, Fig 1B) [16].
In addition to C. elegans and C. briggsae strains, we also have a relatively large collection of microsporidia-infected Oscheius strains (10 O. tipulae strains and one O. sp. 3 strain). However, none of these strains was found with Nematocida or N. major infections. In line with their natural associations, N. parisii and Nematocida major were not able to infect any Oscheius strains in the laboratory. These specializations may be due to long-term coevolution and adaptation processes [44].
In addition, one new microsporidian species infecting Caenorhabditis was found in clade IV, Pancytospora epiphaga. As this Clade IV microsporidian species can infect C. elegans, it would be interesting to develop its study as a model system for Clade IV species infection.
Microsporidian species that naturally infect Oscheius species are diverse (Fig 10, green entries). N. minor, found from two O. tipulae strains, forms two distinct sizes of spores, similar to N. parisii, N. ausubeli and N. major. N. homosporus was found from one O. tipulae strain and one R. typhae and is the only species tested here that is able to infect species of three genera Caenorhabditis, Oscheius and Rhabditella, suggesting that N. homosporus may be a relatively less specific pathogen for rhabditid nematodes.
The Clade IV Oscheius-infecting microsporidia are separated into two groups: Enteropsectra species, and Pancytospora philotis. None of those could infect Caenorhabditis and their host specificity is even narrower, distinguishing between Oscheius tipulae and its sister species Oscheius sp. 3. The SSU rDNA genetic distances between E. longa and E. breve are quite small and two other closely related Enteropsectra strains are also available (Tables 2 and 3). Overall, Enteropsectra and the Tipulae group of Oscheius species [19,21] provide an interesting case to study the evolution of a narrow host specificity.
Although microsporidia are known to be able to adopt either a horizontal or a vertical transmission [42,45], we here only observed infection in somatic tissues and transmission was horizontal. Most of the infections occurred in host intestinal cells, while two independent instances showed infections elsewhere. As reported previously, Nematocida displodere can infect many tissues and cells in C. elegans, including the epidermis, muscle, coelomocytes and neurons, although it appears to invade all cells by firing its polar tube from the intestinal lumen [23]. The second independent case is Pancytospora epiphaga can be seen in the epidermis, coelomocytes and muscles. Whether it also enters the nematode's cells through the gut remains to be studied.
The most striking variation we observed concerns the cellular exit strategies of the spores (Fig 6). Nematocida parisii spores acquire an additional membrane around the spore wall and thus exit through a vesicular pathway, using the host exocytosis machinery [27]; in addition, clusters of spores with two additional membranes were observed. If the process is similar in N. ausubeli to that in N. parisii, the spore clusters may correspond to re-endocytosis of spores from the lumen [36] or perhaps to autophagy of internal spores using the apical plasma membrane. Of note, the host rough endoplasmic reticulum could often be seen to form concentric patterns in the intestinal cell cytoplasm (S1B and S1E Fig), sometimes wrapping around the sporoblasts (Fig 5E). Whether the reticulum may be a precursor for the additional membranes through an autophagic pathway [46,47], is an alternative possibility.
By contrast, in Enteropsectra longa, the sporoblasts and mature spores were never seen surrounded by an additional membrane, which rules out exocytosis as an exit route. Instead, the spores pushed out and deformed the apical plasma membrane of the host intestinal cell (Fig 6; S6 Fig). Whether the final release step was by pinching of the plasma membrane at the base or by rupturing it is unclear, although the former is more probable, given that the intestinal cells were not seen to leak out. We observed spore sections in the lumen, far from any intestinal cells in the corresponding section, with either an additional membrane around them or none. A possible scenario is that the spores are first released with a membrane, and that the membrane then disintegrates (Fig 6B). Yet because we did not follow by serial sectioning the length of the spores, we cannot know for sure that those with a membrane were not still attached to the epithelial cell. We thus cannot rule out an alternative mechanism whereby the spores are released through a hole in the plasma membrane—although given spore size, this latter exit mechanism would likely lead to host cell rupture and death, an event that was never observed. Of note, another exit mode was noted in the human gastrointestinal microsporidia Enterocytozoon bieneusi (in Clade IV like E. longa), whereby the infected cell itself is extruded in the lumen [48–50]. Presumably, the cell then rapidly dies and the spores are released by disintegration of the enterocyte plasma membrane. In the present case of Enterospectra longa, the epithelial intestinal cell remains overall intact and only the spore exits, possibly with the surrounding enterocyte plasma membrane that then disintegrates.
Beyond access to a diversity of microsporidia, our collection of host-parasite combinations also provides a resource for defining the genetic basis of host resistance. Most current work on C. elegans and N. parisii is performed using the C. elegans reference strain N2 and the N. parisii ERTm1 isolate, yet this strain combination has been shown to lead to a very strong infection where the host does not mount an effective defense response (e.g. in comparison with C. elegans CB4856; [29]), thus making it a difficult system in which to identify immune defense pathways. The present collection offers many further possibilities of genetic screens using induced mutations or natural genetic variation for resistance pathways.
Overall, we here considerably enlarged the resources and knowledge on the microsporidia infecting bacteriovorous terrestrial nematodes. These microsporidia are diverse in terms of phylogenetic relationships, spore size and shape, the presence of vesicles containing spores, host specificity pattern, host tissue tropism, host cell intracellular localization and cellular exit route.
Hundreds of samples, mostly from rotting fruits, rotting stems and compost, were collected worldwide over several years, and nematodes were isolated as described [11]. The nematode species was identified as described [11,33], using a combination of morphological examination (dissecting microscope and Nomarski optics), molecular identification (18S, 28S or ITS rDNA) and mating tests by crossing with close relatives. Isogenic nematode strains were established by selfing of hermaphrodites or for obligate male-female species from a single mated female. Individuals of strains showing a paler intestinal coloration (Fig 1A) were examined by Nomarski optics. Strains with meronts and spores in the intestinal cells or elsewhere were labeled as suspected to harbor a microsporidian infection. Each nematode strain was then frozen and stored at -80°C.
For this study, these frozen nematodes were thawed and maintained on nematode growth media (NGM) seeded with E. coli OP50 at 23°C. The microsporidian strain was identified after the strain identifier of its host nematode strain (itself identified according to C. elegans community rules; http://www.wormbase.org/about/userguide/nomenclature), with an additional "m" between the letters and the numbers for the microsporidia. For instance, a microsporidian strain from the nematode strain JU1762 was named JUm1762. Previously published nematode-infecting microsporidian strains keep their names: ERTm1 (from strain CPA24), ERTm2 (from JU1348), ERTm3 (from JU1247), ERTm4 (from JU1395), ERTm5 (from JU2055) and ERTm6 (from JU1638) [26,29,30,39,51] (Tables 1 and 2). ERTm4 was previously reported to correspond to N. parisii infection [29], but as no sequence data was available in GenBank, we also sequenced the SSU rDNA and β-tubulin genes for this study.
We sought to amplify by PCR and sequence fragments of two microsporidian genes (SSU rDNA and β-tubulin genes) from all potentially infected rhabditid isolates. Ten infected worms were placed in a PCR tube with 10 μl single worm lysis buffer (1X PCR buffer (DreamTaq Buffer 10X, Theromo Fisher), 1 mM MgCl2, 0.45% Tween 20 and 170 ng/μl proteinase K), which was then treated at 60°C for 60 min, followed by 95°C for 15 min. This DNA extract was then used as DNA template. To amplify microsporidian SSU rDNA, primers v1f (5'-CACCAGGTTGATTCTGCCTGAC-3') and 1492r (5'-GGTTACCTTGTTACGACTT -3') [52] were used to amplify strains JUm408, JUm1254, JUm1483, JUm1504, JUm2009, JUm2106, JUm2131, JUm2132, JUm2287, JUm2520, JUm2526, JUm2551, JUm2586, JUm2590, JUm2671 and NICm516. We used v1f and 18SR1492 (5'-GGAAACCTTGTTACGACTT-3') to amplify sequences of JUm1456, JUm1460, JUm1505, JUm1510, JUm1670, JUm2552, JUm2793, JUm2796, JUm2799, JUm2816, JUm2825 and JUm2895. We designed a new pair of primers SPF (5'-GATACGAGGAATTGGGGTTTG-3') and SPR (5'-GGGTACTGGAAATTCCGTGTT-3') for JUm2507, JUm2747, JUm2751 and JUm2772. We failed to amplify SSU rDNA for JUm1501 and NICm1041.
To amplify the microsporidian β-tubulin gene, newly designed forward primer βn1F (5'-ACAAACAGGNCARTGYGGNAAYCA-3') and reverse primer βn1R (5'-TGCTTCAGTRAAYTCCATYTCRTCCAT-3') were used. To obtain the β-tubulin gene sequence of JUm2551 and JUm1456, nested PCR was performed using first primers βn1F and βn1R then βnOF (5'-CCGGACAATATCGTCTTTGG-3') and βnOR (5'-CAGCTCCTGAATGCTTGTTG-3') (S1 Table). PCR products showing a positive signal by gel electrophoresis were sequenced on both strands on ABI 3730XL sequencing machines (MWG). SSU rDNA of five additional N. parisii strains (JUm1248, JUm1249, JUm1253, JUm1762, JUm1893) were provided by Aurore Dubuffet and Hinrich Schulenburg. The results were analyzed using Geneious v7.1.7 [53] and compared by BLAST with the NCBI database (http://blast.ncbi.nlm.nih.gov/Blast.cgi). Note that some PCR products could not be amplified (S1 Table). Especially, we failed to amplify the β-tubulin gene in several Oscheius infections. Both genes fail to amplify for the putative microsporidian infection of the O. tipulae strain JU1501 and this infection could not be characterized.
SSU rDNA and β-tubulin gene sequences have been submitted to GenBank under accession numbers KX352724-KX352733, KX360130-KX360167 and KX378155-KX378171 (S1 Table).
SSU rDNA and β-tubulin gene sequences of microsporidia from this study were analyzed with those of other published microsporidian species and fungi (Rozella spp. for SSU rDNA, Basidiobolus ranarum and Conidobolus coronatus for β-tubulin and concatenated sequences of both genes) as outgroups (Figs 2 and 3; S2 Fig) [34]. For phylogenetic analysis of SSU rDNA genes, 28 out of 33 sequences obtained from this study were compared with 11 sequenced Nematocida isolates (ERTm1, ERTm2, ERTm3, ERTm5, ERTm6, JUm1248, JUm1249, JUm1253, JUm1762, JUm1893, and JUm2807), JUm1396, 60 other microsporidian species chosen from all five major clades of microsporidia [34] and two Rozella species [54]. For analysis of β-tubulin genes, only sequences from the six Nematocida species (ERTm1, ERTm2, ERTm3, ERTm5, ERTm6 and JUm2807) and 18 other published microsporidian species were available to be compared with our 32 sequences (S1 Table). To phylogenetically analyze both genes together, we concatenated the two genes of stains ERTm1~6, JUm2807, our 30 strains, 10 other microsporidia species and two outgroups (B. ranarum and C. coronatus). Sequences were aligned using Geneious v7.1.7 with default parameters and further aligned manually and concatenated if available. The alignments were imported to MEGA 6 [55] to estimate the best DNA evolution models and compute mean genetic distances (1000 bootstrap replicates). Bayesian inference phylogenies were constructed using Mesquite v3.04 [56] and MrBayes v3.2.2 [57], with the same DNA models as above [58] and refined by FigTree v1.4.2 (http://tree.bio.ed.ac.uk/software/figtree/).
Worms were frozen in M9 buffer [35] supplemented with 20% BSA (Type V) in the 100 μm cavity of an aluminium planchette, Type A (Wohlwend Engineering, Switzerland) with a HPM 010 (BalTec, now Abra Fluid AG, Switzerland). Freeze substitution was performed according to [59] in anhydrous acetone containing 2% OSO4 + 2% H2O in a FS 8500 freeze substitution device (RMC, USA). Afterwards samples were embedded stepwise in Epon. To achieve a good infiltration of spores, the infiltration times in pure resin were prolonged for 48 h compared to the published protocol. After heat polymerization thin sections of a nominal thickness of 70 nm were cut with a UC7 microtome (Leica, Austria). Sections were collected on 100 mesh formvar coated cupper grids and poststained with aqueous 4% uranylacetate and Reynold’s lead citrate. Images were taken with a Tecnai G2 (FEI, The Netherlands) at 120 kV and equipped with a US4000 camera (Gatan, USA).
Spore size was measured as described [22]. Briefly, infected nematodes were photographed by Nomarski optics and spores were measured using the Image J software [60]. We only took into account spores with a clear outline within the focal plane. In species with two spore size classes, large spores are less numerous than small ones and they are found in groups. When measuring, the spores were first assigned to a size class, in part based on the spatial clustering of large spores. 20 spores were measured for each spore type; except N. ausubeli, for which 42 small ones and 40 large ones were measures.
For the microsporidian spore preparation, we first tried the methods previously established for N. parisii and N. ausubeli [22,51]. Because wild nematodes naturally live in habitats with various microbes [16,17], the microsporidia-infected nematode cultures generally originally contained other microbes, such as bacteria, fungi, or even viruses. In order to obtain a relatively pure microsporidian spore preparation, we treated the nematode cultures repeatedly with antibiotics (100 ug/ml gentamycin, 50 ug/ml Ampicillin, 50 ug/ml Kanamycin, 20 ug/ml Tetracycline, and 50 ug/ml streptomycin), monitoring the presence of non-E. coli bacteria and fungi on the plate. Nematode strains do not lose the microsporidian infection after antibiotic treatment. After antibiotic treatment, if the appearance of a plate with infected worms looks like those with bleached worms, we considered the plate to be clean and the infected worms were used to extract clean spores. Even though inconspicuous microbes may still be carried over, as we know so far, none of them could prevent the worms from getting infected with microsporidia nor induce similar symptoms as microsporidia.
Antibiotic-cleaned worms without other detectable microbes were harvested in 2-ml microfuge tube and autoclaved silicon carbide beads (1.0 mm, BioSpec Products, Inc.) were added. The tube was then vortexed for 5 min at 2,500 rpm and the lysate of worms filtered through a 5 μm filter (Millipore) to remove large worm debris. Spore concentration was quantified by staining with chitin-staining dye direct yellow 96 (DY96).
This method worked well on N. major and N. homosporus, but spores of Clade IV species extracted this way could not infect any worms. To prepare infectious spores of these species, we used instead a plastic pestle to crush worms manually, and stored these spore preparations at 4°C.
Nematocida species spore preparations could generally be stored at -80°C for later infection tests. However, storage at -80°C could affect the infection efficiency of these spore preparations. Indeed, when we made a fresh N. ausubeli (JUm2009) spore preparation and used it directly for infection tests, it could infect O. tipulae strains JU1510 and JU2552, with meronts and spores found in their intestinal cells at 120hpi. One month later, we used the same batch that had been stored at -80°C to infect C. elegans (N2), O. tipulae (JU1483, JU170, JU1510 and JU2552). At 120 hpi, 100% of N2 adult worms were infected, while none of the O. tipulae strains became infected. These results suggested that this spore preparation became less infectious after being frozen and stored at -80°C for one month, which did not compromise infection in C. elegans but did compromise infection of O. tipulae. For further specificity tests, spore preparations of N. major, N. homosporus and Clade IV species were then used within two hours after extraction, without freezing.
20 uninfected L4 or young adults (i.e. prior to first egg formation) were transferred to a 6 cm NGM plate seeded with E. coli OP50. 5 million microsporidian spores in 100 μl distilled water were placed on the E. coli lawn. The cultures were then incubated at 23°C. The infection symptoms of 20 adults were checked by Nomarski optics at 72 hours after inoculation. If no infection symptoms were found at this timepoint, they were scored a second time at 120 hours post-inoculation.
Two transgenic C. elegans strains, ERT54 jyIs8[C17H1.6p::gfp; myo-2p::mCherry] and ERT72 jyIs15[F26F2.1p::gfp; myo-2::mCherry] were used in infection assays to test infection specificity and transcriptional response of C. elegans to different microsporidian infections. These two lines express a constitutive fluorescent Cherry marker in the pharyngeal muscles and induce GFP upon infection with N. parisii [38]. In the first qualitative assay (23°C), we focused on the ERT54 strain. First, 10 L4 stage animals from seven naturally infected strains (C. elegans JU1762 with N. parisii infection, C. elegans JU1348 with N. ausubeli, C. briggsae JU2507 with N. major, O. tipulae JU1504 with N. homosporus, R. typhae NIC516 with N. homosporus, O. tipulae JU1483 with Enteropsectra, Oscheius sp. 3 JU408 with E. longa) were transferred to new plates and cultured for two days, in order to release microsporidian spores onto the plates. Then 10 L4 stage worms of the ERT54 strain were added onto these plates and onto a clean plate as control. Two days post-inoculation (dpi), a chunk was transferred to new plate to prevent starvation. One day later (3 days dpi), GFP expression of ERT54 animals (visualized using the Cherry reporter in the pharynx) and infection symptoms were scored. 20 worms showing GFP expression (if any, else the Cherry marker was used) were picked and transferred to a new clean plate. GFP expression was monitored on 8 dpi and 14 dpi. In the second quantitative assay (23°C), first, 10 L4 stage animals from five naturally infected strains (C. briggsae JU2055 with N. parisii infection, C. elegans JU2009 with N. ausubeli, C. briggsae JU2507 with N. major infection, R. typhae NIC516 with N. homosporus infection, Oscheius sp. 3 JU408 with E. longa infection) and uninfected C. elegans reference strain N2 (as negative control) were transferred to new plates and cultured for three days. Then 200 L4 stage worms of ERT54 or ERT72 were added. GFP expression of 50 worms (if possible) of reporter strains was monitored at five different timepoints (2 hours post inoculation (hpi), 4 hpi, 8 hpi, 28 hpi, 48 hpi) and infection symptoms were scored at 48 hpi.
For measurements of transcripts levels by quantitative RT-PCR (qRT-PCR) (primers used see S2 Table), 3000 synchronized N2 C. elegans L1 larvae were infected for 4 hours at 25°C with 5.0 x 105 ERTm1 (N. parisii) spores and 1.5 x 106 ERTm2 (N. ausubeli) spores. Prior analysis of serial spore dilutions determined that these ERTm1 and ERTm2 spore doses resulted in an average of 1 sporoplasm per L1 larva at 4 hpi at 25°C as measured by FISH to Nematocida rRNA. At 24 hpi, animals were harvested and RNA was isolated by extraction with Tri-Reagent and bromochloropropane (BCP) (Molecular Research Center). cDNA was synthesized from 175 ng of RNA with the RETROscript kit (Ambion) and quantified with iQ SYBR Green Supermix (Bio-Rad) on a CFX Connect Real-time PCR Detection System (Bio-Rad). Transcript levels were first normalized to the C. elegans snb-1 gene within each condition. Then transcript levels between conditions were normalized to uninfected N2 for C. elegans transcripts or normalized to ERTm1 rRNA for Nematocida rRNA.
|
10.1371/journal.pbio.2004786 | Morphological changes of plasma membrane and protein assembly during clathrin-mediated endocytosis | Clathrin-mediated endocytosis (CME) proceeds through a series of morphological changes of the plasma membrane induced by a number of protein components. Although the spatiotemporal assembly of these proteins has been elucidated by fluorescence-based techniques, the protein-induced morphological changes of the plasma membrane have not been fully clarified in living cells. Here, we visualize membrane morphology together with protein localizations during CME by utilizing high-speed atomic force microscopy (HS-AFM) combined with a confocal laser scanning unit. The plasma membrane starts to invaginate approximately 30 s after clathrin starts to assemble, and the aperture diameter increases as clathrin accumulates. Actin rapidly accumulates around the pit and induces a small membrane swelling, which, within 30 s, rapidly covers the pit irreversibly. Inhibition of actin turnover abolishes the swelling and induces a reversible open–close motion of the pit, indicating that actin dynamics are necessary for efficient and irreversible pit closure at the end of CME.
| Cells communicate with their environments via the plasma membrane and various membrane proteins. Clathrin-mediated endocytosis (CME) plays a central role in such communication and proceeds with a series of multiprotein assembly, deformation of the plasma membrane, and production of a membrane vesicle that delivers extracellular signaling molecules into the cytoplasm. In this study, we utilized our home-built correlative imaging system comprising high-speed atomic force microscopy (HS-AFM) and confocal fluorescence microscopy to simultaneously image morphological changes of the plasma membrane and protein localization during CME in a living cell. The results revealed a tight correlation between the size of the pit and the amount of clathrin assembled. Actin dynamics play multiple roles in the assembly, maturation, and closing phases of the process, and affects membrane morphology, suggesting a close relationship between endocytosis and dynamic events at the cell cortex. Knock down of dynamin also affected the closing motion of the pit and showed functional correlation with actin.
| Cells communicate with the extracellular environment via the plasma membrane and membrane proteins. They transduce extracellular signals and substances into the cellular plasma via cell surface receptors, channels, and pumps, as well as by various endocytic processes [1–3]. Cells also disseminate their intracellular contents to the extracellular space via exocytosis. These dynamic cellular processes are largely dependent on the assembly and catalytic function of various proteins in the plasma membrane. Clathrin-mediated endocytosis (CME) is conducted by more than 30 different proteins. Extensive studies using fluorescence imaging techniques revealed the spatiotemporal dynamics of individual proteins in living cells [4–6]. In addition, a number of in vitro studies revealed unique functions of these proteins in deforming the plasma membrane [7]. For instance, Bin-amphiphysin-Rvs17 (BAR) domain proteins bind to the surface of the lipid bilayer and induce membrane curvature and tubulation and are therefore presumed to be involved in membrane deformation in an early stage of CME [8]. Dynamin also induces membrane tubulation with a smaller diameter and vesiculation via a nucleotide-dependent conformational change, and therefore has been considered to be involved in the vesicle scission process [9,10].
Despite our increasingly detailed knowledge regarding the cellular dynamics of these proteins in vivo and their catalytic activity in vitro, the morphological changes of the plasma membrane during CME in living cells have not been studied. This has mainly been due to a lack of imaging techniques for visualizing the membrane. Electron microscopy (EM) has made a substantial contribution to the study of CME, owing to its high spatial resolution. The detailed morphological changes of the plasma membrane, together with the assembly of proteins, such as clathrin, have been imaged and analyzed in a series of images to understand the entire process of CME [11–15]. However, aligning a thousand EM snapshots still suffers from a large limitation in the time resolution. In contrast to EM, fluorescence labeling and imaging techniques are powerful tools for studying protein dynamics in living cells. Recent advances in these techniques allow time-lapse imaging of a single protein molecule in a living cell with subsecond time resolution. However, it is not suitable for imaging morphological changes of the plasma membrane in a living cell at a submicrometer scale.
Scanning probe microscopies, including atomic force microscopy (AFM), are powerful approaches for characterizing the surface of a specimen at nanometer resolution. Notably, high-speed AFM (HS-AFM) has been utilized to visualize various molecular structures and reactions at subsecond resolution in vitro [16–19]. We recently developed an HS-AFM for live-cell imaging and successfully visualized structural dynamics of the plasma membrane in living cells [20,21]. In this study, we utilize this HS-AFM to analyze the morphological changes of the plasma membrane during CME. To understand the role of specific proteins during the morphological change, HS-AFM is combined with confocal laser scanning microscopy (CLSM) so that we could simultaneously visualize membrane structures and protein localizations during CME in living cells. Overlaying AFM and fluorescence images reveals the dynamics of protein assembly and concomitant morphological changes of the plasma membrane with high spatial resolution. In particular, we elucidate the role of actin in the closing step of CME.
To reveal protein-induced membrane deformation during CME in a living cell, we first established a hybrid imaging system with HS-AFM and CLSM. We previously reported the development of a tip-scanning AFM unit and its combination with an inverted optical microscope with a fluorescence illumination unit [20,21]. In this study, we combined the HS-AFM unit with an inverted optical microscope equipped with a confocal laser scanning unit to increase the optical resolution. To obtain stable imaging, the stage was redesigned. The detailed configuration of the stage is described in Fig 1A and 1B. A cross-shaped movable XY-stage is mounted on the base plate of the inverted optical microscope stage, which allows the specimen to move independently of the AFM unit and the objective lens. The AFM scanning unit now has a 6.0 × 4.5 μm2 scanning area, which is larger than the previous unit (4.0 × 3.0 μm2), to enable imaging of a larger area of the cell surface.
We then established a procedure for aligning confocal fluorescence and AFM images. The details of the alignment method are described in S1 Fig. In brief, the probe was brought to approach and attach on the glass surface without scanning. The z-position of the probe tip and confocal plane was also aligned by setting the glass surface as a reference position (z = 0). The x-y position of the probe tip on the optical axis was determined by imaging an autofluorescence signal of the probe (S1A and S1B Fig). The center position of the probe fluorescence was defined as an origin (x, y = 0, 0), and the x-y position of the AFM image also refers to this scale (S1B Fig). Once the reference position of the probe tip and optical axis was determined, the cross-shaped sample stage allowed the specimen (cell) to move in x-y directions without changing the relative position between the HS-AFM and CLSM.
The spatial accuracy of the hybrid imaging described above was tested by observing a chemically fixed cell. CV-1 in origin with SV40 gene line 7 (COS-7) cells expressing enhanced green fluorescent protein (EGFP)-fused clathrin light chain a (EGFP-CLCa) were fixed and observed by hybrid imaging. Several different membrane structures could be identified in the AFM image of the cell surface (Fig 1): membrane invaginations of different sizes (diameters); ruffle-like short narrow protrusions; and large swellings. When the AFM image was overlaid on the confocal fluorescence image of EGFP-CLCa based on the x-y position alignment, several clathrin spots readily colocalized with membrane invaginations identified in the AFM image (Fig 1C). The section profile analysis revealed that the diameter of the membrane invaginations (pits) identified in the AFM images ranged between 150 and 400 nm, whereas those colocalized with a clathrin spot ranged between 150 and 350 nm (237 ± 78 nm [mean ± SD]; n = 4) (the details of the AFM image analysis are provided in S2 Fig). Because this size range of clathrin-coated pits (CCPs) was larger than the previously reported ones (20–175 nm) [11], we examined the size of the entire CCP by observing the cytoplasmic side of the plasma membrane. COS-7 cells were unroofed by the procedure previously described [22], fixed, and then observed by HS-AFM. As shown in Fig 1D, CCPs with hexagonal network of clathrin were clearly observed. The diameter ranged between 150 and 400 nm (225 ± 49 nm [mean ± SD]; n = 45), which approximately corresponds to 130 to 380 nm after subtracting the tip curvature. Therefore, we concluded that the membrane invaginations with clathrin fluorescence spot observed by our hybrid imaging system were CCPs, and the size of the pit ranged from 150 to 400 nm. This result matches well to previous EM observations of the CCP in COS-7 cells, in which CCPs larger than 150 nm were often observed [23].
To evaluate the accuracy of the image alignment, the position (x, y) of the membrane invagination in the AFM image was compared with that of the corresponding fluorescence spot in the confocal image. The x-y offset between the centroid of the membrane invagination in the AFM image and the fluorescent spot of clathrin was 27 ± 20 nm (mean ± SD; n = 7). Considering the diameter of the membrane invaginations (150–400 nm), we concluded that our image-overlaying procedure is accurate enough to merge the clathrin spot with the AFM image. It should be noted that the number of membrane invaginations in the AFM image was less than that of the fluorescent clathrin spots—i.e., there were some clathrin spots that did not colocalize with membrane invaginations in the AFM image. Such clathrin signals could be due to either clathrin-coated vesicles (CCVs) that had already budded from the plasma membrane or CCPs that formed in the basal surface of the cell.
To follow the entire process of CME, we then performed time-lapse hybrid imaging of living cells. A live COS-7 cell expressing EGFP-CLCa was observed by HS-AFM and CLSM in culture medium containing fetal bovine serum (FBS). The temperature was kept constant at 28 °C, at which the entire CME process—including scission step—was demonstrated to occur [24]. Both AFM and CLSM images were obtained every 10 s, and overlaid. Small membrane invaginations as described in Fig 1 were colocalized with EGFP-CLCa spots over several minutes (Fig 2A and 2B) (see video in S1 Movie). The average x-y offset of the fluorescent spot from the corresponding membrane pit in the AFM image was 38 ± 13 nm (mean ± SD; n = 8) (S3 Fig), which was slightly larger than that in the fixed cell (27 ± 20 nm). This is partly due to the time lag between AFM and CLSM scanning; although AFM and CLSM images were taken with the same frame rate (10 s/frame), the exact timing of data recording at a certain position in the scanning area was not perfectly synchronized because 2 different scanning systems are operated by 2 independent controllers. Considering the fact that CCPs are diffusing on the membrane (S1 Movie), the time lag between 2 scanning systems results in a small offset of each CCP spot between 2 different images. However, because this offset is much smaller than the size of the CCP (150–400 nm), we concluded that our hybrid time-lapse imaging procedure has spatial resolution high enough for assigning individual clathrin spots to the membrane invaginations in the AFM image.
Because the plasma membrane of a living cell is always fluctuating in x, y, and z directions, we carefully examined the effect of scanning parameter on the morphologies of CCPs. When the amplitude of the cantilever was increased by 5%, cortical actin network beneath the plasma membrane became more visible (see S4A Fig), which sometimes occurred during our time-lapse observation. In such a condition, the CCP (an arrow in S4A Fig) was still clearly observed, and the difference in the diameter was less than 9%, indicating that this amount of fluctuation had little effect on the morphological analyses of the CCP during CME (see later sections). In addition, “tip skipping” sometimes occurred near the CCP (S4B Fig). However, the frequency of tip skipping was very low (12 out of 272 CCPs), and tip skipping occurred on 1 or 2 consecutive scanning lines, which corresponds to approximately 37 nm. From these analyses, we concluded that fluctuation of tip–sample interaction does not affect the morphological analyses of the CCPs.
The fluorescence intensity of the clathrin spot and the diameter of the pit in the AFM image were plotted against time (Fig 2B). The clathrin signal appeared 20 to 30 s before the membrane started to deform. The clathrin signal then increased (growing phase) until it reached a stable phase as demonstrated by a previous study [25] (see S5A Fig for the definition of individual phases). During the growing phase, the aperture of the pit in the AFM images also increased (Fig 2B), suggesting that the size of the pit also enlarged during this period. During the stable phase, the aperture also remained almost constant. There were large variations in the duration of the growing and stable phases; the growing phase ranged between 40 and 280 s, and the subsequent stable phase lasted between 0 and 260 s (n = 35) (Fig 2C). Following the stable phase, the closing phase proceeded over a short period of time (20–50 s) (Fig 2B and 2C). Some pits closed in a single frame (<10 s). When these fast-closing events were observed at a higher scanning rate (1 s/frame), the pit closed in as fast as 3 s (S5B Fig). Notably, the clathrin spot remained for another 20 to 30 s after the pit closed and then suddenly disappeared. This could indicate that either the clathrin coat remains on the vesicle and is eventually disassembled by G-associated kinase (GAK) [26,27], or the vesicle eventually moves out of the focal plane of the CLSM. We obtained a similar result when the pit area, instead of the pit diameter, was plotted against time (S5C Fig). The total lifetime of the CCP ranged from 40 to 330 s (n = 113).
Assemblies of other CCP-related proteins were also investigated by time-lapse hybrid imaging. Epsin is known to add bending stress to the lipid bilayer at an early stage of CME, thus changing the membrane curvature [28], and it recruits clathrin to the pit surface. COS-7 cells simultaneously expressing mCherry-fused epsin and EGFP-CLCa were subjected to time-lapse hybrid imaging. Epsin started to assemble on the plasma membrane prior to the membrane invagination, which was similar to clathrin (Fig 3A and 3B; S2 Movie) (see also S6 Fig for other observations). However, statistical analysis of the timing of assembly and membrane invagination revealed that epsin assembles prior to clathrin; fluorescence spots of epsin and clathrin appeared 47 ± 9 (mean ± SD; n = 8) and 34 ± 13 s (mean ± SD; n = 35), respectively, before the membrane started to invaginate (Fig 3E). During the growing phase, fluorescence signals of both proteins increased (Fig 3B). Epsin and clathrin signals peaked at 14 ± 5 s (n = 8) before and 3 ± 7 s (n = 35) after the pit closed, and they disappeared at 13 ± 5 s (n = 8) and 39 ± 13 s (n = 35) after the closure, respectively. All of these results demonstrated that CCPs in COS-7 cells show a large variation in the lifetime (40–280 s), and their opening and closing events are tightly coupled with protein assembly.
Dynamin localizes at the neck of the pit and plays a role in the vesicle scission [15] in the last step of CME. COS-7 cells expressing mCherry-fused dynamin 2 (an isoform ubiquitously expressed in a variety of cells) together with EGFP-CLCa were subjected to the time-lapse hybrid imaging. Although dynamin plays a role in the vesicle scission, a dynamin signal started to assemble on the CCP in an early stage of CME as demonstrated in previous studies [4,29]; in our observations, it assembled 25 ± 12 s (mean ± SD; n = 13) before the membrane invagination began (Fig 3C, 3D and 3E, S3 Movie) (see S7 Fig for other observations). The signal gradually increased in the growing phase, but during the stable phase, the dynamin signal was not clearly defined compared to clathrin. The dynamin signal peaked with the same timing as pit closure and gradually decreased thereafter (it completely disappeared 38 ± 9 s [mean ± SD; n = 13] after the closure), consistent with the notion that it is involved in the last step of CME.
We further confirmed that the membrane pits observed were indeed CCPs and not other types of endocytic structures. Caveolae are found in another endocytic pathway that is mediated by other sets of proteins (caveolin, etc.) but also includes invagination of the plasma membrane. mCherry-fused caveolin1, a major component of caveolae, was expressed in COS-7 cells together with EGFP-CLCa, and live cells were subjected to time-lapse hybrid imaging. The clathrin spots did not colocalize with caveolin1 spots during the observation. Overlaying 3 images (AFM, EGFP-CLCa, and mCherry-caveolin1) clearly revealed morphological differences between CCP and caveolae (Fig 4A, see also S4 Movie). The aperture of caveolae ranged from 80 to 120 nm, whereas that of CCPs ranged from 150 to 400 nm (Fig 4B). Similar to the CCP, the aperture of the caveolae observed by AFM was slightly larger than that observed by EM [30], probably for the same reason as described above. Caveolae had longer lifetimes than CCPs; the average lifetime of a CCP was 81 ± 55 s, whereas caveolae remained open for over 400 s. They also showed different lateral movements in the plasma membrane; the diffusion coefficient of CCPs was 7.3 × 10−9 (cm2 s−1), whereas that of caveolae was 2.1 × 10−9 (cm2 s−1) (Fig 4C and 4D). Taken together, these results indicate that time-lapse hybrid imaging could identify and distinguish the different aperture openings and diffusion kinetics of the 2 types of invaginations.
In contrast to the growing and stable phases, which continue for more than 1 min, the closing and disassembly phase was completed relatively rapidly (<30 s). In many cases, the membrane aperture suddenly disappeared (Fig 2). However, the detailed image analyses revealed several unique membrane structures and dynamics in the closing step of the CCP, which include (i) capping, (ii) two-step, and (iii) re-opening (Fig 5A–5C). The capping motion was frequently observed in more than 50% of the CME events (54.9%, Fig 5D, S1 Table). A small membrane region adjacent to the CCP swelled and eventually covered over the pit (Fig 5A). The section profile analysis (S2 Fig) revealed that the swelling region was 378 ± 62 nm in diameter and 38 ± 10 nm in height (mean ± SD; n = 13), which is comparable to the pit size. The entire closing motion took 23 ± 13 s (mean ± SD; n = 48). This structure is very similar to the membrane protrusion observed by ion-conductance microscopy [31] (see Discussion section for details).
Two-step closing was observed in approximately 20% of the CME events (Fig 5D). The pit aperture first decreased to approximately 120 nm and then disappeared (Fig 5B). The duration of the small-aperture step was <40 s. AFM imaging with higher time resolution (2 s/frame) revealed two-step motions with faster small-aperture steps (approximately 10 s). These results indicate that many CCPs close with a two-step motion, but the duration of the smaller-aperture step varied from several seconds to 40 s. A re-opening motion was also observed in more than 10% of the total CME (Fig 5D). A pit once closed completely then re-opened after several frames of the closure (Fig 5C). The duration of the closed state varied between 10 and 70 s (30 ± 24 s, n = 8). During the closed state, clathrin signal decreased to between 14% and 70% (39 ± 18%, n = 8) and re-increased after the pit re-opened (Fig 5C). The position of re-opening was within 13 to 113 nm (52 ± 31 nm, n = 8) from the closed position. These observations were similar to what was previously described as “hot spot,” in which multiple cycles of assembly and disassembly of endocytic proteins such as clathrin or dynamin occurred in a limited area of membrane [32–34] (see Discussion section for details).
There was a clear distinction between the capping and re-opening motions: pits that closed with capping did not tend to re-open (Fig 5E), suggesting that capping plays a role in irreversible closing. In contrast, the two-step motion was not mutually exclusive to other motions so that we sometimes observed a two-step motion that finally culminated with capping (S8 Fig). The comparison of the total lifetime revealed a wide distribution in capping-ended pits, whereas two-step and re-opening motions showed a narrow distribution of about 100 s (Fig 5F).
Actin and actin-related proteins are also known to contribute to CCP assembly, although their exact role is not fully understood [14,35]. We previously observed and reported the dynamic turnover of the cortical actin network [36]; actin filaments are polymerized near the plasma membrane and descend into the cytoplasm. Therefore, we first examined the effect of actin inhibitors on the CME process. The analysis of the CCP lifetime in the presence of actin inhibitors revealed an inhibitory effect of the cortical actin network on the progress of CME. Cytochalasin B (an inhibitor of actin polymerization) and CK666 (an inhibitor of the Arp2/3 complex, which binds to F-actin and generates a branching point) both shortened the CCP lifetime, whereas jasplakinolide—which inhibits actin depolymerization and stabilizes the cortical actin network [36]—prolonged the lifetime (Fig 6A, S2 Table, S5–S7 Movies). In the presence of cytochalasin B, both the growing and stable phases shortened from 45 ± 34 s to 7 ± 12 s and from 88 ± 29 s to 59 ± 33 s, respectively, whereas there was little effect on the duration of the closing step (31 ± 9 to 22 ± 7 s) (Fig 6B, S3 Table, S8 Movie). This indicates that the collapse of the actin network accelerates CCP assembly and maturation, whereas the stabilization of the network inhibits the process. Dissecting the two-step closing motion also revealed that cytochalasin B and CK666 shortened the duration of the large aperture, whereas jasplakinolide prolonged it (Fig 6A), implying that actin dynamics accelerate the assembly of CCP-related proteins.
In addition to the lifetime of the CCP, actin dynamics are involved in the closing motion of the CCP. The most striking effect of the inhibitors was a reduction of the capping motion and an increase in re-opening motions; cytochalasin B and CK666 drastically reduced the frequency of capping motions (from 56% to 0.4% by cytochalasin B and from 56% to 3% by CK666) and increased the frequency of re-opening motions (from 20% to 67% by cytochalasin B and from 20% to 47% by CK666) (Fig 7A and 7C, S4 Table) (see S9 Fig for other examples). The involvement of actin polymerization in membrane swelling was also demonstrated by ion-conductance microscopy [31] (see Discussion section for details). Jasplakinolide also showed a similar effect but to a smaller extent (Fig 7A and 7C). These observations are in good agreement with the result that the capping and re-opening motions are inversely related (Fig 5E) and that actin polymerization plays a role in an efficient and irreversible closing of the vesicle. In addition to the re-opening motions, two-step motions were also increased by cytochalasin B and CK666 treatments (Figs 7A and 6A), suggesting that two-step motions and actin polymerization are tightly coupled. Blocking actin depolymerization, but not polymerization, affected the frequency of CCP formation as previously reported [35], implying that the cortical actin layer also has an inhibitory effect on CCP formation.
To confirm that membrane swelling in the capping motion was induced by actin, we followed the localization of actin during CME. COS-7 cells simultaneously expressing Lifeact-GFP and mCherry-CLCa were subjected to time-lapse hybrid imaging. There were several variations in the actin signal depending on the basal level of actin around the CCP (Fig 7D and 7E, S9 Movie) (see S10 Fig for other examples). When the basal level was low, a burst of actin assembly was observed when the CCP closed. More precisely, it started to increase before the pit closed (−59 ± 18 s [mean ± SD]; n = 14), and peaked slightly after (2.5 ± 7 s [mean ± SD]; n = 15) the pit closure (Figs 3E and 7D). The burst of actin signal intensity was tightly correlated with the membrane swelling in the capping motion; the actin signal peaked when the membrane swelled. This is in good agreement with the result obtained with cytochalasin B (Fig 7A), in which addition of cytochalasin B reduced the frequency of the capping motion. On the other hand, when the basal actin level around the CCP was high, the signal first decreased during the growing and stable phases, then increased again toward the end of the CME (Fig 7E). In this case, membrane swelling was not observed. Taken together, these results demonstrate that actin depolymerization occurs in the growth and maturation phases of CME, and active actin assembly is required for the irreversible scission of the vesicle from the plasma membrane. Furthermore, the swelling of the membrane sometimes developed into a ruffle-like protrusion, even after the pit closure (S10 Fig), demonstrating that active polymerization of actin occurs around the CCP and generates local forces on the membrane.
The involvement of other CCP-related proteins in morphological changes of the membrane was further investigated using RNA interference. Knockdown of dynamin 2 (Fig 8A)—in which 88% and 92% of dynamin 2 expression was suppressed after 24 and 48 h of transfection, respectively—did not affect the frequency of pit formation (Fig 8B) but reduced the occurrence of the capping motion from 59% to 35% (Fig 8C and 8D, S5 Table, S10 Movie). The effect was similar to what was observed in the presence of cytochalasin B (Fig 7A). However, in contrast to cytochalasin B and CK666, which increased both re-open and two-step motions, dynamin knockdown markedly increased the two-step motion but only slightly increased the re-open motion (Fig 8C). The detailed analysis of the two-step motion revealed a prolonged duration of the small aperture (Fig 8E, S3 Table). These results suggest that dynamin is involved in the capping formation as well as the complete closing of the CCP.
In addition to the prolonged duration of the small aperture (two-step motion), the duration of the large aperture was also slightly prolonged in the dynamin-knockdown cells, which resulted in a significant increase of the CCP population with total lifetime longer than 100 s (13% in control knockdown and 54% in dynamin knockdown) (Fig 8E, S6 Table). This is in good agreement with our observation that dynamin started to appear at the CCP during the growing phase (Fig 3C and 3D). These results suggested that dynamin is involved not only in the closing step but also in the assembly and maturation phases of the CCP, as was suggested in a previous study [25]. It should be noted that dynamin knockdown did not completely block the progress of CME, implying that a small amount of dynamin 2 slowly catalyzes the closing reaction or dynamin 1 is induced to compensate for the reduction of dynamin 2, as was demonstrated in the previous study [37,38] (see Discussion section for details).
Imaging the shape of the plasma membrane together with localization of proteins has been technically challenging. Although AFM was first applied to a living cell in 1992 [39] and a hybrid system of AFM and fluorescence microscope was available [40,41], their low scanning rate did not allow visualization of the endocytic process. Many other microscopic techniques such as high-frequency microrheology [42] and high-speed ion-conductance microscopy (HS-ICM) [31] have also been utilized for cell surface imaging and characterization. The recent development of HS-AFM with a sample-scanning system overcame the problem of limited time resolution and visualized the dynamics of cell surface structures with subsecond time resolution [43]. However, AFM with a sample-scanning configuration is not suitable for combining with high-resolution live-cell imaging of CLSM because this configuration moves the specimen (living cells) while the position of the cantilever is fixed. In this study, we utilized a tip-scanning type of HS-AFM coupled with CLSM and were able to successfully visualize morphological changes of the plasma membrane during CME in a living cell. Our AFM images revealed unique membrane structures at the end stage of CME as well as the role of CME-related proteins, especially actin and dynamin, in such morphological dynamics. Notably, some of these structures are similar to what were previously observed by different approaches (TIRF, ion conductance, etc.), demonstrating the fidelity of our imaging technique.
Recent advances in fluorescence microscopy enabled highly precise spatiotemporal analyses of proteins involved in CME in yeast [44,45] and animal cells [4]. They revealed the timings of protein assembly at the CCP and disassembly after the pit closure. On the other hand, the morphological changes of the plasma membrane during CME were mainly drawn based on EM snapshots of fixed and stained cells. The localization of a specific protein on the CCP has also been revealed by immune EM and by fluorescence EM. For instance, clathrin exists at the place where the membrane is slightly bent [11,12]. However, the time resolution of the snapshot analysis is limited and is not suitable for following membrane dynamics. In time-lapse fluorescence imaging, a complete pit closure was detected by a pH-sensitive fluorescent dye combined with fast exchange of the external medium between high- and low-pH solutions [4,46,47], but other morphological changes of the membrane—such as invagination and protrusion—could not be detected. Our time-lapse hybrid imaging of HS-AFM and CLSM rendered both the membrane morphology and protein localization with a time resolution of several seconds, which was particularly suitable for tracking the morphological changes of the CCP together with the assembly of specific proteins.
Our observations of fixed cells, unroofed cells, and living cells all revealed that the size of the CCPs varied between 150 and 400 nm, which is slightly larger than the CCPs in other cell lines. This is somehow similar to clathrin plaque, which is larger than CCPs and undergoes a different mechanism of endocytosis (50–500 nm in HeLa cells and 50–300 nm in skin melanoma cell line 2 [SK-MEL-2]) [12]. However, we believe that what we observed in this study were CCPs and not clathrin plaques because of the following two reasons. First, although the clathrin plaque is involved in endocytosis, it initially forms a flat sheet of clathrin on the flat plasma membrane without significant membrane invaginations. On the other hand, in our observations, the membrane started to invaginate approximately 30 s after clathrin started to assemble (Figs 2 and 3E). Second, a previous study reported that COS-7 cells contain clathrin plaques at the basal plasma membrane but not at the apical (nonadhering) surface [33].
Our image analysis of CME revealed several unique membrane dynamics at the end of the process, which is coupled with the function of related proteins (Fig 5): capping, a two-step motion, and re-opening. Capping occurred in most of the CME events (approximately 60%) and was mediated by actin (Fig 7) and dynamin (Fig 8). A region of the adjacent membrane swelled and covered over the CCP (Figs 5 and 7). The peak actin signal (GFP-Lifeact) corresponded with the timing of membrane swelling (Fig 7D), and the inhibition of actin polymerization by inhibitors (cytochalasin B and CK666) perturbed the capping motion (Fig 7A), indicating that the membrane swelling is caused by rapid and local actin polymerization. These characteristics of capping seem to be similar to membrane protrusions observed by HS-ICM combined with CLSM [31]; small membrane protrusions (caps) were frequently (101 out of 145 at 28 °C) observed beside the pit at the end of CME. The cap was abolished when actin polymerization was inhibited by latrunculin B [31]. Because such membrane protrusion was reported in several other studies [31,43], this could be a general mechanism of CME. In addition to this observation, we found that this capping motion was related to other closing motion (re-open) (Fig 7) and also to the function of dynamin (Fig 8), suggesting functional interaction between actin and dynamin at the closing step of CME (see later section for further discussion on dynamin). It is noted that most of the membrane swelling occurs at one side of the CCP and moves across the pit towards the opposite side (Figs 5 and 7), which is reminiscent of actin comet tails formed behind the motile Listeria monocytogenes [48,49]. We could not find any preference in the direction of the capping. It might be the case that a sudden burst of actin polymerization at a certain point on the CCP induces membrane swelling. These results clearly support the idea that a short burst of actin polymerization produces membrane swelling beside the pit.
In addition to capping, re-open motion was frequently observed in our AFM images. There are 3 possible interpretations of this events: (i) a new CCP was formed at the same position of the membrane after the previous vesicle was budded off (re-formation), (ii) the CCP is still connected to the plasma membrane with an unresolvable small aperture and reversibly changes the aperture size (re-widening), or (iii) the CCP is completely separated from the plasma membrane and then reversibly fuses back to the original position (re-fusion). In the first case (re-formation), the motion has several similarities to what was previously reported as the “hot spot,” which is approximately 400 nm in diameter and sequentially produces CCPs [34] with multiple cycles of dynamin polymerization [32]. It can be speculated that clathrin and other proteins remained in the hot spot and efficiently recruited proteins necessary for another round of CME. In our observation, clathrin also remained after the closure of the first CCP and started to increase again during the second growing phase (Fig 5C). The rate of the clathrin assembly was very similar between first and second CCP assembly (Fig 5C). These results strongly suggest that the re-open motion we observed in AFM could be re-formation of the CCP at the same spot soon after the preceding pit is closed. However, this could not totally exclude the possibility of re-fusion and re-widening models. Further analyses with higher spatiotemporal resolutions of both AFM and fluorescence signal will provide a clearer answer to this question.
It is noted that inhibition of actin dynamics not only decreased the capping motion but also increased the occurrence of re-opening motions (Fig 7A). There could be several possible explanations for this observation. If the re-open motion is a fusion of the vesicle back to the plasma membrane (re-fusion) as we discussed before, this result suggests that actin dynamics are required for irreversible detachment of the CCV from the plasma membrane. In one possible scenario, actin polymerizes around the vesicle and provides a driving force to push the vesicle into the cytoplasm by interacting with myosin in the cell cortex. There are a number of reports of nonmuscle myosins in the cell cortex [50–52], which are involved in various molecular events at the cell surface. Another scenario is that actin assembles near the plasma membrane but does not attach to the vesicle. Newly assembled actin filaments between the plasma membrane and the vesicle may spatially hinder the reversible fusion of the vesicle to the plasma membrane, which consequently pushes the vesicle into the cytoplasm, ultimately leading to endosome fusion, as was suggested by previous studies [53,54]. If the re-open motion is re-formation of a new vesicle (hot spot), this result suggests that actin dynamics have a negative effect on the formation of a new vesicle. However, our result that the inhibition of actin did not affect the number of newly formed CCPs (Fig 7B) does not support this possibility. To clarify the mechanism of the re-open motion by actin inhibitors, further studies will be required.
In addition to the closing motion, we found a role of actin dynamics in the growing phase of the CCP. Our observation that destruction of the actin network by cytochalasin B or CK666 accelerated the CME process (shortened the lifetime of open apertures) and stabilization of the network by jasplakinolide prolonged the lifetime (Fig 6) suggests that the cortical actin layer has an inhibitory effect on the progress of CME. This could be explained by the following 3 mechanisms. The first possibility is that the assembly of CCP protein components at the plasma membrane is inhibited by the cortical actin layer. Because the cortical layer is supposed to be similar to a hydrogel state, the diffusion of cytoplasmic proteins through the cortex is also supposed to be slow. The second possibility is that the cortical actin layer spatially inhibits the growth of the CCP. As the size of the CCP grows, neighboring actin filaments must be excluded. Although it is not clear whether the exclusion is mediated by physical force or an enzymatic process [55,56], the progress of CME is tightly coupled with the dynamics of the cortical actin network. The third possibility is that the actin cortex generates membrane tension that inhibits the progress of CME. Disruption of actin cortex reduces the membrane tension and accelerates the pit formation by clathrin coating.
The role of actin dynamics in CME has been debated in previous studies [53,57]. The cortical actin layer is known to be actively involved not only in endocytosis but also exocytosis. In a secretion process of neuronal cells, the cortical actin network is involved in (i) tethering secretory vesicles [58–63], (ii) providing a platform for directed movement toward the plasma membrane [64], and (iii) facilitating the generation of new release sites [65–68]. In endocytosis, actin is known to localize to the CCP in both yeast and mammalian cells [44,45]. Although several models have been proposed for the function of actin in CME, many details remain obscure. Our results demonstrate not only the involvement of actin in capping formation but also functional interaction of actin and dynamin in the closing step of CME.
In addition to actin, we demonstrated the role of dynamin in the closing motion (Fig 8). Knock down of dynamin reduced the capping and re-opening motions and increased the two-step motion (Fig 8), which is partly similar to the effect of actin inhibition (Fig 7). This effect can be partially explained by a feedback mechanism between actin and dynamin recruitment, which was previously reported [5]; dynamin knockdown decreased the accumulation rate of actin, which resulted in the decrease of the capping motion. In addition, the loss of dynamin apparently prolonged the duration of a smaller aperture size in the two-step motion (Fig 8), suggesting that dynamin plays a role not in the initial narrowing of the aperture but in the complete scission of the vesicle. This is compatible with the known function of dynamin: it binds to the neck of the CCP and catalyzes membrane scission [69,70]. On the other hand, a recent in vitro study reported that actin and BAR domain proteins, but not dynamin, are essential for membrane scission [71]. Therefore, it might be the case that the initial narrowing of the pit aperture is mediated by actin and BAR domain proteins, and the final scission step might be accelerated by a catalytic function of dynamin. In such a case, the inhibition of individual proteins would not completely abolish the complete process but would only slow down the progress.
We found that knock down of dynamin also prolonged the assembly and/or maturation phases of CME (Fig 8). Indeed, dynamin appeared on the CCP just prior to the initiation of membrane invagination and kept accumulating as the pit grew (Fig 3) [72]. The role of dynamin in the assembly and maturation phases has been debated in previous studies [5,25]. Because the catalytic activity of dynamin is regulated by nucleotide-based mechanisms [70], the assembly of dynamin at the CCP may not, in itself, induce any membrane deformations. Considering the fact that the dynamin knockdown slowed down the progress of CME, it might be the case that dynamin is necessary for recruiting other protein machineries to the CCP. Further analyses are required for elucidating the role of dynamin in the whole process of CME. Another possibility might be an activation of dynamin 1, which was recently reported in non-neuronal cells [38, 39]; the dynamin 1—which is presumed to be a neuron-specific isoform of dynamin but was recently found to be expressed in many non-neuronal cell lines—was activated in non-neuronal cells when CME was dysregulated with the expression of a truncated mutant of adaptor protein. Therefore, at nonphysiological low temperature, dynamin 1 might be activated and compensate for the reduction of dynamin 2. Further investigation is required to reveal functional interaction between 2 isoforms of dynamin in CME.
All of the observations described in this study were conducted at 28 °C. As demonstrated in previous studies, endocytic activity is largely affected by temperature, especially in neuronal cells [73–75]. This is partly due to the reduction of membrane fluidity and of catalytic activity of proteins involved. Actin polymerization is reduced in cultured cells at nonphysiological temperature [76,77]. Because the actin polymerization promoted CME (inhibition of actin dynamic elongated the lifetime, Fig 6), imaging at nonphysiological temperature may elongate the lifetime of CCPs, as was discussed before [31]. Therefore, it is highly intriguing to observe dynamic morphologies of CCPs at physiological temperature and reveal the involvement of membrane and protein dynamics in the CME process. For this purpose, the establishment of a stable imaging system of HS-AFM at 37 °C, as well as a higher scanning rate, is necessary.
COS-7 cells were purchased from DS Biopharma Medical (EC87021302-F0). Cytochalasin B was purchased from Sigma-Aldrich (St. Louis, MO), and jasplakinolide was purchased from Abcam (Cambridge, United Kingdom). The reagents were added to the culture medium at final concentrations of 2 μM for cytochalasin B and 1 μM for jasplakinolide. HEPES-NaOH (pH 7.0–7.6), which was used to maintain a constant pH of the medium during AFM observation, was purchased from Sigma-Aldrich. The mammalian expression vectors encoding Dyn2-pmCherryN1 and epsin 2-pmCherryC1 were gifts from Christien Merrifield (Addgene #27689 and #27673, respectively), and the vector for EGFP-Lifeact expression was a kind gift from Dr. Mineko Kengaku (Kyoto University, Kyoto, Japan). Silencer select siRNA targeting DNM2 (#s4212) and siRNA targeting Luciferase (#12935–146) were purchased from Ambion (Waltham, MA) and ThermoFisher Scientific (Waltham, MA), respectively. Transfection reagents Lipofectamin2000 and Effectene were purchased from ThermoFisher Scientific and Qiagen (Hilden, Germany), respectively. Anti-dynamin 2 antibody was from Cell Signaling Technology (Danvers, MA), and PVDF membrane was from Bio-Rad Laboratories (Hercules, CA).
At 1 or 2 d before AFM imaging, COS-7 monkey kidney–derived fibroblast-like cells were seeded on a poly-L-lysine–coated glass slide and grown at 37 °C with 5% CO2 in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% FBS. AFM imaging was performed in DMEM supplemented with 10% FBS and 10 mM HEPES-NaOH (pH 7.0–7.6). cDNAs encoding human CLCa (CLTA, NM_007096) and caveolin1 (CAV1, NM_00172895) were amplified by RT-PCR and cloned into the vector pEGFP-C3 (Clontech, Fremont, CA) to create fused proteins with EGFP. The plasmids were introduced into cells using Effectene Transfection Reagent according to the manufacturer’s protocol. At 24 to 48 h after transfection, the cells were used for AFM imaging. Expression of the fusion protein was confirmed by fluorescence signals from the cells. For experiments with fixed cells, the cells were fixed with 5% paraformaldehyde in PBS for 15 min at room temperature and washed with PBS. For unroofing cells on a cover glass, cells were mildly sonicated, as described in the previous study [22], and then fixed.
Cells were transfected with siRNAs using Lipofectamine 2000 (Invitrogen, Carlsbad, CA) following the manufacturer’s instructions and were harvested after 24 to 48 h for analysis by SDS-PAGE and immunoblotting, using PVDF membranes. Each membrane was cut into 2 halves at the protein size of 60 kDa; the upper half was blotted with anti-dynamin 2 antibody, and the bottom half was detected with anti-β-actin antibody.
BIXAM (Olympus Corporation, Tokyo, Japan)—which is a tip-scan–type HS-AFM unit combined with an inverted fluorescence/optical microscope (IX83; Olympus) equipped with a phase contrast system and a confocal unit (FV1200; Olympus)—was used for this study. The tip-scan HS-AFM imaging system was developed based on a previous study [20]. In brief, the modulation method was set to phase modulation mode to detect tip–sample interactions. An electron beam–deposited sharp cantilever tip with a spring constant of 0.1 N m−1 (USC-F0.8-k0.1, a customized cantilever from Nanoworld [Neuchâtel, Switzerland]) was used. All observations were performed at 28 °C. The AFM tip was aligned with confocal views as described in the Results section. The images from the confocal microscope and AFM were simultaneously acquired at a scanning rate of 10 s/frame. The captured sequential images were overlaid by using AviUTL (http://spring-fragrance.mints.ne.jp/aviutl/) based on the tip position. The fluorescence intensity was quantified by Image J software (http://rsbweb.nih.gov/ij/). The lifetime of the pit was analyzed with Metamorph imaging software (Molecular Devices, San Jose, CA). The diameter/height of the pit or membrane swelling region was obtained using AFM Scanning System Software Version 1.6.0.12 (Olympus).
For the measurement of the x-y offset between pit and clathrin fluorescence spots, the distance between the centroids of the AFM invagination and the fluorescence spot were measured in ImageJ. A diffusion coefficient of the pit was calculated based on the position of the pit. In the calculation, the effect of drift defined as a displacement of caveolin fluorescence spots in the fluorescence movie was corrected by subtracting a drift from pit position. For time-lapse analysis of AFM and fluorescence images of the CCPs, the time when the plasma membrane started to invaginate and when the pit completely closed on AFM images are defined as t = 0, and the fluorescence signal intensity was plotted against this time scale.
Data presented as graphs are from 3 independent experiments. The number of total CCPs analyzed for each analysis is specified in figure legends or the main text. Statistical analysis was performed by two-way analysis of variance followed by Student t test.
|
10.1371/journal.pcbi.1004373 | Dynamics of the Heat Stress Response of Ceramides with Different Fatty-Acyl Chain Lengths in Baker’s Yeast | The article demonstrates that computational modeling has the capacity to convert metabolic snapshots, taken sequentially over time, into a description of cellular, dynamic strategies. The specific application is a detailed analysis of a set of actions with which Saccharomyces cerevisiae responds to heat stress. Using time dependent metabolic concentration data, we use a combination of mathematical modeling, reverse engineering, and optimization to infer dynamic changes in enzyme activities within the sphingolipid pathway. The details of the sphingolipid responses to heat stress are important, because they guide some of the longer-term alterations in gene expression, with which the cells adapt to the increased temperature. The analysis indicates that all enzyme activities in the system are affected and that the shapes of the time trends in activities depend on the fatty-acyl CoA chain lengths of the different ceramide species in the system.
| The heat stress response in yeast is a model system for elucidating how cells organize and execute complex tasks. While a genomic response to heat is necessary, it is by itself too slow for immediate means of protecting the cell against damage. However, one observes changes in the physiology of the cells within a few minutes, which raises the question of how these changes are facilitated. The present study highlights one aspect of this fast response to heat stress. It deciphers the strategies with which the baker’s yeast Saccharomyces cerevisiae changes the concentrations of certain sphingolipids, which subsequently lead to targeted alterations in gene expression. The study is based on de novo time series measurements of sphingolipid concentrations, which are analyzed with a complex combination of dynamic modeling, reverse engineering, and optimization.
| Decades of research on sphingolipids have documented the enormous importance of this class of lipids in mediating a variety of critical cell functions. Sphingolipids exist in eukaryotic cells, where they serve not only as constituents of membranes but also as second messengers in different signaling transduction pathways. These can trigger specific changes in gene expression in organisms like baker’s yeast and aid the control of cell proliferation, differentiation, cell trafficking and apoptosis in mammals [1–4]. Different sphingolipids often mediate overlapping but distinct cell functions, and it is frequently the balance between different sphingolipid species that evokes a critical response. In particular, the balance among ceramide, sphingosine, and sphingosine-1-phosphate is critical for regulating stress responses, programmed cell death, cell proliferation, differentiation, and cancer survival [5].
The biosynthesis of sphingolipids has been characterized quite well in yeast as well as mammalian cells. It is governed by a complex and highly regulated network of pathways that synthesize and degrade the various sphingolipids and incorporate them into membranes or retrieve them from membranes as the situation demands [6,7]. The complexity of sphingolipid metabolism renders it close to impossible to understand intuitively how sphingolipid-mediated responses to different environmental stresses are launched and coordinated. It is clear that any type of genomic response would require at least 15 to 20 minutes, if not hours, for transcription, translation, and protein activation, depending on the particular species in question. However, the first aspects of many stress responses are observed within a few minutes, if not seconds. This important fact leaves no doubt that some important regulatory mechanisms must be activated immediately upon the onset of stress, and it has been shown that sphingolipids and trehalose belong to the first responders and react to stresses very quickly [8–12]. The sphingolipid-based stress response strategy is highly dynamic and difficult to measure experimentally, but we recently showed that it can be inferred computationally from metabolic profile data in the form of time series, if they are supported by a mathematical model that is analyzed with a customized optimization strategy. In the case of heat stress in Saccharomyces cerevisiae, the sphingolipid response seems to be a consequence of temperature-induced conformational changes in essentially all enzymes of the pathway system. The resultant alterations in enzyme activities lead to changes in the metabolic profile of sphingolipids, which secondarily induce the expression of specific genes that are beneficial for an effective heat stress response.
Interestingly, the inferred enzymatic profiles throughout the 30-minute heat stress response show distinct differences, but are similar for enzymes that are grouped into functional modules within the larger pathway system [13]. For example, the key enzymes ceramide synthase, sphingoid base kinase, and hydroxylase, which are all directly involved in sphingolipid biosynthesis and internal material distribution, show similar dynamics. These functional modules imply an intricate coordination of enzyme activities within the heat stress response. Our earlier computational analysis also revealed a rapid initial uptake of substrate and increased sphingolipid biosynthesis, which however lasts only for a few minutes. Afterwards, the key enzymes of de novo biosynthesis exhibit decreased activity or stop catalyzing altogether. By contrast, the enzyme Isc1 (IPCase) which is involved in converting inositol phosphorylceramide (IPC) compounds to ceramide, exhibits increased, persistent activation, suggesting that a few minutes after the heat stress begins, sphingolipid materials are retrieved from the membranes and redistributed to serve the cell’s needs. Finally, our computational results showed that after 30 minutes of heat stress, the metabolic profile has essentially returned to its initial state under optimal, cooler temperature, while many enzyme activities are still substantially altered. This observation suggests that the cells assure that the sphingolipid metabolites, as important regulators of cell function, are maintained in, or returned to, a particular metabolic state, which apparently is optimal in some sense, while the corresponding enzyme activities seem to be a means to this end and in themselves may be rather different under optimal and stress conditions.
Recent experimental studies have demonstrated that not only the main sphingolipid metabolites in yeast differentially respond to heat stress, but that even variants of the same key sphingolipids, which differ in the lengths of their lipid backbones and fatty acyl head groups, can play distinct roles in cell signaling [14,15]. These alternate signaling roles of sphingolipids have been receiving increasing attention from academia and the biomedical industry. They also lead to the question of how exactly these different sphingolipid species are synthesized and how cells regulate and control the concentration of each sphingolipid variant.
Again, an experimental investigation of the control strategies used by the cells seems difficult, and we therefore develop computational methods here that allow us to shed light on these strategies. Using the results from our previous analysis quasi as boundary conditions, we narrow our focus on the smaller system of ceramides and design a much more detailed model that accounts for different ceramide species. As in earlier studies, we use Biochemical Systems Theory (BST) as the modeling framework and design a combinatorial modeling approach that includes data preprocessing, dynamic flux estimation, and a modified multiple shooting optimization method. This custom-tailored method allows us to infer changes in enzymatic activities that lead to observed responses of the various ceramide species following heat stress.
Ceramides form a class of sphingolipids with a lipid backbone and a fatty acyl head group. Distinct variants of ceramides result from different backbones and fatty acyl CoAs. For example, C16-dihydroceramide (C16-DHC) is a ceramide with a DHS backbone and a C16 fatty acyl CoA (palmitoyl CoA) head group. Recent research has revealed that DHC species with long fatty acyl chain lengths, such as C18-DHC, as opposed to very long chain based DHC species, such as C26-DHC, have distinct signaling roles [15,16]. To understand and explain these subtle differences, it is necessary to characterize the metabolic mechanisms of ceramide biosynthesis and utilization.
Ceramide can be generated via two paths, namely de novo biosynthesis and conversion of inositol phosphoceramide (IPC). De novo biosynthesis of sphingolipids is initiated by the condensation of serine and palmitoyl CoA, a reaction which is catalyzed by serine palmitoyltransferase (SPT) (Fig 1). The product, 3-keto-dihydrosphingosine (3KDHS) is quickly reduced by KDHS reductase to dihydrosphingosine (DHS). DHS is a main source of ceramide backbone compounds. It can be converted into different dihydroceramides (DHC), due to multiple options for fatty acyl CoAs, which can serve as substrates for ceramide synthase [17,18]. The reverse reaction, from DHC to DHS, is catalyzed by dihydroceramidase [19]. DHC and DHS are key branch points in the sphingolipid biosynthesis pathway, because hydroxylases [20] can irreversibly convert these compounds into phytoceramide (PHC) and phytosphingosine (PHS), respectively. PHC and PHS may undergo reversible reactions catalyzed by ceramide synthase (PHS to PHC) and phyotoceramidase (PHC to PHS); the forward reaction (PHS → PHC) requires one from among several different fatty acyl CoAs as substrate[8].
De novo biosynthesis is not the only option for making sphingolipids available when needed. Equivalent to sphingomyelin in mammalian cells, yeast complex sphingolipids, including inositol phosphoryl-ceramides (IPCs), refer to a class of ceramides with one or two inositol groups attached. They can be formed from DHC and PHC substrates via catalysis by the enzyme IPC synthase. IPC can be irreversibly converted into mannose inositol phosphorylceramide (MIPC), which can furthermore form mannose di-inositol phosphorylceramide (MIP2C). Importantly, IPC, MIPC and MIP2C all can serve as sources of DHC and PHC, through catalysis by the enzyme IPCase (Isc1). Thus, utilization of IPC compounds serves as the second path of ceramide production [21].
The reactions described in the previous paragraphs form a complex metabolic network, in which different enzyme systems exert subtle control over the proper concentration profile of the various sphingolipid species. Focusing on the ceramide species, namely, DHC and PHC, we can describe the biosynthesis and utilization of DHC and PHC as a cooperation among five enzyme systems consisting of ceramide synthases, ceramidases, IPC synthases, IPCases, and DHC hydroxylases. Ceramide synthases and IPCases mediate the formation of DHC and PHC, while ceramidases and IPC synthases use DHC and PHC as substrates. The hydroxylation reaction, catalyzed by DHC hydroxylases, converts DHC into PHC.
Our previous results [13] indicate that the enzymatic activities within the sphingolipid pathway can be grouped by their responses to heat stress, and thereby suggest that the various sphingolipid species are components of highly coordinated enzyme modules. An important detail we discovered is that the de novo biosynthesis of sphingolipids is activated immediately upon heat stress, but sustained only for a few minutes. This phase is followed by the activation of IPC utilization, which leads to the accumulation of sphingolipids that are retrieved from complex sphingolipids in the membranes. In this earlier, larger model, a single variable represented the concentration of all dihydroceramides (DHCs), irrespective of their fatty-acyl chain lengths, and a second variable represented the collective concentration of all phyotoceramides (PHCs). The DHC and PHC levels are controlled by five enzyme classes, namely ceramide synthase, dihydroceramidase and phyotoceramidase, IPC synthase, IPCase (Isc1), and dihydroceramide hydroxylase. A simplified representation is shown in Fig 1; for details of the sphingolipid pathway at large, see [13,22,23]. Ceramide synthase and DHC hydroxylase showed a similar pattern of activity changes, with a strong initial uptake of material followed by a weak profile toward the end. By contrast, dihydro- and phytoceramidase showed entirely different patterns. While dihydroceramidase exhibited very strong activation in the first 10 minutes, which then receded, phytoceramidase exhibited less pronounced peaks of activation. IPC synthase and IPCase had similar activities, again beginning with a sharp spike, which decreased much slower than for ceramide synthase.
While these inferred activity patterns shed some light on how cells regulate their sphingolipid or ceramide contents, they are too coarse to account for changes in the regulatory patterns that govern distinct ceramide species. The present study fills this gap.
The enzyme activities inferred computationally in our previous study [13] demonstrate how yeast resets its ceramide concentrations under heat stress. The alterations in activities are globally coordinated and grouped, and show distinct regulatory patterns across different enzymes and even ceramides with different fatty acid chains. In the following, we first briefly summarize pertinent findings from this previous research, which focused on heat-stress induced enzyme responses within the broader sphingolipid pathway system, and then zero in on details of the responses among enzymes directly associated with ceramides of different fatty acyl chain lengths.
As mentioned above, the dynamics of ceramide species is governed by five groups of enzymes, which are shown in Fig 1 along with SPT and KDHS, which catalyze prior reactions steps. Our overall strategy for zeroing in on this sub-pathway is similar to our previous approach, although the detailed account here requires several methodological modifications, which are detailed in the section Materials and Methods. As an illustrative example, consider dihydroceramidase. In our previous study, this enzyme catalyzed a single reaction step from DHC to DHS. Now, dihydroceramidase is involved in five reactions, from C14 and C16 DHC to DHS, C18 DHC to DHS, and so on, to the conversion of C26 and C26:1 DHC into DHS. The relationships between fluxes in the previous and in the present model are shown in the right panel of Fig 2. References regarding the enzymatic reactions can be found in the section Pertinent Details of Sphingolipid Metabolism. The refined ceramide model discussed here shares many reactions with the previous sphingolipid model, but is expanded toward details regarding the availability of specific substrates.
The reactions of this detailed subsystem are presented in the left panel of Fig 2. The left side of the diagram shows the different DHC species, categorized by chain lengths, and the right side the corresponding PHC species. The processes in the center are associated with elongation and desaturation. We designed a dynamic pathway model of this system with methods of Biochemical Systems Theory [24–27], which have been widely documented in the literature.
Using the computational approach described in the Materials and Methods Section, we set out to infer detailed changes in enzymatic activities from time series measurements that had been generated specifically for the purpose of better understanding the heat stress response [14]. The data demonstrate that during the 30-minute heat stress response, different ceramide species respond in dramatically different ways, with some accumulating, others being reduced over time, and yet others maintaining a relatively stable concentration. We used these data to parameterize the model, while allowing the enzyme activity levels to change every three minutes. Custom-tailored methods, reminiscent of multiple-shooting optimization, were used for this purpose.
As a first diagnostics of the parameterization, we tested the goodness of fit of the model with time varying enzyme activities for the smoothed, interpolated ceramide data, which are given in Fig 3; the raw data are shown in the Materials and Methods Section. Specifically, every three minutes the enzyme activities in the model were allowed to change, and this procedure yielded a reasonably good fit. The results are not entirely smooth due to the abrupt changes in enzyme activities between modeling windows. Even shorter windows were not considered necessary, considering the magnitude of noise in the data and our intent merely to characterize trends in enzyme activities, whereas longer windows (5 or 10 minutes) did not lead to satisfactory fits. As a consequence, the comparatively minor changes in sphingolipids do not seem to affect their levels much. Therefore, much smaller fluctuations in these major metabolites during stress environment should be expected in order for cells to minimize side effects in other pathways.
The model with these optimized parameter values fits the smoothed, interpolated metabolite data quite well, and large-scale simulations (see Materials and Methods) confirmed that the model settings are robust to moderate perturbations. Our computational inferences of enzyme activities, derived from these metabolic measurements, reveal the dynamic patterns that generate the various ceramide species with different carbon chain lengths.
One should note that the shapes, rather than the magnitudes, of the inferred fold changes are more indicative of actual changes. The reason is that, due to insufficient experimental information, our method only infers the product of a rate constant and the corresponding enzyme activity, but cannot assign numerical values to the two factors separately. This situation is analogous to determining Vmax in a Michaelis-Menten reaction, rather than kcat and the total enzyme concentration.
Recent research has indicated that yeast ceramides of different types and with different fatty-acyl chain lengths can exert distinct signaling functions [15]. It is therefore important to investigate how yeast cells manage to make the different ceramide variants available when needed. Our results shed light on this question. The computational inferences of dynamic enzymatic activities in ceramide biosynthesis are in agreement with previous insights into the control of sphingolipids under heat stress, but have a much higher resolution. The earlier results had suggested that essentially all enzymes in the pathway are involved in the adjustment of sphingolipid metabolism under heat stress, rather than just a few key enzymes, as one might have expected. The same observation holds again for the results obtained here, where we zero in on the much smaller but very important ceramide subsystem. Our simulations suggest that, for each ceramide species, the accumulation or utilization is governed by all associated enzymes. The results also render it evident that the ceramide heat stress response is not a haphazard panic reaction by the cell, but a well-coordinated, highly cooperative adaptation whose implementation is shared among all associated enzymes.
The experimental metabolic time series data and our customized optimization method allowed us to infer trends in all five major enzymes: ceramide synthase, ceramidase, IPC synthase, IPCase, and DHC hydroxylase. Interestingly, these enzyme systems exhibit distinct dynamic activity patterns that strongly depend on the carbon chain lengths of their fatty acids.
The initial activation of ceramide synthase confirms experimental and computational results indicating that heat stress induces the de novo biosynthesis of ceramides. In particular, our earlier results had shown that the production of dihydrosphingosine (DHS), the precursor of DHC, is immediately and strongly triggered by heat stress. The results here indicate that some of this newly synthesized DHS is directly channeled into long chain DHC. After just a few minutes, the activity of ceramide synthase for C14-C24 returns to the former baseline under optimal temperature conditions. Especially for C18:1 and C24, a second peak is detected after about 20 minutes, which coincides with the time when long chain DHC ceramidases cease to be active. The synthesis of the corresponding PHC increases more slowly and peaks about six to eight minutes into the heat stress. This short delay between DHC and PHC synthesis may be explained with the fact that phytosphingosine, the substrate of PHC synthase, must first be produced from DHS. In contrast to these patterns, the activities for very long chain substrates (C26 DHC and PHC) only flare up very briefly, but strongly, and then remain rather low. Although this activity trend is short-lived, a comparison with our earlier results suggests that this pattern actually dominates the overall trend in ceramide synthase. Indeed, this suggestion aligns well with the fact that C26 PHC is by far the most prevalent ceramide variant under normal conditions (Fig 3).
By catalyzing the utilization of complex sphingolipids, IPCase is the second source for ceramides. In contrast to de novo biosynthesis, these processes initially increase much more slowly and exhibit a strong and long lasting activity peak between about 5 and 20 minutes (Fig 7). Subsequently, their activities essentially cease. Interestingly, the trends are very similar for DHC and PHC and for chain lengths up to 24, which may suggest that the same enzyme could catalyze all reactions for both DHC and PHC of chain lengths up to 24. The dynamics for C26 DHC and PHC is distinctly different.
The DHC hydroxylase reaction facilitates an internal redistribution between DHCs and PHCs. It is activated instantly for long chain DHCs, with a subsequent decrease in activity, whereas C24 activity occurs mostly between 10 and 20 minutes (Fig 8). The activities with respect to C26/C26:1 DHC show a mixed pattern with a very small magnitude.
The utilization of ceramides follows two routes, namely toward the sphingosine backbone via ceramidase, and toward IPC via IPC synthase. The long chain DHC ceramidases arguably exhibit the most striking pattern (Fig 5). Their activities rise immediately with heat stress and are sustained for about ten minutes, after which the activities drop quickly and cease altogether. As these activities are much higher than for the corresponding ceramide synthases, it appears that the cells are preferably channeling long chain material to DHS. The corresponding activities for PHC are not as clear-cut. They also increase, but not as quickly or strongly, and return to their baseline over the entire 30-minute time period. This difference between DHC and PHC substrates may again be explained with the fact that phytosphingosine and PHC must first be synthesized from dihydrosphingosine and DHC, respectively. Intriguingly, the activities for very long DHC substrates and for C26 PHC increase more gradually and peak between 5 and 10 minutes. As C26 PHC is the most prevalent ceramide species, this pattern dominates the overall trend in ceramidase.
Finally, IPC synthase incorporates ceramides into complex sphingolipids. Here, the activities for long chain DHC and PHC substrates rise instantly, and the activities essentially cease after about 20 minutes (Fig 5). By contrast, the highest use for very long chain DHC and PHC occurs at about 10 minutes of heat stress. As in other cases, these trends are in line with the overall trends we observed in our previous analysis.
Taken together, it appears that the immediate response to heat stress is the de novo synthesis of long chain DHC, its conversion into the corresponding PHC, and the return of some material to DHS. Also, some long chain DHCs and PHCs are incorporated into complex sphingolipids. Between 5 and 10 minutes of heat stress, long and very long chain PHCs are generated. Around 10 minutes into heat stress, complex sphingolipids are used to generate DHC and PHC of all lengths. Also at this time, C24/C24:1 DHC is converted into PHC. Most activities are back to normal at the end of the 30-minute period, which is consistent with our earlier findings [13].
The molecular and cell-physiological reasons for the differences in activity patterns towards substrates of different lengths are unknown. The most straightforward hypotheses might be that the differences are due to:
the existence of specific enzymes or isozymes for different substrates;
different affinities of the same enzymes to substrates with different N-acyl chain lengths;
compartmentalization of substrates and/or enzymes, which would allow the same enzyme to be regulated differently in its action on distinct substrates.
In mammalian cells, at least five ceramide synthases were identified [31], and they perform different but overlapping functions with respect to different fatty acyl CoAs. In yeast, LIP1, LAC1 and LAG1 are known as genes coding for subunits of ceramide synthase, but otherwise not much is known about the reactions associated with different ceramide species. Our computational inferences indicate that reactions using substrates with different fatty acyl chain lengths are grouped, often into long chain and very long chain classes, which could suggest that ceramide associated enzymes are regulated in a “fatty acyl chain length specific” manner. However, there is so far no evidence that different species of the enzymes in question, which seems to suggest that explanation (3) might be more likely than (1) and (2). Further experimental research will be needed to support or refute this conclusion.
Beyond the particular application to ceramide metabolism, the computational modeling and inference methods in this work demonstrate how metabolite profiles, obtained as time series data, may be used to decipher in vivo strategies with which cells organize their responses to environmental stimuli.
Ceramide time series data for the present study were obtained de novo, as described in [32]. S. cerevisiae cells were cultured overnight at an optimal temperature of 30°. In duplicate experiments, the cells were moved to a water bath that was kept 39°C, which causes heat stress in yeast, and sampled every 5 minutes between 0 and 30 minutes. The samples were analyzed with High Performance Liquid Chromatography and Mass Spectrometry (HPLC-MS) to yield heat stress time series data of the following dihydro- and phyto-ceramide species: C14, C16, C18, C18:1, C20, C22, C24, C26; here the numbers refer to fatty acyl chain lengths and “:1” refers to an unsaturated fatty acid with one double bond. Considering that some datasets were missing or measurements fell below the detection limit, we used C14/C16, C18, C18:1, C24/C24:1 and C26/C26:1 DHC/PHC to construct our mathematical model.
As in our previous study [13], we found it beneficial to interpolate the data in a smooth fashion. It seems reasonable to assume that the heat stress response is a continuous phenomenon, and that a minimally biased spline technique would reflect the true dynamics in acceptable approximation. The raw and smoothed, interpolated data are exhibited in Fig 10. One should note that the Y-axes in Fig 10 are quite different, which indicates a large variation in prevalence of the various ceramide species.
Yeast responds to heat stress within minutes. Among the different aspects to this response, the concentration profile of sphingolipids starts to change in less than two minutes. Due to the complexity of the pathway, these changes in synthesis and degradation, which result from activity changes in enzymes, can hardly be inferred with intuition alone. To shed light on the response strategy, we recently proposed a customized computational approach to infer the dynamic changes in enzymatic activities from sphingolipid time series data [13]. These data consisted of time course measurements that were taken under heat stress conditions every 5 minutes until the end of a 30-minute interval and contained concentration measurements of dihydrosphingosine (DHS), dihydrosphingosine 1-phosphate (DHS1P), phytosphingosine (PHS), phytosphingosine 1-phosphate (PHS1P), dihydroceramide (DHC) and phytoceramide (PHC).
In order to infer the heat-induced changes in enzymatic activities, we used a Generalized Mass Action (GMA) model, in which every process was represented as a product of a rate constant and of all variables that directly affected the process, raised to a power [24]. The specific model was adopted from our earlier work [22,23] and included 31 metabolites as dependent variables and 64 enzymes or cofactors as independent variables. With this model as base structure, we developed a piecewise optimization approach.
First, the time series measurements were interpolated by smoothing splines and then re-sampled to produce time series values for every minute during the experimental time period. These values were entered into the GMA model. For each time interval, we formulated and solved the optimization problem of finding a set of enzymatic activities to establish the lipid profiles in that specific time frame. Given 31 time points, we thus found 30 corresponding sets of enzymatic activities that generated the observed dynamic metabolic profiles. An additional randomization scheme allowed us to infer solution spaces and confidence bands rather than point estimates. Several validation studies confirmed the results. The trajectories of the computed enzymatic activities revealed interesting regulatory mechanisms of sphingolipid metabolism, as described in the introduction.
The basic concepts of the modeling method were taken from our previous study [13]. As before, we smoothed the heat stress time series data of the six DHC and PHC species with different chain lengths, considered one-minute time intervals, and developed a customized, piecewise optimization approach that allowed us to infer changes in enzyme activities in a step-by-step manner. Also as in earlier studies, the system was represented as a GMA model. The pathway system under investigation is depicted in Fig 2. It consists of three major parts: synthesis and utilization of DHC, synthesis and utilization of PHC, and fatty acid elongation.
In comparison to the previous inference of all sphingolipid enzyme activities, the subsystem we consider here is relatively small. However, by accounting for the different chain-length variants, the system is much more detailed and leads to surprising complexity. In particular, the fatty acid elongation process becomes critical, whereas it was modeled only coarsely in the earlier analysis. It is our goal here to tease out the details of fatty acid elongation and the synthesis and degradation of variant ceramide species with different fatty acyl groups.
All responses outside the subsystem addressed here are expected to be the same as in the larger system, within normal biological variability, which allows us to use the previous model as a large set of dynamic boundary constraints that govern the sphingolipid system at large. For instance, fluxes entering the ceramide subsystem are directly taken from the large sphingolipid model. Similarly, the heat stress concentration of palmitoyl fatty acyl CoA can be directly imported from the sphingolipid model. The table in Fig 2 summarizes constraints on fluxes in the present model, imposed by the prior model.
The resulting ceramide subsystem has 53 fluxes and 15 dependent variables (Fig 2). X1 to X5, X6 to X10, and X11 to X15 represent the corresponding species of C16-DHC to C26-DHC, C16-PHC to C26-PHC, and C16-fatty acyl CoA to C26-fatty acyl CoA, respectively.
Details of the two core components of our approach, namely the estimation of dynamic fluxes and of enzyme activities, are shown in the flowchart of Fig 11. The first task, as shown in the upper panel of Fig 11, consists of checking the mass balances within the system and to construct the stoichiometric matrix that describes the production and degradation rates of the dependent metabolites. Because the system has considerably more fluxes than metabolites, we are faced with a highly underdetermined system. We are dealing with this situation by solving the system in 30 pieces, starting from the initial steady state to time 1, from time 1 to time 2, all the way to the end of the heat stress experiment (30th minute). The following describes in more detail a customized optimization strategy with which we determine the flux distribution in each time point.
Each flux is determined such that the model as a whole matches the observed data within a sufficiently small range of noise. Furthermore, all subsets of these fluxes, for instance, those representing the DHC hydroxylases, must collectively be consistent with the known fluxes of our previous model (see table in Fig 2). Also, all fluxes should change relatively smoothly from one time interval to the next. Finally, all fluxes are subject to upper and lower bounds. These tasks are formulated as a constrained nonlinear optimization program. Specifically, at each time point, we use two types of constraints. First, we constrain the system by ensuring that the slopes of the dependent variables (X) at a given time point t are sufficiently close to X(t) − X(t − 1), which we accomplish by minimizing the sum of squared errors between these differences and the corresponding slopes of these X variables. Second, we constrain the system by ensuring that the fluxes entering the system from the outside are collectively equivalent to those of the former model (cf. Table in Fig 2). These two constraints can be formulated as a combined, single objective function. To achieve robustness of the solution, the system is solved repeatedly by assigning for all unknown fluxes initial values that are drawn randomly from the uniform distribution U(0.01, 100). The results suggested that 1,000 simulations return sufficiently diverse ensembles.
One could surmise that the rather strong variability in the time series data (Fig 10) would unduly affect the flux estimation. To test this hypothesis, we estimated fluxes based on interpolated concentration data that were perturbed in either direction by a random factor sampled from U(1/1.5, 1.5). We compared these estimated fluxes with those obtained from noise free interpolated data. The flux distributions show very similar patterns (Fig S4.1 and S4.2 in S5 Text); detailed procedures are provided in the Supplements.
As an example, the resulting fluxes of reactions catalyzed by ceramide synthase are given in Fig 12; other fluxes are shown in the Supplements (Fig S1 in S2 Text). Once the flux distributions are computed for each time point, we examine the histogram of each flux in each time point to ensure that the solutions are well constrained; details are presented in the Supplements (Fig S3 in S4 Text). The analysis revealed that most of the flux distributions at any given time point were rather tightly bell shaped, suggesting the use of the mean value of each flux at each time point as an appropriate, time-dependent estimate (Fig 13). To validate this result further, we compared the sum of squared errors (SSEs) between the 1,000 individual fluxes and the corresponding averaged fluxes at each time point. The SSE of the averaged flux always falls within the range of SSEs from individual fluxes (Fig S5 in S6 Text); further details are provided in the Supplements. We also redid the analysis with medians, but the results were essentially the same.
The method of Dynamic Flux Estimation [33] allows us to obtain important hints for how material flows within the system. The resulting flux estimates are also important for constructing a dynamic model of ceramide synthesis and utilization upon heat stress. For later purposes, we need to identify how substrates, enzyme, modulators and kinetic parameters contribute to the magnitude of a flux. Because the actual enzyme amounts and rate constants are not known, we assume, as it is commonly done, that enzyme activities enter a flux representation in a linear manner and that substrates contribute with a power of 1, which corresponds to a mass-action formulation. Given these assumptions we obtain coarse estimates of the product of the rate constant and the enzyme activity. This product is similar to a Vmax value, which by definition consists of the product of kcat and the total enzyme concentration. Some of these estimates are shown in Fig 14 as functions of time. (All estimates are presented in Fig S3 in S4 Text)
Based on these time dependent estimates of flux magnitudes, we employed smoothing splines with proper degrees to obtain smooth time trends in enzyme activities (please refer to lower panel in Fig 11). The functional representations of enzyme activities were entered into the ODE model in order to obtain trends in the different ceramide species. Under ideal conditions, these trends should match the observed concentration profiles. By using the residual errors between the results of the described modeling strategy and the data, we created an optimization strategy that iteratively refined the trends in enzyme activities. This strategy was gleaned from the method of Multiple Shooting, which is a well-documented methodology for fitting dynamical data in boundary value problems [34,35]. While most optimization methods that correspond to a single shooting strategy aim at searching for one parameter set to fit the observed dynamic trajectory in its entirety, multiple shooting splits the time series into successive time frames and initially searches for independent parameter sets that match the data one frame at a time. In many cases of complex dynamic systems, this type of multiple shooting has demonstrated a better performance than single shooting. Here, we address a separate initial value problem for each time frame, and the last data point in each time frame is not defined as a condition for the subsequent frame.
We applied the multiple-shooting inspired strategy to fit the various ceramide time series data. However, instead of searching for entirely new parameter sets for each time frame, the task here is simpler, because the algorithm searches merely for slight adjustments of the coarse functional forms of enzyme activities that we had previously derived for each time frame to fit the ceramide data. Specifically, we associated unknown coefficients C to the enzyme activities in the ODEs. For example, from the 3rd to the 6th minute of heat stress, the functional representation of the enzyme activity, f3−6minute (t), was replaced with C3−6minute * f3−6minute (t). With all coefficients set equal to one, the ODEs are unchanged. However, by using the coefficients as free parameters, we are now capable of adjusting the system dynamics in each time frame. Thus, we subdivided the 30-minute time frame of the heat stress experiment into 3-minute intervals and fit the data separately in each interval. Furthermore, to minimize bias, we executed the search algorithm with many random initial settings for each coefficient, each enzyme, and each timeframe so that we obtained ensembles of solutions within a larger solution space. The slightly adjusted enzyme activities were then collected for biological inference.
|
10.1371/journal.pntd.0004828 | Dengue Virus Nonstructural Protein 1 Induces Vascular Leakage through Macrophage Migration Inhibitory Factor and Autophagy | Dengue virus (DENV) is the most common mosquito-borne flavivirus; it can either cause mild dengue fever or the more severe dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS). One of the characteristic features of DHF/DSS is vascular leakage; although DENV nonstructural protein 1 (NS1) has been proved to induce vascular leakage after binding to Toll-like receptor 4, the down-stream mechanism has not yet been fully understood. In the sera of DENV-infected patients, the concentrations of DENV NS1 and inflammatory cytokine macrophage migration inhibitory factor (MIF) are positively correlated with disease severity, but whether DENV NS1 induces vascular leakage through MIF secretion remains unknown. We demonstrated that recombinant NS1 induced vascular leakage and MIF secretion both in human endothelial cell line HMEC-1 and in mice. Furthermore, these phenomena were inhibited in the presence of anti-NS1 antibodies both in vitro and in vivo. DENV NS1 also induced LC3-I to LC3-II conversion and p62 degradation in endothelial cell line, which indicated the formation of autophagy. To clarify whether MIF or autophagy mediated DENV NS1-induced vascular leakage, various inhibitors were applied. The results showed that DENV NS1-induced vascular leakage and VE-cadherin disarray were blocked in the presence of MIF inhibitors, anti-MIF-antibodies or autophagy inhibitors. An Atg5 knockdown clone further confirmed that autophagy formation of endothelial cells was required in NS1-induced vascular leakage. Furthermore, DENV NS1-induced LC3 puncta were also decreased in the presence of MIF inhibitors, indicating that MIF mediated DENV NS1-induced autophagy. Taken together, the results suggest a potential mechanism of DENV-induced vascular leakage and provide possible therapeutic targets against DHF/DSS.
| Dengue is a viral disease transmitted by mosquitoes. The symptoms of dengue are often mild; however, severe dengue is one of the leading causes of hospitalization and death among children in Asian and Latin American countries. A symptom of severe dengue is vascular leakage, which can result in fluid accumulation, hypotension, circulatory collapse, and even death. For dengue and severe dengue, there is no specific treatment, and the only supportive treatment is to maintain a patient’s body fluids at normal levels. As a result, investigating the mechanism of how dengue virus (DENV) causes vascular leakage is an important and urgent issue. In this study, we demonstrated that DENV nonstructural protein 1 (NS1) induced vascular leakage through the secretion of macrophage migration inhibitory factor (MIF) and the formation of autophagy. Inhibition of MIF or autophagy formation effectively reversed NS1-induced vascular leakage both in vitro and in mice. These results provide possible therapeutic targets for treating vascular leakage in severe dengue.
| Dengue virus (DENV) is the most common mosquito-borne flavivirus that spreads in tropical and sub-tropical areas. The World Health Organization estimates that more than 2.5 billion people, over 40% of the world’s population, are now at risk of dengue infection [1, 2]. DENV infection generally causes dengue fever (DF), which is often asymptomatic or results in a mild flu-like illness with intense joint pain and fever. However, a small proportion of cases develop into severe illness termed dengue hemorrhagic fever (DHF). DHF is characterized by vascular leakage, thrombocytopenia, and coagulopathy [3]. Among these characteristics, vascular (plasma) leakage results in hemoconcentration and serious effusions, which can lead to circulatory collapse and life-threatening dengue shock syndrome (DSS) [4, 5]. It has been estimated that there are 50–100 million infections and approximately 500,000 people with severe dengue requiring hospitalization each year globally. The mortality of DF is less than 1% with adequate treatment; however, severe disease carries a mortality rate of 26%. Despite the high mortality of DHF/DSS, there are still no effective drugs or vaccines available because of a limited understanding of the pathogenic mechanism [6].
DENV nonstructural protein 1 (NS1) is a 48 kDa glycoprotein that can be expressed on the cell surface as a dimer and secreted as a hexamer into the blood circulation of dengue patients. The NS1 hexamer is composed of three dimers, which forms a detergent-sensitive hydrophobic central cavity that carries a cargo of ~70 lipid molecules; the composition is similar to that of high-density lipoprotein [7–9]. The concentration of NS1 in the sera of DHF/DSS patients can reach 50 μg/ml, which is positively correlated with disease severity [10–12]. The secreted NS1 may bind to cell membranes via interactions with heparin sulfate and chondroitin sulfate [13]. NS1 can also interact with prothrombin to interrupt the coagulation cascade [14]. In addition, NS1 can activate complement to elicit complement-dependent cytotoxicity in endothelial cells or to escape from innate immunity attack [15–17]. Recently, NS1 has been shown to be able to induce vascular leakage via binding to Toll-like receptor 4 (TLR4) [18, 19]. Therefore, investigating the downstream effectors of NS1-induced vascular leakage may provide potential targets for treating DHF/DSS.
Vascular permeability is normally maintained by the well-regulated endothelial barrier structure, which plays a crucial role in the control of exchange of small solutes and macromolecules between the intravascular and interstitial space [20, 21]. The integrity of endothelial permeability is regulated by many factors. Under pathological conditions such as infection, vascular leakage may occur because of damage to endothelial cells or loss of endothelial barrier function [22]. The physical damage to endothelial cells can be a result of cell apoptosis, which will take time to repair. In contrast, dysfunction of the endothelial barrier is reversible and may occur because of exposure to various vasoactive mediators or cytokines leading to the disruption of cell-cell junctions [23]. Vascular leakage in DHF/DSS patients occurs on days 3–7 of the illness and will resolve within 1 to 2 days in patients who receive appropriate fluid resuscitation [24, 25]. Therefore, it is generally believed that a mechanism that induces vasoactive cytokines rather than structural destruction of endothelial cells may be the major factor responsible for vascular leakage in DHF/DSS [6, 26, 27].
In a previous study, we found that DENV infection can induce macrophage migration inhibitory factor (MIF) secretion, which can cause an increase in vascular permeability both in vitro and in vivo [28]. Using recombinant MIF, we further demonstrated that MIF induces endothelial hyperpermeability through autophagy and that this process is related to the degradation of junction proteins [29]. MIF is a 12.5 kDa protein that is widely expressed in different cells, including immune cells, platelets, hepatocytes, and endothelial cells. Under physiological conditions, MIF exists in cells as a trimer consisting of three identical subunits, resulting in a catalytic site located in the intermonomeric pocket. Under stress conditions, such as inflammation and hypoxia, MIF is secreted into the blood circulation to modulate both innate and adaptive immune responses [30]. Secreted MIF can bind to cell surface receptors such as CXCR2, CXCR4 and/or CD74 [31, 32], inducing downstream signals such as the phosphoinositide 3-kinase (PI3K)/Akt pathway or the mitogen-activated protein kinases (MAPK)/extracellular signal-regulated kinase (ERK) pathway [33, 34]. It is known that MIF secretion can also be induced upon TLR4 stimulation [35]. Therefore, it is possible that MIF-induced by DENV NS1 may play an important role in DENV-induced vascular leakage.
Autophagy is a degradation pathway that occurs when cells are under stress conditions such as starvation, hypoxia, and infection [36–38]. Autophagy begins with the sequestration of the area of the cytoplasm inside double-membrane vesicles called autophagosomes [39, 40], which subsequently fuse with lysosomes to form autolysosomes or late endosomes to produce amphisomes [41]. Two ubiquitin-like conjugation of autophagy proteins (Atg5 and Atg12) are essential for autophagosome formation. Atg5 and Atg12 promote lipidation of a cytosolic form of light chain 3 (LC3; LC3-I) to form the LC3-phosphatidylethanolamine conjugate (LC3-II). The lipidated LC3-II, which is tightly associated with the autophagosomal membranes, can be observed by immunofluorescence staining to monitor autophagy, in which LC3 puncta formation reflects the existence of autophagosomes. In addition, after fusion with lysosomes, adaptor protein p62 will be degraded in the autophagolysosomes. As a result, autophagy formation can be determined by the decrease of p62 or the increase of LC3-I to II conversion by immunoblotting analysis. It has been demonstrated that DENV infection promotes the formation of autophagy, which can enhance virus replication [42]. However, the role of autophagy in DENV-induced vascular leakage has not been studied. Therefore, we proposed and tested the hypothesis that dengue NS1 increases vascular permeability through MIF secretion and autophagy formation.
To assess the role of DENV NS1 protein in vascular permeability, recombinant serotype 2 DENV NS1 derived from human 293T cells (293T-NS1) and Drosophila S2 cells (S2-NS1) were used in this study. Different concentrations of 293T-NS1 were incubated with human endothelial cell line (HMEC-1) for 6 h and the endothelial permeability was determined by the transwell permeability assay. This result showed that 293T-NS1 increased endothelial permeability in a dose-dependent manner. At least 5 μg/ml of 293T-NS1 was required to increase the permeability (Fig 1A). The kinetic changes of 293T-NS1-induced endothelial hyperpermeability were also measured. Endothelial hyperpermeability was induced 3 h after incubating with 293T-NS1 (20 μg/ml), which persisted to 24 h (Fig 1B). Similar effects were found using S2-NS1 (Fig 1C). To determine whether NS1 caused vascular leakage in vivo, protein extravasation in the abdominal cavity of mice was measured 6 h after i.p. injection of bovine serum albumin (BSA) or S2-NS1 (Fig 1D). Protein concentrations in the abdominal lavages of S2-NS1-injected mice were significantly increased compared to those in BSA-injected mice, suggesting that S2-NS1 was able to induce vascular leakage in mice (Fig 1D).
To confirm that the vascular leakage was specifically induced by NS1, we co-treated different anti-NS1 antibodies with 293T-NS1 and examined whether 293T-NS1-induced vascular leakage could be blocked. In addition, a real-time cell analysis (RTCA) system was used to monitor the kinetic change of endothelial permeability. These antibodies alone did not have any effect on the endothelial permeability of HMEC-1 cells either measured by RTCA or transwell assay (S1A and S1B Fig). However, 293T-NS1-increased endothelial permeability was inhibited in the presence of monoclonal antibodies (mAb) or polyclonal antibodies (pAb) against NS1 as measured by RTCA (Fig 2A). It was noted that different NS1 mAbs showed different blocking effect of which mAb 2E8 was better than mAb DN5C6 (Fig 2A). On the other hand, isotype control mouse IgG (CTRL mIgG) did not block 293T-NS1-increased permeability (Fig 2A). Similar results were also observed using the transwell permeability assay (Fig 2B). Likewise, in vivo experimentation also showed that anti-NS1 mAb 2E8 and pAb could block S2-NS1-induced protein extravasation in mice nearly to the basal value of the abdominal cavity, whereas CTRL mIgG could not (Fig 2C).
To test whether MIF secretion was induced upon DENV NS1 stimulation of endothelial cells, the amount of MIF in the cell culture supernatant was determined by ELISA. As shown in Fig 3A, MIF secretion was induced by incubating 293T-NS1 with HMEC-1 cells. Anti-NS1 mAb 2E8 and pAb completely reversed 293T-NS1-induced MIF secretion, mAb DN5C6 showed partial inhibitory effect, while CTRL mIgG had no effect (Fig 3A). In addition, these antibodies alone did not alter the basal level of MIF secretion of HMEC-1 cells (S1C Fig). Similar to what we found in in vitro study, intraperitoneal or intravenous injection of S2-NS1 but not PBS into mice increased MIF concentrations in peritoneal lavage or plasma of mice, respectively (Fig 3B and 3C). Furthermore, anti-NS1 mAb 2E8 and pAb, but not mAb DN5C6 or CTRL mIgG, significantly inhibited S2-NS1-induced MIF secretion in mice (Fig 3B).
In our previous study, we found that MIF was involved in DENV-induced vascular leakage [28]; therefore, we tested whether inhibition of MIF could block DENV NS1-induced endothelial hyperpermeability. Inhibition of MIF by its inhibitors, ISO-1 or p425, decreased 293T-NS1-increased permeability as shown by RTCA (Fig 4A) and the transwell permeability assay (Fig 4B). In addition, anti-MIF pAb could also block 293T-NS1-increased endothelial permeability (Fig 4B). Rabbit IgG isotype control (CTRL RaIgG) was used as a negative control of anti-MIF pAb, which did not inhibit 293T-NS1-increased endothelial permeability. In addition, all these chemical inhibitors or antibodies alone did not have any effect on endothelial permeability, as shown in the supporting information (S1A and S1B Fig).
MIF was reported to induce vascular leakage through autophagy formation [29], so we assessed whether DENV NS1 could induce autophagy of HMEC-1 cells. PBS- or 293T-NS1-treated HMEC-1 cell lysates were collected. Western blot analysis showed that 293T-NS1 induced p62 degradation and LC3-I-to-LC3-II conversion, which indicated autophagy formation in HMEC-1 cells (Fig 5A). Furthermore, 293T-NS1 also decreased the protein level of VE-cadherin, which might result in endothelial hyperpermeability (Fig 5A). Because the function of autophagy is to digest or degrade organelles or proteins, we wondered whether autophagy mediate DENV NS1-induced VE-cadherin degradation. Immunofluorescence staining was thus applied. Double staining of VE-cadherin and LC3 showed that the number of LC3 puncta was increased after 6 h of 293T-NS1 treatment (Fig 5B and 5C). In addition, cytosolic VE-cadherin colocalized with the LC3 puncta was found in 293T-NS1-stimulated HMEC-1 cells, indicating that some of the VE-cadherin proteins were embedded by autophagosomes (Fig 5B and 5D). Inhibiting MIF by its inhibitor ISO-1 decreased 293T-NS1-induced autophagy formation, LC3 puncta and the colocalization of LC3 puncta with VE-cadherin (Fig 5B–5D).
To clarify whether autophagy mediated 293T-NS1-induced vascular leakage, autophagy inhibitors were used. RTCA results showed that 293T-NS1-induced endothelial hyperpermeability was inhibited by co-treatment with PI3K inhibitor 3-methyladenine (3-MA) or the reactive oxygen species (ROS) scavenger N-acetyl-L-cysteine (NAC) (Fig 6A). The results from transwell permeability assay also showed that both 3-MA and NAC inhibited 293T-NS1-increased endothelial permeability (Fig 6B), whereas neither 3-MA nor NAC alone had effect on endothelial permeability in vitro (S1A and S1B Fig). The importance of autophagy in NS1-induced endothelial hyperpermeability was further supported by the stable Atg5 knockdown HMEC-1 cells (shAtg5), which, unlike the control shLuc cells, were resistant to S2-NS1-induced endothelial hyperpermeability (Fig 6C). In vivo permeability assay was also applied to test whether inhibition of MIF or autophagy could rescue DENV NS1-induced vascular leakage in mice. The results showed that either inhibiting MIF or autophagy could rescue DENV NS1-induced vascular leakage in mice, indicating that both MIF and autophagy are involved in DENV NS1-induced vascular leakage (Fig 6D).
Because MIF was previously shown to increase vascular permeability through the disarray of endothelial junction proteins ZO-1 and VE-cadherin, we sought to determine whether NS1 alters the alignment of endothelial junction proteins [29]. The immunofluorescence staining results showed that 293T-NS1 increased the ratio of cytosolic/barrier VE-cadherin of HMEC-1 cells (Fig 7A and 7B). To determine whether MIF and autophagy are involved in NS1-induced VE-cadherin disarray, we treated HMEC-1 cells with NS1 in the presence of MIF inhibitor ISO-1 or autophagy inhibitor 3-MA. The results showed that 293T-NS1-induced VE-cadherin translocation was inhibited in the presence of MIF or autophagy inhibitors and these inhibitors alone has no effects on VE-cadherin distribution (Fig 7A and 7B).
Little was known about the pathogenic roles of NS1 during DENV infection until recently. Two independent groups published papers which demonstrated that DENV NS1 can induce vascular leakage via TLR4 [18] and anti-NS1 antibodies or that NS1 vaccination can block this effect [19]. In this study, our results confirmed their findings and further suggests that MIF-induced autophagy of endothelial cells may mediate NS1-induced vascular leakage. The hypothetical model of the pathway by which DENV NS1 increases vascular permeability is shown in Fig 8.
In this study, we found that 5 μg/ml 293T-NS1 was sufficient to induce endothelial hyperpermeability at 6 h (Fig 1A). In in vivo experiments, we injected 50 μg S2-NS1 into mice. Because the total blood volume of a mouse is approximately 2–3 ml, the sera concentration of NS1 in mice is approximately 20 to 25 μg/ml. Furthermore, because the serum concentration of NS1 in dengue patients is estimated to range from 0.01 to 50 μg/ml [10], our experiments mimic the pathological condition in dengue patients. Even though further study is required to understand the contribution of NS1 in vascular leakage of dengue patients, these results suggest that NS1 can directly bind to endothelial cells to cause vascular leakage in dengue patients.
To further understand the interaction between NS1 and endothelial cells, we used different NS1 antibodies to block its effect. It is known that NS1 can also induce pathogenic antibodies that can cross-react with endothelial cells and induce endothelial cell apoptosis through molecular mimicry [43, 44]. Some of these anti-NS1 antibodies can also recognize platelets, resulting in thrombocytopenia [45]. Other anti-NS1 antibodies can cross-react with thrombin and plasminogen, resulting in inhibition of thrombosis and enhanced fibrinolysis [46]. Therefore, we used two different anti-NS1 mAbs 2E8 and DN5C6. Both of which did not bind to endothelial cells. We found that anti-NS1 mAb 2E8 showed better effect than mAb DN5C6 to block the activities of 293T-NS1 and S2-NS1 to stimulate endothelial cells. Similar results were also found by Beatty et al. which demonstrated that not all anti-NS1 antibodies can inhibit NS1-induecd vascular leakage [19]. Therefore, certain regions of DENV NS1 are more important for NS1 to interact with endothelial cells to induce vascular leakage. Identification of these regions may shed light to generate antibodies or vaccines to block NS1-induced vascular leakage.
It is known that MIF knockout mice show lower hemoconcentration and lethality compared with normal mice during DENV infection [47]. In sepsis, knockout or inhibition of MIF also increased survival rate of mice [48–50]. Previously, we demonstrated that MIF could mediate DENV-induced junction disarray and increase permeability in endothelial cells [28]. In this study, we further demonstrated that MIF is involved in DENV NS1-induced vascular leakage. Inhibition of MIF by its inhibitors can prevent DENV NS1-induced vascular leakage both in vitro and in mice. It is known that in addition to endothelial cells, other cells such as peripheral blood mononuclear cells (PBMC) can secrete MIF during DENV infection. Therefore, MIF secretion can be induced by either DENV infection or NS1 stimulation of different cells in dengue patients. However, in addition to MIF, other cytokines may also contribute to vascular leakage during DENV infection. Modhiran et al. showed that the expression of several cytokines including IL-6, TNF-α, IL-8, and MCP-1 were up-regulated in PBMC after DENV NS1 stimulation [18]. Many of these cytokines can also increase endothelial permeability [51–54]. Furthermore, culture supernatants from DENV-infected macrophage can induce endothelial cell apoptosis which is blocked by anti-TNF-α antibodies [55]. Even though it is known that MIF can augment the secretion of TNFα and counteracts the anti-inflammatory action of glucocorticoids [56, 57], DENV-induced vascular leakage may involve different mechanisms and the importance of MIF as therapeutic target against DENV-induced vascular leakage should be further studied.
It is known that autophagy is induced by DENV to prevent cell death and enhance viral replication during infection in human hepatoma cell lines [42, 58, 59]. Autophagy not only provides isolated environment but also provides energy and materials required for DENV replication by regulating lipid metabolism [60]. In addition, recent study showed that autophagy plays an important role in the antibody-dependent enhancement response in Fc receptor-bearing cells [61]. However, the role of DENV-induced autophagy in endothelial cells has not yet been discussed extensively. It has been reported that DENV NS4A is able to induce autophagy [62], but whether NS1 can also induce autophagy has not yet been reported. In this study, we showed that DENV NS1 induced autophagy, which mediated NS1-induced vascular leakage. As autophagy is required during DENV infection, inhibition of autophagy may prevent vascular leakage as well as suppress DENV replication. However, further studies are required to validate the therapeutic effects of autophagy inhibitors as anti-DENV drugs.
Taken together, our results suggest NS1-induced MIF secretion and autophagy may represent potential therapeutic targets for preventing vascular leakage in DHF/DSS. Our study highlights DENV NS1 as an important pathogenic factor in DHF/DSS. NS1-induced MIF secretion and autophagy may contribute to vascular leakage in DHF/DSS. Even though NS1 purified from DENV-infected cells or patients should be used to further confirm the pathogenic effects of NS1 on endothelial cells in the future, NS1-induced vascular leakage may represent a disease model in mice to develop potential therapeutic drugs and vaccines against dengue [63–66].
All experiments were performed in conformity with the Guide for the Care and Use of Laboratory Animals (The Chinese-Taipei Society of Laboratory Animal Sciences, 2010) and were approved by the Institutional Animal Care and Use Committee (IACUC) of National Cheng Kung University (NCKU) under the number IACUC 99057.
HMEC-1 cells were cultured in Medium 200 (Thermo Fisher Scientific, Waltham, MA) supplemented with 10% fetal bovine serum (FBS; HyClone Laboratory, Logan, UT) at 37°C in a 5% CO2 atmosphere. Stable clones of luciferase (Luc)-knockdown HMEC-1 cells were generated by a lentivirus-based short hairpin RNA (shRNA) system (National RNAi Core Facility, Academia Sinica, Taipei, Taiwan) targeting sequence 5’-GCCACAACATCGAGGACGGCA-3’. A stable clone of Atg5-silenced HMEC-1 cells was a kind gift from Dr. Chiou-Feng Lin. Both the shLuc and shAtg5 HMEC-1 cells were selected with 2 μg/ml of puromycin (MDBio, Inc., Taiwan).
In this study, we used two different commercialized recombinant NS1 proteins which were expressed in non-bacterial systems. Mammalian recombinant DENV serotype 2 NS1 protein, 293T-NS1 (The Native Antigen Company, UK), was engineered and expressed in the human 293T cell line. Another recombinant DENV serotype 2 NS1 protein, S2-NS1 (CTK biotech, San Diego, CA), was expressed in Drosophila S2 cells. Recombinant NS1 proteins were tested for endotoxin contamination by the Limulus amebocyte lysate (LAL) assay using the LAL Chromogenic Endotoxin Kit (Thermo Fisher Scientific, Waltham, MA) and shown to be endotoxin-free. Background endotoxin concentration of 0.036 EU/ml was found in 20 μg/ml 293T-NS1 and 0.018 EU/ml in 20 μg/ml S2-NS1, respectively.
In this study, BALB/c mice were purchased from and maintained at the Laboratory Animal Center of NCKU. Purified recombinant DENV2 NS1 was used to immunize 6- to 8-week-old female BALB/c mice at a dose of 50 μg as previously described [67]. The first dose was administered in complete Freund's adjuvant (CFA), and the following three doses were administered in PBS. After sacrifice, mice splenocytes were fused with FO cells (Taiwan Medical Cell and Microbial Resources). The resultant hybrid cells were selected in hypoxanthine-aminopterin-thymidine medium. An ELISA was performed to screen for the specific antibodies against NS1. After the hybridoma clones 2E8 and DN5C6 were established, the hybridoma cells were i.p. injected into pristine-primed BALB/c mice to produce monoclonal antibodies in ascites. The antibodies were then purified using a Protein G column (GE Healthcare). Rabbit polyclonal anti-NS1 antibodies were purified from purchased recombinant DENV2 NS1-immunized antiserum (GeneTex, Inc., Irvine, CA). Endotoxin concentrations in these antibodies were also measured by LAL assay. Endotoxin concentrations in 30 μg/ml mAb 2E8, mAb DN5C6 and CTRL mIgG (Leadgene Biomedical, Taiwan) were 0.082, 0.092 and 0.028 EU/ml, respectively. Nevertheless, none of these antibodies alone could alter endothelial permeability nor induce MIF secretion as shown in the supporting information (S1 Fig).
In the in vitro experiments, 20 μg/ml (~ 400 nM) 293T-NS1 or S2-NS1 was applied. In the in vivo experiments, 50 μg S2-NS1 was applied by i.p or i.v. injection. Different anti-NS1 antibodies (mAb 2E8, DN5C6 and pAb) were used to block recombinant NS1-induced effects. The concentration of antibodies utilized in the studies was at 30 μg/ml (~ 200 nM) for in vitro experiments (Fig 2A, 2B, 3A and S1 Fig), and at 40 μg per mouse for in vivo setting (Figs 2C and 3B). Since IgG has two antigen binding sites, it can bind to more than one antigen by binding identical epitope carried on the surfaces of these antigens. Therefore, the amount of IgG antibodies used in current studies should be able to bind to most of the NS1 (~ 400 nM) we added for in vitro experiments. To inhibit MIF activity, the MIF tautomerase inhibitor (S,R)-3-(4-hydroxyphenyl)-4,5-dihydro-5-isoxazole acetic acid methyl ester (ISO-1) (50 μM; Calbiochem, La Jolla, CA) and 6,6'-[(3,3-Dimethoxy[1,1'-biphenyl]-4,4'-diyl)bis(azo)]bis[4-amino-5-hydroxy-1,3-napthalenedisulphonic acid] (p425)(100 μM; Calbiochem) were mixed with 293T-NS1 or S2-NS1 before treatment. Polyclonal anti-MIF antibody was purified using a protein G column (GE Healthcare), and 30 μg/ml was used to block MIF in 293T-NS1-treated cells. The endotoxin concentration in 30 μg/ml anti-MIF antibody and CTRL RaIgG (GeneTex) were 0.065 EU/ml and 0.02 EU/ml as determined by LAL assay, respectively. To inhibit autophagy, 5 mM of 3-MA (Sigma-Aldrich, St. Louis, MO) or 5 mM of NAC (Sigma-Aldrich) was used.
To measure the permeability of endothelial cells in vitro, we used two different methods in this study: the transwell permeability assay and real-time cell analysis (RTCA) [68]. For the transwell permeability assay, cells (2 x 105) were grown on a Transwell insert (0.4 μm; Corning B.V. Life Sciences, The Netherlands) until a monolayer was formed. The upper chambers were reconstituted with 10% FBS-containing medium with 293T-NS1, S2-NS1 and the inhibitors. At the indicated time points, the media in the upper chambers were changed to 300 μl of serum-free media containing 4.5 μl streptavidin-horseradish peroxidase (HRP) (R&D Systems, Minneapolis, MN). The medium (20 μl) in the lower chamber was collected 5 min after adding streptavidin-HRP and was assayed for HRP activity by adding 100 μl 3,3',5,5'-tetramethylbenzidine (TMB) substrate (R&D Systems). Color development was detected by a VersaMax microplate reader (Molecular Devices, Sunnyvale, CA) at 450 nm.
RTCA was used to test cell-cell or cell-matrix adhesion by detecting the electric resistance of the monolayered endothelial cells. High resistance indicates strong endothelial barrier function. Using this device allowed us to detect the kinetic changes in endothelial permeability. For RTCA, 1 x 104 HMEC-1 cells were grown on a 96-well E-plate to form a confluent monolayer. After 293T-NS1 and the inhibitors were added, the resistance of the monolayer was recorded by an xCELLigence Real-Time Cell Analysis System (Cambridge Bioscience, UK) for 24 h.
The method for testing vascular leakage in the peritoneal cavity was described previously [28]. Briefly, 8- to 12-week-old BALB/c mice were injected intraperitoneally with 50 μg of S2-NS1, which was solubilized in 500 μl of PBS with or without the inhibitors. The mice were sacrificed 6 h after the treatments and the abdominal cavity was washed with 5 ml PBS after sacrificing the mice. The concentration of protein in the abdominal lavage was determined using the BCA method (Pierce Biotechnology, Rockford, IL). Mean concentration was calculated with 3 to 5 mice in each condition.
The MIF concentration in the cell culture medium was tested by using an ELISA kit (R&D System, Minneapolis, MN) following the manufacturer’s instructions. The MIF concentration in the peritoneal lavage fluid or plasma of mice was tested by using another ELISA kit (BlueGene Biotech, China).
For Western blot analysis, VE-cadherin (BD Biosciences, Franklin Lakes, NJ), p62 (Santa Cruz, Dallas, TX) and LC3 (GeneTex) were detected using a 1:1,000 dilution of antibodies followed by a 1:6,000 dilution of HRP-conjugated anti-mouse or anti-rabbit immunoglobulin antibody (Leadgene Biomedical). The β-actin (Table 1) antibody (Sigma-Aldrich) was used at a 1:10,000 dilution as an internal control. Bound HRP-conjugated antibodies were detected using the Luminata Crescendo Western HRP substrate (Merck Millipore, Germany). The Western blot results were quantified using the Image J software program.
Cell monolayers were seeded onto microscope cover glass. After treatment, the cells were fixed in 4% paraformaldehyde for 5 min, followed by three washes with PBS. The cells were then blocked with SuperBlock blocking buffer (Thermo Fisher Scientific) for 1 h at room temperature. To detect VE-cadherin and LC3 localization, specific antibodies against VE-cadherin (Beckman Coulter, Brea CA), and LC3 (Genetex) (1:200 dilutions in PBS) were incubated with cells overnight at 4°C. After three washes with tris-buffered saline and Tween 20, the cells were treated with an Alexa 488-conjugated goat anti-mouse IgG monoclonal antibody (Invitrogen, Carlsbad, CA) (1:500 dilution) and Alexa 594-conjugated goat anti-rabbit pAb (Invitrogen) (1:1,000 dilution) for 1 h, followed by three washes with tris-buffered saline-Tween 20. Images were obtained using a confocal microscope (Olympus FluoView FV1000, Melville, NY).
For quantifying barrier/marginal VE-cadherin, 3 μm across the cell border was defined as barrier/marginal area. And the remaining area within a cell was defined as cytosolic/perinuclear area. 50 cells in each condition were quantified by using Image J software.
The data are expressed as the mean ± standard error of the mean (SEM) from more than three independent experiments. One-way ANOVA and Bonferroni's multiple comparison test as post-test, two-way ANOVA or Student’s t-test was used to analyze the significance of the difference between the test and the control groups by GraphPad Prism 5 software. P values < 0.05 were considered statistically significant.
|
10.1371/journal.pbio.2004328 | The Plasmodium falciparum transcriptome in severe malaria reveals altered expression of genes involved in important processes including surface antigen–encoding var genes | Within the human host, the malaria parasite Plasmodium falciparum is exposed to multiple selection pressures. The host environment changes dramatically in severe malaria, but the extent to which the parasite responds to—or is selected by—this environment remains unclear. From previous studies, the parasites that cause severe malaria appear to increase expression of a restricted but poorly defined subset of the PfEMP1 variant, surface antigens. PfEMP1s are major targets of protective immunity. Here, we used RNA sequencing (RNAseq) to analyse gene expression in 44 parasite isolates that caused severe and uncomplicated malaria in Papuan patients. The transcriptomes of 19 parasite isolates associated with severe malaria indicated that these parasites had decreased glycolysis without activation of compensatory pathways; altered chromatin structure and probably transcriptional regulation through decreased histone methylation; reduced surface expression of PfEMP1; and down-regulated expression of multiple chaperone proteins. Our RNAseq also identified novel associations between disease severity and PfEMP1 transcripts, domains, and smaller sequence segments and also confirmed all previously reported associations between expressed PfEMP1 sequences and severe disease. These findings will inform efforts to identify vaccine targets for severe malaria and also indicate how parasites adapt to—or are selected by—the host environment in severe malaria.
| Infection by Plasmodium falciparum—the parasite responsible for malaria in humans—can result in a severe disease that can be fatal or in an uncomplicated disease that can be resolved by the host immune system. However, whether the parasites causing severe disease differ from those causing uncomplicated disease is unknown. Several strands of evidence have suggested that parasites causing severe disease may express a restricted set of the Plasmodium falciparum Erythrocyte Membrane Protein 1 (PfEMP1) proteins. PfEMP1 proteins are expressed on the surface of the infected red blood cells and elicit protective immunity. We compared the transcriptomes of parasites causing severe and uncomplicated malaria to determine whether these parasites differed in the genes they expressed. We found that the parasites causing severe malaria had altered expression of genes involved in basic metabolism, nuclear processes, and surface expression of PfEMP1. The parasites causing severe malaria had up-regulated expression of a set of PfEMP1 proteins. Some of these PfEMP1s had been previously implicated in severe malaria, lending support to our data. Multiple associations identified between severe malaria and expressed PfEMP1 sequences were novel. These novel, severe disease–associated PfEMP1 sequences could be useful for informing design of vaccines targeting severe malaria disease.
| P. falciparum is the leading cause of fatal malaria and is responsible for the death of over 400,000 people annually, primarily in sub-Saharan Africa [1]. However, severe disease also occurs in Southeast Asia, and Papua is the Indonesian province with the highest prevalence of malaria [1]. Severe malaria due to P. falciparum can manifest as multiple, diverse clinical syndromes [2], but a critical common feature is the sequestration of erythrocytes infected with mature parasites in the microvasculature (reviewed in [3]).
Comparative genome-wide analyses of parasite isolates that cause severe and uncomplicated malaria can be used to identify genes associated with parasite virulence and pathology. This knowledge could inform therapy and the design of vaccines targeting severe disease. Although previous microarray studies of ring-stage–and/or ex vivo–cultivated mature parasites [4,5] showed no differences between parasites causing severe and uncomplicated malaria, transcriptomic differences indicative of fundamental metabolic variations between clinical isolates have been reported. These differences were apparent in clinical isolates segregated by transcriptional profile alone [6] or by direct [7] or surrogate measures of parasitemia [8]. Limitations of these studies included the need to cultivate the isolates prior to analysis, the absence of clinical severity phenotype [8], and the inability to directly compare severe and uncomplicated malaria in the same study population [7]. In the current study, we used massively parallel sequencing technology to undertake comparative analysis of transcriptomes from parasites associated with uncomplicated and severe malaria in the same population. We found a unique parasite transcriptional profile that was associated with severe malaria. While elements of this profile were congruent with reported profiles of clinical isolates [6,8], these have not been previously linked with severe malaria phenotype. Genes deregulated in severe malaria were involved in pathways including central carbon metabolism, folate biosynthesis, histone methylation, chaperone function, and surface expression of P. falciparum Erythrocyte Membrane Protein 1 (PfEMP1).
PfEMP1 is the immunodominant, variant surface antigen of P. falciparum [9]. PfEMP1 is expressed on the surface of the infected erythrocyte (IE), where it mediates adhesion to diverse host receptors. PfEMP1 binding to receptors on endothelium leads to the pathogenic sequestration of IEs in the microvasculature (reviewed in [10]). The resulting obstruction is probably exacerbated by IEs binding receptors on uninfected erythrocytes to form ‘rosettes’ [11]. By switching between single, expressed PfEMP1 variants, the parasite can change receptor specificity and also avoid the acquired immune response, leading to chronic and recrudescent infections. A parasite’s genome contains approximately 60 var gene copies that code for PfEMP1 [12], and immune pressure has driven evolution of extreme diversity in PfEMP1s such that there is very little overlap in var repertoires [13–16].
Even with such large sequence diversity, var genes can be classified into three broad groups based on their upstream sequence (UPS; A, B, and C) [12,17]. Group A var genes appear to have diverged from groups B and C in their binding properties [18]. Expression of group A and B var genes has been associated with clinical malaria in Papua New Guinea [19,20], severe malaria in Africa [21], and cerebral malaria in Africa [22–24]. The PfEMP1 ectodomain contains multiple, semiconserved Duffy binding-like (DBL) domains and cysteine-rich interdomain regions (CIDRs) that mediate adhesion to host receptors [25]. These domains have been classified into major types—DBLɑ, β, γ, δ, ε, ζ, and x and CIDRα, β, γ, and δ [14,26,27]—and into 147 further subtypes, e.g., CIDRα1.1 [14]. Multiple domain cassettes (DCs) containing conserved, sequential arrangements of 2 or more domain subtypes have also been identified [14].
A conserved group of variant surface antigens that are presumably a subset of PfEMP1s appear to be expressed by parasites causing severe disease. These antigens are encountered early in life and are recognised more widely by sera from semi-immune children than antigens expressed by parasites causing uncomplicated disease [28,29]. The expression of a conserved subset of PfEMP1s by parasites that cause severe malaria probably explains why immunity to severe malaria is acquired more rapidly than immunity to uncomplicated malaria [30,31]. The conserved PfEMP1 VAR2CSA is expressed by parasites causing malaria in pregnancy, but associations with entire PfEMP1s have not been detected for other severe malaria disease syndromes.
However, at a finer resolution than whole PfEMP1s, some of the PfEMP1 domains that bind specific host receptors and/or are expressed in severe malaria have been identified. CIDRα1 binds endothelial cell protein C receptor (EPCR) [32], and its expression has been linked to severe malaria in children and adults [33–35], whilst rosetting is associated with severe malaria and expression of the DBLα1-CIDRβ/γ/δ head structure [11,36,37]. DBLβ5 from group B var genes and specific motifs in DBLβ1 and DBLβ3 from group A var genes bind intercellular adhesion molecule 1 (ICAM1) [38–42], and cerebral malaria has been associated with ICAM1 binding [43,44] and expression of group A carrying tandem CIDRα1-DBLβ1/3 domains [42]. DBLβ12 binds the host receptor gC1qR, and its expression is also associated with severe malaria [45]. Elevated expression of a number of DCs has also been associated with severe malaria; these included DC8 (DBLα2-CIDRα1.1-DBLβ12-DBLγ4/6) [34,45–49], DC13 (DBLα1.7-CIDRα1.4) [32,49], DC4 (DBLα1.4-CIDRα1.6-DBLβ3) [39], DC5 (DBLγ12-DBLδ5-CIDRβ3/4) [48], and DC6 (DBLγ14-DBLζ5-DBLe4) [34].
Due to the immense diversity seen in var gene domains, attempts have been made to investigate them by concentrating on more conserved sequence or homology blocks [14,50]. Few studies have attempted to link these conserved blocks with disease severity, although one study found an association between homology blocks 219 and 486 and rosetting, whereas homology block 204 was associated with impaired consciousness [51].
All of the previously reported associations between severe disease and var gene expression relied on PCR using primers derived from var sequences of laboratory isolates. In contrast, RNA sequencing (RNAseq) of clinical samples can be used to assemble all expressed var sequences, regardless of their homology to the var genes of sequenced laboratory isolates. In the current study, innovative bioinformatic approaches were used to identify multiple novel associations between severe disease and differential expression of var gene sequences at the multi-, single-, and sub-domain levels. Furthermore, we recapitulated all previously described associations between expressed var gene sequences and severe malaria. These novel, severe malaria–associated var sequences have relevance to efforts to design vaccines targeting severe disease.
Parasites were isolated from the venous blood of 23 patients with severe malaria and 21 patients with uncomplicated malaria (Table 1). Patients with severe malaria tended to be older than those with uncomplicated malaria, but there were no significant differences in P. falciparum density, haemoglobin (Hb) concentration, or gender. Among patients with severe malaria, 19 had presented with a single diagnostic criterion [2], including 4 with cerebral malaria, 3 with jaundice, 8 with hyperparasitaemia, 3 with prostration, and 1 with acute renal failure. Four patients had 2 or more manifestations of severe malaria: 1 patient with jaundice and acute renal failure, 1 with acute renal failure and acute respiratory distress syndrome, 1 with jaundice and hyperparasitaemia, and 1 with hyperparasitaemia and prostration. The parasite biomass marker Histidine Rich Protein 2 (HRP2) was present at higher concentrations in the plasma of patients with severe malaria than those with uncomplicated malaria (p = 0.02). None of the patients had severe malarial anaemia (defined as Hb < 5 g/dL in children <12 years old; Hb < 7 g/dL in adults; Table 1) [2]. These findings suggest that severe P. falciparum malaria in these patients was associated with sequestration rather than anaemia due to repeat infections [52].
RNA quality was assessed using the BioRad Experion system (Fig A in S1 Fig). The median RNA Quality Index (RQI) value was 7.75, and the interquartile range (IQR) was 7.175 to 8.55. Transcriptome libraries were constructed for 44 patient samples (Arrayexpress accession: E-MTAB-5860). Library sizes ranged from 17,054 to 247,859,790 sequence reads (Fig B in S1 Fig). The libraries were aligned to the Homo sapiens (GRCh38), P. vivax (PlasmoDB-11.1 Sal1), and P. falciparum (PlasmoDB-11.1 3D7) reference genomes, and the proportion of P. falciparum in the libraries ranged from 0.11% to 88.44% (S1 Table). To identify significant features distinguishing severe and uncomplicated malaria transcriptomes, the transcriptome libraries were subjected to a series of sequence and expression analyses (Fig C in S1 Fig).
A pipeline for the de novo assembly of var genes from RNAseq data was developed and verified using a P. falciparum ItG clone (ItG is the parent line of the It4 sequenced clone) for which the var repertoire is known. Expression profiles of the assembled transcripts were compared with those obtained by quantitative PCR (qPCR) and were found to correlate significantly (Pearson correlation coefficient R = 0.88) (Fig 1A). The pipeline used the SoapDeNovo-Trans/Cap3 method of [53], which is robust to chimeric assemblies and minimises redundant transcripts. Non-var P. falciparum, P. vivax, and H. sapiens reads were filtered out prior to assembly.
As proof of concept, the pipeline correctly assembled an ItG subclone E8B that expressed predominantly the IT4var04 var gene. Additionally, the ItG subclone CS2—with a recombination event between IT4var04 and IT4var08 var genes [54]—was correctly assembled (Figs A and B, respectively, in S2 Fig). Alternative approaches were investigated (S2 Table), with the SoapDeNovo-Trans/Cap3 pipeline chosen because it assembled the known samples correctly, was sensitive to low-expressed transcripts, and produced minimal redundancy. The pipeline is available at https://github.com/PapenfussLab/assemble_var.
The assembly pipeline was run separately for each of the 44 patient samples in addition to a pooled sample assembly where all the reads from each patient sample were combined (European Nucleotide Archive [ENA] accession: PRJEB20632). S3 Table indicates the number of assembled transcripts constructed for each sample along with the major N50 and maximum-length values after discarding transcripts shorter than 500 nt in length. For the remainder of this paper, we refer to these 2 assemblies as the separate and combined assemblies, respectively. The assembled var genes were analysed at the transcript, domain, and segment or homology block level (Fig 1B). Three of the severe malaria samples had a low percentage of reads mapping to P. falciparum: SFC025, SFD001 (both cerebral malaria), and SFM009 (hyperparasitemia) (Fig B in S1 Fig, S1 Table). These samples were used for var gene assemblies and sequence clustering but were omitted from the differential gene expression analysis, both for var and non-var genes.
Two patients (SFU2 and SFU3) were drug treated at admission prior to blood collection, and 4 patients (SFC023, SFM007, IFM012, IFM021) were treated with antimalarials for previous Plasmodium infections more than 2 weeks but less than 4 weeks prior to admission. These patients were omitted from the differential expression analyses of the total transcriptomes. Significant differences were identified in the expression of genes between severe and uncomplicated cases of malaria. After accounting for library size, parasite life cycle, and other unwanted sources of variation, 358 genes were found to be differentially expressed after multiple testing correction (p = 0.1, limma/Voom pipeline [55,56]). A full list of genes with relevant log fold changes and p-values can be found in S1 Data.
A mixture model was used to account for parasite life cycle. A constrained linear model was fit using published data [57] to estimate the proportion of ring, early trophozoite, late trophozoite, schizont, and gametocyte stages present in each sample (Fig 2A, S2 Data). This approach returns similar results to the maximum likelihood approach of [5] and is comparable to the approach of [58], which focused on microarray data. The mixture model correctly identified sample SFC21 as having a higher proportion of gametocytes, a finding that was confirmed by microscopy. Trimmed mean of M values (TMM) normalisation [59] was used to account for library size, with samples SFC025, SFD001, and SFM009 excluded due to insufficient coverage.
The proportion of parasites present at the ring stage—as well as 3 factors of unwanted variation estimated using the R package ruv [60]—were used to account for life cycle and other unwanted batch effects. Differential expression testing was conducted using the limma/Voom pipeline [55,56]. The impact of including these covariates in the model is evident in the Principal Component Analysis (PCA) plots (Fig 2B and 2C, S2 Data). The choice of covariates strikes a balance between testing power and accounting for unwanted variation. The PCA plots indicate that the outlying SFC21 sample has been accounted for. Furthermore, the separation between the severe and uncomplicated cases shows that, after accounting for variations due to parasite life cycle, significant differences exist between the phenotypes.
The severe malaria transcriptomes could be separated by profile of differentially expressed genes into 2 principal clusters—S1and S2 (S3 Fig), which was consistent with previous reports of clinical isolates and severe malaria [6,7]. This suggests that severe malaria can be caused by parasites in different physiological states. A previous report also found that median parasitemias differed between severe malaria clusters [7]; the median parasitemias in the clusters in this study were also suggestive of a difference (p = 0.0755 Mann Whitney test; parasites/μl median, IQR, S1: 43,040; 5,880; 259,378; S2: 786,316; 212,708; 1,095,789). However, the severe malaria transcriptomes did not cluster by clinical syndrome (Fisher’s exact test, all p > 0.12).
The 358 genes differentially expressed in severe malaria were from diverse functional pathways and revealed a distinct severe malaria parasite transcriptome (S3 Fig, S1 Data). Biological pathways annotated as Gene Ontology (GO) biological process terms or Kyoto Encyclopedia of Genes and Genomes (KEGG) were ranked using a hypergeometric test, and those with p < 0.1 are considered. Terms relating to glycolysis, histone methylation, folate metabolism, and protein folding ranked highly and included genes down-regulated in severe malaria, whilst pathways relating to the tricarboxylic acid (TCA) cycle, nucleoside diphosphate (pyrimidine) metabolism, and regulation of guanosine triphosphatase (GTPase) activity included genes up-regulated in severe malaria (Fig 3A, S3 Data). In addition, genes involved in PfEMP1 transport and a gene involved in regulation of var genes were down-regulated in severe malaria. This suggested that var gene expression was modulated but PfEMP1 surface presentation was reduced. Several GO and KEGG pathways that ranked highly included deregulated genes that were not functionally related in a coherent manner and will not be discussed further.
Parasites isolated from patients with severe malaria had significantly down-regulated genes included in the KEGG pathway ‘Glycolysis/gluconeogenesis’. Significant decreases were observed in transcript levels of 3 glycolytic enzymes (0.52- to 0.62-fold the levels in parasites that caused uncomplicated malaria) (all adjusted p < 0.1) (Fig 3B, Fig 3A, S1 Data, S3 Data). These were aldolase, glyceraldehyde 3-phosphate dehyrogenase, and mitochondrial dihydrolipoyl dehydrogenase (LPD1) that converts glycolytic pyruvate to acetyl coenzyme A (acetyl-CoA). Expression of most other enzymes in this pathway trended down (with 3 adjusted p ≤ 0.12) (Fig 3B). The lactate transporter (also known as the formate nitrite transporter [61]) was also down-regulated in parasites causing severe malaria (0.33-fold p = 0.009). Together, these data suggest that parasites associated with severe malaria have decreased transcription of genes involved in aerobic glycolysis.
Our results confirm and extend previous analyses on the transcriptional regulation of enzymes involved in central carbon metabolism in clinical isolates [8]. In particular, Daily et al. [6] described a cluster of P. falciparum clinical isolates that exhibited a distinct, starvation-like response, characterised by decreased transcription of genes involved in glycolysis and increased transcription of genes encoding enzymes involved in the TCA cycle, which is similar to the transcriptional signature we observed from parasites linked to severe malaria samples (S4 Table). In contrast, parasites isolated from patients with severe malaria in a subsequent study had a transcriptional profile that was more consistent with a glycolytic phenotype [7]. However, neither study directly compared transcriptional profiles from parasites causing severe and uncomplicated malaria [6,7].
To determine whether nutrient availability was contributing to the reduced expression of genes encoding glycolysis enzymes by the parasites causing severe malaria, metabolite levels in plasma samples from the severe and uncomplicated malaria patients were analysed by untargeted liquid chromatography–mass spectrometry (LC-MS) analysis. Thirty-five metabolite peaks differed significantly between the plasma of patients with severe and uncomplicated malaria (p < 0.01, Benjamini-corrected; S4 Data). These included 7 metabolites—provisionally identified as lipids—and citrulline (confirmed with an authentic standard), which was depleted in the patients with severe malaria. Citrulline recycling to arginine contributes significantly to nitric oxide (NO) synthase substrate availability and thereby NO bioavailability in malaria. Low citrulline is therefore likely to contribute to the hypoargininenia, impaired NO bioavailability and endothelial dysfunction found in both adults and children with severe malaria, in both Melanesian [62] and African [63–65] populations. The plasma levels of glucose and lactate were similar in patients with uncomplicated and severe malaria (Fig 3C), suggesting that the down-regulation of parasite glycolysis in the patients with severe malaria is not a direct response to reduced availability of blood glucose. Blood glucose concentrations were also similar in individuals harboring parasites with or without the proposed starvation transcriptional pattern described by Daily et al. [6].
Glucose-starved yeast [66] and clinical P. falciparum isolates with the proposed starvation response-like transcriptome both increased transcription of TCA cycle enzymes [6]. Similarly, in severe malaria, the GO category ‘tricarboxylic acid cycle’ included 2 genes up-regulated more than 2-fold in severe malaria (both adjusted p < 0.097): the Fe2S subunit of the mitochondrial TCA cycle enzyme succinate dehydrogenase and the putative succinyl CoA synthetase β subunit. Aconitase was also up-regulated more than 1.6-fold (adjusted p = 0.156); however, no significant differences in expression of the other TCA cycle enzymes were observed (p-values > 0.2). We have previously shown that P. falciparum asexual blood stages primarily sustain TCA cycle fluxes and low-level oxidative phosphorylation by catabolizing glutamine [67] up-regulating the TCA cycle. However, key enzymes in glutamine utilisation were either down-regulated (glutamate dehydrogenase down 0.4-fold, p = 0.012) or unchanged (NADP-specific glutamate dehydrogenase, aspartate transaminase, glutamate synthase, malate dehydrogenase, phosphoenolpyruvate carboxylase, and branched chain ketoacid dehydrogenase complex [BCKDH] subunits E1β and E2) in isolates from patients with severe malaria, indicating that these parasites are unlikely to exhibit a significant switch to mitochondrial respiration. Overall, these data suggest that parasites associated with severe malaria were not compensating for decreased glycolysis by increasing oxidation of pyruvate in the TCA cycle and may be metabolically less active than parasite isolates associated with uncomplicated malaria.
The GO term ‘methylation’ and a number of subsidiary GO terms relating to histone methylation included genes down-regulated in parasites causing severe malaria. The down-regulated genes included a putative histone S-adenosyl methyltransferase and 2 of the 10 SET-domain lysine methyl transferases found in P. falciparum (SET2 and PfSET7). Levels of SET3 and a putative protein arginine N-methyltransferase 1 (PfPRMT1) were also suggestive of down-regulation (both adjusted p < 0.11, <0.73-fold). PfSETvs or SET2 plays an important role in regulating expression of var genes (see below). PfSET3 and PfSET7 appear essential for blood-stage growth [68], and PfSET7 can methylate H3 but is localised to the cytoplasm in asexual blood stages [69]. PfPRMT1 probably methylates cytoplasmic and nuclear proteins including methylations of histone 4 that are involved in gene activation [70]. These data suggest that histone methylation pathways involved in gene regulation were down-regulated in severe malaria. Genes involved in chromatin modification were also deregulated by parasites previously reported to have caused high parasitaemia infections [8]. Severe malaria is known to elicit gametocytogenesis, and heterochromatin structure dependent on histone methylation is known to repress the gametocytogenesis transcription factor ApiAP2G [71], so down-regulation of histone methylation would be consistent with induction of gametocytogenesis in severe malaria.
The KEGG term ‘folate biosynthesis’ and a number of related GO terms included 2 genes down-regulated in severe malaria: dihydropteroate synthetase (DHPS) and guanosine triphosphate (GTP) cyclohydrolase. GTP cyclohydrolase is the first and rate-limiting enzyme in the folate pathway and therefore is essential for DNA and protein synthesis. Aspartate carbamoyltransferase (ATCase) was also down-regulated; it is the second enzyme in the pyrimidine biosynthetic pathway; and whether it is rate limiting in P. falciparum is unknown, but it is so in bacteria [72]. These changes suggest that nucleoside biosynthesis may be decreased in severe malaria, consistent with lower growth rate and/or reduced metabolism. A number of GO pathways related to nucleoside diphosphate and pyrimidine metabolism included 2 genes up-regulated in severe malaria: nucleoside diphosphate kinase (NDK) and the putative small subunit of ribonucleotide reductase. These 2 genes are central to ribonucleoside triphosphate (NTP) and deoxyribonucleoside triphosphate (dNTP) synthesis. Although up-regulation of these enzymes suggests increased DNA synthesis, the down-regulation of key enzymes in the folate and pyrimidine pathways instead indicates that a diminished nucleoside pool is subject to increased flux through ribonucleoside diphosphate (NDP) to dNTP metabolism.
The GO term ‘translational elongation’ included 3 down-regulated elongation factor genes suggesting decreased protein production. The GO term ‘protein folding’ included 9 down-regulated genes, including the HSP70 interacting protein (HIP), the peptidyl-prolyl cis-trans isomerase cytochrome P450 52 (CYP52)—which has in vitro chaperone activity (Marin-Menendez, 2012)—an FK506 binding protein (FKBP)-type peptidyl-prolyl isomerase, and the PfEMP1 transport–associated KAHsp40. Functionally related proteins outside this pathway were also down-regulated, including the hsp70/hsp90 organising protein (HOP), HSP70-x (see below), and the essential PfHsp110c, which is important for preventing heat-induced aggregation of the many P. falciparum Asn repeat rich proteins during fever [73]. Overall, down-regulation of these genes indicated a decreased stress response or generalised, decreased protein processing.
The GO category ‘regulation of GTPase activity’ and related GO categories included 3 GTPase-activating protein genes that were up-regulated in severe malaria, 2 of which were specific for Rab GTPases. This would be consistent with decreased Rab GTPase trafficking regulatory activity and therefore decreased vesicular transport. Two genes involved in vesicle transport were down-regulated; these were SNAP proteins, which is involved in dissociation of the Soluble NSF (N-ethylmaleimide-sensitive factor) Attachment Protein Receptor (SNARE) complex, and choline-phosphate cytidylyltransferase (CCT), which is rate limiting for synthesis of the major P. falciparum membrane phospholipid, phosphatidylcholine [74]. Four genes involved in vesicle transport were up-regulated. These included 2 proteins involved in endoplasmic reticulum (ER) to Golgi transport, the trafficking protein particle complex subunit 5 (TRAPPC5), and the SNARE protein PfGS27; the retrieval receptor for ER membrane proteins, which is required for anterograde vesicular transport; and the vacuolar protein sorting–associated protein 45 that is implicated in vesicle transport from the Golgi to endosomes or the food vacuole. Overall, the probable decreased trafficking activity of several Rabs and deregulated vesicle transport processes suggest deregulated protein trafficking in severe malaria.
Multiple genes involved in PfEMP1 biology were down-regulated in severe malaria. These included PfSETvs—which methylates lysine 36 on histone 3, is required for var gene silencing [68], and is involved in normal var switching [75]. The knob-localised KAHRP—which binds PfEMP1 and the cytoskeleton—and the lysine-rich, membrane-associated PHISTb protein (LyMP) (PF3D7_0532400) are both required for optimal binding of PfEMP1 to (some) receptors [76] and were amongst the most down-regulated genes in severe malaria. Also down-regulated were the following: the Maurers cleft proteins SBP1 and REX1, which are required for proper Maurer’s cleft organization and PfEMP1 transport to the erythrocyte surface [77–79]; Heat shock protein 70-x, which forms a complex with Hsp40 in the red blood cell cytosol and is possibly involved in PfEMP1 transport [80]; and KAHsp40, which binds PfEMP3 and KAHRP and colocalises with knob-associated proteins [81]. Therefore, we observed probable mechanistic drivers of modulated var regulation and decreased transport of PfEMP1 to the parasite surface. GO categories that were highly ranked due primarily to inclusion of deregulated 3D7 var genes were not reported because 3D7 var genes were not present in the clinical isolates.
Several parasite surface proteins with established functions unrelated to ring-stage parasites were highly up-regulated in severe malaria. The second most up-regulated gene encoded the glycosylphosphatidylinositol-anchored cysteine-rich protective antigen (CyRPA) that anchors the critical invasion protein Plasmodium falciparum reticulocyte-binding protein homolog 5 (PfRh5) to the surface of the merozoite [82]; the seventh most up-regulated gene was the merozoite surface-located 6-cysteine protein P41 [83], and the 14th most up-regulated gene was sporozoite invasion-associated protein-2 (SIAP-2). This gene is expressed at low levels in blood-stage cultures but at high levels on the surface of sporozoites, and it appears to be important for hepatocyte traversal [84]. The serpentine receptor 10 was also up-regulated. It is most closely related to receptors that transduce external stimuli in other organisms [85].
There was no difference between severe malaria and uncomplicated malaria in total var gene expression, i.e., the number of reads that mapped to de novo–assembled var genes (normalised for number of total reads that mapped to all genes; Welch 2-sample t test, p = 0.28). Differential expression analysis was conducted at the var multidomain transcript, individual domain, and segment levels because associations between var expression and severe disease have been previously detected separately at each of these resolutions. At each level, significant, differentially expressed sequences were identified. Additionally, the resulting sequence transcripts, domain classification, and segments were found to better distinguish severe and nonsevere cases of malaria than previous var gene classifications [14,39,46,48].
Fig B in S4 Fig illustrates a PCA plot of normalised read counts annotated to the transcripts from the combined sample var gene assembly. By comparing it to the all-gene PCA plot (Fig 2C), it is evident that var gene expression differentiates severe cases of malaria. The severe cases are more tightly clustered together than the nonsevere.
S5 Data lists all the separate sample assembly transcripts along with whether they were significant at the transcript, domain, or segment level. A number of transcripts had domains and segments that were significantly associated with disease severity when the transcript itself was found not to be significantly associated with severe disease. This highlights the importance of investigating the var gene sequences at multiple resolutions.
In the combined sample assembly (S6 Data), 53 transcripts were found to be differentially expressed using the default DESeq2 pipeline [86]. Of these, 17 are up-regulated in severe malaria (p < 0.05) (Fig 4A). The expression profiles of the up-regulated transcripts from the combined assembly differentiated the samples based on severity (Fig 4A, S4 Fig, panel B). Amongst the transcripts up-regulated in severe malaria, the extracellular domains most highly expressed in severe malaria were a DBLζ4 (284084_soapGraphK61), a DBLε3 (274611_soapGraphK61), and a DBLε12 (Contig1811) (Fig 4B). The up-regulated transcripts included a transcript that contained DBLβ5-DBLγ14 (298068_soapGraphK61); DBLγ14 has only been found in DC6 [14], and its expression was recently associated with severe disease [34]. In 7 P. falciparum genomes [14], the tandem combination DBLβ5-DBLγ14 was detected only in the 3D7 gene PFL0020w, which is expressed by 3D7 parasites selected for adhesion to ICAM1 [87]. Another of the up-regulated transcripts contained DBLγ18-DBLε14 (Contig3067); this tandem domain arrangement was only detected twice in the 7 sequenced genomes but was not part of any DCs. The remaining transcripts were either single domains or common tandem domain arrangements. A transcript incorporating DC5 (DBLδ5-CIDRβ3-DBLβ7-[DBLγ4]) (contig12688) was also up-regulated in severe malaria (p = 0.0537). DC5 was up-regulated in severe malaria in Africa [48] and expressed in a cerebral malaria case in Papua New Guinea [88].
Corset [89] groups transcripts together based on the number of reads that multi-map between them whilst ensuring transcripts are not combined if they have significantly different expression profiles. We used Corset to detect transcripts associated with severe disease in the separate sample assemblies. Associations between severe disease and var transcripts can be inferred with greater confidence if identified using multiple approaches. Corset identified 82 differentially expressed clusters in total, of which 5 were clearly up-regulated in severe disease (Fig 5A, S7 Data). These clusters included overlapping, multidomain contigs that spanned DC4 (N-terminal sequence A [NTSA]-DBLα1.2/1.5/1.4-CIDRα1.6-DBLβ3-[DBLγ11-DBLδ1-CIDRβ1/2]) (cluster-10.1182) and DC11 (CIDRβ4-DBLγ7-DBLε11-DBLζ2-DBLε11-DBLε3) (cluster-10.1147). The contigs spanning DC4 were the most abundantly expressed of the clustered contigs up-regulated in severe malaria (Fig 5B). DC4 expression has previously been associated with severe malaria [39], and the DC4 cluster included 2 DC4 transcripts from the cerebral malaria sample SFC15. For each transcript in the separate assembly, its closest basic local alignment search tool (BLAST) [90] hit in the combined assembly was identified. Of the 5 Corset clusters up-regulated in severe malaria, 2 included transcripts with their closest BLAST hit in the 17 up-regulated transcripts from the combined assembly. These were the DC4 cluster—which was homologous to the DBLδ1-CIDRβ1 combined assembly transcript 297752_soapGraphK61—and cluster-10.839 (N-terminal sequence B [NTSB]-DBLα0.5-CIDRα2.6/3.4-DBLβ5/8/13-DBLδ1-CIDRβ5)—which was homologous to the combined assembly transcript 284128_soapGraphK61 (CIDRα2.6-DBLβ8) (Fig 6). The 2 remaining clusters contained transcripts spanning NTSB-DBLα0.1/0.4-CIDRα3.1/4-DBLδ1-CIDRβ1/7 (cluster-10.583) and NTSB-DBLα0.5-CIDRα2.2/2.3/2.6/2.8-DBLδ1-CIDRβ1 (cluster-10.548). These 2 clusters and the DC4 and DC11 clusters were all homologous to additional transcripts that were up-regulated in the combined assembly at an adjusted p-value of no more than 0.153 (Fig 6). The elevated p-values in the combined assembly analysis can be explained by the heavier multiple testing penalty due to the larger number of transcripts.
For simplicity, we restricted our analysis at the type level to distinguishing between UPS types A and B/C combined. Expression of the conserved NTS segments allows for these 2 groups to be identified. HMMER3 [91] was used to align the profile hidden Markov models of the domains defined in [14] to the transcripts built from the separate assemblies. Reads that aligned to the regions annotated as either NTSA or NTSB were then used as counts for the respective var types. NTSA was more highly expressed in the severe malaria samples than in the uncomplicated malaria samples (Fig A in S4 Fig). This is consistent with previous studies [19–23,34,92].
The domain models of [14] were first investigated using the same approach as the type-level analysis. Of the 149 domain classifications identified in the transcripts, 16 were found to be significantly up-regulated in severe malaria using the default pipeline of DESeq2 [86] (S8 Data, Fig 7A). Some previously described associations between expressed var sequences and severe malaria were confirmed, adding confidence to our analysis. These included up-regulation in severe malaria of CIDRα1.1 and CIDRα1.6, which bind EPCR [35] and are often found in DC8 and DC4, respectively [32,34,45,46,48,49]; DBLα2, which is restricted to DC8; and DBLβ12, which binds gC1qR [45] and is invariably found in DC8. DBLβ3 was also up-regulated; it can bind ICAM1 and is found in—but is not restricted to—severe malaria–associated DC4 [39] and DC8. The domain subtypes DBLβ3, CIDRα1.6, DBLα1.2, DBLα1.5, and DBLγ11 that were all up-regulated in this analysis were also part of the up-regulated DC4 transcript in the Corset analysis (Fig 6, Fig 5A, S5 Data, S7 Data). NTSA that is restricted to group A var genes was also up-regulated. Despite these clear differences, many of the domains were still expressed in a large number of the uncomplicated samples, e.g., DBLβ3 was far more abundantly expressed than CIDRα1.6 and DBLα1.5, but the latter 2 domains were more clearly differentially expressed in severe malaria (Fig 7A, Fig 8B). Comparing the domain-level PCA plot in Fig 7B with the transcript-level PCA plot in Fig B in S4 Fig showed that the differentiation between severe and uncomplicated malaria samples was less evident at the domain level and suggested that a more accurate classification could be made.
A novel hierarchical approach was developed to identify domains that are associated with severe malaria. Domain regions were first defined using HMMER3 domain models based on the domain sequences identified in [14]. The identified domain regions were then hierarchically clustered using USEARCH [93] as described in the Materials and methods section at sequence identity levels 50, 55, 60, 65, 70, 75, 80, 85, 90, 95, and 97. The counts for each cluster were aggregated up the hierarchical tree, and differential expression was tested at each node. Benjamini-Yekutieli [94] multiple testing correction was performed before the most significant node was successively chosen and added to the list of significant clusters. The children and parents of each node were removed from the list of potential clusters before the next node was chosen in the tree. A more detailed description is given in the Materials and methods section.
This approach attempts to identify the point, or node, in the hierarchical tree that best distinguishes domains associated with severe disease from those that are not. By looking at different levels of the tree, we are able to identify potentially important domain groups that would otherwise remain elusive. Additionally, by grouping domains at various identity levels, we are increasing the sensitivity to domain groups with higher sequence variation such as the DBLδ domain class.
Fig 7C illustrates the advantage of this approach by focusing on a tree related to the DBLε3 domain of [14]. At 70% identity, no clusters are significantly associated with severe disease. However, at 65%, 1 cluster (DBLε3.s.1) becomes significant. This difference is then lost at 50% identity. A PCA plot of the clusters at 50% identity is shown in Fig 7D. It provides a much clearer grouping of the severe samples than the previous domain definitions. Similar groupings are seen at all levels of the tree (S5 Fig).
The tree-building approach identified 70 differentially expressed domain clusters, of which 15 are up-regulated in severe malaria (S9 Data). To investigate possible associations between these 15 clusters, they were grouped based on their expression across the samples using average linkage hierarchical clustering. The dendrogram was then cut using the dynamic height-cutting algorithm of [95]. This identified 4 groups, labelled in the heatmap diagram of Fig 8A. Compared to Fig 7A, these domains provide a clearer distinction between parasites causing severe and uncomplicated malaria. However, the expression levels of these domain clusters were, in general, lower than the levels of the domains identified using the HMMER3 approach of Rask et al. (Fig 8B). This is consistent with the higher level of sequence identity within a domain cluster identified by the hierarchical tree approach. Therefore, fewer sequences per isolate were captured by each domain cluster than were assigned to each domain by the HMMER3 domain models, and the more diverse sequences captured by a HMMER3 model included many that were highly expressed in uncomplicated malaria (Fig 8B).
Examining Fig 8A in more detail reiterated elements of the original domain analysis (Fig 7A and Fig 6) and confirmed previous associations between expressed PfEMP1 domains and severe malaria. Two domain clusters similar to DBLβ3 were up-regulated in severe disease. Domain transcripts from the DBLβ3 clusters were aligned with published DBLβ3 sequences [39] using MUSCLE [96] and clustered using FastTree [97]. Notably, the DBLβ3 domain 4 from DC4 in the var gene PFD1235w [39] clustered tightly with 2 of the 4 transcripts from the DBLβ3s.2 cluster we identified but none of the 4 transcripts from the DBLβ3s.1 cluster. The turquoise group of domain clusters includes a cluster with sequences similar to DBLβ12 domains. These DBLβ12 sequences were compared with domains from DC8 but formed a separate cluster to the 2 clusters formed by the DBLβ12 domains found in DC8 [14] (S2 Text). The turquoise group of domain clusters in Fig 8A also contained the domain cluster DBLδ1.s.4, which did cluster closely with 2 DC8 DBLδ1 domains, although DC8 DBLδ1 domains cannot be differentiated from other DBLδ1 domains. Nonetheless, the clustering by expression profile of DBLδ1.s.4 and DBLβ12.s.1 (turquoise group in Fig 8A) suggests that these domains could be part of a single DC8-like var in these patients. Unlike all the other analyses employed in this study, the hierarchical approach did not associate up-regulated CIDRα1 sequences with severe malaria. This is presumably a consequence of the low conservation of the CIDRα1 sequences [35] that would not be grouped at the minimum 50% identity threshold employed in the hierarchical analysis. All alignments and trees made with published sequences described above are further described in S2 Text and are available in the Github repository https://github.com/gtonkinhill/falciparum_transcriptome_manuscript.
Fig 8A also identifies domain clusters that have not been previously associated with severe disease. These included DBLγ3.s.1 and DBLζ4.s.1 from the pink group of domain clusters in Fig 8A. Four of the 5 DBLγ3 and 11 of the 12 DBLζ4 sequences previously described were found in tandem in DC9 [14]. These domain clusters were among the most abundant up-regulated in severe malaria (Fig 8B) and were also up-regulated in the combined transcript assembly (Fig 6), although not as a single transcript, so there is no direct evidence that DC9 itself is associated with severe malaria. DBLε3 was also amongst the most abundant domain clusters up-regulated in severe malaria in this analysis (Fig 8B). DBLε3 was also up-regulated in every analysis we performed (Fig 6) and was part of DC11 in the separate assembly transcript Corset analysis, although the DC11 transcripts were all different from the up-regulated DBLε3 transcripts from the individual domain analysis. A cluster similar to DBLεpam5 from the pregnancy malaria-associated gene var2csa was also up-regulated.
DBLδ1-CIDRβ1/2/3/5/7 are common arrangements and were up-regulated in severe malaria in both the combined assembly and the Corset analysis of separately assembled transcripts. These domain subclasses are highly variable and thus difficult to distinguish using the previous classifications of [14]. However, the hierarchical approach identified the CIDRβ1.s.1 domain cluster that includes a number of identical domain sequences from different isolates. This finding differs from the high variability noted in previous studies [14]. A single CIDRβ1 from a gene containing DC8 formed a phylogenetic cluster with the conserved severe malaria–associated CIDRβ1 sequence, but CIDRβ1 from other DC8 genes did not (S2 Text). The DBLδ1.s.1 and DBLδ1.s.2 domain clusters did not form phylogenetic clusters with DBLδ1 sequences from DC8 genes (S2 Text) but did cluster by expression profile with the uniquely conserved CIDRβ1.s.1 (purple group Fig 8A) and with the non–DC4-like DBLβ3.s.1, suggesting the existence of var genes that carry a unique pathogenic arrangement of domains, including a highly conserved CIDRβ1 sequence. These 4 domains were not the most abundantly expressed in severe malaria but did discriminate strongly between severe and uncomplicated malaria, suggesting that they were strongly associated with severe malaria in a subset of cases (Fig 8A, Fig 8B). Images of the hierarchical trees that make up each of these newly identified domain clusters are provided in the Github repository along with their respective multiple sequence alignments.
The sequence of domains associated with severe malaria by RNAseq was confirmed by Sanger sequencing 34 sequences that were cloned from genomic DNA (gDNA) of patient samples. These included domains identified by hierarchical analysis (13 domains up-regulated and 11 down-regulated in severe malaria), HMMER3 analysis (9 domains up-regulated in severe malaria), or corset analysis (the CIDRα2.6-DBLβ8 tandem arrangement up-regulated in severe malaria) (S10 Data). Every one of these cloned sequences was 100% identical to the cognate sequence assembled from RNAseq. RNAseq quantitation of domains in severe malaria was corroborated by quantitative reverse transcription PCR (Q-RT-PCR) of 10 up-regulated and 3 down-regulated domains. Insufficient RNA was available to test all patient samples, so a subset of patients was tested that included several patients for each domain that had high levels of RNAseq expression of that domain. Q-RT-PCR data correlated with RNAseq Reads Per Kilobase of transcript per Million mapped reads (RPKM) for 12 of the 13 domains (Spearman r all greater than 0.53, all p < 0.03) (S6 Fig). Q-RT-PCR and RNAseq of DBLε3.s.1 did not correlate because a number of uncomplicated malaria samples contained high levels of expression by Q-RT-PCR but not by RNAseq. Because Q-RT-PCR works best on small sequences (in this case a 64 bp product) and detects hybridisation rather than actual sequence, the most probable explanation for this discordance is cross-reactive amplification of non-DBLε3.s.1 by the Q-RT-PCR.
Due to the highly variable nature of var genes, it is common to focus on the most conserved segments—or blocks—of var gene sequence [14,50]. To investigate these conserved regions, we examined 628 homology blocks that were previously defined [14]. Of these, 613 were available for download from the VARDOM server. An approach was also developed comparable to that of [98] to divide the var sequences into conserved and variable regions.
The previously defined Homology blocks [14] were clustered and examined using the same approach as for the previously defined domains [14]. HMMER3 [91] profile hidden Markov models were used to annotate the separate assembly transcripts, and read counts were obtained from the aggregate of these annotations for each block. Overall, 16 homology blocks were identified as being differentially expressed (S11 Data). Ten of these (homology blocks 47, 97, 121, 126, 141, 142, 150, 183, 219, and 582) were up-regulated in severe disease. The heatmap in Fig A in S7 Fig indicates that homology blocks 219 and 582 are the most distinct in their expression profiles. Homology block 219 is located in the DBLα1 domain class found in group A var genes and has previously been associated with severe malaria and rosetting [51]. Block 582 is usually found after a DBLζ4 domain in DC9. DBLζ4 domains were found to be up-regulated in severe disease in the domain-level analysis. Homology blocks 126 and 142 are found mainly within DBLε5 but also other DBLε subtypes, whilst block 97 is found in DBLε4–8,12,14,PAM5 and DBLγ6,12,16,17 domains. Some of these DBLε subtypes were also identified in the domain-level analysis. Blocks 121 and 150 are found in CIDRα1 domains, and homology block 141 is found at the junction between CIDRα1 and DBLβ1,3,7,12 domains, whilst block 183 is found in DBLβ1,3–5,10–12, domains. As mentioned previously, DBLβ3, DBLβ12, and CIDRα1 domains are associated with DC4 and DC8, which have been associated with severe disease. Finally, homology block 47 is found within the acidic terminal sequence (ATS) of var genes. This region does not code for the extracellular part of the protein.
Although differentially expressed blocks are identified, it likely that, as in the domain analysis, a better classification can be made by making use of the novel transcripts. The homology blocks used for this analysis were defined based on conserved recombining regions in the var gene genome [14] and not on their relationship to disease severity. This may have obscured conserved regions that are related to severe disease. Furthermore, by focusing on only the most conserved regions, we are potentially ignoring informative—but more variable—regions. Finally, the homology blocks of [14] were inferred from laboratory strains and may not include conserved segments that are unique to severe disease types.
As an alternative, an approach similar to [98] was used to divide up multiple sequence alignments of the major domain classes. Domains identified using HMMER3 [99] were grouped into their major domain classes and aligned using Gismo [100]. Sequence logos of the resulting alignments were generated using skylign [101] (see S8 Fig). Gismo [100] was found to handle the large diversity in the var domains better than other aligners. The resulting alignments were then segmented into regions of high and low occupancy. If 7 or more consecutive columns within an alignment had an occupancy greater than 95%, these columns were considered a conserved region. The columns in between these conserved regions were considered variable regions. Regions of high variability are harder to align and consequently result in more gapped alignments. The results were found to be robust to the choice of the occupancy threshold as well as the choice for the number of consecutive conserved columns. This approach produces interleaved regions of higher conservation and diversity. The approach is similar to that proposed by [98]; however, we focus on both the conserved and variable regions. Each domain sequence was then split into segments based on the regions identified. We refer to these segments by their location within the domain from which they originate. For example, DBLα_block2 is the second interleaved region of the DBLα domain class. The segments were then hierarchically clustered within their respective regions and analysed for differential expression in a similar manner to that used for the domains. Due to the short nature of these segments, CD-HIT [102] was used in place of USEARCH [93] because it accounts for the terminal gaps in its definition of pairwise sequence identity. Aside from identifying segments associated with severe disease, an advantage of this approach is that the resulting segments can easily be understood in terms of their relationship to the var domains and gene sequence.
DESeq2 [86] was used to investigate the differential expression of the segments, and Benjamini-Yekutieli [94] correction was used to correct for the multiple dependent tests. Overall, 26 clusters of segments were identified as being differentially expressed, of which 21 were up-regulated in severe disease (S12 Data). Fig 9 indicates the expression levels for each segment cluster across the samples. One lies in segment 4 of the NTSA region and 3 in regions 1, 5, and 6 of the DBLα1 domain class of group A var genes. A single up-regulated cluster (170183_0.9 DBLα_block5) lies in segment 5 of DBLα0.1 from non–group A var genes.
A cluster from region 6 of the CIDRα domain class had a similar expression profile to those segment clusters from the DBLα domain class. The cluster was also mostly made up of segments from CIDRα1.4 and 1.8 domains that have been associated with severe disease [32–34,46,49]. Two DBLε segment clusters from regions 6 and 8 derived primarily from DBLε5 subtypes also had similar expression profiles to the DBLα segment clusters. Furthermore, the clusters from region 6 of the DBLε domain often appear in conjunction with homology blocks 126 and 142, suggesting they are identifying similar domains. The DBLε5 subtype has only been described in var1.
A striking difference in expression profile was observed between the moderately high levels of expression in both uncomplicated and severe malaria samples of the grouping of clusters from NTSA, DBLα1, CIDRα1, and the DBLε region 6 and 8 compared to the markedly lower levels of expression in the uncomplicated samples of all other segment clusters that were up-regulated in the severe samples (Fig 9). This is consistent with the presence of NTSA and DBLα1 on all group A var genes and therefore their expression in both uncomplicated and severe malaria. Similarly, var1 is ubiquitously expressed by laboratory isolates and is not subject to the same program of gene regulation as other var genes, and so might be expected to be expressed in both uncomplicated and severe malaria. However, these data also suggest that expression of CIDRα1 that can bind EPCR does not distinguish between severe and uncomplicated malaria as well as other var regions, which presumably mediate adhesion to other receptors.
The relationships between the segments, homology blocks, domains, and transcripts are illustrated in Fig 6. The Clusters from region 2 of DBLγ and regions 12 and 15 of DBLζ contained 15 transcripts, including 13 from the domain clusters DBLγ3.s.1 and DBLζ4.s.1, suggesting that we have identified similar disease-associated sequences at both the domain and segment level. Five of the transcripts included region 2 of DBLγ and both or either of regions 12 and 15 of DBLζ and 5 of the transcripts included regions 4 of CIDRγ and/or region 11 of the DBLδ. DBLδ-CIDRγ tandem domains invariably precede the DBLγ-DBLζ tandem domains of DC9. Seven of the 11 transcripts containing regions 12 and 15 of the DBLζ domain class also contained published homology block 582 [14], indicating they may be detecting similar signals (S5 Data).
Significant clusters from regions 1 and 3 of the DBLε domain class often appear in both of the previously identified domain clusters DBLε2.s.1 and DBLε9.s.1 as well as a number of other DBLε9 domain sequences. As the segment clusters collapse, 2 previous domain clusters along with a number of other sequences; this indicates that these segments may have better captured the sequence elements associated with severe disease. Due to the high diversity of the DBLε domain class at the domain level, it is hard to accurately define which sequences are associated with severe disease, and consequently this highlights the virtue of investigating these sequences at multiple resolutions.
Eight out of 11 clustered DBLβ region 4 sequences are also part of the DBLβ12 domain class. In 4 occurrences, it appears in a transcript that includes DC8. Two of the 10 DC4 transcripts clustered by Corset also contained DBLβ region 4 sequences that were part of DBLβ3. Therefore, the up-regulated DBLβ region 4 sequence collapses 2 DBLβ subtypes that are independently associated with severe malaria and implicated in different adhesion phenotypes. Five of the 13 transcripts from the cluster containing DBLβ region 8 also contained the DC4-like homology block 141, whilst 2 of the 8 transcripts from the cluster containing region 5 of DBLβ also contain the non–DC4-like DBLβ3.s.1 domain.
Transcripts from clusters of regions 5 and 7 of the DBLγ domain class don’t appear with other segments significantly associated with severe malaria, with the exception of 2 region 7 segments that appear in transcripts that include DC6. These segments may represent signal lost in the analysis of larger sequence elements.
To investigate the utility of using the different feature levels to differentiate severe and nonsevere disease, we fit a logistic regression model with lasso regularisation. A model was generated for each level of the var gene analysis (transcript, domain, and segment) using the features found to be up-regulated in severe disease. We made use of crossvalidation to determine the optimal lambda value for the regularisation and to give an indication of how well the features distinguish severe and nonsevere disease. Overall, the segment level provided the best discrimination, with misclassification error of 9.76% and 12.20% for the homology block and segment clusters, respectively. The misclassification for the domain-level analysis was 21.95% when using either the Rask et al. domains or the hierarchically clustered domains as features. Notably, by making use of the domains defined using the novel hierarchical approach, fewer features were required to achieve a similar classification accuracy. Distinguishing between phenotypes using a smaller number of features is important when investigating possible targets for vaccines. The transcript-level features provided the least discrimination, giving misclassification errors of 31.71% and 43.90% for the combined assembly transcripts and transcript clusters, respectively. It should be noted that these classification rates cannot be generalised to new samples because the cross validation was used to determine the lambda value as well as the misclassification rates. The code for this regression analysis is available in the Github repository.
The relationship between the segments, domains, and transcripts discussed is available in S5 Data and Fig 6. Tree diagrams, like those produced for the domains, are available for each significant segment cluster in the Github repository.
Transcriptional profiling of parasites isolated from patients with severe malaria indicated a shift towards a less glycolytic phenotype. Previous studies have also reported decreases in glycolytic transcripts in some clinical isolates [6], including those from patients with higher temperatures [8] as well as in parasites cultivated in vitro at a high density that inhibits subsequent growth [103]. Down-regulation of genes encoding key enzymes in folate and pyrimidine biosynthesis is also consistent with decreased nucleotide production and reduced parasite growth. The down-regulation of genes involved in histone methylation was similar to deregulation of genes involved in chromatin and RNA biology that was observed in clinical isolates from patients with an elevated surrogate measure of parasitaemia [8].
Our data suggest that parasites causing severe malaria have a more metabolically quiescent phenotype than parasites causing uncomplicated malaria. It remains to be determined whether parasites with the severe malaria transcriptional profile are more resilient and therefore able to cause severe malaria, or whether the host environment in either severe malaria or uncomplicated malaria could have selected or elicited the differing transcriptional profiles. Modulation of parasite growth in response to host environment might be consistent with previous reports of P. falciparum density sensing in malaria [103–105] and protracted maturation of P. berghei and P. yoelli in response to an acute host immune response [106]. In the latter study, more mature, circulating P. berghei and P. yoelli were detected in semi-immune than naive mice. This was consistent with our observation that the circulating parasites were older in the uncomplicated than the severe malaria patients because we previously showed that the uncomplicated malaria patients had more immunity to PfEMP1 than the severe malaria patients [88].
The parasites causing severe malaria had also down-regulated genes involved in PfEMP1 surface expression. This differed from the reported increased expression of genes encoding exported proteins involved in PfEMP1 surface expression in severe malaria from a posthoc comparison [107] of separately published transcriptomes of parasites causing severe [7] and uncomplicated [58] malaria. This difference probably relates to the difficulty of posthoc inference of differential gene expression when the compared samples are from different populations and different studies and were analysed using different microarrays. None of the 87 genes identified in up-regulated gene sets by Pelle et al. were up-regulated in severe malaria in the current study; however, 17 of these genes were down-regulated, including skeleton-binding protein 1, which was the only gene directly involved in PfEMP1 surface expression identified by Pelle et al. We previously showed that the severe malaria patients in the current study had antibodies to PfEMP1 that were generally present at lower levels and that recognised fewer PfEMP1s than the antibodies from the uncomplicated malaria patients [88]. This suggests that humoral immunity to PfEMP1 did not select for decreased PfEMP1 surface expression in the parasites causing severe malaria. However, loss or decrease of many of the proteins involved in PfEMP1 surface expression causes decreased cytoadherence [76,108,109], so the parasites infecting patients with severe malaria at the time of sampling might have had a decreased cytoadherent capacity.
The unique var transcriptional profile we describe in severe malaria recapitulates all of the previously described associations as well as uncovering multiple, novel sequence associations. These findings are remarkable considering that all of the associations that have been observed previously in children with severe malaria from multiple sites across Africa were found in 23 adults with severe malaria from Papua. This suggests that the same conserved var genes are associated with severe disease in nonimmune individuals regardless of geography or patients’ age. Furthermore, a consistent pattern of expression of restricted subsets of var genes, domains, and/or segments was observed despite heterogeneous presentations of severe disease. Similarly, the severe malaria non-var transcriptome clusters also did not segregate by specific severe malaria syndromes. These observations suggest that common mechanisms of disease may cause the varied syndromes of severe malaria. This could have therapeutic implications, although the analyses should be confirmed with larger sample sets.
These findings emphasise the strength of the association between severe malaria and DC8, DC4, DC6 CIDRα1, DBLβ3, and DBLβ12 sequences, which were each shown to be up-regulated in multiple analyses of the de novo var assemblies. However, they also uncover significant, novel associations with other var sequences at the transcript, domain, and segment level. Some of these were found at multiple levels of analysis, e.g., DC11, CIDRα2.6-DBLβ8, DBLε3, DBLγ3, DBLζ4, and DBLε2/9, and in the individual domain analysis, the latter 4 domains were expressed at least as highly in severe malaria as the EPCR-binding CIDRα1 sequences. We cannot exclude the possibility that some of these domains were present on the same PfEMP1 as a CIDRα1 sequence; however, CIDRα1 was not present on the significantly up-regulated transcripts that carried these other domains in either the combined assembly or the corset analysis of the individual isolate var assemblies.
We developed a novel analytical approach testing sequences for associations with disease at multiple levels of sequence homology. This revealed domain subtypes that were strongly associated with disease, including a highly conserved CIDRβ1 subtype and a DBLδ1 subtype that clustered in the same patients. The diversity of the parent CIDRβ1 and DBLδ1 subtypes prevented detection of an association using the established subtype classifications. Finally, we revealed striking associations between smaller var sequence segments and severe disease again by testing for associations at multiple levels of sequence identity. These small segments were limited in number, and many of the findings recapitulated our domain analysis. Some of these segments collapsed multiple domain subtypes, e.g., DBLβ_block4 collapsed DBLβ3 and DBLβ12, raising the possibility that a single segment may elicit cross-reactive immunity against different domain subtypes that are independently associated with severe disease. These segments may help identify critical, fine-scale details of the var sequences expressed by parasites that cause disease and may be of great utility in designing vaccines for severe malaria. The association of these sequences with severe malaria should be validated in other populations from across the world and the encoded proteins tested for adhesion phenotype and for seroreactivity consistent with protection from severe malaria.
Written, informed consent was provided by all participants. The study was approved in Indonesia by the Eijkman Institute Research Ethics Commission (project number 46), in Australia by the Melbourne Health Human Research Ethics Committee (project number 2010.284) and Human Research Ethics Committee of the NT Department of Health & Families and Menzies School of Health Research, Darwin, Australia (HREC 2010–1396).
The data sets generated and/or analysed during the current study are available in the Arrayexpress repository accession: E-MTAB-5860 (sequenced libraries for each sample) and the ENA repository accession: PRJEB20632 (de novo var gene assemblies for combined and individual samples).
Venous samples were collected from patients with severe (n = 23) and uncomplicated (n = 21) malaria attending a healthcare facility in Timika, Papua Province, Indonesia. This area has unstable malaria transmission, with estimated annual parasite incidence of 450 per 1,000 population and symptomatic illness in all ages [110]. Severe malaria was defined as peripheral parasitaemia with at least one modified World Health Organization (WHO) criterion of severity [111]. All of the 23 patients with severe malaria had parasitemias greater than 1,000 per μL, which is a previously derived threshold that predicts clinical disease in northern Papua [49]. Therefore, incidental parasitaemia is unlikely in these severe malaria patients.
White blood cells were depleted from the blood by retention on CF11 cellulose (Whatman-no longer available) using a modification of a previously described protocol [112] (S1 Text Supplementary methods). RNA was extracted from erythrocytes in TRIzol using a modified RNeasy mini (QIAGEN, Hilden, Germany) protocol (S1 Text Supplementary methods). Purified RNA 1 to 3 μg was depleted of Hb mRNA using the Globinclear human Hb RNA depletion kit (Ambion, Thermo Fisher Scientific, Waltham, MA) and a modified protocol (S1 Text supplementary methods).
mRNA was oligo dT purified from the total RNA using the NEBNext Poly(A) mRNA magnetic isolation module (New England Biolabs, Ipswich, MA) and mRNA fragmented, reverse transcribed, and used for library synthesis using the NEBnext ultra directional RNA library prep kit for Illumina (New England Biolabs) as per the manufacturer’s instructions but with modifications (S1 Text supplementary methods), including a high AT tolerant PCR amplification [113]. Libraries were 100 bp paired end sequenced on a 2500-HT Hiseq (Illumina, San Diego, CA) using RapidRun chemistry (Illumina).
Briefly, de novo assembly of var genes was performed by running the SoapDeNovo-Trans [114] and Cap3 [115] pipeline described in [53] (Fig C in S2 Fig). Non-var reads were first filtered out by removing reads that aligned to the H. sapiens, P. vivax, and non-var P. falciparum reference genomes. The resulting contigs were filtered for contaminants and translated into the correct reading frame. A more thorough description of the assembly methods is available in S1 Text. Additionally, the code used to run the pipeline is available on Github at https://github.com/PapenfussLab/assemble_var.
Reads were first aligned to the H. sapiens and P. falciparum reference genomes using Subread-align v1.4.6 [97] with parameters -u -H. FeatureCounts v1.20.2 [116] was used to obtain read counts for each gene. To account for parasite life cycle, each sample is estimated as a mixture of 6 parasite life cycle stages from [57], excluding the ookinete stage. We aimed to choose the proportions π for each sample to minimise
∑i=1N(gi,sample−∑s∈Sπsgi,s)2
subject to the constraints
∑s∈Sπs=1
And
πs≥0
such that gi,s represents the expression of the ith gene in stage s of the [57] data.
Three factors of unwanted variation were estimated using the RUV4 function from the R package ruv v0.9.6 [117] using the 1,009 genes with the lowest p-values from [118] as controls. The choice of control genes was compared to using the least differentially expressed genes of [8], which was found to give similar results. Finally, the gene counts along with the estimated ring-stage factor, and 3 factors of unwanted variation estimated by RUV4 were fed into the Limma/Voom [55,56] differential analysis pipeline. For a detailed outline of the specific commands run in the all-gene analysis, see S1 Text and rmarkdown S1 Text available in the Github repository https://github.com/gtonkinhill/falciparum_transcriptome_manuscript.
For all levels of var expression analysis, library size was normalised using the median ratio method [119] in the default DESeq2 pipeline [86].
Differential expression analysis at the var transcript level was performed using 2 distinct approaches. The first made use of the separately assembled transcripts by first aligning reads to the transcripts allowing for multiple mapping using Bowtie v0.12.9 [120]. The transcripts were then clustered using Corset v1.03 [89]. The resulting cluster read counts were analysed using the default DESeq2 pipeline [86]. An alternative strategy made use of the combined assembly transcripts. Reads were aligned and transcript level counts obtained using Subread and Featurecounts, respectively, before analysing differential expression using DESeq2. Rmarkdown texts S2 and S3 in the Github repository provide a more thorough description of this analysis along with the code https://github.com/gtonkinhill/falciparum_transcriptome_manuscript.
HMMER3’s hmmsearch v3.1b1 [91] was used to search the NTS, DBL, and CIDR domain models of [14] against the assembled transcripts from each sample. The most significant domain was annotated first and then successively less significant domains, with the requirement that 2 domains do not overlap. An E-value threshold of 1e-8 was chosen to minimise spurious annotations.
FeatureCounts was used to allocate reads to domains using a SAF file built from the HMMER3 annotations. The resulting counts were then aggregated using the previous domain classification of [14] as well as a novel hierarchical approach. The annotated domains were hierarchically clustered using USEARCH [93] by first clustering by length and then by successively lower identity thresholds. The read counts for each domain are then aggregated up this hierarchical tree, and the default DESeq2 pipeline was used to identify differentially expressed nodes. After multiple testing correction [94], we iteratively reject the null hypothesis (p < 0.05) of the most significant node in the hierarchy before removing its ancestor and children nodes. This ensures that we select the most significant grouping of domains from which to form clusters. DESeq2 was also run on the domains aggregated using the previous classification.
|
10.1371/journal.pntd.0004299 | Genomic and Proteomic Studies on the Mode of Action of Oxaboroles against the African Trypanosome | SCYX-7158, an oxaborole, is currently in Phase I clinical trials for the treatment of human African trypanosomiasis. Here we investigate possible modes of action against Trypanosoma brucei using orthogonal chemo-proteomic and genomic approaches. SILAC-based proteomic studies using an oxaborole analogue immobilised onto a resin was used either in competition with a soluble oxaborole or an immobilised inactive control to identify thirteen proteins common to both strategies. Cell-cycle analysis of cells incubated with sub-lethal concentrations of an oxaborole identified a subtle but significant accumulation of G2 and >G2 cells. Given the possibility of compromised DNA fidelity, we investigated long-term exposure of T. brucei to oxaboroles by generating resistant cell lines in vitro. Resistance proved more difficult to generate than for drugs currently used in the field, and in one of our three cell lines was unstable. Whole-genome sequencing of the resistant cell lines revealed single nucleotide polymorphisms in 66 genes and several large-scale genomic aberrations. The absence of a simple consistent mechanism among resistant cell lines and the diverse list of binding partners from the proteomic studies suggest a degree of polypharmacology that should reduce the risk of resistance to this compound class emerging in the field. The combined genetic and chemical biology approaches have provided lists of candidates to be investigated for more detailed information on the mode of action of this promising new drug class.
| The mode of action of a new class of boron-containing chemicals (the oxaboroles), currently under development for the treatment of human African trypanosomiasis, is unknown. Here we identify a number of potential candidate proteins that could be involved either in the mode of action of these compounds or in the mechanism of resistance. This information could prove critical in protecting the compounds against resistance emerging in the field as well as opening up new avenues for drug discovery.
| Human African trypanosomiasis (HAT) is caused by two subspecies of the unicellular parasite Trypanosoma brucei, an infection which is transmitted by the bite of a tsetse fly. HAT progresses through a haemo-lymphatic stage into a meningo-encephalitic stage [1] and has a fatality rate close to 100% if left untreated [2]. The disease is also a key factor in maintaining the poverty cycle, and patients are often discriminated against or abandoned [3]. The reporting of new cases of HAT has fallen to below 7,000 in 2011 [4]. However, the disease has previously resurged from even lower levels in the 1980s and 1990s [5]. Current estimates place 70 million people at risk with more than 5 million living in areas of high or very high risk for contracting HAT [4].
T. brucei gambiense is responsible for around 98% of reported cases [5], and has been targeted by the World Health Organization for elimination by 2020. However, elimination of T. brucei rhodesiense, which has epidemic potential, is not feasible due to its animal reservoir [5]. The tsetse fly vector also presents significant risks to disease control in that climate change may allow the vector to invade new geographical regions [6], and sexual recombination, which occurs within the vector, could allow rapid transfer of drug resistance and virulence factors [7]. Hence, whilst improvements in control have been achieved, there are several risk factors that could lead to resurgence of the disease [8,9].
Existing drugs are highly unsatisfactory due to toxicity, mode of administration and efficacy [8]. The ease of developing resistance to both components of the nifurtimox / eflornithine combination therapy (NECT, the newest treatment to enter the clinic) [10] is also a major concern [11,12]. Moreover, due to its status as a neglected disease of declining incidence, the current drug discovery pipeline for HAT is far from robust [13]. Thus, development of new drugs remains a critical task.
Recent advances have included the entry of fexinidazole into phase II/III trials against HAT (ongoing) [14] and the identification of oxaboroles as a class of compounds active against T. brucei by a collaboration between the Drugs for Neglected Disease initiative, Anacor Pharmaceuticals and SCYNEXIS [15]. One member of this class, SCYX-7158, shown to be effective in the meningo-encephalitic stage of HAT [16], entered phase I clinical trials in March 2012 and studies, including safety profiling, are ongoing (DNDI diseases and projects portfolio accessed 14/08/15 www.dndi.org/diseases-projects/portfolio/oxaborole-scyx-7158]).
Oxaborole compounds have been demonstrated to act via inhibition of leucyl RNA synthetase as anti-pneumococcal agents [17] and anti-fungal agents [18]. They can also form adducts with cis-diols in sugars and have been shown to inhibit other enzymes such as phosphodiesterases, β-lactamases and kinases (see review [19]). However, the mode of action against African trypanosomes has not been determined. This information would inform the selection of appropriate partner compounds to protect against resistance, and could also open up novel areas of drug discovery.
Our objective was to apply genomic sequencing and chemo-proteomic approaches [20] to facilitate mode of action studies on the oxaborole series, an approach which has been successful with other antitrypanosomal compounds [21,22]. Here, we report the use of two orthogonal methods (proteomic studies using affinity chromatography and stable isotope labelling by amino acids in cell culture [SILAC]; and whole genome sequencing of sensitive and drug-resistant cell lines) to produce lists of candidate targets for oxaborole compounds that will be pursued in future work. Our experiments suggest a high level of polypharmacology that could protect the oxaborole class from resistance emerging in the field.
All chemicals were obtained from Sigma-Aldrich (Gillingham, UK) unless otherwise indicated. Amino acids for SILAC labelling (4,4,5,5-D4 L-Lysine and U-13C6 L-Arginine) were obtained from CK Gas Products (Hampshire, UK). PBS was formulated in-house. Foetal bovine serum for HMI9T was obtained from PAA Laboratories (Yeovil, UK), dialysed foetal bovine serum for SILAC-labelling was obtained from Life Technologies (Paisley, UK). IMDM for SILAC–labelling (lacking Arginine and Lysine) was obtained from Thermo Scientific (Basingstoke, UK). The cOmplete EDTA-free protease inhibitor cocktail was obtained from Roche Diagnostics (West Sussex, UK).
Bloodstream-form T. brucei ‘single marker’ cells [23] were cultured at 37°C with 5% CO2 in HMI9T medium [24]. Cells were counted using a CasyCounter model TT (Roche Innovatis, Reutlingen) and maintained at densities below 5×106 ml−1, sub-culturing as necessary. EC50 determinations were carried out using a resazurin-based assay, and means weighted to the standard error calculated as previously described [25,26].
SILAC-labelling was carried out using an adapted HMI11 [27]; T. brucei log-phase cells in HMI9T medium were washed in PBS and seeded at 1×104 ml−1 into HMI11-SILAC + R6K4 as described previously [28]. Following 3 days growth to ~1×106 ml−1, cells were harvested by centrifugation, washed in ice-cold PBS, and resuspended in PBS at 1.25×109 ml−1. Four parts cell suspension was mixed with one part 5× lysis buffer (25% glycerol; 30 mM MgCl2; 4% IGEPAL CA-630 (octylphenoxy poly(ethyleneoxy)ethanol); 5 mM DTT), cOmplete protease inhibitor cocktail added to 1× concentration and the sample freeze-thawed three times. DNase-I was added to 1 μg ml−1, the mixture incubated on ice for 5 min, vortexed for 10 s and clarified by centrifugation at 20,000 g for 1 h at 4°C. The supernatant was divided into 500 μl aliquots, adjusted to 5 mg ml−1 with 1 × lysis buffer, snap frozen, and stored at -80°C prior to subsequent processing.
SCYX-6759 and oxaborole-1 were prepared using previously published methodology [29]. Full experimental details for the synthesis of the oxaboroles utilised in this study are given in the Supporting Information (S1 Text). Beads derivatised with an oxaborole, or control compound were prepared as follows. The storage solvent was removed from commercial NHS functionalised magnetic beads (Thermo Scientific) and the beads washed and resuspended in anhydrous DMSO (150 μl [mg resin]−1). The amine-containing compound (7 nmol [mg resin]−1) and DIPEA (14 nmol [mg resin]−1) were then added and the resin gently agitated for 24 h at room temperature. After which the reaction solvent was removed, and the beads washed and resuspended in anhydrous DMSO (150 μl [mg resin]−1). Ethanolamine (70 nmol [mg resin]−1) and DIPEA (70 nmol [mg resin]−1) were then added and the resin gently agitated for 24 h at room temperature. The reaction solvent was then removed and the resin washed with DMAc, prior to storage in the same solvent. Note, the incubation step with an amine-containing compound is omitted when preparing ‘blank’ ethanolamine-capped beads.
Lysates were pre-cleared by incubation with ethanolamine-capped paramagnetic beads (0.2 mg) for 30 min at 4°C, after which the supernatant was transferred to a new sample tube along with a 50 μl wash. For competition experiments, oxaborole-1 in DMSO (1 μM final concentration) or a DMSO control was added (0.5% DMSO final) and incubated with mixing for 30 min at 4°C. Subsequently, 0.2 mg of oxaborole-resin (Fig 1) was added to each sample and incubated for a further 60 min at 4°C. The beads were isolated using a magnet, washed twice with lysis buffer and united into a single sample tube. The beads were further washed three times with PBS, and bead-bound proteins were eluted with NuPAGE LDS buffer (Invitrogen) containing 50 mM DTT for 5 min at 95°C. For comparison of the oxaborole resin and control resin, pulldowns were performed in a similar manner in the absence of soluble compound.
Eluted samples were subjected to electrophoresis on a NuPAGE bis-Tris 10% acrylamide gel until the dye front had entered about 1 cm into the gel. The proteins were stained with InstantBlue (Expedeon), and the entire stained area excised and subjected to in-gel digestion for 18 h at 37°C with 12.5 μg ml−1 trypsin gold (Promega) in 10 mM NH4HCO3, 10% MeCN. Tryptic peptides were recovered in 45% MeCN, 1% formic acid and lyophilized prior to analysis.
Liquid chromatography tandem mass spectrometry was performed by the Fingerprints Proteomic Facility at the University of Dundee, as described previously [28]. Data was processed using MaxQuant [30] version 1.3.0.5 which incorporates the Andromeda search engine [31]. Proteins were identified by searching a protein sequence database containing T. brucei brucei 927 annotated proteins (Version 4.0, downloaded from TriTrypDB [32], http://www.tritrypdb.org/) supplemented with the VSG221 sequence and frequently observed contaminants (porcine trypsin, bovine serum albumin and mammalian keratins) that contains a total of 10,081 protein sequences. Search parameters specified an MS tolerance of 6 ppm, an MS/MS tolerance at 0.5 Da and full trypsin specificity, allowing for up to two missed cleavages. Carbamidomethylation of cysteine was set as a fixed modification and oxidation of methionine residues, N-terminal protein acetylation and N-pyroglutamate were allowed as variable modifications. Peptides were required to be at least 7 amino acids in length and a MaxQuant score >5, with false discovery rates (FDRs) of 0.01 calculated at the levels of peptides, proteins and modification sites based on the number of hits against the reversed sequence database. SILAC ratios were calculated using only peptides that could be uniquely mapped to a given protein group, and required a minimum of two SILAC pairs. To account for any errors in the counting of the number of cell numbers mixed, the distribution of SILAC ratios was normalised within MaxQuant at the peptide level so that the median of log2 ratios is zero [30]. Data were visualized using Perseus 1.3.0.4 (www.perseus-framework.org) and further information on the identified proteins was obtained from TriTrypDB [32] (http://www.tritrypdb.org).
T. brucei cultures (50 ml) were seeded at 5×105 ml−1 in the presence of 225 nM Oxaborole-1 (5× EC50). Samples (4 ml) were taken at 0, 14, 20 and 28 h, collected by centrifugation at 850 g for 10 min and processed essentially as previously described [33]. Briefly, cells were washed in 1 ml PBS containing 1% FBS, the supernatant was removed and cells resuspended in the residual volume. Cells were fixed with 1 ml ice cold 70% ethanol, adjusted to 5×105 ml−1 and washed twice with PBS containing 1% FBS. Cells were resuspended in 400 μl staining solution (PBS containing 1% FBS, 50 μg ml−1 propidium iodide, 50 μg ml−1 RNase A), stained for 20 min at room temperature before being analysed by flow cytometry as previously described [33].
Oxaborole-resistant T. brucei were generated in three independent flasks by sub-culturing in the presence of increasing concentrations of Oxaborole-1. Beginning at the sub-lethal concentration of 20 nM, the process was continued until the cells were growing in 500 nM (~10 × original EC50). Throughout the process, increasing concentrations of Oxaborole-1 were attempted once the cells were displaying cell growth and motility similar to a control grown in the absence of the drug. After 180 days, cells were cloned by limiting dilution in the presence of 500 nM Oxaborole-1 to yield independent clones from each of the three flasks.
The three resistant clones were diluted 1000-fold into media without Oxaborole-1 and sub-cultured as necessary over a two month period. EC50 determinations were carried out to indicate whether resistance had been maintained in the absence of exposure to Oxaborole-1.
Genomic DNA was prepared from five T. brucei lines, i.e. the parental clone Lister 427 (SM), three oxaborole-resistant clones (clone 1, 2, and 3), and a drug-resistance revertant clone (clone 1R). For each sample, 0.6–2 μg of genomic DNA was used to produce standard Illumina libraries of 400–600 base pairs (bp) [34]. Sequencing was carried out on an Illumina HiSeq 2000 sequencer according to the manufacturer’s standard sequencing protocol and yielded 22.8–29.6 million reads of 100 bp length per library. These data sets represented a nominal sequencing coverage of the T. brucei genome (35Mb) of approximately 65.2- to 84.6-fold. The Illumina data were aligned against the T. brucei brucei TREU927 reference genome [35] assembly using SMALT v0.7.4 (http://www.sanger.ac.uk/resources/software/smalt/). For variant calling, the alignment was run employing an exhaustive search (-x) and with parameters wordlen = 13 (-k), skipstep = 1 (-s), minscor = 0.8 (-m), and insertmax = 1000 (-i). To assess relative read coverage and copy number variations (CNVs), the alignment runs were repeated using the above parameters with repetitive mapping (-r) enabled which results in read pairs with multiple equally good alignment positions being aligned to one of these locations at random. Variants were called using SAMtools v0.1.19 mpileup (-Q 15 for baseQ/BAQ filtering) and BCFtools [36]. To exclude the hypervariable subtelomeric regions, only variants found in the following chromosomal core regions were included in the downstream analyses: Tb927_01_v4:202,695–988,120; Tb927_02_v4:259,723–1,161,408; Tb927_03_v4:146,614–1,602,829; Tb927_04_v4:80,380–1,467,268; Tb927_05_v4:72,088–1,366,595; Tb927_06_v4:111,409–1,414,033; Tb927_07_v4:26,571–2,177,541; Tb927_08_v4:135,192–2,476,033; Tb927_09_v4:325,850–2,394,987; Tb927_10_v5:55,698–3,993,940; Tb927_11_01_v4:36,585–4,482,610. SNP calls were further filtered for all of the following: for a minimum of 8 "high-quality" base calls ("DP4"); for a minimum phred-scale QUAL score of 20; for a maximum phred-scale likelihood of the best genotype call of 5 ("PL1"); for a minimum phred-scale likelihood of the second best genotype call of 10 ("PL2"); for a minimum strand bias P-value of 0.01 (first of "PV4"); for a maximum ratio of conflicting base calls for homozygous genotypes of 5%; and, for positions with a minimum and maximum read depth of three times the median read depth observed for that chromosome: for a minimum mapping quality of 20; and for a minimum distance of 10 nucleotides from the nearest INDEL call. Files in Variant Call Format (VCF) listing all 206,417 genomic positions at which any one of the five sequenced parasite lines had a variant call are available in the supplementary material (S1–S5 Datasets). The illustrations showing the location of genes along chromosomal regions in Figure CNVs was generated using Web-Artemis (http://www.genedb.org/web-artemis/).
The raw sequence data are available under the following accession numbers at the European Nucleotide Archive (http://www.ebi.ac.uk/ena): parent: ERS136142; resistant clone 2: ERS136134; resistant clone 3: ERS136135; resistant clone 1: ERS136145; revertant clone of clone 1: ERS136137. The raw and processed mass spectrometry data have been deposited with the ProteomeXchange Consortium [37] (http://www.proteomexchange.org/) via the PRIDE partner repository under the identifier PXD002848.
The synthesis of SCYX-6759 and Oxaborole-1 was readily achieved using published procedures [29]. Initial attempts to immobilise the oxaborole scaffold involved the direct attachment of biotin (for use in conjunction with a streptavidin resin) to the aniline functionality to give Oxaborole-2 (Fig 1 and S1 Fig). However, subsequent biological assay demonstrated that Oxaborole-2 was only weakly active against T. brucei (EC50 15 μM) compared to SCYX-6759 (EC50 0.16 μM), Oxaborole-1 (EC50 0.064 μM) or SCYX-7158 (EC50 0.79 μM, data from [16,38]). Therefore, a small number of analogues retaining the benzamide functionality of SCYX-6759 and SCYX-7158 were prepared (one of which is shown in S2 Fig). Oxaborole-3, which contains a polyethyleneglycol linker in the meta position of the benzamide was found to retain activity in the bloodstream form T. brucei assay (EC50 0.086 μM). The carbamate protecting group of Oxaborole-3 was subsequently removed and the resultant primary amine reacted to prepare an amide of biotin (Oxaborole-Biotin, S2 Fig), which in this case was found to be bioactive (EC50 0.40 μM). An analogue of SCYX-6759, where the oxaborole bicycle was replaced with a phthalide bicycle (Control-1) was prepared and found to be inactive against T. brucei. Therefore, a control biotin conjugate (Control-Biotin) was prepared in an analogous fashion to Oxaborole-Biotin (S3 Fig). Pilot chemical proteomics studies suggested that the use of a biotin-conjugate/streptavidin bead system was sub-optimal. As a result, the linker containing oxaborole and control analogues were instead attached to paramagnetic beads via an amide linkage to give an Oxaborole-Resin and a Control-Resin respectively (Fig 1 and S2 and S3 Figs).
In order to directly profile the proteins that bound to the Oxaborole-Resin, chemical proteomic profiling was undertaken using two orthogonal strategies that utilised SILAC quantitation to eliminate non-specific binding proteins. In these experiments parasites are grown in identical media where one contains “light” and the other “heavy” amino acid isotopes (in this case arginine and lysine). After several rounds of cell division, the two populations are identical except for the differential labelling of the proteome with either light or heavy isotopes. After undergoing differential processing, the two samples are combined and the ratio of heavy to light peptide from each individual protein determined by mass spectrometry. Proteins that are specifically enriched by the differential treatment will have a heavy to light ratio not equal to 1, whereas proteins affected equally with have a ratio = 1 (or binary logarithm of 1 = zero).
In the first strategy, the profile of proteins from T. brucei cell lysates that bind the beads in the presence (heavy label) or absence (light label) of soluble inhibitor was quantified (Fig 2). Non-specific binders will be unaffected by the presence of soluble compound, thus will produce an equal heavy to light ratio (log2 H/L = 0). In contrast, the specific binders will bind Oxaborole-1 in the pre-incubation step, making them unavailable to bind to the immobilised oxaborole, resulting in a low heavy to light ratio (log2 H/L < 0) (Fig 2A, upper pair).
In the second strategy, an inactive Control-Resin was prepared (Fig 1 and S3 Fig). The profile of proteins from T. brucei cell lysates that bind the Control-Resin (heavy label) or Oxaborole-Resin (light label) can then be quantified (Fig 2). Proteins that bind non-specifically, or whose binding is not related to activity, produce an equal heavy to light ratio (log2 H/L = 0), whereas proteins whose binding correlates with activity will have a low heavy to light ratio (log2 H/L < 0) (Fig 2A, lower pair).
The results of the two orthogonal strategies are shown in Fig 2 and Table 1, with the full data presented in the supplementary material (S1 Table). The binding of a subset of proteins was prevented by the presence of the soluble compound (Fig 2B), with 42 proteins displaying greater than a four-fold reduction in binding (log2 H/L < -2). In the orthogonal strategy, a subset of 24 proteins displayed greater than a four-fold reduction in binding (log2 H/L < -2) to the Control-Resin compared to the Oxaborole-Resin (Fig 2C). Comparing the profile of the proteins quantified in both experiments revealed a strong correlation (Pearson 0.841) between the proteins that are displaced by Oxaborole-1 and those that bind only the Oxaborole-Resin and not the Control-Resin (Fig 2D). The 14 proteins that display greater than four-fold selectivity in each experiment, presented in Table 1, can be considered to be specific targets of Oxaborole-1. The number of specific targets identified and the lack of discernible commonality strongly suggest that oxaboroles display considerable polypharmacology, and provide too great a number to investigate systematically as individual targets.
FACS analysis of T. brucei cells incubated with 225 nM Oxaborole-1 (5× EC50) indicated a statistically significant increase in the proportion of G2 and >G2 cells compared to the untreated control (Fig 3). These increases probably result from re-replication of DNA in the absence of cytokinesis. DNA re-replication has been seen in a variety of mutant T. brucei cell lines; however, it is possible that the cytokinesis defect is an indirect effect [40]. Indeed perturbation of several processes results in inhibition of cytokinesis including flagellar attachment, GPI biosynthesis, Golgi duplication and kinetoplast duplication [41,42]. Given such an impact on DNA fidelity, we wanted to investigate the genomic effects of resistance to the oxaborole.
In order to investigate the ease with which resistance to the oxaborole could occur, we generated three independent clones of T. brucei able to sustain growth in 500-nM Oxaborole-1 (Fig 4A). This process took 180 days for all three cell lines to achieve the target of growth in 500 nM Oxaborole-1. A single clone was chosen for each resistant line, and sensitivity to the oxaborole measured by EC50 (Fig 4B). The resulting EC50 shifts were between 5–8 fold compared to the sensitivity of the parental cell line to Oxaborole-1. A similar process using nifurtimox generated T. brucei able to grow in >20× EC50 after 140 days [11], although after cloning, the shift in sensitivity to nifurtimox was 8-fold. T. brucei resistant to eflornithine, pentamidine and the methionine tRNA synthetase inhibitor 1433, have all been generated to grow at 32× their EC50 concentrations within 120 days or less [43]. Whilst a major motive for investigating mode of action was to aid the protection of the oxaborole class from resistance in the field, these results indicate greater resilience to resistance to the oxaborole class than drugs currently used in the field as well as compounds in development.
Following two weeks incubation in the absence of Oxaborole-1, an EC50 determination showed loss of resistance from cell line 1 (EC50 value of 123 ± 12 nM compared to 530 ± 21 nM). This cell line was cloned by limiting dilution and EC50 determinations were carried out on five clones. All of the clones showed loss of resistance, the clone with greatest loss of resistance (termed clone 1R) had an EC50 value of 83 ± 2 nM (weighted mean of two determinations). Culture of resistant cell lines 2 and 3 failed to show any loss of resistance over eight weeks in the absence of Oxaborole-1.
The apparent greater instability of resistance in resistant cell line 1 was consistent with problems encountered when attempting to revive frozen cells. Stabilated cells revived from resistant cell lines 2 and 3 grew at the same rate as the parental cell line. However, cells from resistant cell line 1 showed little motility, although there were no abnormalities in gross morphology by light microscopy. After 7–14 days growth was regained, however after a single passage and three days of growth to select for healthy cells, resistance had been lost. This suggests at least two routes of resistance, an unstable mechanism and one or more stable mechanisms. Greater stability could be conferred by a gene segment being totally lost rather than silenced, or genomic amplification carrying a significant fitness cost compared to one with no such cost.
To identify genetic determinants that may be involved in drug resistance to the oxaborole class, we sequenced the genomes of the susceptible parental strain Lister 427, the three drug-resistant clones and the revertant cell line. We found striking copy number variations (CNVs) between the parasite clones, ranging from apparent whole chromosome duplications to CNVs affecting regions of approximately 5 kb to 15 kb in length (Fig 5). For example, chromosome 1 occurs in three instead of the usual two copies in the genome of clones 1 and 1R (Fig 5A), while chromosome 4 displays an elevated copy number in clone 2 (Fig 5B). In addition, a short region of chromosome 4 of approximately 5.5 kb is further duplicated in this cell line (Fig 5C), thereby providing further complete copies of the two genes CPSF3 (a putative cleavage and polyadenylation specificity factor subunit, Tb927.4.1340) and glx2-2 (a glyoxalase, Tb927.4.1350) (Table 2). In contrast, two short regions on chromosome 6 and chromosome 10 in drug-resistant clone 3 have lost one of their two alleles (Fig 5D and 5E). Interestingly, the deleted regions are flanked in both cases by shorter regions with nearly 100% sequence identity: on chromosome 6 the central, deleted region of 5.1 kb is flanked by two near-identical regions of approximately 4.8 kb each, whereas on chromosome 10 it is a central region of 12.9 kb that is flanked by two near-identical regions of approximately 3.1 kb each (Fig 5D and 5E). This suggests homologous crossover as the mechanism of DNA deletion in these cases. These deletions directly affect over a dozen genes (Table 2) and render the affected regions hemizygous, an observation that is confirmed by the loss of a second allele in the genotype of some of these genes (S2 Table).
The comparison of genotype assignments between the drug-resistant lines and those of the parental strain uncovered in total 78 single nucleotide polymorphisms (SNPs) in 66 genes. Of these, 41 in 38 genes are predicted to result in non-synonymous amino acid changes that could potentially contribute to the observed drug resistance phenotypes (S2 Table). Only one SNP was common to all four clones (receptor-type adenylate cyclase GRESAG 4, putative), but no SNP was common to all 3 resistant clones and absent from the revertant clone, as might be expected if a single point mutation in a single gene was responsible for resistance. Likewise, the small ubiquitin-related modifier (SUMO) contained different SNPs in all 4 clones. SUMOylation regulates a wide variety of cellular processes, including transcription, mitotic chromosome segregation, DNA replication and repair and ribosomal biogenesis [44,45] and offers an attractive explanation for the chromosomal abnormalities described above. Knock down of SUMO in procyclic forms of T. brucei results in arrest in G2/M phase of the cell cycle as observed here [46]. In the case of clone 3, the SNP results in replacement of the initiator methionine residue with an isoleucine. Inspection of the flanking region of SUMO revealed no upstream in-frame methionine and the next downstream methionine is at residue 50 in this 114 amino acid protein. Based on the solution structure of T. brucei SUMO [47], the truncated protein is likely to be non-functional. In the other two clones, the SNPs are located at the C-terminus of the 114 residue peptide close to (Ala101Gly, clone 2) or adjacent to (Thr106Ile, clone 1) the site of cleavage by ULP1/SENP which reveals a C-terminal di-glycine motif required for activation of SUMO by the E1 activating complex [45]. Ala101 maps to a region predicted to interact with the SUMO conjugating enzyme E2 (Ubc9) [47], but it is difficult to predict whether or not such a conservative substitution with a glycine would significantly alter the interaction of the enzyme with its substrate. Cells expressing Thr106Arg or Thr106Lys SUMO mutants have been used in a proteomic study to successfully identify SUMO targets in T. brucei [48] so it appears possible that an isoleucine would also be tolerated at this position. Furthermore, this SNP is retained in the revertant clone, suggesting this mutation is not involved in resistance. Nevertheless, it is possible that different resistance mechanisms may have arisen in each of these lines. None of the genes potentially involved in SUMOylation [49] were found in common with either our proteomic or our genomic studies. However, of the 44 proteins identified as SUMOylated in a previous study [48], one gene Tb927.4.1330 (DNA topoisomerase 1B, large subunit) was identified as duplicated in resistant clone 2 (Table 2). This 90 kDa protein has 4 SUMOylation sites [48] and forms a functional heterodimer with a 36 kDa catalytic subunit and is essential for growth of the parasite [50]. However, it is noteworthy that topoisomerase-IIα, a SUMOylated protein in other organisms, is essential in bloodstream form T. brucei for centromere-specific topoisomerase cleavage activity [48], but was not present in any of our candidate lists
Another candidate in the genome sequencing data, T. brucei homoserine kinase, has recently been studied in our laboratory in relation to de novo synthesis of threonine [51]. The recombinant enzyme was completely insensitive to inhibition by Oxaborole-1 (up to 50 μM) and therefore is not the target for this compound. To investigate if the deletion of a copy homoserine kinase (CNV in resistant line 3, Table 2) was implicated in resistance, the sensitivity to Oxaborole-1 in wild-type (WT), single knockout (SKOPAC) and double knockout (DKO) bloodstream forms was determined. The resulting EC50 values were all within experimental error of each other (41.1 ± 1.6, 36.1 ± 1.3 and 37.5 ± 1.5 nM for WT, SKOPAC and DKO, respectively). Taken together, we can conclude that HSK is neither the target nor a resistance determinant for oxaborole compounds.
This list of candidate mode of action genes, affected either by CNVs or the presence of SNPs, is too long to be systematically investigated. Since no genes are common to both the proteomic studies and the resistant studies, genes involved in resistance mechanisms would appear to be distinct from candidate proteins implicated in the mode of action. In addition, the genomic variations we have observed could be the result of either oxaborole exposure or an unidentified resistance mechanism resulting in a general loss of DNA fidelity.
In conclusion, genetic analysis of laboratory-generated resistant lines has been an effective technique when the field can be narrowed to particular genes of interest as in the case of resistance of T. brucei to tRNA synthetase inhibitors resulting from overexpression of the target [43]. However, taking an unbiased whole genome sequencing approach alongside the analysis of oxaborole-binding proteins in the current study, has revealed too many candidates to embark on a systematic appraisal.
Our SILAC-based analysis suggests considerable polypharmacology consistent with the unusually long time taken to develop resistance, apparent multiple routes to resistance and lack of stability in at least one of those routes. It should be borne in mind that resistance and/or mode of action may involve several candidates acting in concert.
The surprising number of large-scale genomic aberrations in our resistant cell lines (Fig 5), and the accumulation of cells in G2/>G2 (Fig 3) suggest DNA fidelity as an area of specific interest. The presence of SNPs in the gene for SUMO (S2 Table) is particularly striking as its repertoire of targets includes proteins involved in chromatin structure and DNA repair [52]. Future work will involve selecting candidates to test by protein modulation and sensitivity to oxaboroles in the whole cell.
|
10.1371/journal.pcbi.1000852 | The Construction and Use of Log-Odds Substitution Scores for Multiple Sequence Alignment | Most pairwise and multiple sequence alignment programs seek alignments with optimal scores. Central to defining such scores is selecting a set of substitution scores for aligned amino acids or nucleotides. For local pairwise alignment, substitution scores are implicitly of log-odds form. We now extend the log-odds formalism to multiple alignments, using Bayesian methods to construct “BILD” (“Bayesian Integral Log-odds”) substitution scores from prior distributions describing columns of related letters. This approach has been used previously only to define scores for aligning individual sequences to sequence profiles, but it has much broader applicability. We describe how to calculate BILD scores efficiently, and illustrate their uses in Gibbs sampling optimization procedures, gapped alignment, and the construction of hidden Markov model profiles. BILD scores enable automated selection of optimal motif and domain model widths, and can inform the decision of whether to include a sequence in a multiple alignment, and the selection of insertion and deletion locations. Other applications include the classification of related sequences into subfamilies, and the definition of profile-profile alignment scores. Although a fully realized multiple alignment program must rely upon more than substitution scores, many existing multiple alignment programs can be modified to employ BILD scores. We illustrate how simple BILD score based strategies can enhance the recognition of DNA binding domains, including the Api-AP2 domain in Toxoplasma gondii and Plasmodium falciparum.
| Multiple sequence alignment is a fundamental tool of biological research, widely used to identify important regions of DNA or protein molecules, to infer their biological functions, to reconstruct ancestries, and in numerous other applications. The effectiveness and accuracy of sequence comparison programs depends crucially upon the quality of the scoring systems they use to measure sequence similarity. To compare pairs of DNA or protein sequences, the best strategy for constructing similarity measures has long been understood, but there has been a lack of consensus about how to measure similarity among multiple (i.e. more than two) sequences. In this paper, we describe a natural generalization to multiple alignment of the accepted measure of pairwise similarity. A large variety of methods that are used to compare and analyze DNA or protein molecules, or to model protein domain families, could be rendered more sensitive and precise by adopting this similarity measure. We illustrate how our measure can enhance the recognition of important DNA binding domains.
| Protein and DNA sequence alignment is a fundamental tool of computational molecular biology. It is used for functional prediction, genome annotation, the discovery of functional elements and motifs, homology-based structure prediction and modeling, phylogenetic reconstruction, and in numerous other applications. The effectiveness of alignment programs depends crucially upon the scoring systems they employ to evaluate possible alignments. For pairwise alignments, scores typically are defined as the sum of “substitution scores” for aligning pairs of letters (amino acids or nucleotides), and “gap scores” for aligning letters in one sequence with null characters between letters in the other. Substitution and gap scores may be generalized to multiple alignments, i.e. those involving three or more sequences.
Most useful local pairwise alignment algorithms allow gaps and explicitly assign them scores [1]–[4]. However, many local multiple alignment algorithms do not allow gaps, or allow them only implicitly as spacers between distinct ungapped alignment blocks. Indeed the alignments recorded in some protein family databases are explicitly constructed with ungapped alignment blocks separated by variable length spacers [5], and it has been argued that this formalism corresponds well to the observed relationships imposed by protein structure [6]. Short ungapped blocks are also used in the DNA context, to represent, for example, transcription factor binding sites.
Many pairwise substitution scores have been developed for protein [7]–[20] and DNA [21], [22] sequence comparison, and a statistical theory for substitution scores has been developed for local alignments without gaps [23], [24]. It is not trivial to generalize pairwise scoring systems to multiple alignments, and the following four principal approaches have been proposed to this long-standing problem: A) Tree scores. An evolutionary tree can be defined relating the sequences in question, with each sequence residing at one leaf of the tree. By reconstructing letters at the internal nodes of the tree, the score for an aligned column of letters is defined as the sum of pairwise substitution scores for all edges of the tree [25], [26]. B) Star scores. As a special case of tree-scores, a single “consensus” letter can be defined for an alignment column. The column score is defined as the sum of pairwise scores for the consensus letter to each letter in the column. The tree in question reduces to a star, with the consensus at the central node. C) Sum-of-the-Pairs or SP scores. A column score can be constructed as the sum of substitution scores for all pairs of letters in the column [27], [28]. D) Entropy scores. Scores can be based on the entropy of the letter frequencies observed in a column [29]; these scores have become particularly popular for DNA alignments. All these approaches are open to refinement, for example by weighting the pairwise scores of the sequences involved.
All reasonable substitution scores for pairwise local alignment are implicitly log-odds scores [23], [30], which compare the probabilities of aligning two letters under models of relatedness and non-relatedness, and the most popular are explicitly so constructed [7], [8], [14]. We argue that multiple alignment column scores should be similarly constructed, based upon explicit target frequency predictions for columns from accurate alignments of related sequences. For this purpose, we propose, the method with the strongest theoretical foundation relies upon the specification of a Bayesian prior, over the space of multinomial distributions for describing alignment columns representing true biological relationships [31], [32]. We call column scores based on such a formalism “Bayesian Integral Log-odds” or BILD scores. Although these scores are implicit in earlier work, their full generality and utility has not been recognized. They may be calculated efficiently, and may be generalized to allow for the differential weighting of sequences in a multiple alignment. We also consider an alternative approach that allows log-odds column scores to be derived from any pairwise substitution matrix.
Given their form, multiple alignment log-odds scores can be used directly to define the proper extent of multiple alignment blocks, and to derive natural scores for profile-profile comparison. We show that they also arise from the perspective of the Minimum Description Length Principle [33], which allows them to be combined naturally with other information theoretic measures. Other direct applications are specifying when a sequence should be included in a multiple alignment at all, and when an alignment of many related sequences is better split into several alignments each involving fewer sequences.
Efficient methods for calculating BILD scores allow them to be incorporated into Gibbs sampling algorithms for ungapped local multiple alignment. Most practical protein applications, however, require provisions for gaps. We describe two methods for extending an ungapped local multiple alignment produced by the Gibbs sampling strategy to a gapped alignment, the first using asymmetric affine gap costs, and the second hidden Markov models. In the latter, column BILD scores inform the construction of position-specific gap costs, and yield gapped alignments in greater conformity with considerations of protein structure. We illustrate the applications of the programs by using them to uncover previously undescribed Api-AP2 domains of Toxoplasma gondii and Plasmodium falciparum.
Multiple sequence alignment comprises a diverse set of problems and approaches. Many sophisticated statistical inference techniques have been applied to the multiple alignment problem and to the related problem of phylogenetic reconstruction, e.g. [34]–[37]. It is not our purpose here to develop a new multiple alignment program. Rather, we seek only to argue that the “substitution scores” for multiple alignment columns which lie at the core of most multiple alignment methods can in many cases be improved. Although many statistical alignment methods are Bayesian-based, the BILD scores directly implied by Bayesian reasoning have been heretofore unrecognized.
Log-odds pairwise substitution scores can be written . Here, is the frequency with which residues and correspond in accurate alignments of related sequences, and is the background probability with which residue occurs. The base of the logarithm is arbitrary, and merely defines a scale for the scoring system. We henceforth assume that unless the natural logarithm is specified, all logarithms are base , and the resulting scores are therefore in the units of bits [30]. Note that no target frequencies are uniquely optimal for pairwise sequence alignment, because different are appropriate for comparing sequences diverged by different amounts of evolution [7], [8], [13], [30]. This perception gives rise to families of substitution matrices, such as the PAM [7], [8] and BLOSUM [14] series for protein comparison.
To generalize log-odds scores to multiple alignments, we first develop some notation. We consider the alphabet from which the letters in our sequences are drawn to consist of elements, which for convenience we represent by the numbers 1 through . An ungapped column from a multiple alignment of sequences is a vector , each of whose components through takes on a value in . In essence, the log-odds approach compares two theories, one in which all the letters aligned are related or homologous, and the other in which none are. Each theory implies a probability for observing any given set of data. For the alignment column , we define as the probability of observing the data under the assumption of relatedness, and under the assumption of non-relatedness. Then the log-odds score for this column is defined as(1)Assuming background probabilities through for the various letters, is given simply by(2)We will consider one primary strategy for deriving . As with pairwise scores, all sets of multiple alignment column scores with negative expected value are implicitly log-odds scores [23], [30]. However, unless their values for are explicitly constructed in a sensible way, log-odds scores are unlikely to perform well in the applications suggested below.
For alignments of more than two sequences, there are of course other possibilities than for all or none of the sequences to be related. However, as we will describe below, scores of the form of equation (1) can be applied to the comparison of sequences where only a subset are related, by adding indicator variables to include or exclude sequences.
Log-odds scores for alignment columns immediately suggest substitution scores for aligning two different columns of letters. Specifically, letting be the concatenation of the vectors and , define(3)These column-column alignment scores may be used consistently in progressive alignment algorithms, which proceed by aligning the most closely related sequences first [38], [39], although as will be discussed below problems may arise in the definition of gap scores. They may also be used for profile-profile alignment, a topic of considerable recent interest [40]–[48].
For multiple alignments, perhaps the best approach to defining and calculating is a Bayesian one [31], [32]. (An alternative approach, based on pairwise scoring matrices, is described in Text S1.) Assume that the letters in a specific column from an accurate alignment of related sequences are generated independently, but with probabilities through that in general differ from the background probabilities. Assume further that it is possible to assign a prior probability distribution to the multinomial distributions associated with columns of related letters. This prior can be derived from a detailed study of related protein or DNA sequences.
Although the data associated with a specific column generally have no temporal or other privileged order, assume for convenience that they are observed sequentially, in the order to . Then we may apply Bayes' theorem to transform the prior distribution to a posterior , after the observation of . More generally, each subsequent observation can be seen to transform the prior into a posterior distribution . We may then use the chain rule to write(4)The individual terms in this product may be calculated by integrating over all possible multinomial distributions :(5)Finally, combining equations (1), (2) and (4) yields(6)
We call scores defined in this way Bayesian Integral Log-odds or BILD scores. They can be understood simply as the sum of log-odds scores for the individual letters observed in a column, with the “target frequency” for each letter calculated based upon the prior distribution , and the “previously observed” letters through . Even though, by this formula, the log-odds score for a letter varies with its position in the column, the total column score is nevertheless invariant under permutation of the column's letters.
BILD scores have some conceptual connections to star- and entropy-based multiple alignment scoring systems. The simplest generalization of star scores imposes a prior probability distribution on the consensus letter, but still assumes a probabilistic pairwise substitution model. As we describe in Text S1, this yields a class of log-odds scores we call MELD scores. BILD scores arise, in contrast, by thinking of the “consensus” not as an ancestral letter, but rather as a generative probabilistic model, and by integrating over a prior distribution placed on this model.
Given observed and background letter distributions and , entropy scores have been defined variously, and conceptually distinctly, as: i) , the entropy difference between and ; ii) , the entropy difference between a uniform distribution on letters and ; and iii) , the relative entropy of and . Definitions i) and ii) differ only by a constant. One may refine any of these definitions by taking to be a posterior letter distribution, derived from a prior and a set of observations. Both BILD and entropy-based scores can be viewed as the sum of scores derived from the probabilities for individual observations. The central distinction is that BILD scores estimate the probability for a given such observation using only “earlier” ones, whereas entropy scores estimate this probability using the complete collection of observations.
Although the definition of BILD scores is valid for any prior distribution one wishes to specify, it is in general impractical to calculate the , or the integral in equation (5), except when takes the form of a Dirichlet distribution [49], or a mixture of a finite number of Dirichlet distributions [31], [32]. In this case, as described below, all the are also Dirichlet distributions, or Dirichlet mixtures, and is easily calculated. Therefore, for mathematical as opposed to biological reasons, we always assume that BILD scores are defined using a Dirichlet or Dirichlet mixture prior. The family of Dirichlet mixtures, however, is rich enough that it can capture well much relevant prior knowledge concerning relationships among the various amino acids or nucleotides.
We review here the essentials of Dirichlet distributions. A multinomial distribution on letters is specified by an -dimensional vector , within the simplex defined by , and . The requirement that the sum to 1 renders the space of multinomials dimensional. A Dirichlet distribution, defined over this space, is parametrized by an -dimensional vector with all positive. We shall sometimes refer to such a distribution by its parameters , and we define as the sum of the . The Dirichlet distribution is given by the probability density function(7)where the normalizing scalar ensures that integrating over its domain yields 1. Here , is the Gamma function, and for positive integral . The uniform density is a special case that arises when all the are 1.
Dirichlet distributions have two convenient properties. First, the expected frequency of letter implied by is . Second, the posterior distribution yielded by Bayes' theorem, after the observation of the letter , is a Dirichlet distribution with , but with all other parameters equal to those of .
To illustrate how to calculate BILD scores using these properties, consider the case of DNA comparison (with the numbers 1 through 4 identified respectively with the nucleotides A, C, G and T), with uniform background probabilities , and a Dirichlet prior given by the parameter vector (1,1,1,1). By equation (4), the target frequency associated with the alignment column “AATC” is given by . Thus the score for the column is bits. In contrast, for the column “AAAC”, , and the score for this column is bits.
The essence of a Dirichlet distribution is perhaps best understood through the alternative parametrization (; ), where , and . Because the must sum to 1, there are still only independent parameters. The vector describes the center of mass of the distribution, while indicates how concentrated the distribution is about this point. Large values of correspond to distributions with most of their mass near , whereas values of near 0 correspond to distributions with most of their mass near the boundaries of the simplex. It is frequently sensible, although not necessary, to choose a prior whose is identical to the background frequencies . In this case, , and the first summand in equation (6) is always 0. In other words, no letter in a column, considered in isolation, carries any information as to whether the column represents a true biological relationship.
Single Dirichlet distributions frequently are adequate for capturing prior knowledge concerning “true” alignment columns of related DNA sequences, but this is not the case for proteins. Most simply, distinct regions of multinomial space, representing different collections of amino acids, should have high prior probabilities. In order to address the deficiency of single Dirichlet distributions, Brown [31] proposed the use of Dirichlet mixture priors. A Dirichlet mixture is simply the weighted sum of distinct Dirichlet distributions. It is specified by positive “mixture parameters” through that sum to 1, and a set of standard Dirichlet parameters, through , for each of the component Dirichlet distributions. (It will be useful later to define as .) In all, because of the restriction on the sum of the , a Dirichlet mixture has independent parameters. The Dirichlet components of a mixture generally are thought of as describing various types of positions (e.g. hydrophobic, charged, aromatic) typically found in proteins.
Bayes' theorem implies that, given a -component Dirichlet mixture as a prior, the posterior distribution after the observation of a single letter is also a -component Dirichlet mixture [31], [32]. Brown [31] proposed Dirichlet mixture priors in the context of deriving “substitution” scores for aligning amino acids to columns from a multiple protein sequence alignment. This restricted context can be understood as comprehending a single summand from equation (6). BILD scores extend Brown 's sequence-profile alignment scores to comprehensive scores for multiple alignment columns.
Generalizing the development above, we describe here how to calculate the probability of a particular observation given a Dirichlet mixture prior , and how to calculate the posterior resulting from this observation. First, given a Dirichlet mixture, with parameters and , the probability of observing letter is given simply by(8)which follows directly from the definition of Dirichlet mixtures, and the result for single Dirichlet distributions. Second, given the observation of letter , and a Dirichlet mixture prior parametrized as above, the parameters and of the posterior distribution may be calculated as follows:(9)In short, first multiply the mixture parameters by the Bayesian factors and normalize, and then add 1 to each . Mathematics establishing the validity of this procedure appears in [32]. Their development is more complex than we require here, because we modify the Dirichlet mixture parameters only one observation at a time. We note that given the and , it is simple to invert procedure (9) to determine the and . This is useful for applications such as the Gibbs sampling algorithm discussed below.
Many multiple alignment problems involve subsets of sequences that are much more closely related to one another than to the other sequences being considered, and this may yield suboptimal results, because a large number of closely related sequences can “outvote” a few more divergent sequences. One remedy has been to assign each sequence a numerical weight, with closely related sequences down-weighted [50]–[61]. Also, subsumed in such weights may be the recognition that the total number of effective observations represented by an alignment column may be smaller than the number of sequences it comprehends [4], [62], [63]. Thus, for certain applications it may be desirable to generalize BILD scores to weighted sequences. To do so, we need to define the concept of the probability of a “fractional observation” of a letter, and describe as well how a posterior distribution is calculated after such a fractional observation. Arguments supporting how this may be done can be extracted from the mathematical development in [32]. Both equation (8) and the first step of procedure (9) involve multiplication by the factors . For the fraction of an observation of letter , these factors must be replaced by the alternative factors(10)Also, in the last step of procedure (9), the quantity rather than 1 must be added to each . The factors are identical to the original factors when , and all approach 1 as approaches 0, as some reflection shows they must.
Finally, note that equation (10) may be applied to as well as , and may be useful even when all observations are unitary. Thus, by aggregating observations, the BILD score for a column containing unique letters may be calculated with summands, rather than the summands of equation (6). For a single Dirichlet prior, reduces to the simple formula(11)where is the count of letter , and is the total count of all residues. Only the numerator inside the product varies from column to column within an alignment, yielding further efficiency for calculation.
Only the research team that first proposed Dirichlet mixtures for protein sequence comparison has derived, from analyses of large protein alignment collections, sets of Dirichlet mixture prior parameters [31], [32]. Twelve such sets, involving various numbers of Dirichlet components, can currently be found at http://compbio.soe.ucsc.edu/dirichlets/index.html. We list five of these in Table 1, which we refer to as through .
Proteins diverged by different degrees of evolutionary change are best studied using pairwise substitution matrices with different relative entropies [30], and the analogous claim should hold for Dirichlet mixture priors. A Dirichlet mixture prior implies a background amino acid frequency distribution , as well as a symmetric pairwise substitution matrix, by means of the formula . The relative entropies of the substitution matrices implicit in the priors through range from 1.44 bits, roughly equivalent to that of the PAM-80 matrix [7], [8], which is appropriate for fairly close evolutionary relationships, to 0.18 bits, roughly equivalent to that of the PAM-360 matrix, which is appropriate only for extremely distant relationships (Table 1).
As well as , one may calculate the mean relative entropy of the multinomial distributions described by a Dirichlet mixture prior to the background frequencies (see Text S2). For to , ranges from to bits (Table 1). That has a much greater value than indicates that on average much more information is available per position from an accurate multiple alignment of many related sequences than from a single sequence. We note that, in lieu of using different priors, the effective relative entropy of a particular Dirichlet mixture may be tuned by scaling the weights of the sequences to which it is applied [43].
Standard pairwise substitution matrices are constructed from sets of proteins with certain background amino acid frequencies , and are non-optimal for the comparison of proteins with compositions that differ greatly from [64]. Similarly, a Dirichlet mixture prior has an implicit background amino acid composition , and should not be optimal when applied to proteins with compositions that differ greatly from . It is possible to adjust standard matrices for use with non-standard compositions [64], [65], and we will discuss elsewhere an analogous strategy that can be applied to adjust Dirichlet mixture priors.
Single Dirichlet priors may be appropriate for DNA sequence comparison. The uniform density, arising when all (), has frequently been advocated in the absence of prior knowledge, and “Jeffreys' prior” [66], which is uninformative in a deeper sense, corresponds to all () [33]. When specific prior knowledge concerning an application domain is available, however, there is generally not a strong argument for using uninformative priors. For related DNA sequences, the columns of accurate alignments are sometimes dominated by one or two nucleotides, suggesting that all should be smaller than . Furthermore, it usually makes sense for the to be proportional to the background frequencies . If this is stipulated, the specification of a Dirichlet prior reduces to the specification of . Assuming a uniform nucleotide composition, the values of and implied by from to are given in Table 2. An empirical study of transcription factor binding sites [67] concludes that, at least for the analysis of such sites, should be or lower.
A direct application of multiple alignment log-odds scores is to determining local alignment width. As formulated by Smith and Waterman [1], an optimal local alignment is one that maximizes an alignment score but is of arbitrary width. Such scores should fall on the log side of the “log-linear phase transition” [68], which implies that for ungapped local alignments, substitution scores must be of log-odds form [23], [30].
Equation (1) explicitly generalizes pairwise log-odds scores to the multiple alignment case. They are positive for some alignment columns, negative for others, and must have negative expected value. Therefore it is appropriate to define an optimal ungapped multiple alignment as one with maximal aggregate log-odds score. This immediately allows one to define the proper width or extent of an ungapped multiple DNA or protein alignment, without resorting to the ad hoc principles frequently required for other scoring systems [69]. Although the Smith-Waterman algorithm can be applied to optimize log-odds-scored local multiple alignments, it is too slow for most purposes. Nevertheless, once relative offsets have been fixed for a set of sequences, it is trivial to determine an optimal ungapped local multiple alignment along the single implied diagonal.
The ungapped local multiple alignment problem may be formulated as seeking segments of common width within multiple DNA or protein sequences that, when aligned, optimize a defined objective function. We take this function here to be the aggregate log-odds score for the aligned columns. One way to approach this optimization is by means of a Gibbs sampling strategy, as described by Lawrence [69]. Log-odds scores can be used to adjust dynamically, by applying the Smith-Waterman algorithm to the diagonal implied by a provisional alignment, without the need for an arbitrary parameter or an ad hoc optimization. They may also be used to determine dynamically whether or not a sequence should participate in the multiple alignment at all, for which purpose it is useful first to consider log-odds scores from the perspective of the Minimum Description Length Principle.
The Minimum Description Length (MDL) Principle provides a criterion for choosing among alternative theories for describing a set of data [33], [49]. To simplify greatly, it suggests that given a set of alternative theories to describe a set of data , that theory should be chosen which minimizes , defined as the sum of , the description length of the theory, and , the description length of the data given the theory. By convention, description lengths are measured in bits.
From information theory [70], the information associated with an event of probability is bits. Focusing on actual encoding schemes for probabilistic events can unduly complicate MDL analyses. Accordingly, we here follow the approach of section 3.2.2 of [33], in which description lengths are allowed to be non-integral, and are identified with negative log probabilities. Thus, if the data can be described probabilistically, . The length of the theory is defined as the number of bits needed to specify the free parameters of , i.e. those that are fitted to the data [33].
For local multiple alignment, the theory that the input sequences are unrelated has only the background probabilities as parameters, whose description length we will call . The data is comprised of sequences, with lengths through , and consisting of the letters . Then . The theory states that segments of width beginning at positions within the various sequences are related, and that the probability of the data within each column of the implied alignment is ; the probability of the rest of the data may be described with the background frequencies . The free parameters are , the vector of starting positions , and . Each may take on one of values, so its description length is approximately , if is not too large compared to . Thus, we have , where is the description length of . (If all feasible widths are taken to be equally likely, is just . Other encodings have grow slowly with [33], [49].) It is apparent that , where the latter sum is taken only over those letters not participating in the local multiple alignment. Everything simplifies when we consider the difference in the total description lengths of the two theories:(12)where is simply the log-odds score for the implied alignment. In other words, is preferred whenever exceeds . As described in Text S3, this prescription is related to the statistical theory for ungapped local alignments [23].
To allow one or more sequences to be excluded from the multiple alignment, we consider not 2, but theories, distinguished by binary indices , which take on the value 1 to indicate that sequence participates in the alignment, and otherwise. These theories need not be a priori equally likely; if necessary, for from 1 to we can specify prior probabilities that sequence contains a segment related to segments in the other sequences. Let us consider the difference in the description lengths of two theories, and , that differ only in their index . Theory incurs the cost for the prior probability that , and also requires describing the location of the related segment, which costs bits. In contrast, theory incurs only the cost , so costs more bits to describe than . Thus, for to be preferred, the log-odds score of the multiple alignment must increase by at least when the segment from the th sequence is added. If is close to 1, can be negative, and is if . In short, the greater the prior probability that a given sequence contains a relevant segment, the lower the score of such a segment need be for inclusion in the alignment.
The change in the log-odds score with the addition of a segment from the th sequence depends upon which other sequences, and which of their segments, participate in the alignment. Consequently, the values of the indicator variables must be part of the larger optimization, and their selection can be readily incorporated into a Gibbs sampling algorithm. The MDL Principle can also be extended to the case where a single sequence may contain more than one copy of a pattern, and, as previously described [62], [71], [72] and discussed in Text S4, to the clustering of multiple alignments into subfamilies.
Although our central concern is to define a new type of multiple alignment substitution score, many important applications require the construction of gapped multiple alignments, and these generally entail scores for insertions and deletions. Multiple alignment gap scores should be defined in a manner consistent with the substitution scores used [73], so we will consider what type gap scores might fruitfully be paired with BILD scores.
Just as the log-odds perspective places pairwise substitution scores in a probabilistic framework [7], [8], [23], [30], so pairwise gap scores can be viewed as specifying probabilities for insertions and deletions within biologically accurate alignments [74]–[82]. For pairwise alignments, “affine” gap scores, of the form for a gap of length [83]–[85], are those most commonly used [3], [4], although more complex gap scores have frequently been proposed [86]–[89]. When there is an essential asymmetry between the sequences being aligned, differing scores may be assigned to gaps within the two sequences. Furthermore, when substitution and gap scores are properly integrated and both expressed in the units of bits, the two parameters of affine gap scores can be understood to specify jointly the average frequencies and lengths of gaps in the alignments sought [82]. If gaps are to be introduced into the BILD score formalism, an immediate problem is which, if any, letters from individual sequences should be understood as insertions with respect to the “canonical” pattern. In other words, it appears a canonical width for the multiple alignment must somehow be chosen, with respect to which gaps arising in the alignment of individual sequences can be assessed.
For simplicity, suppose we have a “canonical” multiple alignment , i.e. one with a specified number of columns, to which we wish to align a single sequence , to produce a new multiple alignment . It is reasonable to define the alignment score of as the pre-existing alignment score for plus the incremental pairwise score for aligning and . This pairwise alignment involves substitutions (letters from aligned to columns from ), insertions (runs of letters from that are not aligned to any columns from ), and deletions (runs of columns from that are not aligned to any letters from ). BILD scores for the columns of arise naturally when one defines the substitution scores for aligning to as incremental BILD scores. It remains then only to define gap scores for insertions and deletions in the alignment of and .
There is an essential asymmetry in gap scores for aligning to , relevant in many biological applications. For proteins, the columns of represent canonical positions, present in most sequences of a protein family, and it should accordingly be very costly to delete any of these columns. In contrast, individual proteins often contain long loops not present in the great majority of related sequences [90], [91], so even long insertions should not be very costly. Uniform but asymmetric affine insertion and deletion scores can capture this simple idea, and we have implemented them in one program described in the Results section below. These scores can be derived from the average frequencies and lengths [82] of insertions and deletions with respect to canonical protein family multiple alignments.
Just as incremental BILD substitution scores change as more sequences are added to a multiple alignment, so it is possible to let insertion and deletion scores change as well, and vary by position. In the context of Hidden Markov Models [76]–[81], many methods for doing this have been described. Below, we implement one simple procedure that depends only upon the BILD scores of multiple alignment columns, and not upon the relatively sparse gaps observed in any particular alignment.
Formula (3) permits BILD substitution scores to be used for progressive multiple alignment. However, as described above, gaps scores pose a particular problem, because to define insertions and deletions one needs to construct a canonical alignment, and this is difficult for a small number of sequences. For example, when just two proteins are aligned, it is quite possible that gaps in both sequences would ultimately be seen as insertions with respect to a model describing the whole protein family, but there is no obvious way to determine this in advance. (The problem does not arise when substitution and gap scores are defined using the sum-of-pairs or SP formalism [27], [28], for which no canonical alignment is necessary [73].) Accordingly, the approach we take below is eschew gaps at first, and thereby construct a canonical multiple alignment whose columns represent positions present in the majority of sequences. Only then do we realign individual sequences to this model, allowing gaps.
There has been considerable recent interest in aligning profiles that describe different protein families [40]–[48]. If BILD substitution scores, defined by equation (3), are to be used for this purpose, it would seem that we face the same problem for gaps that we do for progressive multiple alignment. Specifically, an insertion with respect to one profile is seen as a deletion with respect to the other, so how may one determine which, if either, perspective to adopt in a model describing both? However, so long as this goal is only to compare pairs of profiles, and not to proceed further, this problem may be elided. It is consistent to define pairwise gap costs for the alignment of two profiles, just as one would for the alignment of two sequences, without reference to a canonical alignment, and the substitution scores of equation (3) can be used sensibly with such gap costs. The gap costs chosen may depend upon the profiles being aligned, and may therefore be asymmetric and position specific. We leave for elsewhere the comparative evaluation of profile-profile alignment using substitution scores defined by equation (3), and those defined in other ways [40]–[48].
Substitution scores for multiple alignment columns form only one element of successful multiple alignment programs. Depending upon their specific purposes, such programs may also employ gap scores, sequence weights, heuristic optimization algorithms, low-complexity filters, discontiguous patterns, provisions for no or multiple copies of a pattern within a sequence, the search for multiple distinct patterns, statistical assessments, etc. It is not our purpose here to develop a fully realized program to outperform existing state-of-the-art programs that involve multiple alignment. Rather, we seek only to argue that the use of explicitly constructed log-odds substitution scores can in many cases add values to these methods.
The programs we consider below have been constructed for evaluation purposes, to isolate the contribution of log-odds scores as much as possible. These programs are parsimonious in their complexity and use of free parameters, and employ various ideas that have appeared frequently elsewhere, and for which no novelty is claimed.
BILD scores find perhaps their purest application in the ungapped local alignment problem described above, so it is worth studying them in this restricted context. The Gibbs sampling approach to finding optimal local multiple alignments was introduced by Lawrence et al. [69], and this algorithm can easily be modified to employ BILD scores. Potential advantages are improved sensitivity and the automatic definition of domain boundaries. Evaluation ideally requires a set of proteins with ungapped domains whose correct alignment is structurally validated, but such sets are unfortunately very rare. Nevertheless, the collection of ungapped helix-turn-helix (HTH) domains in [69] provides a limited test set for analyzing BILD scores in the absence of gaps. As we describe in Text S5, with Tables S1 and S2, BILD scores achieve success on two fronts. First, they have greater average sensitivity than the entropy-based scores proposed by Lawrence et al. [69], in yielding accurate alignment from fewer sequences; second, they recognize with good precision the extent of the structurally-defined domains, and therefore do not require a prior specification of alignment width.
Local multiple alignment programs generally must allow for gaps, either implicitly or explicitly. However, even for aligning gapped domains, the search for ungapped local alignments can be a fruitful first step. BILD scores can play an important role at this stage in defining the common core of a protein family, and can be adapted in subsequent stages to score gapped multiple alignments. As a proof of principle, we here develop a relatively simple algorithm, Program 1, that uses BILD scores as part of a gapped multiple alignment strategy. We describe this program's architecture and motivation below, and use a standard artificial test set to evaluate its ability to recognize the boundaries of local motifs, and to properly construct gapped local alignments. We then describe in section C how Program 1 may be refined through the consideration of features of protein structure, and illustrate the application of our methods to the delineation of a protein domain family.
As mentioned above, real protein domains are subject, on average, to much longer insertions than deletions, and this implies the utility of asymmetric affine gap costs for Program 1. The particular costs that are best will depend upon the statistical properties of gaps, and a possible refinement of Program 1 would be to adjust gap costs dynamically. From the analysis of a variety of protein families, we have found empirically that reasonable gap scores to use in conjunction with Dirichlet mixture priors are bits for a deletion of motif positions (corresponding [82] to an initiation frequency per motif position of 0.28%, and a mean length of 2.0), and bits for an insertion of length into the motif (corresponding to a frequency of 0.87%, and a mean length of ).
Protein structure implies more than an asymmetry between the frequency and length statistics of insertions and deletions. Reflecting the evolution of secondary structure elements and loops, certain motif positions are much less likely to be deleted than others and, similarly, insertions are much less likely to occur between certain pairs of motif positions than others. We describe below an extension of Program 1 to an HMM-based Program 2 that relies only upon column BILD scores to calculate position-specific gap score parameters. We then apply Programs 1 and 2 to the detection of Api-AP2 domains.
We have described a natural generalization of log-odds substitution scores for pairwise alignments to substitution scores for multiple alignment columns. Multiple alignment log-odds scores probably are best derived using a Bayesian approach, yielding what we have called BILD scores. Log-odds scores imply scores for aligning multiple alignment columns to one another, or for aligning multiple alignment columns to single sequences, and it was in this latter context that the Bayesian approach was first formulated by Brown [31]. In conjunction with the Minimum Description Length Principle, log-odds scores provide a means for determining the proper width or extent of a local multiple alignment, and for deciding whether a segment should be included in the alignment. They may also be used to cluster a set of related segments into subclasses; see Text S4 and [62], [71], [72].
One may compute rapidly the BILD score for a multiple alignment column, as well as the new score that results from the addition or subtraction of a single letter. This permits BILD scores to be used practically in Gibbs-sampling local multiple alignment programs. They can improve the performance of such programs, and remove the need for specifying the width of a pattern sought.
The proper description of protein domains in most cases requires a provision for gaps. We have implemented two relatively simple programs for extending a core ungapped pattern or profile to a gapped local multiple alignment. There are several key elements to our approach. First, the initial maximization of aggregate BILD scores using Gibbs sampling yields a core pattern and pattern length for further refinement. Second, the semi-global alignment of this pattern to the input sequences recognizes the importance of complete occurrences of the pattern. Third, the use of asymmetric affine gap costs (Program 1) recognizes that, with respect to the core pattern, long deletions generally are much more deleterious than long insertions. The placement of gaps can be refined using position-specific gap costs derived from column BILD scores (Program 2). Fourth, greedy alignment allows multiple instances of a pattern to be found within a single sequence. In conjunction with length-dependent gap costs, it discourages alignments spanning more than one instance of a pattern, but can still uncover long insertions. Fifth, iteration permits the core model to be refined, improving the discrimination of true relationships from chance similarities. This strategy, informed by considerations of protein structure, has proved a rapid and effective method for delineating protein families. Although our programs were developed only for research purposes, with the limited goal of testing the impact of BILD scores, their code is available upon request.
We have sought here primarily to describe the construction and potential uses of log-odds scores in the multiple alignment context. However, many avenues for further research, involving the development and benchmarking of complete multiple alignment programs, remain. To what extent can BILD scores improve the accuracy of profile-profile comparison programs? How does Erickson-Sellers semi-global alignment [92], with uniform asymmetric affine gap costs, compare to HMM [80], [81] and other methods [6] in recognizing related sequence in database searches? We look forward to investigating some of these questions.
|
10.1371/journal.pntd.0006260 | Taenia solium, Taenia saginata, Taenia asiatica, their hybrids and other helminthic infections occurring in a neglected tropical diseases' highly endemic area in Lao PDR | Most part of Southeast Asia is considered endemic for human-infecting Taenia tapeworms; Taenia solium, T. saginata, and T. asiatica. However, until now there was no report of the occurrence of human cases of T. asiatica in Lao PDR. This study, conducted in Savannakhet Province, Lao PDR, microscopically examined a total of 470 fecal samples by Kato Katz method and found 86% of people harboring at least one helminth. Hookworms were detected in 56% of the samples besides Opisthorchis like eggs (42%), Trichuris trichiura (27%), Ascaris spp. (14%), and Taenia spp. (4%) eggs. Serology for cysticercosis showed 6.8% positives with results varying from 3% to 14.3% in Ethnic School students and Kalouk Kao village respectively. Species-specific PCR targeting mitochondrial DNA (mtDNA) of 28 tapeworms, recovered from 16 patients, revealed T. solium (n = 2), T. saginata (n = 21), and T. asiatica (n = 5). Two patients were confirmed to be coinfected with T. saginata and T. asiatica, indicating the endemicity of the 3 human Taenia in Lao PDR. However, nucleotide sequencing of a nuclear DNA gene, DNA polymerase delta (pold) revealed that all the tapeworms identified as T. asiatica using mtDNA had T. saginata type allele at pold locus, demonstrating that they are not “pure T. asiatica” but the hybrid descendants between the two species, confirming the wide distribution of hybrids of T. saginata/ T. asiatica in Southeast Asia. The high prevalence of several helminthic NTDs in east Savannakhet area even with conventional control measures indicates the importance to establish wide and multifaceted health programs to sustainably improve the quality of life of the populations living in these communities.
| Southeast Asian Countries are endemic for several foodborne and soil-transmitted helminths occurring in different levels and areas, depending on environmental and cultural conditions. This study aimed to study the soil-transmitted helminths (STHs) and foodborne parasites in Savannakhet Province of Lao PDR, bordering with Vietnam. We found people infected with hookworms, roundworms, whipworms, intestinal/liver flukes, and tapeworms. We also detected antibodies against cysticercosis, an infection caused by eating the eggs of Taenia solium, the pork tapeworm. Focusing on human tapeworm infection, using molecular techniques based on mitochondrial DNA, we detected the three species of human tapeworms T. solium, T. saginata and T. asiatica. Interestingly, when we checked the same material using nuclear gene markers we noted that T. asiatica found in the region were the hybrid descendants of T. saginata and T. asiatica. The causes of hybridization may include people harboring different tapeworm’s species at the same time, allowing the exchange of genetic material but, the consequences of this hybridization are unknown including the seriousness of disease in people. Despite the deworming programs, there is a high prevalence of STHs and foodborne parasites in east Savannakhet area, therefore wide and sustainable health programs are an urgent task to improve the quality of life of the people living in the area.
| Only human are the definitive hosts for Taenia solium, Taenia saginata, and Taenia asiatica, which are referred as the human-Taenia. The distribution of each of the 3 species of human Taenia depends on peoples’ cultural characteristics which involve the consumption of undercooked meat or organs of intermediate hosts infected with viable metacestodes [1–3]. Swine are the intermediate hosts for T. solium and T. asiatica. However, the metacestodes of these species present different tropism: usually muscle and brain for T. solium, and viscera, mainly liver, for T. asiatica [4–6]. Domestic bovine are the main intermediate hosts for T. saginata with cysticerci establishing predominantly in the muscles [7].
Southeast Asia is considered an endemic area for the 3 species of human Taenia with several reports of occurrence in human and animals [8–11]. However, there is no report of the occurrence nor evidence of T. asiatica in human in Lao PDR despite its localization, surrounded by endemic countries [12–14]. Antibody serosurveillance of four provinces in the northern area of Lao PDR in 2011 indicated high frequency of contacts with adult (46.7%) and larval parasites (66.7%) [14]. The existence of T. solium was confirmed by DNA sequencing of copro-PCR positive fecal samples, but no T. saginata or T. asiatica were detected [14]. Furthermore, in a recent study, 15 haplotypes of T. saginata were obtained from 30 isolates from Khammouane Province, central Lao PDR [15]. An extensive study on the prevalence of taeniasis in Lao PDR with whole country coverage reported the presence of mainly T. saginata found in all Lao PDR’s provinces and T. solium in Luang Prabang, northern area [10].
In this study, we report a high prevalence area for foodborne parasites and STHs. Furthermore, we could detect worm carriers of T. solium, T. saginata and T. asiatica by mitochondrial DNA in east Savannakhet Province, suggesting that Lao PDR as an endemic country for the 3 human-Taenia species. Moreover, we verified hybridization of T. saginata and T. asiatica is likely to be occurring in the region.
The study was conducted in Sepon District, Lao PDR in March and December 2013. The area is located in the eastern part of Savannakhet province and is bordering with Quang Tri province of Central Vietnam (Fig 1) and it is covered by subtropical forests in its majority. Ancient human occupation is reported in the actual area of Lao PDR, and according to the last classification, there are 49 different ethnical groups in the country, with more than 4 groups living in Savannakhet area [16]. The participants joined the study on a voluntary basis, from an estimated population reported by the Basic Health Center of Sepon district of 743 people in the study area, 396 (53%) males and 347 (47%) females aged from 3 to 74 years old, and living in the 3 studied villages (Kalouk Kao, Poung, and Ayay Yay) besides the residents of an ethnic college located in Sepon district area, at the eastern Savannakhet province (Fig 1). A detailed explanation of the study was done in the local language for proper understanding. Adult subjects provided written informed consent, and a parent or guardian of any participant child provided informed consent on the child’s behalf and, after approving the informed consent, the participants received instructions for collecting and transporting the fecal samples. Fecal examinations were conducted at the Sepon District Hospital by Kato-Katz modified cellophane thick smear method (KK) [17]. Each slide was examined under a microscope, helminth eggs were counted, and the number of eggs per gram of feces (EPG) was calculated as previously described [18]. Blood sampling was conducted from 235 persons for serological diagnosis of cysticercosis. Feces and serum samples were then brought back to the laboratory in Thailand and Japan for further analysis.
Fecal samples were submitted to KK and molecular procedures. Samples for copro PCR were added sufficient volume of RNAlater stabilization solution (Life Technologies, USA) and brought to the laboratory for Copro PCR analysis. Copro DNA technique was done as previously established [19] with some modification, briefly; fecal samples were homogenized, the volume estimated and 300mg of each sample was used for DNA extraction. The fecal material was disrupted with a μT-12 beads crusher (TAITEC Co., Koshigaya, Japan) using 3 stainless steel beads of 4mm plus 200mg of 0.2mm glass beads in each tube. DNA was extracted from the homogenized solution using the QIAamp DNA Stool Mini Kit (QIAGEN, Hilden, Germany) following the manufacturer’s instructions. Final DNA elution was done in 30 μl of elution buffer.
PCR was conducted using a T100 Thermal Cycler DNA thermocycler (BIO-RAD, Hercules, CA, USA). The reaction was carried out in a final volume of 25 μl containing PCR reagent (TOYOBO, Osaka, Japan) and 1 `l of DNA preparation as template. The DNA samples were initially denatured at 94°C for 4 min, followed by 30 amplification cycles of denaturation at 94°C for 1 min, annealing at 60°C for 30 s, and elongation at 72°C for 2 min. PCR products were electrophoresed on a 2.0% agarose gel with positive samples showing amplicons of the proper size for each Taenia species [19].
The participants presenting Taenia eggs on KK or voluntarily looked-for treatment were given a single oral dose of niclosamide (Yomesan, Bayer AG, Germany) according to the fabricant recommendations, followed by purgation with 60 ml of saturated magnesium sulfate solution. After treatment, attention for worm expulsion was dispensed for all participants, including the people who showed the Taenia eggs in fecal examination or the people who requested treatment even with no positive results in KK exams [20]. Discharged parasites recognizable by naked eye were separated, identified, washed in saline solution (NaCl 0.85% w/v) and preserved individually in 70% ethanol or RNAlater stabilization solution (Life Technologies, USA). The samples were transferred to a laboratory in Japan for further DNA analysis.
The genomic DNA of each ethanol preserved tapeworm was extracted using QIAamp DNA Mini Kit (Qiagen) and subsequently used as a template for polymerase chain reaction (PCR). For the differentiation of three human Taenia species, multiplex PCR was performed as previously described [19], with a slight modification. Briefly, one reverse and four forward primers were used to amplify different sizes of amplicons, specific for the cytochrome c oxidase subunit I (cox1) gene sequences of T. solium Asian genotype, T. solium Afro-American genotype, T. saginata and T. asiatica, respectively. The forward primer specific to T. asiatica was newly designed as TasiMpF (5’- TTA TTT ATT TAC GTC AAT CTT ATT G -3’), instead of the originally used primer.
For some of the tapeworms identified as T. saginata or T. asiatica, nucleotide sequencing of nuclear DNA gene markers was performed to examine whether they are the hybrid descendants of the two Taenia species. Partial sequences of two nuclear genes, ezrin-radixin-moesin-like protein (elp) and DNA polymerase delta (pold), were amplified by PCR and directly sequenced [21,22]. In the case of double peaks in the sequencing electropherogram, PCR products were cloned using pGEM-T vector (Promega) transformed into Escherichia coli DH5α and plated on LB agar containing X-Gal (20mg/ml) and ampicillin (100ug/ml). At least 10 positive clones from each PCR product were used for nucleotide sequence confirmation.
ELISA and immunoblotting using LMWAgs were performed as previously described [23]. Briefly, for ELISA, 100 μl of 1 μg/ml of T. solium LMWAgs in PBS were loaded in 96-well microplates (Maxisorp, Nunc, Copenhagen) overnight at 4°C, blocked with 300 μl of blocking buffer (20 mM Tris–HCl, pH 7.6, 150 mM NaCl, 1.0% casein, 0.1% Tween 20) at 37°C for 1 hour and washed twice with PBS containing 0.1% Tween 20 (PBST). Serum samples diluted 1:100 in blocking buffer, were applied (100 μl/well) in duplicates and incubated at 37°C for 1 h. After washing twice in PBST, the plates were incubated with 100 μl/well of recombinant protein G conjugated with peroxidase (Invitrogen, USA) diluted 1:2000 in blocking buffer at 37°C for 1 h. For color development, the plates were incubated with 100 μl of peroxidase substrate ABTS (2,2’-azino-di(3-ethyl-benzothiazoline-6-sulfonate)) (Sigma-Aldrich) in 0.2 M citric acid buffer (pH 4.7) for 30 min at room temperature. Optical density (OD) was determined at 405 nm for each well using microplate reader (Immuno Mini NJ-2300, Biotec, Japan). The cut-off value was determined as the mean of OD plus 4 standard deviations of sera from 37 healthy donors.
Immunoblot was used for confirmation of the ELISA positive serum samples and serum samples with OD value close to the cut-off value. Briefly, LMWAgs (60 μg/mini gel) in SDS sample buffer (62.5 mM Tris-HCl, pH 6.8, 2.0% SDS, 50 mM dithiothreitol and 10.0% glycerol) were loaded in a 15.0% polyacrylamide gel. The separated proteins were transferred onto a polyvinylidene difluoride (PVDF) membrane sheet (Millipore) and then blocked with blocking buffer. Serum samples diluted 1:20 in blocking buffer were incubated at room temperature for 1 h. After washing 3 times for 5 minutes in blocking buffer, the membranes were incubated with the peroxidase-conjugated recombinant protein G (Invitrogen, USA) diluted 1:2000 in blocking buffer. The substrate (4-Chloro-1-Naphthol/Phosphate) was used for color development at room temperature for 30 min, and the reaction was stopped by washing with water.
All the data collected in paper forms were tabulated and the analyzed in Excel 2016 (Microsoft) and EpiInfo version 7.1.5.0 (Centers for Diseases Control and Prevention, Atlanta, GA, USA). DNA sequencing data were edited and analyzed using MEGA 6 software [24]. In this cross-sectional study, the sample size was determined at a confidence level of 95% with confidence limits of 5% considering the entire population, gender and age calculated with the expected frequencies of cysticercosis and taeniasis. The data were analyzed using descriptive statistics and chi-square test to determine association between the results of the tests used in each category of data. P values <0.05 were considered statistically significant.
This study was approved by the National Ethical Committee for Lao Health Research of the Ministry of Health, Lao PDR (172/NECHR).
The participants of this study that provided fecal samples totalized 470 individuals (259 males and 211 females aging from 5 to 72 years old), corresponding to 63% (470/743) of the total estimated population, with fecal samples and 235 individuals out of 743 (32%) provided blood samples.
The information provided from Sepon Health Center regarding the cultural aspects of the region showed no restrictions for food within the ethnic groups living in the area, with all the analyzed population consuming different types of meat and vegetables including raw or cooked beef and pork meat or viscera and cooked and fresh vegetables mainly produced for subsistence in the local area or acquired from local markets. The residents of the ethnical college were from different ethnical groups (Makong, Tri, Katang, Ta Oi). The students enrolled in this study totalized 66 people with serum and fecal samples, 32% (16) female and 68% (50) male, were aged from 10 to 19 years old. The students preserved their cultural traditions while using the infrastructure provided by the school and frequently went back to their hometown, located in different areas of Sepon district.
Stool samples analysis by KK were performed on 470 fecal samples. Analyses of the results revealed 86% of people harboring at least one species of parasite: 42% (196/470), 31% (146/470), 11% (50/470) and 2% (11/470) harbored 1, 2, 3 and 4 helminth species, respectively. The prevalences of the detected helminth eggs were 56% (265/470) for hookworms, 42% (199/470) for Opisthorchis like eggs, 27% (129/470) for Trichuris trichiura, 14% (66/470) for Ascaris spp., and 4% (19/470) for Taenia spp.
Copro-PCR to detect human Taenia DNA were performed in 163 fecal samples revealing 9.8% (16/163) of T. saginata, 3.1% (5/163) of T. solium and 1.8% (3/163) T. asiatica, indicating the endemicity of the 3 human Taenia in the Sepon district.
Eighteen people received treatment with niclosamide, a total of 28 tapeworms were recovered from 16 taeniasis patients in 4 villages and the ethnic school. Two worms were identified as T. solium and 26 worms appeared to be T. saginata or T. asiatica morphologically. All the specimens were submitted to DNA examination. People receiving treatment for taeniasis and seropositives for cysticercosis by village, tapeworm expulsion and the results of the different diagnostic tests done are shown in Table 1.
Firstly, using multiplex PCR targeting mtDNA cox1 gene, based on mtDNA sequences, 75% (21/28) of the tapeworms were identified as T. saginata, 18% (5/28) were T. asiatica and 7% (2/28) T. solium. Two patients from Kalouk Kao village harbored multiple tapeworms. One patient had a mixed infection with three tapeworms (two T. saginata and one T. asiatica). The other had two worms (one T. saginata and one T. asiatica). T. asiatica was confirmed only in Kalouk Kao village, and T. solium was found in Kalouk Kao village (n = 1) and in the Ethnic School (n = 1). Nuclear DNA analysis was carried out to clarify whether the tapeworms identified as T. asiatica by mtDNA were pure T. asiatica or hybrid descendants between T. saginata and T. asiatica. The tapeworms identified as T. saginata or T. asiatica in Kalouk Kao (n = 9) were used for nuclear DNA study.
Partial sequences of elp and pold genes (1164 bp and 1097 bp) were obtained from all except three samples by direct sequencing. After cloning and sequencing, two different nucleotide sequences were obtained from each of those three samples. In the pold locus, all the alleles obtained from nine tapeworms were T. saginata types (poldA or poldB) as previously found [22], and one worm was heterozygous of pold A/B alleles (Table 2). On the other hand, both T. asiatica type (elpA) and T. saginata type (elpC) alleles [21, 25] were confirmed in elp locus. The genotype of elp locus mostly corresponded to the species identification by mtDNA, however, two worms with T. saginata type mtDNA were heterozygous of elp A/C alleles (Table 2).
Serum samples collected totalized 235 samples, corresponding to 50% of the fecal samples, once not all the individuals that provided fecal samples accepted to provide blood samples.
Results of ELISA confirmed by immunoblot showed the presence of positive cases of cysticercosis in all the villages with a prevalence of 7.2% (17/235) as shown in Fig 2. Considering the results by village, 2 locations, Kalouk Kao and Poung presented higher prevalence with 14.3% (7/49) and 10.7% (6/56) respectively. The serum samples from Ayay Yay village and the Ethnic School students presented lower seropositives to cysticercosis when compared to the other villages (p<0.05) with 2 cysticercosis positives samples each, corresponding to 3.1% (2/64) and 3.0% (2/66) of the studied people respectively.
This is the first report of T. asiatica in Lao PDR, a country with considerable differences in latitude from the south at 13oN in Champassack to Phongsali (Lat.22oN) the northern province, presenting great climatic diversity. The geographic particularities of the country, as the Boulavean plateau and the Mekong basin, created a basis for the development of different ethnicities, with many cultures and eating habits. A total of 49 ethnic groups have been recognized in Lao PDR [16], consequently, as one could expect, parasitic infection prevalence patterns also might differ according to different areas of the country and may explain some parasitic infection particularities as we observed in Sepon, now considered an endemic area for the three human Taenia. The absence of restrictions on food consumption can contribute to the parasite infection in those communities. Additionally, we detected rates as high as 86% of the studied population harboring at least one species of helminth. In general, there is a lack of sanitary infrastructure, toilet access and other issues like no schooling and basic hygiene, factors that would be considered causes for the high infection rates of parasitic infection, as pointed out in Lao rural communities [11].
The villages of Poung and Kalouk Kao, where higher prevalence for all helminths was recorded, presented no structures for sewage treatment and the lowest education level according to the data provided by the Sepon Health Center. The lower prevalence of cysticercosis was found within the students, and it can be due to the access to education and basic infrastructure, as toilet, positive points on the protection for infection as previously reported [11], indicating that basic and general hygiene practices could diminish the infection levels in endemic areas. Another issue that could interfere with the knowledge or understanding of the importance of hygiene habits is the adhesion of the target population to health promotion initiatives, perceived when only 49 of 129 people listed collaborated with this study in Kalouk Kao village. That is, the village with less sanitary conditions presented the lowest adhesion to the health project, as we verified with only 38% of people joining the study. Bringing the interest of more people for health actions is an important issue to be improved in health programs for better results.
The occurrence of cysticercosis in the studied area was detected by serology without additional supportive data. Despite no clinical symptoms of neurocysticercosis were reported by the health center staffs or the villagers, suggesting there might be non-clinical cysticercosis cases, follow-up studies for the confirmation of such cases is an urgent task for early evaluation and treatment.
In the present study, all the three human Taenia were confirmed from taeniasis patients by the conventional PCR method targeting mtDNA. However, all the tapeworms identified as T. asiatica had T. saginata type pold allele, indicating that they are not “pure T. asiatica”. It has been shown that most of the tapeworms identified as T. asiatica using mtDNA have genetic traces of T. saginata in some nuclear DNA loci, and possible “pure T. asiatica” has been confirmed only in Taiwan and Philippines until now [21, 22, 25]. Those tapeworms showing nuclear-mitochondrial discordance are considered to be derived from the hybrid descendants between “pure T. asiatica” and “pure T. saginata”. Although the infection of T. asiatica in humans has been associated with eating raw or undercooked pork viscera, it is uncertain which animals, cattle and/or pigs, are the intermediate hosts of the hybrids. To clarify the host affinity and tissue tropism of T. saginata, T. asiatica, and their hybrids, it is necessary to examine the cysticercus from domestic animals by using both mtDNA and nuclear gene markers.
The reasons for the endemicity of the hybrids between T. asiatica and T. saginata in this region could be the proximity and the commuting of people and goods from Vietnam, a reported endemic area for T. asiatica with several human cases [26, 27]. Ethnic overlapping occurs on all borders of Lao PDR with neighboring countries especially the Austro-Asiatic groups on both sides of the Laos-Vietnam border as well as the cultural behavior of ethnical communities which are overlapping in this Laos-Vietnam border region [16] with habits of eating pork liver in several dishes. Moreover, in the west area of Savannakhet province which is bordering with Thailand, also a known endemic country for T. asiatica [9], there are reports for T. saginata and T. solium only [10, 28] reinforcing the idea of the Laotian T. asiatica or its hybrids origin from Vietnam, though more studies in this issue are necessary.
The causes of hybridization may include people harboring different tapeworm’s species as observed in this (Table 1) and other studies [21, 22, 25]. Multi-species parasitism may facilitate the exchange of genetic material allowing the occurrence of hybrids. However, the consequences of this hybridization as the seriousness of disease in people and even if the hybrid descendants can produce viable generations are unknown. For further studies, to detect hybrid cysticerci in intermediate hosts, confirm hybridization, and its infectivity to intermediate hosts is necessary.
MDA is the recommended strategy of the World Health Organisation to control or eliminate NTDs in endemic areas. It has been implemented widely in Southeast Asian countries and its success is related to the improvement of sanitation and education programs [29]. The Sepon region is included in the MDA programs for elimination of parasitic diseases with treatment using praziquantel. The detection of seropositive individuals for cysticercosis leads to a point of concern in this issue: the occurrence of seizures and other collateral effects after treatment due to the inflammation caused by the sudden damage or death of cysts in the brain [30–33]. Side symptoms after treatment, including seizures, were reported by the population submitted to praziquantel MDA in Lao PDR, leading the people to reject subsequent treatments and the stop of the program. Nowadays only children under 5 years of age will take mebendazole when going for vaccination at primary school under a WHO project (Dr. Pongvongsa personal communication). The strict calculation of doses to be administrated should be considered in prevalent areas for cysticercosis, furthermore, a program for early diagnosis of cysticercosis in asymptomatic patients could be designed in these areas for improvement of MDA as accidental death after treatment may occur due to the extensive use of praziquantel in MDA of Asian countries [34].
Another issue in this endemic area for 3 species of human Taenia is to detect worm carriers. As we could observe in the results of worm recovering after treatment, the number of worm carriers would be higher if we combine the results of KK, Copro-PCR, and self-detection (Table 1). Unfortunately, Copro-PCR was not done in the field, so its results were not used for treatment protocols. Therefore, copro-PCR is a feasible option to be done in province’s central laboratories/health centers, where a safe treatment can be prescribed and conducted. In this study, we could note self-diagnosis as an educational alternative for detection of worm carriers; after the explanation of the study and the description of the fecal aspects and symptoms, 4 individuals who could not supply fecal samples, voluntarily came to expel worms, with 100% of worm expulsion (Table 1). This could be an excellent method for detection of worm carriers with a high rate of success as reported in a Mexican endemic area for T. solium [35].
Differently from the other studies in the central, north and northeastern areas of Lao PDR, where cases of T. asiatica were not found [12, 14, 15], in the east part of Savannakhet province (Fig 1) we could detect the presence of 3 species of human Taenia species in Sepon district that border Vietnam. The prevalence of Taenia eggs in KK was not high as presented by Okello et al. [14] which detected percentages as high as 46%. This result raises suspicion that other "hotspots" of T. solium hyper endemicity may exist in some regions, particularly in communities where the consumption of raw or undercooked pork is common, associated with the lack of health education.
Regarding other helminths observed in this study, the high prevalence of Opisthorchis like eggs (43%) is also a point of concern once Opisthorchis viverrini infection is a major cause of cholangiocarcinoma in endemic areas [36]. O. viverrini is a fishborne fluke and, like Taenia and its infection to humans is related to the consumption of raw fish meat (cyprinid fish) [37] and to domestic dogs, natural reservoirs of O. viverrini, which can be an important source of aquatic environment contamination because of its routine behavior to assess water sources [38]. Moreover, a high prevalence of hookworm (57%) was found, a parasitic disease where dogs play an important role [38, 39]. Dog meat is eaten in Asian countries including Lao PDR. However, considering the dog as an intermediate host of T. solium [40], dogs survey for cysticercosis in addition to pigs’ survey may also be important to screen the risk factor for human infections. More studies on the ecological aspects of NTDs, as carried out in other localities including checking reservoir animals and using environmental DNA detection [38, 41], would be an interesting way to determine the level of exposure of the people living in endemic areas to agents causing diseases like O. viverrini, hookworms and other STHs found in this study as Ascaris and Trichuris.
This study revealed a highly endemic area for helminthic diseases in east Savannakhet, Lao PDR including the high occurrence of STHs and foodborne trematodes. The existence of T. solium in Savannakhet province was confirmed in this study, moreover hybrids descendants between T. saginata and T. asiatica were detected, indicating the presence of 3 human-Taenia species in Lao PDR. The situation points out the importance of establishing new, wide and multifaceted health program to sustainably improve the quality of life of the populations living in these communities.
|
10.1371/journal.pntd.0000791 | A Major Role for Mammals in the Ecology of Mycobacterium ulcerans | Mycobacterium ulcerans is the causative agent of Buruli ulcer (BU), a destructive skin disease found predominantly in sub-Saharan Africa and south-eastern Australia. The precise mode(s) of transmission and environmental reservoir(s) remain unknown, but several studies have explored the role of aquatic invertebrate species. The purpose of this study was to investigate the environmental distribution of M. ulcerans in south-eastern Australia.
A range of environmental samples was collected from Point Lonsdale (a small coastal town southwest of Melbourne, Australia, endemic for BU) and from areas with fewer or no reported incident cases of BU. Mycobacterium ulcerans DNA was detected at low levels by real-time PCR in soil, sediment, water residue, aquatic plant biofilm and terrestrial vegetation collected in Point Lonsdale. Higher levels of M. ulcerans DNA were detected in the faeces of common ringtail (Pseudocheirus peregrinus) and common brushtail (Trichosurus vulpecula) possums. Systematic testing of possum faeces revealed that M. ulcerans DNA could be detected in 41% of faecal samples collected in Point Lonsdale compared with less than 1% of faecal samples collected from non-endemic areas (p<0.0001). Capture and clinical examination of live possums in Point Lonsdale validated the accuracy of the predictive value of the faecal surveys by revealing that 38% of ringtail possums and 24% of brushtail possums had laboratory-confirmed M. ulcerans skin lesions and/or M. ulcerans PCR positive faeces. Whole genome sequencing revealed an extremely close genetic relationship between human and possum M. ulcerans isolates.
The prevailing wisdom is that M. ulcerans is an aquatic pathogen and that BU is acquired by contact with certain aquatic environments (swamps, slow-flowing water). Now, after 70 years of research, we propose a transmission model for BU in which terrestrial mammals are implicated as reservoirs for M. ulcerans.
| Mycobacterium ulcerans is the causative agent of Buruli ulcer (BU), a destructive skin disease found predominantly in sub-Saharan Africa and south-eastern Australia. The mode of transmission and environmental reservoir remain unknown, but several studies have explored the role of aquatic insects, such as water bugs, and biting insects, such as mosquitoes. In the present study we investigated possible environmental source(s) of M. ulcerans in Victoria, Australia. Our results revealed that although M. ulcerans DNA could be detected at low levels in a variety of environmental samples, the highest concentrations of M. ulcerans DNA were found in the faeces of two species of possums, common ringtails and common brushtails. Possums are small arboreal marsupial mammals, native to Australia, and these particular species occur in both urban and rural areas. Examination and sampling of live captured possums in an area endemic for BU revealed that 38% of ringtail possums and 24% of brushtail possums, respectively, had laboratory-confirmed M. ulcerans lesions and/or M. ulcerans PCR-positive faeces. The finding that large numbers of possums in a BU-endemic area are infected with M. ulcerans raises the possibility that mammals are an environmental reservoir for M. ulcerans.
| Buruli ulcer (BU) is caused by the environmental mycobacterium, Mycobacterium ulcerans. Infection with M. ulcerans often leads to extensive necrosis of the skin and soft tissue with the formation of large ulcers, usually on the leg or arm, due to the production of the destructive polyketide toxin, mycolactone [1]. Although rarely fatal, BU causes serious morbidity and frequently results in permanent disability [2]. The disease has been reported in more than 30 countries worldwide; however, cases mainly occur in regions with tropical and subtropical climates. The majority of cases are found in West and sub-Saharan Africa. Cases of BU often cluster around particular water bodies and are highly focally distributed, with endemic and non-endemic communities often separated by only a few kilometres [2].
Australia is the only developed country reporting significant local transmission of M. ulcerans. In 1948, a cluster of cases linked to the Bairnsdale region in Gippsland was described by McCallum et al. [3]. Since then, foci of infection have been reported in tropical far north Queensland [4] and temperate coastal Victoria, where there have been several outbreaks over the past two decades: Phillip Island (1992–1995), the Frankston/Langwarrin region (1990–1997), St Leonards (2001–2002) and Point Lonsdale (2002-present) (Fig. 1) [5], [6]. The present outbreak in Point Lonsdale, a small coastal town approximately 60 km south-west of the Victorian capital Melbourne, is the largest on record in Australia, with over 100 laboratory-confirmed cases diagnosed since 2002. Geographically, the town is close to sea level, and there are several natural and man-made swamps and water features in the area [6]. Cases of BU have also been described in both native wildlife and domestic mammal species in Victoria, including koalas (Phascolarctos cinereus) [7], common ringtail possums (Pseudocheirus peregrinus) [8], a mountain brushtail possum (Trichosurus cunninghami), a long-footed potoroo (Potorous longipes) (J. Fyfe, unpublished), two horses [9], two dogs (O'Brien et al., manuscript in preparation), an alpaca [8] and a cat [10]. All animal cases were identified in locations where human cases of BU have been reported.
The precise mode(s) of transmission and environmental reservoir(s) of BU are unresolved and continue to be the subject of intense research. Proximity to marshes and wetlands is a recognised risk factor for infection and several studies have explored the role of aquatic invertebrate species as potential vectors and/or reservoirs [6], [11]–[13]. Detection of M. ulcerans in environmental samples is mainly achieved using PCR, as culturing M. ulcerans directly from the environment is extremely difficult [14]. In Australia, M. ulcerans DNA was detected in water and detritus from swamps during the outbreak of BU on Phillip Island in the mid-1990s [15], [16] and more recently in five species of mosquitoes (Aedes sp., Coquillettidia sp. and Culex sp.) captured from Point Lonsdale (infection rate, 4.3/1,000 mosquitoes) [6]. In West Africa, M. ulcerans DNA has been detected in water and aquatic plants [17], insects (Belastomatidae, Naucoridae, Hydrophilidae), crustaceans and molluscs (Bulinus sp. and Planorbis sp.) and small fish (including Tilapia sp.) [11], [13], [18]–[21]. Recent studies of the distribution of M. ulcerans in aquatic sites in Ghana found evidence of M. ulcerans DNA in insects, water filtrate, biofilm and soil [12], [13]. In 2008, Portaels et al. described, for the first time, the cultivation and characterisation of an M. ulcerans strain obtained from an aquatic Hemiptera (common name Water Strider, Gerris sp.) from Benin [14].
Analysis of the whole genome sequence of M. ulcerans has provided further insights into the elusive environmental reservoir and mode of transmission [22]. Complete sequencing of an M. ulcerans strain isolated from a patient in Ghana revealed a 5,631,606 bp circular chromosome with 4160 genes, 771 pseudogenes and a 174,155 bp virulence plasmid pMUM001 that is required for the production of mycolactone [23], [24]. Comparison of the M. ulcerans genome with the genome of M. marinum confirmed the very close relationship between these species; however, it also revealed that there are some striking differences, mostly due to the presence of the plasmid pMUM001 and the many chromosomal deletions and rearrangements that have occurred in M. ulcerans [23]. It is therefore likely that M. ulcerans has evolved from an M. marinum-like ancestor by lateral gene transfer and reductive evolution, through the acquisition of a pMUM001-like plasmid, expansion of the two high copy number insertion sequence elements IS2404 and IS2606, extensive gene disintegration (formation of pseudogenes), genome rearrangements and DNA deletion. These characteristics suggest that M. ulcerans has recently passed through a so-called “evolutionary bottleneck” and is adapting to a new, niche environment.
In this study, we investigated potential environmental reservoirs of M. ulcerans in south-eastern Australia with the aim of developing a more comprehensive model of its life cycle and mode of transmission. Specifically, using semi-quantitative real-time PCR and culture to test for the presence of M. ulcerans, we investigated a range of potential abiotic and biotic reservoirs (selected using emerging information in the literature and our own ongoing field based research) in areas of varying BU endemicity. Our findings have led us to propose that M. ulcerans is able to infect small mammals, survive and potentially replicate within their gastrointestinal tracts and raises the possibility that mammals play a major role in the ecology of M. ulcerans.
A case of BU was defined as a human patient with a suggestive clinical lesion from which M. ulcerans was identified by PCR [26] or culture from January 2005 to December 2009 inclusive. The likely geographic origin of infection was determined on the basis of the patient's residential address and/or travel history. A patient was considered as having acquired BU from a particular geographic area if he/she was a resident of, or a visitor to, that area and had not reported recent contact with any other known BU endemic area. Due to the large seasonal fluctuations in the population of endemic areas (most of which are summer holiday destinations), and the difficulty in estimating the number of visitors to a particular area, the average annual incidence of BU in each geographic area over the five-year study period was calculated by dividing the average annual number of cases in residents only (that is, cases in visitors were excluded) by the resident population of the specified geographic area. Resident population numbers were obtained using Australian Bureau of Statistics data derived from the 2006 Census of Population and Housing [33].
Statistical analyses were performed using STATA version 10.0 (STATA Corporation, College Station, TX). Proportions were compared using the two-sample test of proportion.
Testing of environmental samples commenced in mid-2004, just prior to the peak of the Point Lonsdale outbreak. The initial focus was low-lying, wet areas in which mosquitoes were likely to breed, such as drains, soak pits (covered concrete pits into which storm water and street runoff flows and sits until it gradually seeps into the ground), man-made lakes and natural water bodies. In Point Lonsdale, low levels of M. ulcerans DNA (that is, weak positive real-time PCR signals for IS2404, IS2606 and KR) were detected in sediment from a man-made lake; soil, sediment and detritus from a number of different soak pits and drains; biofilm; aquatic plants; and residue from filtered water (Table 1). The estimated bacterial loads for these samples ranged from 10–100 organisms/ml for residue from filtered water and 103–104 organisms/gram for biofilm. In contrast, only four samples (two soil and two vegetation) from low endemicity areas were positive for M. ulcerans DNA (Table 1).
In late 2006, the scope of our environmental testing expanded to samples in dryer areas at higher elevations, including leaf litter, leaves, tree bark, flowers, seeds, stems and faeces from brushtail possums (Table 1). The rationale for this was: (i) soil collected outside drains had previously tested positive for M. ulcerans DNA, (ii) BU patients have reported an association between small penetrating injuries, sustained from vegetation, and subsequent ulcers [34], and (iii) cases of BU are known to occur in arboreal marsupial mammals, including koalas [7] and ringtail possums [8]. Testing revealed that while M. ulcerans DNA could be detected at low levels in some samples of leaf litter and bark from trees (estimated bacterial load 102–103 organisms/gram), much higher levels of M. ulcerans DNA were detected in brushtail possum faeces (estimated bacterial load ≥106 organisms/gram). This important discovery led to the large scale, systematic testing of possum faeces in Point Lonsdale, as well as low and non-endemic sites.
Over a two-year period (2007–09), systematic collection of faeces from brushtail and ringtail possums was carried out across Point Lonsdale, nearby low endemicity areas and non-endemic areas (Table 2). A total of 589 faecal samples from ringtail possums and 250 samples from brushtail possums were tested. The difference in the number of samples collected from each geographic location and from each species reflected the relative population densities, with ringtail possums being much more abundant than brushtail possums in many areas sampled (K. Handasyde and A. Legione, unpublished data).
In Point Lonsdale, M. ulcerans DNA (IS2404) was detected in 43% of ringtail possum and 29% of brushtail possum faecal samples (Table 2). All samples tested for the presence of IS2606 and KR were PCR-positive for these additional targets. Furthermore, the ΔCt (IS2404-IS2606) was always in the range expected for M. ulcerans (2.17–2.79), rather than another MPM (6.94–8.07) [26]. The median estimated bacterial load was 104 organisms/gram (range: 102–108 organisms/gram) for ringtail possums (Fig. 3), with 17% of positive samples having an estimated bacterial load >106 organisms/gram. The median estimated bacterial load for brushtail possum faeces was 102–103 organisms/gram (range: 102–106 organisms/gram).
In low endemicity areas, the proportion of PCR-positive faecal samples varied by location. For example, in Barwon Heads, where 15 human cases of BU have been reported since 2005, the proportion of positive ringtail and brushtail faecal samples was relatively high (26% and 19% respectively) compared with the other locations where fewer cases of BU have been reported (Table 2). The median estimated bacterial load of positive faecal samples from low endemicity areas also varied. In Barwon Heads the median estimated bacterial load for ringtail possum faeces was 104 organisms/gram (with 16% of the positive samples having an estimated bacterial load >106 organisms/gram). As in Point Lonsdale, the estimated bacterial load of the positive brushtail possum faeces in Barwon Heads was generally lower than for the ringtail possum faeces, with a median estimate of 102–103 organisms/gram. Similarly low M. ulcerans bacterial loads of 102–103 organisms/gram were estimated for faeces (ringtail possum only) collected in Queenscliff [Fig. 3] and Phillip Island. Only one sample collected from a non-endemic area (Torquay) was positive for M. ulcerans DNA and the estimated bacterial load of this sample was low (102–103 organisms/gram).
Mapping of the samples collected in Point Lonsdale revealed that M. ulcerans DNA could be detected throughout Point Lonsdale and did not appear to be concentrated in one particular area or limited to one particular point source (Fig. 3). However, in Barwon Heads, positive faecal samples were only detected in the southern part of the town (data not shown). No seasonal trends were observed, with the number of positive samples, and the estimated bacterial loads of those samples, consistent between summer, autumn, winter and spring (data not shown).
All attempts at culturing M. ulcerans from possum faeces were unsuccessful. PCR-positive and PCR-negative possum faeces were inoculated into MGIT and onto Brown and Buckle and 7H10 slopes with antibiotics. The MGIT broths and Brown and Buckle slopes exhibited extensive fungal contamination after two weeks and were discarded. Despite the absence of fungal contamination on the 7H10 slopes, no growth of M. ulcerans was detected after 16 weeks incubation.
Over a 20-month period from February 2008 to November 2009, 42 ringtail possums and 21 brushtail possums were captured in Point Lonsdale and examined for BU disease. Among the ringtail possum cohort, 16 (38%) animals had laboratory-confirmed (PCR ± culture) M. ulcerans lesions and/or M. ulcerans PCR-positive faeces. Of the 11 animals with BU disease, nine had M. ulcerans PCR-positive faeces, one had M. ulcerans PCR-negative faeces and we were unsuccessful in collecting a faecal sample from the remaining animal (Table 3). Notably, five of the ringtail possums that did not have BU skin lesions had M. ulcerans PCR-positive faeces. Interestingly, as shown in Table 3, there was little difference in the median estimated bacterial loads of faeces from animals with BU skin lesions and animals without BU lesions. The incidence of M. ulcerans infection among the 21 brushtail possums was lower. One animal had a BU skin lesion and M. ulcerans PCR-positive faeces (estimated bacterial load, 103–104 organisms/gram) and four animals without BU lesions were found to be shedding low levels of M. ulcerans DNA in their faeces (102 organisms/gram) (Table 3).
The most common site for BU lesions was the tail (Fig. 2E). Amongst the 12 possums with BU disease, nine had lesions on the tail and four had lesions on the toe/foot (Table 4). Five of the ringtail possums had multiple lesions, with one animal having severe ulcerative and oedematous lesions on her nose (Fig. 2F), left upper lip, both fore paws, right hock, left hind leg and tail. Three of these animals were euthanased and full necropsies performed to determine the extent of the M. ulcerans infection. The results of these necropsies, along with the results of the other clinical samples taken from all 63 possums captured (including blood, buccal swabs and nasal swabs and urine), are described in a separate report (manuscript in preparation).
The two multiplex real-time PCR assays used in this study to detect M. ulcerans in environmental samples distinguish between M. ulcerans and other MPM that also harbour IS2404 and IS2606 [26]. However, we also sought to determine whether the DNA detected in environmental samples was from the same strain of M. ulcerans that causes disease in humans in Victoria. PCR reactions for 10 VNTR loci and three MIRU loci were performed on a subset of DNA extracts from possum faeces (estimated bacterial load 105–106 organisms/gram), aquatic plant biofilm (estimated bacterial load 103–104 organisms/gram) and water filters (estimated bacterial load 103–104 organisms/filter). The concentration of M. ulcerans DNA in the other sample types (for example, soil) has previously been shown to be insufficient for PCR amplification of these single copy loci [35]. DNA extracted from the possum faeces generated PCR amplicons of the same size (Fig. 4) and sequence as the Victorian human outbreak strain at all loci. As predicted by the lower concentration of M. ulcerans DNA in the samples, DNA extracts from the aquatic plant biofilm and water filter generated PCR amplicons at one locus only (VNTR locus 6 and 19, respectively). In each case the sequence was identical to the Victorian outbreak strain. These data provide evidence that the strain of M. ulcerans detected in these samples is the same as the strain which causes disease in humans in this region. The results also confirm that this method of analysis can only be applied successfully to samples (clinical or environmental) with an estimated M. ulcerans load of ≥105 organisms/gram and should only be used as a confirmatory/epidemiological tool and not as the primary method by which all environmental samples are screened for the presence of M. ulcerans DNA [35].
Illumina high-throughput short-read sequencing was used to compare the genome of an M. ulcerans isolate from a ringtail possum captured in Point Lonsdale (M. ulcerans JKD8170) and a human clinical isolate from Point Lonsdale (M. ulcerans JKD8049) obtained during the period of the M. ulcerans outbreak. This process generated 31,028,581 reads for JKD8170 and 10,921,914 reads for JKD8049. Bioinformatic analysis involved read mapping to the reference genome M. ulcerans Agy99 and reciprocal comparisons to consensus sequences derived from de novo sequence assemblies of each data set. These analyses revealed that both the possum and human isolates shared 5455 SNP differences compared to the reference genome (an African strain) but were differentiated from each other by only two SNPs (confirmed by PCR and Sanger DNA sequencing) across 5.6 Mb of chromosomal DNA sequence. These data confirm the extremely close genetic relationship between the human and possum isolates.
Elucidation of the mode of transmission and environmental reservoir(s) of M. ulcerans is essential for the development of strategies to control and prevent BU outbreaks. Early epidemiological studies from Uganda in the 1970s suggested that M. ulcerans may be associated with certain grasses growing at the edges of permanent swamps [36], [37], and that transmission to humans was via contact with this environmental source. However, attempts to culture M. ulcerans from a range of plants were unsuccessful [38]. The possible role of rodents in the ecology of M. ulcerans was also considered over 30 years ago [39], however the presence of the organism in the organs of 700 animals from a BU endemic area in Uganda could not be confirmed by culture. The development of IS2404 PCR in the 1990s [16], [40] enabled researchers to detect the DNA of M. ulcerans and other MPM in a range of different samples, leading to a renewed search for the environmental reservoir(s). The PCR detection of M. ulcerans DNA in waterbugs from Benin and Ghana [18] and subsequent culture of M. ulcerans from a waterbug [14], focussed the search to aquatic habitats. Currently, the prevailing dogma is that the environmental reservoir of M. ulcerans is an abiotic or biotic component of aquatic, rather than terrestrial, ecosystems. Indeed, numerous epidemiological and environmental studies support this view [5], [11], [12], [14], [15], [17]–[21], [26], [41]–[43], including some of the data from our current study. We found that M. ulcerans could be detected in various aquatic samples including aquatic plants, biofilm and residue from filtered water (Table 1). The major strength of our study, however, was the use of a suite of real-time PCR assays targeting multiple regions in the M. ulcerans genome which, in addition to being highly sensitive, specific and less prone to contamination than conventional gel-based PCR [12], [13], enabled us to estimate the relative numbers of M. ulcerans in the various samples tested by determining the relative concentrations of M. ulcerans DNA among the different sample types.
By following this gradient of M. ulcerans DNA, we discovered that the faeces of two marsupial mammals (ringtail and brushtail possums), contained higher concentrations of M. ulcerans DNA than the other samples tested. The large-scale testing of possum faeces in BU high-, low- and non-endemic sites, and the subsequent capture and examination of possums in Point Lonsdale, generated a number of important findings. Firstly, we discovered that there is a high density of ringtail possums throughout Point Lonsdale that are excreting copious amounts of faeces, almost half of which are estimated to contain M. ulcerans, into the environment (Table 2, Fig. 3). Secondly, we observed a strong positive correlation between the BU endemicity of an area and the proportion and DNA concentration of M. ulcerans-positive possum faeces, with 41% of faecal samples collected in Point Lonsdale testing positive for M. ulcerans compared with less than 1% of faecal samples collected from non-endemic areas (p<0.0001). Similar results were obtained in Benin with a correlation between BU endemicity in patients and environmental results. Environmental studies detected variations in M. ulcerans DNA positivity rates of aquatic insects over time, and these changes were reflected in corresponding alterations of frequency of BU patients in the same foci [44]. Thirdly, 38% of captured ringtail possums and 24% of captured brushtail possums were found to have laboratory-confirmed M. ulcerans skin lesions, mostly on the tail or feet, and/or M. ulcerans PCR positive faeces (Table 3). One explanation for the observation that most lesions occurred on the extremities is that these sites have lower temperatures favouring the growth of M. ulcerans. Another possibility is that, because these sites have less fur, they are more susceptible to insect bites or skin trauma via contact with vegetation or fighting with other possums, which may lead to inoculation of M. ulcerans. Fourthly, we observed that five of the 14 ringtail possums, and four of the five brushtail possums, that were shedding M. ulcerans DNA in their faeces did not have BU skin lesions, indicating that the presence of M. ulcerans DNA in faeces is not limited to clinically diseased animals (Table 3). However, we noted that animals with multiple lesions tended to have higher estimated faecal loads of M. ulcerans than animals with single lesions (data not shown). Finally, whole genome sequencing confirmed the extremely close genetic relationship between the human and possum isolates.
Taken together, these findings suggest that possums may be an environmental reservoir for M. ulcerans in south-eastern Australia. If so, the biology of possums prompts a new interpretation/understanding of the life cycle of M. ulcerans. In particular, ringtail possums are exclusively arboreal, feeding on a variety of leaves of both native and introduced plants, as well as flowers and fruits [45], hence are unlikely to be exposed to M. ulcerans in soil or water. They are also caecotrophic. Caecotrophy is the ingestion of soft faeces of high nutritive value derived from caecal contents and is a critical factor in the ringtail possum's ability to utilise eucalypt foliage as a whole or major food source [46]. This behaviour may also favour gastrointestinal persistence of M. ulcerans. Brushtail possums are semi-arboreal, spending a considerable portion of their foraging time on the ground and, although mainly folivorous, have a more varied diet than ringtail possums [45]. The ecology of these species, which occur in strictly terrestrial habitats, contradicts the idea that the environmental host(s) of M. ulcerans are likely to reside primarily in aquatic environments, although the presence of M. ulcerans in aquatic habitats within the same location is also likely, based on data presented in this study. Thus, in light of our data, we suggest that reservoir species could include terrestrial mammals, and that the association of the disease with low-lying, wetter areas might be driven by the dependence of a vector species (such as mosquitoes [47]) on moist habitats.
A disease reservoir may be defined as: “one or more epidemiologically connected populations or environments in which a pathogen can be permanently maintained and from which infection can be transmitted to the target population. Populations in a reservoir may be the same or a different species as the target and may include vector species” [48]. Our findings from Point Lonsdale suggest that at least one free-ranging mammal species (the ringtail possum), which can be very abundant in urban environments, forms part of a transmission cycle (Fig. 5) for M. ulcerans that could explain human outbreaks of BU in south-eastern Australia, although they may not necessarily be true maintenance hosts (that is, be able to maintain the organism in the absence of other environmental sources). However, bovine tuberculosis, caused by Mycobacterium bovis, and Johne's disease, caused by Mycobacterium avium subsp. paratuberculosis, are both maintained in wildlife reservoir species. In the United Kingdom, badgers (Meles meles) contribute to the spread of M. bovis between herds of cattle [49]. In New Zealand, where bovine tuberculosis is a major problem, the principle wildlife host for M. bovis is the common brushtail possum, which was originally imported from Australia and now occurs at such a high population density that it is a major agricultural and conservation pest [49].
The way in which M. ulcerans might be transmitted from an animal to humans is not clear. A similar epidemiology to leptospirosis, the most common zoonosis worldwide [50], in which rodents are reservoirs but the disease is acquired by contact with contaminated water, should be considered. We envisage that the transmission pathway for M. ulcerans may involve vegetation, vertebrate hosts and invertebrate vectors in both terrestrial and aquatic ecosystems (Fig. 5). Such a model represents a fundamental change to the existing views on the ecology of M. ulcerans, although the idea that M. ulcerans is not confined to low-lying swampy areas is not new [36]–[39], [51], [52]. While we lack important information about whether mosquitoes are productive or simply mechanical vectors, and have only limited information on the site of carriage/colonisation, either on or within mosquitoes, a number of lines of evidence implicate mosquitoes as vectors of M. ulcerans in Victoria [6], [53]–[55]. Given that we found active M. ulcerans lesions in 26% of captured ringtail possums, transmission to humans might occur when an adult mosquito that has fed on a diseased possum, or rested on vegetation contaminated by a possum lesion, subsequently bites a human. Another possibility is that heavy environmental contamination with possum faeces containing M. ulcerans would enable mosquitoes (either as larvae or adults) to come into contact with M. ulcerans, in contaminated soil/water in roof gutters or drains (Fig. 5). This is supported by a study by Tobias et al. which showed that, in a feeding experiment where mosquito larvae were fed possum faecal material spiked with M. ulcerans or M. marinum, M. ulcerans accumulated within the mouth and midgut whereas M. marinum did not [55].
Key to determining which of these potential routes of transmission is most likely (or possible) is the demonstration of viable M. ulcerans organisms in possum faeces. We acknowledge that the detection of M. ulcerans DNA in possum faeces does not necessarily indicate the presence of viable organisms. However we, like many others who have attempted to culture M. ulcerans from environmental samples [14], have currently been unable to culture M. ulcerans from possum faeces. This was despite the fact that some of the samples had real-time PCR signals equivalent to those obtained for the lesion swabs from which culture of M. ulcerans was successful (data not shown). We believe that this has been largely due to the presence of fungi or fungal spores in the faecal samples which, despite decontamination methods, rapidly grew in broth cultures and on Brown and Buckle slopes and inhibited the growth of slower growing organisms such as M. ulcerans. Furthermore, on the basis of subsequent real-time PCR studies, it has become evident that the organisms are tightly associated with the particulate matter and that homogenising faeces in bead bottles results in very few bacteria in the suspension that would normally be used to inoculate the culture media (C. O'Brien, unpublished). We have also found that intact DNA can be recovered from possum faeces many months after sampling and that DNase treatment of the faecal homogenate does not lead to a reduction in the PCR signal (data not shown). This suggests that intact M. ulcerans organisms are present (though not necessarily viable), rather than just free M. ulcerans DNA.
There is also the question of whether mammals could act as reservoirs in sub-Saharan Africa, where the majority of BU cases occur. Recent studies in Ghana failed to detect M. ulcerans in the organs or faeces of rodents and shrews [17], [56]. However these authors did not reject the hypothesis that these, or other species of small terrestrial mammals, may be part of the reservoir of M. ulcerans in this setting. Recent work conducted by our group, including the post-mortem examination of ringtail possums and rats (Rattus rattus) with and without clinical BU disease, has shown that M. ulcerans can be present in the gastrointestinal tracts of animals but not in the organs of the same individual (manuscript in preparation). We are currently investigating the potential role of other mammal species as hosts for M. ulcerans in the Australian setting.
This study has led to a major a shift in our understanding of the environmental distribution of M. ulcerans in south-eastern Australia. It is hoped that the results presented here, along with our continuing laboratory and field research, will take us closer to elucidating the mode of transmission and environmental reservoir(s) of M. ulcerans and in turn the development of strategies to control and prevent this important yet often neglected human disease.
|
10.1371/journal.pcbi.1000966 | Biosensor Approach to Psychopathology Classification | We used a multi-round, two-party exchange game in which a healthy subject played a subject diagnosed with a DSM-IV (Diagnostic and Statistics Manual-IV) disorder, and applied a Bayesian clustering approach to the behavior exhibited by the healthy subject. The goal was to characterize quantitatively the style of play elicited in the healthy subject (the proposer) by their DSM-diagnosed partner (the responder). The approach exploits the dynamics of the behavior elicited in the healthy proposer as a biosensor for cognitive features that characterize the psychopathology group at the other side of the interaction. Using a large cohort of subjects (n = 574), we found statistically significant clustering of proposers' behavior overlapping with a range of DSM-IV disorders including autism spectrum disorder, borderline personality disorder, attention deficit hyperactivity disorder, and major depressive disorder. To further validate these results, we developed a computer agent to replace the human subject in the proposer role (the biosensor) and show that it can also detect these same four DSM-defined disorders. These results suggest that the highly developed social sensitivities that humans bring to a two-party social exchange can be exploited and automated to detect important psychopathologies, using an interpersonal behavioral probe not directly related to the defining diagnostic criteria.
| Human social interaction is exquisitely complex, and perturbed social interaction is a hallmark of psychological pathogy. When someone has a psychological disorder the focus is generally on their behavior, but this behavior is rarely something displayed in isolation and typically induces profound changes in the people interacting with the disturbed individual. In this work we asked if the behavior of one person in a simple two-person economic exchange game is sensitive to features that could classify the pathology of their partner. We analyzed a large group of previously recorded interactions involving healthy persons and people diagnosed with a variety of psychological disorders, and found that a healthy person's behavior is indeed quantitatively and systematically influenced by their partner's pathology. These results could ultimately lead to a different way of understanding and diagnosing psychological disease.
| Social interactions among humans reflect the execution of some of the most important and complex behavioral software with which humans are endowed. Consequently, we should expect the computations involved in human social exchange to be subtle and perhaps even difficult to expose and study in controlled settings. However, exposing these computations is crucial if we are to improve our characterization and understanding of normal human cognitive function and dysfunction.
In recent years, the components of social exchange in healthy subjects have been probed using interactive economic exchange games [1]–[8]. These games typically involve two subjects interacting for one or multiple rounds through the exchange of monetary gestures to one another. For our purposes here, these games require three classes of computation be intact and functioning in the minds of the interacting subjects. They require that each subject can (1) compute norms for what is fair in each exchange, (2) detect deviations in monetary gestures that deviate from these norms, and (3) choose actions predicated on such deviations [9]–[15]. These experimental probes have been used previously in the area of behavioral economics and neuroeconomics, but here we show that the behavioral gestures elicited in the context of economic exchange games can be used to classify certain psychopathologies. The twist in our effort here is that we use a data-driven approach examining the reactions of the healthy partner as a kind of biosensor while playing an exchange game with a subject possessing a psychopathology.
In this paper, we used a multi-round fairness game played by pairs (“dyads”) of interacting humans to extract behavioral phenotypes defined by the dynamics of play exhibited over the 10 rounds of a complete game [6], [7], [16]. The game we employ is called a trust game [17]–[19]; see Figure 1A. In the 10-round trust game, one player (called the investor or the proposer) is endowed with 20 monetary units and chooses to send some fraction to their partner (called the trustee or the responder). The amount sent is tripled to on the way to the trustee. The trustee decides which fraction to return in response to the investor, thus each round is represented by two numbers: the investment fraction and the repayment fraction . All the rules are transparent to both players. The game is played for 10 rounds and the repeated exchanges allow the players to build models of what to expect from their partner providing that their capacity to sense, model, and respond to their partner's decisions is intact.
In most of the dyads, the subjects were given no information about their partner and did not meet or speak to the partner before, during, or after the task. Following [16], we also included “personal” dyads, in which the partners met before the task, were instructed together, and saw a picture of their partner during each round.
The basic approach of this paper derives from our prior work showing that this same game elicits unique behavioral phenotypes when a game is played between a healthy investor and a trustee diagnosed with a range of DSM-defined disorders – Autism Spectrum Disorder (ASD) [20], Borderline Personality Disorder (BPD) [6], Major Depressive Disorder (MDD), and Attention Deficit Hyperactivity Disorder (ADHD) [20]. In all these studies, we noticed that the behavioral differences affect not only the trustee, but also a healthy investor who plays with this trustee.
A similar conclusion that a healthy subject is sensing the psychological nature of the opponent during play was obtained in a recent paper [21], where it was shown that a subject can gauge the strategic sophistication of the opponent in repeated play of a complex stag hunt game.
These observations suggested the hypothesis that the healthy (or typical) investor's behavior might be used to ‘read out’ features that could characterize the psychopathology group playing in the trustee role. This possibility was also suggested by the nature of the interpersonal interaction enforced by the game. In any multi-round interaction with another human, a player's choices are rather dramatically entangled with those of her partner. In addition, although the game is characterized by two numbers per exchange (investment and repayment ratio), it does require players to have several cognitive capacities intact to accomplish a ‘normal’ exchange. These include short-term and working memory, sufficiently accurate models of what to expect from another human in this exchange, appropriate sensitivity to positive and negative social signals, and intact capacity to respond to such signals. Collectively, these observations support the basic hypothesis that humans bring highly developed social sensitivities to two-party interactions that might be profitably exploited as a biological “sensor” (biosensor) – first using a human proposer (investor) and later capturing this behavior in a computer agent.
We analyze the results of 287 dyads, in which healthy participants play against healthy trustees, as well as against the trustees that have four different disorders: ASD, BPD, MDD, and ADHD. Each subject played only one game. With the exception of some patients with BPD, participants with disorders were not medicated. A detailed description of the data is given in Table S1.
We sought to classify the dynamics using only the numbers exchanged in the game between players (investment and repayment ratios), the number of “types” or styles of play (number of clusters), and the functional dependence of the next investment on preceding investment and repayment ratios. In short, we sought a heavily data-driven approach.
We extended a previously published method [22], [23] to cluster available trust game data. This method uses a regression approach to the functional dependence that clusters individuals based on coefficients of the regression. This method has advantages over traditional clustering approaches: (i) the number of types in our population is estimated directly from the data, and (ii) classification uncertainty is captured by probabilities rather than categorical cluster assignments. An investor is not classified as either within or not within a cluster, but instead a probability of being in a cluster is computed. This allows us to identify clusters where a style of behavior (a type) is over-represented (in comparison with what is expected by chance), under-represented, or neither (see below for details of this calculation).
The basic model is determined directly from the numbers exchanged by the two players during the game. We model the healthy proposer's investment at time as a function of preceding investment and repayment ratios. In this “black-box”, regression approach [22]–[23], we assume that we can capture meaningful variations in types of investor play by using a regression model based only on previous investment and return ratios, in contrast to other approaches [21], [24] which commit to more explicit models of how these values are used in mental processes to generate behavior.
It is known that an arbitrary continuous function can be approximated, with any given accuracy, by a polynomial of an appropriate order. As a result, a widely used approach to describe such functions is to try polynomial dependence of increasing order. For a first order dependence of the current investment on previous investments and repayments the model is:(1)where indexes the subject and indexes the current round of the game. For a second order dependence of the current proposer investment on previous investments and repayments, this expression would accrue all possible second order terms in lagged investments and repayments, including terms of the type that describe interaction between investments and repayments. Such terms acknowledge that the current choice by the investor is entangled with their previous interactions with their partner. Although expression 1 depicts a first order dependence on previous investments and repayments extending back two rounds of the game, in this paper, we do not pre-commit to the exact functional dependence for the current proposer investment nor to the number of exchanges into the past that best predict the person's current decision. Instead, we assume a general polynomial dependence of the current investment ratio on previous investments and repayments, and determine the order of this polynomial dependence directly from the data. Similarly, we determine the number of rounds into the past required to predict optimally the person's current investment ratio from previous investment and repayment ratios. The details of this general approach follow.
Formally, we model the behavior as a mixture of regressions. For a fixed order of polynomial dependence , a fixed look-back window , and a fixed number of clusters , we assume a single investor's data is given byHere, is an investor's (8-dimensional) vector of investments (we consider models looking back as many as two rounds; to make the models comparable we only consider 8 investments), is the model matrix of independent variables defining the regression (all less than or equal to -degree monomials in lagged investments and repayments going back rounds), is the -th cluster's regression coefficients, is the variance of the error term in the -th cluster, is the weight assigned to the -th cluster, the multivariate normal, and is the identity matrix.
This behavior model is applied to the data from the whole group, with the data itself determining both the appropriate subdivision into clusters and the regression coefficients within each cluster.
We use the data augmentation approach [25], defining latent variables which assign investors to clusters, to form the complete data . We then get the joint posterior of the parameters and the latent variables by combining the complete data likelihood with priors over the parameters. We choose for priorswhere Dir denotes the Dirichlet distribution and the inverse gamma distribution [26]. These are the same priors that were used by Houser-Keane-McCabe in their work [23]. As our independent variables all lie on the interval , we chose the prior variance of the coefficients to be proportional to this range.
For the above model, we use a two-stage Gibbs sampling algorithm to estimate the parameters [27]:
Start with initial parameters then repeat:
Step 1: Sample allocations given : , where Mult is a multinomial distribution, and .
Step 2: Sample given 's:for to Here, is the pooled investment data over cluster , is the pooled model matrix over cluster , and is the normal multivariate density with mean and covariance . The sequences of samples can then be used to estimate parameters. To avoid possible adverse effect of potential outliers on this Gaussian-based (hence outlier-sensitive) method, we check that the empirical distribution of the differences between the observed and predicted values is indeed consistent with the normality hypothesis. Finally, the optimal number of clusters, polynomial order, and look-back window can be determined by computing the marginal likelihood of each model (see the Methods section for details) and selecting the model with the largest value.
The method described above identified 4 clusters. In terms of the relevant parameters, two rounds were found to be the optimal number of previous moves for predicting the influence of past investments and repayment ratios on the current investment ratio made by the investor. To connect our clusters to the DSM-IV phenomenology, we determined which groups of subjects defined by DSM-specific criteria were over- or under-represented in each cluster and the number of standard deviations by which they were over- or under-represented.
The results of the clustering are shown in Figure 2 (see Table S2, Table S3 and Table S4 for a detailed description). Cluster 1 over-represents individuals with ADHD. Although 54% of these individuals would be expected to fall into this cluster by chance, 89% of them end up in this cluster. Cluster 2 significantly over-represents individuals with Autism Spectrum Disorder. By chance, 23% of these individuals should fall in this cluster; however, we see 44% of them in the cluster. In Cluster 3, medicated and non-medicated individuals with Borderline Personality Disorder are over-represented. By chance, 15% of individuals from each group should fall into this cluster. However, 36% of medicated and 27% of non-medicated Borderline Personality Disorder individuals belong to this cluster. Cluster 4 should by chance represent 8% of individuals with MDD, but 20% of them fall into this cluster. The chi-square analysis confirms the statistical significance of this over-representation (see Methods section).
For two disorders, there are known scores describing its severity: for ASD, there is a score on the Autism Diagnostic Interview-Revised [28] Repetitive behavior subscale, and for BPD, there is a score on the Interpersonal Trust Scale [29]. In both cases, we found a statistically significant correlation between these scores and the probability of belonging to the corresponding cluster ( = percent match of the dyad in this cluster from 30,000 draws from the posterior): and for ASD (Figure 3) and and for BPD (Figure 4).
With the clusters defined as described, we sought to characterize the kinds of social gestures (signals sent across rounds and between players) that define them. In Figure 5, we summarize the across-round social gestures for each cluster in terms of the regression coefficients for the investment and repayment ratios and the constant term (see also Figure S1 and Figure S2). We discuss the potential importance of these findings below, but here we summarize in Figure 5 the average social gesture of each cluster by plotting the average regression coefficients for each restricting the number of rounds back to two – the optimal number that predicts the investors next investment ratio (Figure 5B). Notice that in Cluster 4, the dependence is dominated by the constant term; this term reflects universally high investments. In Cluster 4, investors playing subjects with major depressive disorder are over-represented. The other over-represented group in Cluster 4 are investors playing trustees that they meet before the game and whose pictures they see each round of the exchange. It is interesting to note that investors playing subjects with ASD end up over-represented in the same cluster (Cluster 2) as investors playing subjects in an impersonal version of the game – where subjects do not meet nor see each other.
The above results provide evidence that examining investor-side behavior provides a new kind of ‘readout’ for some important psychopathology groups studied under the probe of the multi-round trust game. The game itself, although simple (in each round only two numbers are exchanged), requires a number of intact cognitive functions including working memory, short-term memory, the capacity to model and predict the partner's likely response, the capacity to sense deviations from these expectations, good a priori models of human trade instincts (reflected by round one offers and responses), and so on. One value of this approach is that it utilizes a probe that is not directly related to the symptom lists that define DSM classifications, and therefore provides a possible alternative method of classifying some psychopathologies – or at least identifying or isolating some of their malfunctioning computations.
To verify the robustness of the clustering algorithm we employed a previously described computer agent designed to play the trustee role.
The possibility to design agents of this type was shown in our previous work [6]. The corresponding “-nearest neighbor” agents use the database containing the results of all the rounds of all the dyads. A healthy trustee agent, to describe how much to repay, looks at the vector of 6 previous choices (last 3 investments and last 3 repayments) and finds, of all the records with healthy trustees, situations in which corresponding previous choices were the closest (in the Euclidean distance). Out of the 6 recorded outcomes of these closest situations, the agent selects one with equal probability. A BPD trustee agent similarly selects from dyads with a BPD trustee.
These trustee agents were validated in [6]: in interaction with healthy human investors, the BPD agent was shown to reproduce accurately ruptures in cooperation normally observed when a healthy investor plays a BPD trustee. Such ruptures were not observed in healthy investors playing a healthy computer trustee.
In our case, we need to supplement these agents with a similar investor agent that select the investment value based on the 6 closest dyads. Our hypothesis is that the same correlation with disorders can be detected by players playing against the investor agent.
Since it was already shown that the trustee agents adequately describe the trustee behavior, we had healthy investor agent play either the healthy or BPD trustee agent in the trust task for ten rounds (Figure 6A) 1,000 times. These interactions were then assigned to the previously determined clusters using the posterior distribution of parameters generated from the analysis of the human dyads (see details in the Methods section). Notably, interactions between the BPD trustee and healthy investor agent were statistically significantly over-represented by 7.19 standard deviations in Cluster 3 – the same cluster in which investors playing both medicated and non-medicated individuals with Borderline Personality Disorder are over-represented. On the other hand, interactions between the healthy investor and healthy trustee agents were not statistically significantly over-represented in this same cluster; see Figure 7 and Figure S3.
Thus, for BPD, the same correlation between the statistical clustering and disorders can indeed be achieved by using the investor agents (For the ASD group, there were insufficient data () to develop an analogous trustee agent and so no validation along this psychopathology was possible at this time).
We have used a data-driven, Bayesian regression approach to cluster the healthy investor behavioral data from a large set of 287 trust interactions, which included trustees from several DSM mental-illness groups. The Bayesian approach allowed us to determine in a principled way the number of clusters in our population (four) and probabilities for each dyad to belong to each cluster. Next, we used a chi-square criterion for over/under-representation to determine which pre-defined DSM-IV groups are statistically significantly over- and under-represented in each cluster. We found that there is a one-to-one correspondence between the resulting clusters and the DSM-IV disorders: namely, dyads in which trustees have a certain DSM-IV disorder are over-representedin the corresponding cluster.
Moreover, there is a correlation between the severity of each disorder and the probability of belonging to the correpsonding cluster.
The finding that a trustee's disorder can be detected based on the investor's behavior is in line with the fact that in any multi-round interaction with another human, a player's choices are dramatically entangled with those of her partner. Humans bring highly developed social sensitivities to two-party interactions. Our results show that these sensitivities can serve as a biosensor – the quantitative behavioral dynamics of a healhy person can capture the subtle behavioral abnormalities (abnormalities that are difficult to capture by the usual statistical analysis) of her partner.
To further validate our approach, we used a previously described -nearest neighbor sampling agent, as well as its implementation in the investor role, to simulate healthy vs healthy and healthy vs BPD interactions. We showed that healthy vs. BPD agent interactions were over-represented in the same cluster as healthy vs BPD individuals, whereas the healthy vs. healthy agent interactions were under-represented in this cluster.
Having arrived at an initial validation of our clustering, one can ask what further information can we extract from our method. Specifically, what do the patterns of play (Figure 5A) and polynomial coefficients used to predict investor behavior (Figure 5B) tell us about the behavior of individuals in each group?
We start with the fourth, and smallest, cluster. This cluster over-represents (i) dyads who met before playing the trust game as well as (ii) healthy investors playing trustees with Major Depressive Disorder. In this cluster, investment ratios are very high, and return ratios, in comparison to other clusters, are also high. For this “trust cluster”, the constant term effectively dominates the polynomial predicting the investment ratio.
The third, next largest, cluster, over-represents both medicated and non-medicated individuals with Borderline Personality Disorder. In this cluster, both investment and return ratios are relatively low.
The second cluster over-represents adolescents with Autism-Spectrum Disorder. The difference in pattern of play between this cluster and cluster one is difficult to detect by simply looking at the round by round average investment and repayment levels. Notably, the two clusters separate individuals with Autism-Spectrum Disorder from individuals with ADHD, two disorders that are often difficult to separate because they share several symptoms. One of the advantages of our method is that we arrive not only at clusters, but also at polynomial coefficients that can be used to predict investment ratios in each cluster. By looking at these coefficients, one can see a characterizing feature of cluster two - specifically, the current investment ratio depends strongly on the ratios of investment and return one round back. It is known that reciprocity is a driving signal in the trust game [25], and that the sensitivity to reciprocity of individuals with Autism-Spectrum Disorder is blunted [28]. The investor behavior in cluster may be an adaptation to this diminished sensitivity.
While our results show the statistically significant biosensing of certain disorders, the resulting clustering does not provide us with a clear diagnostics – since each cluster contains, in addition to individuals with the corresponding disorder, also a large number of healthy individuals; see Table S5. The fact that we did not get a clear separation between normal participants and participants with disorders (i.e. we find healthy participants scattered across the cluster) points to two distinctly different ways to approach psychopathology [30]. One possibility is that psychopathology groups are reflections of “quantitative” differences along normal cognitive dimensions (and their correlations) that are probed by our interpersonal exchange game. The second is that the first possibility holds but is augmented by the fact that psychopathology groups bring extra (or fewer) or different cognitive dimensions to the responses elicited by the game (Figure S4). To shed light on this issue in the context of this task, we clustered the healthy dyads alone and then assigned the disordered dyads to these clusters. The algorithm again selected 4 as the optimal number of clusters, 1 as the polynomial order, and 2 as the lookback length, but the assignment of the disordered dyads to the clusters is somewhat different than in the main result (Table S6). For BPD dyads the overrepresentation result is stronger, but for the other groups it is weaker. Also, while the betas of the regression (Table S7) are quite similar in three of the clusters, the fourth (Figure S5) is substantially different. Finally, the cluster assignments of the healthy dyads are in good concordance across the two clusterings (adjusted Rand index of .94 [31]–[33]; see also Table S8). Taken together, these facts suggest that the second view of psychopathology mentioned above is to be favored in this task, and that as far as this behavioral probe is concerned the disordered individuals are qualitatively different.
Interestingly, a seemingly more direct classification – based directly on the return values – does not lead to such a statistically significant correlation between clusters and disorders: many differences between healthy and pathological trustees cannot be detected against the background of other behavioral differences; see Table S9. This shows that humans acting as biosensors have the ability to “filter out” the important differences – and thus, help in diagnosing psychopathologies.
To summarize: we have data from 287 dyads involved in one such task - the trust game. We use a data-driven, agnostic method [22], [23] to arrive at (i) the number of clusters, (ii) the order of the polynomial that predicts investment ratios, and (iii) the number of rounds prior on which investor decisions depend directly from the data. We then arrived at a probabilistic clustering of these dyads, and analyzed over-representation of initial groups in the new clusters. We found that, by clustering dyads based on investor decisions, we were able to over-represent trustees with different disorders in separate clusters. Further, we used previously described k-nearest neighbor sampling agents [6] to generate 1000 interactions each for healthy vs healthy and healthy vs BPD agents. By clustering these interactions based on the polynomial coefficients from our initial clusters, we found that simulated healthy vs BPD interactions are statistically significantly over-represented in the same cluster as real healthy vs BPD interactions, but that simulated healthy vs healthy interactions are statistically significantly under-represented in the same cluster. We believe that these results constitute a significant step forward in quantitative diagnosis of psychiatric illness. The fact that brain images have helped in the analysis of human behavior in fairness games [2]–[8] makes us believe that our diagnoses can be further refined by using the corresponding brain imaging data.
Current psychiatric diagnoses are based on the DSM [34]. Essentially, these are lists of criteria used by a trained physician to characterize whether or not a person has a specific disorder. Such clinical, experience-based classification schemes provide a valuable understanding of psychiatric and neurological disorders. However, to uncover genetic underpinnings of various psychiatric disorders and to provide quantitative behavioral and neural measures, it is desirable to have quantitative measures of normal social interactions that can expose computations perturbed in various psychopathologies. Such measures could then be used to quantify abnormalities in social exchange, to diagnose psychiatric and neurological disorders, and to probe the genetic basis of such disorders. The results presented in this paper show some of our first steps in this direction; however, as more data on this and similar parametric social exchange tasks becomes available it should help to construct a quantitative understanding of mental disorders.
Intuitively, one might expect the investment on the next round to be an interactive function of both previous investment and the repayment the investor received, rather than independent effects of each. However, our analysis shows that the optimal clustering corresponds to polynomials of order , i.e., to the linear dependence (1). This means that, contrary to this intuition, the second-order terms – in particular, interaction terms between investments and repayments (such as ) – do not lead to a statistically significant improvement of the model's explanatory power.
For patients diagnosed with a DSM-IV disorder, medication is an important potential confound. In our study, only some BPD patients were medicated. According to Figure 2, both medicated and non-medicated BPD patients were statistically significantly over-represented in the corresponding Cluster 3. Thus, the presence or absence of medication does not affect our classification.
In this paper, we use a purely data-driven approach to data analysis. This approach is important from the foundational viewpoint, since it enables us, in particular, to further confirm the objective nature of the existing psychopathology classification. From the practical viewpoint, once this classification is established, we can improve the diagnostic efficiency if we explicitly use the known diagnoses in classification and regression analysis. For example, this may make it possible to find the markers that identify healthy subjects with superior discriminatory power.
Informed consent was obtained for all research involving human participants, and all clinical investigation were conducted according to the principles expressed in the Declaration of Helsinki. All procedures were approved by the Baylor College of Medicine Institutional Review Board.
The game is described in the previous section. Healthy participants were invited to the Human Neuroimaging Laboratory at Baylor College of Medicine. Prior to playing the game, each participant was instructed they would earn between $20 and $40, scaled by number of monetary units (MU) each player individually accrued. Following the game, each participant was compensated as follows: <68 MU = $20, 68–133 MU = $25, 134–200 MU = $30, 201–300 MU = $35, and >300 MU = $40.
We discarded 1,000 draws as burn-in, sampled 30,000 draws from the posterior, and assessed convergence using the Raftery-Lewis test [35]. We used the R Bayesian Output Analysis program to perform these calculations [36]. We repeated our analyses using 8,000 cycles total as per Houser, Keane, and McCabe [23] and 1,000; 3,000; and 5,000 cycles as burn-in and arrived at similar over-representation results.
To check that the empirical distribution of the differences between the observed and predicted values is indeed consistent with the normality hypothesis, we normalize each difference by subtracting the sample mean of the differences from the corresponding cluster and then divide by the sample standard deviation of these differences. We then compute the sample skewness and the sample kurtosis of the collection of all these differences, and use Matlab's Jarque-Berra test to check normality. Normality has been confirmed with . Since the null hypothesis of normality is rejected when , our value of indicates a strong empirical support for the normality hypothesis.
We used the Laplace-Metropolis estimator of the marginal likelihood [37], as described in Houser, Keane, and McCabe [23], to compare models with different values of the number K of clusters, order P of the polynomials, and the number D of past rounds on which the model depends. We did not include any results in which 2 of 3 samplers arrived at at least one empty type in the mode of the last 5,000 of 8,000 draws from the posterior. To maximize marginal likelihood (i.e., to find a posterior mode), we used component-wise optimization (also known as conditional maximization or step-wise ascent; see, e.g., p. 312 of [38]), the use of which is well-established for Bayesian problems such as maximizing the posterior mode, and arrived at the same answer when comparing the maximum log marginal likelihoods for different models. As a result, we concluded that the optimal model has clusters, a first order polynomial , and a dependence of ratios of investment on ratios of investment and return rounds into the past.
We found that, in contrast to the simpler case described in [23], our marginal likelihood values are sometimes fairly close to one another in many cases and thus, the results of comparing these values can potentially change if we repeat the same computational experiment. To makes sure that our selection of 4 clusters does not change, we supplemented the conditional maximization by the exhaustive analysis of all possible triples with up to 10 clusters, polynomials of order 1 to 3, and a time dependence of 1 or 2 rounds into the past. For each such model, we used several samplers and got several values of marginal likelihoods; when we compare two models, we select the simpler one (the one with fewest overall parameters) unless the other one has a statistically significantly larger mean. Since for the same model , the distribution of marginal likelihood values is sometimes not Gaussian (see Figure S6B), we could not use the usual t-test. Instead, we used the Wilcoxon rank-sum test at the 5% significance level [39]. The results (shown on Figure S6A and detailed in Table S4) confirm that the model with = (4, 1, 2) is optimal.
To check whether the observed over-representation of participants with disorders in different clusters is statistically significant, we apply the chi square test corresponding to a null hypothesis that the participants of different disorders are randomly distributed in different clusters .
Let be the number of elements in the -th cluster, the number of elements of -th group in this cluster, the cluster corresponding to the group , and the ratio of group in the population as a whole. Under the null hypothesis, due to the central limit theorem, the value is asymptotically normally distributed, with mean and variance . Thus, the ratio is normally distributed with mean and variance . Thus, to test the null hypothesis, we can form the test statistic , where is a relative over-(under-) representation of the group in the cluster . For , the null hypothesis is rejected with when . Thus, when each of the four terms in the sum satisfies the inequality , the null-hypothesis is rejected. We therefore consider the groups which are over-represented at the level .
Please note that when , already the over-representation of the group in cluster is statistically significant with .
Our clustering is based on iterations of Gibbs sampling. Every additional vector (e.g., of agents playing) is then classified as follows. For each recorded iteration of the Gibbs sampling (after the burn-in), based on the recorded values of and , we compute the probabilities of belonging to different clusters (we use the same formula as in the subsection “Estimating the parameter”). Then, we select a cluster with the probability . After all these selections, we assign the dyad characterized by the vector to the cluster to which, among all the iterations, this vector was assigned the largest number of times.
|
10.1371/journal.ppat.1006253 | A self-perpetuating repressive state of a viral replication protein blocks superinfection by the same virus | Diverse animal and plant viruses block the re-infection of host cells by the same or highly similar viruses through superinfection exclusion (SIE), a widely observed, yet poorly understood phenomenon. Here we demonstrate that SIE of turnip crinkle virus (TCV) is exclusively determined by p28, one of the two replication proteins encoded by this virus. p28 expressed from a TCV replicon exerts strong SIE to a different TCV replicon. Transiently expressed p28, delivered simultaneously with, or ahead of, a TCV replicon, largely recapitulates this repressive activity. Interestingly, p28-mediated SIE is dramatically enhanced by C-terminally fused epitope tags or fluorescent proteins, but weakened by N-terminal modifications, and it inversely correlates with the ability of p28 to complement the replication of a p28-defective TCV replicon. Strikingly, p28 in SIE-positive cells forms large, mobile punctate inclusions that trans-aggregate a non-coalescing, SIE-defective, yet replication-competent p28 mutant. These results support a model postulating that TCV SIE is caused by the formation of multimeric p28 complexes capable of intercepting fresh p28 monomers translated from superinfector genomes, thereby abolishing superinfector replication. This model could prove to be applicable to other RNA viruses, and offer novel targets for antiviral therapy.
| Superinfection exclusion (SIE) is employed by many viruses to guard their host cells against secondary invasions by the same or highly similar viruses. We describe a transformative discovery that self-perpetuating coalescence of a virus-encoded replication protein serves as a novel mechanism for SIE. Our findings further suggest that the primary targets of SIE may very well be progeny viral genomes rather than superinfectors, with the goal of blocking renewed multiplication of progeny viruses in the cells resided by “parental” viruses, thereby minimizing the proliferation of replication errors. Consequently, the general mechanistic framework revealed in this study may be highly conserved among viruses with diverse genome structures and host organisms, and could become a potent target for antiviral therapy.
| Superinfection exclusion (SIE) refers to the ability of a pre-existing virus (the primary invader) to exclude secondary infections by the same or closely related viruses (superinfectors) at cellular and/or organismal levels. SIE has been observed with many human and animal pathogenic viruses, including the reverse-transcribing human immunodeficiency virus (HIV), positive sense (+) RNA viruses such as hepatitis C virus (HCV) and West Nile virus (WNV), as well as negative sense (-) RNA viruses like vesicular stomatitis virus (VSV) [1–5]. More recent studies reported examples of SIE occurring during infections of large, double-stranded DNA viruses including herpesviruses and poxviruses [6, 7]. In all these examples SIE exerted by primary invaders acted in single cells to prevent the multiplication of highly homologous superinfectors in the same cells.
SIE has also been documented for many plant viruses including citrus tristeza virus (CTV) and tobacco mosaic virus (TMV) [8, 9]. Studies using mutants of CTV, plum pox virus, soilborne wheat mosaic virus, and apple latent spherical virus that were labelled with different fluorescent proteins, established that co-introduced variants of the same virus occupy adjacent, yet non-overlapping cell clusters in the same leaves or tissue niches [10–13]. These studies clearly demonstrated that SIE among closely related plant virus variants likewise exclude each other at the cellular level, thus drawing strong parallels between SIEs occurring in plant and animal virus infections.
SIE may also be mechanistically related to the well-documented cross protection phenomenon observed in virus-infected plants [14–16]. Cross protection refers to the protection gained by plants against a more damaging virus variant, through pre-inoculating these plants with a mild variant of the same virus. Although cross protection has been adopted for plant virus disease management for at least 50 years [15], exactly how the protection is achieved remains to be satisfactorily explained. In theory the cellular level SIE could account for at least some aspects of cross protection, as the plant-wide spread of the pre-inoculated variants could block the cells already occupied by these variants from being invaded by the more severe strains [17]. Nevertheless, there is some evidence to suggest that additional mechanisms might act at the organismal level to augment cross protection [10].
This widespread functional conservation of the cellular level SIE suggests that, once better understood, it could become a promising target for antiviral therapy. However, the molecular mechanisms of SIE remain little elucidated for most viruses. While for some viruses interference with entry of the superinfector appears to be critical [18], for most others the essential inhibitory step is clearly post-entry [2, 3, 19]. A striking feature of SIE shared by the viruses examined so far is its dependence on one or a few virus-encoded proteins and relatively little involvement of host cell components, highlighting SIE as a primarily virus-driven phenomenon [6, 20].
We report here the characterization of molecular mechanism of SIE in turnip crinkle virus (TCV) infected cells. TCV is a small icosahedral plant virus with a (+)-strand RNA genome that encodes five proteins. The 5’ proximal p28 and its C-terminally extended derivative—p88—are both required for viral genome replication (Fig 1A). The two small proteins encoded in the middle of the TCV genome, p8 and p9, are translated from a subgenomic RNA (sgRNA1) and function as cell-to-cell movement proteins (MPs; Fig 1A). Finally, the 3’ proximal p38 is translated from a second sgRNA (sgRNA2) and serves both as the capsid protein (CP) and the TCV-encoded suppressor of RNA silencing (Fig 1A).
TCV p28 is known to play an indispensable role in viral RNA replication, presumably by rearranging mitochondrial outer membrane into partially enclosed viral replication complexes (VRCs) that house itself, p88, viral RNA, as well as cellular factors required for optimal genome amplification [21–24]. In the current study, we uncover a novel role of p28 by showing that it exclusively determines SIE among TCV variants. Strikingly, it does so by forming multimeric intracellular inclusions that inactivate p28 translated from superinfectors, thereby denying superinfectors the chance to replicate. This novel, simple mechanism of SIE could prove to be applicable to many other viruses, and predicts a novel target for antiviral therapy.
To adopt TCV as a model for investigating SIE, we first tagged TCV replicons with two different fluorescent proteins (Fig 1B). Previous studies showed that deletions within the CP coding region of TCV, while compromising viral cell-to-cell and systemic movement, did not affect its replication in single cells [25, 26]. We hence created two TCV replicons, TCV_sg2G and TCV_sg2R, by replacing the 5’ two thirds of the CP coding region with that of GFP and mCherry (Fig 1B). The replicon cDNAs were flanked by the duplicated 35S promoter and terminator (2X35S and T35S) of cauliflower mosaic virus, and inserted into a binary plasmid destined for Agrobacterium tumefaciens (pPZP212) [27]. The resulting recombinant plasmids were transformed into A. tumefaciens strain C58C1 in order to initiate TCV replication in Nicotiana benthamiana leaf cells via agro-infiltration. Note that expression of GFP and mCherry proteins from these constructs strictly depends on TCV replication, as the sgRNA2 transcript is only produced during replication (hence the “_sg2” designation in the names of replicon constructs). A carnation mottle virus (CarMV)-derived replicon named as CarMV_sg2R was used as a non-TCV control (Fig 1B), as CarMV and TCV share similar genome organizations yet limited pair-wise sequence identities of encoded proteins (approximately 50%) [28]. Finally, a construct that facilitates non-replicative expression of p19, the tomato bushy stunt virus-encoded suppressor of RNA silencing, was included to counteract RNA silencing-mediated degradation of the primary, 2X35S-driven replicon RNAs [27, 28].
N. benthamiana leaves infiltrated with the replicon constructs were then inspected by confocal fluorescence microscopy at four days post infiltration (4 dpi). As shown in Fig 1C, leaf patches co-infiltrated with the two TCV replicons (left panels) contained cells that expressed either GFP or mCherry, but were completely devoid of cells that expressed both. Indeed, among thousands of cells inspected in multiple repeat experiments, less than 0.1% of the fluorescent cells fluoresced both green and red. In contrast, approximately 80% of the fluorescent cells observed following co-infiltration of TCV_sg2G and CarMV_sg2R expressed both GFP and mCherry (Fig 1C, right panels). These data illustrate that (i) agro-infiltration efficiently and simultaneously introduced multiple viral constructs into the same leaf cells; and (ii) intracellular, mutual exclusion occurred readily between variants of the same virus (TCV), but not between two distantly related viruses (TCV and CarMV).
Was the mutual exclusion between two TCV replicons caused by SIE, which by definition depends on a temporal lag between the entry of primary invader and superinfector? We speculated that co-introduced replicons might experience varying lengths of post-entry delays before they could initiate replication, thus allowing the ones that replicate first to exert SIE against others in the same cells. To test this idea, we compared the timing of fluorescence emergence in cells treated with 2X35S-GFP, a construct that expresses GFP independent of virus replication, and ΔMP_sg2R, a TCV replicon that expresses mCherry only when viral replication occurs (S1 Fig, panel A in Supporting Information). ΔMP_sg2R differs from TCV_sg2R by harboring a 92 nucleotide (nt) deletion within the MP region, thus restricting its replication to primary infected cells (S1 Fig, panel A). As shown in S1B Fig, replication-independent GFP fluorescence emerged in a few cells at 24 hours post infiltration (hpi), but quickly filled more than 70% of cells by 36 hpi. By contrast, replication-dependent mCherry expression first occurred in a few isolated cells at 48 hpi (white arrow in top right panel), and expanded only gradually thereafter, so that approximately 10%, 23%, and 33% of the cells became red fluorescent at 72, 96, and 120 hpi, respectively (S1 Fig, panel B). To summarize, replication-independent expression of GFP occurred early and synchronously in most cells, whereas replication-dependent expression of mCherry was seriously delayed, and then commenced stochastically in a limited number of cells, over an extended time span.
These observations were confirmed with Western blotting (WB) of leaves treated with the GFP and ΔMP_sg2R constructs separately, using an antibody that reacts with both fluorescent proteins (S1 Fig, panel C). Together these results indicate that replication initiation by a TCV replicon in host cells is delayed relative to replication-independent expression, and the length of delay varied dramatically from cell to cell. While the reason for delays will become evident later in this report, their varying lengths likely permitted one of the co-introduced replicons to commence replication earlier than others in the same cell, thus creating the time lag required for SIE.
We sought to further confirm SIE as the underlying reason for the mutual exclusion using a sequential delivery procedure. As shown in Fig 1D, pre-introduction of either TCV_sg2G or TCV_sg2R dramatically reduced the number of cells that replicated the reciprocal superinfector delivered with a 16-hour delay. As demonstrated in S1 Fig, panel D, delayed introduction did not in itself reduce the chance to initiate replication by replicons, as both replicons introduced with a 16-hour delay relative to the p19 construct initiated replication in a similar number of non-overlapping cells. Together, we conclude that SIE likely accounts for the inability of simultaneously introduced TCV_sg2G and TCV_sg2R to co-replicate in the same cells (Fig 1C).
We next assessed whether any of the TCV-encoded proteins could induce SIE by transiently expressing each of the TCV-encoded proteins, along with the TCV_sg2G replicon, in N. benthamiana cells. TCV p38 (CP) was not included in this set of testing as constructs without CP (TCV_sg2G and TCV_sg2R) still displayed SIE (Fig 1). To facilitate the verification of their expression, these TCV-encoded proteins were all fused to a C-terminal 2XHA tag, permitting WB detection with an anti-HA antibody (S2 Fig, panel A). Expression of p8-HA or p9-HA did not affect TCV_sg2G replication, as evidenced by robust GFP fluorescence in whole leaves (Fig 2A). In contrast, expression of p28-HA eliminated GFP fluorescence. We further established that p28-mediated repression of TCV_sg2G replication depended on the p28 protein, because a frame-shift mutation of p28 (p28fs) abolished repression. Importantly, the repression of TCV_sg2G by p28-HA was highly specific as p27 of CarMV (CarMV-p27HA) was ineffective. Interestingly, p88-HA partially repressed TCV_sg2G accumulation, presumably attributable to its N-terminal region identical to p28 (Fig 1A). We confirmed these results with Northern blots (NB) and WB (Fig 2B). Both TCV_sg2G RNA and GFP protein accumulated to high levels in leaves co-infiltrated with p8-HA, p9-HA, p28fs, or CarMV-p27HA constructs (lanes 5–8), but to much lower levels in leaves expressing p88-HA (lane 4), and were almost undetectable in samples expressing p28-HA (lane 3). Together these data indicate that p28-HA suffices for highly specific, potent repression of TCV replication.
To further test if p28 produced during TCV replication likewise represses the replication of a co-delivered replicon, we generated Δp88ΔMP_sg2R, a derivative of ΔMP_sg2R that contained a 4-nt deletion in the p88 ORF (Fig 2C). As a result, this defective replicon encodes p28 as the sole TCV protein, whose replication function was previously shown to depend on its expression in cis [22]. Functional p88 protein must then be supplied through the co-delivered ΔMP_sg2G (Fig 2C). We wish to emphasize that this latter replicon can replicate by itself as it encodes both wild-type p28 and p88. Also keep in mind that the ΔMP deletion in both constructs restricts them to single cells, thus successful complementation must depend on the co-entry of both constructs into the same cells. As expected, cells infiltrated with Δp88ΔMP_sg2R alone did not express mCherry, confirming its inability to replicate (Fig 2C, bottom left). Conversely, cells infiltrated with ΔMP_sg2G alone expressed only GFP (Fig 2C, middle panel). However, when the two constructs were co-infiltrated into the same cells, approximately 10% of fluorescent cells expressed mCherry, indicating that in these cells the replication of Δp88ΔMP_sg2R was successfully rescued in trans by the p88 protein translated from ΔMP_sg2G transcripts (Fig 2C, bottom right; see S2 Fig, panel B for images of individual channels). Strikingly, these mCherry-expressing cells always lacked GFP fluorescence, indicating that the replication of ΔMP_sg2G was abolished in these cells (S2 Fig, panel B). This result was consistently observed in multiple repeat experiments, and indicates that Δp88ΔMP_sg2R, once becoming replicationally active, turned around to block the replication of ΔMP_sg2G in the same cells, even though it must rely on the p88 provided by the latter for replication. Together these data reveal two vital insights into TCV SIE: (i) p28 acts both as a replication facilitator for Δp88ΔMP_sg2R itself, and as a repressor in trans for ΔMP_sg2G in the same cell, (ii) p28-mediated repression likely involves a post-translational interaction with p28/p88 encoded by a different RNA.
To understand how p28 represses TCV replication, we next examined its subcellular localization by expressing it as a fusion protein with GFP attached to its C-terminus (p28-GFP). As shown in S3 Fig (GFP panel), most of the transiently expressed p28-GFP within each cell coalesced into large (5–10 μm in diameter), intensely fluorescent foci. Interestingly, these foci did not co-localize with endoplasmic reticulum (ER, visualized with ER-mCherry; mCherry and Merge panels), or nuclei, or cell wall (both visualized with DAPI; DAPI and Merge panels. Note the white arrows). Since TCV replication was shown to alter the structure of mitochondria (MT) [21, 23], we further evaluated whether p28-GFP foci co-localized with MT using an MT-mCherry marker. As shown in Fig 3A, MT-mCherry appeared as tiny, elliptical dots (mCherry panel). p28-GFP foci and MT-mCherry dots were occasionally next to each other but rarely overlapped (Merge panel). Therefore, these p28-GFP foci appear to be physically distinct from sites of TCV replication.
Intriguingly, many of the p28-GFP foci underwent dynamic changes in their shape, size, and intracellular location. In Fig 3B, we highlight three of such foci, designated a, b, and c, by using orange and white arrowheads to mark their start and ending positions within a two minute timeframe. Focus a overlapped with the nucleus of the resident cell at the beginning but became clearly separated two minutes later. Similarly, focus b migrated approximately 20 μm downwards and changed shape along the way. Finally, focus c underwent similarly remarkable shape and size changes, despite of its less dramatic movement. Changes of foci a, b, and c, as well as several smaller p28-GFP foci, are further illustrated in a time lapse movie (S1 Video). Together these data clearly demonstrate that p28-GFP coalesces into highly mobile, dynamic inclusions in the cells of its expression that are likely not part of VRCs.
We next tested whether p28-GFP repressed TCV replication by co-expressing p28-GFP and ΔMP_sg2R. As shown in Fig 3C, left panel, most cells treated with both p28-GFP and ΔMP_sg2R constructs at a 1:1 ratio acquired the large irregular inclusions characteristic of p28-GFP. Only a few cells were observed to express mCherry, and these cells were always devoid of GFP, suggesting that they failed to take in the p28-GFP construct. The number of mCherry-expressing cells could be increased by reducing the p28-GFP to ΔMP_sg2R ratio to 0.1:1 (Fig 3C, right panel), and again the red fluorescent cells were free of GFP. We confirmed these findings with NB of TCV RNA, and WB of p28 protein (Fig 3D). While ΔMP_sg2R gRNA and sgRNAs were detected at similar levels in the absence or presence of unfused GFP (lanes 3 and 4), they became undetectable in the presence of p28-GFP (lane 5). Reducing p28-GFP input to one-tenth of ΔMP_sg2R largely released this repression (lane 6). Since both constructs were restricted to the cells they initially entered, these results strongly suggest that p28-GFP, similar to p28-HA, dominantly repressed ΔMP_sg2R replication in the same cells.
The 2X35S promoter-driven transcription of primary ΔMP_sg2R transcripts occurs independently of viral replication. As a result, some p28 proteins are expected to be translated from these primary transcripts even if replication does not take place. This experimental set-up provided us with the opportunity to test whether p28-GFP exerts its repressive activity by interfering with the translation of p28 from a replicon RNA. As shown in Fig 3D, when subjected to WB with a p28 antibody, leaves treated with either p28-GFP or ΔMP_sg2R alone accumulated p28-GFP and p28, respectively (ca. 55 or 28 kilo-daltons [kDa]) (lanes 2 and 3). Surprisingly however, leaves treated with both of them accumulated both proteins to easily detectable levels (lane 5). The detection of p28 in the co-inoculated leaves is particularly noteworthy because its corresponding ΔMP_sg2R transcript level was at least 10-fold lower than in cells with ΔMP_sg2R alone (compare lanes 3 and 5). Thus, p28-GFP-mediated repression of replication must have occurred at a step downstream of p28 translation. These results are consistent with those of Fig 2C implying normal translation of p88 from the repressed ΔMP_sg2G replicon. Together they indicate that p28-GFP blocked ΔMP_sg2R replication post-translationally, probably by disrupting the replication function of p28 (and p88) translated from ΔMP_sg2R RNA.
The results presented in previous sections, while consistent with a pivotal role of p28 in SIE induction, also beg the immediate question of whether p28 expressed independently of replication, without any C-terminal tags (i.e. 2XHA or GFP), still represses a TCV replicon. This question needed to be resolved because if it does, it would be puzzling that any of our replicon constructs could initiate TCV replication at all, as their primary transcripts would all be expected to translate p28 as the first protein (e.g. S1 Fig; Fig 3D). On the other hand, if it does not, then neither p28-HA nor p28-GFP would reflect an inherent function of p28, which would in turn contradict the observation that the p28-only defective replicon (Δp88ΔMP_sg2R; Fig 2C) could repress a different replicon (ΔMP_sg2G; Fig 2C).
To resolve this puzzle, we produced three new constructs: the first would express an untagged p28 in agro-infiltrated cells. The second, G11-p28, should express a p28 variant tagged at the N-terminus with a 25-aa “G11” tag derived from the 11th β-strand of GFP (Fig 4A). The value of G11 tag will become apparent later (Fig 5). Note that these two constructs express the untagged p28 and G11-p28 proteins independent of TCV replication. Finally, the third construct, [p28fs]_sg2R, harbors a non-replicating TCV mutant that contains a one-nt deletion at position 106, causing p28 (and p88) translation to stop after 14 aa. However, the p88 function of [p28fs]_sg2R was little affected, as the replication of this mutant was restored by providing only the p28 (Fig 4A and 4B). An N-terminally truncated p88 could be produced from [p28fs]_sg2R through translational re-initiation at another AUG codon 36 aa downstream.
Contrary to the prevailing assumption that replication function of p28 is cis-acting [22], the transiently expressed, untagged p28 was able to partially complement the replication of the p28-defective [p28fs]_sg2R. As shown in Fig 4B, [p28fs]_sg2R did not replicate by itself (panel 1; also Fig 4C, lane 2), but replicated in an average of 25% cells in the presence of untagged p28 (panel 3; Fig 4C, lane 4). Notably, no complementation was observed with p28-GFP (Fig 4B, panel 2), suggesting that the C-terminal GFP tag abolished the replication function of p28. Unexpectedly, the N-terminally tagged G11-p28 complemented the p28-defective replicon twice as efficiently as p28 (compare panels 3 and 4 of Fig 4B, and lanes 4 and 5 of Fig 4C), suggesting that the N-terminal G11 tag rendered p28 more replication-active, possibly by overcoming certain repressive state of p28 (see later). Importantly, complementation by untagged p28 weakened only modestly when [p28fs]_sg2R was introduced with a16-hour delay (Fig 4B, panel 5; and Fig 4C, lane 7), suggesting that the replication state of p28, once established, was minimally affected by extended pre-accumulation. By contrast, complementation by G11-p28 actually strengthened when the defective replicon was delayed, further confirming G11-p28 as more replication-active than untagged p28 (compare Fig 4B, panels 4 and 6; Fig 4C, lanes 5 and 8).
However, when co-introduced with the wild-type (wt) p28-encoding TCV_sg2R replicon, transiently expressed p28 turned repressive. As shown in Fig 4B, untagged p28 caused the number of red fluorescent cells to decrease to about 1/4 of the TCV_sg2R only control (compare panels 7 and 9), indicating a partial repression of the replicon by untagged p28. Furthermore, this repression by p28 became more potent when the TCV_sg2R replicon was introduced with a 16-hour delay, reducing the number of cells with replication-dependent mCherry fluorescence to 1/15 of the control (panel 11). Notably, the N-terminally tagged G11-p28 was substantially less repressive than untagged p28 (Fig 4B, panels 10 and 12). This contrasts with the C-terminally tagged p28-GFP that caused a complete loss of TCV_sg2R replication (Fig 4B, panel 8; also Figs 2 and 3), suggesting that these tagged forms of p28 represent two extremes of the repressive activity of wt p28.
These data were further corroborated with NB (Fig 4C). The reduction in TCV_sg2R RNA levels caused by untagged p28 was less dramatic than the numbers of red fluorescent cells, possibly reflecting the limitation of confocal microscopy in detecting cells with very low mCherry expression (Fig 4B and 4C). Together these results strongly suggest that transiently expressed, untagged p28 exists in two different states—one replication-active, the other repressive. Importantly, the replication-active state, once established early on, did not readily transit to the repressive state, as evidenced by the sustained ability of p28 (and G11-p28) to complement sequentially delivered [p28fs]_sg2R (16 hour delay; Fig 4B, panels 3 and 5). Intriguingly, the repressive state of p28 that repressed the co-introduced TCV_sg2R must have established itself fairly early as well. The co-existence of two p28 states with opposite functions in turn suggests that their corresponding protein fractions probably exist in separate cellular compartments.
Results presented above strongly suggest that untagged p28 exists in both replication-active and –repressive states. Since the replication-repressing p28-GFP formed large, dynamic, and mobile intracellular inclusions, we next wondered if the repressive state of untagged p28 also existed in similar inclusions. Because untagged p28 does not fluoresce by itself, we first determined whether G11-p28, being replication-competent but repression-deficient, were capable of forming punctate inclusions. Detection of G11-p28 was in turn facilitated by co-expression of G1-10, the first 10 β-strands of GFP, which interacts with the G11 tag to generate green fluorescence (Fig 5A) [29, 30]. As shown in Fig 5B, panel 1, GFP fluorescence reconstituted by G11-p28 and G1-10 co-expression showed a clearly diffuse distribution without discrete foci. Since G11-p28 efficiently complemented the replication of a p28-defective mutant (Fig 4), this result suggests that the replication function of p28 does not require the formation of punctate foci.
In contrast, while untagged p28 alone did not cause detectable changes in cell morphology (panel 2, cell boundaries visualized with ER-mCherry), its co-expression with (G11-p28 + G1-10) led to the coalescence of green foci in approximately 70% of fluorescent cells, and a reduction of diffuse GFP in the same cells (Fig 5B, panel 3; note the top, left, and right cells contained green foci but little diffuse green fluorescence, whereas the bottom middle cell contained mostly diffuse green fluorescence). We infer that these intensely fluorescent foci must have arisen from the coalescence of the otherwise diffuse G11-p28, induced by the co-expression of untagged p28. This in turn suggests that untagged p28 could self-associate into punctate inclusions capable of coalescing G11-p28. Consistent with this inference, p28-mCherry, which like p28-GFP formed visible inclusions, coalesced G11-p28 much more efficiently than untagged p28, leading to dominance of mostly yellow foci in co-infiltrated cells, and concomitantly near complete loss of diffuse G11-p28 (Fig 5B, panel 4; note yellow foci). Together these results reveal two important insights: (i) transiently expressed, untagged p28 induced the formation of punctate inclusions in a substantial fraction of cells; and (ii) the p28 (and p28-mCherry) inclusions, once formed, are capable of seeding the coalescence of a non-coalescing p28 mutant (G11-p28).
Notably, complementation of the p28-defective [p28fs]_sg2R by G11-p28 (plus G1-10) was accompanied by diffuse distribution of both GFP and mCherry, resulting in cells that emit diffuse, brown fluorescence (Fig 5B, panel 5). By contrast, co-expression of G11-p28 (plus G1-10) with TCV_sg2R encoding wt p28 led to the appearance of green foci in approximately 50% of cells, indicating that wt p28 expressed from the replicon RNA likewise formed punctate inclusions that seeded G11-p28 coalescence (Fig 5B, panel 6). Strikingly, some of the focus-harboring cells contained little or no red fluorescence, suggesting that in these cells the replication-repressive state of p28 was established very early, thus preventing the initiation of replication by TCV_sg2R. This outcome mirrors the inefficient replication initiation by TCV replicons documented in S1B Fig, suggesting that at least in some cells, p28 translated from replicon transcripts turned repressive before the transcripts had chance to commence replication. Most importantly, similar to p28-GFP inclusions, these replicon-induced inclusions were highly mobile, often racing across the entire cell in less than ten minutes (time lapse S2 and S3 Videos in Supporting Information).
Here we summarize several key observations concerning the role of p28-induced punctate inclusions in SIE. (i) The inclusions are not required for replication function of p28. (ii) Untagged p28, expressed either transiently or replication-dependently, is capable of forming inclusions, though less efficiently than C-terminally tagged p28 derivatives. (iii) Once formed, these inclusions nucleate the otherwise non-aggregating p28 derivatives. (iv) The inclusions are highly mobile, likely enabling them to capture monomeric p28, sequestering them from the alternative functionality of p28 (replication). Together they support the concept that these inclusions correspond to a repressive state of p28, and they exert the repressive function by trapping freshly translated p28 molecules.
Having shown that the replication-competent, yet repression-deficient G11-p28 by itself maintained a diffuse distribution, but was converted to punctate inclusions by p28, we next asked whether G11-p28 expressed from a replicon could support TCV replication and SIE. To address this question, we generated two new replicon constructs (Fig 6A). The first, [G11-p28]_sg2G1-10, fused G11 to the N-terminus of p28, and incorporated G1-10 in the sgRNA2 coding region. Since sgRNA2 can only be synthesized following successful replication, the expression of G1-10 from this replicon, and resulting green fluorescence, will only occur if G11-p28 can function in replication. The second construct, [G11-p28]_sg2R, was identical to the first except that G1-10 was replaced by mCherry (Fig 6A).
When introduced individually into cells, the [G11-p28]_sg2G1-10 and [G11-p28]_sg2R replicons each produced fluorescence in approximately 10% of cells (Fig 6B, panel 1 and 2), suggesting that the G11 insertion compromised, but did not abolish, the replicability of the replicons. This was also confirmed with NB (Fig 6C, lanes 1 and 2). As expected, the green fluorescence reconstituted by G11-p28 and G1-10 co-expression was diffusely distributed (Fig 6B, panel 1). To our surprise, co-introducing these two constructs into the same cells led to approximately 50% of fluorescent cells to express both GFP and mCherry, indicating a loss of SIE that allowed the simultaneous replication of both the [G11-p28]_sg2G1-10 and [G11-p28]_sg2R replicons. We conclude that the replicon-borne G11-p28 was incapable of exerting SIE, despite of retaining p28’s activity in replication. Hence, the SIE function of p28 in the replicon background was likewise successfully decoupled from its replication function by the N-terminal G11 tag.
Our data thus far revealed that G11-p28 simultaneously weakens p28’s propensity to form inclusions and compromise its ability to exert SIE, leading us to hypothesize that the p28 inclusions are directly responsible for SIE. To further test this hypothesis, we asked if untagged, wt p28 expressed from a TCV replicon could exert dominant SIE to another replicon encoding G11-p28. To do so, we further generated the TCV_sg2G1-10 replicon that encoded wt p28 but expressed G1-10 from sgRNA2, which could not fluoresce unless G11-p28 translated from a different construct is present in the same cell (Fig 6A). N. benthamiana leaf cells were then treated with a mixture of TCV_sg2G1-10 and [G11-p28]_sg2R constructs. As expected, most of the fluorescent cells contained green but not red fluorescence (Fig 6B, panel 4), demonstrating preferential replication of TCV_sg2G1-10 encoding wt p28, and simultaneous repression of [G11-p28]_sg2R encoding G11-p28 (which nevertheless could translate G11-p28 independent of replication). Importantly, the green fluorescence existed as dense foci in these cells, confirming the coalescence of G11-p28 through an interaction with wt p28 that exerted SIE to [G11-p28]_sg2R. Although a few isolated cells showed red fluorescence indicative of [G11-p28]_sg2R replication (Fig 6B, panel 4, lower left), these cells never contained any GFP fluorescence, suggesting that they took in only the [G11-p28]_sg2R construct. Together these data further support the idea that a repressive state of p28 exerts SIE by trans-aggregating fresh, diffuse form of p28 translated from superinfecting TCV genomes.
SIE acts in the cells already occupied by a virus to deny the chance of secondary infections by the same or closely related viruses, but has no effect on more distantly related viruses. While SIE has been observed in infections of a wide range of viruses with diverse genome structures and host tropisms, its molecular mechanism remain poorly understood. Studies using CTV (different from TCV!) found that SIE can be abolished by replacing three virus-encoded proteins, namely the replication-related L1 and L2 proteases, as well as movement-related p33, with their counterparts from a distantly related CTV strain [8, 31]. Similarly, SIE of WNV and HCV was found to be reversed by mutations in a few viral proteins, among them WNV-encoded 2K and NS4A, and HCV-encoded E1, p7, and NS5A [4, 19]. More recently, Tatineni and French [20] revealed that proteases and capsid proteins encoded by two plant-infecting potyviruses were responsible for SIE of cognate viruses. These studies suggest that SIE is exerted by a relatively few proteins encoded by the primary virus. Mechanistic dissections using several (+) ssRNA animal viruses (WNV, HCV, and BVDV) further suggest that SIE functions primarily by blocking the replication of the superinfector [2–4, 19]. However, exactly how this blockage is achieved remains a mystery.
In the current study, we demonstrate that SIE among TCV variants is adequately explained by the action of p28, one of the TCV-encoded replication proteins, that exists in two distinct functional and structural states in infected cells. Collectively our data support a model through which p28 acts both as a replication facilitator and a repressor, depending on its cellular concentration (Fig 7). First and foremost, we hypothesize that both the replication and SIE functions of p28 derive from its inherent ability to self-interact to form oligomers. These oligomers become seeds onto which additional p28 molecules are coalesced, ultimately leading to structures responsible for replication or SIE. At the very early stage of a TCV invasion when p28 concentration is relatively low, the rate of p28 polymerization is slow, as the polymerization itself necessarily lowers p28 concentration in the cytoplasm. This rate is probably further constrained by various steps of virus replication complex (VRC) formation, including incorporation of p88 and/or viral RNA into p28 polymers, and enclosure of the complex by mitochondrion outer membrane [21, 23] (Fig 7, steps 1 and 2). On the other hand, in cells where a resident TCV is undergoing active replication, the large amounts of progeny TCV genomes template p28 translation to very high levels. This could in turn accelerate the polymerization of p28, leading to the formation of oversized p28 complexes that escape membrane enclosure, permitting easy trapping of freshly translated p28 monomers, including those translated from superinfecting TCV genomes, thus preventing the formation of new VRCs, leading to SIE (Fig 7, steps 3 and 4). Note that the structures of p28 complexes depicted in Fig 7 are hypothetical as we currently do not know their structural details. It is further possible that high levels of p28 might also dilute out the host proteins needed for VRC assembly, thus favoring the coalescence of p28 inclusion bodies.
Importantly, the formation of large cytoplasmic p28 complexes does not prevent the p28 molecules already sequestered inside VRCs from functioning in viral replication. This arrangement ensures that the primary invader continues to replicate inside VRCs, while superinfectors are subdued by the cytoplasmic p28 complexes. That p28 can exist both in membrane-associated VRCs and repressive cytoplasmic complexes is not unprecedented, as similar partitioning was reported for the tomato mosaic virus (ToMV)-encoded 130K replication protein, which exists in a membrane-associated, replication-competent form, as well as a cytoplasmic form capable of suppressing RNA silencing [32–34].
Our model is highly consistent with the observation that the strongly SIE-inducing p28 variants, including p28-GFP/mCherry/HA, form large, dynamic, and mobile cytoplasmic inclusions that trans-aggregate the otherwise diffusely distributed, yet replication-active G11-p28. Furthermore, our experiments indirectly implicated untagged p28, expressed either transiently or during replication, in similar cytoplasmic inclusions. It can be envisioned that the C-terminal tags could enhance the stability of p28 oligomers, thus favoring the formation of oversized polymers, whereas the N-terminal G11 tag could destabilize p28 oligomers, thus slowing down the polymerization.
Our model provides an immediate explanation for the inefficient initiation of replication by agro-delivered replicon constructs (S1B Fig). The strong 2X35S promoter in these constructs drives the transcription of primary TCV RNA to levels that are higher than natural viral invasions (typically a few viral genomes), but lower than progeny genomes produced by active replication. As a result, p28 oligomerization likely takes place at an intermediary rate that blocks most primary TCV transcripts from replication, yet allows occasional escape from this blockage in a stochastic manner (S1B Fig). Our model further asserts that the ones that do evade this blockage would in turn produce vastly more abundant p28, thus solidifying the repression against other co-introduced replicons.
Previous studies of other virus models appear to link large cytoplasmic inclusions formed by viral replication proteins with active replication [35, 36]. Based on our results, this view might warrant a second look. Specifically, we show that the N-terminally tagged G11-p28 robustly complemented the replication of a p28-defective replicon, yet did not form large cytoplasmic inclusions, and was incapable of SIE. By contrast, the C-terminally tagged p28-GFP/mCherry and p28-HA were completely inactive at replication, but formed large inclusions and exerted powerful SIE. These results suggest that the cytoplasmic inclusions represent the repressive form of p28, and that replication-active VRCs might be rather orderly structures discernable only with electron microscopy [37]. Presence of viral RNA in the cytoplasmic foci does not necessarily signal active replication, as viral RNA could become trapped in the foci through interactions with p28 and its analogs. Finally, our results also call for caution when interpreting experiments using tagged proteins, as these tags could alter the balance among various functional state(s) of original proteins.
Many virus-encoded replication proteins are cis-acting or at least cis-preferential, meaning they exclusively or preferentially benefit the very viral RNA molecule from which they are translated [22, 38–41]. In most (+) ssRNA viruses, the cis-acting protein typically corresponds to the replication protein that is more highly expressed, yet does not possess RNA-dependent RNA polymerase (RdRP) signature, thus analogous to p28 of TCV. Kawamura-Nagaya and colleagues [40] found that the replication-competent form of TMV p126 bound to TMV genomic RNA co-translationally, thus proposing this p126-RNA binding as the underlying mechanism for p126 cis-preference. However, it is uncertain as to how widely applicable this mechanism is, as it does not explain how the 1a protein encoded by the tripartite brome mosaic virus (BMV), being cis-preferential relative to the 1a-coding RNA1, is nevertheless able to function in trans to facilitate the replication of the other two BMV genome segments (RNA2 and RNA3) [39].
Similar to TMV p126 and BMV 1a, TCV p28 is also known to be cis-preferential [22] (also see Fig 2C). However, our new findings suggest that cis-preference of p28 likely reflects a strong trans-repression rather than cis-action. We show that transiently expressed p28 complemented a mutant replicon incapable of producing its own wt p28. Therefore, p28 cis-preference is readily overcome once the following two conditions are met: (i) p28 is expressed from a non-replicating mRNA; and (ii) the defective replicon does not encode a functional p28. It should be noted that this experimental system differs from that of White and colleagues [22], where the two mutant replicons used, ΔAPA and RT, produced either p28 (ΔAPA) or p88 (RT), but not both. As a result, the preferential amplification of the p28-encoding ΔAPA replicon could be explained by the fact that the p28 produced by a replicating ΔAPA could strongly repress the p88-encoding RT, whereas p88 produced by the RT mutant would be only weakly repressive (e.g. Fig 2A). This difference in repressive activity between p28 and p88 could cause p28 to appear to be cis-preferential. While it remains to be determined whether this mechanism applies to cis-acting proteins of other viruses, we note that it allows the replication-active form of p28 (and its analogs in other viruses) to be accessed by non-p28-coding RNAs like satellite RNAs, defective interfering RNAs and, in viruses like BMV, other genome segments. Of course, this model does not rule out the possibility that certain cis-acting RNA elements could function to de-aggregate p28 complexes, thus permitting their re-entry into the replication-competent pool.
Multi-variant populations of many plant viruses are known to undergo dramatic bottlenecking during systemic infections [28, 41–44]. Recent studies strongly suggest that this population bottlenecking is likely caused by cellular level SIE exerted by viral variants that arrive at certain parts of the plants (e.g. vascular bundles and systemic leaves) ahead of other variants [17]. The SIE mechanism proposed in this study additionally provides an alternative explanation for the intracellular population bottlenecking observed by Miyashita and colleagues [45]. These authors found that individual tobacco cells receiving a viral population comprising thousands of ToMV variants allowed a tiny fraction of the incoming variants (fewer than 10) to replicate. The authors invoked active but stochastic degradation by host cell nucleases as a possible reason of observed population bottlenecking [45]. Based on the SIE mechanism proposed here, we think that it is likely that, as soon as a few variants start to replicate, they would swiftly produce large amounts of replication-repressing viral proteins (or proteins like p28 that exert both replication and repression functions) that actively block the replication of other variants. This model is further consistent with the authors’ observation that the handful surviving variants in each cell accumulated varying numbers of progenies [45]–these variants could have initiated replication at different post-entry time points prior to the full establishment of the repressive state.
We noted earlier that SIE was observed in diverse viruses with distinct genome structures and host tropisms, suggesting it as an evolutionarily advantageous trait for viruses. With our current understanding of the TCV system as the basis, we next consider the underlying rationale for the preservation of SIE by viruses. In our model, we hypothesize that p28 translated from the newly replicated TCV genomes mostly serves to reinforce the repressive state of p28. The logical extrapolation of this idea is that these newly replicated TCV RNAs are also targeted by the repressive state of p28. Indeed, we speculate that they are the primary or “intended” target—the superinfector genome becomes targeted because it is “mistaken” as one of the newly synthesized viral RNAs. Why is then a trait that represses the replication of progeny genomes selected for? We reason that such a trait would ensure random mutations accidentally incurred during the viral replication process are isolated in a minimal number of progeny genomes. This is particularly important for RNA viruses as virus-encoded RdRPs are known to be error-prone, estimated to introduce close to one error per genome per replication cycle for TCV-sized genomes [46]. Consequently, allowing progeny genomes to repeat the replication cycle inside the cells of their origin would not only cause the proliferation of these errors, but also exacerbate their potential damages by compounding them with additional errors. By denying the progeny genomes the chance to re-replicate, SIE constitutes a powerful adaptive constraint that maximizes the genome stability of RNA viruses, and possibly also viruses with DNA genomes.
In summary, the current study reveals a novel mechanism of SIE that engages a repressive state of a replication protein to capture its counterparts translated from superinfected (and newly synthesized) viral genomes, thereby preventing the latter from entering replication cycles. While substantial additional research is needed to resolve the detailed structure and functional mode of the repressive form of the protein involved, we are confident that this general mechanistic framework will prove to be applicable to other viruses, possibly leading to the development of novel antiviral strategies that target this mechanism.
The TCV_sg2G construct was from a previous study (formerly named TCV-GFP) [47, 48]. Briefly, this construct is based on a binary plasmid (pPZP212) that replicates in both E. coli and Agrobacterium tumefaciens (strain C58C1). The TCV cDNA, in which the 5’ 2/3 of the CP coding sequence was replaced by that of GFP, was placed under control of 2X35S to permit the transcription of primary TCV RNA by RNA polymerase II of plant cells. Both TCV_sg2R and CarMV_sg2R were produced in a similar manner. Note that the mCherry coding sequence was modified to eliminate sites of common restriction enzymes while maintaining the amino acid sequence (the modified mCherry sequence, designated mCherry2, is available upon request). ΔMP_sg2G and ΔMP_sg2R were derivatives of TCV_sg2G and TCV_sg2R, respectively, that incorporated a 92-nt deletion within the MP coding region [26, 49]. Δp88Δ92_sg2R was in turn a derivative of ΔMP_sg2R that contained a 4-nt deletion within the p88 coding region, as a result of digestion with ApaI, followed by blunt-ending and relegation [22].
Replicon constructs containing G11-tagged p28 were generated through a three-step manipulation. First, in order to maximize the number of unique restriction enzyme sites, the 2X35S-TCV_sg2R-T35S cassette was switched to pAI101, a pCambia1300 derivative with a modified multiple cloning site, to create pAI-TCV_sg2R. Next, to produce [G11-p28]_sg2R, a gBlock fragment (IDT, Coralville, IA) consisting of TCV 5’ UTR plus the p28 start codon (66 nt), the coding sequence of the 11th β-strand of GFP connected to an 8-aa flexible linker (GRDHMVLHEYVNAAGITDGGSGGGS), and the first 63-nt of p28 coding sequence was synthesized, and used to replace the first 129 nt of TCV cDNA (released by XhoI and XmaJI sites) in pAI-TCV_sg2R using Gibson Assembly cloning (NEB, Ipswich, MA). Finally, the [G11-p28]_sg2G1-10 construct was created by replacing the mCherry coding sequence with that of the first 10 GFP β-strands (G1-10) [31].
Constructs 2X35S-GFP and 2X35S-p19 were described in an earlier study [27]. Other transient expressing constructs were produced similarly. The ER-targeting signal was derived from mGFP5 [50]. The mitochondrion-targeting signal was derived from the yeast cytochrome c oxidase IV [51].
Agro-infiltration experiments were conducted as described in ref [27].
Confocal microscopic observations were carried out using a Leica Confocal microscope (TCS SP5) available through Molecular and Cellular Imaging Center at the Ohio Agricultural Research and Development Center, The Ohio State University.
The quantitative data in S1B Fig were obtained by counting fluorescent cells in viewing fields containing approximately 50 cells. For each treatment, five such fields were counted and the percentages of cells with GFP or mCherry fluorescence were calculated. The numbers presented were midpoint values with range of variations. For all other figures, the fluorescent cells were quantified as a percentage of the total number of cells in a given viewing field by dividing a minimally magnified (10X) image of the viewing field to 100 (10 X 10) equal-sized mini-squares using the View/Show/Grid option of Photoshop, and counting the number of mini-squares within each image that are at least 50% fluorescent. For each treatment, at least four images were quantified in this manner and the range of variations was provided in the relevant figures.
NB and WB were carried out as described [27, 47]. A 32P-end labelled oligo complementary to the 3’ UTR of TCV was used to detect TCV genomic as well as subgenomic RNAs. A p28 antiserum was used to detect p28 protein in Western blots. A GFP antibody (Life Tech, Carlsbad, CA) was used to detect both GFP and mCherry.
|
10.1371/journal.ppat.1006668 | Merkel cell polyomavirus recruits MYCL to the EP400 complex to promote oncogenesis | Merkel cell carcinoma (MCC) frequently contains integrated copies of Merkel cell polyomavirus DNA that express a truncated form of Large T antigen (LT) and an intact Small T antigen (ST). While LT binds RB and inactivates its tumor suppressor function, it is less clear how ST contributes to MCC tumorigenesis. Here we show that ST binds specifically to the MYC homolog MYCL (L-MYC) and recruits it to the 15-component EP400 histone acetyltransferase and chromatin remodeling complex. We performed a large-scale immunoprecipitation for ST and identified co-precipitating proteins by mass spectrometry. In addition to protein phosphatase 2A (PP2A) subunits, we identified MYCL and its heterodimeric partner MAX plus the EP400 complex. Immunoprecipitation for MAX and EP400 complex components confirmed their association with ST. We determined that the ST-MYCL-EP400 complex binds together to specific gene promoters and activates their expression by integrating chromatin immunoprecipitation with sequencing (ChIP-seq) and RNA-seq. MYCL and EP400 were required for maintenance of cell viability and cooperated with ST to promote gene expression in MCC cell lines. A genome-wide CRISPR-Cas9 screen confirmed the requirement for MYCL and EP400 in MCPyV-positive MCC cell lines. We demonstrate that ST can activate gene expression in a EP400 and MYCL dependent manner and this activity contributes to cellular transformation and generation of induced pluripotent stem cells.
| Merkel cell carcinoma (MCC) is a highly aggressive, neuroendocrine cancer of the skin. MCC frequently contains integrated copies of Merkel cell polyomavirus DNA and expresses two viral transcripts including a truncated form of Large T antigen (LT) and an intact Small T antigen (ST). While LT binds the Retinoblastoma protein and inactivates its tumor suppressor function, it is less clear how ST contributes to MCC tumorigenesis. Here we show that ST specifically recruits the MYC homolog MYCL (L-MYC) to the 15-component EP400 histone acetyltransferase and chromatin remodeling complex. The ST-MYCL-EP400 complex binds to specific gene promoters to activate their expression. Both MYCL and EP400 are required for maintenance of MCC cell line viability and can cooperate with ST to promote gene expression. We demonstrate that ST enhances the interaction between MYCL and the EP400 complex interaction and this activity contributes to transcriptional activation, oncogenesis and reprogramming of MCC.
| Merkel cell carcinoma (MCC) is an aggressive skin cancer with a high rate of mortality. Risk factors for developing MCC include immunosuppression and UV-induced DNA damage from excessive exposure to sunlight [1]. Recognition of the immunosuppressive risk for MCC prompted a search to identify pathogens and led to the discovery of Merkel cell polyomavirus (MCPyV) [2]. MCPyV-positive MCC tumors contain clonally integrated copies of viral DNA and express small T antigen (ST) and a truncated form of large T antigen (LT). Genome sequencing of virus-negative MCC revealed an extremely high number of single nucleotide polymorphisms containing the C>T transition consistent with UV damage [3, 4]. In contrast, MCPyV positive tumors contain very few somatic mutations suggesting that MCPyV ST and LT contribute the major oncogenic activity to MCC development. In all virus-positive MCC cases reported to date, LT has undergone truncations that disrupt viral replication activities but leave the LXCXE, RB-binding, motif intact [5]. While LT can bind and inactivate RB, it is less clear how ST contributes to MCC tumorigenesis [6, 7].
MCPyV ST has oncogenic activity and can contribute to tumor development in mouse models [8, 9]. Similar to ST from other polyomaviruses, MCPyV ST can bind to the A and C subunits of protein phosphatase 2A (PP2A) [10]. However, ST binding to PP2A may not be necessary for its oncogenic activity. A unique domain of ST known as the LT stabilizing domain (LSD) has been reported to bind FBXW7 and CDC20, functions to increase levels of LT, and contributes to its transforming function [11, 12]. ST can activate MTOR signaling and promote the phosphorylation of EIF4EBP1 (4EBP1) [13]. ST can also perturb NFκB and inflammatory signaling [14, 15]. When expressed in fibroblasts, ST led to increased levels of glycolytic genes including the lactate transporter SLC16A1 (MCT1) and induction of aerobic glycolysis also known as the Warburg effect [16]. However, it is not clear how ST contributes to so many distinct activities.
The EP400 histone acetyltransferase complex is involved in multiple biological events including transcription, stem cell maintenance and DNA damage response. The mammalian EP400 complex contains at least 15 distinct components including the large subunits EP400 (also known as p400) and TRRAP plus ACTL6A, BRD8, DMAP1, EPC1 (and its homologue EPC2), ING3, KAT5 (also known as Tip60), MBTD1, MEAF6, MORF4L1 (and MORFL2), MRGBP, RUVBL1 (and RUVBL2), VPS72 and YEATS4 [17–20]. The EP400 complex contains several intrinsic enzymatic activities including EP400 chaperone activity for histone variants H3.3 and H2AZ, KAT5 mediated acetylation of histones H2A and H4, and the DNA helicase activity of RUVBL1 and RUVBL2. TRRAP can bind directly to MYC and has been reported to bind equally well to the homologue MYCN and poorly to MYCL (L-MYC) [21, 22].
Here we demonstrate that MCPyV ST recruits MYCL to the EP400 complex to activate specific gene expression, promote cellular transformation and contribute to its oncogenic potential.
To understand how MCPyV T antigens contribute to MCC oncogenesis, we performed a large-scale immunoprecipitation with a monoclonal antibody (Ab5) specific for the shared N-terminal region of LT and ST to identify associated cellular proteins from lysates of virus-positive MCC cell lines MKL-1 and WaGa (Fig 1A) [23]. Identification of the immunoprecipitated proteins by multi-dimensional protein identification technology (MudPIT) revealed MCPyV LT and ST (Fig 1B and S1 Table) [24]. RB1 and VPS39 were detected as expected given their previously reported association with LT [5, 25]. Both homologues of the PP2A scaffold (PPP2R1A, PPP2R1B) and catalytic (PPP2CA, PPP2CB) subunits were also detected, likely due to association with ST [10, 14]. Unexpectedly, Ab5 also co-precipitated MYCL and MAX as well as all known subunits of the EP400 complex listed above including ACTL6B, a homologue of ACTL6A, and the recently reported MBTD1 [20]. In contrast, MudPIT using Ab3, specific for LT only, identified LT, RB1 and VPS39 and none of the components of the EP400 complex.
To validate the interactions with endogenous proteins, MKL-1 cell lysates were immunoprecipitated with antibodies to MAX, EP400, ACTL6A, EPC1 and VPS72. Each of these antibodies co-precipitated ST, PPP2CA and MYCL as well as several components of the EP400 complex (Fig 1C). MudPIT with antibodies for EP400 identified all 15 subunits including homologs of the EP400 complex plus MYCL, MAX, ST and PP2A (Fig 1B and S1 Table). MudPIT with antibodies for MAX enriched for MYCL, ST, PP2A, all components of the EP400 complex plus several MAD and MAD-associated proteins [26, 27]. MudPIT with an IgG control antibody detected small amounts of RUVBL1, RUVBL2, MEAF6 and ACTL6B but none of the other EP400 complex components. Therefore, antibodies for MAX, EP400 and MCPyV ST each specifically co-precipitated MYCL, the EP400 complex, ST and PP2A (Fig 1B and S1A Fig).
To determine if ST could form a single complex with MYCL and the EP400 complex, we performed gel filtration of MKL-1 cell lysates [28]. Fractions #5–7 contained protein complexes of ≥ 2 MDa with ST, MYCL, MAX, several EP400 complex components and EP400 itself (Fig 1D). EP400 was only detected in the large complex fraction while other subunits of the complex including TRRAP, KAT5, RUVBL2, DMAP1 and ING3 were present in the large complex and in fractions with smaller sized complexes. Of note, MYCL1 isoform 1 (i1) was present in the ST-containing fractions #5–7 whereas the larger MYCL i3 was detected in intermediate sized fractions and the shortest form (i2) in smaller size fractions (Fig 1D and 1E, S1B and S1C Fig). An immunoprecipitation for MAX with lysates from fraction #5 co-precipitated EP400, TRRAP and ST (Fig 1F). In contrast, MAX co-precipitated TRRAP and ST but not EP400 from fraction #13 and neither TRRAP or EP400 from fraction #21. This indicates that a specific fraction of MAX binds to EP400, a key component of the ST-MYCL complex [29].
To determine the contribution of MCPyV ST binding to MYCL and the EP400 complex in MCC, we transduced MKL-1 cells with lentiviral shRNAs targeting both LT and ST (shPanT) or ST only (shST) [13, 30]. Expression of either shRNA but not scrambled shRNA (shScr) led to reduced levels of ST and MYCL (Fig 2A, S2A Fig). Reduced levels of ST led to decreased ability of MAX to co-precipitate EP400, TRRAP, DMAP1 and YEATS4 and reduced the ability of EP400 to co-precipitate MYCL and MAX. Of note, when ST levels were reduced, EP400 retained the ability to bind to other components of the EP400 complex including TRRAP, DMAP1 and YEATS4.
To determine if MCPyV ST could increase the ability of MAX to bind to the EP400 complex, we introduced ST or C-terminal HA-tagged ST into HCT116 cells and a virus-negative MCC cell line UISO. Immunoprecipitation for MAX from parental HCT116 and UISO cell lysates readily co-precipitated MYC but not any EP400 complex components (Fig 2B). However, in the presence of MCPyV ST, MAX efficiently co-precipitated TRRAP, EP400, DMAP1 and KAT5. Of note, MYCL levels increased in HCT116 and UISO cells when ST was expressed (Fig 2B, Input). Stable expression of ST in primary human foreskin fibroblasts (HFF) also increased levels of MYCL (S2B Fig).
To determine if ST interaction with the EP400 complex was specific, we generated a series of ST mutants and stably expressed them in HCT116 cells. We focused the mutagenesis on a region of MCPyV ST within the unique domain that is not well conserved with ST from other human polyomaviruses and includes the LSD motif (residues 91–95) (Fig 2C, S2C Fig). Immunoprecipitation for ST containing alanine substitutions of residues 83 to 88 (83-88A) showed decreased binding to EP400 complex components while retaining strong binding to PP2A (Fig 2D, S2D Fig). Within this region, substitution of E86 and E87 with serine (E86S, E87S referred to as 2M) led to reduced EP400 complex binding yet retained some PP2A binding. In contrast, alanine substitution of residues 83 to 95 (83-88A, 90-95A, 93-95A) or 102 to 105 (102-105A) led to increased levels of ST relative to WT ST (Fig 2D, S2D Fig). Within this region, substitution of K92 and D93 with serine (K92S, D93S, referred to as 3M) led to increased levels of MCPyV ST (Fig 2D, S2D Fig). Combining the 2M and 3M mutants to create 4M (E86S, E87S, K92S, D93S) resulted in a ST construct that expressed at levels higher than WT ST and retained PP2A binding, but was unable to co-precipitate the EP400 complex. To test the ability of these ST constructs to promote MAX binding to the EP400 complex, we performed an IP with MAX antibodies. WT, 102-105A, and 3M ST led to increased ability of MAX to co-precipitate the EP400 complex and PP2A, while the 2M and 4M mutants were not co-precipitated by MAX and did not enable MAX to co-precipitate the EP400 complex or PP2A (Fig 2D).
We observed that MCPyV ST binds specifically to MYCL and the EP400 complex. However, it was not clear if any of these factors were required for proliferation. To identify essential genes in MKL-1 cells, we performed a CRISPR-Cas9 screen of 18,493 genes using two pooled sgRNA libraries H1 and H2, each containing 5 unique sgRNAs for each gene. Using the MAGeCK-VISPR analysis pipeline, Gene Set Enrichment Analysis (GSEA) of known human housekeeping genes revealed that these genes were significantly negatively correlated with the results of the CRISPR screen [31] (S3A Fig). After accounting for copy number variations in MKL-1 cells, we identified 481 genes that were negatively selected in the CRISPR-Cas9 screen with a false discovery rate (FDR) < 0.05 [32], of which 276 have been classified as housekeeping genes (S3B and S3C Fig, S2 Table) [33]. Among the 205 genes not classified as housekeeping genes, 79 genes were identified with FDR < 0.01 and the remaining 126 genes were identified with FDR > 0.01 but < 0.05. MYCL, EP400 and RUVBL2 were identified as essential (FDR < 0.05, Fig 3A). Additional components of the EP400 complex were identified in the CRISPR-Cas9 negative selection screen with p-values < 0.05 but with higher FDR values and included KAT5, TRRAP, DMAP1, ING3 and YEATS4.
Given the requirement of MYCL for viability of MKL-1 cells in the CRISPR-Cas9 screen and the presence of MYCL in the ST-EP400 complex in both MKL-1 and WaGa cell lines (Fig 1B and S1 Table), we examined the levels of the three MYC family members in MCC cell lines. Six different virus-positive MCC cell lines that expressed ST and LT also expressed MYCL while some had low levels of MYCN and none expressed full length MYC (Fig 3B). For controls, we tested HCT116 cells that predominantly expressed MYC and Kelly neuroblastoma cells that expressed MYCN. The virus-negative MCC cell line UISO did not have detectable levels of MYCL until a C-terminal epitope tagged ST (ST-CT) was introduced (Fig 3B).
To determine if MYCL was required for ST to binding to the EP400 complex, we generated MKL-1 cells that contained doxycycline (Dox) inducible shRNA (shMYCL) or miRNA (mirMYCL) that specifically targeted MYCL. Expression of shMYCL or mirMYCL led to reduced levels of MYCL and decreased MAX co-precipitation of EP400, TRRAP, KAT5 and ST (Fig 3C). Notably, depletion of MYCL reduced the ability of ST to co-precipitate the EP400 complex and reduced EP400 binding to ST (Fig 3C).
Omomyc is a modified fragment of MYC that can bind to MAX and disrupt endogenous MYC-MAX heterodimers [34]. To test if MYCL-MAX heterodimers were necessary for ST interaction with the EP400 complex, we introduced a Dox-inducible, HA-tagged, Omomyc construct into MKL-1 cells. When expressed, HA-Omomyc co-precipitated MAX as expected but not MYCL, ST or subunits of the EP400 complex and led to decreased levels of both MAX and MYCL (Fig 3D). However, ST retained the ability to co-precipitate components of the EP400 complex but not MYCL or MAX when Omomyc was expressed, indicating that ST can bind to the EP400 complex independent of the MYCL/MAX heterodimer. The viability of MKL-1 cells was decreased when MYCL levels were depleted by shMYCL or mirMYCL and when the MYCL-MAX heterodimer was disrupted by Omomyc (Fig 3E).
To determine regions of MYCL that contributed to ST and EP400 binding, a series of HA tagged C-terminal constructs of MYCL was stably expressed in HCT116 cells in the presence or absence of ST. When ST was present, MYCL robustly co-precipitated TRRAP, EP400, YEATS4, DMAP1 as well as ST (S3D Fig). In the absence of ST, WT MYCL co-precipitated MAX but only weakly bound to EP400 complex components. In contrast, the N-terminal 165 residues of MYCL, unable to bind MAX, could co-precipitate ST and the EP400 complex in the presence or absence of ST (S3D Fig). The N-terminal 165 residues of MYCL contains several highly conserved MYC homology boxes (MB) that function to bind MYC modifying proteins [35]. MB1 binds to FBXW7 and MB2 contributes to TRRAP binding (S1 Fig) [21, 36]. We generated HCT116 cells that stably expressed HA-tagged MYCL full length constructs with small in-frame deletions of MB1 or MB2. An HA IP for ΔMB2 MYCL co-precipitated ST and MAX but not the EP400 complex, while the ΔMB1 MYCL co-precipitated MAX but neither ST or the EP400 complex (S3D Fig). These data indicate that MB1 and MB2 of MYCL contribute to ST and EP400 complex binding.
To test the requirement for EP400 in virus-positive MCC, we generated MKL-1 cell lines containing three different dox-inducible shRNAs targeting EP400 (Fig 4A). In the presence of dox, levels of EP400 were reduced and an immunoprecipitation for EP400 was unable to co-precipitate DMAP1 or MAX (Fig 4A). Of note, knockdown with shEP400-1 led to decreased levels of ST and MYCL in addition to lower levels of EP400 (Fig 4B). In contrast, shEP400-2 and shEP400-3 reduced EP400 levels but did not affect ST and MYCL levels. When EP400 levels were reduced by shEP400-2 or -3, MAX and ST retained the ability to co-precipitate each other, as well as MYCL and TRRAP but not DMAP1 or YEATS 4 (Fig 4B). The viability of MKL-1 cells decreased significantly upon depletion of EP400 with each of the 3-inducible shRNA (Fig 4C).
To determine if the effect of EP400 depletion was specific to virus-positive MCC cell lines, we introduced the 3 inducible shEP400 constructs into the virus-negative MCC line UISO. We confirmed that knockdown of EP400 led to reduced levels of EP400 in UISO cells (Fig 4D). In contrast to the MCPyV-positive MCC cell line MKL-1, the viability of UISO cells was unaffected by EP400 knock-down (Fig 4E). Given that the Kelly neuroblastoma cell line is dependent on continued expression of MYCN, we tested its sensitivity to EP400 depletion [37]. Depletion of EP400 in Kelly cells by shEP400-1 led to reduced binding of MAX to DMAP1 and YEATS4 while retaining binding to MYCN and TRRAP (Fig 4F). Of note, shEP400-1 did not affect MYCN levels in Kelly cells. As shown in Fig 4G, Kelly cells had reduced viability when EP400 levels were reduced.
Expression of MYC or MYCL together with OCT4, SOX2 and KLF4 (OSK) can generate induced pluripotent stem (iPS) cells from a variety of somatic cell types [38, 39]. Furthermore, MYC interaction with the EP400 complex has been implicated in the generation and maintenance of embryonic stem (ES) and iPS cells [40, 41]. Given that MCPyV ST can bind to MYCL and the EP400 complex, we tested its ability to contribute to iPS cell generation. Since keratinocytes have higher reprogramming efficiency compared to other cell types due to lower p53 and p21 protein levels [42], we generated hTERT-immortalized human keratinocytes with an inducible OSK expression vector and stably introduced MYCL, ST or ST mutants. Expression of OSK in the presence of MYCL, ST or 3M led to the appearance of flat human ES cell-like colonies with defined borders that could be stained by alkaline phosphatase and ES cell surface markers TRA-1-60 and TRA-1-81 (Fig 5A–5C) [43]. In contrast, the ST EP400-binding defective 2M and 4M mutants were unable to generate iPS cells. These results indicate that ST binding to MYCL and the EP400 complex was able to cooperate with OSK to promote the generation of iPS cells.
We examined if ST mediated transformation was dependent on interaction with MYCL and the EP400 complex. When a MCC tumor-derived MCPyV early region (E) that encoded truncated LT and wild type ST was expressed in IMR90 human diploid fibroblasts, we observed a senescent phenotype with elevated levels of p53 and p21 [23]. To suppress this phenotype, we stably expressed a dominant negative form of p53 (p53DD abbreviated as P) and hTERT (H) in IMR90 cells to generate PH cells [44]. The PH cells tolerated MCPyV early region with wild type ST (PHE), 3M ST (PH3) or 4M (PH4) mutant ST, and exogenous MYCL (PHL) without undergoing senescence. Immunoprecipitation of ST with Ab5 from PHE cell lysates revealed a weak interaction with DMAP1, a component of the EP400 complex (Fig 5D). However, when MYCL was co-expressed with wild type ST in PHEL cells, ST and MAX readily co-precipitated the EP400 complex. The 3M ST mutant could efficiently co-precipitate the EP400 complex even without exogenous MYCL expression (PH3). In contrast, the 4M ST mutant (PH4) was unable to co-precipitate the EP400 complex.
We tested the ability of these fibroblasts to grow in an anchorage-independent manner when cultured in soft agar. IMR90 cells expressing p53DD and hTERT (PH) alone or with MYCL (PHL) were unable to form colonies. Cells expressing the MCPyV early region (PHE) formed a few colonies, while co-expression of MYCL (PHEL) led to an increased number of soft agar colonies (Fig 5E and 5F). The highly expressed 3M ST mutant (PH3) could induce anchorage-independent growth while the 4M mutant (PH4) failed to form soft agar colonies. The number of colonies formed for each cell type reflected the relative binding of ST and MAX to the EP400 complex in the presence of the various ST constructs (compare Fig 5D and 5F).
Given the known chromatin binding activities of MYCL and the EP400 complex, we tested if MCPyV ST could bind specifically to DNA. We performed chromatin immunoprecipitation followed by sequencing (ChIP-seq) with the mass spectrometry-validated antibodies to EP400 and ST. Since no antibody suitable for immunoprecipitation or ChIP was available for MYCL, we performed ChIP-seq with the MAX antibody. Although it has been reported that MCPyV truncated LT does not bind to chromatin, we considered the possibility that ChIP with Ab5 might also enrich for chromatin-bound LT [6]. To account for this possibility, we generated a MCC cell line (MKL-1) derivative that stably expressed MCPyV ST with a C-terminal HA epitope tag and performed ChIP with an HA antibody. Replicas of MAX and EP400 ChIP-seq identified many peaks that were also identified by anti-ST (Ab5) and anti-HA ChIP-seq. Common gene targets were identified by assigning peaks to the nearest genes (Fig 6A and S4A Fig). De novo DNA motif analysis identified the MYC target E-box sequence CACGTG as the most frequently observed motif with Z-scores -42.1726, -20.0773, -23.9634, -19.137 for MAX, EP400, ST-HA and Ab5 antibodies respectively (Fig 6B). Peaks were highly enriched for promoters and 5’UTR sequences (Fig 6C and S4B Fig). ChIP for MAX, EP400 or ST followed by re-ChIP for these three factors indicated that they could bind simultaneously to DNA (S4C Fig).
Given the strong enrichment for promoters, we performed ChIP-seq with antibodies to histone H3 modified by lysine 4 trimethylation (H3K4me3), a histone mark enriched at actively transcribed gene promoters [45]. H3K4me3 ChIP-seq identified 20,222 peaks with MAX, EP400 and ST centered on the same peaks (Fig 6D and 6E). These results indicate that MAX, EP400 and MCPyV ST bind as a complex specifically to E boxes near the transcription start sites (TSS) of actively expressed genes.
We validated the ChIP-seq experiments on several promoters. We prepared chromatin from MKL-1 cells after transduction with vectors expressing shMYCL, mirMYCL or controls and performed ChIP with Ab5. As shown in S5A–S5C Fig. we observed ST binding to the MYCL gene as well as three additional gene promoters that were significantly reduced by MYCL depletion. We also prepared chromatin from MKL-1 cells containing the inducible shEP400-1 before and after dox addition. We observed strong MAX binding to several gene promoters that was reduced upon EP400 depletion (S5D Fig).
To identify genes and associated biological functions that are controlled by the ST-MYCL-EP400 complex, we performed RNA-seq of MKL-1 cells containing inducible shMYCL, shEP400-2, shEP400-3 and shScr with RNA isolated from cells treated with dox for 5 days. The differentially expressed genes (DEG) list consists of 2157 genes that passed the cutoff Padj < 0.001 in all three comparisons (shEP400-2, shEP400-3 and shMYCL vector, relative to shScr control). To create heatmaps, counts were normalized separately for the two experiments (shEP400 and shMYCL) and then corrected for batch effect using ComBat (S6 Fig) [46]. These genes were first grouped into 62 clusters using model-based clustering [47]. The average expression profiles of each cluster were then merged into four general patterns of behavior using hierarchical clustering (Fig 7A and S3 Table; see Methods for more details). The genes in each of the four merged clusters were evaluated for statistical enrichment in Gene Ontology (GO) biological process terms. Cluster membership and all results of the GO term analysis are presented in S3 Table. We observed that genes upregulated by shEP400 and shMYCL fell into the cluster DEG-CL2 and were enriched in neurogenesis, skin development and hair cycle. DEG-CL4 contained genes downregulated by EP400 and MYCL and were enriched in cellular component biogenesis, RNA processing and amide biosynthetic process. Two smaller clusters represent genes that behaved differently under shEP400 and shMYCL conditions. DEG-CL1 genes were decreased by shEP400, increased by shMYCL and enriched for actin cytoskeleton and regulation of signaling. DEG-CL3 exhibited the opposite pattern of expression and was enriched in nerve development and liposaccharide biosynthesis. These results show that both MYCL and EP400 support cell growth by upregulating bulk synthesis of biomolecules including ribosomes and proteins while simultaneously repressing cell adhesion and developmental programs in neurogenesis and skin.
To integrate expression profiling with the aforementioned ChIP-seq experiments (Fig 6), we performed Binding and Expression Target Analysis (BETA) that links the proximity of the ChIP-seq binding peaks to the TSS with expression level changes in the corresponding genes to predict activating and repressive activities of transcription factors [48] (Fig 7B). We performed BETA analysis for MAX, EP400 and ST ChIP-seq studies with RNA-seq analysis for shEP400-2, shEP400-3 and shMYCL. We observed that the genes whose levels decreased (downregulate) upon EP400 or MYCL depletion were significantly enriched for MAX, EP400 and ST chromatin binding (S7A Fig, S4 Table). In contrast, genes whose levels increased (upregulate) with EP400 depletion were not significantly associated with the MAX, EP400 and ST ChIP-peaks. This indicates that the ST, MYCL/MAX and EP400 complex binding contributes to specific gene activation. We compared the target genes identified for each ChIP-seq analysis with the RNA-seq analysis for shEP400-2, shEP400-3 and shMYCL and identified 951 shared target genes of MAX, EP400 and ST whose levels went down upon EP400 or MYCL depletion and had significant evidence for direct ChIP binding by BETA analysis (BETA3, Fig 7C & S7B Fig). When the RNA-seq data for shEP400-1 was also included in the analysis, a total of 379 target genes were identified (BETA4, S7A–S7C Fig).
We then examined 951 genes identified in BETA3 that were downregulated by shEP400 and shMYCL with evidence of direct binding according to BETA analysis of the ChIP-seq data. We note that these genes exhibited a wide range of fold changes upon depletion of EP400 and MYCL, with 136 of the 951 genes showing greater than 2-fold downregulation due to shEP400 (shEP400 inverse fold change), and 62 out of 951 genes showing greater than 2-fold downregulation due to shMYCL (S8 Fig and S5 Table). To find global patterns of expression that reflected functional regulation, we centered and scaled their expression profiles and created model-based clusters and merged clusters, using the same procedure as in the analysis of the DEG list. The final merged clusters were then evaluated for GO term enrichment. Cluster membership of the BETA3 genes and full GO term enrichment results are listed in S3 Table. We found that these genes naturally divide into two groups: genes that were more strongly affected by shEP400 (BETA3-CL1 and 2) and genes that were more strongly affected by shMYCL (BETA3-CL3 and 4) (Fig 7D). The shEP400 clusters are enriched for nucleobase-containing compound metabolic process and translation initiation and elongation whereas the shMYCL clusters are involved in RNA processing and peptide metabolic processes.
Among the target genes identified in the shEP400-2, -3 and shMYCL depletion analyses was the translational control factor 4EBP1 that has been reported to be upregulated by ST [13]. To test this effect, lysates were generated from MKL-1 cells before or after depletion of EP400 and MYCL were blotted for 4EBP1. As expected, levels of MYCL were depleted by shMYCL and EP400 by shEP400-2 and -3 (Fig 7E). Of note, levels of 4EBP1 and the phosphorylated serine residue 65 form (pS65-4EBP1) were reduced upon EP400 or MYCL knockdown. In addition, we have recently reported that levels of the lactate transporter MCT1 (SLC16A1) increase upon expression of ST [16]. Levels of MCT1 were also decreased upon depletion of EP400 or MYCL (Fig 7E).
We compared the effect of depleting EP400 or MYCL in the virus-positive MKL-1 cells to the effect of expressing ST in normal cells. We examined RNA-seq profiles from IMR90 human fibroblasts with inducible expression of GFP or MCPyV ST over the course of 4 days [16]. As shown in the heatmap in S9 Fig, the genes that were downregulated by shEP400 and shMYCL in MKL-1 cells tend to be upregulated by ST in IMR90 cells consistent with the model that ST activates functional interactions with EP400 and MYCL and their transcriptional targets.
Here, we demonstrate that MCPyV ST specifically recruits the MYCL and MAX heterodimer to the 15-component EP400 complex. These interactions are essential for the transforming function of MCPyV ST, the viability of virus-positive MCC cells and likely to be a major contributor to the oncogenic potential of MCPyV in MCC. Consistent with this model, a genome-wide CRISPR-Cas9 screen revealed that MYCL and several components of the EP400 complex were essential for viability of the virus-positive MCC cell line MKL-1. The interaction of MCPyV ST with MYCL and the EP400 complex is unique to the family of polyomaviruses. To date, no other polyomavirus ST has been reported to bind the EP400 complex or a MYC homolog [15]. Several other viruses have been reported to engage the EP400 complex. Perhaps most similar to the results reported here, the adenovirus E1A oncoprotein binding to MYC and the EP400 complex contributes to its transforming potential [49, 50].
We observed a striking relationship between MYCL and MCPyV ST. MYCL was expressed in all 6 virus-positive MCC cell lines tested. Furthermore, ST appeared to regulate MYCL levels. For example, introduction of ST into several naïve cell lines led to increased levels of MYCL. Conversely, depletion of ST from MKL-1 cells led to decreased levels of MYCL. ST together with EP400 and MAX could bind to the MYCL promoter. In addition, the virus-positive MKL-1 cell line was sensitive to Omomyc expression indicating that the MYCL-MAX heterodimer was required for viability as well as ST interaction. These results are consistent with a positive feedback loop where ST binding to the MYCL promoter contributes to transcriptional activation of MYCL leading to increased levels of MYCL that in turn binds to ST and the EP400 complex.
MYCL has been the forgotten MYC species, since it is not required for normal mouse development and expression is highly restricted in adult tissues [51]. Despite its relative obscurity, MYCL can function as a bona fide oncogene [52, 53]. For example, amplification of MYCL or the presence of the RLF-MYCL1 fusion gene are mutually exclusive of amplifications of MYC or MYCN in small cell lung cancer. Better still, amplification of the 1p34 locus containing the MYCL gene has been reported in MCC suggesting that Merkel cell carcinogenesis is dependent upon excessive MYCL function [54]. Consistent with this report, we estimate that the copy number of MYCL was 3.5 copies in MKL-1 cells based on analyzing ChIP-seq input DNA (S3B Fig, S2 Table). In keeping with this notion, MCPyV ST shows a strong preference for recruiting MYCL to the EP400 complex. It is possible that, despite the widespread presence of MCPyV in healthy individuals, only cells capable of expressing MYCL are susceptible to the oncogenic potential of MCPyV [55, 56].
MYC and MYCL can cooperate with the OSK reprogramming factors to induce a pluripotent state in somatic cells [39, 57]. Comparison of the contributions of MYC to transformation and iPS cell generation show significant overlap with the interaction with the EP400 complex as a key component [41]. Here we find that MCPyV ST can substitute for MYCL in iPS cell generation and that this activity is strictly dependent upon ST interaction with the EP400 complex. Our data indicate that, at least in part, MCPyV ST functions similarly to MYC by binding to the EP400 complex, recruiting it to specific promoters to transactivate gene expression and thereby promoting the generation of iPS cells. These functions may also prove to be critical in establishing and maintaining the oncogenic state of MCC.
Our data reveal that the ST-MYCL-EP400 complex functions, at least in part, to activate specific gene expression. Depletion of MYCL and EP400 led to significant changes in gene expression and cell viability. Those genes whose levels decreased upon MYCL and EP400 depletion were significantly associated with ST, MAX and EP400 binding to their promoters and include classic MYC targets involved in RNA processing, ribosome biogenesis, nitrogen compound and peptide metabolic processes. Additional target genes are involved in cell morphogenesis and signaling in the TNF, WNT, NFκB and DNA damage pathways. Importantly, a large number of metabolic genes were activated by the ST-MYCL-EP4000 complex including a number of transporters including SLC16A1 and SLC7A5 and the MYC-metabolism genes MLX and MLXIP (Mondo) [16, 58]. Factors that promote transcription elongation were also highly enriched including EIF4E, EIF4EBP1, EIF5A [13].
Interestingly, genes whose levels increased upon MYCL or EP400 depletion were involved in neurogenesis, axon guidance, wound healing and cell-cell adhesion. These results can be interpreted to indicate that ST-MYCL-EP400 complex serves to repress differentiation markers and induce a more primitive, progenitor or embryonic state, consistent with its ability to generate iPS cells.
The MYC family functions to activate gene expression at least in part by interaction with a variety of chromatin factors. In addition to the EP400 complex, MYC can bind to the TRRAP-containing STAGA (SPT3-TAF9-GCN5 acetylase) complex that in turn interacts with Mediator [59]. MYC binds to BRD4 and the pTEFb complex to facilitate transcriptional elongation by release of paused RNA polymerase II [60, 61]. The conserved Myc Boxes contribute to transformation with the Myc Box 3b (MB3b) binding to WDR5 and Myc Box 4 (MB4) binding to HFCF1 (S1 Fig) [62, 63]. MB3a, or simply referred to as MB3, found only in MYC and MYCN and not MYCL, is required for tumorigenic activity of MYC in vitro and in vivo [64] and contributes to transcriptional repression by recruiting HDAC3 [65]. At oncogenic expression levels, MYC interacts with MIZ-1 (ZBTB17) to repress transcription, which can be disrupted by mutating valine 394 (V394) in the helix-loop-helix (HLH) domain [66]. We only detected the EP400 complex and did not detect any of these other MYC binding factors in any of the ST complexes. Both MB1 and MB2 of MYCL contribute to ST and MYCL binding. Of note, it appears that ST and MYCL bind directly to TRRAP as evidenced by co-precipitation of TRRAP with ST and MAX antibodies after EP400 depletion (Fig 4B).
In contrast to MCPyV ST, transformation by SV40 ST is strictly dependent on its interaction with PP2A. SV40 ST binding to PP2A perturbs its ability to de-phosphorylate certain substrates including MYC that in turn leads to higher levels of MYC. While it has been reported that MCPyV ST interaction with PP2A is not required for its transforming function, it is possible that PP2A contributes to activity of the MYCL-EP400 complex.
These results highlight an important mechanism for MCPyV ST mediated transformation. The ST-MYCL-EP400 complex functions as a powerful engine to transactivate gene expression and promote oncogenesis. Important questions to be pursued include whether any of the specific downstream transcriptional targets of the ST-MYCL-EP400 complex contribute to MCC oncogenesis and if any of these target genes provide a therapeutic opportunity for virus-positive MCC. In addition, the ST-MYCL-EP400 complex itself may provide a therapeutic opportunity to disrupt interactions between ST, MYCL and the EP400 complex or with any of its components to DNA. In addition, these results may help to explain why the oncogenic activity of MCPyV ST is limited to MCC because of its dependency on MYCL or if this virus is capable of inducing MYCL and cancers in other tissue types.
Human Subject Research performed in this study complied with all relevant federal guidelines and institutional policies. The Dana-Farber Cancer Institute Institutional Review Board (IRB) approved the study. All subjects were adults and provided informed written consent.
MCC cell lines MKL-1, MKL-2 and MS-1 were gifts from Masa Shuda (University of Pittsburgh, PA); MCC cell lines WaGa and UISO from Jürgen Becker (Medical University Graz, Austria); MCC cell lines PeTa and BroLi from Roland Houben (University of Wuerzburg, Germany). Kelly neuroblastoma cell line was a gift from Rani George (Dana-Farber Cancer Institute, MA). 293T, HCT116 and IMR90 cells were obtained from ATCC. HFK-hTERT cells were a gift from Karl Münger (Tufts University, MA).
MCPyV early region was PCR amplified from DNA extracted from a Merkel cell carcinoma sample [23]. The cDNA for ST was modified to eliminate the LT splice donor by introducing silent mutations (GAG|GTCAGT to GAa|GTCtcc). Additional ST mutants were generated using QuikChange Lightning Site-Directed Mutagenesis Kit (Agilent).
The EP400, MYCL shRNA target sequence was designed using Block-iT RNAi Designer (Life Technologies) and annealed forward and reverse oligos of hairpin sequence were cloned between AgeI/EcoRI sites of the doxycycline inducible shRNA vector Tet-pLKO-puro (a gift from Dmitri Wiederschain, Addgene #21915) [67]. The MYCL miRNA target sequence was designed using Block-iT RNAi Designer and cloned into pcDNA 6.2-GW/EmGFP-miR vector (Life Technologies) and the pre-miRNA expression cassette targeting MYCL was transferred to pLIX_402 Dox-inducible expression vector via consecutive BP and LR recombination reactions to generate pLIX-mirMYCL plasmid. shRNAs constitutively expressed from lentiviral PLKO vector targeting MCPyV LT/ST (shPanT), ST (shST) or scramble (shScr) have been published before [13, 30, 68].
pMXs-Hu-L-Myc was a gift from Shinya Yamanaka (Addgene # 26022) [39]. MYCL was PCR amplified with C-terminal 3xHA tag or with original stop codon and cloned into pLenti-CMV gateway vector. Omomyc was a kind gift from Sergio Nasi (Sapienza University of Rome, Italy), modified by PCR amplification to include C-terminal HA tag and cloned into pLIX_402. The OCT4-2A-SOX2-2A-KLF4 polycistronic coding sequence was PCR amplified from pKP332 Lenti-OSK1 (Addgene #21627) [69] and cloned into pLIX_402.
Expression vectors include pLenti-CMV (a gift from Eric Campeau, Addgene #17451) [70], doxycycline inducible lentiviral gateway expression vector pLIX_402 (a gift from David Root, Addgene #41394). Lentiviral packaging plasmid psPAX2 and envelope plasmid pMD2.G were gifts from Didier Trono (Addgene #12260, #12259). Retroviral packaging plasmid pUMVC3 was a gift from Robert Weinberg (Addgene # 8449) [71] and envelope plasmid pHCMV-AmphoEnv from Miguel Sena-Esteves (Addgene # 15799) [72]. Retroviral plasmids pBabe-neo-p53DD and pBabe-hygro-hTERT were previously described [44].
Packaging and envelope plasmids were co-transfected with lentiviral or retroviral expression vectors into 293T cells using Lipofectamine 2000 (Life Technologies). Two days after transfection, 293T cell supernatant was purified with 0.45 μm filter and supplemented with 4 μg/ml polybrene before transducing recipient cells. Stable cell lines were generated after selection with 1–2 μg/ml puromycin, 5–10 μg/ml blasticidin, 500 μg/mL neomycin, and 100 μg/mL hygromycin as required by each vector.
CellTiter-Glo Luminescent Cell Viability Assay was performed according to the protocol from Promega. Basically, 3000 MKL-1 parental or dox-inducible cells were plated in 96 well plate. Fresh medium was supplemented every two days with or without doxycycline. The number of days that cells had been treated with doxycycline was labelled on X-axis. At the end of time course, CellTiter-Glo reagents were added to lyse cells. For each cell line, doxycycline treated samples were normalized to untreated samples.
Anchorage independent growth was performed as described [23] using 6-well dishes with SeaPlaque Agarose (Lonza) at concentrations of 0.3% top and 0.6% bottom layers. Agarose was diluted with 2X MEM (Gibco) supplemented with 2X Gluta-max (Gibco), 2X pen-strep (Gibco), and 30% FBS. IMR90 cells (105) were seeded in triplicate in the top agarose layer. Wells were fed with top agarose twice per week. After 4 weeks, cells were stained with 0.005% crystal violet (Sigma) in PBS and colonies were counted. Statistical significance was determined by ordinary one-way ANOVA for multiple comparisons with p < 0.05.
HFK-hTERT cells were transduced with pLIX-OSK and selected with puromycin to establish the parental cell line (P) followed by transduction with MYCL or ST in pLenti-CMV vector and selection with blasticidin. 200,000 cells were seeded in Matrigel (BD Biosciences) coated 6-well plate in triplicate on day 0 in Keratinocyte-SFM medium (Gibco) supplemented with 0.5 μg/ml doxycycline. On day 3, medium was changed to mTeSR1 (Stemcell Technologies) supplemented with doxycycline. iPS colonies were visible under microscope after 3 weeks and stained with StainAlive TRA-1-60 or TRA-1-81 antibodies (Stemgent) and Alkaline Phosphatase Detection Kit (Millipore).
The following antibodies were used: Ab5 and Ab3 [23, 73]; HA (Abcam); EP400, RUVBL2 (Bethyl); MAX, KAT5, DMAP1, MNT (Santa Cruz); MYCL (R&D Systems); ING3 (Sigma); PPP2CA (BD Biosciences); and H3K4me3 (Millipore; 07–473).
Cell lysates were prepared in EBC Lysis buffer (50 mM Tris pH 8.0, 150 mM NaCl, 0.5% NP-40, 0.5 mM EDTA, 1 mM β-Mercaptoethanol and freshly added protease inhibitor and phosphatase inhibitor cocktail). Immunoprecipitations were performed with protein G Dynabeads (Life Technologies) mixed with immunoprecipitation antibodies or anti-HA magnetic beads (Pierce Biotechnology). After overnight incubation on a rotating apparatus at 4°C, magnetic beads were washed with high salt wash buffer (50 mM Tris pH 7.4, 300 mM NaCl, 0.5% NP-40, 0.5 mM EDTA) five times. Bound proteins were eluted from magnetic beads with 2x Laemmli sample buffer (Bio-Rad). After electrophoresis, the separated proteins were transferred to PVDF membrane and blotted. Immunoblots were developed using Clarity Western ECL substrate (Bio-Rad) and imaged with G:BOX Chemi system (Syngene).
MudPIT was performed with MKL-1 or WaGa suspension cells (30 x 15-cm diameter plates) harvested in 30 ml EBC lysis buffer. Clarified cell extract (100–300 mg) was incubated overnight at 4°C with 30 μg antibodies crosslinked to 30 mg protein G agarose beads by dimethyl pimelimidate (DMP). Beads were washed with high salt wash buffer five times, then eluted with 0.2 M glycine pH 3 and neutralized with 1 M Tris pH 8.0. Proteins were precipitated with 1/5 TCA overnight at 4°C and washed with cold acetone twice and analyzed by MudPIT as described [74]. The triple-phase fused-silica microcapillary column was packed with 8–9 cm of 5-μm C18 Reverse Phase (Aqua, Phenomenex), followed by 3 to 4 cm of 5-μm Strong Cation Exchange material (Partisphere SCX, Whatman) and 2 to 3 cm of C18 RP and equilibrated with Buffer A (5% ACN, 0.1% Formic Acid). A10-step chromatography run was performed with the last two chromatography steps consisting of a high salt wash with 100% Buffer C (500mM Ammonium Acetate, 5% ACN, 0.1% Formic Acid) followed by an acetonitrile gradient to 100% Buffer B (80% ACN, 0.1% Formic Acid). 2.5 kV voltage was applied distally to electrospray the eluting peptides. Full MS spectra were recorded on the peptides over a 400 to 1,600 m/z range, followed by five tandem mass (MS/MS) events sequentially generated in a data-dependent manner on the first to fifth most intense ions selected from the full MS spectrum (at 35% collision energy).
A frozen pellet of MKL-1 cells was resuspended in mammalian cell lysis buffer (MCLB; 50mM Tris pH 7.8, 150 mM NaCl, 0.5% NP40) in the presence of protease and phosphatase inhibitors (Roche Complete, EDTA-free Protease Inhibitor Cocktail and 25 mM sodium fluoride, 1 mM sodium orthovanadate, 5 mM β-glycerophosphate). The lysate was incubated on ice for 15 minutes then clarified by centrifugation in a refrigerated microfuge for 10 minutes at top speed. The supernatant was further clarified using 0.45 μM Durapore PVDF spin filters (Millipore). Approximately 7 mg of total cellular protein was applied to a Superose 6 10/300 GL column run in an AKTA pure FPLC (GE Healthcare) with MCLB as the running buffer. The injection volume was 500 μl, the flow rate was 0.5 ml/minute, and 0.5 mL fractions were collected from 0.2 column volumes to 1.5 column volumes. The molecular weights were estimated by loading 1 mg of individual protein standards from the Gel Filtration Markers Kit for Protein Molecular Weights 29,000–700,000 Da (Sigma-Aldrich).
MCV T antigens were knocked down in MKL-1 cells using shRNAs previously published [13, 30]. The shRNAs were cloned into pLKO.Puro vectors, lentivirus was generated in 293T cells using psPax2 and pVSV.G vectors, and MKL-1 cells were infected using spinoculation (centrifugation at 800g for 30 mins with viral supernatants) followed by infection overnight in the presence of 1 μg/ml Polybrene. 24 hours post infection, MKL1 cells were spun down and resuspended in medium containing puromycin (1 μg/ml). Cells were harvested after 72 hours and processed for immunoblotting and immunoprecipitation.
CRISPR lentiviral libraries H1 and H2 each contain 92,817 pooled sgRNAs targeting 18,493 human genes. CRISPR screen was performed by following a previous protocol (https://doi.org/10.1101/106534). Briefly, 2x108 MKL-1 cells were transduced with H1 and H2 CRISPR libraries separately at MOI 0.3 to ensure single sgRNA incorporation per cell. After 6 days of 1 μg/ml puromycin selection, surviving cells from each sgRNA library transduction were split in half, 3x107 cells were saved as initial state controls, the rest were cultured for a month with at least 3x107 cells maintained and used as final state samples. Genomic DNA was extracted and 200 μg from each sample were used to PCR amplify integrated sgRNAs and to generate 4 libraries for next generation sequencing. 50 million reads were obtained for each sequencing library. To filter out false positive targets due to strong correlation between decreased cell viability and increased gene copy number in CRISPR/cas9 screens [32], copy numbers of every 50-kb segment of MKL-1 genome were called from the input of ChIP-seq experiments using QDNAseq software. Segmented copy numbers were converted to copy numbers per gene based on gene coordinates. MAGeCK-VISPR pipeline was used to assess data quality, correct copy number variation effect and identify statistically significant targets [75].
MKL-1 cells or a derivative stably expressing MCPyV ST with a C-terminal 3xHA tag were used for ChIP. For MAX, EP400, Ab5 and HA antibodies, ChIP was performed as described [76] with the modification that cells were dual cross-linked with 2 mM disuccinimidyl glutarate (DSG) and 1% formaldehyde [77] and sonicated at 4°C with a Branson Sonifier 250 at 20% duty cycle for 1 minute with 1 minute rest in between for 15 cycles. ChIP- reChIP was performed using the Re-ChIP-IT kit (Active Motif). For ChIP-seq, 10 ng of DNA from ChIP experiments or input DNA were prepared for sequencing with NEBNext ChIP-seq Library Prep Reagent Set for Illumina (New England BioLabs). Amplified libraries were cleaned up using AMPure XP beads (Beckman Coulter) and checked on a Bioanalyzer (Agilent) to confirm a narrow distribution with a peak size around 275 bp. Diluted libraries were used for 50 cycles single-end sequencing on HiSeq 2000 system (Illumina) at the Center for Cancer Computational Biology (CCCB) at Dana-Farber Cancer Institute following the manufacturer’s protocol.
H3K4me3 ChIP-seq was performed as described with minor changes [78]. 0.5xE06 MKL-1 cells were cross-linked with 1.1% formaldehyde and sonicated at 4°C with a Bioruptor (Diagenode). Samples were sonicated on the high setting for 30 seconds with 30 seconds rest in between. Libraries for Illumina sequencing were prepared using the ThruPlex FD DNA-seq kit (Rubicon Genomics). Amplified libraries size-selected using a 2% gel cassette in the Pippin Prep system (Sage Science) to capture fragments between 200–700 basepairs. Libraries were run in Illumina Nextseq.
ChIP-seq mapping was performed using Bowtie (version 0.12.7) against human genome version hg19 allowing only uniquely mapping reads. Peak calling was done using MACS2 (version 2.1.0.20140616) on either single replicate mapped files or replicates merged as mapped bam files using the samtools (version 0.1.18-dev (r982:313)) merge function. Top ranking peaks (5000 most significant, p-val as reported by MACS2) and the Macs2 generated tag pileup output was visualized using the Meta Gene signal distribution function of the CEAS (version 0.9.9.7) analysis package [79].
Significantly enriched transcription factor binding element motifs where found using the Cistrome SeqPos tool using the 1000 most significant (Macs2 p-value) peaks [80].
To calculate genome-wide overlap, all enriched H3K4me3 peaks were extended 5kb in each direction, divided into 250 bins and the read density was calculated in each bin. Density was normalized to the largest value observed in each experiment genome-wide and plotted either as an average of all regions (meta plot) or as a heat map.
MKL-1 cells containing tet-PLKO-shEP400, tet-PLKO-shMYCL and tet-PLKO-shScramble were used to perform RNA-seq. Cells (107) were collected before and 5 days after dox addition. Total RNA was purified using RNeasy Plus Mini Kit (Qiagen). mRNA was isolated with NEBNext Poly(A) mRNA Magnetic Isolation Module (New England BioLabs). Sequencing libraries were prepared with NEBNext mRNA library Prep Master Mix Set for Illumina (New England BioLabs) and passed Qubit, Bioanalyzer and qPCR QC analyses. 50 cycles single-end sequencing was performed on HiSeq 2000 system.
Reads were mapped to the Hg19 genome by TOPHAT. HTSeq was used to create a count file containing gene names [81]. The R package DESeq2 was used to normalize counts and calculate total reads per million (TPM), and determine differential gene expression. QC was performed to generate a MA plot to display differentially expressed genes.
To create the heatmaps, counts were normalized separately for the two experiments (shEP400 and shMYCL) using “voom” from the R/Bioconductor package limma. Data were then corrected for batch effect using ComBat from the R/Bioconductor package sva. In ComBat, the normalized gene expression data were fit to a linear model capturing the effects of basal expression, sample conditions, batch variation, and noise [46]. In our case, the sample conditions corresponded to shScr, shEP400, and shMYCL, and the batch variable corresponded to the experiment in which the data were measured. In the final step of ComBat, the best-fit parameters from the linear model were used to subtract only the effect of the batch variable from the data. Principal Components Analysis (PCA) plots of the data before and after ComBat show that shScr samples from the two different batches cluster together after ComBat (S7 Fig). We then subsetted the batch-adjusted data using the “BETA3” list, defined as previously described in the section on BETA analysis of the ChIP-Seq data, or a “DEG” list of genes that were differentially expressed with Padj < 0.001 in all three comparisons: shP400-2 vs. shScramble, shEP400-3 vs. shScramble, and shMYCL vs. shScramble. Differential expression was determined using DESeq2. Note that the DEG list includes both up- and down-regulated genes, whereas the BETA3 list includes only down-regulated genes.
For each gene list, the batch-adjusted expression values were first standardized across all 15 samples by mean-centering and scaling so that standard deviations are all set to 1. Genes were then clustered using model-based clustering as implemented in the R package mclust. An average profile was created for each gene cluster by taking the mean over the standardized expression values for all the genes in the cluster. Next, the average profiles were merged using complete linkage hierarchical clustering with a Euclidean distance metric. By cutting the tree at a height of 3.5 (for the BETA3 list) or 5 (for the DEG list), we merged the model-based clusters into larger patterns of gene expression. Gene Ontology (GO) term enrichment was run on the final merged clusters using the R/Bioconductor package GOstats with the following parameters: the background set consisted of all the genes from the original RNA-seq alignment, the Benjamini-Hochberg method was applied for multiple testing correction, and the conditional hypergeometric test was used to take into account relationships between GO terms. Heatmaps depict the average standardized expression profiles and were created using the “heatmap.2” function from the R package gplots.
The IMR90 ST and GFP RNA-seq data is available from the Gene Expression Omnibus (GEO) with accession number GSE79968. The IMR90 data were processed using Tophat and Bowtie, and the log-transformed FPKM values were used for all analysis, as described [16]. The genes in the DEG list that also had non-zero expression values across all IMR90 expression profiles were used to create the final heatmap. To visualize both datasets in the same setting, the IMR90 profiles were each subtracted by a corresponding control, which was defined as the average expression level in the IMR90 GFP cell line at the same time point. The MKL-1 shEP400 profiles were subtracted by the average expression level in the shScr samples from the shEP400 batch. Likewise, the shMYCL profiles were subtracted by the average expression level in the shScr samples from the shMYCL batch. Finally, for each gene, all its expression values across both IMR90 and MKL-1 datasets were centered and scaled to the same standard deviation to create the final heatmap. Complete linkage hierarchical clustering with Euclidean distance was used to create the row dendrogram.
MAX, EP400, ST ChIP-seq data were integrated with individual differential expression data from shEP400–1, -2, -3 and shMYCL RNA-seq using Binding and Expression Target Analysis (BETA) software package, which infers activating or repressive function of MAX, EP400, ST and predict the target genes based on rank product of binding potential and differential expression [48]. Shared targets of all three factors were termed shEP400-1 BETA, shEP400-2 BETA, shEP400-3 BETA and shMYCL BETA respectively. Common targets of all four aforementioned datasets were termed BETA4, or BETA3 if shEP400-1 BETA was excluded.
|
10.1371/journal.ppat.1007978 | Structural Insights into Curli CsgA Cross-β Fibril Architecture Inspire Repurposing of Anti-amyloid Compounds as Anti-biofilm Agents | Curli amyloid fibrils secreted by Enterobacteriaceae mediate host cell adhesion and contribute to biofilm formation, thereby promoting bacterial resistance to environmental stressors. Here, we present crystal structures of amyloid-forming segments from the major curli subunit, CsgA, revealing steric zipper fibrils of tightly mated β-sheets, demonstrating a structural link between curli and human pathological amyloids. D-enantiomeric peptides, originally developed to interfere with Alzheimer’s disease-associated amyloid-β, inhibited CsgA fibrillation and reduced biofilm formation in Salmonella typhimurium. Moreover, as previously shown, CsgA fibrils cross-seeded fibrillation of amyloid-β, providing support for the proposed structural resemblance and potential for cross-species amyloid interactions. The presented findings provide structural insights into amyloidogenic regions important for curli formation, suggest a novel strategy for disrupting amyloid-structured biofilms, and hypothesize on the formation of self-propagating prion-like species originating from a microbial source that could influence neurodegenerative diseases.
| Atomic resolution structural insights into the biofilm-associated curli amyloid fibril secreted by Enterobacteriaceae revealed elements of fibrillar architecture conserved between bacterial and human amyloids. This inspired us to repurpose anti-amyloid drugs designed to target human pathological amyloids as a novel class of anti-biofilm agents. Moreover, the results provide a molecular basis for understanding interspecies cross-seeding of amyloids through the generation of prion-like agents by molecular mimicry. This raises concerns regarding human exposure to exogenous sources of amyloids, such as contaminated food and amyloid-secreting microbes. Overall, we provide a novel framework for investigating interspecies amyloid interactions at the molecular level and offer novel insights into mechanisms which may underlie the evolutionary and etiological relationships between the human and microbial amylomes.
| Amyloid formation has traditionally been viewed as a hallmark of protein misfolding diseases, such as amyloidosis, Alzheimer’s and Parkinson’s [1]. In a groundbreaking study, Chapman and coworkers discovered that Escherichia coli secrete extracellular fibers called curli, that biochemically and biophysically resemble human pathological amyloids [2]. Curli proteins are produced via dedicated and tightly regulated cellular processes [3–5], and their fibrils mediate host cell adhesion and invasion, lead to immune system activation, and scaffold bacterial communities known as biofilms [6–9]. Functional amyloids have since been identified in all kingdoms of life, demonstrating a ubiquitous role for amyloids in physiological processes [2, 6, 9–13]. In microbes, the amyloids often serve as virulence determinants involved in aggressive infections and are thus attractive drug targets [9, 14].
Advances in structural biology methods have substantially contributed to the understanding of the remarkable stability of eukaryotic amyloids, ascribed to their shared cross-β structural feature, composed of tightly mated β-sheets with β-strands situated perpendicular to the fibril axis [15–26]. In contrast, the study of bacterial amyloids has proceeded in a long-standing vacuum of high-resolution structural knowledge. We have previously determined the first crystal structure of a bacterial amyloid fibril, i.e., the phenol soluble modulin α3 (PSMα3) peptide secreted by the pathogenic Staphylococcus aureus [27]. The structure revealed a fundamental polymorphism of the amyloid fold, showing that α-helices can replace the β-strands stacked perpendicular to the fibril axis, forming cross-α fibrils [27]. Fibrillation was shown to facilitate PSMα3 cytotoxicity against human cells [27, 28], supporting its classification as a functional bacterial amyloid involved in pathogenicity [27–29]. Furthermore, the secondary structure polymorphism of the PSMα3 cross-α fibrils was particularly striking and indicative of structurally-encoded functional specificity, considering that homologous family members, PSMα1 and PSMα4, form the canonical cross-β amyloid [30]. These PSMs are involved in S. aureus biofilm structuring, and their fibrils, composed of β-sheets mated through tight steric zipper interfaces [30], likely contribute to the stability of the biofilm matrix [31]. In addition, recent structural work in Bacillus subtilis live biofilms showed a swift structural change in TasA, another suspected amyloid protein, from a globular shape toward β-sheet–rich fibrils [32].
The two key players in curli fibril formation are CsgA and CsgB, the major and minor curli subunits, respectively [5]. CsgB nucleates CsgA fibrillation in-vivo via interactions with soluble and unstructured CsgA monomers secreted to the outer bacterial membrane [3–5]. This specific nucleation process likely ensures fibril homogeneity and integrity [33]. CsgA, a 151-residue protein, consists of five imperfect sequence repeats (R1-R5) (S1 Table), defined by regularly spaced serine (Ser), glutamine (Gln) and asparagine (Asn) residues. The first and the last repeats (R1 and R5) form amyloid fibrils in isolation and are critical to CsgA seeding and nucleation by CsgB [34], while the other repeats (R2-R4) contain 'gatekeeper' residues that temper the amyloidogenicity of CsgA [35–37]. CsgA fibrils are resistant to chemical and proteolytic degradation [2, 35, 38, 39], bind the amyloid indicator dyes thioflavin T (ThT) [40] and congo red (CR) and can be visualized, via transmission electron microscopy (TEM), surrounding the bacteria [2, 37, 41]. The X-ray fibril diffraction pattern of CsgA shows reflections at 4.6–4.7 Å and 8–10 Å, indicating a structural spacing typical of cross-β architecture [39, 41] reminiscent of human pathological amyloids. Solid-state NMR data have previously suggested that recombinant CsgA adopts the same structure as native curli isolated from the bacteria, validating much of the biophysical information obtained from recombinant CsgA [42].
Molecular structures of curli-associated proteins are needed to better understand how curli biogenesis is controlled. Structures are available for few curli accessory components, including solution NMR structures of the periplasmic accessory protein CsgE [43] and the adaptor protein CsgF [44], as well as crystal structures of the periplasmic chaperone CsgC [45], which inhibits intracellular CsgA fibrillation [4, 46], and the outer membrane secretion channel CsgG [47, 48]. Despite attempts to obtain structural information on curli fibrils, atomic resolution structures of CsgA and CsgB have remained elusive due to the limitations of some structural methods in contending with the intrinsic properties of full-length amyloid fibrils, which are insoluble and often polymorphic and partially disordered in nature [19, 25, 42, 49]. We thus adopted the reductionist approach of looking for amyloidogenic spine segments which are suggested to nucleate and form the structured backbone of amyloid fibrils [25], and explored segments of CsgA.
We revealed structural similarity between fibrillar spine segments derived from CsgA and those derived from human pathological amyloids, which prompted us to investigate whether fibrillation inhibitors designed against human amyloids could also inhibit curli formation. Accordingly, we found that two D-enantiomeric peptides, originally designed to interfere with the formation of oligomers of Alzheimer’s disease-associated amyloid-β (Aβ) [50–60], inhibited the fibrillation of CsgA spines as well as of full-length CsgA, and reduced biofilm formation in curli-expressing Salmonella typhimurium. Furthermore, in accordance with the structural similarity, fibril seeds of CsgA and of its amyloidogenic segments facilitated fibrillation of Aβ. The results provide structural insights into a biofilm-related amyloid, which could have bearing on amyloid diseases by cross-seeding and the creation of transmissible agents [61], as well as pave the way for the rational development of anti-microbial drugs targeting amyloid-structured biofilms.
To investigate structural features of CsgA, we identified potential amyloid-forming segments that may function as structured spines of CsgA fibrils. These segments were selected by combining computational data predicting regions of amyloidogenic propensity [62–66]. We focused on 45LNIYQY50 and 47IYQYGG52 from the R1 repeat, 137VTQVGF142 from the R5 repeat, and 129TASNSS134 from the R4-R5 loop (sequence positions are indicated as subscript based on the sequence of CsgA with a Uniprot accession number P28307). 129TASNSS134 was selected as a control sequence as it was computationally predicted to be amyloidogenic but is located in a region not implicated in fibrillation [34]. TEM micrographs demonstrated that all four segments formed fibrils (S1A Fig), in accordance with the predictions of their amyloidogenic propensities. However, while 45LNIYQY50 (R1), 47IYQYGG52 (R1), and 137VTQVGF142 (R5) formed unbranched and elongated amyloid-like fibrils, the 129TASNSS134 (R4-R5 loop) segment formed scarce and more amorphous structures (S1A Fig). The 45LNIYQY50 (R1), 47IYQYGG52 (R1), and 137VTQVGF142 (R5) segments bound the amyloid indicator dye ThT and demonstrated dose-dependent amyloid fibrillation curves with short lag times (S2 Fig). 45LNIYQY50 polymerized with the shortest lag time. The 129TASNSS134 segment from the R4-R5 loop did not bind ThT (S2 Fig).
To obtain atomic-resolution structural insight into curli fibrils, we solved the crystal structures of the four segments (Fig 1, S3–S6 Figs, S2 Table). 45LNIYQY50 and 47IYQYGG52, which are overlapping segments from the R1 repeat, adopted very similar structures. Both segments, as well as 137VTQVGF142 (R5), adopted a classical amyloid steric zipper architecture, with two possible dry interfaces between paired β-sheets (Fig 1). The β-strands were oriented parallel to each other along the β-sheets. These three structures are class 1 steric zippers, as defined by Sawaya and Eisenberg according to the organization of the β-strands and β-sheets [24, 25]. In each of these dry interfaces, the chemical properties governing fibril stability, i.e., buried surface area and shape complementarity between sheets, resembled those of eukaryotic steric zipper structures (S3 Table). Correspondingly, the three segments formed fibrils that bound ThT (S1 and S2 Figs). Unlike the three spine segments that adopted tightly packed steric zipper architectures, 129TASNSS134 adopted an anti-parallel β-sheet structure lacking a dry interface between mated sheets. The packing of the β-sheets most closely resembled a class 8 steric zipper [24], with a truncated interface between the two facing β-sheets.
The structural features of full-length CsgA fibrils were analyzed using attenuated total-internal reflection Fourier transform infrared (ATR-FTIR) spectroscopy and showed a main peak at 1617 cm-1, corresponding to a rigid cross-β amyloid architecture [67–69] (S7 Fig). Furthermore, in accordance with a previous report [70], CsgA seeds accelerated the fibrillation of Alzheimer’s disease-associated Aβ1–40, which have been shown to adopt canonical cross-β fibril architectures [20], resulting in massive fibril formation compared to Aβ1–40 alone (S8 Fig). The R1 segment 45LNIYQY50 and the R5 segment 137VTQVGF142 also induced fibrillation of Aβ1–40 (S8 Fig). The R1 segment 47IYQYGG52 and the control segment 129TASNSS134 showed a milder effect on the fibrillation rate of Aβ1–40 (S8 Fig).
We next sought to investigate whether known inhibitors of pathological human amyloid formation can effectively inhibit CsgA fibrillation on the basis of this proposed structural similarity. We tested a group of synthetic D-enantiomeric peptides (referred to here as D-peptides) designed against Aβ [50–60]. Two of the D-peptides tested, ANK6 and DB3DB3 [55], were found to inhibit CsgA fibrillation in a dose-dependent manner. Specifically, freshly purified recombinant CsgA showed a characteristic ThT amyloid-fibrillation kinetics curve with a very short lag time, followed by rapid aggregation, while the addition of the D-peptides resulted in a lower fluorescence signal and a longer lag-time, indicative of delayed fibril formation (Fig 2A and S9 Fig). At 1:5 molar ratios of CsgA to inhibitor, ANK6 induced a longer lag time in CsgA fibrillation than DB3DB3, but both reaction mixes eventually reached similar maximal fluorescence intensities, which were significantly lower than that of CsgA without the D-peptides (Fig 2A). These data may reflect the ability of the D-peptides to induce different morphologies of the CsgA fibrils or interfere with one or more microscopic processes underlying the kinetics of CsgA fibrillation. Dose-dependent inhibition of CsgA fibrillation by D-peptides is shown in S9 Fig, where a significant inhibitory effect was already observed at 1:1 molar ratio of CsgA to the D-peptides. The TEM micrographs correspondingly showed mostly amorphous aggregates or co-precipitates of CsgA in the presence of ANK6 or DB3DB3 compared to fibrils of CsgA alone (Fig 2C). Both ANK6 and DB3DB3 were more potent fibrillation inhibitors than D3 (Fig 2A and S9 Fig), a prototype D-peptide previously shown to reduce amyloid deposits and inflammation and improve cognition in transgenic mouse models for Alzheimer’s disease [51]. In agreement with the hypothesis that the steric zipper segments serve as spines of CsgA fibrils, we found that the D-peptides also inhibited the fibrillation of the R1 45LNIYQY50 and R5 137VTQVGF142 spine segments in a dose-dependent manner (S10 Fig). Moreover, the same inhibitory series for the D-peptides (ANK6>DB3DB3>D3) was observed for the spines. Interestingly, these two segments showed the largest effect on enhancing fibrillation of Aβ1–40 (S8C Fig). In contrast to CsgA and its spine segments, ANK6 and DB3DB3 did not affect the fibrillation of the PSMα3 peptide secreted by the pathogenic S. aureus bacterium, which forms cross-α amyloid-like fibrils [27] (Fig 2B), suggesting that inhibition is dependent on the secondary structure of the fibril. Of note, TEM micrographs showed that these three D-peptides did not form fibrils (S1B Fig).
The effect of ANK6 and DB3DB3 on the secondary-structure transition of CsgA during fibrillation was assessed using time-dependent circular dichroism (CD) spectroscopy (S11 Fig). The CD spectra of freshly purified recombinant CsgA displayed a typical random coil configuration, with a detected minimum around 195 nm. After six hours of incubation, the spectra of CsgA showed a transition to a well-ordered β-sheet structure, with a distinctive maximum near 198 nm and minimum near 218 nm. The timescale of this transition was similar to that observed in a previous work [41]. The CD spectra of CsgA incubated with ANK6 (at 1:5 molar ratio) showed that CsgA retained a random coil configuration throughout the 18 hours of incubation (S11 Fig), indicating inhibition of CsgA fibrillation.
CsgA fibrils are resistant to sodium dodecyl sulfate (SDS) solubilization [41, 71]. We therefore examined the effect of the D-peptide inhibitors on CsgA fibrillation by assessing its migration on a polyacrylamide gel (Fig 3 and S12 Fig). We first assessed the oligomeric state of freshly purified recombinant CsgA using size exclusion chromatography coupled with multi-angle light scattering (SEC-MALS) (S13 Fig). This analysis showed that the majority of freshly purified CsgA was in the monomeric state. A minor population existed as hexameric oligomers. The presence of some oligomers at these early time points was not surprising, considering the rapid fibrillation kinetics observed for CsgA, but it was still unclear whether these hexamers play a role in CsgA fibrillation. Interestingly, the formation of pentamers or hexamers as the smallest populated assembly species, has been described for Aβ as well [72]. As expected, freshly purified soluble CsgA, but not incubated CsgA, migrated in SDS polyacrylamide gel electrophoresis (SDS-PAGE) (Fig 3 and S12 Fig). In the presence of DB3DB3, and to a greater extent ANK6, the soluble CsgA state was stabilized, suggesting a robust effect of the D-peptide inhibitors on CsgA fibrillation in accordance with the ThT, TEM (Fig 2 and S9 Fig) and CD results (S11 Fig).
Treatment with ANK6 and DB3DB3 reduced the total static biofilm biomass formation of S. typhimurium MAE52 (a constitutive curli fimbria and cellulose producing strain) in a dose-dependent manner, as shown by lower crystal violet staining of the biofilm (Fig 4). Both DB3DB3 and ANK6 showed a significant inhibitory effect at a dose of 10 μM, with DB3DB3 producing a more pronounced effect. The D3 peptide only slightly affected biofilm formation at a concentration of 20 μM and was less effective than the other two D-peptides, in accordance with its in-vitro inhibition of CsgA fibrillation. No effect on bacterial growth was observed in the presence of up to 150 μM of any of the three peptides, demonstrating that the observed impact on biofilm mass was unrelated to a bacteriostatic or bactericidal effect (S14 Fig). Confocal microscopy images showing biofilm cells stained with propidium iodide confirmed the significant reduction in the formation of an otherwise abundant surface-attached biofilm of S. typhimurium by the addition of 10 μM DB3DB3 or ANK6 (Fig 5). D3 (10 μM) had no significant effect on the biofilm, in agreement with the crystal violet assay.
In order to confirm that the effect of the D-peptides on biofilm biomass is related to inhibition of amyloid formation, S. typhimurium MAE52 cells were grown on agar supplemented with congo red (CR), a dye which is known to stain amyloids, including curliated whole cells [8]. The resulting biofilm was reddish, indicating adsorption of the dye from the agar (S15A Fig). Pre-addition of ANK6 and DB3DB3 to the bacteria showed a dose-dependent discoloration at the center of the biofilm colony (where the suspension was initially placed on the agar), indicating less CR adsorption (S15A Fig) and hence less fibril formation. Since CR also stains secreted cellulose in biofilm [73], the S. typhimurium MAE150 mutant (MAE52 derived strain that does not express cellulose) was grown on agar under the same conditions, to verify the effect of the D-peptides on curli fibril formation. To comparatively quantify S. typhimurium whole-cell curliation in the presence and absence of the D-peptides, the colonies were removed from the CR-supplemented agar and the concentration of CR that was not adsorbed by the curli-producing bacterial cells but remained in the agar, was measured (Fig 6 and S15B Fig). DB3DB3 was most effective in reducing curli fibril formation, followed by ANK6, while D3 only elicited an effect at a dose of 40 μM. These results are in agreement with the effect of the D-peptides on static biofilm formation (Fig 4). Overall, ANK6 and DB3DB3 showed inhibitory effects on CsgA fibrillation in-vitro, and on curli fibrillation and biofilm formation in the bacteria.
The curli biogenesis machinery is designed to secrete, nucleate and elongate extracellular amyloid fibrils that participate in biofilm formation and stability [3–9]. While the amyloidogenic properties of the major curli subunit, CsgA, have been investigated in detail, high-resolution structural information on CsgA fibrils is lacking. In this study, we elucidated atomistic structural features of CsgA spine segments. Specifically, the R1 45LNIYQY50 and 47IYQYGG52 segments and the R5 137VTQVGF142 segment formed fibrils that bound the amyloid indicator dye ThT and formed canonical amyloid class 1 steric zippers (Fig 1 and S1–S5 Figs). These segments contain Gln49 or Gln139 (marked in bold in 45LNIYQY50, 47IYQYGG52, and 137VTQVGF142), which are critical for fibrillation and cannot be mutated to asparagine without interfering with curli assembly [34]. In contrast, 129TASNSS134 does not contain any residues that have been shown to act as sequence determinants of CsgA fibrillation [34], and moreover, Ser133 is nonessential for fibrillation [34]. Furthermore, in agreement with its atypical crystal structure, 129TASNSS134 formed scarce and more amorphous structures (S1A Fig), and did not bind ThT (S2 Fig). These observations demonstrated that not all segments that are predicted to have high aggregation propensities actually form steric zippers. Here, only those segments which were considered likely to play a role in nucleating fibrillation according to biochemical evidence, formed steric zippers. We postulate that the three steric-zipper forming spine segments from R1 and R5 contribute to scaffolding and stabilizing the robust curli amyloid architecture.
The CsgA steric zipper spine structures closely resemble those of pathological human amyloids (S3 Table). Extensive structural studies on pathological amyloids have suggested that their cross-β signatures result from steric zipper-like structures in the fibril spine formed from one or more short segments of the protein [74–76]. We therefore hypothesize that the CsgA steric zipper spine segments contribute to the structured core of the CsgA fibril and allow for the formation of cross-β fibrils by full-length CsgA [2, 37, 39, 41]. Cross-β ultra-stable structures have been mechanically characterized to be as stiff as silk and as strong as steel [77]. In bacteria, these fibrils likely stabilize and structure biofilms, thereby rendering the bacterial communities more resilient and resistant to antibiotics [2, 6, 9, 10]. Similarly, steric zipper structures of spine segments were suggested to form the cores of the cross-β fibrils of PSMα1 and PSMα4 found in the biofilm of S. aureus [30]. Together, our findings suggest that segments capable of forming steric zippers may be a structural hallmark of biofilm-associated microbial amyloids as well as their disease-associated counterparts, supporting a structural building block that is conserved from bacteria to human.
A previously proposed structural model of full-length CsgA suggested that the fibrils adopt a β-helix or β-solenoid-like fold. This was based on a model suggested for Salmonella CsgA [35] and on information obtained from solid-state NMR and electron microscopy [39, 42] that were more consistent with a β-helix-like structure rather than with in-register parallel β-sheets, although the authors note that the data were insufficient to definitively confirm the adoption of such a structure [39]. This arrangement was also supported by a computational model suggesting that in such a β-helical fibril, each turn corresponds to one repeat sequence of CsgA, forming two β-strands connected by a loop, which ultimately creates a “rectangular” hydrophobic core [78]. The elongation of the structure is achieved by intermolecular stacking along the fibril axis mediated via the R1 and R5 repeats [78]. We herein suggest an alternative model in which fibril formation is nucleated and stabilized via several spine segments in the R1 and R5 repeats. In this model, only specific regions form β-sheets and structure the fibril in contrast to the entire protein being structured as in the β-helix model. The lack of CsgA fibril polymorphism [33] supports both the β-helix model and the spine-based model. In the β-helix model, polymorphism may be averted as the entire protein is involved in structuring the fibril. In the spine-based model, polymorphism may be avoided due to the sequence specificity of nucleation. The gatekeeping residues in the R2-R4 repeats [36] might prevent segments within these repeats from serving as spines, thereby allowing for specific nucleation sites to mediate homogenous fibril assembly.
ATR-FTIR spectroscopy was used to obtain more insight on the structural properties of full-length CsgA fibrils. Previous analyses of amyloids [67, 79–82] revealed a signature FTIR spectral peak between 1611 and 1630 cm−1, while native β-sheet-rich proteins [67] showed a peak at higher wavelengths of 1630–1643 cm−1. Often, a shift from higher to lower wavelengths is observed while monitoring fibril formation [80], which indicates the assembly of longer and planar sheets [67, 83], an increase in the number of β-strands in the β-sheet and/or the formation of stronger hydrogen bonding, typical of extremely stable amyloid fibrils [68, 80]. In contrast to canonical amyloids, mature fibrils of the yeast prion HET-s(218–289) fragment, which was shown by solid-state NMR to form a β-solenoid-like fold [84], showed an FTIR peak at 1630–1631 cm-1 [85]. This peak wavelength is at the threshold between those defining amyloids and native β-sheet proteins [67]. CsgA fibrils showed a peak at 1617 cm-1 (S7 Fig), which aligns with the range of peaks corresponding to rigid cross-β amyloid fibrils [67–69]. A main FTIR peak at 1617 cm-1 was also shown for amyloid fibrils of apomyoglobin [86] and of γD-crystallin [87]. A previous work showed a main FTIR peak for CsgA at 1623 cm-1, also within the lower wavelength range [41]. Overall, the FTIR spectra of CsgA fibrils support the formation of stable cross-β fibrils of in-register, tightly mated sheets. Moreover, we expect that a β-helix fold, which encompasses the entire CsgA sequence, would originate from a folded or partially folded monomers, yet, freshly purified recombinant CsgA is disordered in nature [41] (S11 Fig). Nevertheless, high-resolution structural work on full-length CsgA is needed to accurately describe the architecture of the fibrils, considering the possible existence of novel types of fibril organizations.
The D-enantiomeric peptide D3 was identified from a mirror image phage display selection against monomeric Aβ42 with the intention to stabilize monomeric Aβ42 and to shift equilibria between various Aβ assembly species away from toxic Aβ oligomers. In vivo, treatment with D3 led to reduction of amyloid plaque load and, more importantly, to improvement of cognition in transgenic AD mice. In vitro, D3 converts toxic Aβ oligomers into non-toxic amorphous co-precipitates of D3 and Aβ [53, 54, 88]. DB3DB3 and ANK6 are D3 derivatives designed to stabilize Aβ monomers more efficiently than D3. They were indeed shown to bind Aβ monomers with increased affinity and to eliminate oligomers more efficiently compared to D3 [55, 58]. Inspired by the structural resemblance of spine segments of CsgA to those of human pathological amyloids, we tested the D-enantiomeric peptides [50–55], and found ANK6 and DB3DB3 to inhibit fibrillation of full-length CsgA, slowing its transition from an unstructured soluble configuration to insoluble fibrils (Figs 2 and 3 and S9, S11 & S12 Figs). Since CsgA fibrillation is extremely fast, we assume that the D-peptides bind and stabilize CsgA monomers, and possibly forming amorphous co-aggregates (Fig 2C) that are SDS-soluble (Fig 3). These D-peptides also inhibit the fibrillation of the CsgA spine segments from the R1 and R5 CsgA repeats (S10 Fig), suggesting that these regions can serve as binding sites within the full-length CsgA. As these segments contain positions important for fibrillation and have high aggregation propensity, binding of the D-peptides to these regions could occlude nucleation and delay fibrillation. We thereby suggest that the steric-zipper spine segments serve as critical determinants of fibrillation. The ability of the D-peptides to inhibit both CsgA and Aβ provides further support for the claimed structural similarity between their targets. Moreover, ANK6 and DB3DB3 did not affect the fibrillation of the cytotoxic S. aureus PSMα3, which forms cross-α amyloid-like fibrils [27], demonstrating specificity in structural properties targeted by the inhibitors.
ANK6 and DB3DB3 significantly reduced static biofilm formation of Salmonella typhimurium in a curli-dependent manner (Figs 4–6 and S15 Fig), in agreement with their effects on CsgA fibrillation in-vitro. Similarly, other small molecules and peptidomimetics that have been shown to interfere with the in-vitro assembly of amyloids secreted by E. coli, B. subtilis and other bacteria, prevent biofilm formation and pilus biogenesis [6, 71, 89–95]. Moreover, curlicides (compounds acting against the curli amyloid) attenuated uropathogenic E. coli in a murine model of a urinary tract infection [90]. Furthermore, two such identified compounds also inhibited conversion of the yeast New1 protein to the prion state [95]. Here, we offer a novel strategy for disrupting curli amyloid formation, and prompt further investigation into the promiscuity of a class of anti-amyloid therapeutics as antivirulence agents targeting amyloid-structured biofilms.
Amyloid protein folding and aggregation patterns are highly conserved through evolution and appear in all kingdoms of life [61]. Amyloids have even been suggested to serve as prebiotic replications of information-coding molecules [96]. The best-known pathological manifestation of amyloid self-assembly is in association with neurodegenerative diseases, which involve the formation of transmissible, self-propagating prion-like proteins [97, 98]. Molecular structures shared between amyloids of different species may be involved in the creation of these prion-like agents through molecular mimicry [61], raising concerns regarding the exposure of humans to various food sources and microbes that contain amyloids [99–101]. Indeed, microbial amyloids interact with the amyloids of host systems [6, 102], putatively providing some immune-evasive and survival strategies [11, 103], and have been suggested to contribute to the pathology of aggregation diseases [102, 104–113]. This phenomenon may underlie the ability of seeds of amyloid fibrils from one species to nucleate monomers from another species [70, 114–116]. For instance, infecting the brains of transgenic mice with Salmonella Typhimurium elicited rapid Aβ deposition closely co-localized with the invading bacteria [117]. Exposure to curli-producing E. coli enhanced α-synuclein aggregation involved in Parkinson’s disease in aged rats and Caenorhabditis elegans [116]. Silk from Bombyx mori, Sup35 from Saccharomyces cerevisiae, and curli from E. coli were shown to promote secondary amyloidosis disease in mice [115]. Murine amyloidosis was also accelerated by dietary ingestion of both purified amyloid fibrils and tissue homogenates that contain amyloid fibrils [99]. Several microorganisms, including herpes simplex virus type 1 and various periodontal pathogens, were associated with dementia and brain lesions in Alzheimer’s disease and islet lesions in type 2 diabetes [109, 111, 118–123]. These and other reports on the connection between microbes and amyloid diseases call for comprehensive molecular-level analyses of the interactions between microbial and human amyloids. Here, we showed that CsgA and spine segments from R1 and R5 can cross-seed fibrillation of Alzheimer’s disease-associated Aβ (S8 Fig), in accordance with a previous report [70].
Epitaxial heteronucleation necessitates some degree of structural similarity between participating amyloids, suggesting that full-length CsgA adopts a fibril architecture sufficiently similar to that of Aβ to effectively template fibril elongation [70, 124]. Nucleation is expected to be specific, as for example, islet amyloid polypeptide protein fibrils were unable to seed CsgA [125]. The influence of curli-producing bacteria on human neurodegenerative diseases could originate from the nucleation, acceleration or deterred fibrillation of human amyloids. Alternatively, the bacteria may affect the immune system, leading to inflammation and stress that are correlated with neurodegenerative diseases [126, 127]. Future work is needed to characterize additional microbial amyloids at the molecular and atomic levels, study their inter-species interactions and modulate their activities [128].
Curli peptide segments LNIYQY, IYQYGG, VTQVGF, and TASNSS from CsgA (UniProt accession number P28307), and TAIVVQ from CsgB (UniProt accession number P0ABK7) were used. The peptides were synthesized with unmodified termini for crystallography, or with fully or semi-capped termini (acetylated in the N-terminus and amidated in the C-terminus), as specified, for the other assays. The curli segments, PSMα3 (UniProt accession number P0C805), Aβ1–40 (all custom synthesis) at >98% purity were purchased from GL Biochem (Shanghai) Ltd. The tested D-peptide fibrillation inhibitors consisted of D-enantiomeric amino acids, and were C-terminally amidated. The sequences were as follows: ANK6: RKRIRLVTKKKR-NH2, DB3DB3: RPITRLRTHQNRRPITRLRTHQNR-NH2 and D3: RPRTRLHTHRNR-NH2 (S1 Table). The D-peptides (custom synthesis) at >95% purity were purchased from either GL Biochem (Shanghai) Ltd., peptides&elephants (Potsdam, Germany), or JPT Peptide Technologies (Berlin, Germany). Thioflavin T (ThT), congo red and crystal violet were purchased from Sigma-Aldrich. Dimethyl-sulfoxide (DMSO) was purchased from Merck. Ultra-pure water was purchased from Biological Industries.
D-peptides inhibitors (ANK6, DB3DB3 and D3) were solubilized in ultra-pure water and their concentrations were calculated using a spectrophotometer Nanodrop 2000c instrument (Thermo), at an absorbance of 205 nm, with the specific extinction coefficient calculated for each peptide by the ‘protein parameter calculator’ [129] (http://nickanthis.com/tools/a205.html).
Amyloidogenic propensities of CsgA and CsgB segments were predicted using combined information from several computational methods, including ZipperDB [62], Tango [63, 64], Waltz [65] and Zyggregator [66].
The protocol for CsgA expression and purification was adapted from Wang et al. [36]. A plasmid containing the CsgA sequence cloned into pET11d with a C-terminal His6-tag, was kindly provided by the Chapman lab (University of Michigan, USA) [36]. The plasmid was transformed to E. coli BL-21 cells, which were then grown overnight in 25 ml Luria-Bertani (LB) medium supplemented with 50 μg/ml ampicillin, and further diluted into 700 mL of the same medium, and then incubated at 37°C with 220 rpm shaking, until OD600 was 0.8–0.9. CsgA expression was induced with 0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) for 1 h. Bacterial cell pellets were harvested by centrifugation at 4,500 rpm for 25 min and stored at -20°C. Thawed cell pellets were resuspended in 25 ml lysis buffer (8 M guanidinium HCl, 50 mM potassium phosphate buffer pH 7.3) and incubated at room temperature (RT), with agitation, for 18–24 h. The supernatant was separated by centrifugation at 10,000 g for 25 min and incubated with 1.6 ml HisPur cobalt resin beads (Thermo scientific) equilibrated with lysis buffer, at RT with agitation, for 1 h. The mixture was loaded on a disposable polypropylene column at 4°C and washed with 10 ml of 50 mM potassium phosphate buffer pH 7.3, followed by another column wash with the same buffer supplemented with 12.5 mM imidazole. CsgA was eluted with 125 mM imidazole in 50 mM potassium phosphate buffer pH 7.3. Freshly purified CsgA was filtered using a 30 kDa cut-off column at 4°C (Amicon Ultra-4, Sigma-Aldrich) to remove insoluble protein aggregates and seeds. Imidazole was removed by desalting the protein solution at 4°C using a Zeba spin 7K desalting column (ThermoFisher Scientific), into 50 mM potassium phosphate buffer pH 7.3. CsgA concentration was determined by measuring absorption at 280 nm, calculated with a molar extinction coefficient of 11,460 M-1 cm-1, as determined via the Expasy server (http://web.expasy.org/cgi-bin/protparam/protparam). The identity of CsgA was confirmed by Western blot, using anti-6X His tag antibody.
Multi-angle light scattering size exclusion chromatography (SEC-MALS) analysis of freshly purified CsgA was performed to determine the accurate molecular weight and the oligomerization state of soluble CsgA. CsgA was concentrated to 1.8 mg/ml, using a 3 kDa cut-off spin column at 4ºC (Amicon Ultra-4, Sigma-Aldrich). SEC was performed over a size-exclusion column (Superdex 75 10/300) operated by AKTA avant. MALS was performed with miniDAWN TREOS (WYATT Technology) and its companion Optilab T-rEX (WYATT Technology) dRI detector. Characterization and analysis of the SEC-MALS results were performed using ASTRA software (WYATT Technology).
CsgA fibrils are insoluble in sodium dodecyl sulfate (SDS), even after boiling, and are thus unable to migrate in SDS polyacrylamide gel electrophoresis (SDS-PAGE), in contrast to soluble CsgA [71]. Therefore, this method can be used to monitor CsgA fibrillation and test potential inhibitors. Freshly purified recombinant CsgA was diluted in 50 mM potassium phosphate buffer, pH 7.3, mixed with diluted D-peptides into 1 mL tubes. Final concentrations used were 10 μM CsgA and 50 μM D-peptides (1:5 molar ratio). Samples were then incubated for 24 h at 25ºC, with 300 rpm shaking. Samples (20 μL) were then mixed with 10 μL 3× SDS sample buffer supplemented with dithiothreitol (DTT), and incubated at 95°C for 10 min. Samples were loaded on a 15% SDS-PAGE gel, and a known concentration of BSA solution was added to each lane to quantify the amount of protein in each band. Coomassie Stain (Expedeon—InstantBlue Coomassie Protein Stain) staining was performed to visualize the migration of soluble CsgA in the gel. Imaging was performed using a gel documentation system (Gel Doc–BioRad).
Circular dichroism (CD) was used to monitor the secondary structure transitions of CsgA, in the presence or absence of the D-peptide inhibitor ANK6. Immediately prior to the CD experiment, freshly purified recombinant CsgA (in 50 mM potassium phosphate buffer, pH 7.3) was dialyzed at 4°C, using a Zeba spin 7K desalting column (ThermoFisher Scientific), into 2 mM potassium phosphate buffer, pH 7.3, to reduce background signal during CD measurements. CsgA was directly diluted into the CD cuvette to a final concentration of 9 μM. In parallel, 9 μM CsgA was mixed in a second cuvette with 45 μM ANK6 (1:5 molar ratio) diluted from a 50 mM stock, prepared in ultra-pure water. The signal from blank solutions of either 2 mM potassium phosphate buffer pH 7.3 alone, or of the same buffer containing 45 μM of ANK6, were recorded just before the addition of CsgA into the appropriate cuvette. CD measurements were performed at several time point over 18 h, with cuvettes being incubated at RT between measurements, and mixed thoroughly before each measurement.
Far UV CD spectra were recorded on the Applied Photophysics PiStar CD spectrometer (Surrey, UK) equilibrated with nitrogen gas, using a 0.1 mm path-length demountable quartz cell (Starna Scientific, UK). Changes in ellipticity were followed from 290 nm to 180 nm, with a 1-nm step size and a bandwidth of 1 nm. The measurements shown are an average of three scans for each time point, or two scans for blanks, captured at a scan rate of 1 sec per point.
Freshly purified recombinant CsgA in 50 mM potassium phosphate buffer, pH 7.3, was frozen in liquid nitrogen and lyophilized overnight to complete dryness. The dry CsgA powder was dissolved in deuterium oxide (D2O) to remove water residues interfering with the FTIR signal, and further lyophilized. This procedure was repeated twice. Immediately prior to measurements, the treated sample was dissolved in D2O to 20 mg/ml. Samples (5 μl) were spread on the surface of the ATR module (MIRacle Diamond w/ZnSe lens 3-Reflection HATR Plate; Pike Technologies) and allowed to dry under nitrogen gas. Absorption spectra were recorded using a Tensor 27 FTIR spectrometer (Bruker Optics). Measurements were performed in the wavelength range of 1500–1800 cm-1, in 2 cm-1 steps, and averaged over 100 scans. Background (ATR crystal) and blank (D2O) were measured and subtracted from the final spectra. The amide I’ region of the spectra (1600–1700 cm-1) is presented in the graphs.
Peptides synthesized with free (unmodified) termini were used for crystallization experiments to facilitate crystal contacts. 45LNIYQY50 and 129TASNSS134 formed crystals that diffracted well only when mixed with the 134TAIVVQ139 segment from the R5 repeat of the nucleator protein CsgB prior to crystallization. TAIVVQ was not present in the crystals but aided in crystallization. It is possible that this phenomenon is relevant to the mechanism of heteronucleation of CsgA by CsgB in-vivo [5]. IYQYGG (10 mM) was dissolved in water. VTQVGF (10 mM) was dissolved in 100% DMSO. LNIYQY crystals were grown in a mixture of 10 mM LNIYQY and 10 mM TAIVVQ dissolved in 82% DMSO. TASNSS crystals were grown in a mixture of 30 mM TASNSS and 10 mM TAIVVQ, dissolved in 82% DMSO. Peptide solution drops (100 nL) were dispensed onto crystallization screening plates, using the Mosquito automated liquid dispensing robot (TTP Labtech, UK) located at the Technion Center for Structural Biology (TCSB). Crystallization using the hanging drop method, was performed in 96-well plates, with 100 μL solution in each well. The drop volumes were 150–300 nL. All plates were incubated in a Rock imager 1000 robot (Formulatrix), at 293 K. Micro-crystals grew after few days and were mounted on glass needles glued to brass pins [130]. No cryo protection was used. Crystals were kept at RT prior to data collection. Structures were obtained from drops that were a mixture of the following peptide and reservoir solutions: IYQYGG: 10 mM IYQYGG, 0.1 M sodium acetate, pH 4.6, and 2.0 M sodium formate. LNIYQY: 10 mM LNIYQY, 10 mM TAIVVQ, 0.1 M HEPES, pH 7.5, and 20%v/v Jeffamine M-600. VTQVGF: 10 mM VTQVGF, 3.0 M sodium chloride, 0.1 M BIS-Tris, pH 5.5. TASNSS: 30 mM TASNSS, 10 mM TAIVVQ, 0.2 M lithium sulfate, 0.1 M Tris-HCl, pH 8.5, and 30% (w/v) polyethylene glycol 4000.
X-ray diffraction data were collected at 100 K, using 5˚ oscillation. The X-ray diffraction data for VTQVGF were collected at the micro-focus beamline ID23-EH2 of the European Synchrotron Radiation Facility (ESRF) in Grenoble, France; wavelength of data collection was 0.8729 Å. The X-ray diffraction data for LNIYQY, IYQYGG and TASNSS were collected at the micro-focus beamline P14 operated by EMBL at the PETRAIII storage ring (DESY, Hamburg, Germany); wavelength of data collection was 0.9763 Å. Data indexation, integration and scaling were performed using XDS/XSCALE [131]. Molecular replacement solutions for all segments were obtained using the program Phaser [132] within the CCP4 suite [132–134]. The search models consisted of geometrically idealized β-strands. Crystallographic refinements were performed with the program Refmac5 [134]. Model building was performed with Coot [135] and illustrated with Chimera [136]. There were no residues that fell in the disallowed region of the Ramachandran plot. Crystallographic statistics are listed in S2 Table.
The Lawrence and Colman’s shape complementarity index [137] was used to calculate the shape complementarity between pairs of sheets forming the dry interface (S3 Table). The buried surface area was calculated with Chimera (UCSF), with a default probe radius and vertex density of 1.4 Å and 2.0/Å2, respectively. The number of solvent accessible buried surface areas was calculated as the average area buried of one strand within two β-sheets (total area buried from both sides is double the reported number in S3 Table).
Salmonella typhimurium bacterial strains used: the MAE52 strain displays a constitutive multicellular morphotype mediated by the expression of the agfD operon, leading to the production of an adhesive extracellular matrix consisting of curli fimbriae (previously called thin aggregative fimbriae (agf)) and cellulose [138]. The MAE150 (ΔbcsA) strain sustained a deletion of the bacterial cellulose synthesis (bcs), a gene encoding cellulose synthase [139]. S. typhimurium bacterial MAE52 and MAE150 strains were grown on LB agar plates without NaCl, at 30°C. For liquid media growth, a single bacterial colony was picked from an agar plate and dipped in LB medium (lacking NaCl), followed by incubation at 30°C with vigorous shaking.
The evaluation of static biofilm production was adapted from a previous study [140]. Bacterial suspensions of S. typhimurium MAE52, at OD600 of 0.05–0.13, were incubated in 200 μl LB medium containing D-peptide inhibitors (0, 5, 10, 20, or 40 μM) for 48 h, at 30°C, in a 96-well plate (nunclon, Roskilde, Denmark). The plate was then washed twice with 0.9% saline (w/v).
Curli (along with cellulose) production is associated with CR staining in S. typhimurium [73], thus the lack of CR adsorption by the bacterial cells is evident in the residual agar below the colony [8]. S. typhimurium MAE150 was grown overnight in LB medium and then diluted to obtain a bacterial suspension of OD600 of 0.05–0.13. D-peptide inhibitors (0, 10, 20, or 40 μM) were added to the medium and 3 μl of the suspension was spotted on a LB-agar plate supplemented with 20 μg/ml CR. The bacterial biofilm colonies were allowed to grow for 48 h at 30°C, after which, the colony was immersed in 0.4% paraformaldehyde, followed by the removal of the colony with double distilled water washes. In order to measure the amount of CR that remained on the agar after the removal of the colonies, the agar was dissolved as previously described [140]. In brief, the agar underneath the removed colony was cut out, and incubated at 55°C for 10 min, in 1 ml of binding buffer solution (Hylabs, Rehovot, Israel). The residual CR in the melted agar was measured by absorbance at OD510 (optical density was calibrated after scanning the optimum wave length for CR detection), using nanodrop 2000c (Thermo).
Statistical analyses were carried out for the total biofilm biomass assay (crystal violet staining) and for curli production analysis (CR staining). The data were not normally distributed according to the Shapiro Wilk test. The analyses were performed using the non-parametric Kruskal-Wallis test for comparing three independent groups and the Mann-Whitney non-parametric test for comparing two independent groups (all variables calculated had a small sample size (N ≥10)). All tests applied were two-tailed, and a p-value of 5% or less was considered statistically significant. Analyses were performed using SPSS software, version 24.
|
10.1371/journal.ppat.1004503 | IFI16 Restricts HSV-1 Replication by Accumulating on the HSV-1 Genome, Repressing HSV-1 Gene Expression, and Directly or Indirectly Modulating Histone Modifications | Interferon-γ inducible factor 16 (IFI16) is a multifunctional nuclear protein involved in transcriptional regulation, induction of interferon-β (IFN-β), and activation of the inflammasome response. It interacts with the sugar-phosphate backbone of dsDNA and modulates viral and cellular transcription through largely undetermined mechanisms. IFI16 is a restriction factor for human cytomegalovirus (HCMV) and herpes simplex virus (HSV-1), though the mechanisms of HSV-1 restriction are not yet understood. Here, we show that IFI16 has a profound effect on HSV-1 replication in human foreskin fibroblasts, osteosarcoma cells, and breast epithelial cancer cells. IFI16 knockdown increased HSV-1 yield 6-fold and IFI16 overexpression reduced viral yield by over 5-fold. Importantly, HSV-1 gene expression, including the immediate early proteins, ICP0 and ICP4, the early proteins, ICP8 and TK, and the late proteins gB and Us11, was reduced in the presence of IFI16. Depletion of the inflammasome adaptor protein, ASC, or the IFN-inducing transcription factor, IRF-3, did not affect viral yield. ChIP studies demonstrated the presence of IFI16 bound to HSV-1 promoters in osteosarcoma (U2OS) cells and fibroblasts. Using CRISPR gene editing technology, we generated U2OS cells with permanent deletion of IFI16 protein expression. ChIP analysis of these cells and wild-type (wt) U2OS demonstrated increased association of RNA polymerase II, TATA binding protein (TBP) and Oct1 transcription factors with viral promoters in the absence of IFI16 at different times post infection. Although IFI16 did not alter the total histone occupancy at viral or cellular promoters, its absence promoted markers of active chromatin and decreased those of repressive chromatin with viral and cellular gene promoters. Collectively, these studies for the first time demonstrate that IFI16 prevents association of important transcriptional activators with wt HSV-1 promoters and suggest potential mechanisms of IFI16 restriction of wt HSV-1 replication and a direct or indirect role for IFI16 in histone modification.
| HSV-1, a ubiquitous human pathogen that establishes a life-long infection, has evolved several mechanisms to evade host immune detection and responses. However, it is still subject to regulation by cellular factors. Recently, a host nuclear protein, IFI16, was shown to be involved in the innate defense response to HSV-1 infection. Here, we provide the first evidence that IFI16 inhibits wild-type HSV-1 replication by repressing viral gene expression independent of its roles in the immune response. We show that IFI16 binds the HSV-1 genome at the transcription start sites of several HSV-1 genes. Using a permanently IFI16-negative cell line that we generated, we demonstrate that IFI16 reduces the association of important transcription factors. IFI16 also promotes global histone modifications by increasing the markers of repressive chromatin and decreasing the markers for activating chromatin on viral and cellular genes. These insights into the role of IFI16 in HSV-1 biology suggest that stabilization of IFI16 is an attractive avenue for antiviral drug development.
| Herpes simplex virus type I (HSV-1) is a ubiquitous and highly contagious virus that establishes a life-long infection in host organisms. It typically enters the host through mucosal epithelia and causes a lytic, productive infection in many cell types, including fibroblast, epithelial, and endothelial cells, during which more than 80 gene products are produced from the nuclear viral genome. After primary infection, HSV-1 spreads to neuronal cells in the trigeminal ganglia where it establishes latent infection, during which only the Latency Associated Transcript (LAT) is produced. Periodically, HSV-1 is reactivated from latency and causes recurrent lytic infection at the site of primary infection [1]. HSV-1 typically causes oral lesions but can cause much more severe pathologies, including blindness and fatal encephalitis, due to its infection of corneal cells and the central nervous system [2]–[4].
During lytic infection, HSV-1 genes are transcribed by cellular RNA polymerase II (RNA pol II), assisted by cellular transcription factors, including TATA-binding protein (TBP), in a highly regulated temporal cascade. Transcription from the immediate early (IE) gene promoters of HSV-1 begins as soon as the viral genome enters the nucleus and is initiated by the virion tegument-associated protein, VP16, in conjunction with the cellular transcription factors, Oct1 and HCF. Most IE genes regulate viral and cellular gene expression. The next temporal class of HSV-1 genes, early (E) genes, is expressed around 2–8 hours post-infection (h p.i.) and is largely involved in DNA replication. Expression of these genes is dependent on the viral IE regulatory proteins, ICP0 and ICP4, and cellular RNA pol II, TBP, and other transcription factors. The final broad category of HSV-1 gene expression, late (L) genes, is further categorized into leaky late (DNA replication-independent) and true late (DNA replication-dependent) genes. Late genes encode predominantly structural proteins and their expression is also dependent on viral ICP4, and host RNA pol II, and TBP proteins [1], [5], [6]
Because of its lifelong infection of human hosts, HSV-1 has necessarily evolved a complex set of interactions with host cell factors to modulate host and viral gene expression and to evade immune detection and responses. In addition to the gene regulation outlined above, HSV-1 inhibits cellular gene expression to better co-opt the gene expression machinery for itself. The virion host shut off (vhs) protein inhibits protein synthesis by causing degradation of host and cellular mRNAs [7], [8] and, conversely, enhancing expression of late viral mRNAs [9]. Similarly, ICP34.5, a late protein, dephosphorylates host eIF2α and inhibits protein synthesis [10].
The virion-associated HSV-1 genomic DNA associates with the histones and nucleosome proteins, leading into its chromatinization and the epigenetic control of viral genes [11]–[16]. Histones can be shifted along DNA or modified, leading to the condensation (heterochromatin) or relaxation (euchromatin) of chromatin, resulting in the suppression or activation of gene expression, respectively. Many of these functional histone modifications occur on histone H3, including markers for heterochromatin, including trimethylated H3 lysine 9 (H3K9me3) or lysine 27 (H3K27me3), and markers for euchromatin, such as trimethylated lysine 4 (H3K4me3) or acetylation of lysines K9, K14, or K27 [17]. Several viral gene products, including ICP0, VP16, and LAT have been implicated in viral chromatin remodeling during HSV-1 infection [12]–[14], [16], [18].
Along with these basic gene expression and replication interactions with host factors, HSV-1 regulates the host immune response by several mechanisms. ICP0, a transcriptional regulator and E3 ubiquitin ligase, induces the degradation of antiviral factors [19]–[21] and inhibits type 1 IFN expression [20], [22], [23]. HSV-1 has also evolved mechanisms to inhibit IFN signaling [24]–[27]. Our recent studies have shown that HSV-1 activates and then represses the inflammasome response [19].
IFN-γ inducible protein 16 (IFI16) is a DNA binding protein [28], [29] that was first described as an IFN-inducible protein involved in the differentiation of human myeloid cells [30], [31]. It is a transcriptional modulator [32] through mechanisms that are not yet fully defined. IFI16 interacts with p53 [33] and regulates p53 target gene expression [34]. Interestingly, both loss and gain of IFI16 induces p53 checkpoints; overexpression leads to the apoptotic p53 checkpoint and loss of IFI16 leads to cell cycle arrest [35], [36]. IFI16 is acetylated by the histone acetyltransferase, p300 [37], and may play a role in regulating gene expression by modulating chromatinization [38].
Several immunomodulatory roles have been described for IFI16. It recognizes nuclear herpesviral genomes, including those of HSV-1, Kaposi's sarcoma-associated herpesvirus (KSHV), and Epstein-Barr virus (EBV), and responds to human immunodeficiency virus (HIV) infection, leading to association with apoptosis-associated speck-like protein containing a caspase activation and recruitment (CARD) domain (ASC) through its CARD domain to induce inflammasome activity, resulting in the maturation of caspase-1 and IL-1β [19], [39]–[42]. IFI16 is also necessary for HSV-1-, human cytomegalovirus (HCMV)-, and Vaccinia-virus-induced STING-mediated IFNβ expression [20], [28], [43]. Interestingly, HSV-1 induces the specific degradation of IFI16 at late times post-infection (after 4 h), dependent, at least in part, on ICP0 [19], [20], [44].
Recently, IFI16 has been described as a restriction factor for herpesviral lytic replication [38], [44], [45]. It restricts HCMV replication by displacing transcription factors on E and L but not IE gene promoters [45] and restricts HSV-1 replication [38], [44], [45], particularly replication of ICP0-deficient HSV-1 [38], [44]. IFI16 promotes association of repressive histone modifications with the ICP4, ICP27, and ICP8 HSV-1 promoters during infection with this mutant virus that lacks ICP0 but had no apparent effect on histone modifications associated with viral promoters during infection during infection with a rescue virus [38]. In addition, viral gene expression was repressed, somewhat, during infection with this ICP0-null virus in IFI16-positive cells compared with that in IFI16-depleted cells, but viral gene expression of ICP0-competent rescue HSV-1 was not affected [38]. Though this study suggested that IFI16 recognizes and regulates unchromatinized DNA [38], observations such as the recognition of chromatinized KSHV and EBV DNA by IFI16 during latency, persistence of latent KSHV and EBV gene expression in the presence of IFI16 and the IFI16-inflammasome [39], [40], [42] and the repressive effect of IFI16 on wild-type (wt) HSV-1 replication observed previously [45] or in this study suggest that the mechanisms of IFI16 restriction of HSV-1 are complex.
The study showing repression of HCMV replication by IFI16 demonstrated IFI16-mediated repression of HSV-1 replication but did not pursue mechanisms, thereof [45]. There are multiple differences in cell tropism, replication kinetics, and immune evasion strategies evolved by HSV-1 and HCMV [1], [37]. Notably, HSV-1 causes the specific degradation of IFI16 while HCMV does not [19], [20], [37], [44]. Therefore, we speculated that the mechanisms of HSV-1 restriction by IFI16 may be distinct from those of HCMV restriction.
Here, we show that IFI16 restricts wt HSV-1 replication and gene expression in multiple cell types. It binds to HSV-1 transcription start sites (TSS) of all temporal classes of HSV-1 gene expression, and prevents association of transcription factors, including RNA pol II, TBP, and Oct1 with viral promoters but not cellular promoters. We also show for the first time that IFI16 induces increased euchromatin markers and decreased heterochromatin markers associated with both wt HSV-1 and cellular DNA. These data suggest that IFI16 plays a multi-level role in the modulation of HSV-1 gene repression and that development of drugs to stabilize the function of IFI16 may potentially lead into an effective anti-HSV-1 treatment.
Gariano et al. showed that IFI16 depletion in wt HSV-1-infected cells significantly increased viral yield [45]. However, other studies showed that depletion of IFI16 increased replication of an ICP0-null virus but had no effect on the ICP0-rescue virus or wt HSV-1 strain 17 [38], [44]. To confirm that IFI16 restricts wt HSV-1 (KOS strain) replication, we depleted IFI16 from HFF cells using microporated siIFI16, which resulted in ∼91% knockdown of IFI16 compared with that of siControl (siCtrl) RNA (Figure 1A). To determine the effect of IFI16 depletion, 48 h after microporation we infected the cells with HSV-1 at a multiplicity of infection (moi) of 0.1 or 1.0 plaque forming units (pfu) per cell. At 24 hours post infection (h p.i.), cell culture supernatant was collected and viral yield was determined by plaque assay. Depletion of IFI16 resulted in a significant 5- or 6-fold increase in viral yield (Figure 1B), after infection at an moi of 0.1 or 1.0 pfu/cell, respectively. This inhibition is consistent with that observed previously for wt HSV-1, but further suggests a different mechanism of HSV-1 repression by IFI16 than that involved in repression of HCMV, which was moi-dependent [45]. Interestingly, this reduction in ICP0-positive HSV-1 yield observed here and previously [45] is inconsistent with other reports, showing IFI16-induced inhibition of an ICP0-null virus but not the ICP0-positive rescue virus or wt strain 17 virus [38], [44].
To determine the effect of IFI16 depletion on HSV-1 gene expression, we infected HFF cells microporated with siCtrl or siIFI16 with HSV-1 at an moi of 1 pfu/cell for 2, 4, or 8 h and determined the relative expression of IE (ICP0 and ICP4), E (ICP8 and TK) and L (gB and Us11) gene mRNA by qRT-PCR and normalized data to GAPDH in each sample and to siCtrl at 2 h p.i. (Figure 1C) or to concurrent siCtrl samples (Figure 1D), using the ddCt method [46]. Expression of each gene increased over the course of infection; however, the increase was significantly further augmented in the absence of IFI16 (Figure 1C). Compared to the expression in siCtrl-transfected cells, IFI16 depletion led to an increase of expression of all gene classes: at 8 h p.i., ICP0 expression increased 1.5-fold, ICP4 expression was increased by about 24-fold, ICP8 increased by 8-fold, TK increased by 12-fold, gB increased 150-fold, and Us11 increased by 12-fold (Figure 1D).
The role of innate immune proteins can be cell-type specific [47]. To determine if our observed effects of IFI16 on HSV-1 replication and gene expression, shown in figure 1 for HFF cells, are also cell-type specific and if over-expression of IFI16 restricts HSV-1 replication and gene expression, we transduced human osteosarcoma U2OS cells with IFI16, which was expressed 2.25-fold over GFP-transduced cells (Figure 2A). HSV-1 yield, measured as in figure 1, was significantly inhibited by >5-fold in U2OS cells overexpressing IFI16 (Figure 2B). HSV-1 gene expression was determined and normalized as above. Consistent with the IFI16 knockdown experiments (Figure 1), though expression of each gene increased over time regardless of the presence of overexpressed IFI16 (Figure 2C), IFI16 overexpression significantly inhibited expression of HSV-1 genes from all gene classes over the course of 8 h of infection: compared to GFP-transduced cells, in IFI16-transduced cells at 8 h p.i., ICP0 was inhibited by >50%, ICP4 was inhibited by 76%, ICP8 was inhibited by 50%, TK was inhibited by 87%, gB was reduced by 74%, and Us11 was inhibited by 77% (Figure 2D).
Human breast epithelial cancer MCF-7 cells are naturally IFI16-deficient (Figure 3A) [48]. To determine the effect of IFI16 expression on HSV-1 replication and gene expression in these IFI16-negative cells, we transduced MCF-7 cells with lentiviruses expressing IFI16 or GFP (Figure 3A). HSV-1 replication was inhibited ∼1.6-fold by IFI16 overexpression in MCF-7 cells (Figure 3B). Similar to the results observed in U2OS cells, though expression of each HSV-1 gene increased over the course of infection (Figure 3C), viral gene expression was relatively diminished in IFI16-expressing MCF-7 cells over the course of 8 h of infection. In comparison to expression in GFP-transduced, IFI16-negative MCF-7 cells, in IFI16-transduced cells at 8 h p.i., ICP0, ICP4, ICP8, TK, gB, and Us11 were inhibited by 50%, 70%, 62%, 87%, 46%, and 65%, respectively (Figure 3D).
Together, the results shown in figures 1, 2, and 3 confirmed that IFI16 restricts HSV-1 replication, as was first published by Gariano et al. in HFF cells [45], and suggested that the mechanism of restriction is at the level of HSV-1 gene expression and is not cell-type dependent.
Recently, we showed that HSV-1 infection induces the IFI16 and NLRP3 inflammasome, causing IFI16 and NLRP3 to associate with the inflammasome adaptor protein, ASC, and resulting in the maturation of caspase-1 and IL-1β [19]. IFI16 also has a role in the expression of HSV-1-induced type I IFN [20]. To determine if the inflammasome or IFN response play roles in the innate inhibition of HSV-1 replication and gene expression, we transduced HFF cells with shRNA-expressing lentiviruses targeting IFI16, ASC, or IRF3 resulting in 92%, 81%, and 97% knockdown of IFI16, ASC, and IRF3, respectively (Figure 4A, lanes 2, 3, and 4, respectively, compared with lane 1). None of the knockdown conditions tested affected STING levels (Figure 4A), showing knockdown specificity. Knockdown of ASC did not affect HSV-1 viral yield, as measured by plaque assay (Figure 4B). Consistent with previous reports showing no effect on wt- or ICP0-null HSV-1 replication in the absence of IRF3 [49], knockdown of IRF3 also did not affect HSV-1 yield (Figure 4B). In contrast, consistent with our previous results, knockdown of IFI16 resulted in a significant >5-fold increase of HSV-1 yield (Figure 4B). To further confirm functional knockout of IFI16 and IRF3, we assayed HSV-1-infected HFF cell culture supernatant IFNβ levels at 6 h p.i., infected 48 h post-transduction at an moi of 1 pfu/cell. A robust IFNβ response was detected from HSV-1-infected cells that had been transduced with shCtrl or shASC but not from cells transduced with shIFI16 or shIRF3 (Figure 4C). To confirm functional knockout of ASC, we determined procaspase-1 cleavage in shRNA-transduced HFF cells infected with HSV-1 at an moi of 1 pfu/cell at 6 h p.i., 48 h post transduction. Procaspase-1 was cleaved to active caspase-1 in HSV-1-infected cells transduced with shCtrl, shIRF3, and, to a lesser extent, with shIFI16, likely due to the HSV-1-induced activation of the nucleotide binding and oligomerization (NOD)-like receptor family pyrin domain-containing 3 (NLRP3) inflammasome [19], but not in mock infected cells or in HSV-1-infected cells transduced with shASC (Figure 4D, compare lanes 1 and 3 with lanes 2, 4, and 5). These studies suggested that the role of IFI16 in the inhibition of HSV-1 viral replication is independent of its role in HSV-1-induced inflammasome activation and interferon induction.
Transfection and transduction of cells leads to activation of innate immune responses and adversely effects HSV-1 replication [50]–[52]. In addition, exogenous DNA introduced into cells activates the absent in melanoma 2 (AIM2) and/or NLRP inflammasome responses [19], [39], [40], [42], [53]. To eliminate potential artifacts from these effects and to more thoroughly investigate the effects of IFI16 on HSV-1 replication and gene expression, we used Clustered Regularly Interspaced Short Palindromic Repeat (CRISPR) Cas9-mediated genome editing [54], a highly specific method for targeted eukaryotic genome editing [55]–[57], to create a permanent IFI16-negative cell line. We designed guided RNA to target the Cas9 endonuclease to a region within the coding sequence for the IFI16 Pyrin domain (PYD), the first functional domain of IFI16 (Figure 5A). Wt U2OS cells were cotransfected with 3 plasmids encoding the guided RNA, Cas9, and GFP (a marker for transfection) at a ratio of 4∶1∶1, respectively. After 48 h, cells were sorted for GFP expression and grown clonally before screening for IFI16 expression by dot blot and western blot (Figure 5B). The IFI16-negative U2OS clones 45 and 67 (Figure 5B, lanes 1 and 4, compared with lane 5) were further characterized.
IFI16-negative U2OS cell growth was moderately slower than that of the wt U2OS parental cells; doubling time for wt cells was approximately 31.8 h and those of clones 45 and 67 were approximately 40 h (Figure 5C). In addition, the deletion of IFI16 in U2OS cells caused a significant change in cellular morphology, leading to rounded, elongated cells when compared with the parental wt cells (Figure 5D). We selected clone 67-U2OS for further experiments.
To ensure that the stimulatory effects of IFI16 depletion on HSV-1 replication occurred in these newly generated IFI16-negative U2OS cells, we assayed viral yield at 24 h p.i. from wt U2OS, IFI16-negative clone 67, and clone 67 cells transduced with IFI16 to rescue the effects of IFI16 in IFI16-negative cells. When compared with that in parental wt U2OS cells, HSV-1 yield from clone 67 cells was increased over 6-fold in IFI16-negative clone 67 U2OS cells (Figure 6A). Transduction of IFI16 into clone 67 cells resulted in HSV-1 yield 30% lower than that from wt U2OS cells (Figure 6A), consistent with higher levels of IFI16 expression in these cells (Figure 6B), confirming that IFI16 is the factor responsible for HSV-1 restriction and further suggesting specificity of the Cas9-mediated IFI16 deletion in clone 67 cells.
HSV-1 gene expression at 2, 4, and 8 h p.i. was also determined in these cells. Consistent with our experiments with transient knockdown of IFI16, permanent IFI16 depletion led to increased HSV-1 gene expression from all temporal classes of genes over the course of 8 h of infection, though expression of all the genes increased over the course of infection in both conditions (Figure 6C). Compared with that in IFI16-positive parental U2OS cells, in clone 67 cells at 8 h p.i., ICP0 was increased 3-fold, ICP4 was increased 7-fold, ICP8 was increased 5-fold, TK was increased 3-fold, gB was increased 5-fold, and Us11 was increased 4-fold (Figure 6D).
In addition, we examined IFI16 protein levels in wt U2OS cells and clone 67-U20S cells transduced with empty vector or with IFI16 expression vector in the absence of HSV-1 infection or at 24 h p.i. (Figure 6B). Though previous results showed stability of IFI16 in HSV-1-infected U2OS cells up to 12 h [44], IFI16 protein was decreased 80% in wt-U2OS cells (Figure 6B, lane 4 compared with lane 1) and 87% in IFI16-transduced clone 67 cells (Figure 6B, lane 6 compared with lane 3) after a 24 h infection with HSV-1 at an moi of 1 pfu/cell. To ensure IFI16 stability in HSV-1-infected U2OS cells over the course of our gene expression analysis, we performed western blot analysis of HSV-1-infected U2OS and clone 67 cells. IFI16 was stable in U2OS cells up to 8 h p.i. and was, in fact, induced between 2–4 hours (Figure 6E, lane 3 compared with lanes 1–3), consistent with the induction seen in HSV-1-infected HFF cells [19]. ICP0 protein expression, as measured by western blot (Figure 6E, top) and quantified by densitometry (Figure 6E, bottom), was increased in clone 67 cells when compared with that in wt U2OS cells (Figure 6E, compare lanes 7 and 8 with lanes 3 and 4), consistent with our mRNA results (Figure 6C).
Together, these results demonstrated that our permanent IFI16-negative cell line is an appropriate tool to further study the inhibition of HSV-1 replication and gene expression by IFI16.
Previously, we showed by FISH analysis and co-immunofluorescence that IFI16 colocalized with the HSV-1 genome at 1 h p.i. in HFF cells [19]. To determine HSV-1 genome recognition by IFI16 and colocalization in U2OS cells, we infected U2OS cells with EdU-labeled HSV-1 for 30 or 60 min, performed immunofluorescence for IFI16, and costained for EdU-labeled HSV-1 genome (Figure 7A). We also performed proximity ligation assays (PLA), a fluorescence-based assay that uses DNA-oligonucleotide-linked secondary antibodies to detect closely associated proteins. If two protein epitopes are within 40 nm of each other, the antibody-linked oligonucleotides can ligate with adaptor oligonucleotides to form complete circles that are replicated via rolling-circle DNA replication and detected with fluorescent sequence-specific probes. PLA provides a method for the detection of weak or transient interactions [58]. It can also provide amplified, very distinct localization of a single protein. Here, we did PLA with two antibodies to IFI16, and costained for EdU-labeled HSV-1 (Figure 7B). In uninfected cells, IFI16 was exclusively nuclear. By 30 min p.i., EdU-HSV-1 was detected in approximately 30% of cell nuclei and colocalized with IFI16 in small nuclear puncta (Figure 7A and B, enlarged images of B, yellow arrows). By 60 min, consistent with previous reports from our laboratory and others [19], [37], some IFI16 was detected in the cytoplasm (Figure 7A and B, white arrows), and nuclear IFI16 largely colocalized with the HSV-1 genome (Figure 7A and B, yellow arrows). Pixel intensity plots for the red and green channels were generated for each of the PLA figures (Figure 7B, bottom panels). These show quite clearly the yellow signals that indicate colocalization between IFI16 (green) and HSV-1 genomic DNA (red). These data are the first to distinctly show IFI16 colocalization with the HSV-1 genome at such early times post-infection in U2OS cells and suggest that the association of IFI16 with HSV-1 occurs very shortly after viral DNA enters the nucleus.
HSV-1 gene expression occurs in the nuclei of host cells [1]. To determine if the different morphologies of HFF, U2OS, and clone 67 U2OS cells affected the time post-infection that HSV-1 genomic DNA enters host cell nuclei, we mock infected or infected HFF (Figure S1A), U2OS (Figure S1B), or clone 67-U2OS (Figure S1C) cells with HSV-1 genome-labeled with 5-bromo-2-deoxyuridine (BrdU) at an moi of 1 pfu/cell for 15, 30, or 90 min and immunostained for BrdU. Cytoplasmic BrdU staining showed punctate spots, likely representing nucleocapsid-bound HSV-1 genomes [1]. BrdU-HSV-1 genomes first appeared in the nuclei of approximately 5% of all three cell types at 15 min p.i. (Figure 8A). Once in the nucleus, BrdU staining was still somewhat punctate in many cells but was, overall, much more diffuse than cytoplasmic BrdU staining, likely reflecting the relative expansion of nuclear HSV-1 DNA compared with that condensed in nucleocapsids, consistent with previous studies [59], [60]. The percent of cells with nuclear BrdU increased to approximately 25% at 30 min and to 80% at 90 min p.i. These results demonstrated that the kinetics by which the HSV-1 genome enters host cell nuclei is consistent between HFF, U2OS, and clone 67-U2OS cells in that the genome begins to enter host cell nuclei within 15 min of exposure to the virus, and nuclear HSV-1 DNA increased steadily in a consistently measurable proportion of cells at 30 min p.i. These data suggest that the kinetics of HSV-1 genome entry into cell nuclei is not affected by the morphology of these cell types.
Because the number of nuclei with HSV-1 changed over time, we more quantitatively determined the relative nuclear HSV-1 levels by qPCR of the ICP4 promoter region in nuclear extracts of HSV-1-infected U2OS and clone 67 cells from 30 min to 4 h p.i., and normalized to GAPDH and to DNA levels in U2OS cells at 30 min p.i. using the ddCt method (Figure 8B). We observed that there was variation in both U2OS and clone 67-U2OS cells over time, consistent with previous studies [13], but no difference between cell types at any time (Figure 8B). Based on these results, we used 30 min p.i. as the earliest time point in all our subsequent HSV-1 gene expression studies.
IFI16 binds the sugar-phosphate backbone of DNA in a purportedly sequence-independent manner in vitro [61] and has been shown to interact with oligonucleotides derived from HSV-1 [28], [37] and colocalize with HSV-1 in infected cell nuclei (Figure 7) [19]. There is some preference for cruciform or super helical DNA in vitro [29], but otherwise, little is known about the location of IFI16 binding DNA sites on viral or cellular genomes and no studies have shown, by chromatin immunoprecipitation (ChIP), association of endogenous IFI16 with HSV-1 genomic DNA during infection. Our results in Figures 1, 2, 3, 4, and 6 showed that IFI16 reduced HSV-1 gene expression. We therefore chose to examine its presence at the transcriptional start sites (TSS) of HSV-1 genes.
To determine whether IFI16 binds HSV-1 gene TSS, we performed ChIP analysis of HSV-1 infected cells. IFI16 was immunoprecipitated from the nuclei of HFF and wt-U2OS cells infected with HSV-1 (1 pfu/cell) for 30 min, 1, 2, or 4 h. DNA associated with IFI16 was analyzed by qPCR with primers corresponding to regions flanking the TSS of HSV-1 ICP0, ICP4, ICP8, gB, and Us11 genes, as well as the cellular GAPDH and p21 genes. Because levels of nuclear HSV-1 DNA change up to 2-fold over the course of infection, we normalized our ChIP data concerning HSV-1 DNA to input HSV-1 (ICP4 promoter region) to avoid artifacts from the different sample inputs. Cellular gene DNA was normalized to input GAPDH. For each promoter tested data were further normalized to the 30 min time point using the ddCt method [46].
IFI16 antibodies coprecipitated each of the viral TSS tested in both HFF and wt-U2OS cells (Figure 8C). The association of IFI16 with these regions increased over the course of the infection for nearly all of the genes in both cell types, most dramatically between 30 min and 1 h p.i. Interestingly, in HFF cells, IFI16 accumulated most heavily onto the Us11 promoter whereas it accumulated least on the Us11 promoter in U2OS cells (Figure 8C).
In an earlier study, IFI16 was not found to be bound to cellular promoters in HCMV infected HFF cells [43]. However, we found IFI16 bound to the GAPDH and p21 promoters in HFF and U2OS cells that were mock infected or infected with HSV-1 (Figure 8D and E). This association did not significantly change between mock infected or HSV-1 infected cells or over the course of the 4-hour HSV-1 infection (Figure 8D and E).
To ensure specificity of our ChIP assays, we performed additional ChIP analysis of HSV-1-infected U2OS (U) or clone 67 U2OS (67) cells with IFI16 antibody or control IgG. Promoter sequences were amplified and analyzed by agarose gel electrophoresis. IFI16 antibodies, but not control IgG, precipitated viral and cellular promoters from wt U2OS cells but not IFI16-negative clone 67 U2OS cells (Figure 8F, compare lane 1 with lanes 2–4). Importantly, there was no amplification of DNA from these regions in no-antibody controls at 4 h p.i. (Figure 8G).
These data are the first to show endogenous IFI16 binding to HSV-1 DNA during infection and suggested that IFI16 targets all temporal classes of HSV-1 genes and does not specifically affect the immediate early and early regulatory genes.
HSV-1 genes are transcribed by cellular RNA pol II [1]. Our data show that IFI16 binds HSV-1 DNA (Figure 8). To determine whether the presence of IFI16 affects the accumulation of RNA pol II on the various HSV-1 TSS, we infected wt and clone 67 U2OS cells with HSV-1 at an moi of 1 pfu/cell for 30 min, 1, 2, or 4 h and performed ChIP analysis with an antibody to RNA pol II and primers to HSV-1 TSS, as described above, normalizing once again to input ICP4 levels because of the differences observed in nuclear HSV-1 DNA levels (Figure 8B). RNA pol II accumulation at HSV-1 TSS was increased significantly in IFI16-negative cells for all viral genes tested compared with that in parental U2OS cells (Figure 9A). Interestingly, RNA pol II accumulation at the cellular GAPDH gene promoter was not affected by the presence or absence of IFI16 (Figure 9A). To confirm that RNA pol II association was unchanged on cellular promoters in the presence or absence of IFI16, we repeated the RNA pol II ChIP, in U2OS and clone 67 cells that were infected with HSV-1 for 30 min or 2 h, with primers for the GAPDH and p21 promoter regions. RNA pol II association with each of these regions was the same, regardless of infection time or the presence of IFI16 (Figure 9B). HSV-1 and GAPDH amplicons were detected in samples from U2OS and clone 67 cells immunoprecipitated with RNA pol II but not with beads alone at 4 h p.i. (Figure 9C).
These results suggested that the presence of IFI16 restricts the association of RNA pol II with HSV-1 TSS of all temporal classes, consistent with IFI16-mediated HSV-1 gene repression (Figures 1–3).
HSV-1 gene expression is dependent on many viral and cellular transcription factors [1], [59], [62]–[64]. Some of these factors, including the TATA binding protein (TBP) and Oct1 are necessary to recruit RNA pol II to all HSV-1 promoters, in the case of TBP, and only to immediate early promoters, in the case of Oct1 [65], [66]. Because IFI16 has been shown to prevent the association of some transcription factors with the HCMV promoter [45], we hypothesized that IFI16 prevents association of TBP and Oct1 with HSV-1 promoters, thereby inhibiting the recruitment of RNA pol II for gene expression that we observed (Figure 9).
To determine whether IFI16 reduces the association of TBP with HSV-1 promoters, we infected wt and clone 67 U2OS cells with HSV-1 at an moi of 1 pfu/cell for 30 min, 1, 2, or 4 h, performed ChIP analysis with antibodies to TBP and primers to HSV-1 TSS (Figure 10), and normalized as described above. The absence of IFI16 expression led to significantly increased (3- to 10-fold) association of TBP with HSV-1 promoters but no change in association of TBP with the GAPDH promoter (Figure 10A). To confirm that TBP association with cellular promoters was unchanged by the presence of IFI16, we repeated the TBP ChIP in both cell types infected with HSV-1 for 30 min or 2 h, and amplified the GAPDH and p21 TSS. TBP association with each of these regions was unchanged, regardless of infection time or the presence of IFI16 (Figure 10B). DNA was amplified only from samples from both cell types immunoprecipitated with antibodies to TBP but not from beads alone controls (Figure 10C).
To determine if IFI16 reduces association of Oct1 with HSV-1 promoters, we infected wt and clone 67 U2OS cells with HSV-1 at an moi of 1 pfu/cell for 30 min, 1, 2, or 4 h and performed ChIP analysis with antibodies to Oct1 and primers to HSV-1 TSS as described above. The absence of IFI16 expression led to significantly increased (3- to 17-fold) association of Oct1 with HSV-1 immediate early ICP0 and ICP4 promoters, a very modest increase in association of Oct1 with ICP8 and Us11 promoters, and no significant change in the association of Oct1 with gB or GAPDH promoters (Figure 11A). To confirm that Oct1 association with cellular promoters was not altered by the presence or absence of IFI16, we repeated the Oct1 ChIP in both cell types infected with HSV-1 for 30 min or 2 h, with primers for GAPDH and p21 promoters. Oct1 association with each of these regions was the same, regardless of infection time or the presence of IFI16 (Figure 11B). Amplified DNA was observed only from samples immunoprecipitated with antibodies to TBP but not in the control beads alone samples at 4 h p.i. (Figure 11C).
Together, these results demonstrated that IFI16 does, indeed, prevent and/or reduce association of transcription factors with HSV-1 promoters but not cellular promoters and provide a potential explanation for the decrease in RNA pol II association with HSV-1 promoters in IFI16 positive cells (Figure 9).
Orzalli et al. [38] suggested that IFI16 may induce changes in histone modifications associated with HSV-1 promoters in HFF cells at 6 h p.i. during infection with an ICP0-null virus, but saw no effect on histone modification with their rescue virus. Because we observed IFI16-induced differences in HSV-1 gene expression and replication during infection with wt virus, we hypothesized that there were IFI16-induced differences in the histone modifications associated with wt HSV-1 DNA.
To determine if IFI16 affected the histone modifications in chromatin associated with HSV-1 TSS, we infected wt and IFI16-negative clone 67 U2OS cells with HSV-1 at an moi of 1 pfu/cell for 30 min, 1, 2, or 4 h and performed ChIP analysis with antibodies to total histone H3, H3K4me3 (a marker for active or euchromatin), or H3K9me3 (a marker for repressive or heterochromatin), primers for HSV-1 TSS, and normalized to input ICP4 TSS levels (for HSV-1 TSS) or GAPDH (for cellular TSS), as described above.
As determined previously [38], the presence of IFI16 had no effect on the levels of total histone H3 associated with HSV-1 promoters or the cellular GAPDH or p21 promoters over the course of infection (Figure 12A).
In the presence of IFI16, H3K4me3 (active chromatin) was detected associated with IE, E, and L HSV-1 promoters at 30 min p.i. This association decreased 3- to 5-fold by 4 h p.i. (Figure 12B). At 30 min p.i., there was no effect of IFI16 depletion on H3K4me3 association with viral promoters (Figure 12B). In contrast, IFI16 depletion led to a marked increase in the association of the active H3K4me3 with HSV-1 promoters at later times p.i. (Figure 12B). Interestingly however, the absence of IFI16 also led to an increase in the levels of active H3K4me3 on the GAPDH and p21 promoters (Figure 12B).
IFI16 also had little to no effect on the association of H3K9me3 (repressive chromatin) with IE, E, and L HSV-1 promoters at the very early times post infection but at later times, the presence of IFI16 led to increased association of the repressive chromatin marker with HSV-1 promoters (Figure 12C). In the absence of IFI16, HSV-1 promoter occupancy by H3K9me9 did not change significantly over the course of the 4 hour infection. Again, interestingly, the association of H3K9Me3 with the cellular GAPDH and p21 promoters was increased in the presence of IFI16 at all times p.i. (Figure 12C).
These data corroborate the previous findings that IFI16 does not affect the association of total histones with HSV-1 DNA [38] but demonstrated that there is a global decrease in repressive heterochromatin markers and an increase in active euchromatin markers associated with wt HSV-1 and cellular promoters in the absence of IFI16, which was not observed in previous studies.
IFI16 has several antiviral properties, including roles in IFNβ induction [20], [28], [67], activation of the inflammasome response to nuclear viral DNA [19], [39]–[41], [68], [69], and modulation of HCMV replication and gene expression [45], [70]. In this study, we show that HSV-1 replication and gene expression are inhibited by IFI16. Our data lead us to propose a model whereby, in the presence of IFI16, histone H3 associated with viral and cellular genes is modified by repressive trimethylation at lysine 9 and important transcription factors, including TBP and Oct1 (in the case of immediate early genes), bind to HSV-1 transcription start sites, along with IFI16, recruiting RNA pol II to induce transcription at low levels (Figure 13A). However, in the absence of IFI16, there is a global shift from repressive chromatin markers to the active trimethylation at lysine 4 and binding of the transcription factors and, therefore, RNA pol II, to viral promoters is increased (Figure 13B), thereby significantly increasing gene expression.
Interestingly, the repressive effect of IFI16 on wt HSV-1 observed here and previously [45] was not observed for an ICP0 rescue virus [38] or wt HSV-1 strain 17 [44]. To a point, similar methodologies were used for all: plaque assay to determine HSV-1 replication in fibroblasts depleted for IFI16 by sh- or siRNA. Here, we show derepression of HSV-1 replication and gene expression HFF and U2OS cells using siIFI16, shIFI16, and CRISPR technology to delete IFI16. We also show repression of HSV-1 replication and gene expression in U2OS and MCF7 cells after overexpression of IFI16. Currently, it is difficult to resolve these differences. We suspect that nuanced and dynamic interplay between IFI16 and ICP0 is involved in the repression of HSV-1. Further studies are necessary to fully understand these nuances.
Though IFI16 is known to be involved in transcriptional regulation [32]–[34], [48], [70]–[76], the mechanisms of this regulation have been largely undetermined. IFI16 may modulate transcription through association with transcription factors and/or blocking their association with promoters [33], [34], [45], [48], [70], [72]–[74] or it may promote the formation of heterochromatin on promoters, resulting in their repression [38], [76].
Here, we show evidence for both effects during HSV-1 infection; IFI16 reduces association of RNA pol II, TBP, and Oct1 specifically with HSV-1 promoters and not with the promoters of the cellular genes, GAPDH and p21 (Figures 9–11). This suggests that IFI16 can specifically prevent association of transcription factors with HSV-1 DNA while allowing them to associate with cellular genomic DNA.
We also show evidence that IFI16 directly or indirectly impacts histone modifications at the TSS of both HSV-1 and cellular genes (Figure 12). Interestingly, there was an increase in total histone association with only the TSS of ICP4 (Figure 12A). This could be due to the necessity for high levels of ICP4 early during infection to stimulate E and L gene expression and the relatively little ICP4 needed at later times. Additionally, at 30 minutes p.i., there was little difference in histone modifications between cell types, but the increase in H3K4me3 and decrease in H3K9me3 at HSV-1 TSS in IFI16-negative cells compared with that in IFI16-positive cells suggests that these modifications are being added to virally-associated histones largely between 30 min and 1 h p.i. rather than histones being modified prior to association with viral DNA. This alteration of the histone modifications at extremely early times post-infection could be due to ICP0, VP16, IFI16, or other viral or cellular factors, or a combination.
Previous studies have suggested that IFI16 promotes the formation of heterochromatin and reduces the formation of euchromatin on the promoters of an HSV-1 mutant genome, which does not express the viral E3 ubiquitin ligase, ICP0 [38]. However, that study did not show a change in chromatin markers associated with their wt-like rescue virus and did not examine the association of these markers with cellular genes [38]. The difference between our findings and the earlier study is likely due to their normalization of data to GAPDH promoters associated with specific chromatin markers and our normalization to input viral genomes. Normalizing our data to co-precipitated GAPDH sequences would also yield results suggesting no difference between chromatin markers on wt HSV-1 genomes because we found that association of these histone modifications with cellular DNA was also altered in the presence of IFI16 (Figure 12). We postulate that IFI16 may have a global effect on the formation of hetero- and euchromatin and not an effect strictly on viral genome-associated chromatin. This is consistent with previous studies linking IFI16 with histone modifications and histone modification machinery [37], [76] and suggests that IFI16 may play a direct or indirect role in modulating the activity of these enzymes or their association with HSV-1 genes. The role of ICP0 in chromatin remodeling [13] and possible nuclear interaction between IFI16 and ICP0 [19] further suggest a role for IFI16 in histone modification. Further studies are required to fully understand the involvement of IFI16 in chromatin remodeling complexes in the basal state or during infection with HSV-1 or other pathogens.
We found IFI16 associated with GAPDH promoters in HFF cells and U2OS cells (Figure 8). However, previously Li et al. showed that, during HCMV infection, IFI16 associated with viral genes but did not associate with cellular genes [43]. This discrepancy could be due to differences in ChIP protocol; we immunoprecipitated IFI16 from cell nuclear extracts, which reduces background. Li et al., immunoprecipitated from total cell lysate [43], which would lack such nuclear enrichment and, because IFI16 is exported from the nuclei to the cytoplasm of herpesvirus-infected cells [19], [37], [39], [40], [42], [77], immunoprecipitating from total cell lysate could capture ligands associated with cytoplasmic IFI16, possibly reducing the threshold of detection for nuclear associations. However, given that Li et al., show exclusively nuclear IFI16 at the time of their IFI16 ChIP experiments, differences in gene detection by ChIP may be based on the proximity of the primer-amplified DNA region with the binding site of IFI16.
It is also remarkable that IFI16 was found associated with promoters from each temporal class of HSV-1 expression (Figure 8). This suggests that the inhibition of HSV-1 by IFI16-mediated transcriptional repression is not the result of IFI16 associating with only IE promoters causing the inhibition of expression of downstream classes of genes by inhibiting their viral activators. This provides IFI16-expressing cells with redundant means of HSV-1 gene expression inhibition. However, the binding of IFI16 to each temporal class of HSV-1 gene does not exclude the possibility that it prevents other stages of HSV-1 replication. IFI16 is involved in STING-mediated type 1 IFN induction [20], [28], [67], STING is associated with the ER and trans-Golgi network [78], and HSV-1 nucleocapsids bud through the Golgi during egress [1]. IFI16 may inhibit budding of progeny HSV-1 virions. Further studies are required to investigate this possibility and also to determine the extent and localization of IFI16 binding sites to HSV-1 DNA.
We show that IFI16 prevents the association of important transcription factors with HSV-1 gene promoters (Figures 9–11). However, HSV-1 gene expression is not completely abrogated in the presence of IFI16, as shown here, and by the permissiveness of most cell types to infection with HSV-1 [1]. Therefore, the effect of IFI16 on HSV-1 gene repression cannot be absolute. This could be due to the stoichiometry of the interaction between IFI16 and HSV-1 promoters: there may not be sufficient IFI16 to promote IFN and inflammasome induction and concurrently be present on all HSV-1 promoters. The induction of IFI16 expression at early times p.i. in HFF cells [19] and U2OS cells (Figure 6) could be a cellular response to simultaneously promote all of these activities.
In addition, the HSV-1-induced degradation of IFI16 occurs with significantly slower kinetics than that of another nuclear foreign DNA sensor and viral restriction factor, PML [44]. This suggests that IFI16 has a nuanced role in the regulation of HSV-1 genes and may be useful for HSV-1 at early times p.i. Perhaps its viral gene repression activity facilitates HSV-1 replication, in vivo, by preventing uncontrolled viral replication and undue stress on host cells and the infection microenvironment, which may expedite immune cell recruitment. It is also possible that IFI16 may act to promote the expression of some viral genes, as it is a positive regulator of some cellular genes [48].
IFI16 is involved in cell-cycle regulation and the DNA damage response [48], [79], [80]. It could interfere with the HSV-1 DNA replication process, making it critical for the virus to decrease IFI16 protein levels prior to the bulk of DNA replication. Further studies are needed to fully appreciate the consequences of IFI16 degradation during the HSV-1 replication cycle.
It is clear that IFI16 can discriminate between host and foreign DNA [20], [28], . Some studies suggest that this is due to differences in chromatinization state [38]. Because of the global differences in chromatin modifications we observed and the activation of IFI16 during EBV and KSHV latency, during which viral DNA is chromatinized [40], [42], we believe that there must be other factors involved. Perhaps the topography of viral DNA is different from that of the host genome, leading to increased affinity, or IFI16 may bind DNA in concert with other DNA binding proteins, thereby increasing the affinity or avidity of an interaction between IFI16 and viral DNA when compared with those of IFI16 and cellular DNA.
ICP0 is a multifunctional IE alphaherpesvirus protein that is important in reactivation of HSV-1 from latency [1]. Though mechanisms of HSV-1 latency establishment are not yet well understood, ICP0 has long been considered a confounding factor. Recently a neuron-specific microRNA, miR-138, was shown to be important in the establishment of HSV-1 latency by targeting ICP0 mRNA for degradation [81]. ICP0 is necessary for the degradation of IFI16 in non-ICP0-complementing cells [19], [20], [44] and here we show that IFI16 inhibits HSV-1 gene expression. If, in neurons, miR-138 effectively disposes of ICP0, IFI16 would remain stable and able to carry out its role in gene repression. Our studies suggest that a balance between ICP0 and IFI16 may play a crucial role in determining the outcome of infection. Additionally, because the restriction of HSV-1 by IFI16 is independent of the roles IFI16 plays in the inflammasome and interferon responses (Figure 4), this offers a potential mechanism for IFI16 control of HSV-1 lytic gene expression during latency maintenance that does not lead to inflammation. IFI16 is, indeed, expressed in human neurons (Johnson, unpublished data) and further studies could elucidate its potential role in HSV-1 latency establishment.
Like that of other innate immune factors [47], the role of IFI16 in different innate responses may be cell type-specific and modulated by HSV-1 proteins such as ICP0 and the Us3 protein kinase [44]. We showed here that IFI16 inhibits HSV-1 replication in multiple cell types (Figures 1–2), suggesting that this effect is general and not likely to be cell-type dependent.
Recent studies that point to the relative stability of IFI16 in HSV-1-infected U2OS cells [44] and more established studies suggesting that replication-defective HSV-1 ICP0 mutant viruses can successfully replicate in U2OS cells [82] caused us some concern. However, a) similar effects of IFI16 on HSV-1 replication and gene expression in HFF and U2OS cells (Figures 1 and 2), b) IFI16 degradation in HSV-1-infected U2OS cells (Figure 6) albeit at a much later time point than in HFF cells [19], [20] and, indeed, a later time point than was tested previously in U2OS cells [44], and c) enrichment of IFI16 on HSV-1 promoters in HFF cells and U2OS cells (Figure 8), demonstrate that the mechanisms of IFI16-mediated inhibition of HSV-1 gene expression are shared between cell types.
The decreased association of IFI16 with the late HSV-1 Us11 promoter in U2OS cells compared with that in HFF cells may provide insights into the increased viral yield in U2OS cells compared with that in HFF cells (Figures 1 and 2). Us11 has many diverse pro-HSV-1 functions, including promotion of protein synthesis, intracellular trafficking, inhibition of RIG-like receptor signaling, and inhibition of autophagy [83]–[88]. Further studies are essential to clarify these effects.
Importantly for the establishment of our CRISPR-mediated IFI16-negative cell line, a clonal population of U2OS cells can be grown from a single cell (Figure 5), which ensures identical genotypes for all cells tested. We were not able to grow such clonal populations of HFF cells. Given the consistency between U2OS and HFF cells described above, the U2OS and IFI16-negative clone 67 provide a valuable tool for further studies of the antiviral roles of IFI16. The slower cell growth rate and change in morphology of IFI16-negative U2OS cells could be due to the interactions of IFI16 with p53, which affect cell cycle dynamics [35]. p53 has important roles in cell morphology [89]–[92], which may also be modulated by its association with IFI16.
Several antiviral activities have now been described for IFI16 to counter infection by a broad range of viruses, including α- and γ-herpesviruses, HIV, and Vaccinia virus [19], [20], [28], [38]–[42], [45], [93], [94]. Developing drugs that stabilize the foreign gene repressive functions of IFI16 (this study and [19], [38], [45]) and transiently support the IFN-inducing and inflammasome-activating functions of IFI16 without allowing for constitutive innate immune signaling could lead to effective, broad range antiviral therapeutics. To safely design such a drug, future studies need to be done to further characterize the nuances of IFI16-DNA binding, IFI16-mediated IFN induction, and HSV-1-induced IFI16 degradation.
Human osteosarcoma cells (U2OS), human foreskin fibroblasts (HFF cells), MCF7 (breast epithelial cancer) cells, human embryonic kidney (HEK293T), and African green monkey (Vero) cells were from American Type Culture Collection (ATCC, Manassas, VA). These cells were propagated in Dulbecco's modified Eagle Medium (DMEM) supplemented with Glutamax (Gibco, Grand Island, NY), 10% fetal bovine serum (FBS; Atlanta Biologicals, Lawrenceville, GA), and 1% penicillin/streptomycin (Gibco, Grand Island, NY). They were routinely tested for mycoplasma using the Mycoalert kit (Lonza, NJ), according to the manufacturer's instructions, and were found to be negative.
KOS strain HSV-1 was propagated and titered by plaque assay on Vero cells, as described [95]. Briefly, Vero cells were infected with HSV-1 at an moi of 0.001 pfu/cell until cells began to round up and could be shaken from the flask (3–5 days). At 4–6 h prior to harvest, 50 µg/mL heparin was added to cell culture supernatant. Cells were removed from cell culture supernatant by centrifugation at 1,000 rpm for 10 min at 4°C. Virus was isolated by further centrifugation of the supernatant at 20,000×g for 2 h at 4°C. Pellet was resuspended in PBS-AB/15% glycerol and stored at −80°C.
To generate 5-ethynyl-2′doxyuridine- (EdU) and 5-bromo-2-deoxyuridine (BrdU)-labeled infectious HSV-1 virus, a modified protocol described to produce BrdU-labeled HCMV [96] and KSHV [93] was used. Briefly, while propagating KOS strain HSV-1, EdU Labeling Reagent (Life Technologies, Camarillo, CA) was added to flasks at 50 µM and BrdU Labeling Reagent (Life Technologies) was diluted 1∶100 and added to the culture medium at 8, 24, and 48 h p.i. Flasks with media containing BrdU were kept in darkness or dim light during incubation to avoid photolysis of BrdU residues. Viral purification was carried out as described [95].
Cells were incubated with HSV-1 for 2 h or until the time indicated in serum free DMEM, washed with PBS, and incubated in DMEM supplemented with 2% FBS until the times indicated.
Viral yield at 24 h p.i. was determined by titering on Vero cells. Briefly, infected cell supernatants were cleared of cell debris by centrifugation. Vero cells were infected in duplicate or triplicate with serial dilutions of supernatants for 2 h in serum free DMEM, washed with PBS, overlaid with 1× DMEM/1% agarose, and incubated at 37°C until plaque formation was observed (48–72 h). Cells were fixed by overlaying the agarose layer with 4% paraformaldehyde in PBS for 20 min and then stained with 0.2% crystal violet in 50% methanol for 20 min. Dye was washed off and plaques counted. Figures shown are representatives of three or more experiments, each.
A plasmid was constructed by cloning IFI16 guided RNA (target sequence: GAAAAGTTCCGAGGTGATGCTGG synthesized within the guided RNA scaffold [54]) into pGEMT, using NheI sites. Using Lipofectamine LTX and Plus reagent (Life Technologies), U2OS cells were transfected with 3 plasmids encoding: guided RNA, Cas9 (Addgene plasmid 41815, a generous gift from Dr. George Church [54]), and GFP at a ratio of 4∶1∶1, respectively. At 48 h post-transfection, GFP-positive cells were sorted individually into 96-well plates containing complete growth media. Lack of IFI16 expression in each clone was screened by dot blot and confirmed by western blot.
The following antibodies were used in Western blot and immunofluorescence analysis: mouse anti-IFI16 and anti-BrdU (Santa Cruz Biotechnology, Inc, Santa Cruz, CA), TBP (Abcam, Cambridge, MA), ASC (MBL Laboratories, Woods Hole, MA), and rabbit-anti IFI16 and anti-actin (Sigma, St. Louis, MO).
Antibodies used for chromatin immunoprecipitation assays (ChIP) were: IFI16, RNA polymerase II, TBP, Oct1, total histone H3, H3K9me3, H3K4me3, and HSV-1 VP16 (Abcam, Cambridge, MA), normal mouse IgG (Santa Cruz Biotechnology, Inc).
To create an IFI16-expression lentiviral vector, the IFI16 coding region (NM_001206567.1, nucleotides 291–2482) was cloned into pcpsppw [97] using primers (Table 1) and the BamHI and ApaI sites. IFI16 lentiviral vectors were produced using a four-plasmid transfection system, as previously described [97]. Briefly, HEK293T cells were transfected with IFI16 expressing vector and packaging plasmids and the media was changed 16 h after transfection. Supernatants containing the lentiviral vectors were collected 24 h later, passed through a 0.44 µm filter and used to transduce cells in the presence of polybrene (5 µg/ml, Pierce, Rockford, IL). ShIFI16, shASC, and shCtrl were obtained (Santa Cruz Biotechnologies) and HFF cells were transduced according to the manufacturer's instructions. No selection was done. Western blot analysis was performed to confirm the level of knockdown at 48 h post-transduction.
Scrambled siRNA and siIFI16 were transfected into HFF cells using the Neon transfection system (Life Technologies), according to the manufacturer's instructions and as described [93]. Briefly, subconfluent cells were harvested, washed once with PBS, and resuspended in resuspension buffer R (provided) at a density of 1×107 cells/ml. 10 µL of this cell suspension was mixed with 100 pmol siRNA and microporated at room temperature using a single pulse of 1350 V for 30 ms. After microporation, cells were distributed into complete medium and placed at 37°C in a humidified 5% CO2 atmosphere. 48 h post-transfection, cells were analyzed for knockdown efficiency by western blot. siRNA oligonucleotides (siGenome SMARTpool) for IFI16 (catalog number MHSXX0020) were purchased from Thermo Fisher Scientific (Waltham, MA).
Cells were lysed in radioimmunoprecipitation assay (RIPA) buffer (15 mM NaCl, 1 mM MgCl2, 1 mM MnCl2, 2 mM phenylmethylsulfonyl fluoride) supplemented with protease inhibitor cocktail (Sigma), sonicated, and clarified by centrifugation at 16,000×g for 10 min. Equal amounts of protein were separated by SDS-PAGE and electrophoretically transferred to nitrocellulose membranes. Membranes were incubated with primary antibodies and secondary antibodies conjugated to horseradish peroxidase (KPL, Gaithersburg, MD). Immunoreactive bands were visualized using ECL western blotting substrate (Pierce). Blots were scanned using FluorChemFC2 software with the AlphaImager system (Alpha Innotech Corporation, San Leonardo, CA). Figures shown are representatives of three or more experiments each.
Cells in 8-chamber slides were infected as indicated before fixation for 10 min with 4% paraformaldehyde. They were washed with PBS, permeabilized with 0.2% Triton X-100 for 10 min, and blocked with Image-IT FX signal enhancer (Life Technologies) for 10 min. Cells were incubated with primary antibodies for 1 hour at room temperature in PBS, 1% BSA, washed, and incubated with Alexa fluor-conjugated secondary antibodies for 1 hour at room temperature in PBS, 1% BSA. They were washed again and mounted onto slides using Slowfade gold mounting reagent with DAPI (Life Technologies). BrdU-labeled virus was detected as described [93]. Briefly, cells were fixed for 10 min with 4% paraformaldehyde then treated with 4N HCl for 10 min at room temperature to expose BrdU to antibody staining. Cells were washed with PBS, permeablized with 0.2% Triton X-100 for 10 min, and blocked with Image-IT FX signal enhancer (Life Technologies) for 10 min. Cells were stained with a rabbit-derived antibody to BrdU and Alexa Fluor-conjugated secondary antibodies (Life Technologies) and mounted onto slides as above. Cells were imaged using a Nikon Eclipse 80i fluorescence microscope and Metamorph software (Molecular Devices, Silicon Valley, CA). Figures shown are representatives of two or more experiments each.
Total RNA was extracted from cells using Trizol (Life Technologies), according to the manufacturer's instructions. Briefly, cells were homogenized in Trizol and mixed with chloroform to separate the proteins and nucleic acids. RNA was precipitated from the aqueous layer using isopropanol and washed in 75% ethanol before resuspension in RNase-free water. RNA was DNase treated for 30 min at 37°C (Promega, Madison, WI) and reverse transcribed using Multiscribe Reverse Transcriptase (Life Technologies) with random primers, according to the manufacturer's instructions.
To perform ChIP assays, we used a protocol modified from two previous studies [98], [99]. Briefly, 90–95% confluent T-150 flasks of cells were cross-linked for 10 min by adding formaldehyde to a final concentration of 1%. Crosslinking was stopped by adding glycine to 125 mM for 5 min. Cells were collected and resuspended in cell lysis buffer (5 mM PIPES, pH 8.0; 1 mM EDTA; 1% SDS; protease inhibitors) and incubated on ice for 15 min before being passed through a 27.5 gauge needle and a 30 gauge needle 5 times, each. Nuclei were pelleted by centrifugation and then resuspended in nuclear lysis buffer (50 mM Tris, pH 8.1; 10 mM EDTA; 1% SDS; protease inhibitors), incubated on ice for 15 min, and sonicated at an amplitude of 40, 10 seconds on, 10 seconds off for 20 min. Debris was cleared by centrifugation and supernatant was flash frozen in liquid N2 and stored at −80°C overnight. Nuclear lysates were diluted in ChIP dilution buffer (0.01% SDS; 1.1% Triton X-100; 1.2 mM EDTA; 16.7 mM Tris, pH 8.1; 167 mM NaCl; protease inhibitors). Diluted lysates were precleared for 30 min at 4°C with salmon sperm DNA/protein G agarose slurry (Millipore, Billerica, MA) and then incubated overnight at 4°C with 1.5 µg of the indicated antibody. Immune complexes were collected with salmon sperm DNA/protein G agarose slurry for 1 hour at 4°C and washed with low salt wash (0.1% SDS; 1% Triton X-100; 2 mM EDTA; 20 mM Tris, pH 8.1; 150 mM NaCl), then high salt wash (0.1% SDS; 1% Triton X-100; 2 mM EDTA; 20 mM Tris, pH 8.1; 500 mM NaCl), and LiCl wash (0.25 M LiCl; 1% NP-40; 1% deoxycholate; 1 mM EDTA; 10 mM Tris, pH 8.1). Complexes were eluted in elution buffer (1% SDS; 0.1 M NaHCO3). The crosslinking was reversed by adding 1 µL RNase A and NaCl to a final concentration of 0.3 M NaCl and incubating at 65°C for 5 hours. Protein was removed by incubating lysate with proteinase K at 55°C for 1 hour. DNA was precipitated with ethanol and resuspended in nuclease-free water before real-time PCR with TSS primers (Table 1).
Real-time PCR was performed using SYBR green (Life Technologies) and primers (Table 1) according to the manufacturer's instructions. Briefly, 2 µL of cDNA (for mRNA experiments) or immunoprecipitated DNA (for ChIP experiments) was added to real-time PCR reaction mixtures containing SYBR green reaction mixture (final concentration of 1×) and the appropriate primers (final concentration of 0.25 µM, forward and reverse). An ABI Prism 7500 real-time PCR system was used to amplify and detect cDNA. cDNA data were normalized to GAPDH Ct levels and ChIP data were normalized to input samples, as indicated, using the ddCt method [46]. Figures shown are representatives of three or more experiments each.
PLA was performed using the DuoLink PLA Kit (Sigma-Aldrich) to detect protein–protein interactions using fluorescence microscopy as per manufacturer's protocol. Briefly, HFF cells were cultured and infected with EdU-labeled HSV-1 (1 pfu/cell) for the indicated times in 8 chamber microscopic slides, fixed with 4% paraformaldehyde for 15 minutes at room temperature, permeabilized with 0.2% Triton X-100 and blocked with DuoLink blocking buffer for 30 minutes at 37°C. Cells were then incubated with primary antibodies against IFI16 (mouse monoclonal and rabbit polyclonal), diluted in DuoLink antibody diluents for 1 hour, washed and then further incubated for an hour at 37°C with species-specific PLA probes under hybridization conditions and in the presence of 2 additional oligonucleotides to facilitate the hybridization only in close proximity (<40 nm). Hybridized oligonucleotides were ligated to form a closed circle, which served as a template for rolling-circle amplification after adding an amplification solution to generate a concatemeric product extending from the oligonucleotide arm of the PLA probe. Fluorescently labeled oligonucleotides were hybridized to the concatemeric products and the signal was detected as distinct fluorescent dots in the Texas red channel and analyzed by fluorescence microscopy as above.
IFNβ secretion was detected using the Verikine™ Human IFN Beta ELISA kit (PBL Interferon Source, Piskataway, NJ) according to the manufacturer's instructions. Infected cell supernatant was collected at 6 h p.i., diluted 1∶1 with sample dilution buffer and attached to the assay wells by incubation at room temperature for 1 h. Wells were washed and incubated 1 h at room temperature with diluted antibody solution, then washed again and incubated 1 h at room temperature with diluted HRP solution. Wells were washed again and incubated 15 min at room temperature with TMB substrate solution in the dark. The reactions were stopped by the addition of stop solution and the absorbance at 450 nm was read using a Synergy2 Biotek plate reader (Biotek, Winooski, VT).
|
10.1371/journal.pntd.0004668 | Underestimation of Leptospirosis Incidence in the French West Indies | Leptospirosis is a neglected zoonosis affecting mainly tropical and subtropical regions worldwide, particularly South America and the Caribbean. As in many other countries, under-reporting of cases was suspected in the French West Indies because of inadequate access to diagnostic tests for the general population.
In order to estimate the real incidence of leptospirosis in Guadeloupe and Martinique, a study was performed in 2011 using the three prevailing available biological tests for diagnosis: Microscopic Agglutination Test (MAT), IgM ELISA and PCR. The study investigated inpatients and outpatients and used active case ascertainment from data provided by a general practitioners’ sentinel network. The epidemiology of the disease was also described in terms of severity and demographic characteristics. Leptospirosis incidence was estimated at 69.4 (95%CI 47.6–91.1) and 60.6 (95%CI 36.3–85.0) annual cases per 100 000 inhabitants in Guadeloupe and Martinique, respectively, which was 3 and 4 times higher than previous estimations.
Inclusion of PCR and IgM ELISA tests for diagnosis of leptospirosis resulted in improved sensitivity in comparison with MAT alone. Our results highlighted the substantial health burden of the disease in these two territories and the importance of access to appropriate laboratory tests. Based on our results, PCR and IgM ELISA tests have now been included in the list of tests reimbursed by the national system of social security insurance in France. Our results also underline the relevance of implementing an integrated strategy for the surveillance, prevention and control of leptospirosis in the French West Indies.
| Leptospirosis is a common disease in tropical regions around the world. It is caused by a bacteria excreted in environmental waters by mammals, especially rodents, through their urine. Leptospirosis has symptoms similar to other tropical diseases, including dengue fever, and early laboratory diagnosis is crucial to provide both appropriate treatments for patients and rapid control measures when an outbreak occurs. In 2011, we undertook a study to determine the incidence of leptospirosis in two territories (Guadeloupe and Martinique) in the French West Indies by establishing a surveillance network and implementing new diagnostic assays in order to ensure an exhaustive diagnostic analysis. We concluded that leptospirosis was previously significantly under-reported in the French West Indies and we recommended: 1- access to these new diagnostic tests for the entire population for a better detection of leptospirosis patients and, 2- the implementation of an integrated surveillance, alert and prevention system for the disease in this region. Our findings had raised the awareness of this neglected disease in the French West Indies and, as another consequence; new diagnostic tests are now reimbursed by the social security insurance in France.
| Leptospirosis is a zoonotic bacterial disease which is particularly widespread in tropical and subtropical regions. It produces a wide array of clinical symptoms, ranging from an undifferentiated mild fever to severe multi-organ failure [1]. None of these symptoms is specific to the disease. Until recently, diagnosis was mostly based on serological tests, as antibodies are detectable in blood by the second week after the onset of symptoms. PCR-based methods are becoming more widely used for the detection of bacterial, in part because of their superior sensitivity and ability to establish an early diagnosis. The disease can usually be cured in humans within a few weeks without sequelae using appropriate antibiotic therapy [2]. However, it often requires hospitalization during the acute phase and complications related to the disease can be fatal. Leptospirosis disease burden estimates have been recently updated by Costa et al. [3]. The authors estimate that worldwide 1.03 million cases and 58,000 deaths occur annually. In addition, the social cost in years of potential life lost and hospital costs associated with leptospirosis are high when compared with the cost of early treatment and prevention of the infection [4,5]. However, leptospirosis is underdiagnosed worldwide, especially in low-resource tropical countries [6].
The bacteria Leptospira [1] are maintained in nature through the chronic renal infection of host animals, and are excreted in the hosts’ urine. Leptospirosis is transmitted through contact of abraded skin or mucous membranes either directly with infected urine or organs or indirectly with contaminated soil or water. Accordingly, activities which bring humans into contact with such contaminated environments increase the risk of contracting the disease. These include farming, gardening, building work, animal husbandry, hunting, fishing and water sports in fresh water environments.
This study focused on the two French overseas territories of Guadeloupe and Martinique (approximately 400,000 inhabitants in each) which are located in the French West Indies (hereafter FWI). Guadeloupe is an archipelago which includes two main islands, Grande-Terre and Basse-Terre (hereafter jointly covered by the name “Guadeloupe”). Guadeloupe and Martinique have a similar tropical climate with a rainy season between July and December. Furthermore, the populations are quantitatively and qualitatively similar as far as ethnic and socio-economical characteristics are concerned The population of the FWI is mainly of African or mixed descent. There are also Europeans, Indians, Lebanese, Syrians, Chinese, and Amerindians (remnants of the original pre-European population). Life expectancy at birth for males in 2013 was 76 and 79 years, respectively, in Guadeloupe and Martinique, and 85 for females in both territories. The service sector dominates the economy of the two territories while the public sector is the major employer accounting for 42% of total salaried workers. The economy is very dependent on France for subsidies and imports. In 2013, unemployment rates were 25.5% and 22%, respectively, in Guadeloupe and Martinique, 2.5 times higher than in mainland France.
Because of the warm climate in the FWI, outdoor activities are common throughout the year and are easier to undertake without the use of protection (boots, gloves, etc.). The tropical climate promotes the survival of leptospires and their proliferation in wet environments. In addition, numerous Caribbean mammals are hosts to pathogenic Leptospira species including rodents (most frequently), opossums, mongoose, bats, pigs, bovines, goats and dogs [7]. The Leptospira genus includes ten pathogenic species [8]. The most frequent serogroups in Guadeloupe and Martinique are Icterohaemorrhagiae, Canicola and Sejroe. The serogroup Ballum is also frequently reported in Guadeloupe [9].
Between 2002 and 2008, the estimated annual incidence of leptospirosis in Guadeloupe (22.5 per 100,000 inhabitants) and Martinique (13.9 per 100,000 inhabitants) was much higher than that observed in mainland France (0.47 per 100,000 inhabitants) [10]. At that time, biological diagnosis in Guadeloupe and Martinique was performed by sending blood samples to the National Reference Center for leptospirosis in Paris (Pasteur Institute) for microscopic agglutination test (MAT). Because of these logistical and technical limitations, as well as a lack of epidemiological surveillance, it was expected that the disease was underdiagnosed and its public health burden underestimated in the FWI.
In this context, an incidence study was performed in 2011: i) to obtain reliable data and assess the “real” disease burden of leptospirosis in the FWI; ii) to provide scientific evidence for the hypothesis that leptospirosis diagnosis should be reinforced by including tests capable of reliably confirming diagnosis immediately after the onset of the disease in the NABM (Nomenclature des actes de biologie médicale, which is the list of clinical pathology tests which social security insurance in France covers) and iii) to contribute to the implementation of an integrated management strategy based on an epidemiological surveillance, warning and response system.
The study covered the entire population of Continental Guadeloupe and Martinique. Between January 1, 2011 and December 31, 2011 (i.e., the study period), the number of incident cases of leptospirosis was counted in both territories using two sources: i) public hospitals (2 in Guadeloupe and 3 in Martinique) and ii) General Practitioners’ (GPs) sentinel surveillance networks (one on each island). In these two networks, GPs are included on a voluntary basis, according to a sampling strategy based on geographical localization and population density. The activity (annual number of consultations) of the GPs participating to these sentinel surveillance networks represents 20.4% and 22.4% of the total activity of all GPs in Guadeloupe and Martinique, respectively.
Patients were eligible for inclusion in the study if they had lived permanently for one year in Martinique or Continental Guadeloupe and had consulted either a sentinel GP or a healthcare professional in a public hospital for a suspected clinical case of leptospirosis, defined as the acute onset of fever ≥38°C which then continued for less than 14 days, without any other infectious diagnosis and with at least one of the following symptoms: headache, myalgia, arthralgia and lower back pain.
The strategy used for the diagnosis of leptospirosis depended on the time of sampling, as illustrated in the accompanying supplementary document (S1 Fig). Between the first and ninth day of illness (acute phase), a real-time PCR test [11] was locally performed. If it tested negative the IgM ELISA test was performed (also locally) [12–14] and if the latter tested positive (i.e., single titer ≥1:400), definitive confirmation was obtained using a Microscopic Agglutination Test (MAT) which included a panel of 17 antigens [15]. After the ninth day of illness (immune phase), an IgM ELISA test was performed and, if it proved positive, confirmation was then obtained with a MAT. The MAT, which was performed at the National Reference Center for Leptospirosis (Institut Pasteur, Paris, France), was considered positive when the titer was ≥ 1:400 for at least one antigen (except antigen L biflexa serovar Patoc) [16].
If the first blood sample tested negative for all the tests, a second blood sample was recommended two weeks after the first in order to repeat the IgM ELISA. If the latter tested positive, the MAT was repeated.
Finally, leptospirosis was confirmed if the real-time PCR or IgM ELISA and MAT tests tested positive for at least one sample.
Information on gender, date of birth, city of residence, date of the onset of symptoms, date of blood sampling, laboratory test results, and, if relevant, information on hospitalization duration and disease severity, was recorded for each patient, whether the case was confirmed or not, using a standardized form. A case was defined as severe if the person died or was admitted to an intensive care unit or underwent renal dialysis or mechanical ventilation, or when a combination of these criteria was met.
The overall incidence of leptospirosis was estimated using a sampling approach, stratified on the two data sources—hospitals and GPs (Fig 1). All hospital confirmed cases (inpatients and outpatients) were taken into account for the overall incidence calculation. The number of cases confirmed outside of hospital was estimated from sentinel GPs figures, using a random two-level sampling method calculation as follows: 1) the number of cases reported by the sentinel GPs as eligible cases was extrapolated to the whole population over both territories, based on the total weekly activity rate of participating GPs (first level); 2) the number of blood samples collected by each sentinel GP for eligible cases was considered as a random sample (second level) and the positivity rate of these blood samples was applied to the previous extrapolated number.
When a patient was diagnosed by a sentinel GP and then hospitalized, he/she was considered as an hospitalized case and therefore subtracted from the GPs cases numbers.
The calculation of the confidence interval estimate took into account both the variance of the number of eligible cases reported by GPs, and the variance of the above positivity rate (S1 Protocol. Calculations for the estimation of incidence).
Differences between positivity rates according to the source of data and to the specific territory (i.e. either Guadeloupe or Martinique) were tested for their statistical significance with χ2 test. P values < 0.05 were considered significant.
Statistical analyses were performed using Microsoft Office Excel 2003 and Intercooled Stata 08.
This study was part of national public health surveillance program of the Institute for Public Health Surveillance (Institut de Veille Sanitaire, InVS), a governmental agency reporting to the French Ministry of Health. Therefore, consultation with ethics committee was not required. Information on leptospirosis was distributed and each participant agreed verbally an informed consent to participate as a volunteer in the study and could withdraw anytime without further obligation (S1 Consent form). Diagnostic test results was provided free of charge to the participants. The study was approved by France's data protection commission (CNIL) under number DR-2011-96 (S1 Authorization). All data used in the study was anonymized.
A total of 1,305 suspected cases were included in the study in both territories, 1,167 being recruited in hospitals and 138 through the sentinel networks. The total number of hospitalized-confirmed cases was 126 in Guadeloupe and 108 in Martinique (Table 1). By extrapolating the data reported by the sentinel networks, the total estimated number of cases was 267 and 240 in Guadeloupe and Martinique, respectively (Table 2). The corresponding overall incidence (per 100 000 inhabitants) was 69.4 and 60.6 in Guadeloupe and Martinique, respectively.
In total, in both territories, more than a third of biological confirmations were obtained using PCR. In Guadeloupe, secondary blood sample testing—using IgM ELISA and MAT—confirmed 31% of diagnosed cases.
The number of included patients (i.e. suspected cases) and the number of confirmed cases were higher in hospitals than in the sentinel networks (the latter being samples of all the GPs of each territory). Positivity rates for biological diagnosis ranged from 13 to 22% according to the data source and territory. No difference was observed between positivity rates of sentinel GPs’ patients from Guadeloupe and Martinique (13%), or between positivity rates of hospital patients and sentinel GPs’ patients (15 vs 13%) in Martinique (p>0.05). A moderate difference (22 vs 15%) was observed between the positivity rates of hospital patients from Guadeloupe and Martinique (p = 0.02), and between the positivity rates of hospital and sentinel GPs’ patients (22 vs 13%) in Guadeloupe (p = 0.05).
In Table 2, the estimated incidences in Guadeloupe and Martinique in 2011 are compared with the figures for the reference period 2002–2008. In our study in 2011, the overall estimated incidence of leptospirosis was three times higher for Guadeloupe and four times higher for Martinique compared with the reference period 2002–2008. In addition, a difference of 12% was observed between Guadeloupe and Martinique in 2011, compared with almost 40% for the reference period (p < 0.01).
In both territories, the positivity rates was most commonly observed in adults aged 20–59, than in the over-60 population (Table 3). The study also showed, for the first time in Guadeloupe and Martinique, that leptospirosis also occurs in children, with cases confirmed in persons younger than 10. Men were six times more likely than women to be affected by leptospirosis in both territories, the sex ratios of confirmed cases being similar in Guadeloupe (6.4) and Martinique (6.2). This trend was observed across all age groups, being statistically significant among adults and, in Guadeloupe, among people over 60 years.
Disease severity indicators are displayed in Table 4. In both Guadeloupe and Martinique, these indicators confirm the leptospirosis disease burden. The eight deaths which occurred in Guadeloupe were directly attributed to leptospirosis by hospital specialists in infectious diseases.
In 2011, the estimated number of confirmed cases of leptospirosis in Guadeloupe was 267, comprising 115 hospital cases and an estimated 152 GP cases. In Martinique, 240 cases were confirmed, comprising 101 hospital cases and an estimated 139 GP cases. The total estimated number of cases for each territory–approximately 250 cases per year—was close to that observed in mainland France, where the population is approximately 120 times greater [15], indicating that the burden of leptospirosis is much higher in the FWI. The incidence in FWI remained high in the last few years (2012–2014) with a similar number of laboratory-confirmed cases in both territories.
Approximately seventy and sixty cases were estimated per 100,000 inhabitants in 2011, respectively, in Guadeloupe and Martinique. These estimates were respectively three and four times higher than each territory’s average incidence during the reference period 2002–2008 (reported by the National Reference Center) (Table 2). Our study shows that a significant higher number of cases was detected when both IgM ELISA and PCR tests are used, further indicating that the lack of adequate diagnostic tests contributes to under-reporting of cases [3].
However, some limitations of our study must be considered. First, sentinel GPs are not randomly selected. However, these n sentinel GPs networks have been widely used for a decade to estimate the numbers of suspected cases of other diseases in Guadeloupe and Martinique (including dengue and, more recently, chikungunya) and the estimates obtained from these networks were coherent with those obtained from other surveillance systems, as laboratory-based data or hospital emergency department data [17].
Second, 22% of eligible patients were not included in the study because sentinel GPs did not test them for leptospirosis. The reasons for non-inclusion were diverse (GPs in holidays, misinterpretation of the case definition or inclusion criteria, etc.) and may not introduce bias. We therefore considered that eligible patients were included at random to receive a prescription for biological testing.
Leptospirosis belongs to the group of neglected tropical diseases [3] which comprises some of the most common infections in Latin America and Caribbean countries. The 7.5% rate of severe leptospirosis (ratio of the number severe cases / total estimated number of cases) observed in this study in 2011 is much higher than the 0.3% recorded for dengue during the most recent epidemic in 2010 [17–18]. The annual incidence of leptospirosis has been shown to be associated with climate and meteorological conditions. Thus, heavy rainfall results to increased survival of Leptospira in the environment and increased exposure of humans to water. In 2011, both territories experienced heavy rainfalls, but no cyclone [19,20]. A long-term surveillance system would therefore be required to accurately describe annual variations in the disease incidence, for example during the El Niño Southern Oscillation periods [21].
The incidence of severe cases of leptospirosis was 5.2 in Continental Guadeloupe and 3.3 in Martinique per 100,000 inhabitants in 2011. This reported incidence is similar to the one recorded in Réunion Island in 2011where the incidence of cases transferred to intensive care (only) was 2 per 100,000 inhabitants [22].
Case fatality rates (CFR) observed in Guadeloupe (3%) and in Martinique (0%) were also similar to that observed in Réunion Island for the period 2004–2008, where it ranged from 0% to 7% depending on the year, except in 2006 when leptospirosis lethality reached 38% during the chikungunya epidemic [23]. Possibly due to the relatively higher income status of FWI and to access to early and free testing during our study period, estimated case fatality rates in FWI are lower than the global estimations of Costa et al. [3].
The demographic characteristics of the cases in our study match those described in the literature, albeit with an increased proportion of older persons in the FWI [24, 25]. The predominance of male cases is generally attributed to the hypothesis that men are potentially more exposed than women due to more frequent at-risk activities. However, we also observed this ratio in the older age group where women are more numerous than men, and probably have similar daily activities as the latter, suggesting that other hypothesis needs to be evaluated.
Improving patient care is a priority. Access to diagnosis is crucial, because treatment for leptospirosis patients is much more effective if antibiotics are administered as early as possible following the onset of disease [5,26]. Early diagnosis of acute leptospirosis by real-time PCR would prevent potential complications and limit periods of stays in hospital. This assay, which was not available in Guadeloupe before the study, is now routinely used in both territories. During the immune phase of the disease (from the end of the first week), the IgM ELISA, which usually becomes positive earlier than MAT in the course of the illness, can also offer useful support to physicians to make good treatment decisions. The IgM ELISA is a simple and rapid method which is not requiring the use of sophisticated laboratory equipment or trained personnel. Partly because of the results of this study, in September 2014 the French social security insurance decided to reimbursed the cost of both PCR and ELISA for the diagnosis of leptospirosis in mainland France and French overseas territories.
Implementation of an epidemiological surveillance system including the systematic collection and analysis of data should allow the public health community to respond more quickly to a given epidemiological situation (clustered cases, seasonal outbreaks, evaluation of prevention and control measures, etc). These results advocate for an integrated surveillance, early warning and management strategy to reduce the incidence and severity of leptospirosis. The experience of dengue, which prompted the implementation of an integrated management strategy promoted by the WHO in the FWI, could serve as a model [18, 27].
|
10.1371/journal.pntd.0005362 | Schistosome infectivity in the snail, Biomphalaria glabrata, is partially dependent on the expression of Grctm6, a Guadeloupe Resistance Complex protein. | Schistosomiasis is one of the most important neglected tropical diseases. Despite effective chemotherapeutic treatments, this disease continues to afflict hundreds of millions of people. Understanding the natural intermediate snail hosts of schistosome parasites is vital to the suppression of this disease. A recently identified genomic region in Caribbean Biomphalaria glabrata snails strongly influences their resistance to infection by Schistosoma mansoni. This region contains novel genes having structural similarity to known pathogen recognition proteins. Here we elaborate on the probable structure and role of one of these genes, grctm6. We characterised the expression of Grctm6 in a population of Caribbean snails, and performed a siRNA knockdown of Grctm6. We show that this protein is not only expressed in B. glabrata hemolymph, but that it also has a role in modulating the number of S. mansoni cercariae released by infected snails, making it a possible target for the biological control of schistosomiasis.
| Schistosomiasis is one of the most prevalent parasitic diseases in the world. Though treatments for schistosomiasis infection exist, there is no vaccine, and reinfection is common in areas where the parasite occurs. One possible way to mitigate schistosomiasis is by controlling the transmission of the parasite larvae from the snails that carry them. Understanding the snail-parasite relationship is essential for the development of new means to interrupt transmission of the parasite from snails to humans. Snails possess immune mechanisms for fighting infection, most of which are based in hemolymph tissue. Here we characterize a novel protein, Grctm6, in a snail host of schistosome parasites. Grctm6 is structurally similar to certain other immune proteins and is present in snail hemolymph. Importantly, we demonstrate that the release of schistosomes by infected snails is exacerbated when this protein is experimentally suppressed in live snails. These results support the suitability of Grctm6 as a possible target for reducing the transmission of this human disease.
| The World Health Organization (WHO) has estimated that schistosomiasis, a detrimental parasitic helminth disease, affects approximately 258 million people, making it one of the most important parasitic diseases in the world [1, 2]. Millions of people are chemotherapeutically treated for schistosomiasis, but in areas where this parasite is endemic there are high rates of reinfection and persistent debilitating illness. Schistosomiasis-attributed mortality in sub-Saharan Africa alone exceeds 250,000 per year and, with no effective human vaccine, alternative control methods are vital for reducing the burden of this neglected tropical disease [3].
Schistosome miracidia must infect a compatible aquatic snail host in order to produce the cercarial stage that is capable of infecting human hosts. Controlling this intermediate snail host is a primary method for the alternative control of schistosomiasis [4]. Biomphalaria glabrata is the intermediate freshwater snail host of Schistosoma mansoni in the Americas, and has been a target for the successful control of schistosome transmission since the 1950s [5, 6]. Though the initial success of this biological control strategy was limited to Puerto Rico [5], contemporary attempts have expanded to other Caribbean island systems, East Africa (on another snail host), and South America [4–10]. Efforts to control B. glabrata populations have commonly employed the introduction of carnivorous or competitive snail species, but molluscicides have also been heavily exploited [4–12]. Both of these measures can have negative ecological impacts [11]. Despite these consequences, snail control has been shown to be the most successful approach to reduce the prevalence of schistosomiasis, particularly if it is paired with human pharmacological treatment [12]. Recent efforts have begun to focus on determining the relative importance of individual B. glabrata genes on schistosome-infection resistance, with the goal of characterizing snail immune responses to infection, and eventually manipulating snail populations so that they are more naturally resistant to schistosome infection [13–15].
Allelic variation in the Guadeloupe Resistance Complex (GRC), a recently discovered novel gene region in B. glabrata, has been shown to strongly influence Guadeloupean B. glabrata (BgGUA) resistance to Guadeloupean S. mansoni (SmGUA) infection [13]. Allelic variation in this genomic region has an 8-fold effect of infection odds, greater than for any other known snail locus [13, 16, 17]. Resistance is dominant, suggesting a mechanism of parasite recognition and/or clearance by the host, rather than host recognition by the parasite [13]. There are three distinct haplotypes in the GRC region (with 15 coding genes), which we designate R (for the dominant allele that confers increased resistance), S1 and S2 (for the two alleles that confer increased susceptibility; S1 and S2 are equivalent in their effects). The GRC region contains several genes having structural similarity to membrane-bound, pathogen recognition molecules and receptors such as Toll-like receptors and Fc receptors. The region also appears to be under balancing selection, again consistent with a role in pathogen recognition [13]. Determining the functions of these genes, and their potential immunological roles during schistosome infection, is vital for understanding schistosome-infection resistance by this snail species. Given that Biomphalaria species are major intermediate hosts for human schistosomiasis, understanding how schistosome infections can be controlled in these snails may provide insights into ways to proactively limit schistosomiasis transmission. In the present study, we chose one of the GRC genes for in-depth functional analysis: grctm6, which encodes the Guadeloupe Resistance Complex Transmembrane 6 (Grctm6) protein. grctm6 is a particularly compelling candidate locus because the resistant allele at this locus shows high non-synonymous substitution relative to the two susceptibility alleles (particularly in the predicted extracellular domain), susceptibility is not correlated with mRNA levels, and because bioinformatic structural analyses confirms that Grctm6 is a potential candidate for immunological activity due to its predicted transmembrane structure. We report that this gene is expressed at the protein level in hemolymph, and demonstrate that a short interfering RNA (siRNA) knockdown of Grctm6 increased the number of cercariae released into the environment by treated snails.
B. glabrata (BgGUA: “snails”) and S. mansoni (SmGUA: all miracidia or cercariae described) were collected in 2005 in Guadeloupe and maintained as previously described [13, 18]. The SmGUA strain of S. mansoni was cycled through BgGUA and hamsters, and parasite eggs were isolated from rodent livers. BgGUA snails were genotyped based on their GRC locus as previously described [13]. From the outbred BgGUA population we isolated 6 independent, partially-inbred lines that were homozygous at the GRC locus (2 RR, 2 S1S1, and 2 S2S2 lines). We used these lines to verify the baseline resistance (percentage infected) and levels of constitutive expression of grctm6 in each of the three genotypes. All RNAi studies were done on a single RR line. Snails for all experiments were size matched (~7 mm) and housed identically. The Oregon State University Institutional Animal Care and Use Committee, which adheres to Public Health Service Domestic Assurance for humane care and use of laboratory animals (PHS Animal Welfare Assurance Number A3229-01), approved this research as Animal Care and Use Proposal 4360.
Alignment of the protein products of the three alleles of Grctm6 found in BgGUA were calculated previously from RNA-sequencing [13]. We calculated protein molecular weights using Science Gateway (http://www.sciencegateway.org/tools/proteinmw.htm). We examined secondary structure using PSIPRED (http://bioinf.cs.ucl.ac.uk/psipred/). In addition, the signal peptide (http://www.cbs.dtu.dk/services/SignalP/), transmembrane domain (http://www.cbs.dtu.dk/services/TMHMM/), and asparagine glycosylation (http://www.cbs.dtu.dk/services/NetNGlyc/) were predicted using the Center for Biological Sequence Analysis’ prediction servers. Homology searches were performed with DELTA-BLAST (https://blast.ncbi.nlm.nih.gov/) and Pfam (pfam.xfam.org/).
Parasite challenges were carried out as previously described [13]. In brief, snails were placed in 2 ml of dechlorinated water in individual wells of a 24 well dish containing 10 or 20 miracidia for 24 h, and subsequently transferred into tubs containing 10 snails each to be monitored for infection. All infections were conducted at 1 pm in the afternoon following animal sacrifice at 11 am. Two independent lines of each RR, S1S1, and S2S2 snails were challenged and pooled by genotype for the verification of GRC locus susceptibility. This was done on two separate occasions using a minimum of 30 snails each time (n = 64 S1S1, 68 S2S2, 86 RR). These snails were examined for cercarial shedding, and scored as either infected or uninfected. Starting 5 weeks post challenge, and for 5 subsequent weeks, snails were placed in 2 ml of dechlorinated water in individual wells of a 24 well dish and exposed to light for 3 h beginning at 9 am. For the siRNA experiments, RR snails were treated with enhanced green fluorescent protein, (GFP, Sham injected) or grctm6 oligonucleotides (oligos) and then challenged with miracidia. We challenged using 10 or 20 miracidia, and did two independent trials for each number of miracidia. We used a minimum of 30 snails per trial (Snails that survived for analysis: n = 55 snails using GFP oligos, and n = 37 using grctm6 oligos for 20 miracidial challenges; n = 65 using GFP oligos, and n = 63 using grctm6 oligos for 10 miracidial challenges). These were examined for cercarial shedding as described above [14, 19, 20]. siRNA treated snails were challenged at the beginning of the third day post siRNA injection. Cercariae were enumerated by taking three equal aliquots from a 2 ml sample (or counting all cercariae in a well if density was low). All shedding snails were individually marked with nail polish so that a cumulative count over the 5 week scoring period could be achieved for each snail.
Specific siRNA oligos for grctm6 were designed and produced by Integrated DNA technologies (IDT). Oligos aligned to the more conserved 3′ intracellular domain. Three oligos (GUUAGGACACCGUCAAUU, CACUGCUGACAUUGGCAG, UUUCAUUUGCAUUGCUUG) were suspended according to the manufacturer’s instructions and injected into live ~7 mm RR snails at 2 μg/μl as previously described [14, 19]. In brief, either enhanced GFP IDT control oligos (GFP /Sham) or the grctm6 oligo mixture was suspended in Xfect transfection reagent with nanoparticles (Clontech) according to the manufacturer’s instructions and each snail received a single 10 μl injection distal to the heart [14, 15]. The siRNA-mediated knockdown was assessed at the mRNA level over 4 days (the end of day 0–4), and additionally confirmed by Western blot analysis of protein levels 2–4 days post-injection. Mortality was also compared, within each single day, between GFP control oligos and the grctm6 oligo mixture to ensure that the grctm6 oligo mixture was not inducing additional mortality.
Quantitative RT-PCR (qPCR) was used to quantify mRNA transcripts of grctm6 in whole snail lysates of BgGUA snails, and used to detect the extent of siRNA knockdown of grctm6 mRNA in RR snails. Constitutive levels of grctm6 mRNA were assessed in 2 independent homozygous lines of each RR, S1S1, and S2S2 snails and pooled within a genotype (2 RR lines pooled, 2 S1S1 lines pooled, and 2 S2S2 lines pooled). RR snails were also assessed for their grctm6 mRNA levels 0, 1, 2, 3, and 4 days post injection of GFP or grctm6 oligos. In brief, whole snails were snap-frozen in liquid nitrogen, total RNA extracted using the Direct-Zol RNA miniprep kit (Zymo Research) and cDNA synthesized using iScript Reverse Transcriptase Supermix for RT-qPCR (BioRad). Additionally, the head-foot, albumen gland, or hemolymph were removed and snap frozen for tissue analysis of R snails. Hemolymph was collected by head-foot retraction as previously described [14]. qPCR was performed, as previously described [21–23]. All primers were prepared at 300 nM, had a single melt curve, had efficiencies between 90–100%, and were designed or verified using Primer 3 (National Center for Biotechnology Information). grctm6 primers aligned to the more conserved 3′ intracellular domain and showed 100% identity to all three alleles. B-actin (F: 5’-GCTTCCACCTCTTCATCTCTTG -3’; R: 5’-GAACGTAGCTTCTGGACATCTG-3’) was used as an internal control, and did not vary across treatments. grctm6 (F, 5′-TGTTGAGTACGCTGCTGTCAATAAG -3′; R, 5′- ATTCATATCCTTGTTGCTTGGGTCC-3′) was used with the following PCR conditions (in a Applied Biosystems 7500 fast qPCR thermocycler): 95°C for 5 min; 40 cycles of 95°C for 15s and 60°C for 15s. All mRNA levels of GRC lines were normalized to β-actin expression and presented relative to S2 snails. All mRNA levels of oligo injected RR snails were normalized to β-actin expression and presented relative to GFP control samples.
Rabbit polyclonal Grctm6 antibodies were produced against a peptide epitope (within RR Grctm6, isolated, purified, and validated by Genscript custom antibody services (Genscript). Re-validation of the antibody was performed in house and it was found to be effective at a concentration 1:2000 for western blot detection of synthetically generated Grctm6 peptides and native Grctm6 isolated from hemolymph. Genscript Anti-rabbit IgG secondary HRP conjugate (1:2000) was used for detection.
Western blots were used to detect the presence of Grctm6 protein in RR snail hemolymph following siRNA knockdown as previously described with the modifications described below [15]. Snail tissue preps (whole snail, albumen gland, head-foot, hemolymph +/- hemocytes, and hemocytes) were examined from pooled untreated samples from RR snails. Hemolymph preparations were obtained using the head-foot retraction method and either directly added to lysis buffer (Bolt LDS sample and reducing buffer (Thermo)), or cells were removed by centrifugation at 1200g before protein extraction was performed. Unmodified hemolymph produced a consistently detectable band when interrogated with a Grctm6 polyclonal antibody, so this tissue was extracted and used for knockdown analysis. Snail hemolymph was collected, pooled (4–10 snails per sample), and mixed with Bolt LDS sample and reducing buffer (Thermo), and homogenised using a 25G needle before being heated to 95°C for 7 minutes. Tissue samples (whole snail removed from the shell, albumen gland, and head-foot) were homogenized in BOLT LDS sample and reducing buffer using a 18G needle followed by a 25G needle and then treated identically to hemolymph samples. Total protein levels in each sample were quantified using absorbance at 280 nm (Nanodrop, Thermo), additional loading buffer was added to more concentrated samples to ensure all samples had equivalent concentrations of total protein, and 750 ng of total protein was used for each sample per well (500 ng for Fig 3B) [15]. This was the maximum protein concentration that could be used across all samples. Electrophoresis was performed in a 10% pre-cast Bolt Bis-Tris gel for 20 minutes at 200 V. Samples were blotted using a Pierce power blotter cassette (Thermo), and western blot detection was done using an iBind Solutions kit and iBind western device (Thermo). Detection was achieved via Supersignal West Pico luminol solution (Thermo) and chemiluminescence was acquired using a MyECL imager (Thermo). Densitometry was performed post image acquisition using ImageJ software (NIH), and was calculated on the only clear band, which coincided with the predicted 68 kDa size of Grctm6. Three independent blots were run, each inclusive of all 3–5 samples/treatment (5 on Day 3). Relative density normalized to BgActin was calculated independently for each western blot and then averaged as a triplicate for each sample (comparisons within a day only).
Statistical analyses were completed as indicated, and were generally completed by one-way ANOVA (or unpaired Student’s t-test) with a Tukey post-test unless otherwise specified (p<0.05). If a Barlett’s test (or F-test) for equal variance failed, then data underwent a natural log transformation (ln) before reanalysis. Analyses of the susceptibility of snail populations were done by calculating the Z score (standard score) of the population. Analyses were completed using GraphPad Prism software (La Jolla, CA, USA).
In the GRC region there are seven genes coding for putative membrane spanning proteins that have structural similarities [13]. We examined one of the two genes that were identified by Tennessen et al. [13] as most likely to be responsible for schistosome resistance. Our further analysis of Grctm6 indicates that it is likely a single-pass transmembrane protein of ~68 kDa (Fig 1). The predicted extracellular sequence of the R allele protein differs substantially from those of the two S alleles, but the signal sequence and the transmembrane and cytosolic domains are far less variable between the R and S alleles (Fig 1). This highly variable extracellular domain also has the potential to be glycosylated, which is common in B. glabrata [24]. Interestingly, both the suspected extracellular and intracellular stretches of this protein are long (>200aa) and are likely to support substantial secondary structure (extracellular: 9–11% α helices, 30–39% β strands; intracellular: 6–13% α helices, 7–13% β strands). Therefore, it is likely that both contain unidentified functional domains (Fig 1). However, consistent with previous results [13] we found no significant homology to any known protein domains [13].
Among our inbred lines, RR snails were ~25% as likely to be infected by SmGUA as either S1S1 or S2S2 snails (p <0.01; Fig 2A ~0.2 vs ~0.8). These same snail lines were examined for mRNA expression of grctm6. RR snails showed no consistent corresponding increase or decrease relative to susceptible snails, although S2S2 snails had ~2–3 fold higher mRNA levels of grctm6 than S1S1 or RR snails (p<0.01; Fig 2B). These findings verify, for our inbred lines, that the constitutive mRNA expression of grctm6 in whole snails does not explain the resistance of the different genotypes at the GRC locus, and that amino acid sequence divergence may be biologically important for this gene’s function [13]. grctm6 mRNA transcripts were detected in all of the snail tissues that were isolated. Transcript levels appear to be slightly elevated in hemolymph, although a statistical difference was only found between hemolymph and the head-foot (p = 0.03; Fig 3A). When Grctm6 protein levels were examined, the only preparations that had consistently detectable Grctm6 protein were from whole hemolymph lysates (Fig 3B). Grctm6 protein was possibly present in isolated hemocytes and cell free hemolymph, but only unmodified hemolymph (hemolymph that received no manipulations prior to protein extraction) produced a consistently detectable band at ~68kDa at various total protein concentrations (Fig 3B). A more sensitive antibody and immunohistological analysis would be required to definitively determine the tissue specific/cellular location of Grctm6. It is puzzling that we were only able to consistently detect Grctm6 in unmodified hemolymph. Perhaps some part of the cell separation protocol modifies or destroys the epitope our antibody binds. It is possible that spinning the cells triggers an intracellular trimming of Grctm6 and loss of the epitope our antibody recognises. It is also possible that, because we are using a novel polyclonal antibody, we were unable to detect smaller amounts of Grctm6 in other tissues because of low sensitivity. Regardless, this is the first evidence that Grctm6 exists at a protein level in any tissue or species of Biomphalaria.
Using RR snails, we knocked down grctm6 mRNA via siRNA. Injections of oligos caused approximately 25–30% mortality over the first days post-injection (Fig 4A). However, our siRNA knockdown of Grctm6 did not increase mortality beyond that of the control (GFP), so the initial dip in survival is likely the result of physical damage to the snail from the injection procedure.
grctm6 mRNA was significantly reduced by up to ~60% during the first 3 days post siRNA injection in whole snail lysates (p = 0.02; Fig 4B). Given that grctm6 mRNA levels were reduced after 3 days, but normalize by the end of day 4, we examined extent of the protein knockdown surrounding the third day. The amount of Grctm6 protein was significantly reduced by ~30% in the hemolymph 3 days post siRNA injection, but was unmodified on any other day (p = 0.03; Fig 4C). Since we hypothesise that Grctm6 may have some recognition function, we chose to infect snails at the beginning of day 3, given that the protein levels of Grctm6 are reduced from the control on that day, and may be sub-physiological [13].
Two independent doses of miracidia were used so that potential changes to schistosome susceptibility were not overlooked by using a single dose of miracidia [18]. Resistant snails treated with grctm6 oligos or GFP oligos exhibited equivalent susceptibility to infection by SmGUA (Fig 5A and 5B). However, grctm6 injected snails had significantly higher (~3–4 fold) cercarial shedding of infected individuals that shed at least one cercariae (p = 0.04, 0.03; Fig 5C and 5D). This pattern was apparent when snails were challenged with either 10 or 20 miracidia, although overall cercarial shedding was ~150-fold lower in snails challenged with 10. This is the first evidence that this protein is directly involved in snail host defense to any pathogen, and specifically to schistosomes.
In the last decade, there has been a flurry of breakthroughs elucidating snail-schistosome interactions that have exploited a pharmacological, or an RNAi knockdown of a protein of interest in either the snail host or the schistosome [14, 15, 19, 20, 25–36]. Generally, these RNAi knockdowns have targeted proteins that have been relatively well described in other species, or are homologous to another group of defined-immunologic targets. We used this technique to assess the importance of a completely uncharacterized gene/protein, with no known homologs in other species, on snail-schistosome compatibility following its discovery by linkage mapping.
Tennessen et al. [13] described a novel genomic region (the GRC) with alleles that are strongly associated with snail resistance to schistosome infection. Seven transcriptionally expressed, coding GRC genes, designated the grctm loci, are particularly promising candidates in this genomic region. Though these transcripts (including grctm6) bear little resemblance to any known proteins, they share some characteristics common to immunologic membrane bound receptors (single-pass transmembrane proteins with a highly variable extracellular domain, and sequence divergence between disease relevant alleles) [13]. Grctm6 is of particular interest because of its putative structure and because the R allele at the grctm6 locus has high non-synonymous substitution relative to the S alleles. In this study, we verified that GRC genotypes are strongly correlated with resistance and that constitutive transcript variation does not explain resistance differences between the various alleles in BgGUA snails [13]. Previous RNA-Seq data from outbred snails suggested that both S alleles show approximately 2-fold higher expression of grctm6 than R [13]. Our qPCR results on inbred lines indicate that S2, but not S1, has at least 2-fold higher expression than R, supporting the notion that mRNA levels of grctm6 are not a likely explanation for GRC locus-associated resistance in BgGUA (Fig 2B). We also show that Grctm6 is expressed at the protein level in hemolymph (but were unable to determine if it is specific to hemolymph). Although this partial and transient knockdown of Grctm6 did not significantly change the proportion of hosts infected, it did increase cercarial shedding, indicating that Grctm6 has a role in modulating the extent/burden of the schistosome infection in BgGUA. Increasing the number of miracidia used to challenge snails from 10 to 20 has little effect on the proportion of snails infected, but a huge effect on the number of cercariae shed by infected snails (Fig 5). Thus, it is plausible that grctm6 has a role in controlling the number of miracidia that successfully infect the snail, but that we only observed an effect on cercarial shedding because this trait may be more sensitive, than proportion infected, to the number of parasites that successfully established. Alternately, grctm6 may help to regulate some subsequent larval stage leading to cercariogenesis or cercarial release into the environment. Regardless, we have demonstrated that the Grctm6 protein has an important effect on the extent of schistosome infection.
We speculate, based on amino acid sequences, that Grctm6 may be a membrane bound receptor. R and S alleles exhibit substantive sequence divergences (15–45% amino acid differences, Fig 1) in the extracellular region of this protein, which could indicate variation in extracellular domain stabilities, target ligands, and/or binding affinities [24, 37]. The more conserved region of Grctm6 is located in the putative transmembrane and cytoplasmic regions, which could serve to preserve potential outside-in signaling functions of this protein as in other immune receptor proteins [38]. The high expression of Grctm6 in hemolymph, relative to other tissues, (Fig 3) is noteworthy because that is the location of crucial snail-schistosome interactions [39–41]. A rigorous immunological analysis would be required to determine if any of these speculations regarding the mechanistic or immunological role of Grctm6 on parasite infection or immunity are accurate. Although we have clearly shown that this protein influences the numbers of cercariae shed, whether Grctm6 actually functions as an immune receptor, and at what stage in the infection process it acts, remain to be shown. Grctm6 could have negative impacts on any stage of sporocyst growth or development, and further immunohistological and functional analysis are required to determine how Grctm6 is mechanistically involved in schistosome infection.
It is interesting that knocking down Grctm6 expression affected cercarial shedding but not susceptibility (percentage infected). However, this effect was achieved with just a 30% protein knockdown. Perhaps a full knockdown would yield a much stronger phenotype, including an increased susceptibility. A CRISPR/Cas9 knockout would be one way to conclusively test the role of this locus [42]. It would also be pertinent to examine the cellular location of this protein by immunohistochemistry using monoclonal antibodies from recombinant Grctm6. This technique could provide vital future information pertaining to the biological mechanism of Grctm6. We also note that there are six other transmembrane loci in this region, and can’t rule out the possibility that other loci act in concert with grctm6 to produce the observed susceptibility phenotypes.
Here we have shown that a modest reduction in Grctm6 protein levels affects cercarial output, which suggests that this protein may be involved in a pathway which is important during a challenge by S. mansoni. More importantly, this study provides a potentially new type of target for controlling transmission of schistosomes at the snail stage. If, in the future, the resistant allele of genes like grctm6 could be gene-driven into a population of B. glabrata[43], then it is possible that their resistance to schistosomes could be improved without completely immunocompromising the snail. Using these types of modern molecular methods, we would anticipate fewer deleterious ecological consequences than those following contemporary snail control methods, which involve introduced predators/competitors or molluscicides [44, 45]. Even a genetic manipulation that only reduces the number of cercariae shed into the environment could have epidemiological consequences. Further exploration of these genes, their physiological functions, and potential roles for the control of schistosomiasis will be important for combatting this widespread and destructive disease.
|
10.1371/journal.pbio.1002243 | Insulin and mTOR Pathway Regulate HDAC3-Mediated Deacetylation and Activation of PGK1 | Phosphoglycerate kinase 1 (PGK1) catalyzes the reversible transfer of a phosphoryl group from 1, 3-bisphosphoglycerate (1, 3-BPG) to ADP, producing 3-phosphoglycerate (3-PG) and ATP. PGK1 plays a key role in coordinating glycolytic energy production with one-carbon metabolism, serine biosynthesis, and cellular redox regulation. Here, we report that PGK1 is acetylated at lysine 220 (K220), which inhibits PGK1 activity by disrupting the binding with its substrate, ADP. We have identified KAT9 and HDAC3 as the potential acetyltransferase and deacetylase, respectively, for PGK1. Insulin promotes K220 deacetylation to stimulate PGK1 activity. We show that the PI3K/AKT/mTOR pathway regulates HDAC3 S424 phosphorylation, which promotes HDAC3-PGK1 interaction and PGK1 K220 deacetylation. Our study uncovers a previously unknown mechanism for the insulin and mTOR pathway in regulation of glycolytic ATP production and cellular redox potential via HDAC3-mediated PGK1 deacetylation.
| Phosphoglycerate kinase (PGK1) catalyzes the reversible phosphotransfer reaction from 1, 3-bisphosphoglycerate (1, 3-BPG) to ADP to form 3-phosphoglycerate (3-PG) and ATP. By controlling ATP and 3-PG levels, PGK1 plays an important role in coordinating energy production with biosynthesis and redox balance. In contrast to the extensive investigation of the transcriptional regulation of PGK1, little is known about its post-translational regulation. Here, we report that PGK1 is acetylated at lysine 220 (K220) and this acetylation inhibits PGK1 activity by disrupting the binding with its substrate, ADP. We have identified KAT9 and HDAC3 as the acetyltransferase and deacetylase, respectively, for PGK1. Moreover, we show there is molecular crosstalk between mTOR-mediated HDAC3 S424 phosphorylation and PGK1 K220 acetylation. Our study uncovers a previously unknown mechanism for the insulin and mTOR pathway in regulating glycolytic ATP production and cellular redox potential via HDAC3-mediated PGK1 deacetylation.
| Phosphoglycerate kinase (EC 2.7.2.3; PGK) catalyzes the reversible phosphotransfer reaction from 1, 3-bisphosphoglycerate (1, 3-BPG) to ADP to form 3-phosphoglycerate (3-PG) and ATP. The PGK-catalyzed reaction is the first ATP-yielding step of glycolysis and is essential for energy generation by the glycolytic pathway of aerobes and the fermentation of anaerobes in most living cells [1]. Besides ATP, the other product of PGK-catalyzed reaction is 3-PG, which can not only serve as a glycolytic intermediate but also be oxidized by phosphoglycerate dehydrogenase (PHGDH) to form 3-phosphohydroxypyruvate and thus enter one-carbon metabolism [2]. It is known that one-carbon metabolism involving the folate and methionine cycle integrates carbon units from amino acids, including serine and glycine, and creates various outputs, such as the maintenance of redox status by affecting glutathione biosynthesis and NADPH production, the synthesis of lipids, nucleotides, and substrate for methylation reactions [3–5]. By controlling ATP and 3-PG levels, PGK therefore plays an important role in coordinating energy production with biosynthesis and redox balance.
Regulation of PGK has been studied extensively, with research mostly focused on the transcriptional level. In yeast cells, PGK is one of the most highly expressed genes and accounts for approximately 5% of the total mRNA and protein [6]. PGK gene expression can be regulated by diverse carbon sources, with glucose induction and pyruvate suppression having been observed in yeast cells [7–9]. PGK gene expression is also up-regulated by oxidative stress [10]. Overexpression of PGK can suppress the apoptotic phenotypes induced by high ROS and restore normal aging of yeast cells [11]. In humans, PGK has two isoforms (PGK1 and PGK2), which share 87% identity in amino acid sequence and are structurally and functionally similar, but have different expression patterns [12–14]. PGK1 (Gene ID: 5230) is broadly expressed in most cell types, while PGK2 (Gene ID: 5232) is uniquely expressed in meiotic and postmeiotic spermatogenic cells [13]. PGK1 gene as the target of the hypoxia-inducible transcriptional factor HIF-1α has been reported to be selectively up-regulated by oxidants in cultured human colon carcinoma cells [15] and hepatoblastoma cells [16,17]. In contrast to the extensive investigation on the transcriptional regulation of PGK1, little is known about its post-translational regulation.
Protein acetylation has recently been discovered as an evolutionarily conserved post-translational modification in the regulation of a wide range of cellular processes, particularly in nuclear transcription and cytoplasmic metabolism [18–20]. Together with several recent acetylome proteomic studies [21–23], more than 4,500 acetylated proteins, including human PGK1, have been identified by the mass spectrometric analyses. In this study, we investigate the regulatory mechanism and functional consequence of PGK1 acetylation.
Previous proteomic studies have identified that PGK1 is acetylated on multiple lysine residues [18–20]. Western blotting with a pan anti-acetyl lysine antibody demonstrated that PGK1 was indeed acetylated, and its acetylation was elevated by 3.2-fold in HEK293T cells after treatment with nicotinamide (NAM), an inhibitor of the SIRT family deacetylases [24,25], and trichostatin A (TSA), an inhibitor of histone deacetylase (HDAC) I and II (Fig 1A) [26,27]. By performing enzyme activity assay in vitro, we found that the specific enzyme activity of PGK1 was decreased by as much as 63% after NAM and TSA treatment (Fig 1A), suggesting that acetylation negatively regulates PGK1 activity. Notably, the effect of TSA on increasing PGK1 acetylation and decreasing PGK1 activity was more potent than that of NAM (Figs 1A, S1A and S1B), implying that a HDAC I and/or II is the major deacetylase for PGK1.
Given that PGK1 is a highly conserved protein [28], we speculated that important regulatory sites in PGK1 targeted by acetylation might also be conserved. Sequence alignments from diverse species revealed that of twenty putative acetylated lysines identified by the different mass spectrometric analyses, fifteen lysine residues (K11, K30, K48, K86, K91, K97, K131, K146, K156, K192, K199, K264, K267, K291, K323) are not conserved, while five (K75, K139, K141, K220, K406) are invariably conserved (S2 Fig). To determine which lysine residue(s) plays a major role in the regulation of PGK1, we mutated each of the five conserved putative acetylated lysine residues in PGK1 to arginine (R) or glutamine (Q) and assayed their enzyme activity individually. The K to R mutation is often used as a deacetylation mimetic, whereas the K to Q mutation may act as a surrogate of acetylation [29]. We found that substitution at K220, but not the other four lysine residues (K75, K139, K141, and K406), by arginine substantially reduced PGK1 enzyme activity by 82% (Fig 1B), indicating K220 has an important role in controlling PGK1 activity. Moreover, both the K220R and K220Q mutants exhibited a negligible response in changing acetylation level and enzyme activity upon NAM and TSA treatment (Figs 1C and S3), re-affirming that K220 is a major acetylation site in PGK1.
PGK1 exists as a monomer comprising two nearly equal-sized N- and C-terminal domains. This extended two-domain structure is associated with large-scale “hinge-bending” conformational changes, bringing the two substrates into close proximity, with 3-PG or 1,3-BPG binding to the N-terminal domain and the nucleotide substrate ADP binding to the C-terminal domain of the enzyme [30,31]. PGK1 in complex with 3-PG, ADP, and tetrafluoroaluminate (ALF) results in a fully active and closed conformation [14]. Molecular modeling predicts that K220 acetylation disturbs PGK1’s binding with ADP (Fig 1D), suggesting that K220 acetylation may inhibit PGK1 enzyme activity by blocking the substrate ADP binding.
To test this hypothesis, we employed an expression system genetically encoding Nε-acetyllysine to prepare recombinant PGK1 protein that was completely acetylated on K220 in Escherichia coli [32,33]. Briefly, the K220 codon in PGK1 was mutated to an amber stop codon. An amber suppressor tRNA and a tRNA synthetase that could conjugate the acetyllysine to the amber tRNA suppressor were also expressed in the bacteria. Therefore, in the presence of acetyllysine added in the culture medium, the amber stop codon at the position of K220 was replaced by an acetyllysine when PGK1 was expressed in the genetically engineered E. coli strain. This expression system produced PGK1 proteins with 100% acetylation at the lysine residue 220. Moreover, we generated and verified an antibody that specifically recognizes the K220 acetylated PGK1 [α-acPGK1(K220)] (S4A, S4B and S4C Fig). The K220 acetylation of the recombinant PGK1 was confirmed by immunoblotting with the site-specific α-acPGK1(K220) antibody (Fig 1E). Consistent with the structural prediction, the recombinant PGK1K220ac protein was catalytically inactive when compared to recombinant wild-type PGK1. Importantly, isothermal titration calorimetry (ITC) analysis demonstrated that the recombinant PGK1K220ac protein was defective in binding with ADP (Fig 1F). These results provide direct and unequivocal biochemical evidence to support a model that K220 acetylation in PGK1 inhibits its enzymatic activity by blocking the substrate ADP binding.
The acetylation state of a given protein is controlled by the action of lysine acetyltransferases (KATs) and deacetylases (KDACs), enzymes that catalyze the addition and removal, respectively, of an acetyl group from a lysine residue. To search for potential KAT(s) which are responsible for PGK1 K220 acetylation, we generated a siRNA library with three siRNAs targeting each of the 19 human KAT genes [34]. The knockdown efficiency of each siRNA against corresponding KAT genes was determined by quantitative RT-PCR (S5 Fig and S1 Table). We found that knockdown of ATF2 (Gene ID: 1386), KAT5 (Gene ID: 10524), KAT6A (Gene ID: 7994), KAT9 (Gene ID: 55140), KAT12 (Gene ID: 9329), or KAT13B (Gene ID: 8202) led to an increase in cellular PGK1 enzyme activity (Fig 2A), indicating that these KAT enzymes may play a direct or indirect role in the regulation of PGK1 activity. Among these six candidate KAT genes, knockdown of KAT9 (also known as ELP3), but not the other five KATs, significantly decreased the K220 acetylation level of endogenous PGK1 without changing its protein expression in HEK293T cells (Fig 2B and 2C). In a converse experiment, co-overexpression of HA-KAT9 with Flag-PGK1 increased the PGK1 K220 acetylation level by 2.2-fold and decreased PGK1 activity by 36% (Fig 2D). Collectively, these results indicate that KAT9 is a potential acetyltransferase of PGK1.
Our earlier observation that TSA is more potent than NAM to increase PGK1 acetylation and inhibit PGK1 activity (Fig 1A) led us to search for the HDAC enzyme(s) that mediates PGK1 deacetylation. We found by binding assay that PGK1 interacted with HDAC3 (NP_003874.2), but not the other six HDACs, when co-expressed in HEK293T cells (Fig 3A). The endogenous protein interaction between HDAC3 and PGK1 was readily detected in HEK293T cells (Fig 3B). Furthermore, based on the protein amount and the immunoprecipitation efficiency, we found that ~16% of endogenous HDAC3 interacted with endogenous PGK1 (Fig 3B). Co-overexpression of HA-HDAC3 with Flag-PGK1 decreased the acetylation level of PGK1 by 75%, and increased PGK1 activity by ~1.6-fold (Fig 3C). In contrast, co-overexpression of HA-HDAC3 with the K220R/Q mutants of PGK1 did not change PGK1 acetylation or enzyme activity (Fig 3C). When PGK1 was co-overexpressed with a catalytic inactive mutant HDAC3Y298H [35], neither PGK1 acetylation nor enzyme activity was altered (Fig 3D). Conversely, depletion of HDAC3 increased the K220 acetylation level of endogenous PGK1 by >2-fold and decreased PGK1 activity by >50% in HEK293T cells (Fig 3E). More importantly, we performed in vitro pull-down assay using purified recombinant proteins of His-PGK1 and GST-HDAC3, and found that HDAC3 directly binds with PGK1 in vitro (Fig 3F). Furthermore, in vitro deacetylation assay using Flag-tagged PGK1 and GST-HDAC3 confirmed that PGK1 is a direct substrate of HDAC3 and that HDAC3-mediated K220 deacetylation increases PGK1 activity (Fig 3G).
Previous studies have shown that PGK1 gene expression is regulated by nutrient availability in yeast [7–9], indicating that PGK1 may respond to energy status to maintain cellular energy homeostasis. We found that PGK1 activity was dose-dependently stimulated by insulin, an important metabolism and energy regulator, and this activation was associated with a concomitant reduction of PGK1 K220 acetylation (Fig 4A). The effect of insulin on PGK1 K220 acetylation and activity was suppressed by TSA treatment (Fig 4A). Moreover, the K220R mutant PGK1 displayed a negligible change in K220 acetylation and enzyme activity upon insulin stimulation (Fig 4B), indicating that K220 is a vital site for PGK1 deacetylation and enzymatic activation by insulin.
Next, we determined the K220 acetylation level of endogenous PGK1 in HEK293T cells. By using the recombinant PGK1K220ac protein purified from E. coli as the standard, we found that ~37% of endogenous PGK1 was acetylated at K220, and that insulin treatment decreased PGK1 acetylation to 11% (Fig 4C). The fact that a significant fraction of endogenous PGK1 undergoes deacetylation upon insulin treatment suggests that K220 acetylation plays a physiologically relevant role in PGK1 regulation.
Furthermore, we assessed PGK1 K220 acetylation in mouse tissues after insulin injection. As expected, intraperitoneal injection of insulin (5 U/kg) resulted in a transient drop in blood glucose levels (Fig 4D). Interestingly, the K220 acetylation level of endogenous Pgk1 in mouse livers was significantly (p < 0.01) decreased and bottomed at 60 min after insulin injection, followed by a period of recovery in parallel with the blood glucose levels (Fig 4D and 4E). Similarly, Pgk1 K220 acetylation was dynamically changed in mouse kidneys after insulin injection (S6 Fig). Together, these findings suggest that K220 acetylation plays a signaling role in regulating PGK1 function both in cultured cells and in mouse tissues upon insulin treatment.
We next set out to investigate how insulin signal regulates PGK1 K220 acetylation and activity. We found that insulin signal did not change KAT9 gene expression in HEK293T cells (S7A Fig). The PGK1-KAT9 protein association was readily detected in cells co-overexpressing Flag-PGK1 and HA-KAT9, leading to an increase in PGK1 K220 acetylation (S7B Fig). However, neither the KAT9-PGK1 protein interaction nor PGK1 K220 acetylation was affected by insulin treatment in cells co-overexpressing PGK1 and KAT9 (S7B Fig). These findings suggest that KAT9 may not contribute to insulin-regulated PGK1 acetylation.
On the other hand, we observed that the effect of insulin on reducing the K220 acetylation level of endogenous PGK1 was diminished in HEK293T cells when HDAC3 was depleted by siRNA (Fig 5A). Insulin treatment did not change HDAC3 gene expression in HEK293T cells (S8 Fig), but could enhance endogenous PGK1-HDAC3 association and decrease PGK1 K220 acetylation in a dose-dependent manner (Fig 5B). In agreement, insulin injection reduced Pgk1 K220 acetylation and increased Pgk1-Hdac3 interaction in mouse livers (Fig 5C). Together, these results suggest that insulin promotes PGK1 K220 deacetylation likely through enhancing PGK1-HDAC3 association.
Recent studies have identified HDAC3 as a phosphorylated protein, with Ser424 being one of the key phosphorylation sites [36,37]. Protein kinase CK2 has been reported to be responsible for HDAC3 S424 phosphorylation [38]. Mutation of serine (S) 424 to alanine (A) (mimics dephosphorylation) inhibits HDAC3 enzyme activity without affecting its expression or subcellular localization [38]. In accord, we found that the HDAC3S424A mutant displayed a ~50% reduction in enzyme activity (S9 Fig). Strikingly, we observed that HDAC3S424A mutant exhibited impaired association with endogenous PGK1 in HEK293T cells (S9 Fig), suggesting that Ser424 phosphorylation is critical for not only the deacetylase activity of HDAC3 but also its interaction with PGK1. To provide further evidence to support this notion, we transfected HEK293T cells singularly with plasmid expressing Flag-PGK1 or with plasmids co-expressing Flag-PGK1 and HA-HDAC3. Lysates from transfected cells were first immunoprecipitated with Flag antibody and then the remaining supernatant was precipitated with HA antibody. We found that HA-HDAC3 was co-precipitated with Flag-PGK1 (Fig 5D). Importantly, HA-HDAC3 in the PGK1 immune complex was highly phosphorylated on Ser424, whereas the free HDAC3 was weakly phosphorylated. These results suggest that Ser424 phosphorylation of HDAC3 may enhance its interaction with PGK1.
Previous studies have shown that HDAC3 phosphorylation is regulated by phosphoinositide-3-kinase/AKT (PI3K/AKT) pathway [39]. In addition, HDAC3 has also been identified as a downstream target of mTOR [36,37]. This led us to hypothesize that the PI3K/AKT/mTOR signaling pathway may regulate HDAC3 S424 phosphorylation and PGK1 K220 acetylation in response to insulin. To test this hypothesis, we treated HEK293T cells co-overexpressing Flag-PGK1 and HA-HDAC3 with either LY294002 or Wortmannin, two specific PI3K inhibitors. As expected, either PI3K inhibitor profoundly attenuated AKT S473 phosphorylation (S10A and S10B Fig). Notably, both LY294002 and Wortmannin dose-dependently increased K220 acetylation of ectopically expressed Flag-PGK1 and decreased the interaction between ectopically expressed PGK1 and HDAC3 (S10A and S10B Fig). In accord, both LY294002 and Wortmannin dose-dependently increased K220 acetylation of endogenous PGK1 and decreased endogenous PGK1-HDAC3 association in HEK293T cells (Fig 6A and 6B). In addition, MK-2206 2HCI, a specific AKT inhibitor, produced similar effects on changing PGK1 K220 acetylation and PGK1-HDAC3 association as the PI3K inhibitor LY294002 or Wortmannin (Fig 6C). Furthermore, we observed that the mTOR inhibitor Rapamycin profoundly attenuated S6K T389 phosphorylation as expected, reduced HDAC3 Ser424 phosphorylation, impaired endogenous PGK1-HDAC3 interaction, and increased PGK1 K220 acetylation (Fig 6D). Collectively, these findings demonstrate that the PI3K/AKT/mTOR pathway can regulate PGK1 K220 acetylation, possibly through controlling HDAC3 S424 phosphorylation and HDAC3–PGK1 protein interaction.
The PI3K/AKT/mTOR pathway is regulated by a wide variety of cellular signals, including insulin [40]. We then studied the role of the PI3K/AKT/mTOR pathway in mediating insulin signal to control PGK1 K220 acetylation. As shown, insulin increased HDAC3 Ser424 phosphorylation and enhanced the interaction between ectopically expressed proteins of HDAC3 and PGK1 (Fig 6E). The dephosphorylation-mimicking HDAC3S424A mutant exhibited weak binding to PGK1 even upon insulin treatment (Fig 6E), supporting the notion that HDAC3 S424 is a key site for regulating HDAC3-PGK1 association upon insulin stimulation. Rapamycin treatment decreased HDAC3 Ser424 phosphorylation, impaired protein interaction between endogenous HDAC3 and PGK1, and increased PGK1 K220 acetylation (Fig 6F). Moreover, rapamycin abolished the effect of insulin on changing HDAC3 Ser424 phosphorylation, HDAC3-PGK1 association, and PGK1 K220 acetylation (Fig 6F). Besides insulin, other growth factors, such as EGF, may also enhance the PI3K/mTOR pathway [41]. We found that EGF stimulation increased ERK1/2 phosphorylation, but not S6K T389 phosphorylation, in HEK293T cells, suggesting that EGF cannot potently activate mTOR in our experiment. As a result, EGF treatment did not change HDAC3 S424 phosphorylation and PGK1 K220 acetylation (S11 Fig). Collectively, these data suggest that the mTOR pathway regulates PGK1 K220 acetylation through controlling HDAC3 S424 phosphorylation.
To address the physiological significance of PGK1 regulation by K220 acetylation, we generated stable HEK293T cells in which endogenous PGK1 was depleted, and wild-type or K220 mutant PGK1 was re-introduced at a comparable level (knockdown and put-back, S12 Fig). We found that ATP production did not differ between wild-type and PGK1K220Q mutant put-back cells (Fig 7A). We then treated these stable cells with rotenone, a chemical which inhibits the complex I of the respiratory chain and thus mitochondrial ATP production [42,43]. A short-term treatment with rotenone greatly decreased cellular ATP production without inducing substantial cell death (S13 Fig). Notably, PGK1K220Q mutant put-back cells displayed a remarkable reduction (by 76%; p < 0.01) in glycolytic ATP production compared to wild-type rescued cells (Fig 7A). Liquid chromatography-mass spectrometry (LC-MS) analysis demonstrated that the level of 3-PG, which is the product of PGK1 catalysis, was dramatically reduced, by 89% (p < 0.05), in PGK1K220Q mutant put-back cells compared to wild-type rescued cells (Fig 7B). 3-PG can be oxidized to 3-phosphohydroxypyruvate for de novo (that is originated from glucose) serine biosynthesis [44]. In accord with reduced 3-PG, the serine level was significantly decreased (by 64%; p < 0.01) in PGK1K220Q mutant put-back cells (Fig 7B). Moreover, extracellular acidification rate (ECAR) analysis revealed that several parameters reflecting the glycolytic function, such as glycolysis, glycolytic capacity, and glycolytic reserve, were lower in PGK1K220 mutant put-back cells than wild-type rescued cells (Figs 7C and S14). Glucose consumption was significantly decreased (~60% less; p < 0.01), while glycogen storage was significantly increased (by ~2.2-fold; p < 0.01) in the PGK1K220Q mutant put-back cells when compared to wild-type rescued cells (S15A and S15B Fig). These results suggest that K220 acetylation plays an important role in regulating PGK1 activity and/or function to modulate glycolytic ATP production and glucose metabolism.
We also observed that stable knockdown of PGK1 led to a ~65% reduction in the NADPH/NADP+ ratio in HEK293T cells (Fig 7D), suggesting that PGK1 is an important contributor to NADPH pools besides its well-known role in glycolytic ATP production. Moreover, higher ROS production was detected in PGK1 knockdown cells subjected to menadione, a quinone compound that induces the production of superoxide radicals (Fig 7E). ROS has been extensively implicated in signaling cascades, which function as important cell survival mechanisms in response to oxidative stress [45]. We found that PGK1 knockdown cells exhibited higher levels of cleaved PARP, an indicator of apoptosis (S16A Fig), as well as higher levels of p38 MAPK phosphorylation, a stress-responsive kinase (S16B Fig). Importantly, re-expression of wild-type PGK1, but not the acetylated mimetic K220Q mutant, restored the NADPH/NADP+ ratio and suppressed ROS production in PGK1 knockdown cells subjected to menadione (Fig 7D and 7E). Putting-back wild-type PGK1, but not the K220Q mutant, reduced the levels of cleaved PARP and p38 MAPK phosphorylation in PGK1 knockdown cells when subjected to menadione (S16A and S16B Fig). As a result, PGK1 knockdown cells exhibited a higher incidence of cell death in response to menadione, and re-expression of wild-type PGK1, but not the K220Q mutant, could rescue cells from menadione-induced cell death (Figs 7F and S17). These results suggest that K220 acetylation plays an important role in regulating PGK1 activity/function to modulate NADPH redox and cellular oxidative response.
To further illustrate the role of K220 acetylation in controlling glucose metabolism and redox potential in the cell, we generated PGK1 knockdown and put-back stable HEK293T cells, in which endogenous PGK1 was depleted and the K220Q mutant was re-introduced at a higher protein level in order to reach an equivalent PGK1 enzyme activity between the cells re-expressing wild-type and K220Q mutant PGK1 (Fig 7G). By monitoring the reduction of NADH, we found that the activity of wild-type PGK1 was significantly (p < 0.05) stimulated by insulin treatment (Fig 7G), leading to a remarkable increase of 3-PG (by 3.2-fold; p < 0.01) (Fig 7H). In contrast, cells with PGK1K220Q mutant put-back cells displayed a negligible change in PGK1 enzyme activity and 3-PG upon insulin stimulation (Fig 7G and 7H). These findings strongly support the notion that K220 acetylation plays an important role in regulating PGK1 activity/function in cells upon physiological stimulus, such as insulin.
The current study uncovers a biochemical mechanism for how acetylation controls the activity/function of PGK1. We have identified KAT9 and HDAC3 as the potential acetyltransferase and major deacetylase of PGK1, respectively. To the best of our knowledge, we show for the first time that acetylation of PGK1 plays an important role in modulating glycolytic energy production and cellular oxidative response.
We have identified K220 as an important regulatory acetylation site within the PGK1 protein. Human PGK1 exists as a monomer containing two nearly equal-sized domains corresponding to the N- and C-termini of the protein [30]. 1, 3-BPG binds to the N-terminal domain whereas ADP binds to the C-terminal domain of PGK1 [46]. Though the binding of either substrate 1,3-BPG or ADP triggers a conformational change, only through the binding of both substrates does domain closure occur, bringing the two substrates into the proper vicinity for phosphotransfer [47,48]. Of note, K220 locates in the C-terminus of PGK1 and interacts with the nucleotide substrate ADP [48,49]. Our data show that acetylation mimetic K220Q substitution significantly reduces PGK1 catalysis. This is best illustrated by the observation that the recombinant PGK1K220ac protein is defective in ADP binding, implying that neutralization of the positive charge of K220 by acetylation may disrupt ADP binding and thus inhibit PGK1 catalysis. It has to be noted that both acetylation mimetic K220Q substitution and deacetylation mimetic K220R substitution can significantly reduce PGK1 activity. This raises another possibility, that is, the steric hindrance that occurs when K220 is mutated to Q or R, and this will generate different interaction force and/or steric hindrance, thereby abolishing ADP binding and inhibiting PGK1 catalysis.
Our study has provided novel insights into the role of K220 acetylation in controlling PGK1 activity/function in response to insulin. To the best of our knowledge, this is the first evidence linking the regulation of PGK1 to mTOR downstream of PI3K/AKT and insulin. In addition, we show a molecular crosstalk between mTOR-mediated HDAC3 S424 phosphorylation and PGK1 K220 acetylation.
We have identified HDAC3 as the major deacetylase of PGK1. HDAC3 can form multi-protein complexes with the co-repressors SMRT and N-CoR and deacetylates histones, thereby regulating the transcription of a plethora of genes [50,51]. In addition, many non-histone substrates of HDAC3 have been identified, including the NF-ƘB protein RelA [52], sex-determining region Y (SRY, a master regulator of testis organogenesis) [53], and several transcription factors such as p53 [54], myocyte enhancer factor-2 (Mef2) [55], and glial cell missing (GCMa) [56]. Very recently, it was reported that HDAC3 deacetylates methionine adenosyltransferase IIα (MAT IIα) in the methionine cycle [57], indicating that HDAC3 can deacetylate a metabolic enzyme and may play an important role in regulating metabolic pathway(s). Supporting this notion, we show in this study that another metabolic enzyme, PGK1, is a direct substrate of HDAC3 and that HDAC3-mediated PGK1 deacetylation plays a signaling role in regulating PGK1 activity/function upon insulin stimulation. Moreover, we demonstrate that the PI3K/AKT/mTOR pathway regulates PGK1 K220 acetylation, in part, via affecting HDAC3 Ser424 phosphorylation, suggesting a potential mechanism for HDAC3 Ser424 phosphorylation regulating PGK1 K220 acetylation. Upon insulin stimulation, the PI3K/AKT/mTOR pathway induces HDAC3 Ser424 phosphorylation, which increases the deacetylase activity of HDAC3 and/or enhances the protein association between HDAC3 and PGK1, leading to PGK1 K220 deacetylation and enzyme activation (Fig 7I). mTOR is a central cell growth controller and is potently activated by insulin [58]. In addition, mTOR is also regulated by nutrient availability and cellular energy status to control cellular metabolism. Our data provides a direct link of mTOR activation to glycolysis, which is achieved by mTOR-mediated HDAC3 phosphorylation and PGK1 deacetylation, thus leading to PGK1 activation. These results also provide an example illustrating how cells integrate different pathways such as extracellular growth signaling and intracellular metabolic flux by a crosstalk involving different type of protein modifications such as phosphorylation and acetylation.
In addition, we have also identified KAT9 as a potential acetyltransferase of PGK1. KAT9/ELP3, which encodes the catalytic subunit of the histone acetyltransferase elongator complex, has previously been identified as an α-tubulin acetyltransferase in mouse neurons [59]. Previously, we reported that KAT9 is the potential acetyltransferase of glucose-6-phosphate dehydrogenase (G6PD), which is a key enzyme in the pentose phosphate pathway and plays an essential role in the oxidative stress response by producing NADPH [60]. In this study, we show that KAT9 as the potential acetyltransferase of PGK1 interacts with and increases PGK1 acetylation, thereby inhibiting PGK1 activity. However, KAT9 may not contribute to cells sensing insulin signal to regulate PGK1 acetylation and function, as neither KAT9 expression nor KAT9-PGK1 interaction is changed upon insulin treatment.
Our study also links the regulation of PGK1 activity by acetylation to cellular response to oxidative stress. We show that replacement of endogenous of PGK1 with an acetylation-mimetic K220Q mutant results in a significant decrease in NADPH production and higher susceptibility of cells to oxidative stress. The underlying mechanism for PGK1 controlling NADPH production remains unclear. We propose that a general inhibition of glycolysis may at least in part explain the reduction in NADPH production. In addition, we also show that both 3-PG and serine levels are significantly reduced in cells with PGK1 knockdown and put-back of the acetylation mimetic K220Q mutant as compared to wild-type rescued cells (Fig 7B). 3-PG is essential for de novo serine biosynthesis [44]. Serine can be converted to glycine by serine hydroxymethyl transferase, a reaction that yields one carbon units, which enter the tetrahydrofolate cycle and are critical for NADPH production [61,62]. Whether PGK1 K220 acetylation inhibits PGK1 catalytic activity, leading to reduced production of 3-PG, disturbed serine biosynthetic flux, and subsequently reduced NADPH production, still needs further investigation.
Recent work has suggested a critical role of serine metabolism in cancer pathogenesis [63–65]. Supporting this notion, PHGDH gene expression is up-regulated in diverse cancer cells, such as esophageal adenocarcinoma, triple-negative breast cancer, and melanoma [66]. Moreover, cancer cells with PHGDH amplification were found to exhibit increased metabolic flux into serine biosynthesis, which is known to play a dual role in redox balance and providing nucleotide units to support cancer cell proliferation [5,67]. As noted above, K220 acetylation inhibits PGK1 activity/function and reduces the 3-PG and serine levels, implying that modulation of PGK1 activity/function by acetylation may serve as a promising anti-cancer strategy through regulating serine metabolism.
Moreover, the clinical implication of PGK1 dysfunction has been highlighted by chronic haemolysis with progressive neurological impairment in PGK1-deficient patients [68–70]. Two metabolic alterations, a decreased steady-state level of ATP and an increased 2,3-BPG, in red blood cells with human PGK1 deficiency have been proposed to cause hemolytic anemia [71]. It is known that hemolytic anemia also contributes to the inability of erythrocyte cells to produce NADPH and withstand harmful oxidants [72]. Likely, PGK1 dysfunction-associated chronic haemolysis and neurological impairment are in part caused by oxidative stress due to increased levels of oxidative damage and decrease levels of antioxidants, such as reductant NADPH. In this study, we have provided evidence showing that HDAC3-dependent acetylation regulates the function of PGK1 in both glycolytic ATP production and NADPH redox. Therefore, future therapeutic intervention(s) to modulate PGK1 activity via HDAC3-mediated deacetylation may serve as a potential target for treating related diseases, such as hemolytic anemia and brain disorders associated with PGK1 dysregulation.
All the animal experiments were carried out in accordance with the National Institutes of Health guidelines for the Care and Use of Laboratory Animals and the regulations of Fudan University for animal experimentation. MaleBALB/c mice (6–8 wk old, 20–25 g body weight) were purchased from the Fudan Animal Center. Animals were given unrestricted access to a standard diet and tap water. Blood was collected from tail vein at different time points post intraperitoneal injection of insulin (5 U/kg body weight), and blood glucose levels were determined by using a glucose detection kit (Roche). Before being humanely killed, mice were anesthetized with sodium pentobarbital (25 mg pentobarbital/kg body weight, ip). After that, mouse livers and kidneys were harvested and then homogenized using the Tissuelyser-24 (Shanghai JingXin) in 0.5% NP-40 buffer containing protease inhibitor cocktail, and lysed on ice for 30 min. To determine the acetylation level of endogenous Pgk1, tissue lysates were incubated with the Pgk1 antibody (Santa Cruz) for 1 hr, followed by incubating with Protein-A beads (Upstate) for another 2 hr at 4°C.
The Rabbit anti-pan-acetyllysine antibody was generated as previously described [73]. To generate acetyl-lysine 220 specific antibody of PGK1, synthetic peptide VADKIQLINNMLDK was coupled to KLH as antigen to immunize rabbit (Shanghai Genomic Inc). For more detail information about the other antibodies used in this study, please refer to the S1 Text.
Flag-tagged PGK1 protein was overexpressed in HEK293T cells, immumoprecipitated and eluted by 250 mg/ml Flag peptides (Gilson Biochemical) dissolved in PBS (pH 7.5). The PGK1 activity assay was carried out as previously described [74]. The reactions were started by adding enzyme into the buffer containing 80 mM pH 7.6 triethanolamine, 8.0 mM MgCl2, 0.25 mM NADH, 2.4 mM ATP, 12 mM 3-phosphoglycerate and 50 μg/ml glyceraldehyde-3-phosphate dehydrogenase in a total volume of 0.3 ml, and assayed at 25°C. By monitoring the reduction of NADH fluorescence (Ex350 nm, Em470 nm), the specific PGK1 activity was measured using HITACHI F-4,600 fluorescence spectrophotometer.
The K220 site-specific acetylated PGK1 was expressed in E. coli as previously described [32,33]. In short, the ORF of PGK1 was cloned into pTEV-8 vector with amber codon being incorporated at lysine 220 (AAG to TAG by site-directed mutagenesis). The E. coli strain BL21 (DE3) was transformed with three plasmids, pAcKRS-3, pCDF PylT-1, and pTEV-8-PGK1. Cells were grown overnight in LB containing spectinomycin (50 μg/ml), kanamycin (50 μg/ml), and ampicillin (150 μg/ml) at 37°C till OD600 reached 0.6–0.8. The culture was added with 20 mM NAM and 10 mM N-acetyllysine (Sigma-Aldrich). Protein expression was induced at 37°C by addition of 0.5 mM of isopropyl-1-thio-D-galactopyranoside (IPTG) for 3 hr. Afterward, the cells were harvested, the K220-acetylated PGK1 protein was purified and then stored at −80°C till further analysis.
ITC assay was performed by using a MicroCal VP-ITC type microcalorimeter (MicroCal Inc.). Briefly, temperature equilibration was allowed for 1–2 hr till to 20°C prior to the experiment. PGK1 protein and ADP (Sigma) were dialyzed against 40 mM Tris-HCl (pH 7.4), 50 mM KCl, and 10 mM MgCl2. All solutions were thoroughly degassed before being used by centrifugation at 13,000 rpm for 20 min. The experiment was conducted by consecutively injecting 50 μM ADP solution into the calorimetric cell containing 50 μM purified PGK1. The titration enthalpy data was corrected for the small heat changes in control titrations of ADP solution into the dialysis buffer.
To generate stable PGK1 knockdown cell pools in HEK293T cells, shRNA targeting PGK1 was constructed, and retrovirus was produced using a two-plasmid packaging system as previously described [75]. The shRNA targeting sequence for PGK1 is 5′-GCTTCTGGGAACAAGGTTAAA-3′. The pMKO.1-puro shRNA construct was co-transfected with vectors expressing the gag and vsvg genes into HEK293T cells. Retroviral supernatant was harvested 36 hr after transfection, and mixed with 8 μg/mL polybrene to increase the infection efficiency. HEK293T cells were infected with the retrovirus and selected in 1 μg/ml puromycin for 1 wk.
To generate PGK1 knockdown and put-back stable cell pools in HEK293T cells, two silent nucleotide substitutions were introduced into Flag-tagged human wild-type or K220 mutant PGK1 in the sequence corresponding to the shRNA targeted region. Both shRNA resistant PGK1 were cloned into the retroviral pQCXIH-hygro vector and co-transfected with vectors expressing the gag and vsvg genes in HEK293T cells to produce retrovirus. Retroviral supernatant was harvested 36 hr after transfection, and mixed with 8 μg/mL polybrene to increase the infection efficiency. HEK293T cells with PGK1 knockdown were infected with the retrovirus and selected in 1 μg/ml puromycin and 2 μg/ml hygromycin for 4 wk.
The intracellular NADPH/NADP+ was measured by enzymatic cycling methods as previously described [76,77]. Briefly, cells were counted and seeded in 10 cm dishes at a density of 1.5 × 106. After 24 hr, cells were lysed in 400 μl of extraction buffer (20 mM NAM, 20 mM NaHCO3, 100 mM Na2CO3) and centrifuged at 1,200 g for 15 min. 150 μl of the supernatant was incubated in a heating block for 30 min at 60°C for NADPH extraction. And then, 20 μl of the cell extract with 160 μl of NADP-cycling buffer (100 mM Tris-HCl, pH8.0; 0.5 mM thiazolylblue; 2 mM phenazine ethosulfate; 5 mM EDTA) containing 1.3 U of G6PD was added to a 96-well plate. After incubation for 1 min at 30°C in darkness, 20 μl of 10 mM G6P was added to well, and measured the change of absorbance at 570 nm every 30 s for 10 min at 30°C by using SpectraMax M5 Microplate Reader (Molecular Devices). Subtracting NADPH (heated sample) from the total of NADP+ and NADPH (unheated sample) was the level of NADP+.
Statistical analyses were performed with a two-tailed unpaired Student's t test. Almost all data shown represent the results obtained from triplicated independent experiments with standard deviation of the mean (mean ± S.D.). The values of p < 0.05 were considered statistically significant. The numerical data and statistical analysis used in all figures are included in S1 Data.
|
10.1371/journal.pgen.1003396 | Robust Prediction of Expression Differences among Human Individuals Using Only Genotype Information | Many genetic variants that are significantly correlated to gene expression changes across human individuals have been identified, but the ability of these variants to predict expression of unseen individuals has rarely been evaluated. Here, we devise an algorithm that, given training expression and genotype data for a set of individuals, predicts the expression of genes of unseen test individuals given only their genotype in the local genomic vicinity of the predicted gene. Notably, the resulting predictions are remarkably robust in that they agree well between the training and test sets, even when the training and test sets consist of individuals from distinct populations. Thus, although the overall number of genes that can be predicted is relatively small, as expected from our choice to ignore effects such as environmental factors and trans sequence variation, the robust nature of the predictions means that the identity and quantitative degree to which genes can be predicted is known in advance. We also present an extension that incorporates heterogeneous types of genomic annotations to differentially weigh the importance of the various genetic variants, and we show that assigning higher weights to variants with particular annotations such as proximity to genes and high regional G/C content can further improve the predictions. Finally, genes that are successfully predicted have, on average, higher expression and more variability across individuals, providing insight into the characteristics of the types of genes that can be predicted from their cis genetic variation.
| Variation in gene expression across different individuals has been found to play a role in susceptibility to different diseases. In addition, many genetic variants that are linked to changes in expression have been found to date. However, their joint ability to accurately predict these changes is not well understood and has rarely been evaluated. Here, we devise a method that uses multiple genetic variants to explain the variation in expression of genes across individuals. One important aspect of our method is its robustness, in that our predictions agree well between training and test sets. Thus, although the number of genes that could be explained is relatively small, the identity and quantitative degree to which genes can be predicted is known in advance. We also present an extension to our method that integrates different genomic annotations such as location of the genetic variant or its context to differentially weigh the genetic variants in our model and improve predictions. Finally, genes that are successfully predicted have, on average, higher expression and more variability across individuals, providing insight into the characteristics of the types of genes that can be predicted by our method.
| Variation in gene expression among human individuals plays a key role in the susceptibility to different diseases and phenotypes [1]–[7]. Some of this variation has been linked to genetic variation that exists among individuals [8]–[16]. For this reason, a major goal is to predict gene expression changes among individuals based on their genetic sequence variation. Ideally, if we could predict the expression profile of genes for each individual, we could estimate this individual's risk for developing certain diseases; but unfortunately, our current ability to make such predictions is poor.
A popular approach to study the relationship between genetic variation and expression variation is by using a mapping approach, where the expression of every gene is treated as a quantitative trait and one searches for genetic changes that are significantly correlated with the expression changes of each gene. A dominant method is to fit a linear regression model from the minor allele count of single nucleotide polymorphisms (SNPs) to the expression level of a gene across different individuals. Under some assumptions, a p-value for the resulting correlations can be computed and after correcting for multiple hypotheses testing, the set of SNPs that are significantly associated with the expression changes of every gene can be identified. Using this approach researchers were able to examine all of the SNPs measured in the human genome for significant associations in an unbiased way, resulting in many potential effectors of gene expression [9], [10], [17], [18].
A major issue in the above analyses is low statistical power for detecting associations that arises from the large number of SNPs examined. This means that SNPs with a low correlation to expression will not be detected as possible effectors. However, since the degree of correlation of SNPs to expression is associated with their genomic features such as location (e.g., distance from the transcription start site) or function [9]–[16], [18]–[20] (e.g., within transcription factor binding sites), several studies used genomic features of SNPs as a prior on the probability of their functional importance within a Bayesian framework [13], [19], [20], resulting in additional statistical power to detect associated SNPs in particular genomic regions.
A second issue is that the above methods examine each SNP in isolation. Thus, they cannot detect combinatorial associations between multiple SNPs and expression changes. Several methods tried to address this issue by examining multiple SNPs at the same time using ridge-regression [21], a Bayesian approach [22], or multiple interval mapping [23], and identified statistically significant epistatic interactions that are not found using the single SNP approaches.
Collectively, the above methods focused on mapping SNPs that are significantly correlated with expression and have identified many interesting associations between expression changes across genes and genotype data. However, since their goal was to identify statistically significant associations of single SNPs, they do not consider multiple-SNP models, where combinations of SNPs and SNPs with small effects could play a role in generating the predictive model. One work attempted to devise a multiple-SNP predictive model using a cross validation scheme, whereby a model whose parameters were fitted from the data of a subset of the individuals is tested for its ability to predict the expression of the remaining unseen individuals [24]. This work also integrated genomic features of SNPs, using them as priors in a regularized linear model, and found SNPs located in certain regions of the gene to be more predictive of gene expression variation. However, in addition to the SNP information, this work also used the expression value of other genes for the prediction task, making it difficult to assess which part of the reported predictive power comes from SNP information and which from the expression of correlated genes.
Here, we set out to develop an algorithm that predicts changes in gene expression among human individuals using only their genetic variation. Notably, this goal is different from that of the above studies that focus on mapping associated single SNPs. Accordingly, rather than providing p-values for SNP associations, we evaluate the quality of our algorithm by its ability to predict the gene expression levels of unseen human individuals given only their genotype information. Contrary to the single SNP association studies mentioned above, we use multiple SNPs to predict the expression of every gene, but to avoid overfitting only SNPs that are in the genomic vicinity of the predicted gene are used. We also incorporate a comprehensive set of genomic and functional features for each SNP, by allowing the algorithm to use this meta-information to weigh the importance of each SNP in the prediction.
We evaluated the ability of several algorithms to predict gene expression data from immune precursor cells of four different human populations for which corresponding SNP data is available [8], [10]. Clearly, this prediction task is very difficult for several reasons, including noise in the data and missing SNPs, and because the expression of many genes is determined by environmental factors and by trans-acting SNPs that we do not consider. Nevertheless, comparing two different multi-SNP algorithms and one single-SNP algorithm, we find that both a variation of the K-nearest neighbor (KNN) algorithm [25], [26] and a regularized linear model achieve high correspondence between their predictions on the training data and those on the held out test data. We also show that combining their predictions into a single model improves the predictions on held-out test data, as the two models explain different aspects of the expression data. As expected, the total number of genes whose expression can be accurately predicted using their proximal SNP data is relatively small, but the high robustness of our algorithm means that it can determine the quality of the predictions in advance with high accuracy.
As we had four different populations in the data, we evaluated the predictions on several different cross validation schemes. Notably, we found that incorporating individuals from other populations improves the predictions of a given population compared to only using the individuals of the predicted population, suggesting that some of the SNPs that are associated with expression changes are shared across populations. We also found that our algorithm achieves similar performance when it attempts to predict the expression of genes from one population using only the data of other populations.
Taken together, our work provides concrete means by which some of the expression variation among human individuals can be predicted with a low false positive rate using only their genotype data, suggesting that the time may be ripe for a broader examination of this important prediction task, perhaps while incorporating additional features such as trans-acting SNPs.
As the data for our study, we used the HapMap Phase II dataset [8], consisting of 210 unrelated individuals from four distinct populations of European (CEU), African (YRI), Chinese (CHB), and Japanese (JPT) origins, for which corresponding expression measurements from immune precursor cells are available for 15,439 genes [10]. We defined a prediction task for each such gene using only its set of proximal cis-SNPs, defined as those SNPs that reside inside the body of the gene or in the 100 kb upstream and downstream regions flanking the gene. This resulted in a total of ∼4.7 M cis-SNPs across all 15,439 genes, for an average of ∼304 SNPs per gene.
To assess whether the predictive power comes from using individuals of the same population or those of other populations, we used three different cross-validation (CV) schemes throughout the paper (Figure 1). In the first Cross-Pop scheme, three populations serve as training data, and the remaining population is used as the test data. In the second Mixed-Pop scheme, the populations are mixed such that the training and test data both contain individuals from all four populations. Finally, in the third Intra-Pop scheme, the partition to training and test is done separately for each population, such that there is a separate test data for each of the four populations that is predicted using a model learned only from the remaining individuals of the tested population. The prediction task of the Cross-Pop scheme may be expected to be the hardest, since the parameters are learned using only the data of other distinct populations, which requires the SNPs affecting expression to be shared across populations. Conversely, in the Intra-Pop scheme, the model can select different SNPs for each population, if each population has a different SNP affecting expression. This may improve the predictions but can also lead to overfitting of the training data. Finally, in the Mixed-Pop scheme, although SNPs have to be shared across populations, having representatives from all populations in the training and test sets may improve the robustness of the model.
Most studies of the relationship between genetic and expression variation have focused on identifying single SNPs that are significantly correlated with gene expression changes [8]–[16]. Thus, the question of whether multiple SNPs can be used jointly to improve predictions of gene expression remains largely unexplored, and hence, we sought to devise an algorithm that can robustly predict the expression of unseen test individuals using multiple SNPs. Given that we restricted ourselves to cis-SNPs and that many expression changes are determined by environmental factors or by trans-SNPs, this prediction task is difficult and even the optimal predictor would not be able to explain most of the observed variation in expression. Rather, we strive for a robust predictor that will have good agreement between the training and test data, such that it can determine its performance on held out individuals in advance using only the individuals given to it during the training phase.
A predictive model based on a single SNP (single-SNP model) will have only two parameters for regression, namely the slope and the intercept. Therefore, we decided to compare this model with two different multi-SNP models that vary in the number of parameters they use. The first model is based on the K-nearest neighbor (KNN) algorithm, which predicts the expression level of an unseen individual based on the measured expression values of the K individuals in the training set whose genotypes are most similar to it [25], [26]. In the KNN algorithm, similarity is measured by a distance metric, which we defined in our case as the absolute difference in minor allele counts summed over all cis-SNPs of the tested gene. We chose the KNN algorithm due to its inherent property of generating its predictions using the measured values of those individuals that are closest to it in feature (genotype) space, rather than by attempting to construct a generative model of its predicted values. This property allows the KNN model to use multiple SNPs in the prediction, while using only a single parameter, K (the number of neighbors) and it is thus similar to the single-SNP model in terms of the parameter space it uses. The second multi-SNP model we consider is a regularized (elastic-net) linear regression model [27] that has one weight parameter for each SNP, thereby increasing the number of parameters to the number of SNPs used by the model. In order to increase robustness, we employed an internal training cross-validation procedure to determine the strength of the regularization and the training performance (see Methods). We also combined the two multi-SNP models into a single model by averaging their predictions for each individual, since we hypothesized that the different assumptions and number of parameters employed by the two models will capture different aspects of the data and hence their combination could improve the predictions.
We tested these different models in each of the three different cross validation schemes. For all models, in each cross validation partition we used the training data to learn the parameters for each gene, which maximize the reduction in variance achieved by the corresponding predictor (i.e., regressor or KNN, see Methods). We measure this reduction using R2, defined as the proportion of variance in the data (either training or test) explained by the predictor.
Notably, for all multi-SNP models, we found high correlations between their predictions on the training data and their predictions on the test data in all three cross validation schemes (e.g., Pearson correlations of 0.57–0.88 between training and test results across mean cross-validation values of 15,439 genes for the combined model). In nearly all cases, these correlations between the training and test predictions were higher than those achieved by the single-SNP model. This improved performance of the multi-SNP models was most notable in the hardest Cross-Pop prediction scheme, where the KNN model had 0.83 Pearson correlation, compared to the 0.21 of the single-SNP model (Figure 2A, Figure S1, Figure S2). The improved performance of the multi-SNP models was also evident when binning the genes according to their training performance in each algorithm and then comparing the R2 that each algorithm achieves on the test data (Figure 2B, 2C). For example, for genes with training R2 between 0.2 and 0.3 (i.e., 20–30% of the expression variation of the training set is explained), the KNN, linear and combined models achieve an average test R2 of 0.21, 0.13 and 0.22, respectively, whereas the single-SNP model achieves test R2 of 0.075 (Figure 2C, Cross-Pop).
The R2 values above represent an average over multiple genes, and thus, they may not be indicative of the R2 of individual genes. As one way to examine performance at the single gene level, we defined a robustly predicted gene as one with test R2 equal to or larger than some threshold fraction of its training R2 (only genes with training R2≥0 are considered). Here too, we found the multi-SNP models to have a much higher fraction of robustly predicted genes across all thresholds and cross-validation schemes as compared to the single-SNP model (Figure 2D). Notably, the differences in this analysis are even more striking than those suggested by the average R2 values, where for example, at a robustness threshold of 0.5 in the Cross-Pop scheme (i.e., test R2 of a gene is at least half of its training R2), 31%, 51% and 75% of the genes are robustly predicted by the KNN, linear and combined models, respectively, compared to only 5% for the single-SNP model.
As expected from this hard prediction task, the overall number of genes that can be well predicted is relatively small (Table 1, Tables S1 and S2, and Figure S3). For example, 364/15,439 (2.4%) and 529/15,439 (3.4%) of the genes have training and test R2 above 0.1 in the Cross-pop and Mixed-pop CV schemes, respectively. However, the robust nature of our algorithms means that the number of false positive predictions will be relatively small, i.e., the algorithm knows in advance which genes will be well predicted. This high degree of robustness persists and is most impressive in the Cross-Pop scheme, where the expression levels of an entire population is predicted without using any individual from that population.
When comparing the overlap between top-predicted genes (in test data) for the different models, we find that ∼50% of the top-predicted genes overlap across all models (Table 2, Tables S3 and S4, top-predicted gene lists appear in Tables S5, S6, S7, S8, S9, S10, S11, S12, S13). This result suggests that some genes can be well predicted regardless of the selected model (i.e., even a single-SNP does relatively well), but for other genes the selection of the model type, i.e., linear (Elastic-Net) or non-linear (KNN) can have a large effect on our ability to correctly predict held-out test data.
Although we found the multi-SNP models to have better robustness in terms of agreement between training and test results, when comparing the best model across all predicted genes, we find that on average, ∼33% of the genes are best predicted by the single-SNP model, while the other ∼66% are best predicted by one of the multi-SNP models (Table S14). This result emphasizes the fact that in different scenarios, different models perform best, and the best approach would probably make combined use of all models.
In conclusion, these results demonstrate that multi-SNP models, including both the KNN-model that uses a single parameter and the regularized linear model that uses multiple parameters, are overall more robust than the single-SNP model, and can outperform it for most genes. In addition, we found that a model that combines the two multi-SNP models can be beneficial in terms of prediction robustness as compared to the individual models.
Despite the robust predictions of our KNN-based algorithm, in the distance metric that it employs all SNPs contribute equally. Since SNPs can vary greatly in their correlation to the expression of the nearby gene, and the degree to which their expression correlations on the training set match their correlations on the test set, we asked whether SNPs with better agreement have particular properties. We reasoned that if that were the case, then we might be able to improve performance by using this information to assign differential weights to SNPs within the distance metric. Motivated by studies demonstrating that SNPs that are closer to the transcription start site or SNPs that are located within 3′ UTRs have higher expression correlations [10], [24], [27], we generated a comprehensive set of 117 genomic features for every SNP (Table S15). In addition to previously examined annotations such as the location of the SNP (e.g., within a UTR, intron, or gene body), and its distance from the transcription start and end sites, we also added features such as the conservation and G/C nucleotide content around the SNP, and whether or not the SNP resides within predicted microRNA or transcription factor binding sites. We also added features based on chromatin immunoprecipitation experiments, such as whether or not a SNP is located in a genomic region marked with a specific histone methylation or acetylation.
Next, we extended the KNN algorithm to integrate genomic features into the model by assigning a weight to each genomic feature representing its relative importance. We then updated the distance metric employed by the KNN algorithm such that for each SNP that it sums over when computing the distance between the genotypes of two individuals, it weighs the absolute difference of the SNP's allele counts between the two individuals by the weighted sum of the SNP's genomic features (Figure 3). The extended algorithm learns the weights of the genomic features in an iterative manner, by updating the weight of each genomic feature in a way that maximizes the training R2 and thus improves the overall expression predictions of the algorithm (Methods). The greedy nature of the algorithm ensures that it improves in each iteration until a local maxima is reached in which changes to the weights of the genomic features do not further improve the training R2.
To test whether this extended algorithm can improve the predictions obtained by the simple-KNN algorithm, we applied it to all genes whose training R2 in the simple-KNN algorithm was above 0.05 in each of the cross-validation schemes. For most genes (65–85%, depending on the cross validation scheme, Figure 4A), integrating genomic features did not result in large improvements of the training R2 obtained by the simple-KNN algorithm. Correspondingly, the test R2 values improved for only ∼45% of these genes (Figure 4B). However, for most (66–90%) of the genes for which integrating genomic features resulted in a substantial (above 0.1) increase in their training R2, the test R2 also increased, and the average test R2 increase of these genes was ∼0.1 in all three cross-validation schemes (Figure 4C, Figure 5). Thus, although integrating genomic features improves the training predictions for only a subset of genes (15–35%, depending on the cross-validation scheme), the extended algorithm maintains the robustness property of the simple KNN algorithm, whereby it can identify in advance which genes will be better predicted on unseen individuals.
Having integrated the SNP genomic features into our extended KNN algorithm, we next examined which genomic features are significantly enriched or depleted in it. The extended KNN model explicitly assigns different weights to the different genomic features, where the weights are learned for each gene separately. Therefore, for each genomic feature, we can compute the significance of its enrichment or depletion in having a non-zero weight across all genes. We find that 44/117 (38%) and 50/117 (43%) of the genomic features are significantly enriched or depleted in the extended KNN model, respectively (Figure 6, P<0.01 in all CV partitions, hypergeometric test, FDR [28] correction). In contrast, the regularized linear model does not use the genomic features information in the learning process. However, as different SNPs were assigned different weights, we can determine whether SNPs that share a certain genomic feature are given a non-zero weight significantly more than expected by chance, or significantly less. We find that 47/117 (40%) and 7/117 (6%) of the genomic features are significantly enriched or depleted in the regularized linear model, respectively.
Since the two multi-SNP models were independently learned, we also examined their intersections of enriched or depleted genomic features. We found that 28 genomic features are enriched in both models, which is a significantly larger intersection than expected by chance (P<10−4, hypergeometric test). Among these enriched genomic features are all distance bins of 15 kb or smaller from TSS or TES, suggesting that sequence variation in these regions is most important for regulation by cis-SNPs, consistent with the higher known expression correlation of proximal SNPs [10], [13] (Figure 6, full list of enriched/depleted genomic features in Table S16). In addition, the genomic feature of high (>0.7) regional GC content (50–100 bp region flanking SNP) was also enriched in both models. High GC content was shown to correlate with higher nucleosome occupancy and regulatory function in humans [29], which may explain why SNPs in these regions are more predictive of expression changes. SNPs in predicted binding sites for both transcription factors and microRNAs were also enriched in both models, as may be expected, since such SNPs may affect the likelihood of binding of the transcription factor [30] or microRNA [31].
Among the various post-translational histone modifications included in our genomic feature set (20 methylations and 18 acetylation marks, Table S15), only 5 histone acetylation marks were significantly enriched in both models (Table S16). Notably, all 5/5 (100%) of these acetylation marks are enriched in enhancers, compared to 11/18 (61%) of all histone acetylation marks [32], suggesting that SNPs in enhancer elements may have a larger effect on gene expression. Next, we examined genomic features that are significantly depleted in both the extended KNN and the linear models. We find only 3 such genomic features, consisting of medium-low regional GC content, and heterochromatin marks. These results suggest that areas with lower regulatory capacity have a smaller effect on variation in gene expression.
To conclude, the analysis of the genomic features that both multi-SNP models selected in an unbiased manner as being more (or less) important for the predictions suggests that genetic variation in active and regulatory regions are more likely to affect expression compared to variation in other regions.
To gain further insight into properties of genes that may allow for good predictions, we searched for commonalities in genes that our algorithms predicted successfully. We arbitrarily defined well-predicted genes as those with both training and test R2 above 0.05 in any of the three-cross validation schemes, resulting in 850 such genes. Notably, we found that the KNN algorithm selected significantly fewer neighbors when predicting these genes (K = 11.9 vs. 15.2 for the predictable vs. non-predictable genes, respectively, P<10−200, Table 3 and Figure S4), and that predictable genes had higher absolute expression levels (8.2 vs. 7.7, P<10−18, Table 3) and were more variable, as measured by their coefficient of variation (standard deviation divided by mean expression, 0.039 vs. 0.027, P<10−40, Table 3). Genes with lower expression and less variability are noisier and their measured variation may thus not represent true variation, which may explain why our algorithm is less successful at predicting them. We also found that predictable genes have on average fewer cis-SNPs (291 vs. 307, P<0.002), but their average information content per SNP, measured by the SNP's entropy, is higher (0.44 vs. 0.42, P<10−7).
Finally, comparing genes that are predicted better by the extended KNN algorithm integrating genomic features (KNN-IGF, 412 genes) to those that are predicted better by the simple-KNN algorithm (90 genes), we found that genes that are better predicted in the extended version have more SNPs (307 vs. 255, P<0.0004) and similar information content per SNP (0.44 in both cases). This suggests that by integrating genomic features and weighting the relevance by various genomic annotations, we can integrate more features (SNPs) without overfitting.
We next asked whether predictable genes are enriched for particular biological functions or processes by testing whether they are enriched with particular GO annotations. After correcting for multiple hypotheses testing using FDR [28], we found 18 GO annotations that were significantly enriched in the 850 genes that were better predicted (see Table S17 for the full list). Of these, the most notable enrichments were immune response categories, such as MHC protein complex and antigen processing and presentation (P<10−4 and P<10−3, respectively), which may be expected given that the expression measurements were done in a lymphoblastoid cell line. This result is consistent with the finding that SNPs can effect expression in a cell-type specific manner [33], and with another study that found enrichment for immune-related phenotypes in the genes associated with SNPs that have high expression correlation in these cell lines [7]. Since our algorithm only used cis-SNPs for its predictions, these results suggest that cis-variation may underlie part of the expression variation of immune response genes.
High expression of a gene in a particular cell type suggests regulation in that cell type. Thus, our ability to better predict highly expressed genes suggests that predicting the expression variation of a gene based on its cis-SNPs requires that gene to be regulated in the measured cell type. Clearly, even if the variation that exists in the cis-regulatory sequence of a particular gene affects its expression, the effect will not be observed in cell types in which that gene is not regulated.
Here, we devised two multi-SNP algorithms for predicting gene expression variation among human individuals using only genotype information. The algorithms use information from multiple SNPs in the local vicinity of the predicted gene and we also present a combination of both algorithms and an extension that incorporates heterogeneous sources of genomic annotations of SNPs and assigns higher relevance to variation in SNPs with particular annotations. Notably, we show that our algorithms can predict the expression of genes of unseen individuals with remarkable robustness even when the training set consists only of data from populations different than that of the predicted individuals. In fact, a subset of genes are predicted better by our algorithm when incorporating information from populations other than that of the predicted individuals, suggesting that part of the predictive sequence variation is shared across different populations and may be hard to distill using only individuals from the same population.
The overall number of genes that can be accurately predicted by our algorithms is relatively small, which is expected given that we only consider cis-SNPs and ignore trans sequence variation and environmental factors. However, the robust nature of our algorithms means that although relatively small in number, the genes that can be successfully predicted can be known in advance, as well as the quantitative degree to which they will be predicted. We show that both a regularized linear regression model and a KNN-based model that use multiple SNPs are more robust than a single-SNP based model, especially when predicting expression across populations. We also show that combining these two models into a single model results in improved predictions on held-out test data. In addition, although multi-SNPs models are more robust, we found different genes to be best predicted by different models (i.e., linear, non-linear, single best-SNP) in different cross-validation schemes, suggesting that perhaps the most promising approach for predicting expression variation from genotype would make some combination of models.
Analysis of the genomic features that both our extended KNN algorithm and the regularized linear model selected as important for prioritizing SNPs provides insight into specific classes of SNPs that may have better predictive power and offers concrete means by which we can efficiently explore the vast space of sequence variation and focus on the more relevant variation. Features that we found to have higher predictive power include proximity to the transcription start and end sites, high regional G/C content, and presence within microRNA or transcription factor binding sites. In contrast, we found that SNPs that are located far away from the gene and SNPs located within heterochromatic regions or low G/C content regions have, on average, lower predictive power. We also found that genes that are successfully predicted by our algorithm have, on average, higher expression and more variability and are enriched for classes of genes that have known functional roles in the measured cell types. This finding provides insight into the types of genes that we may expect to be able to predict in a given cell type.
There are many ways in which our algorithms can be further improved, including incorporating trans genetic variation, additional functional annotations of SNPs, and other types of genetic variation such as insertions, deletions, and copy number variation. Examining our ability to predict gene expression in other tissues using the models learned in this work is an interesting avenue of research, given that it has been shown that eQTLs are shared across different tissues [16], and that the similarity between gene expression profiles across different tissues is dominated by heritable effects at the cis-SNPs [34]. Our ability to provide highly robust predictions suggests that the time may be ripe for working on such problems, and that the ability to accurately predict some of the gene expression patterns of individuals using only their DNA sequence may be within reach.
All data used in the different models and the code generating the predictions presented here are available for download in the following URL: http://genie.weizmann.ac.il/software/gen2exp/gen2exp.html
We obtained genotype data from the HapMap project Phase II [8]. We removed the 60 children from the trios, resulting in 210 unrelated individuals from four distinct populations. We downloaded the entire set of measured single nucleotide polymorphisms (SNPs) for each individual, for a total of ∼3.5 M SNPs per individual. For each gene, we extracted all SNPs located inside the gene and those within 100 kb from the transcription start or end sites. We transformed each SNP to a discrete variable with values of 0, 1, or 2, corresponding to the number of minor alleles that each individual has for the SNP. Therefore, an individual with 0 minor alleles will have a value of 0, etc.
We used expression measurements of 15,439 genes performed in lymphoblastoid cell lines for all HapMap individuals [10]. To remove population specific effects, we separately centered the expression of the gene (i.e., subtracted the mean to achieve zero mean) within each of the four populations.
The full list of SNP genomic features that we used along with their description and corresponding reference or website from which they were acquired are given in Table S15.
For each SNP we compute the entropy of its allele counts across all individuals based on Shannon's entropy calculation. Since different SNPs can have a different number of values (i.e., 0/1/2 vs. only 0/1), we first transform the minor allele count of each SNP into two alleles, A and B, such that 0 is converted to A = 0, B = 0, 1 to A = 0, B = 1 and 2 to A = 1, B = 1. Next, we compute Shannon's entropy for each of the alleles A and B, and average their result per SNP. For a specific gene, we average across all its cis-SNPs to obtain an average SNP entropy value per gene.
|
10.1371/journal.ppat.1002693 | Age of the Association between Helicobacter pylori and Man | When modern humans left Africa ca. 60,000 years ago (60 kya), they were already infected with Helicobacter pylori, and these bacteria have subsequently diversified in parallel with their human hosts. But how long were humans infected by H. pylori prior to the out-of-Africa event? Did this co-evolution predate the emergence of modern humans, spanning the species divide? To answer these questions, we investigated the diversity of H. pylori in Africa, where both humans and H. pylori originated. Three distinct H. pylori populations are native to Africa: hpNEAfrica in Afro-Asiatic and Nilo-Saharan speakers, hpAfrica1 in Niger-Congo speakers and hpAfrica2 in South Africa. Rather than representing a sustained co-evolution over millions of years, we find that the coalescent for all H. pylori plus its closest relative H. acinonychis dates to 88–116 kya. At that time the phylogeny split into two primary super-lineages, one of which is associated with the former hunter-gatherers in southern Africa known as the San. H. acinonychis, which infects large felines, resulted from a later host jump from the San, 43–56 kya. These dating estimates, together with striking phylogenetic and quantitative human-bacterial similarities show that H. pylori is approximately as old as are anatomically modern humans. They also suggest that H. pylori may have been acquired via a single host jump from an unknown, non-human host. We also find evidence for a second Out of Africa migration in the last 52,000 years, because hpEurope is a hybrid population between hpAsia2 and hpNEAfrica, the latter of which arose in northeast Africa 36–52 kya, after the Out of Africa migrations around 60 kya.
| We previously showed that the population history of H. pylori may be used as a marker for human migrations, including the demonstration that humans carried H. pylori out of Africa 60,000 years ago during their recent global expansions. But how long were humans infected by H. pylori prior to the out-of-Africa event? Here we showed that chimpanzees in Central-East Africa do not possess Helicobacter-like bacteria, as would have been expected for pathogen-host co-evolution over millions of years. Using H. pylori gene sequences isolated from San, a group of click-speaking hunter-gatherers, and numerous other sources, we calculated that humans have been infected with H. pylori for at least 88,000–116,000 years. Phylogenetic comparisons showed similar evolutionary histories for human and H. pylori lineages and suggest that this association stemmed from a single host jump. We showed that hpAfrica2, the most divergent H. pylori population, arose in the San and that their progenitors were the source of H. acinonychis which was acquired by large felines approximately 50,000 years ago. Furthermore, our data provided clear evidence for a recent second exodus Out of Africa in the last 52,000 years which was essential for the formation of the hybrid population that currently infects Europeans.
| The Gram-negative bacterium Helicobacter pylori infects the stomachs of at least 50% of all humans, causing gastric inflammation in all infected individuals, gastric or duodenal ulcers in 10–15% and gastric carcinoma or lymphoma of the mucosa-associated lymphoid tissue in ∼1% [1]. H. pylori infection is predominantly transmitted within families [2], suggesting that transmission requires intimate contact. Familial transmission has resulted in strong phylogeographic signals within these bacteria [3] due to the frequent, local dispersion of single nucleotide polymorphisms by homologous recombination [4]. At the global level, H. pylori has been subdivided by population genetic tools such as Structure [5] into multiple, relatively distinct populations that are specific for large geographical areas: hpEurope, hpSahul, hpEastAsia, hpAsia2, hpNEAfrica, hpAfrica1 and hpAfrica2 (Figure 1) [6]–[8]. The partitioning of genetic variation in H. pylori was more discriminatory in determining the ancient sources of human migrants in northern India [9], Southeast Asia [10] and the Pacific [8] than were traditional human genetic measures, such as the hypervariable segment 1 of the mitochondrial DNA control region.
Phylogeographic patterns in H. pylori have been shown to reflect significant demographic events in human prehistory [6], [11]. H. pylori has accompanied anatomically modern humans since their migrations out of Africa some 60,000 years ago (60 kya), and mirrors the human pattern of increased genetic distance and decreased diversity with distance from Africa [7]. However, the age of an association between humans and H. pylori has not been elucidated, other than that it predates 60 kya.
One possible scenario is that H. pylori has infected humans since their origins, possibly even prior to the origins of anatomically modern humans. In that event, we might expect to find H. pylori-like bacteria infecting our closest extant relatives, chimpanzees (Pan troglodytes), with whom humans shared a common ancestor ca. 5.4 million years ago (mya) [12]. We tested this hypothesis with negative results, suggesting that human infection by H. pylori likely post-dated the evolution of humans and resulted from a host jump from a different animal. Host jumps are not necessarily unlikely, because the stomachs of multiple animals are infected by diverse Helicobacter species, whose phylogeny is incongruent with that of their hosts [13]. Indeed, the closest known relative of H. pylori is H. acinonychis, which infects large felines and seems to have arisen by a host jump from humans [14]. And the next closest relative is Helicobacter cetorum, which infects dolphins and whales (Figure 2) [15]. All other Helicobacter species are genetically much more distinct.
If H. pylori infection of humans reflected a host jump prior to 60 kya, genetic traces of that event might be present within populations of H. pylori that are native to Africa, where modern humans originated. However, those populations have not yet been extensively sampled, except in West Africa or among recent migrants to South Africa. Diversity within human mitochondrial DNA (mtDNA) coalesces to a last common ancestor ca. 200 kya [16], [17], the time of divergence of mitochondrial haplogroup L0 from haplogroups L1–6. Haplogroups L0–6 are restricted to Africans, whereas three sub-haplogroups of L3 (haplogroups M, N and R) are found globally. L0d, the earliest sub-division within L0, and L0k, are particularly frequent among the San [16], [18], [19], indigenous people who pursued a nomadic, hunter-gatherer lifestyle until very recently. The San are thought to have been originally distributed throughout large parts of central and southern Africa, but are currently restricted to southern Africa [16], [18], [20]. The San speak variants of the “click language” Khoisan, one of the most ancient human language families, currently consisting of three language sub-groups which were geographically distinct within south-central Africa during pre-colonial times: Northern, Central and Southern Khoisan (Figure 3B) [21]. Around 5,000 years ago the Bantu people, consisting of agriculturalists from southern Nigeria and Cameroon who spoke dialects of Niger-Congo, expanded eastwards and southwards into regions of sub-equatorial Africa that were suitable for their equatorial crops [22]. Bantu populations replaced and/or absorbed most of the original, indigenous hunter-gatherer societies in Africa, and their expansion reached its southern limit in eastern South Africa around 700 AD. During that expansion, mtDNA lineages from the indigenous populations such as the San were assimilated into the Bantu gene pool, which is otherwise predominantly composed of L2 and L3 mtDNA haplogroups [23]–[25]. The San continued to thrive in regions that were even further south and west, which were climatically unsuitable for Bantu agricultural crops, such as modern South Africa, Namibia, Botswana and southern Angola. However, the San have been largely displaced since the arrival of Europeans in the 15th century.
We hypothesized that the San might host descendents of the most ancient H. pylori populations, but until now San have not been screened for H. pylori infection. In particular, we anticipated that they might be infected by hpAfrica2, because that population has previously only been isolated in South Africa and it is very distinct from all other H. pylori populations (Figure 1) [2], [6], [7], [26]. We therefore isolated H. pylori from San individuals and calculated the divergence time (TMRCA) when these bacteria split from other African isolates in order to estimate the minimum age of an association between H. pylori and humans.
Duplicate gastric biopsies from antrum and corpus of the stomach and peripheral blood samples were obtained from 30 San volunteers from the !Xun, Khwe and Khomani communities, which represent all three Khoisan language sub-groupings (Table 1). Similar to previous mtDNA genotyping of 31 DNAs from San [16], [19], our analysis of the blood samples showed that mtDNA haplogroups L0d and L0k are particularly frequent in San (67%), much more frequent than in Bantu (Northern Sotho) from South Africa (25%; p = 0.024, U-test) (Table 2). We also cultivated 131 H. pylori isolates from the San biopsies (Table S1), from which we sequenced the same seven housekeeping gene fragments that had previously been used for global analyses [7]. These sequences were concatenated to yield haplotypes with a total length of 3,406 base pairs, of which 56 were unique. Many duplicate haplotypes were obtained from multiple colonies within individual donors, as expected by clonal expansion from a single source. Identical haplotypes were also found between three pairs of donors, suggesting recent transmission (Table S1). The 56 unique haplotypes were distinct from 234 other haplotypes from Africa and 133 from Europe or the Middle East (Table 2).
In order to assign H. pylori from San to populations, we combined these 56 unique haplotypes with 83 haplotypes from two Bantu ethnic groups in South Africa (Northern Sothos living in the Mpumalanga Province near Pretoria [2], [26], and Xhosa from Cape Town [6]), 37 haplotypes from Cape Coloured in South Africa, 91 haplotypes from other areas of Africa and 133 from Europe and the Middle East (Table 2, Figure 4A,B). (In other analyses (data not shown) we also included a global reference data set of 1040 haplotypes that had previously been assigned to H. pylori populations [7], [8] but they consistently yielded the same population assignments for the reference haplotypes as in previous studies, and did not reveal any novel populations.) Bayesian cluster analysis was performed with the non-admixture model of Structure [5] for estimates of the total number of populations, K, between 2 and 5, which was the highest value of K that yielded consistent clustering and consistent probability estimates between individual runs. Almost half of the San haplotypes (26/56, 46%) belong to hpAfrica2 (Figure 3A, Figure 4A,B, Table 2). hpAfrica2 isolates were found in all three San communities, ranging in frequency from 28% of all haplotypes (!Xun) to 55% (Khwe, Khomani).We also identified 35 hpAfrica2 haplotypes among isolates from the Northern Sotho near Pretoria and from Xhosa and Europeans in Cape Town.
hpAfrica2 is related to H. acinonychis (Hac) from large felines (see below). We therefore performed additional Structure analyses on haplotypes of hpAfrica2, hpAfrica1 and Hac. Under a two population model (K = 2), hpAfrica2 clustered together with Hac, separately from hpAfrica1 (Figure 4C). hpAfrica1, hpAfrica2 and Hac were all distinct at K = 3 whereas at K = 4, hpAfrica2 haplotypes partitioned into two sequence clusters, one associated with southern isolates from South Africa (Khomani and non-San) and the second with northern isolates from speakers of the Northern Khoisan (!Xun) and Central Khoisan (Khwe) language subgroups (Figure 3C, Figure 4C). The only exceptions were two haplotypes from one Khomani San individual, which were assigned to the northern isolates.
The level of genetic diversity, π, was significantly higher among hpAfrica2 haplotypes from San (95% CL [2.83,3.08%]) than from Bantus ([2.52,2.80%]), suggesting that the San isolates might be ancestral. To test this inference, we determined the phylogenetic structure of all 58 hpAfrica2 haplotypes with ClonalFrame, which can discern ancestral relationships even in the presence of homologous recombination [27], [28] (see below). The consensus tree from this analysis shows that the southern (Khomani, Bantu) San haplotypes fell into a young clade which emerged from an more ancestral population of hpAfrica2 haplotypes, all of which were from San and most of which were from the northern Khwe and !Xun (Figure 3C). These observations suggest that hpAfrica2 evolved within the San and was subsequently transmitted to Bantus.
Almost all non-other haplotypes from San were assigned to hpAfrica1. In contrast, to the results described above, these were less diverse (π 95% CL [2.50, 2.82%]) than hpAfrica1 from Bantus ([3.10, 3.20%]), suggesting that the San had acquired hpAfrica1 from Bantu.
The ages of lineages of closely related bacteria that evolved in recent decades can be dated by genomic analyses of isolates known to span that time range [29], [30]. However, it is difficult to accurately date the origins of individual lineages or species of microorganisms over longer time periods [31]. For example, although HIV is thought to be of recent origin, rabbit retroviruses that are related to HIV differentiated from them over seven million years ago [32], raising the possibility that HIV itself is also old but that its diversity was reduced during a recent bottleneck. We have developed a method for dating the origins of H. pylori in the last 60 kyr by calibrating the genetic distances between H. pylori population against the dates of separation of the corresponding human populations [8]. After stripping signals introduced by homologous recombination from the H. pylori sequence data, a linear relationship was found between the genetic distances and the archaeological dates (Figure 5), which allowed the estimation of unknown dates of population splits, such as the original peopling of the Pacific [8]. We have now used this approach to calculate the age of splits between the African lineages of H. pylori plus Hac. Similar to our previous analyses, we used two independent approaches to construct a phylogenetic tree, ClonalFrame [27] and IMa) [33].
ClonalFrame calculates a coalescent whose branch lengths exclude stretches of clustered nucleotide polymorphisms that result from recombination, although these stretches are used to calculate the topology when they are informative [27]. ClonalFrame can retrieve the clonal frame in moderately recombining bacteria such as Bacillus and Salmonella [27], [34]. And in our experience, the phylogenies recovered by ClonalFrame for H. pylori are quite insensitive to the size of the dataset. However, determining an accurate age for the coalescent with such an approach depends on accurate rooting through an outgroup. We were unable to accurately root the tree with housekeeping genes from published genomes of enterohepatic Helicobacter species or from Campylobacter species because none of them contained orthologs of all seven housekeeping genes in our dataset. We therefore shotgun sequenced the genome of H. cetorum strain MIT 99-5665, which represents the closest known relative of H. pylori and Hac [15] (Figure 2), and used the orthologous nucleotide sequences from that genome as an outgroup for rooting the ClonalFrame tree. Independent analyses yielded the same rooting branch point when the tree was rooted with and based on orthologs that were shared between H. pylori and enterohepatic Helicobacter genomes (data not shown).
IMa is a mainstream method for the inference of historical population genetic parameters that were associated with historical splits between pairs of populations [33]. IMa simulates the posterior probabilities for the population parameters theta (Neμ), where Ne is the effective population size and μ is the mutation rate, m (the effective number of migrants per generation) and t (the time since population splitting). IMa accounts for intermittent back-migration after population splits. We therefore identified blocks of recombinant DNA in each pair of populations by the four-gamete test [35] and stripped those blocks from the data. This test assumes an infinite sites model, which is only applicable when the mutation rate is lower than the recombination rate, as is the case for H. pylori [4]. The remaining blocks of sequence were used to estimate dates of splitting by the isolation with back-migration model. Although we do not know of other attempts, except our own [8], to use IMa for the dating of bacterial phylogenies, it has been extensively used for to date population splits among eukaryotic populations [36]–[38].
The models used by ClonalFrame and IMa are fundamentally different, except that both used the same archaeological and molecular calibration points (Table 3). A linear relationship between genetic distance and calibration date with high regression coefficients was found by both sets of analyses (Figure 5), and they estimated overlapping extrapolated dates, with one minor exception (Table 4). These overlapping estimates indicate that our age estimates are primarily dependent on the archaeological calibrations and are independent of method. The TMRCA of all H. pylori plus Hac lineages was 88–116 kya (ClonalFrame: 88–92 kya; IMa: 92–116 kya; Table 4, Figure 6A). The date for the coalescence of non-recombining Y-chromosome lineages in modern humans is similar at 90 kya [39] to 141.5±15.6 kya [40] whereas the date of split between L0 and L1–6 mtDNA haplogroups in humans is older, 194.3±32.5 kya, (Figure 6B) [16], [17]. Despite the different age estimates, the topology and branching pattern of the genealogies are strikingly similar between H. pylori and human mtDNA (Figure 6). The similarity between these two trees could not be compared directly because the numbers of lineages differ between the two genealogies. We therefore performed a quantitative test of whether similar phylogeographic trends exist in both H. pylori and mtDNA data by performing a Mantel regression of the maximum composite likelihood distances between pairs of populations from comparable geographic sources of both humans and H. pylori (Figure 7, Tables S3, S4). The results showed that 60% of the variation in both data sets is distributed similarly (P<0.0001).
We also used these data to estimate the ages of splits between individual lineages within the H. pylori/Hac phylogeny, all of which seem to be later than the Out of Africa migrations of 60 kya [7]. Descendents from the last common ancestors of H. pylori plus Hac diverged into two distinct super-lineages, one of which gave rise to hpAfrica2 plus Hac and the second of which gave rise to all other populations (Figure 6A). The TMRCA for the split between hpAfrica2 and Hac is 43–56 kya (Table 4), and hpAfrica2 subsequently split (32–47 kya) into the northern and southern isolates. We note that a similar date (40 kya) was recently estimated for the TMRCA of Y-chromosome haplogroup A-M51 among the San by Henn et al. [20], which also subsequently split between northern and southern San populations. Within the other super-lineage, the estimated TMRCA was 36–52 kya for the African populations hpAfrica1 and hpNEAfrica (Table 4).
Our calculated TMRCA of 88–116 kya for H. pylori plus Hac might represent the date of a host jump to humans from a different animal host. Due to lineage sorting and bottlenecks, the date of such a host jump may also have been considerably earlier. We therefore attempted to isolate H. pylori from chimpanzees, who are our closest relative. We collected stomach antrum and corpus biopsies from 42 captive, wild-born chimpanzees (Pan troglodytes) that originated from the Great Lakes region of Central-East Africa (Uganda, Rwanda and the Democratic Republic of the Congo) but now live on an isolated island sanctuary in Uganda. Endoscopic examination during esophagogastroduodenoscopy identified a mild gastritis in some animals, suggesting that they might be infected with H. pylori. However it is known that there is a poor correlation between the endoscopic presence of gastritis and the prevalence of H. pylori as gastritis may also be caused by other non-infectious etiologies. The biopsies were taken with single use biopsy forceps (Radial Jaw, Boston Scientific) by an experienced gastroenterologist who routinely obtains aseptic biopsies that allow cultivation of H. pylori, including hpAfrica2. The biopsies were immediately stored and transported in liquid nitrogen. Attempts at cultivation of the biopsies were performed in an H2-containing microaerophilic atmosphere under growth conditions that are routinely successful for the cultivation of H. pylori, H. acinonychis and H. cetorum. Over the past years, these methods have successfully cultivated over one thousand H. pylori strains from multiple geographic locations including remote regions of Siberia, Papua New Guinea and Cameroon. And they routinely succeed with hpAfrica2, which are particularly difficult to grow in the absence of atmospheric H2 (unpublished data). However, we were unable to cultivate Helicobacter-like bacteria from any of the chimpanzee biopsies.
We reasoned that H. pylori-like bacteria from chimpanzees might not grow under these cultivation conditions, and therefore attempted to amplify 16S rRNA sequences from the bacterial DNA in the biopsies. To this end, we designed oligonucleotide primers that should successfully amplify PCR from any Helicobacter species. In independent (unpublished) experiments, these primers have been successful at amplifying rRNA sequences from H. cetorum in fecal samples from dolphins. However, the PCRs performed with DNA extracted from the chimpanzee biopsies all failed to amplify any products, except for two instances of air contamination from previous water controls. These results suggest that H. pylori might be rare in chimpanzees, possibly indicating that it has not coevolved with hominids during the evolution of the great apes.
This project was initiated because we were intrigued by the great genetic distance between hpAfrica2 and other populations of H. pylori as well as by the geographical distribution of hpAfrica 2, which has only been isolated in South Africa. Our data indicate that both of these observations result from an original association between hpAfrica2 and the San. In support of this interpretation, the deepest branches within the hpAfrica2 genealogy are associated with the northern San, represented by the !Xun and Khwe ethnic communities (Figure 3). A further indication that hpAfrica2 evolved in the ancestors of these northern click-speaking people are the results obtained with IMa, according to which migration within the hpAfrica2 lineage has been predominantly from north to south (m2 = 2.20) rather than south to north (m1 = 0.45) (Table 4). Finally, the genetic diversity is greater among hpAfrica2 from San than from Bantu, indicating that it was transmitted to Bantu in the last few hundred years since their arrival in southern Africa.
These conclusions are also relevant to the host jump from humans to large felines which gave rise to H. acinonychis (Hac) [14]. Firstly, our population assignments and phylogenetic reconstructions show that although they are discrete taxonomic species, Hac is part of the same genetic super-lineage of H. pylori as hpAfrica2 (Figures 4 and 6A), and the host jump occurred after H. pylori had sub-divided into two super-lineages, The coalescence of hpAfrica2 and Hac was estimated at 43–56 kya, which provides an estimate of the date of the host jump to large felines. This is later than our prior estimate of the date of that host jump as 100 kya [14], [41]. However, Hac was thought to be phylogenetically distinct from H. pylori, rather than nested within it, and we based our calculation on a comparison of the genomes of strains 26695 (hpEurope), J99 (hpAfrica1) and Sheeba (Hac), which essentially equates to the coalescent for the two super-lineages of H. pylori of approximately 100 kya. In the light of our conclusion that the hpAfrica2-containing super-lineage is associated with the San, the host jump that resulted in Hac may have arisen after the consumption of the stomach contents of an infected ancestor of the San by a large feline.
Our data shows that anatomically modern humans were infected by H. pylori long before their migrations out of Africa of ∼60 kya [7], [42]. We estimate the minimum age of that association to be approximately 100 kyr (range 88–116). This is comparable to the age of the coalescence of the human Y-chromosome and about half of the coalescent for mtDNA. The age of a coalescent is a minimal date estimate because lineage sorting and bottlenecks lead to extinction of older lineages, resulting in a single genealogical source of all subsequent descendents. Indeed, the genealogies of H. pylori and mtDNA are very similar (Figure 6), and a Mantel regression indicated that the geographical distribution of the genetic diversity within both humans and H. pylori is also similar, both within Africa and outside. These results suggest that anatomically modern humans were infected by H. pylori since their origins. We therefore anticipated that we would isolate relatives of H. pylori from wild-born chimpanzees, our genetically closest relatives. However, we failed in this effort, and also failed to PCR amplify Helicobacter rRNA sequences.
Our failure does not provide convincing evidence that chimpanzees are not infected with close relatives of H. pylori. Those close relatives might not have been capable of growth under the conditions we used. Alternatively, unknown technical problems might have affected the sampling, transportation or PCR reactions. We only sampled 42 chimpanzees of subspecies Pan troglodytes schweinfurthii, all of whom were from East and Central Africa, and chimpanzees elsewhere in Africa might be infected with a H. pylori-like organism. We also note that the lack of isolation of Plasmodium spp. from eastern lowland gorillas and bonobos [31] and SIV in eastern chimpanzees [43] was due to variable infection rates among hominid apes from different areas of Africa. And as a final alternative, the human-chimpanzee ancestor might have been infected with Helicobacter precursors, but chimpanzees subsequently lost those bacteria secondarily. Additional analyses involving the other chimpanzee subspecies as well as bonobos and gorillas might help resolve these uncertainties.
The literature contains many reports of the isolation of H. pylori from distantly related primates. These probably reflect either transmission from humans to animals during captivity, or infection with genetically distantly related Helicobacter. H. pylori-like bacteria have been isolated from macaques, including the named species H. nemestrinae [44]. However, the haplotype of H. nemestrinae, which was isolated from pigtailed macaques (Macaca nemestrina) in an American zoo, was subsequently shown to belong to the hpEurope population of H. pylori, which is common in the USA [45]. Widespread H. pylori infection has been reported among rhesus macaques (Macaca mulatta) from China [46] and crab-eating macaques (Macaca fascicularis) from Vietnam and the Philippines [47]. The 16S rRNA sequences of isolates from Philippine macaques were similar to those of hpEurope, which is also frequently isolated from inhabitants of the Philippines [7], [10]. Isolates from Vietnamese macaques belonged to hpEastAsia [47], as do most H. pylori strains from Vietnamese [7]. In fact, macaques are so readily colonized by human H. pylori that rhesus macaques [48], crab-eating macaques [49] and Japanese macaques (Macaca fuscata) [50] are all used as animal models for H. pylori infection and pathogenicity.
Other isolates of Helicobacter from primates are from species that are only very distantly related to H. pylori. In addition to H. pylori of presumptive human origin, macaques are also infected with Helicobacter suis [51], Helicobacter cinaedi [52] and Helicobacter macacae [53]. H. suis has also been isolated from mandrill monkeys and crab-eating macaques in a zoo [51]. However, H. suis is associated with gastritis and ulceration in pigs, and belongs to a parallel lineage to H. pylori, H. acinonychis and H. cetorum in a phylogenetic tree of 16S rRNA sequences (Figure 2). H. cinaedi and H. macacae are even more distinct from H. pylori (Figure 2), and belong to the genetically quite distinct enterohepatic Helicobacters whose primary site of infection is the intestine, colon or liver. Thus, none of these primate isolates are likely candidates for a close relative of H. pylori that might have co-evolved with hominid apes.
We conclude that there is no direct evidence for co-evolution of H. pylori and humans prior to approximately 100 kya. Furthermore, the genealogical relationships within Helicobacter 16S rRNA are consistent with multiple host jumps, as is already indicated by the fact that the closest relative of H. pylori are associated with large felines (Hac) and dolphins/whales (H. cetorum). We therefore propose that the association of H. pylori with humans also reflects a host jump to humans from an unknown species, which occurred approximately 100 kya or earlier. In principle, two later host jumps might explain the existence of two super-lineages of H. pylori, but this seems less likely because the similar phylogeographical patterns of H. pylori and mtDNA haplogroups indicate that they have undergone a parallel evolutionary history.
Despite the general similarities between the genealogy of H. pylori and human mtDNA, there is a striking difference in respect to Europe.
Archaeological differences in the technology of stone tools have been used to justify two out of Africa migrations from two different source populations in Africa, the first spreading “Middle Paleolithic” technology in southern Asia, and the second distributing “Upper Paleolithic” from northern Africa into the Levant and Europe [54], [55]. However, a single successful out of Africa event is indicated by the fact that modern Asian and European mtDNA haplotypes are all derived from a subset of the L3 haplogroup [56], and two independent migrations from Africa were thought to be unlikely due to the greater diversity of mtDNA haplotypes in Africa [17], [25], [57].
The phylogeographic diversity within H. pylori is inconsistent with a single human expansion from Africa. H. pylori accompanied humans on the migration of ∼60 kya [7], reaching Oceania not long thereafter [8]. However, European H. pylori possess distinct properties from most other global populations of these bacteria. H. pylori from Europe, the Middle East, western Asia and India belong to the hpEurope population [6], [7], [10], [58]–[60], which unlike Europeans is typified by great genetic diversity, greater than in Africa except for southern Africa where strong genetic diversity results from the presence of the second super-lineage (hpAfrica2). The great diversity of hpEurope was attributed to the fact that it is a hybrid population which arose from the admixture of AE1 (Ancestral Europe 1) and AE2 (Ancestral Europe 2) (Figure 4B) [6], [7]. AE1 arose in Central Asia after H. pylori was carried out of Africa during the Out of Africa migration of ∼60 kya [7], and its descendants are found among extant hpAsia2. However, the data in Figure 6A indicate that AE2, whose extant descendents in hpNEAfrica are associated with northeast Africa, first split from its sister lineage hpAfrica1 36–52 kya, after the (first) Out of Africa migration. We therefore hypothesize that a second Out of Africa migration in the last 52 kya brought AE2 to the Levant, after which it came into secondary contact with AE1. Subsequent extensive admixture resulted in hpEurope, which subsequently spread to Europe and western Asia (Figure 8). This interpretation differs from classical interpretations based on uni-parental markers (mtDNA, non-recombining Y chromosome) [56], [57] but a secondary colonization of Europe is supported by other archaeological and genetic data. Modern humans spread rapidly from the Levant to most of Europe by 40–46 kya [61]–[63], accompanied by “Upper Paleolithic” or “Mode 4” stone tools, which first occurred in North Africa and Eurasia after 50 kya [54], [63]. During the Last Glacial Maximum 26.5–19 kya [64], Europeans retreated to refugia such as the Iberian Peninsula and the Ukraine, which were the sources of re-colonization of Europe after the end of the ice age [65]–[68]. Signs of this re-colonization are evident in human DNA, e.g. mtDNA haplogroups that are wide-spread among Europeans (HV3, HV4, U4a1) can be traced back to 12–19 kya in eastern Europe, supporting an expansion from an Ukrainian glacial refugium [65]. Similarly, other common European haplogroups (V, H1, H3) arose in the northern Iberian peninsula soon after the Last Glacial Maximum, and dispersed into Europe after a population expansion in Iberia 10–15 kya [66]–[68].
The approach described here allowed estimates of the TMRCA of populations whose ancestry is largely derived from a single ancestral population, but does not allow dating of admixed populations such as hpEurope. hpEurope arose after its parental populations, i.e. within the last 52 kyr. Its near universal presence from Europe through to Western Asia may have been facilitated by any of multiple postulated major human migrations, including the initial colonization of Europe, the re-colonization of Europe from the Ukraine and Iberia and the Neolithic spread of agriculture from the Fertile Crescent into Europe, Western Asia and India in the last 10 kyr [22]. We note however that the migration of farmers from the Near East during the adoption of agriculture was likely very limited because the Near Eastern Neolithic component of the mtDNAs gene pool of modern Europeans is only 15% [69], and the majority of European lineages date back to late glacial and post-glacial times [70]. Similarly, ancient DNA analyses suggest that modern European ancestry is closer to that of the ancestral European hunter-gatherers than that of early farmers [71], [72]. If the initial AE1–AE2 secondary contact occurred as early as 45–52 kya in the Levant, hpEurope might have accompanied the first modern humans into Europe. However, this seems also unlikely because the presence of people from the first “Middle Paleolithic” migration Out of Africa in the Levant is not supported by archaeological evidence [55]. Thus, if initial Europeans were colonized with H. pylori, those bacteria were subsequently replaced by hpEurope, similar to the replacement of hspAmerind strains by hpEurope strains among Amerindians from South America [73]. To illustrate these interpretations, we show approximate routes and timings for a second colonization of Europe based on the properties of H. pylori populations (Figure 8), in which migration waves from North East Africa and Central Asia met and admixed in the Middle East and/or Western Asia sometimes 10–52 kya. The widespread presence of hpEurope in Mediterranean Africa is then attributed to later migrations to northern Africa, including migrations from Iberia (mtDNA haplogroup H1; 8–9 kya) [74], the Near East (mtDNA haplogroup M1; 35 kya) [75]; autosomal DNA; >12 kya [76]), or even as recently as the expansion of the Islamic caliphate in the last 1200 years. Our model also summarizes the dates of other human migrations that have distributed H. pylori from its southern African source (Figure 8).
The results presented here provide a framework for the association of H. pylori with humans over the last 100,000 years, possibly after H. pylori was first acquitted by a host jump from an unknown source. This association began in Africa, where two discrete super-lineages differentiated. One of the super-lineages was predominantly associated until very recently with San (hpAfrica2) and large felines (Hac), whereas the second is widespread throughout Africa (hpAfrica1, hpNEAfrica) and accompanied anatomically modern humans during their first Out of Africa migration, which subsequently resulted in the Asian and Oceanic lineages hpAsia2 hpAsia and hpSahul. Subsequent migrations of ancestors of the African hpNEAfrica and/or the Asian hpAsia2 populations resulted in the admixed hpEurope population which then became the predominant population of extant H. pylori in Europe, the middle East and western Asia. We have provided date estimates for most of these historical events, thus providing a paradigm for the long-term historical reconstruction of the evolutionary path of a bacterial species.
Esophagogastroduodenoscopy was performed with written informed consent at the Interventional GI-Endoscopy Department of the Unitas Hospital in Pretoria, South Africa under ethics certificate 32/2007 (University of Pretoria, Faculty of Health Sciences Ethics Committee), with the permission of the San Council of South Africa and with permission of the ethics committee of the Charité hospital in Berlin, Germany (ethics certificate EA1/071/07). Biopsies were obtained from the antrum and corpus of the stomachs of 30 self-proclaimed San individuals. Among these were 9 Khomani San from South Africa, 11 !Xun from southern Angola, 8 Khwe from northern Namibia, and two individuals of unknown San ethnicity (Table 1, Table 2). Gastric biopsies were also taken from the stomachs of 42 anaesthetized chimpanzees (Pan troglodytes) undergoing annual medical examinations at the Ngamba Island Chimpanzee Sanctuary on Lake Victoria, Uganda. The examinations were conducted in full accordance with guidelines set by the International Primatological Society, Pan African Sanctuaries Alliance (PASA) and standard operating procedures by the Chimpanzee Sanctuary & Wildlife Conservation Trust (CSWCT) all of which practice the highest welfare standards for chimpanzees in captivity. Collection of biopsies was also approved by the Uganda Wildlife Authority (UWA) and ethical approval was obtained from the Uganda National Council for Science and Technology (UNCST certificate NS 71). These chimpanzees were all illegally captured in the wild as infants, but have since been confiscated and donated to the island sanctuary. All biopsies were placed in transfer medium, frozen immediately in liquid nitrogen, and kept at −80°C until transfer by courier in a liquid nitrogen dry shipper to the Max Planck Institute for Infection Biology in Berlin (CITES permit Sn. UG 001944), where we attempted to culture Helicobacter from them.
Gastric biopsy specimens were grown on cultivation plates containing GC agar (Remel, Lenexa, USA) supplemented with 10% (v/v) donor horse serum (Biochrom KG, Berlin, Germany), VITOX vitamin supplement (Oxoid, Basingstoke, UK) and selective antibiotics (10 mg/L vancomycin, 5 mg/L trimethoprim, 5 mg/L amphotericin B and 25000 U/L polymyxin B). The plates were incubated for 3 to 5 days at 37°C in a Forma Series II 3110 Water-Jacketed CO2 incubator (Thermo Scientific) wherein the CO2 concentration was kept at 5% and the O2 concentration was regulated to 5% through a mixture of H2 (10%) and N2 (90%), which facilitates efficient cultivation of microaerophilic bacteria. In case of lack of bacterial growth, the plates were incubated for up to 15 days.
We attempted to cultivate Helicobacter from a total of 81 chimpanzee gastric biopsies, one from the antrum and one from the corpus region of each of the 42 chimpanzees except for 3 individuals from which only one biopsy each was cultured. However, none of the chimpanzee biopsies yielded cultures of Helicobacter-like species. Since some of these Helicobacter-like species may not be detectable through bacterial culturing, a culture-free approach of 16S rRNA amplification was utilised. DNA was extracted from chimpanzee gastric biopsies with the DNeasy Blood and Tissue Kit (Qiagen). Due to the low levels of bacterial DNA that were expected, universal prokaryote 16S rRNA primers F24 (5′-GAGTTTGATYMTGGCTCAG) and F25 (5′-AAGGAGGTGWTCCARCC) were used to perform an initial PCR. This was used as a template for a second round of PCR using Helicobacter-genus specific 16S rRNA primers C97 (5′-GCTATGACGGGTATCC) and C05 (5′-ACTTCACCCCAGTCGCTG). These primers should yield a final amplicon of 1200 base-pairs. The PCR reaction (30 µl) contained 10× PCR Buffer (Qiagen), 330 µM dNTPs, 5 µM of each primer, 5 U of Taq polymerase and 5 µl (Biopsy) or 2 µl PCR product as template DNA. Amplification conditions were as follows: An initial denaturation at 96°C for 5 minutes followed by 35 rounds of denaturation at 96°C for 30 seconds, annealing at 58°C for 30 seconds and extension at 72°C for 90 seconds. A final extension step of 72°C for 10 minutes was then performed. After the initial enrichment PCR, a 2 µl aliquot of the PCR mixture was used as template for the second PCR. A 5 µl aliquot was examined by electrophoresis on a 1% agarose gel containing a 1/10000 dilution of Sybrsafe dye, and visualized under UV light. Using both culture dependent and culture free methods, no Helicobacter-like species were detected in any of the 42 chimpanzees despite intimate association between some chimpanzees and their caretakers.
Since simultaneous infection with multiple, distinct H. pylori strains has been frequently observed among people from Southern Africa [2], four colonies per San individual were analyzed from the stomach biopsies, two from the antrum and two from the corpus. DNA was extracted from cultures grown after single colony isolation using a DNeasy Blood and Tissue kit (Qiagen). The forward and reverse strands of fragments of atpA, efp, mutY, ppa, trpC, ureI, yphC were sequenced from each isolate as previously described [3], [7]. All sequences, primer combinations, PCR conditions and information on isolates are publicly available at http://pubmlst.org/helicobacter, where the new isolates described here are listed as ID numbers 1472–1527. The sequences from all seven housekeeping gene fragments were concatenated to form a 3,406 base-pair sequence. The strains cultured from the 30 San individuals (Table S1) represented 56 unique haplotypes that were used for further analysis (Table 2).
DNA was extracted from San and Bantu (Northern Sotho) blood samples taken under ethics certificates 32/2007 (University of Pretoria, Faculty of Health Sciences Ethics Committee) using the DNeasy Blood and Tissue Kit (Qiagen). Assignment of individual samples to mtDNA macro-haplogroups L0–L6, M, N and R were done using a SNaPshot minisequencing procedure [77]. The mtDNA control region was amplified and sequenced following previously published methods [77], [78]. Sequence data were obtained for hypervariable segments I (HVS I; nucleotide positions 16024–16400) and HVS II (nucleotide positions 57–302), and used to assign individuals to haplogroups and sub-haplogroups according to the nomenclature proposed by Behar et al., 2008 [16].
The 56 unique haplotypes from San were analysed together with a previously described global data set of 1040 haplotypes [7], [8] as well as 83 haplotypes from South African Bantu of Northern Sotho and Xhosa ethnicities [2], [6]. The “no admixture” model of the program Structure V2.0 [5] was used to assign individual strains to the known bacterial populations (Figure 1) [6]–[8]. New populations were not detected. Subsequent analyses were performed exclusively on African isolates, for each of the test number of populations (K) ranging from 2 to 5 (Figure 4). Each set of conditions was tested in ten independent Structure runs, with consistent results.
The linkage model of Structure V2.0 [5] was used to assess the ancestral composition of individual haplotypes in order to differentiate whether populations arose as a result of gradual genetic drift or by hybridisation of two distinct populations that have come into secondary contact. We identified the previously reported populations ancestral EastAsia, ancestral Europe1 (AE1), ancestral Europe2 (AE2), ancestral Africa1, ancestral Africa2 [6], [7] and ancestral Sahul [8]. Runs assuming K = 4 were used to determine the ancestral composition of the European and African haplotypes displayed in the Distruct [79] plot in Figure 4B.
The relatedness among the strains within hpAfrica2 was analysed using the software ClonalFrame v1.1 [27]. This software estimates the clonal (vertical) genealogy of a set of DNA sequences by jointly simulating mutation and homologous (horizontal) recombination events under a neutral coalescent using a Bayesian Markov chain-Monte Carlo (MCMC) framework. Inferred horizontal events at each node are discarded for the calculation of node height, but they are used to infer common ancestry between lineages, further adding to the robustness of the genealogical reconstruction. The resulting phylogeny therefore represents the best estimate of a clonal genealogy that is currently computable. ClonalFrame phylogenies have been used successfully in resolving human demographic events in other parts of the world [8]. Bayesian parameter space was explored with 100,000 iterations, recording the posterior sample every 100 iterations, and discarding the first 10% of iterations as burn-in. This analysis was repeated 100 times, and an 80% majority rule consensus of all the sampled genealogies was computed using Treefinder [80]. Nucleotide diversity and 95% confidence limits (π95) within San and non-San (Bantu) hpAfrica2 strains were calculated in DnaSP4 [81], as were comparisons within San and non-San hpAfrica1 isolates.
H. cetorum is a gastric Helicobacter from dolphins and whales and is the closest known relative of H. pylori according to rRNA sequences (Figure 2) [15]. It was used as an outgroup for genealogical reconstruction. In order to obtain the MLST sequences of H, cetorum, a draft sequence (169 contigs, 20 fold coverage, total contig length 1,744,916 bp) of the genome of H. cetorum strain MIT 99-5665 was obtained by shotgun sequencing with a Roche/454 Genome Sequencer FLX. H. cetorum sequences corresponding to the seven H. pylori housekeeping gene fragments were identified by BLAST searches. All sequences were confirmed by Sanger sequencing of PCR fragments amplified with the primers shown in Table S2. The sequences were submitted to the EMBL database (accession numbers: FB908911–908917).
We used an individual as well as a population approach to determine the structure of the African H. pylori populations relative to other populations distributed in other parts of the world.
We again used ClonalFrame (v1.1) to estimate a clonal genealogy from multilocus sequence data of 91 globally distributed H. pylori strains, rooted with H. cetorum. This global phylogeny was estimated 100 times as above, but with each independent run recovering the same nodal topology (Figure 6A).
Recombination between closely related strains can also introduce no visible change or single nucleotide changes that resemble point mutations. These would tend to bias the ratio of rates of recombination and mutation (the rho/theta parameter) leading to an overestimation of node height. ClonalFrame corrects for this by simultaneously estimating the ratio of rates at which recombination and mutation introduce differences (the r/m parameter) which is less likely to be affected by this kind of recombinational event. To further control for the possibility of a non-linear relationship of ClonalFrame node height with time, we used the software IMa [33] for an independent, population-based estimate of the global structure within H. pylori. We chose Hey and Nielsen's [33] model of isolation with migration to analyze these data because it does not assume that the two populations are at equilibrium for mutation, drift or migrations. Furthermore, the model also assumes that gene flow was possible after the time of population splitting, and using a Bayesian approach, simultaneously estimates the posterior distributions of the following model parameters: time since population split (t), the population parameter theta (θ) and the migration parameter m. Sequences were first processed by the four-gamete criterion [35] implemented in DnaSP4 [81] in order to identify recombinant blocks of DNA sequence between pairs of populations. These blocks were omitted from the data set, resulting in between 42 and 100 non-recombinant blocks, depending on the pair-wise comparison, that were coded as separate loci. Treating these separate blocks as separate loci was chosen because composite likelihoods tend to estimate the true posterior probability of a parameter when the number of loci is high [82]. Bayesian parameter space was heuristically sampled by an MCMC simulation of 1,000,000 iterations, and genealogies were sampled every 100 iterations after a burn-in of 100,000 iterations. Mixing and convergence was stimulated by 100 geometrically-heated Metropolis-coupled chains, with 100 chain swapping attempts between iterations. All estimates were taken after joint parameter maximization. The analysis was repeated 10 times to determine whether MCMC simulations converged to a similar result. Among the analyses of African populations, all four pair-wise comparisons yielded consistently unimodal posterior distributions of the TMRCA, suggesting that these pairings constituted monophyletic groups. All four pairings (Table 4) were consistent with the topology of the global genealogy generated by ClonalFrame (Figure 6A), which confirmed the ancestral branching of hpAfrica2 and showed that hpNEAfrica and hpAfrica1 are sister populations. The highest and lowest values for each set of 10 simulations were regarded as the spread of the mean t.
ClonalFrame and IMa were used to determine lineage and population coalescence respectively, both using the global rate minimum deformation (GRMD) rate-smoothing optimisation in Treefinder [80]. GRMD is a rate-smoothing method that minimises a cost function to maintain rates along different lineages that are as similar to each other as possible, within the imposed time boundaries. This method is appropriate given the linear relationships of ClonalFrame's node height and IMa's t with calibration time (Figure 5). The spread of node heights of the 100 ClonalFrame genealogies and the spread of t was combined with six known calibration points (Table 3), where node height and t values had been previously determined [8], to generate TMRCA estimates. We used a Treefinder [80] script (Text S1) to generate 95% confidence limits from the spread in t values. The ranges of population and individual-based TMRCA dates were found to overlap for all but one case, but the confidence limits returned by the IMa were greater. The upper and lower values described in Figure 6 and in the text are the highest and lowest values that were estimated using both methods.
Pair-wise population divergence estimates were obtained in Mega [83], using maximum composite likelihood distances for both concatenated H. pylori sequences (n = 485, Table S3) and whole genome (or coding region) human mtDNA sequences (n = 447, Table S4). A Mantel test was used to perform a distance matrix regression in GenAlEx [84]. The probability that a random regression co-efficient was greater than or equal to the observed value was determined by 9999 permutations.
We obtained 16SrRNA sequences of the following various gastric and enterohepatic Helicobacter species from Genbank and/or extracted the 16SrRNA sequences from complete genomes: H. pylori (accession number AE000511, human host), H. acinonychis (AM20522, lion), H. cetorum (AY143177, dolphin), H. felis (M37643, cat), H. bizzozeronii (Y09404, dog), H. salomonis (U89351, dog), H. cynogastricus (NR_043457, dog), H. heilmannii (HM625820, cat), H. baculiformis (EF070342, cat), H. suis (AF127028, pig), H. bovis (AF127027, cattle), H. macacae (HQ845265, rhesus monkey), H. canadensis (AF262037, human), H. cholecystus (U45129, hamster), H. cinaedi (M88150, human), H. bilis (U18766, mouse), H. canis (L13464, dog), H. hepaticus (AE017125, mouse), H. muridarum (M80205, rodent), H. pullorum (L36144, chicken), H. trogontum (U65103, rat), H. fennelliae (M88154, human), H. rodentium (U96296, mouse), H. mesocricetorum (AF072334, hamster) plus Wolinella succhinogenes (M88159, cow) and Campylobacter jejuni (AL111168, chicken) as outgroups. The aligned and trimmed sequences were used to generate a Neighbor-joining tree (Figure 2) using the Maximum Composite Likelihood algorithm in Mega [85].
http://pubMLST.org/helicobacter isolate ids:1472–1527.
|
10.1371/journal.ppat.1004087 | Noncanonical Role for the Host Vps4 AAA+ ATPase ESCRT Protein in the Formation of Tomato Bushy Stunt Virus Replicase | Assembling of the membrane-bound viral replicase complexes (VRCs) consisting of viral- and host-encoded proteins is a key step during the replication of positive-stranded RNA viruses in the infected cells. Previous genome-wide screens with Tomato bushy stunt tombusvirus (TBSV) in a yeast model host have revealed the involvement of eleven cellular ESCRT (endosomal sorting complexes required for transport) proteins in viral replication. The ESCRT proteins are involved in endosomal sorting of cellular membrane proteins by forming multiprotein complexes, deforming membranes away from the cytosol and, ultimately, pinching off vesicles into the lumen of the endosomes. In this paper, we show an unexpected key role for the conserved Vps4p AAA+ ATPase, whose canonical function is to disassemble the ESCRT complexes and recycle them from the membranes back to the cytosol. We find that the tombusvirus p33 replication protein interacts with Vps4p and three ESCRT-III proteins. Interestingly, Vps4p is recruited to become a permanent component of the VRCs as shown by co-purification assays and immuno-EM. Vps4p is co-localized with the viral dsRNA and contacts the viral (+)RNA in the intracellular membrane. Deletion of Vps4p in yeast leads to the formation of crescent-like membrane structures instead of the characteristic spherule and vesicle-like structures. The in vitro assembled tombusvirus replicase based on cell-free extracts (CFE) from vps4Δ yeast is highly nuclease sensitive, in contrast with the nuclease insensitive replicase in wt CFE. These data suggest that the role of Vps4p and the ESCRT machinery is to aid building the membrane-bound VRCs, which become nuclease-insensitive to avoid the recognition by the host antiviral surveillance system and the destruction of the viral RNA. Other (+)RNA viruses of plants and animals might also subvert Vps4p and the ESCRT machinery for formation of VRCs, which require membrane deformation and spherule formation.
| Replication of positive-stranded RNA viruses depends on recruitment of host proteins and cellular membranes to assemble the viral replicase complexes. Tombusviruses, small RNA viruses of plants, co-opt the cellular ESCRT (endosomal sorting complexes required for transport) proteins to facilitate replicase assembly on the peroxisomal membranes. The authors show a surprising role for the ESCRT-associated Vps4p AAA+ ATPase during tombusvirus replication. They show that Vps4p is recruited to and becomes a permanent member of the replicase complex through its interaction with the viral replication proteins. Also, EM and immuno-EM studies reveal that Vps4p is required for the formation of single-membrane vesicle-like structures, called spherules, which represent the sites of tombusvirus replication. The authors propose that Vps4p and other ESCRT proteins are required for membrane deformation and replicase assembly.
| Plus-stranded (+)RNA viruses replicate by assembling membrane-bound viral replicase complexes (VRCs) consisting of viral- and host-coded proteins in combination with the viral RNA template in the infected cells. Although major progress has recently been made in understanding the functions of the viral replication proteins, including the viral RNA-dependent RNA polymerase (RdRp) and auxiliary replication proteins, the contribution of many host proteins to VRC assembly is far from complete [1]–[7]. The host proteins contributing to VRC assembly likely include translation factors, protein chaperones, RNA-modifying enzymes, and cellular proteins involved in lipid biosynthesis [8]–[14]. Other host proteins, such as the ESCRT proteins, reticulons and amphiphysins could be involved in membrane deformation occurring during VRC assembly [15]–[17]. However, the actual functions of the majority of the identified host proteins involved in VRC assembly have not been fully revealed.
To assemble their VRCs, RNA viruses take control of cell membranes by interfering with intracellular lipid metabolism, protein regulation, targeting and transport [7], [18]. Viral polymerases of many (+)RNA viruses interact with membranes and build functional VRCs in spherules that are single-membrane vesicles with a narrow opening to the cytosol. Spherules form as invaginations in a variety of cell organelles [7], [18], [19]. Tubulovesicular cubic membranes, double membrane vesicles (DMV) and planar oligomeric arrays are some other classes of membranous structures that can harbor VRCs as documented in the literature [18].
TBSV is a small (+)RNA virus that has recently emerged as a model virus to study virus replication, recombination, and virus - host interactions using yeast (Saccharomyces cerevisiae) as a model host [7], [20]–[23]. Several systematic genome-wide screens and global proteomics approaches have led to the identification of ∼500 host proteins/genes that interacted with the viral replication proteins or affected TBSV replication and recombination [9], [11], [24]–[32]. Subsequent detailed analysis revealed the functions of the two viral replication proteins (i.e., p33 and p92pol), the viral RNA, the host heat shock protein 70 (Hsp70) and the eukaryotic elongation factor 1A (eEF1A), sterols and phospholipids in the assembly of the tombusvirus VRCs [30], [31], [33]–[41]. Hsp70 and eEF1A proteins have been shown to bind to the viral replication proteins [1], [33], [42]. The auxiliary p33 replication protein, which is an RNA chaperone, recruits the TBSV (+)RNA to the site of replication, which is the cytosolic surface of peroxisomal membranes [43]–[48]. The RdRp protein p92pol binds to the essential p33 replication protein that is required for assembling the functional VRC [22], [34], [45], [49], [50].
Interestingly, the genome-wide screens and a proteome-wide over-expression approach for host factors affecting TBSV replication in yeast [9], [11], [26] have led to the identification of 11 ESCRT (endosomal sorting complexes required for transport) proteins involved in multivesicular body (MVB)/endosome pathway [51], [52]. The identified host proteins included Vps27p (ESCRT-0 complex); Vps23p and Vps28p (ESCRT-I complex), Vps25p and Vps36p (ESCRT-II complex); Snf7p and Vps24p (ESCRT-III complex); Doa4p ubiquitin isopeptidase, Did2p having Doa4p-related function; Bro1p ESCRT-associated protein and Vps4p AAA+ ATPase [9]. The identification of ESCRT proteins led to a model that tombusvirus replication depends on hijacking of ESCRT proteins to the peroxisomal membrane. It has been suggested that the protection of the viral RNA is compromised within the VRC assembled in the absence of cellular ESCRT proteins. Altogether, the formation of membranous spherule-like replication structures in infected cells might require co-opted ESCRT proteins. However, the actual functions of the subverted ESCRT proteins in TBSV replication are currently unknown.
A set of 20–30 ESCRT proteins is important for the endosomal/multivesicular body (MVB) protein-sorting pathway in eukaryotic cells, which down-regulates plasma membrane proteins via endocytosis; and sorts newly synthesized membrane proteins from trans-Golgi vesicles to the endosome, lysosome or the plasma membrane [53]–[55]. The ESCRT proteins are involved in membrane invagination and vesicle formation during the MVB pathway. Defects in the MVB pathway can cause serious diseases, including cancer, early embryonic lethality and defect in growth control [53]–[56]. Also, enveloped retroviruses (such as HIV), (+) and (−)RNA viruses (such as filo-, arena-, rhabdo- and paramyxoviruses) redirect cellular ESCRT proteins to the plasma membrane, leading to budding and fission of the viral particles from infected cells [51], [52].
The MVB pathway starts with the recognition of monoubiquitinated cargo proteins in the endosome by Vps27p (ESCRT-0 complex), which serves as a signal for proteins to be sorted into membrane microdomains of late endosomes [54]–[57]. Vps27p in turn recruits Vps23-containing ESCRT-I complex and then the ESCRT-II complex, resulting in grouping the cargo proteins together in the limiting membranes of late endosomes and deforming the membranes that leads to membrane invagination into the lumen [58], [59]. Then, components of the large ESCRT-III complex are recruited from the cytosol, followed by sequential assembly of ESCRT-III monomers into helical lattice on the membrane that leads to the scission of the neck of the invaginated membrane, giving raise to vesicle budding into the lumen of endosome and to MVB formation [60], [61]. Then, Vps4p recycles the ESCRT proteins, whereas Doa4p recycles the ubiquitin, leading to budding of multiple small vesicles into the lumen [57].
The ESCRT-III and the Vps4p AAA+ ATPase together comprise a conserved membrane scission machinery. The ATP-dependent function of Vps4p is to disassemble and remove the ESCRT-III components from the membranes (i.e., recycling them back to the cytosol) [59], [62]. Vps4p is a member of the AAA+ ATPase family, which uses ATP to remodel macromolecular structures in various biological processes, such as protein disaggregation, microtubule severing and membrane fusion. The N-terminal part of Vps4p, termed MIT domain, binds to the ESCRT-III components, while the ATPase domain is involved in ATP hydrolysis. Vps4p is present as an inactive dimer in the cytosol and during activation, Vps4p likely forms a dodecamer with two parallel rings and the ESCRT-III components are likely pulled across the central hole of these rings during the Vps4p-driven recycling event [62].
Our previous works have demonstrated that tombusviruses co-opt selected components of the cellular ESCRT machinery for replication via interaction of the p33 replication protein with Vps23p (similar to Tsg101 in mammals) and Bro1p (ALIX) [15], [63]. The recruitment of these cellular ESCRT proteins to the sites of tombusvirus replication is assumed to lead to the sequential recruitment of additional selected ESCRT factors, such as ESCRT-III and Vps4p AAA+ ATPase. Indeed, in the absence of these cellular factors in yeast or the expression of dominant negative versions of these proteins in N. benthamiana host plant, tombusvirus replication is decreased by 10-to-20-fold [15], [63]. Based on the known functions of the ESCRT proteins, it was suggested that the ESCRT proteins are recruited by TBSV to aid the formation of VRCs, which require membrane deformation to induce spherule-like structures. However, the presented data do not explain how the “neck” of the spherule-like replication structure is stabilized (to maintain an opening towards the cytosol) and why the recruitment of ESCRT factors does not lead to enclosure of the replicase complex (i.e., the VRCs are not converted rapidly into vesicles via scission of the necks of the spherules that bud inside the peroxisomes). The latter event is predicted if TBSV would take advantage of the canonical functions of the ESCRT proteins, which always result in budding of the newly formed vesicles away from the cytosol [57], [59].
In this paper, we identify Vps4p AAA+ ATPase as a major host factor in TBSV replication. We show that the p33 replication protein interacts with Vps4p and three other ESCRT-III proteins. Surprisingly, we find that Vps4p is a permanent member of the tombusvirus VRCs and it also interacts with the viral RNA. EM images revealed that Vps4p is localized in a compartment also containing the viral double-stranded (ds)RNA. Altogether, we propose that the interaction of p33 and Vps4p is critical for spherule formation and efficient tombusvirus replication. These data are consistent with the model that TBSV co-opts ESCRT proteins for its replication and these ESCRT proteins play noncanonical functions in aiding VRC formation and TBSV replication.
The formation of spherule-like structures during TBSV replication on the peroxisomal membrane surfaces and the effect of various ESCRT proteins on tombusvirus replication [15], [48], suggest that some of the co-opted cellular ESCRT proteins might play noncanonical functions. To gain insights into the functions of the co-opted ESCRT proteins during tombusvirus replication, first we analyzed if p33 replication protein could interact with ESCRT-III components or the Vps4p AAA+ ATPase. The split-ubiquitin-based yeast two-hybrid assay (membrane-based MYTH assay) between the tombusvirus p33 and the yeast ESCRT-III components or Vps4p revealed strong interaction between p33 and Vps4p (Fig. 1A). This is surprising, since Vps4p is known to interact with the ESCRT-III components to recycle them from the endosome [59], [62], while recycling of the peroxisome-bound p33 to the cytosol is unlikely to happen and not yet documented. In addition, we observed good interactions between p33 and Vps2p, Vps20p, and Vps24p ESCRT-III factors (Fig. 1A). The most abundant ESCRT-III factor, namely Snf7p, whose deletion greatly affected TBSV replication [15], and Did2p interacted only weakly with p33 (Fig. 1A). Affinity-based co-purification experiments from the membrane fraction of yeast confirmed that Vps4p strongly interacted with p33 (Fig. 1B, lane 12), while the interaction of Vps2p and Vps20p with p33 was also detectable (Fig. 1B, lanes 2 and 8). Unlike in the MYTH assay, the co-purification experiments suggested strong interaction between p33 and Vps24p (Fig. 1B, lane 4). We could not co-purify Snf7p and Did2p with p33 (Fig. 1B, lanes 6 and 10). Altogether, Vps4p showed consistently the strongest interaction with the p33 replication protein and this interaction is unexpected and could play a direct role in TBSV replication.
To confirm that the interaction between the viral replication protein and Vps4p also occurs in plants, we co-expressed the Arabidopsis thaliana AtVps4 in Nicotiana benthamiana leaves together with the FLAG-tagged tombusvirus p33 replication protein (Fig. 1C). After isolation of the membrane-bound replicase from the leaves and solubilization of the membrane fraction with nonionic detergent, we FLAG-affinity purified p33, followed by Western blotting. This approach revealed co-purification of AtVps4 with the tombusvirus p33 (Fig. 1C, lane 1), while AtVps4 was missing after purification in the sample prepared from leaves lacking p33 (lane 2). Thus, similar to yeast, the tombusvirus p33 replication protein also interacts with AtVps4 in plant leaves, suggesting subversion of Vps4 for viral activities.
Since Vps4p showed strong interaction with the tombusvirus p33 replication protein, we tested if Vps4p is part of the tombusvirus VRC. Interestingly, the affinity-purified tombusvirus replicase contained Vps4p (Fig. 2A, lane 2 versus 1). This finding highlights the possibility that Vps4p is a permanent component of the VRC (i.e., not rapidly recycled), such as Hsp70 [40]. To test this, we added cyclohexamide to yeast to prevent new p33 and p92pol translation, thus formation of new VRCs. Then, we measured the level of Vps4p in the affinity-purified replicase at various time points to study if Vps4p is released from the VRCs. As a control, we used Ssa1p Hsp70, whose amount did not change within 150 min, confirming that Ssa1p remained stably associated with the existing tombusvirus VRCs (Fig. 2B, lanes 10–12). In contrast, the amount of Pex19p peroxisomal shuttle protein, which is involved in the delivery of p33 and p92pol to the peroxisomes [64] before its getting recycled to the cytosol, decreased by 60% after 150 min of incubation in the presence of cyclohexamide (Fig. 2B, lanes 2–4). Interestingly, the amount of Vps4p did not change significantly in the affinity-purified replicase preparations during this time period (Fig. 2B, lanes 6–8), suggesting that Vps4p is likely a permanent component of the tombusvirus VRCs. Obviously, this is different from the canonical role of Vps4p, which is quickly recycled from the endosomal membranes to the cytosol after the disassembly of the endosome-bound ESCRT-III structures [59], [62]. Additionally, the N-terminally truncated Vps4p carrying the ATPase domain, but lacking the MIT domain, which is responsible for interaction with the ESCRT-III proteins, was recruited to the VRCs (Fig. 2A, lane 4), suggesting unique interaction between p33 and Vps4p.
To map the binding sites in Vps4p, we used the split ubiquitin assay that revealed that the N-terminal MIT domain, which binds to the ESCRT-III proteins [59], [62], bound efficiently to the full-length p33 (construct 1–100 versus the full-length Vps4 construct 1–437, Fig. 3A). We also observed weaker, but substantial binding between the C-terminal ATPase domain and p33 (Fig. 3A). We confirmed the interaction using the C-terminal ATPase domain and compared with the full-length Vps4p in a pull down assay with p33 (Vps4-ΔMIT, Fig. 3B). The interaction between the ATPase domain and p33 is not abolished by addition of ATP (not shown). Overall, the binding of p33 to Vps4p is different from the binding between Vps4p and the cellular ESCRT-III components that target only the MIT domain and leads to the Vps4p-driven recycling of the ESCRT-III components from the membranes back to the cytosol. We suggest that the unique interaction between the p33 and Vps4p subverts Vps4p for viral replication, leading to association of Vps4p with the membrane-bound replicase, and likely altering the canonical function of Vps4p.
Detailed mapping of p33 sites interacting with the Vps4p or the ATPase domain revealed that the very C-terminus of p33, which contains the p33:p33/p92 interaction sites, is involved in binding to Vps4p (Fig. 3C–F). Binding of the ATPase domain of Vps4p to p33 mostly overlaps with that of the full-length Vps4p in this assay. Altogether, the binding between Vps4p and p33 involves unique interactions that will require high-resolution structural studies.
We have developed an EM-based assay to visualize the tombusvirus-induced spherule-like structures in yeast, which are known to form in infected plant tissues [15], [48]. EM images revealed the presence of single-membrane vesicle-like structures of ∼25–50 nm (Fig. 4A–C) that were missing in wt yeast not expressing the tombusvirus replication proteins (Fig. S1A–B). These vesicles likely represent the spherules seen in TBSV-infected plant tissues [15].
To show if Vps4p is localized to the tombusvirus VRCs, we used immuno-gold EM of yeast cells co-expressing HA-tagged Vps4p, and MT (Metal-binding protein metallothionein)-tagged p33 that was visualized by Metal-Tagging Transmission Electron Microscopy (METTEM) [65] and replicating the TBSV repRNA, which was detected through using a dsRNA-specific antibody [65] (Fig. 4D–G). These samples were processed in the absence of osmium tetroxide that would destroy most protein epitopes and would mask the 1 nm nano-clusters associated to p33-MT. We frequently observed Vps4p in the close vicinity of the dsRNA (present in the VRCs) based on using different sized gold particles for immuno-gold EM. It was difficult to detect Vps4p in the areas of the yeast cells lacking dsRNA (Fig. S2). We suggest that Vps4p is likely concentrated in the VRCs due to binding to p33, thus facilitating detection of Vps4p in yeast membranous compartments replicating TBSV.
For an adequate visualization of gold nanoclusters bound to p33-MT, cells in Fig. 4D and 4E were processed in the absence of osmium tetroxide and contrasting agents as described. Under these conditions intracellular membranes are invisible; however, some membranes can be visualized if these ultra-thin sections are stained with uranyl acetate and lead citrate (Fig. 4G–I) (see Materials and Methods). Stained cells showed that Vps4p and the viral dsRNA were surrounded by membranes (Fig. 4G–I), supporting the model that Vps4p is part of the functional membrane-bound tombusvirus VRCs.
To enhance the visualization of Vps4p, we over-expressed His6-MT-tagged Vps4p in yeast cells replicating TBSV repRNA. This approach facilitated the frequent observation of Vps4p in the close vicinity of the TBSV dsRNA by immuno-gold EM (Fig. 5). Interestingly, the spherule-containing membranous structures are frequently connected to form a complex compartment where Vps4p and dsRNA are clustered together (Fig. 5). Altogether, the co-localization (proximal location based on immuno-EM) of Vps4p and the tombusviral dsRNA in yeast cells suggests that Vps4p is recruited to the VRCs actively involved in TBSV RNA replication.
If Vps4p is part of the tombusvirus VRCs, it is likely associated with the neck structure of tombusvirus-induced spherules, which is predicted to serve as exit places for the newly synthesized (+)RNA from the VRCs. Therefore, we tested if Vps4p is in contact with the tombusviral RNA. Affinity purification of Vps4p from the membrane fraction of yeast containing the tombusvirus VRCs resulted in co-purification of the TBSV (+)repRNA (Fig. 6A, lane 3). Interestingly, affinity purification of Vta1p accessory protein, which facilitates the formation of the functional Vps4p rings with ATPase function from the inactive Vps4p dimers [62], also resulted in co-purification of the TBSV (+)repRNA (Fig. 6A, lane 4). As a positive control, we also used the tombusvirus p33 replication protein (purified via FLAG-tag), which also resulted in co-purification of the TBSV (+)repRNA (Fig. 6A, lane 5). The Ssa1p Hsp70 chaperone, which is another permanent component of the tombusvirus VRCs also resulted in co-purified (+)repRNA, although in a lesser amount than p33 or Vps4p (Fig. 6A, lane 6 versus 3 and 5). The negative control (HF, a peptide sequence containing His6-FLAG sequence expressed from the empty expression plasmid) did not contain any detectable TBSV (+)repRNA, excluding that the viral RNA bound nonspecifically to the affinity resin or the FLAG-antibody.
We also tested the presence of (−)repRNA (which might be present in a dsRNA form within the VRCs) in the above samples. As expected, the p33 replication protein preparation contained (−)repRNA (Fig. 6A, lane 15), while Vps4p, Vta1p, Ssa1p and the negative control samples did not contain (−)repRNA (Fig. 6A). Based on these data, we suggest that the Vps4p/Vta1p complex might be in direct contact with the TBSV (+)repRNA. This contact is unlikely via p33 replication protein since both (+) and (−)repRNAs were co-purified with p33.
To obtain evidence that Vps4p is involved in the formation of tombusvirus VRCs, we took advantage of a cell-free TBSV replication assay based on TBSV-free yeast cell-free extracts (CFE) prepared from wt or vps4Δ yeast [41]. The CFE supports the assembly of the VRCs when purified recombinant p33 and p92 and (+)repRNA are added. Micrococcal nuclease was also added for 15 min (after which it was inactivated) to the assay at various time points to test the nuclease-sensitivity of the viral repRNA within newly assembled membrane-bound VRCs (Fig. 7A). The CFE from wt yeast was able to assemble nuclease-insensitive VRCs in ∼45–60 min (Fig. 7B, lane 5). In contrast, the VRCs assembled with CFE from vps4Δ yeast produced only small amounts of repRNA in the presence of nuclease, suggesting that Vps4p is required for the assembly of the nuclease-insensitive tombusvirus VRCs in vitro. The nuclease-insensitivity of the viral RNA requires cellular membranes and ATP/GTP-dependent VRC assembly and is not due to protein coating of the viral RNA, as shown previously [40], [41].
To visualize the tombusvirus VRCs in yeast lacking Vps4p, we performed EM imaging of vps4Δ yeast replicating TBSV repRNA. Interestingly, unlike the characteristic tombusvirus-induced spherule-like structures in wt yeast (Fig. 8A and B), the membranes from vps4Δ yeast expressing the p33 and p92 replication proteins and the repRNA showed different deformations (Fig. 8C and D). These crescent-shaped structures likely represent incompletely formed spherules with wide openings containing viral replicases. Similar structures (either spherules or open crescent-shaped structures) were not visible in control yeasts not expressing the viral proteins (not shown). Similar crescent-shaped membrane-structures were also visualized by EM in vps24Δ yeast replicating TBSV repRNA (Fig. 8E), suggesting that ESCRT-III components are also likely needed for proper deformation of membranes and VRCs formation. We interpret these data that the tombusvirus replication proteins could not induce the formation of complete spherule-like structures in vps4Δ or vps24Δ yeasts.
To visualize the tombusvirus p33 replication protein in yeast subcelluar membranes, we used Metal-Tagging Transmission Electron Microscopy (METTEM) [66] with yeast expressing MT-tagged p33 replication protein. The MT-p33 molecules were present in elongated structures in vps4Δ yeast (Fig. 9A), while they could form round vesicle-like structures in wt yeast (Fig. 9B). These data are consistent with the model that Vps4p is required for the formation of tombusvirus-induced spherule-like structures in yeast cells. Labeling with anti-dsRNA antibodies revealed weak to moderate signals in vps4Δ yeast (Fig. 9C). These results suggest that, in the absence of Vps4p, the abnormal VRCs are still able to support some repRNA replication. However, the level of repRNA replication is low as shown in the CFE-based replication assay (∼75% less repRNA replication in this strain than in wt) [15]. Thus, replication of repRNA is inefficient in vps4Δ yeast and the repRNAs is less protected in wide open VRCs and much more sensitive to degradation.
Co-opted host proteins, such as Hsp70 and eEF1A, play important roles in the assembly of the tombusvirus VRCs in vitro, in yeast and plant cells [30], [46], [63]. We have also shown previously that TBSV recruits the cellular ESCRT proteins via binding to Vps23p ESCRT-I protein and Bro1p ESCRT-accessory protein, likely leading to subversion of the cellular ESCRT machinery consisting of additional ESCRT-I and ESCRT-III components and Vps4p AAA+ ATPase [15], [63]. Accordingly, deletion of not only VPS23 or BRO1, but several other ESCRT genes in yeast or over-expression of dominant-negative mutants in plant leaves led to 10-to-20-fold reduction in TBSV RNA accumulation [15]. Therefore, a model was proposed that the cellular ESCRT proteins are involved in membrane bending/invagination and viral spherule formation during the assembly of the membrane-bound tombusvirus replicase complexes. However, the recruitment of the ESCRT machinery to the membrane is expected to lead to pinching off the vesicle (scission of the membrane and closure of the neck structure) into the lumen of the organelle as shown during the virion budding of HIV or MVB formation in the cell [57]–[59]. Formation of closed vesicles inside the peroxisomes (if pinching off the spherule would take place, sending the newly formed vesicle into the organellar lumen) should be a problem for tombusviruses. This is because the closed replication vesicles inside the peroxisomes would be deprived from accessing ribonucleotides needed for robust viral RNA synthesis from the cytosol and also face difficulties to release the newly made (+)RNA progeny back to the cytosol for additional viral processes, such as new rounds of translation and replication, cell-to-cell movement and encapsidation. Therefore, it is more plausible that tombusviruses halt the membrane budding process into the organellar lumen and stabilize the neck structure as schematically shown in Fig. 10A. Accordingly, the spherule-like structures could be visualized in tombusvirus infected plant tissues [15], [48] or in yeast (this work). These spherules could have controlled communication with the cytosol [19].
But how the neck of the spherules could be stabilized to prevent the formation of luminar vesicles containing trapped VRCs inside the vesicles? Based on the results from this work, we propose that tombusviruses recruit Vps4p and incorporate it as a permanent component of the VRCs. The supporting data include: (i) the strong binding between the p33 replication protein and Vps4p (Fig. 1); (ii) co-purification of p33 and Vps4p from yeast membrane fraction containing the VRCs (Fig. 1B–2A) or from plant leaves co-expressing the homologous AtVps4 and the tombusvirus p33 (Fig. 1C); and (iii) the extended presence of Vps4p, similar to Ssa1p Hsp70 protein, in the VRCs when new VRC formation is blocked (Fig. 2B). This is unlike the cellular Pex19p shuttle protein, which is recycled from the VRCs back to the cytosol (Fig. 2B). This is an unexpected feature of Vps4p, which should be easily recycled from the membranes after the disassembly of the membrane-bound ESCRT-III structures based on its canonical function [59]. (iv) Co-purification of the viral (+)repRNA with Vps4p from the membrane-bound VRCs (Fig. 6), suggesting that Vps4p might be in contact with the viral RNA. All the above features predict noncanonical functions of Vps4p, which has a role in disassembling and recycling membrane-bound ESCRT-III structures to the cytosol as ESCRT-III protein monomers [57]–[59], [62].
We also found Vps4p in the close proximity of the viral dsRNA using immuno-gold EM (Figs. 4–5). The production of dsRNA likely occurs after the assembly of the VRCs, thus Vps4p is predicted to remain a component of the assembled VRCs. Interestingly, we also observed association of Vps4p with the TBSV(+)repRNA from the membrane fraction of yeast (Fig. 6). Based on these data, we propose that one of the roles of Vps4p could be the stabilization of the neck structure of the spherule preventing scission of the membranes. Vps4p might have additional function(s), such as a role in selective release of the viral (+)repRNA progeny from the replicase during replication using the AAA+ ATPase activity of Vps4p. Additional, high resolution images will be needed to demonstrate this intriguing possibility. We also show the association of Vta1p accessory protein, which facilitates the formation of the functional Vps4p rings with ATPase function from the inactive Vps4p dimers [62], with the viral (+)repRNA (Fig. 6). Although further experiments will be needed to address the role of Vta1p during tombusvirus replication, the intriguing possibility is that Vps4p and Vta1p are part of the tombusvirus replicase and have functional roles.
In addition to Vps4p, some ESCRT-III proteins could also be involved in the putative stabilization of the neck structure, since we find interaction between p33 replication protein and several ESCRT-III factors, such as Vps24p, Vps2p and Vps20p (Fig. 1B). The interactions between p33 and the selected ESCRT-III components could prevent the membrane scission function of ESCRT-III, especially if Vps4p is also recruited by TBSV for noncanonical activities as suggested above. Accordingly, incomplete, crescent-like membrane structures were observed in vps24Δ yeast (lacking the critical Vps24p ESCRT-III component), suggesting the absence of neck-like structures in these membrane deformations. However, it is possible that the ESCRT-III (and ESCRT-I) components are co-opted by tombusviruses for assisting membrane-bending around the replicase, which should occur prior to the neck formation/stabilization.
The highly conserved Vps4p likely plays a similar role in tombusvirus replication in plants. Accordingly, we co-purified AtVps4 with the tombusvirus p33 replication protein from membrane fractions from Nicotiana benthamiana leaves (Fig. 1C). Moreover, we have previously found that expression of a dominant-negative mutant of AtVps4 in N. benthamiana leaves blocked tombusvirus replication and the formation of characteristic spherules [15]. Thus, the emerging picture is that the role of Vps4p and the ESCRT machinery is to aid building membrane-bound VRCs, which are nuclease-insensitive to avoid the recognition by the host antiviral surveillance system and the destruction of the viral RNA. It is likely that other (+)RNA viruses of plants and animals could subvert Vps4p and the ESCRT machinery for formation of VRCs, which require membrane deformation and spherule formation.
In this paper, we document a novel, noncanonical role for the Vps4p AAA+ ATPase during TBSV replication. We propose that tombusviruses not only recruit the cellular ESCRT machinery for assembling the membrane-bound VRCs, but the virus could alter the activities of Vps4p and the ESCRT-III proteins to create new functional structures (spherules acting as VRCs) and possibly new activities. The recruitment of Vps4p and additional ESCRT proteins are needed for the assembly of the replicase complex, which could help the virus evade recognition by the host defense surveillance system and/or prevent viral RNA destruction by the gene silencing machinery.
To study co-purification of host ESCRT proteins with p33 replication protein, the yeast strain BY4741 was transformed with plasmids pGBK-FLAGp33-CUP1 (or pGBK-His33-CUP1 as control, see plasmids used in the Supplementary material) plus the pYES plasmids expressing 6×His-tagged ESCRT proteins from the GAL1 promoter (see supplementary Material and Methods S1 for additional information). The transformed yeasts were pre-grown in SD minimal media plus 2% glucose for 20 h at 29°C, then transferred to SD minimal media plus 2% galactose for 16 h at 29°C to induce over-expression of ESCRT proteins from the GAL1 promoter. Then, the cultures were supplemented with 50 µM CuSO4 to induce expression of FLAG-p33 or His6-p33 from the CUP1 promoter. The yeasts were collected by centrifugation after 8 h, washed in phosphate buffer saline (PBS) and then incubated in PBS plus 1% formaldehyde for 1 h on ice to cross-link proteins. Afterwards, the formaldehyde was quenched by adding glycine to a final concentration of 0.1 M. Then, yeasts were washed in PBS and processed to purify the FLAG-tagged p33 protein as described [31]. The FLAG purified fractions were eluted from the FLAG M2-agarose columns with SDS-PAGE loading buffer (without 2-mercaptoethanol). Then the eluted fractions were supplemented with 2-mercaptoethanol (5%) and boiled for 30 minutes to reverse cross-linking.
To study co-purification of His6-Vps4p and His6-Vps4101–437, the yeast strain BY4741 was transformed with plasmids pGBK-FLAGp33-CUP1/DI72-GAL1 (or pGBK-Hisp33-CUP1/DI72-GAL1 as control), pGAD-His92-CUP1 [67] and pYC-NT-VPS4 or pYC-NT-vps4101–437. Transformed yeasts were pre-grown in SD minimal media plus 2% glucose for 20 h at 29°C, then transferred to SD minimal media plus 2% galactose and 50 µM CuSO4 for 24 h at 29°C. The yeasts are then treated with 1% formaldehyde as above and processed for FLAG-p33 protein purification.
To analyze the time dynamics of host proteins association with p33, BY4741 was transformed with plasmids pGBK-FLAGp33-CUP1/DI72-GAL1 (or pGBK-Hisp33-CUP1/DI72-GAL1 as control), pGAD-His92-CUP1 [67] and either pYES-PEX19, pYES-VPS4 or pYES-SSA1. Transformed yeasts were pre-grown in SD minimal media with 2% glucose for 20 h at 29°C and then transferred to media with 2% galactose for 24 h at 29°C. The cultures were supplemented with 50 µM CuSO4 for 2.5 h to induce expression of the viral proteins. Then cycloheximide was added (100 µg/ml) to stop protein translation and samples were taken immediately (time 0), 60 and 150 minutes afterwards. Yeast cultures were treated with formaldehyde and processed for FLAG-p33 purification as above.
Purified FLAG-p33 was detected by western blot using anti-FLAG antibody followed by anti-mouse antibody conjugated to alkaline phosphatase. Co-purified His6-tagged host proteins were detected with anti-His antibody followed by anti-mouse antibody conjugated to alkaline phosphatase. Detection was performed with NBT and BCIP.
The pMAL c2x-derived plasmids described in the Supplementary materials were transformed into Epicurion Bl21-codon-plus (DE3)-R1L cells (Stratagene). Expression of MBP-tagged proteins was induced by IPTG as described [68]. pGEX-His-VPS4 and pGEX-His-vps4101–437 were also transformed into Epicurion Bl21-codon-plus (DE3)-R1L cells. Expression of GST-His6-tagged Vps4p and Vps4101–437 was essentially done as described [68] with the exception that cultures were incubated at 23°C during IPTG treatment. The cells were broken by sonication as described [68]. The lysates were incubated with GST resin (Novagen) for 1 h at 4°C. The GST resin was washed four times with column buffer [68] and then incubated for 3 h at 21°C with column buffer plus 1 mM 2-mercaptoethanol + 1 mM CaCl2 + 1 U Thrombine (Novagen) to cleave the His6-Vps4p and His6-Vps4101–437 from the column-bound GST protein. The amylose columns containing the MBP or MBP-tagged p33 portions were then incubated with 15 µg of the purified His6-Vps4p or His6-Vps4101–437 for 1 h at 4°C. The columns were then washed five times with column buffer and the MBP-tagged proteins were eluted with column buffer plus 10 mM maltose. The MBP-tagged proteins were analyzed by SDS-PAGE electrophoresis followed by coomassie staining. Bound His6-Vps4p or His6-Vps4101–437 were detected with anti-His antibody followed by alkaline phosphatase- conjugated anti-mouse and NBT/BCIP.
The yeast membrane two-hybrid assay, based on the split-ubiquitin system (Dualsystems) has been described before [31]. The plasmid pGAD-BT2-N-His33 was co-transformed with pPR-N-RE derived constructs into the reporter yeast strain NMY51. Transformed yeasts colonies were suspended in 100 µl of water and serially diluted (10-fold) in water. 6 µl of each dilution were spotted onto TLHA- plates, to score for interaction, or TL- plates, as growth controls.
The yeast strain BY4741 was transformed with plasmids pGBK-Hisp33-CUP1/DI72-GAL1, pGAD-His92-CUP1 and either pYC-HF, pYC-HF-VPS4, pYC-HF-VTA1 or pYC-HF-SSA1. Additionally yeast was transformed with pGBK-FLAGp33-CUP1/DI72-GAL1 and pGAD-His92-CUP1. Transformed yeasts were pre-grown in SD minimal media with 2% glucose for 20 h at 29°C, then transferred to SD minimal media with 2% galactose for 24 h at 29°C. The cultures were then supplemented with 50 µM CuSO4 for 3 h to induce expression of viral proteins. The yeasts were collected by centrifugation and subjected to formaldehyde cross-linking and FLAG purification as described above. The FLAG purified fractions, eluted in SDS-PAGE loading buffer, were treated with phenol/chloroform and ethanol precipitated to recover co-purified RNA. For detection of DI-72(+) RNA, RT reactions were done with SuperScript II (Life Technologies) and primer #22 (GTAATACGACTCACTATAGGGCTGCATTTCTGCAATGTTCC), followed by PCR with primers #1165 (AGCGAGTAAGACAGACTCTTCA) and #927 (TAATACGACTCACTATAGG). For detection of DI-72(−) RNA, the primer used for RT was #18 (GTAATACGACTCACTATAGGAGAAAGCGAGTAAGACAG), followed by PCR with primers #1190 (GGGCTGCATTTCTGCAATG) and #927.
The yeast strains BY4741 and vps4Δ::hphNT1 were transformed with plasmids pGBK-FLAGp33-CUP1 and pGAD-FLAGp92-CUP1 [67]. Transformed yeasts were grown and cell free extracts prepared as described [49] except that the cultures were supplemented with 50 µM CuSO4 1.5 h before harvesting, to induce expression of p33 and p92, and incubated at 37°C for 45 min before harvesting. Replicase reactions were carried out as described [49]. RNA protection was tested by adding 1 mM CaCl2 and 50 ng micrococcal nuclease at selected time points followed by 15 min incubation and then inactivation of the nuclease by addition of 2.5 mM EGTA. All reactions were incubated for a total time of 3 h.
Yeast strains (see Supplementary materials) were pre-cultured from plated single colonies by inoculation in 2 ml of YPG (yeast extract peptone galactose) and shaking overnight at 250 rpm at 30°C. For inducing and maintaining viral replication, yeasts cells were grown for 24 h in YPG at 23°C and shaken at 250 rpm. When OD600 was around 2, cells were collected, centrifuged for 5 min at 4000 g, resuspended in TSD reduction buffer (Tris-sulfate DTT, pH 9.4) and either chemically fixed for electron microscopy (see below) or processed for removing the cell wall and obtaining spheroplasts.
For obtaining spheroplasts, yeast cells were incubated for 10 min at room temperature and treated at 30°C with 0.1 µg/µl zymolyase 20T (AMS Biotechnology) in spheroplast medium A (1× yeast nitrogen base, 2% (w/v) glucose, 1× amino acids, 1 M sorbitol, 20 mM TrisCl, pH 7.5) for 5 or 15 min, depending on the yeast strain. After zymolyase treatment, cells were centrifuged for 5 min, at 1000 g and 23°C and washed once with spheroplast medium B (1× yeast nitrogen base, 2% (w/v) glucose, 1× amino acids, 1 M sorbitol) and twice with spheroplast medium A.
For ultrastructural studies, yeast cells and spheroplasts were processed. Compared to whole yeast cells, spheroplasts are well infiltrated with fixatives and resins allowing an optimal visualization of intracellular compartments. Cells were first fixed for 20 min in suspension at room temperature with 8% paraformaldehyde and 1% glutaraldehyde followed by a second fixation step of 1 h at room temperature with 4% paraformaldehyde and 0.5% glutaraldehyde in HEPES (pH 7.4); fixed cells were then processed by conventional embedding in the epoxy-resin EML-812 (Taab Laboratories) following procedures for an optimal preservation of cell endomembranes [65], [69], [70]. Cells were post-fixed for 1 h at 4°C with 1% osmium tetroxide and 0.8% potassium ferricyanide in water, washed with HEPES, and 40 min with 2% uranyl acetate at 4°C. During post-fixation, samples were protected from light. Cells were submitted to dehydration steps for 20 min each with increasing concentrations of acetone (50, 70, 90, and twice in 100%) at 4°C and incubated with acetone-resin (1∶1) with gentle agitation at room temperature. Cells were infiltrated overnight with pure resin for 1 day and polymerized at 60°C for 3 days. Ultrathin (50–70 nm) sections were collected in 300 mesh cooper grids (G300-C3, Taab) with a plastic layer of 0.25% formvar in chloroform. Then, grids were stained for 20 min with saturated uranyl acetate and for 2 min with lead citrate following standard procedures. Samples were studied in a Jeol JEM 1011 electron microscope operating at 100 kv.
For visualization of MT-tagged p33 in cells and immunogold labeling, cells were processed by embedding in the acrylic resin LRWhite following procedures for an adequate preservation of protein epitopes and optimal visualization of small nano-clusters [66]. Spheroplasts were incubated for 75 min with 0.2 mM HAuCl4 (SIGMA-ALDRICH) in spheroplast medium A. This treatment builds gold nano-clusters in MT-tagged proteins allowing detection of protein molecules in cells with high sensitivity and at molecular scale resolution [66], [71]. Cells were washed with spheroplast medium A before fixation with 4% paraformaldehyde and 0.2% glutaraldehyde in PHEM (20 mM PIPES, 50 mM HEPES, 20 mM EGTA and 4 mM MgCl2, pH 6.9) (1 h at room temperature). Cells were submitted to short dehydration steps of 10 min each in increasing concentrations of ethanol (30, 50, 70, 90 and twice with100%) at 4°C. Spheroplasts were incubated in mixtures of ethanol- LR White acrylic resin (2∶1, 2∶2, 1∶2) with gentle agitation and protected from light and embedded in 100% resin for 24 h. Samples were polymerized for 48 h at 60°C. Ultra-thin sections were collected in 300 mesh Quantifoil holey carbon grids (R 3.5/1 Cu/Rh, Quantifoil Micro Tools) and studied without staining.
For immunogold labeling sections of cells embedded in LR White acrylic resin were processed as described [70]. Briefly, sections were incubated for 6 min with 1% BSA (Bovine serum albumin) in PBS, with primary antibodies diluted in 1% BSA and with secondary antibodies conjugated with 5 or 10 nm colloidal gold particles (from BB International) and diluted in 1% BSA. Dilutions of antibodies were rabbit anti-HA 9110 from ABCAM (1∶200) and mouse anti-dsRNA MAb J2 from English & Scientific Consulting (1∶200). Secondary antibodies were diluted 1∶40.
|
10.1371/journal.pcbi.1005677 | Bow-tie signaling in c-di-GMP: Machine learning in a simple biochemical network | Bacteria of many species rely on a simple molecule, the intracellular secondary messenger c-di-GMP (Bis-(3'-5')-cyclic dimeric guanosine monophosphate), to make a vital choice: whether to stay in one place and form a biofilm, or to leave it in search of better conditions. The c-di-GMP network has a bow-tie shaped architecture that integrates many signals from the outside world—the input stimuli—into intracellular c-di-GMP levels that then regulate genes for biofilm formation or for swarming motility—the output phenotypes. How does the ‘uninformed’ process of evolution produce a network with the right input/output association and enable bacteria to make the right choice? Inspired by new data from 28 clinical isolates of Pseudomonas aeruginosa and strains evolved in laboratory experiments we propose a mathematical model where the c-di-GMP network is analogous to a machine learning classifier. The analogy immediately suggests a mechanism for learning through evolution: adaptation though incremental changes in c-di-GMP network proteins acquires knowledge from past experiences and enables bacteria to use it to direct future behaviors. Our model clarifies the elusive function of the ubiquitous c-di-GMP network, a key regulator of bacterial social traits associated with virulence. More broadly, the link between evolution and machine learning can help explain how natural selection across fluctuating environments produces networks that enable living organisms to make sophisticated decisions.
| How does evolution shape living organisms that seem so well adapted that they could be intelligently designed? Here, we address this question by analyzing a simple biochemical network that directs social behavior in bacteria; we find that it works analogously to a machine learning algorithm that learns from data. Inspired by new experiments, we derive a model which shows that natural selection—by favoring biochemical networks that maximize fitness across a series of fluctuating environments—can be mathematically equivalent to training a machine learning model to solve a classification problem. Beyond bacteria, the formal link between evolution and learning opens new avenues for biology: machine learning is a fast-moving field and its many theoretical breakthroughs can answer long-standing questions in evolution.
| Cells use networks of biochemical reactions to collect cues from the world around them, process that information internally and respond appropriately [1]. Understanding how evolution by natural selection has turned biochemical reactions into information-processing circuits remains a major challenge [2]. The intracellular secondary messenger c-di-GMP (Bis-(3'-5')-cyclic dimeric guanosine monophosphate), ubiquitous in bacteria, is a network hub lying at the core of signaling pathways with dozens of inputs and outputs. This type of network is called a bow-tie because of its shape (Fig 1A) [3]. The key feature of a bow-tie is its ability to compress multiple inputs and command multiple outputs [4]. We find bow-ties in cells that do sophisticated information processing. For instance, macrophages and dendritic cells that integrate toll-like receptor signals to decide on immune responses [5], and a neuron must integrate multiple stimuli into sequences of action potentials which it then delivers to several other neurons [6]. What is the function of the c-di-GMP bow-tie architecture in the bacterial cell?
We investigated this question in Pseudomonas aeruginosa. Like other bacteria [7], P. aeruginosa uses c-di-GMP to decide whether to stay in a place and form a biofilm, or to swarm away in search of better conditions. Biofilm formation is a social behavior in which bacteria attach to surfaces, secrete polymeric substances and form protective communities that make infections hard to treat with antibiotics [8,9]. Swarming is also a social behavior, but swarms are motile and biofilms are sessile; the two behaviors are mutually exclusive and require expressing different sets of genes [10]. A better understanding of how the P. aeruginosa cell commands biofilm and swarming behaviors could lead to anti-biofilm therapies against this major pathogen [11].
P. aeruginosa has dozens of proteins that make and break c-di-GMP. Diguanylate cyclase (DGC) proteins with GGDEF domains synthesize c-di-GMP, and phosphodiesterase (PDE) proteins with EAL or HD-GYP domains degrade c-di-GMP. DGCs and PDEs can respond to diverse stimuli such as contact with a surface or the presence of a chemical attractant. They modulate intracellular levels of c-di-GMP that then regulate expression of downstream genes [7]. According to a well-established model, when c-di-GMP levels are low the enhancer-binding protein FleQ activates flagella genes needed for swarming motility and represses extracellular matrix genes needed for biofilm formation [12,13] (Fig 1B). When c-di-GMP levels are high FleQ forms a complex with another protein, FleN, and the FleN-FleQ complex converts its function to repress flagella genes and de-repress biofilm matrix genes [14] (Fig 1C). The FleN-FleQ is therefore a c-di-GMP-responsive switch that creates an opposed co-regulation of biofilm and motility genes.
Co-regulation is efficient because P. aeruginosa cannot move and stay encased in a matrix at the same time [15], but it comes with a risk: Experimental evolution in swarming conditions selects for FleN mutants with many flagella called hyperswarmers, which are locked in a perpetual motile mode and cannot make proper biofilms [16]. This tradeoff between biofilms and swarming—a dichotomy due to their co-regulation by c-di-GMP—could be exploited in therapies against P. aeruginosa infections. However, two key obstacles remain: First, we lack systems-level understanding of the c-di-GMP network. We know reasonably well how some network components work—for example, physical contact with a solid surface stimulates the Wsp transmembrane complex to synthesize c-di-GMP [12,17]—but we know little about how they work together as a network [18]. Second, we know little about the network’s diversity across the P. aeruginosa species. The link between c-di-GMP, biofilm and swarming was repeatedly validated in isogenic mutants [19] but seems to be absent when compared across different strains [16,20]. Is the tradeoff really absent outside the laboratory, or is it buried by many genetic differences accumulated between strains since their common ancestor? Understanding how selective pressures shape the c-di-GMP network is crucial to new therapies, especially to prevent the emergence of resistance.
Here, we combined genomics, experimental evolution and mathematical modeling to elucidate the function of the c-di-GMP network. We investigated P. aeruginosa isolates from acutely infected cancer patients; this population is distinct from isolates from chronic infections, such as those formed in cystic fibrosis lungs where microbial strains already experienced long-term evolution within the host [21–25]. Against our expectations, we saw no correlation between c-di-GMP, biofilm and swarming levels. To explain these observations, we developed a mathematical model from biochemical reaction principles; we derived a mechanism of how selection across fluctuating environments can tune the c-di-GMP network analogous to machine learning. The model explains why fluctuating environments, such as natural systems and short-term infections, could select for generalist strains but stable environments, such as laboratory evolution or long-term infections, could select for specialists locked in a phenotypic mode. We then applied our knowledge to directed-evolution experiments that revealed new mutations causing loss of biofilm specialism.
We selected a cohort of 28 clinical isolates of P. aeruginosa to investigate associations between c-di-GMP and two social phenotypes—biofilm formation and swarming—that it regulates. The 28 strains originated from a diversity of sample types (blood, urine, etc.) obtained from acutely infected patients at Memorial Sloan Kettering Cancer Center (MSKCC), and belonged to a larger set of P. aeruginosa strains that—we had described before [16]—vary in their capacity for biofilm formation and swarming. To understand how the diverse levels of biofilm formation and swarming relate to c-di-GMP, we measured each strain’s bulk c-di-GMP levels from extracts obtained from dense colonies grown on Petri dishes [26]. The c-di-GMP levels varied significantly between the isolates and from those measured for the laboratory strain PA14 (Fig 2A, p<0.05). We found no association between the c-di-GMP level and the sample type (blood, urine, etc., p>0.05), and also no correlation between c-di-GMP and biofilm formation (quantified by the microtiter crystal violet assay [27]) or swarming motility (quantified by the colony area at 16 h [16], Figs 2B, 2C, S1A and S1B). The two social phenotypes also did not correlate with each other (Fig A in S2 Fig. p>0.05).
The apparent lack of correlations seemed to challenge the well-established notion that c-di-GMP imposes a tradeoff between biofilm and swarming [28,29]. Another explanation, however, was that the 28 strains, despite coming from the same hospital, might be phylogenetically diverse. P. aeruginosa may live asymptomatically with its human host until immune-compromising cancer therapy facilitates opportunistic infection [30]; if the 28 strains spanned a large phylogenetic distance, the tradeoff could be hidden by many genetic differences accumulated during their separate evolutionary histories. To clarify this issue, we sequenced the whole-genomes of the 28 MSKCC isolates and reconstructed their phylogeny (Fig 2D). We included, for reference, the publicly available genome of PA14 and those of two other well-characterized strains, PAO1 and PA7 [31]. The phylogenetic tree confirmed features seen before—PA14 and PAO1 resided in two major clades [32] and PA7 was an outlier [33]—and revealed that the 28 isolates were indeed phylogenetically diverse from each other. Interestingly, the ability to infect a specific body site was not restricted by phylogeny: isolates from different sample types were found in both the PA14 and the PAO1 clades (circle colors, Fig 2D).
We then analyzed c-di-GMP levels, biofilm levels and swarming motility in the light of the reconstructed phylogeny. The sequenced genomes revealed that the strains varied little in the number of genes predicted to be in the c-di-GMP pathway (numbers listed next to each isolate, Fig 2D). A statistical analysis of phylogenetic signal, the Moran I test [34], indicated that the c-di-GMP level had a strong phylogenetic signal (p<0.05; Fig A in S3 Fig) but biofilm and swarming had not (p>0.05; Fig B,C in S3 Fig). We then tried correlating biofilm and swarming using the method of phylogenetic generalized least squares regression (PLSR) [35]—a method that correlates two phenotypes after correcting for phylogeny (see S1 Text). PLSR showed a significant anti-correlation (Fig B in S2 Fig) which would support a tradeoff between biofilm and swarming. But the anti-correlation depended on a subset of three strains—M37351, M55212 and F30658—that were closely related and had strong phenotypic differences among them. The correlation vanished if we excluded those three strains from the analysis (Fig C in S2 Fig), which indicates that the tradeoff between biofilm formation and swarming is hard to detect across large phylogenic distances. We investigated the correlation between biofilm and swarming in three groups of closely related clinical isolates after PLSR (Fig 2D, gray shaded). The genomes in those three subgroups differ in 480, 593 and 1654 SNPs, respectively. The phylogenetically-corrected values of biofilm and swarming showed strong correlations in group I and II (Fig AB in S4 Fig) but not in group III (Fig C in S4 Fig). Other than phylogenetic distance, the correlations also depended on the phenotypic diversity observed in each groups. For example, F30658 in group II was a strong swarmer and weak biofilm-former—the opposite from the other strains of this subgroup. But all of the four strains in group III showed very similar phenotypes to each other. PLSR helped reveal the hidden correlation between biofilm and swarming, and supported that there is a tradeoff between the two co-regulated phenotypes but only among strains that are closely related and have different phenotypes.
We then investigated whether the pattern of c-di-GMP levels, biofilm and swarming observed across the entire phylogenetic tree could be explained by a few genetic variants of large effect in c-di-GMP network genes, or if explaining the pattern required many genetic variants of small effect. We used LASSO technique [36], an algorithm that searches for a small number of features to explain a set of observables (see S1 Text). We selected the smallest subset of genetic variants (the features) as we increased a penalization, λ, for including many features (see S1 Text). According to this analysis, explaining 63% of the phenotype deviance required a model with at least 18 variants in c-di-GMP network genes (Fig 2E). All variants were predicted to have low effect, since even the strongest variant would only explain 23.3% of the phenotypic deviance (Fig 2F). In summary, LASSO showed that c-di-GMP, biofilm and swarming—in addition to being uncorrelated when investigated across the entire tree—have a complex diversity that may not be explained by a small set of genetic alterations of large effect, but was more likely to result from a combination of genetic alterations of small effect.
The lack of correlations between c-di-GMP and the two social phenotypes that it commands—biofilm and swarming—raised an important question: how can the c-di-GMP network co-regulate those phenotypes and, at the same time, allow them to be uncorrelated across the phylogenetic tree? We sought to address this question with a simple theoretical model. The model considers that a bacterial cell has m biochemical sensors that can modulate intracellular c-di-GMP levels in response to environmental stimuli (Fig 3A). Each sensor is either a DGC (which synthetizes c-di-GMP) or a PDE (which degrades c-di-GMP), and we modeled their biochemical kinetics with commonly used methods (e.g. [37]):
Each DGC-based sensor synthetizes c-di-GMP (C) from its substrate—which the model assumes is non-limiting (represented by ∅)—with a basal synthesis rate, Ri,basal. The rate increases to Ri,basal + Ri when the sensor binds to a cognate stimulus Xi, which we modeled as a binary variable (Xi = 0 means the stimulus is absent, Xi = 1 means the stimulus is present). The reaction for DGC-based synthesis of c-di-GMP was therefore
∅→riCwhereri=Ri,basal+RiXi
Similarly, a PDE-based sensor degrades c-di-GMP into a product—which we assumed does not affect the relevant kinetics (again represented by ∅)—at a basal consumption rate Rj,basalC. The degradation rate goes to Rj,basalC + Rj when the sensor binds to a cognate stimulus Xj, which we also modeled as a binary variable. The reaction for PDE-based degradation of c-di-GMP was therefore
C→rj∅whererj=Rj,basalC+RjXj
Considering these two types of biomolecular reactions, we could write a differential equation for the dynamics of c-di-GMP inside the cell as a function of the detected stimuli. This equation considered q proteins of the DGC kind and l proteins of the PDE kind, such that q + l = m:
dCdt=∑i=1q(Ri,basal+RiXi)−∑j=1l(Rj,basalC+RjXj)
[Eq 1]
Then, we used the common steady-state approximation (dC/dt ~ 0) which assumes that the intracellular levels of c-di-GMP stabilize rapidly after sensing new stimuli. This approximation allowed us to write the following mass-balance equation relating the “basal decay”, “basal synthesis” and the “net responsive” rates:
(∑j=1lRj,basal⏟Basaldecay)C=∑i=1qRi,basal⏟Basalsynthesis+∑i=1qRiXi−∑j=1lRjXj⏟Netresponsive
[Eq 2]
With a simple variable substitution we arrived at an equation that determines c-di-GMP levels as a function of a vector of all stimuli sensed by the cell, X = {X1,…,Xm}:
C(X)=α+∑i=1mβiXi
[Eq 3]
Where α≡∑i=1qRi,basal∑j=1lRj,basal, βi≡Ri∑j=1lRj,basal if i is a DGC and βi≡−Ri∑j=1lRj,basal if i is a PDE.
Then, inspired by the FleN-FleQ system, we modeled how an effector module would change its activity depending on the c-di-GMP level. The inverse regulation [29] ensures bacteria express either biofilm genes or motility genes. We modeled this process using a single binary output, Y, such that when the output is Y = 0 the bacterium expresses motility genes and when Y = 1 the bacterium expresses biofilm genes. We defined an effector setpoint CSP, which is the c-di-GMP level at which FleN-FleQ switches from expressing motility genes to expressing biofilm genes. As in previous models of bow-tie networks [4] we used a smooth sigmoidal function (the logistic function) for the effector activity. The probability that a cell expresses biofilm genes depends on c-di-CMP relative to the effector setpoint:
P(Y=1|X)=logistic(C(X)−CSP)
[Eq 4]
Finally, Eq 4 could be re-written with a simple variable change:
P(Y=1|X)=logistic(β0+∑i=1mβiXi)
[Eq 5]
where β0≡∑i=1qRi,basal∑j=1lRj,basal+Rd−CSP.
This model explains how the decision to express biofilm or swarming genes could emerge from simple biochemical reactions (Fig 3B–3D, S5 Fig). Despite its simplicity, the model can describe sophisticated information processing such as conditional gene expression. For example, a network with just two sensors (m = 2), where sensor i = 1 senses mechanical contact with surfaces and sensor i = 2 senses a chemical attractant, can be tuned to form biofilm only when it senses a surface (X1 = 1) but not a chemical attractant (X2 = 0) by having its β’s optimized to express biofilm genes when X1 = 1 and X2 = 0. Importantly, the model also shows that the network behavior can be robust to changes in its biochemical components. Robustness is an important feature of biochemical networks [38]. In the c-di-GMP network this means that two different bacteria could express the same phenotype in a given environment despite having different intracellular c-di-GMP levels, as long the biochemical components were such that the values of the compounded β parameters remained unchanged. The c-di-GMP network of P. aeruginosa has potentially more than 40 DGC and PDE proteins (Fig 2D). This provides many possibilities to integrate different stimuli and regulate biofilm formation or swarming in different environments—a regulatory complexity that explains the phenotypic diversity observed among the 28 clinical isolates.
The next question is how does selection tune the c-di-GMP network depending on the environments experienced? We first sought out to investigate this question using experimental evolution with the laboratory strain PA14. In the past, we had shown that a swarming environment selected for hyperswarmer mutants with single point mutations in FleN [16]. Here, we analyzed a hyperswarmer mutant from that study—mutant FleN(V178G), from hereon called strain fleN*—to understand whether its phenotype could be explained by our model. The mutant fleN* is a poor biofilm former [16]. Its specialist-swarming phenotype could be either due to having a low level of c-di-GMP or a failure of FleN-FleQ to respond to raising c-di-GMP levels since either possibility could cause the bacterium to stay locked in motility mode. We measured c-di-GMP in fleN* and the levels were the same as in the PA14 wild-type strain (Fig 4A). This indicated that the FleN(V178G) mutation decreased the FleN-FleQ response without changing the c-di-GMP level.
To explore whether fleN* could acquire new mutations that recovered its biofilm capabilities, we put this strain under a constant selection for biofilm formation using drip-flow biofilm reactor [39] (Fig 4B). After growing biofilms for a few days (see methods) we could isolate three distinct mutants of fleN* with recovered biofilm capabilities. Two of these had mutations in the dipA gene (DipAL505R, DipAT792P, called respectively dipA*, dipA**) and one had a mutation in the wspF gene (wspFdup776-791, called wspF*). Interestingly, all three mutants had higher c-di-GMP levels than their fleN* ancestor (Fig 4A). We also confirmed—using the Congo red binding assay—that those three mutants indeed decreased their production of extracellular polymers needed for biofilm formation (Fig 4D). To summarize, all mutants had decreased swarming (a mild decrease in dipA*, dipA** and a total loss in wspF*, Fig 4B), higher c-di-GMP levels than both the wild-type and the fleN* (Fig 4A), lower expression of flagella genes (Fig 4E), higher surface attachment (Fig 4C), and higher production of extracellular matrix (Fig 4D). We cloned the dipA*, dipA** and wspF* mutations into the fleN* background and confirmed that these mutations were sufficient to increase capacity for biofilm formation and reduce swarming (S6 Fig). Clean deletions (ΔdipA and ΔwspF) caused similar changes towards more biofilm and less swarming in both the fleN* and wild-type background, indicating (i) that the mutations phenocopied loss-of-function and (ii) that they could work even in the absence of the fleN* mutation (Fig A,B,C in S7 Fig).
The raised levels of c-di-GMP suggested that the mutations in dipA*, dipA** and wspF* could be compensating for the decreased sensitivity of FleN-FleQ and allowing the bacteria to recover their biofilm formation. The two proteins encoded by the mutated genes—DipA and WspF—are however functionally very different. DipA has both a GGDEF and a EAL domain and its loss-of-function can increase biofilm formation and decrease biofilm dispersal [40]; results from a screen suggest that DipA acts as a PDE [41]. WspF does not interact with c-di-GMP directly but does so indirectly; it is a methyltransferase that de-methylates the transmembrane Wsp complex that thereafter activates the c-di-GMP synthase WspR [42]. We created double dipA*wspF* and dipA**wspF* mutants in the fleN* background to determine whether the mutations would conflict with each other (Fig D in S7 Fig).
Our evolutionary experiments produced mutants that—unlike the clinical strains—had large differences in c-di-GMP, biofilm and swarming caused by a few alleles of large effect. How does our model explain these observations? The laboratory strain PA14 is a generalist capable of both biofilm and swarming. Our model says that the interplay between the c-di-GMP level C and the FleN-FleQ setpoint CSP determines the decision to switch the phenotype. In an environment that favors motility—such as a swarming plate—c-di-GMP would stay below the effector setpoint such that C < CSP. In an environment that favors biofilm formation—such as a solid surface—c-di-GMP would raise above the setpoint such that C > CSP (Fig 3B). The fleN* hyperswarmer is a swarming specialist that forms weak biofilms despite having the same c-di-GMP level as the wild-type PA14. According to our model, the hyperswarmer has a higher setpoint, CSP′, which would lock the bacteria in motile-mode even when c-di-GMP levels raise to levels CSP′ > C > CSP (Fig 3C). The three distinct biofilm-recovery mutants dipA*, dipA** and wspF* could compensate for a higher setpoint by producing more c-di-GMP and raising its level to C′ > CSP′. Interestingly, the mutations dipA* and dipA** had milder phenotypes than wspF*; those strains where still capable of both biofilm and swarming despite having higher c-di-GMP levels, whereas wspF* lost its swarming entirely (Fig 3D). This suggests that the two dipA mutants adjusted their c-di-GMP level to regain their generalist behavior, while the wspF mutant became a biofilm specialist (S5 Fig).
The mutants evolved in the laboratory experienced strong selective pressures, and their phenotypes—caused by large-effect alleles—showed strong associations: biofilm and swarming were anti-correlated (Fig 4F, S8 Fig). The clinical strains showed weak phenotype associations and only small-effect alleles, suggesting that they had evolved under weak selection. Can our model help unite our clinical and laboratory observations? The link between small-effect alleles and weak selection, well established in evolutionary theory [43], would be difficult to test empirically: the selection experienced by the clinical isolates during their evolution occurred in the past and is now inaccessible to us. We turned to theory to investigate how the strength of selection across fluctuating environments and the architecture of the c-di-GMP network could lead to the diversity of phenotypes seen across the clinical and laboratory strains.
The bow-tie model in Eq 5—which can be derived from biomolecular reaction principles—is mathematically equivalent to the equation for a logistic regression [44], which is a discrete choice model used for classification problems in machine learning. The analogy immediately suggests that the c-di-GMP network may work as a biochemical classifier that integrates many environmental stimuli and classifies to which of the two categories—motility-favoring or biofilm-favoring—a new environment belongs. The network which gives bacteria the ability to change phenotype when they encounter a new environment results from the environmental changes, or fluctuations, experienced during their evolutionary history. Natural selection exerted in each environment works on the bacteria at the population level in a way that resembles telling bacteria—by killing them or letting them live—whether the action was favorable.
How fast the environment changes relatively to the strength at which natural selection acts on the bacterial population is a critical parameter. We call this parameter n, the effective length of the evolutionary history. In the extreme case of n = 1, selection is so strong that only the last environment matters. A value n > 1, but still small, represents a strong selection where the fittest network consistently outperformed its competitors across a small number of environments. The larger the value of n the weaker the selection in each environment, and the fittest network is the one that consistently outperformed competitors in a long series of environments.
We derived a mathematical analogy between evolution across fluctuating environments and training a logistic regression classifier to investigate how low n (strong selection) can produce specialist networks whereas high n (weak selection) favors generalists. Classifiers learn their task by training with large datasets, for example a matrix m × n of input variables X and their correct output E = (E1,…,En). The likelihood of obtaining the output Yj = Ej is P(Yj = 1|Xj) if Ej = 1, and is 1 − P(Yj = 1|Xj) if Ej = 0. This can be written P(Yj=1|Xj)Ej×(1−P(Yj=1|Xj))1−Ej for brevity. The fitting criterion in a logistic regression is that the values of β = (β0,…,βm) should maximize the likelihood of obtaining output E from input X across the n data points:
maxβ{∏j=1nπjEj(1−πj)(1−Ej)}whereπij=P(Yj=1|Xj)=logistic(β0+∑i=1mβiXijXi)
[Eq 6]
Evolution across fluctuating environments may be described in a similar way. In our case, each environment j ∈ {1,…,n} is either a motility-favoring environment, Ej = 0, or a biofilm-favoring environment, Ej = 1, and the fitness fj in each environment is the agreement between the phenotype favored Ej and the expressed phenotype Yj:
fj=πjEj(1−πj)(1−Ej)whereπij=P(Yj=1|Xj)
[Eq 7]
A classical result from evolutionary theory states that when a diverse population experiences a series of n fluctuating environments natural selection will favor the variant with the highest fitness geometric mean across the n environments [45]:
F=∏j=1nfjn
[Eq 8]
Under these conditions, the fittest network across n environments would be the one that made best use of the array of m stimuli sensed in each environment, Xj = (Xj1,…,Xjm), and expressed—to the extent possible—the right phenotype. This network is the one with β = (β0,…,βm) that maximize geometric mean fitness across the n environments:
maxβ{∏j=1nP(Yj=1|Xj)Ej×(1−P(Yj=1|Xj))1−Ejn}
[Eq 9]
which is the same as the criterion for logistic regression, because maximizing the nth-root of a quantity is the same as maximizing the quantity itself.
To summarize the analogy, a classical result of evolutionary theory [45] allowed us to conclude that the total set of m stimuli sensed during network evolution across n fluctuating environments corresponds to a m × n input matrix, X=(X1T,…,XnT), and the phenotypes favored by each of those n environments correspond to an output vector, E = (E1,…,En)T. The solution of Eq 6 and Eq 9—the set of values β that maximizes the quantities described—is the same and so natural selection across fluctuating environments is mathematically equivalent to training a machine learning classifier.
The analogy above opens the way to investigate how the size of the m × n matrix determines the fitness of a network in future environments, since it is well known in statistical learning that the size of training data determines the performance of a classifier when it encounters new input data. We carried out simulations where we considered a simple scenario: fluctuating environments that selected for either biofilm or motility, and that occurred with the same probability.
We generated the binary vectors of length n to represent the phenotype E favored in each environment (Ej = 0 representing swarming selection and Ej = 1 representing biofilm selection) and we created n × m matrices of noiseless stimuli X (Xij = 0 in a environment favoring swarming and Xij = 1 in an environment favoring biofilm) and then we swapped the values for a fraction 1 − η to add unbiased noise to the stimuli (supporting material). We then derived the analytical solution for the best network in the limit of very long evolutionary histories (n → ∞) as a function of the signal quality, η. This theoretical best network was—by definition—unbiased for biofilm or swarming since the two phenotypes were set to be equally probable. This means that the sensor activities, β1,…,βm, should all be equal (all stimuli are equally informative and should have the same weight on the network’s response), and their values should increase (the sensors should become more sensitive) with increasing signal quality η.
We then investigated how the strength of selection determined the network by calculating the network selected with finite values of n (Fig 5A). This network is the solution of fitting a logistic regression (Fig 5B). In contrast to the theoretical best, the calculated network was typically biased to either biofilm or swarming (Fig 5C). The bias was stronger for small n because it was more likely that the vector of evolutionary histories E with small length n had an overrepresentation of either biofilm or swarming. We then saw that the stronger the network bias was, the worse the fitness in future environments, E′, would be (Fig 5D). This result, while expected from statistical learning, has biological insight: it explains that strong selection, such as in our laboratory experiments, can select for specialist networks biased for biofilm or swarming. Weak selection, more likely outside the laboratory, reduces network bias and produces generalists.
We then investigated how the number of sensors in the network, m, affected fitness (Fig 5E, top). We saw—interestingly—that the future performance of a network increased with the number of sensors m, peaked at an intermediate value m ~ n/2, and then decreased for m > n/2 (Fig 5E, bottom). A network with m > n/2 had too many components and could be tuned to irrelevant features of past environments that were simply due to noise or under sampling, making it incapable of generalizing in future environments. This is related to statistical overfitting, a well-known phenomenon: the more parameters there are in a statistical model, the easier it is to overfit [46,47]. For the c-di-GMP network this means that the optimal number of sensors for a network depends on the strength of selection across fluctuating environments. A network with too few sensors (m < n/2) cannot be properly tuned and will be disfavored by natural selection. Networks with too many sensors (m > n/2), on the other hand, can be over-tuned to the past and maladapted for the future. Optimal networks have a number of sensors m ~ n/2. Consistent with statistical learning, their maximum achievable fitness is limited by the noise in the stimulus (S9 Fig) and increases with the size of the training history, n (Fig 5E).
The analogy between c-di-GMP signaling and a machine learning classifier explains that weak selection favors generalist bacteria; generalists integrate environmental stimuli and decide between biofilm and swarming according to the environmental fluctuations experienced in their evolutionary history. Evolution in strong selection, on the other hand, favors specialists. This is similar to how small data sets tend to produce biased classifiers.
In the light of our model, we sought to exploit strong selection in laboratory environments to search for genetic alterations that might bias the c-di-GMP network towards swarming motility. According to our model, mutations that improve swarming should impact biofilm formation, and could be potentially used as targets against P. aeruginosa virulence [11]. We first noted that the wspF* strain, a biofilm specialist, had a 16 base-pair insert-repeat which functioned as a reversible DNA switch [48]. This strain, when placed under strong swarming selection for longer than 24 h, generated swarming plumes made of mutants that spontaneously lost the insert (Fig 6A, S1 Movie). We repeated this swarming-plume assay with a fleN*ΔwspF strain—a biofilm-specialist that lacked the wspF gene entirely—to search for mutations that could occur elsewhere and cause the phenotype to switch from biofilm specialism back to swarming. The fleN*ΔwspF strain also generated swarming plumes when placed under swarming selection for longer than 24 h (S2 Movie), although this took longer than for the wspF* (Fig 6B; logrank test P = 0.03). Whole-genome sequencing of one plume isolate revealed a 3 bp deletion in the gene wspA(Δ857–859). This fleN*ΔwspFwspA* mutant restored the fleN* phenotype of low biofilm and hyperswarming (Fig 6C). WspA is a critical component of the surface-sensing Wsp complex (Fig 6D–6F); the switch from biofilm specialism to swarming specialism could be due to an inability of raising c-di-GMP when the bacteria touched a surface.
Having found this one Wsp-disabling mutation, we asked whether strong swarming selection applied to the fleN*ΔwspF strain could reveal new Wsp-disabling mutations every time. We repeated swarming-plume experiment 89 times and we used high-throughput sequencing to target-sequence the wspABCDER operon of plume isolates. We identified 43 new distinct mutations affecting the Wsp system: 17 deletions, 5 insertions and 21 single nucleotide variants; some of these mutations occurred multiple times (Fig 6G, Table 1 in S1 Text). All mutations caused the biofilm specialist to regain its swarming, and are therefore potential targets against P. aeruginosa biofilm formation.
Interestingly, two plume isolates apparently had no mutations in wspABCDER. We sequenced their whole genomes to search for mutations elsewhere. Both mutants had point mutations in another predicted c-di-GMP network gene, PA14_03720 (mutations D378G and E506A). This gene has a GGDEF motif but, intriguingly, a previous study had not detected an effect in biofilm or swarming in a ΔPA14_03720 mutant [41]. The point mutations that we identified in PA14_03720 thus provide an unexpected way to impact the c-di-GMP network and cause loss of biofilm specialism.
We presented empirical results and a new mathematical model that provides a new interpretation of the ubiquitous c-di-GMP network of bacteria that computes like a biochemical machine learning classifier. Our analysis of 28 P. aeruginosa clinical isolates revealed diverse levels of c-di-GMP, biofilm formation and swarming motility. The three traits were uncorrelated, and the apparent lack of associations seemed to contradict a well-known dichotomy between biofilm and swarming. Phylogenetic analysis showed evidence of a tradeoff, but only among a few closely related strains.
Explaining a significant fraction of the diversity in c-di-GMP, biofilm and swarming seen in our clinical isolates required many small-effect alleles (Fig 2E). In contrast, mutants evolved in strong-selecting laboratory conditions had large-effect mutations that caused switches from swarming specialism to biofilm specialism and back (Figs 4 and 6). Our mathematical model explains these mutations: Altering the input/output mapping of the c-di-GMP network can lock the bacteria in either biofilm or swarming mode.
We use a classical insight from evolutionary theory—that natural selection across a series of fluctuating environments favors strategies that maximize the geometric mean fitness [45]—to investigate why strains evolved under weak selection (most likely outside the laboratory) have small-effect alleles, whereas strains evolved under strong selection (as we applied in our laboratory evolutionary experiments) have large-effect mutations. We derived a mathematical equivalence between natural selection and training a logistic regression model. This analogy is based on simplifying assumptions and is valid only when the genetic variance within the population is large; in that case selection can choose from wide range of variants and pick the best one. When genetic diversity within the population is low, evolution should resemble reinforcement learning—another learning paradigm, where data is fed online. Mutations in bacteria would correspond to “suggesting” an action, and the environment would “inform” the population whether the action was favorable by killing bacteria or letting them live. Nonetheless, the simplifying assumptions allowed us to investigate the networks with maximum geometric mean fitness and gain biological intuition on the evolution of c-di-GMP. We saw that the strength of selection determines the optimal number of input sensors (Fig 5E). Our simulations also explained why networks evolved in strong selection are more likely to be biased—specialists in either biofilm or swarming. These insights helped us unify our clinical and laboratory observations.
The architecture of biochemical networks determines their function [49]. The bow-tie architecture of c-di-GMP suggests a machine learning classifier whose function is to determine, from a set of stimuli, to which of two categories an environment belongs—biofilm-favoring or motility-favoring. It is likely that some of the stimuli sensed by the c-di-GMP network will be redundant; in that case their integration would improve decision-making by averaging out noise [1]. Some stimuli, however, may be complementary; in that case their integration could enable conditional decision-making. Some of those stimuli may help bacteria determine who their neighbors are to better resist cheating—a constant threat to the stability of social behaviors, including biofilm and swarming [50]. Signal integration in a bow-tie network has therefore many advantages. The reliance on a core molecule, however, has a well-known disadvantage [3]: mutations that improve one output can impair the other output(s). Microbiologists had already noted this phenomenon [29]. The tradeoff also occurred in our experimentally evolved hyperswarmers, which lacked biofilm formation [16]. We saw it again here in the dipA and wspF mutants (Fig 4F, S8 Fig) which improved biofilm formation but decreased swarming. And we leveraged the tradeoff in the plume-isolation assay to find 45 new mutations that caused loss of biofilm specialism (Fig 6).
Our network model—simple on purpose—made several notable assumptions. First, the model assumed deterministic and steady-state biochemical reactions. The model also assumed one single c-di-GMP pool within the cell; some evidence suggests there may be many pools [51] although this is under debate [52]. Our goal, however, was to demonstrate that even a simple biochemical network could compute like a machine learning classifier. Including dynamics, stochasticity and more hidden nodes in the c-di-GMP network could add even more sophisticated computation (S10 Fig) and the network could eventually approach the performance of a deep neural network [4]. Understanding the function and evolution of such biochemical networks is where the concepts of machine learning—already a powerful tool to interpret complex biological data [53]—could help elucidate the evolution of biological systems [54].
Our results shed light on bacterial evolution in three important ways: First, they provide a mechanism of adaptation on a range of timescales, from the second to minutes involved in the swarm/biofilm decision to the timescales involved in evolution.
Second, they suggest that we may be able to estimate the evolutionary history—the number of environments that a bacterium has experienced in its evolution—from the number of sensors in a network. Our model says that well-adapted networks should have a number of sensors (m) that is proportional to the evolutionary history (n). In our simplified model, this relationship is m = n/2. If we know more about the stimuli and dynamics of a biochemical network such as c-di-GMP in P. aeruginosa (m ~ = 53), we should be able to calculate the effective size of the evolutionary history that P. aeruginosa has experienced. This analysis could be made across different species to compare their evolutionary histories and perhaps even predict future fitness.
Third, the idea of “overfitting” to past experiences suggests network weakness that we could exploit. For this application, it will be important to know when is the environment change “extremely rapid” versus “not rapid enough”. The conventional view is that most natural environments change slowly most of the time, as natural environments tend to be smooth, punctuated by rare but large change. Many laboratory settings are “not rapid enough” as well. For example, the drip flow biofilm experiments shown here were “not rapid enough” for all cells to wash away; this was on purpose so we could obtain mutants that recovered biofilm formation. The hygienic environments in hospitals are often “not rapid enough” either, and bacteria can adapt and become resistant to antibiotics. We may already be familiar with the “overfitting” idea: Almost all of our methods to kill bacteria come from knowing that bacteria “overfit” what they experienced in the past, and we need to artificially change the environment fast, such as in a sudden rise in antibiotic concentration or ultraviolet radiation, to effectively kill bacteria. We could take advantage of new knowledge to engineer combinations of environmental stimuli that bacteria never encountered before and trigger a maladapted response—for example biofilm dispersal—in a way that treats infection but prevents resistance.
See supplementary materials for additional methods details.
All strains were grown overnight in lysogeny broth (LB) at 37°C with shaking at 250 rpm. Swarming media consisted of 0.5% agar (Bacto) supplemented with 5g/L casamino acid, 1 mM MgSO4, 0.1 mM CaCl2 and 1X buffer (12 g/L Na2HPO4 (Fisher Scientific), 15 g/L KH2PO4 (Fisher Scientific) and 2.5 g/L NaCl, pH6.7) [55]. Biofilm assays were carried out in 96-well plates in 1% trypton at 25°C for 24 hours and quantified by crystal violet staining [56]. c-di-GMP measurements were obtained from colony biofilms incubated on trypton plates with 1% agar.
The P. aeruginosa clinical isolates were sequenced using PacBio by the Genomics Facility at the Icahn School of Medicine at Mount Sinai (Robert Serba, PI), the genomes were annotated by the PATRIC [57] and the LASSO regression was done with glmnet [58]. Isogenic clones of PA14 were sequenced using Illumina MiSeq platform and mutations were identified using breseq [59].
All data analysis and plotting was conducted in Matlab, except for the Moran test for phylogenetic signal determination conducted in R using package ‘adephylo’ [60]. Mathematical model was implemented in Matlab based on the logistic regression in function mnrfit.m.
|
10.1371/journal.pntd.0001759 | Research on Motor Neuron Diseases Konzo and Neurolathyrism: Trends from 1990 to 2010 | Konzo (caused by consumption of improperly processed cassava, Manihot esculenta) and neurolathyrism (caused by prolonged overconsumption of grass pea, Lathyrus sativus) are two distinct non-infectious upper motor neurone diseases with identical clinical symptoms of spastic paraparesis of the legs. They affect many thousands of people among the poor in the remote rural areas in the central and southern parts of Africa afflicting them with konzo in Ethiopia and in the Indian sub-continent with neurolathyrism. Both diseases are toxico-nutritional problems due to monotonous consumption of starchy cassava roots or protein-rich grass pea seeds as a staple, especially during drought and famine periods. Both foods contain toxic metabolites (cyanogenic glycosides in cassava and the neuro-excitatory amino acid β-ODAP in grass pea) that are blamed for theses diseases. The etiology is also linked to the deficiency in the essential sulfur amino acids that protect against oxidative stress. The two diseases are not considered reportable by the World Health Organization (WHO) and only estimated numbers can be found. This paper analyzes research performance and determines scientific interest in konzo and neurolathyrism. A literature search of over 21 years (from 1990 to 2010) shows that in terms of scientific publications there is little interest in these neglected motorneurone diseases konzo and neurolathyrism that paralyze the legs. Comparison is made with HTLV-1/TSP, an infectious disease occurring mainly in Latin America of which the clinical manifestation is similar to konzo and neurolathyrism and requires a differential diagnosis. Our findings emphasize the multidisciplinary nature of studies on these neglected diseases, which however have not really captured the attention of decision makers and project planners, especially when compared with the infectious HTLV-1/TSP. Konzo and neurolathyrism can be prevented by a balanced diet.
| The irreversible crippling diseases konzo and neurolathyrism with identical clinical symptoms occur among poor subsistence farmers in Africa and Asia. The victims are mostly illiterate and among the poorest section of the population who can only afford the cheapest food in a monotonous diet: bitter cassava roots (Manihot esculenta) in the case of konzo, or the seeds of grass pea (Lathyrus sativus) in the case of neurolathyrism. These neurodegenerations are blamed on the presence of cyanogenic glycosides in cassava and a neuro-excitatory amino acid in grass pea. Both are linked to the deficiency of methionine and cysteine in the diet. These amino acids are needed for the protection of motor neurons against oxidative stress. Both cassava and grass pea are tolerant to adverse environments and become survival foods for the poor during drought-triggered famines and during wartime. The dependency on these crops may increase with global warming. The scientific interest in these neglected diseases is low and little change has been noted over the last two decades. This is especially obvious when compared with the scientific interest in HTLV-1/TSP, a viral disease with similar clinical symptoms which occurs mainly among poor people in Latin America in similar socio-economic settings.
| Konzo and neurolathyrism are toxico-nutritional neurodegenerations that afflict many thousands of people among poor populations in low income countries, especially in Sub Saharan Africa and in the Indian Sub-continent, respectively. In both cases, the victims are mainly illiterate subsistence farmers living in remote rural areas. Those diseases are associated with high dietary exposure from improperly processed cassava roots in the case of konzo or grass pea seeds in the case of neurolathyrism. Both cassava (Manihot esculenta) and grass pea (Lathyrus sativus) are drought tolerant crops that can be survival food after a drought (the underground cassava-roots can even survive fire) [1]. Cyanogenic glycosides linamarin and lotaustralin in cassava and the neuro-excitatory amino acid β-ODAP (β-N-oxalyl-α,β-diaminopropionic acid) in grass pea are substances incriminated in the causation of konzo and neurolathyrism respectively, and to be involved in the depletion of the essential sulfur amino acids provided by the diet. Both foods are deficient in the sulfur containing amino acids methionine and cysteine. These amino acids are required for the detoxification of the cassava cyanogens by the liver enzyme rhodanese [2]. Besides ODAP, a number of anti-nutritional factors present in grass pea as in all legume seeds can also contribute to depleting the sulfur containing amino acids that are essential for the maintenance of the redox homeostasis in the body [3]. The depletion of these amino acids in the body by prolonged consumption of cassava roots or grass pea seeds as staple contributes to oxidative stress.
Clinically, both diseases cannot be differentiated as both are motor neurone diseases with symmetric spastic paraparesis of the legs, an irreversible crippling disability presenting in different stages of severity [4]. Both diseases can be prevented by means of good household food security, availability of safe water, good food practices and handling, balanced and varied diets, and good health status, factors that are in jeopardy during droughts. Fortunately, these two diseases have not yet been reported to occur in the same geographical area. If so, it would be difficult to distinguish one from the other although there are some epidemiological differences. The patient's history with the consumption of either cassava or grass pea is needed for the diagnosis. Both diseases are characterized by sudden onset, with higher incidence during the dry season for konzo or after a drought when other crops except grass pea fail for neurolathyrism. However, the most affected group is different (mainly young children and women at child-bearing age for konzo, mainly young men for neurolathyrism). From the available literature, we can hypothesize that both diseases can occur in epidemics, when three main risk factors occur simultaneously: i) extreme poverty, ii) a high degree of illiteracy and iii) the availability of bitter cassava roots in the case of konzo, or grass pea seed in the case of neurolathyrism, as the cheapest food available. In addition, heavy physical labor was also identified as an epidemiological risk factor in both diseases. The sudden occurrence of konzo in young mothers after childbirth is particularly tragic.
The poorest sections of the population are often illiterate, have little or no political voice and are the most vulnerable for konzo and neurolathyrism. As an apparent consequence, fighting these diseases is of low priority for decision makers or governments. Both diseases involve the irreversible loss of motor neurons for which there is no cure. Some researchers have tried symptomatic treatment with a muscle relaxant ‘Tolperisone’ for neurolathyrism patients. Although the mobility of the less severely affected patients improved, this treatment was not put into practice because of the high cost [5], [6]. Small scale efforts have been made on improving the methods for reducing cyanogens during post-harvest processing of cassava [7] or for reducing β-ODAP during culinary preparation of grass pea [8] but until now these methods could not prevent the occurrence of these diseases. To our knowledge no comprehensive epidemiological studies have yet been conducted or planned. There are no accurate numbers of cases since both diseases are not considered reportable by the World Health Organization (WHO). Poverty and illiteracy of the victims, the knowledge that there is no cure and remoteness of the areas of incidence make underreporting common. There are few reliable statistics on the prevalence of both diseases and at best only estimations have been made. Little attention has been given by governments in afflicted countries [2]. As a consequence, financial support for research on the etiology of these diseases and their prevention has been insufficient and erratic.
The aim of this paper is to analyze research performance and to determine scientific interest in konzo and neurolathyrism in terms of the number of quality articles that have been published over 21 years since the first scientific article [9] that recognized konzo as a distinct disease entity was published in the British journal “Brain”. We also studied the scientific interest in these diseases by analyzing the trend in citation frequency of the scientific articles by other researchers. The publication and citation performance of these toxico-nutritional diseases was compared with the ones of the infectious disease HTLV-1 associated myelopathy/Tropical spastic paraparesis (HAM/TSP). The clinical manifestation of HAM/TSP is similar to konzo and neurolathyrism and this neurological disease has been considered for differential diagnosis with konzo and neurolathyrism. The three diseases occur in similar socio-economic settings. As in konzo and neurolathyrism, environmental factors and geographic locations seem to be important in the etiology of Tropical Spastic Paraparesis [10]. Many cases of TSP are not linked to HTLV-1 infection and some authors consider konzo and neurolathyrism as variants of Tropical Spastic Paraparesis [11].
We searched the Web of Science database (http://apps.isiknowledge.com/) for all articles with the following key words: konzo, neurolathyrism, HTLV-1 associated myelopathy/Tropical spastic paraparesis in topic and in title with the current limits on timespan (from 1990 to 2010). The term ‘lathyrism’ covers both neurolathyrism and osteolathyrism. Osteolathyrism or experimental lathyrism is caused by the bone-deforming β-aminopropionitrile (BAPN) present in Lathyrus odoratus [12]. Because of this ambiguity we have screened those papers on ‘lathyrism’ and the papers dealing with the neurological affection were classified under ‘neurolathyrism’. The results were refined by subject areas, document types, publication years and times cited. No restriction was applied for language and subject areas while correction material was excluded from the document types and the year 2011 was not considered for the citation report. We accessed also The PubMed database (http://www.ncbi.nlm.nih.gov/sites/entrez?otool=ibeuglib) but some overlap was noted on the categorization of the document types and the database has a medical orientation rather than multidisciplinary as in the Web of Science. The website http://Scholar.Google.com was not included because of a great number of subjects which comprise even unpublished or non-quoted publications. Articles listed in Web of Science are classified as A1 papers and used for evaluation purposes in many scientific institutes and universities.
Data were introduced in software package Microsoft Excel 2010 and SPSS 17 to calculate average and proportions, to draw tables and graphics, to analyze regression by means of curve estimation, scatter plots with simple linear regression were displayed for time series and R-squared were calculated to measure how well the regression line approximates real data points. The independent variable was year of publication, ANOVA table was displayed and a P-value of <0.05 was considered statistically significant.
Since the first publication which recognized konzo as a distinct disease entity in 1990 till the end of December 2010, our search in the Web of Science yielded 84, 99 and 504 items published on konzo, neurolathyrism and on HAM/TSP respectively. This represents an average of 4 publications per year for konzo topics, of 5 publications for neurolathyrism and of 25 publications for HTLV-1/TSP topics. The terms konzo, neurolathyrism and HTLV-1/TSP were found in the title of 39.3%, 34.3% and 14.9% respectively of these published topics. Figure 1 shows that there was no significant linear increase over time in the total number of publications on konzo (R2 = 0.006; P = 0.749) and neurolathyrism (R2 = 0.045; P = 0.357) while publications on HTLV-1/TSP increased significantly (R2 = 0.645; P<0.05). A significant increase (P<0.05) was observed in the citations of publications on the three diseases (Figure 2).
When published items were analyzed per document type, more than 75% of these publications (84.5% for konzo; 75.8% for neurolathyrism and 78.4% for HTLV-1/TSP) were categorized as articles, about 6% for konzo and neurolathyrism and 8% for HTLV-1/TSP as reviews, 4.8% for konzo and HTLV-1/TSP and 9.1% for neurolathyrism as proceedings paper, 1.2% for konzo, 6.1% for neurolathyrism and 4.2% for HTLV-1/TSP as meeting abstract and 2% or less as letter or as note or as editorial material (Figure 3).
We compared the keywords used for konzo and neurolathyrism items. The top three of the most frequently used terms are konzo, cassava and cyanide for konzo publications and neurolathyrism, Lathyrus sativus and β-ODAP for neurolathyrism publications. Motor neurone disease is the only term used for these two clinically similar diseases among the top ten keywords (Table 1).
Concerning the subject areas of these publications, we identified 23 areas (see Figure 4) as the most frequently cited, of which 7 are related to publications on the three diseases (Clinical Neurology, Neurosciences, Pharmacology & Pharmacy, Biochemistry & Molecular Biology, Public Environmental & Occupational Health, Tropical Medicine, Medicine General & Internal), 6 to konzo and neurolathyrism (Food Science & Technology, Nutrition & Dietetics, Applied Chemistry, Toxicology, Plant Sciences, and Environmental Sciences), 3 to konzo and HTLV-1/TSP (Microbiology, Infectious Diseases and Immunology), 1 to neurolathyrism and HTLV-1/TSP (Cell Biology), 2 to konzo (Agriculture, Multidisciplinary and Sociology) and 4 to HTLV-1/TSP (Virology, Oncology, Hematology and Medicine Research & Experimental). Clinical Neurology (22.6%), Food Science & Technology (20.2%), Nutrition & Dietetics (17.9%), Neurosciences (13.1%) and Applied Chemistry (13.1%) are the main areas in konzo publications while Neurosciences (25.3%), Biochemistry & Molecular Biology (16.2%), Pharmacology & Pharmacy (16.2%), Clinical Neurology (13.1%) and Toxicology (11.1%) are the main areas in neurolathyrism publications, whereas Virology (26.2%), Immunology (25.4%), Neurosciences (17.3%), Clinical Neurology (16.5%) and Infectious Diseases (14.5%) are the main areas in the HTLV-1/TSP publications.
Our search in the Web of Science shows that over 21 years the number of items published on konzo and neurolathyrism has increased slowly but not significantly when compared to the significant increase in studies on HTLV-1/TSP. HTLV-1/TSP is less hidden from world view because it mainly occurs in Latin America, a higher profile area, whereas the others occur largely in remote parts of sub-Saharan Africa and dry, remote parts of India amongst very poor people. Even in the African countries where konzo and neurolathyrism occur, they are largely hidden diseases, which are (in the case of konzo) often considered to be due to witchcraft. They are neglected and tend even to be ignored by the authorities in these countries, perhaps because they are somehow considered as damaging the reputation of the country in which they occur. HTLV-1/TSP is intrinsically more interesting to medical researchers because it is an infectious disease that has the risk of spreading, whereas neurolathyrism and konzo are non-infectious diseases that are confined to poor and mostly illiterate subsistence farmers and hence not appealing to work on. HTLV-1 TSP has a higher percentage of reviews and notes than konzo and neurolmathyrism as shown in Figure 3 which indicates more interest and funds for research in this area.
This also reflects the low level of interest by political authorities as well as by institutes sponsoring research in these neglected diseases konzo and neurolathyrism which are known to be non-infectious diseases affecting only poor populations in remote rural areas. These diseases do not really capture the attention of decision makers and project planners to make efforts to control these diseases while there is a growing interest and concern in the infectious HTLV-1/TSP.
During the same period there is an increase of the number of citations to published items on the three diseases, potentially indicating that those diseases are becoming important in terms of comparison with neurodegenerative diseases affecting more affluent populations. It may be interesting to note that, not unlike the popular media, the dramatic reports get more attention: the report on a new epidemic in Ethiopia, affecting 2000 patients [13], received 3.92 citations per year; while elaboration of the risk and protective factors that may offer a key to prevention, published in the same journal [14], received far less attention (2.11 citations per year).
There are fewer meetings at international level on those three diseases, which may explain why there is less than 10% of meeting abstracts or proceedings papers. Sponsors may not be well informed or do not see the opportunities to finance projects on konzo and neurolathyrism although our findings emphasize the multidisciplinary nature of studies on these diseases.
While the two neglected diseases konzo and neurolathyrism receive a comparable low level of attention in scientific publications, there is a great difference regarding to the plants incriminated as causes of these diseases. The number of papers dealing with cassava or Manihot esculenta during the same period (5035) is almost 6 times greater than those dealing with grass pea or Lathyrus sativus (859) (Figure 5). This rapid growth in papers on cassava is undoubtedly due to its great importance as a world food source particularly in tropical Africa where it is the staple food. Also the industrial use of cassava roots for the production of chicken pellets, starch and biofuel attracts much more studies when compared with grass pea, although grass pea produces the cheapest dietary protein in arid areas of East Africa and the Indian sub-continent.
The root of bitter cassava is the staple food in konzo-affected areas and is one of the highest yielding starch crops in tropical regions and increasingly becoming an industrial product. For human consumption, a lengthy post harvest processing is necessary to remove the toxic cyanogenic metabolites. For this a stable peaceful environment is required. This processing however does not change the deficiency of the sulfur amino acids methionine and cysteine in the cassava roots. While one meal of unprocessed bitter cassava roots can be lethal and indeed sometimes suicidal, prolonged consumption of insufficiently processed roots as a staple food without variation can give rise to konzo. Konzo can be prevented by better processing of bitter cassava roots and by balancing the diet with cereals and other foodstuffs rich in sulfur amino acids such as fish that is out of reach of the very poor and during social instability. For poor subsistence farmers, cassava can be a survival food during drought and also during military conflicts. Banea et al [15] showed recently that konzo can be prevented in village people by using the wetting method on cassava flour, resulting in an extra reduction of cyanogen intake.
Grass pea is the most drought tolerant legume producing the cheapest protein, but containing a neuro-excitatory amino acid β-ODAP and can give rise to excito-toxicity under certain conditions of prolonged overconsumption, malnutrition and oxidative stress. This same neuro-excitatory amino acid β-ODAP is also identified in seeds and roots of Ginseng (Panax ginseng) [16]. In Chinese traditional medicine Ginseng root is considered a longevity promoting substance. The haemostatic compound “dencichine” extracted from Ginseng is identical to β-ODAP and has been patented as a herbal medicine [17]. Moderate daily consumption of grass pea like other legumes has no deleterious effects, and some authors even mention beneficial effects for human health [18]. However, neurolathyrism often occurs as a consequence of drought-triggered famine. It does not affect longevity or cognitive functions, but the patients become dependent on already deprived and often neglected remote rural communities. The research focus on toxic aspects such as appeared in clinical neurology, neurosciences, pharmacology and toxicology may have contributed to the toxic reputation of Lathyrus sativus, that during the era of the Egyptian pharaoh's was a royal funeral offering. Konzo and neurolathyrism are neglected diseases that occur among poor, often illiterate subsistence farmers. Such socio-economic groups are often neglected by the authorities and these medical problems ignored. Konzo and neurolathyrism can be considered the emanation of an unjust and egoistic world.
One of the risk factors in the epidemiology of konzo and neurolathyrism is the availability of cassava roots or grass pea as the cheapest food and its use as staple in the diet with little or no additional nutrients. Both crops are tolerant to adverse environment and serve as survival foods during droughts and famine and this dependency may increase with further global warming [19]. Both konzo and neurolathyrism can be linked to low concentrations of plasma methionine as a result of a dietary deficiency of this amino acid [20]. As a result, the reduced level of glutathione jeopardises the defense of motor neurons against oxidative stress. Creating market conditions for making alternative food available that is cheaper than cassava roots or grass pea with a better balance of essential amino acids can be a key strategy in the prevention of konzo and neurolathyrism. Genetic improvement of both crops should be aimed at increasing the level of the essential sulfur amino acids methionine and cysteine as long term prevention. This view is recently supported by Ethiopian researchers for the nutritional improvement of grass pea [19], while other authors also document the need for renewed impetus in grass pea research [21]. The present regulation of genetically modified food crops makes the cost for developing more healthy cassava or grass pea varieties exorbitant and contributes to the neglect of these preventable diseases.
|
10.1371/journal.ppat.1003260 | DNA Break Site at Fragile Subtelomeres Determines Probability and Mechanism of Antigenic Variation in African Trypanosomes | Antigenic variation in African trypanosomes requires monoallelic transcription and switching of variant surface glycoprotein (VSG) genes. The transcribed VSG, always flanked by ‘70 bp’-repeats and telomeric-repeats, is either replaced through DNA double-strand break (DSB) repair or transcriptionally inactivated. However, little is known about the subtelomeric DSBs that naturally trigger antigenic variation in Trypanosoma brucei, the subsequent DNA damage responses, or how these responses determine the mechanism of VSG switching. We found that DSBs naturally accumulate close to both transcribed and non-transcribed telomeres. We then induced high-efficiency meganuclease-mediated DSBs and monitored DSB-responses and DSB-survivors. By inducing breaks at distinct sites within both transcribed and silent VSG transcription units and assessing local DNA resection, histone modification, G2/M-checkpoint activation, and both RAD51-dependent and independent repair, we reveal how breaks at different sites trigger distinct responses and, in ‘active-site’ survivors, different switching mechanisms. At the active site, we find that promoter-adjacent breaks typically failed to trigger switching, 70 bp-repeat-adjacent breaks almost always triggered switching through 70 bp-repeat recombination (∼60% RAD51-dependent), and telomere-repeat-adjacent breaks triggered switching through loss of the VSG expression site (25% of survivors). Expression site loss was associated with G2/M-checkpoint bypass, while 70 bp-repeat-recombination was associated with DNA-resection, γH2A-focus assembly and a G2/M-checkpoint. Thus, the probability and mechanism of antigenic switching are highly dependent upon the location of the break. We conclude that 70 bp-repeat-adjacent and telomere-repeat-adjacent breaks trigger distinct checkpoint responses and VSG switching pathways. Our results show how subtelomere fragility can generate the triggers for the major antigenic variation mechanisms in the African trypanosome.
| Previous studies on antigenic variation in African trypanosomes relied upon positive or negative selection, yielding only cells that underwent variation. This made it difficult to define individual switched clones as independent, potentially introduced bias in the relative contribution of each switching mechanism and precluded analysis of cells undergoing switching. We show that DNA double-strand breaks (DSBs) naturally accumulate close to Trypanosoma brucei telomeres. Using the I-SceI meganuclease, we then established a system to trigger breaks in all cells in a population. The specificity, temporal constraint and efficiency of cleavage facilitated the application of a quantitative approach to dissecting subtelomeric break responses and their consequences. Accordingly, we show that the DSB-site determines probability and mechanism of antigenic switching, that DSBs can trigger switching via recombination or transcription inactivation and that a checkpoint-bypass mechanism can explain switching via VSG expression site deletion. Our results provide major new insights into the mechanisms underlying antigenic variation and provide a new model to explain how the repeats flanking VSG genes serve distinct roles in fragility and recombination. The findings are also relevant to telomeric gene rearrangements that control immune evasion in other protozoal, fungal and bacterial pathogens such as Plasmodium, Pneumocystis and Borrelia species, respectively.
| Several important parasites, including those that cause malaria and Human African Trypanosomiasis (HAT), achieve antigenic variation and evasion of the host adaptive immune response through monoallelic expression and clonal phenotypic variation of surface proteins [1], [2]. The African trypanosomes are flagellated parasitic protists of major medical and veterinary importance. They are the causative agents of HAT, and Nagana in cattle, and they proliferate in the mammalian host bloodstream. In Trypanosoma brucei, antigenic variation requires mono-telomeric expression and switching of variant surface glycoprotein genes (VSGs). It is this continuous process of allelic exclusion, transcription of only one telomeric VSG at a time in each cell, which is essential for the persistence of a chronic infection. T. brucei has long been a paradigm for antigenic variation but the molecular triggers and the mechanisms mediating VSG recombination and switching are not fully understood.
Telomeres are specialized structures that cap chromosome ends, consisting of long tracts of T2AG3-repeats in T. brucei and in human cells. T. brucei subtelomeres are the exclusive expression sites (ESs) for VSG genes [3]. One among approximately fifteen bloodstream-form ESs (BESs) is active in each cell and RNA polymerase I drives transcription at an extra-nucleolar site known as the expression site body (ESB) [4], [5], [6]. The BESs are polycistronic transcription units with promoters located up to 60 kbp from the telomere-adjacent VSG [7]. Sequencing of multiple BESs revealed a conserved arrangement, with VSGs flanked by repetitive sequences; the telomeric repeats (up to 15 kbp tracts) downstream and the 70-bp repeats (0.2–7.1 kbp tracts) upstream [7]. The minichromosomes, of which there are up to 100 copies per genome, contain additional archival, non-transcribed VSG genes flanked by telomeric repeats and 70-bp repeats. The BESs typically also encode several Expression Site Associated Genes (ESAGs), but these genes are always separated from the VSG by 70-bp repeats [7]. The single active, transcribed VSG accounts for approximately one-tenth of total cell protein, which forms a dense protective coat on each cell [8], while inactive VSG mRNAs are approximately 10,000-fold less abundant than the active VSG mRNA [9].
Antigenic variation appears to be a stochastic process, typically involving duplicative transposition and replacement of the active VSG [10], [11]. The process can also occur via loss or replacement of the entire active BES [12], [13], [14], [15] or via in situ BES switching, whereby activation of a previously silent BES is coordinated with BES inactivation, typically with no detected DNA rearrangement. The majority of archival VSGs, up to 2,000 subtelomeric genes and pseudogenes [16], [17], are not associated with BES promoters. Thus, recombination and replacement of the active VSG is required to utilize this archive for long-term immune evasion. 70-bp repeat sequences define the 5′ boundaries for VSG recombination [18] and 70-bp repeats are found upstream of most archival VSGs [17], serving as potential templates for homologous recombination; this involves gene conversion or, in the case of telomeric VSGs, break-induced replication (BIR), whereby the template is copied to the chromosome end [10]. The long 70-bp repeat tracts found at active BESs are, therefore, recombination substrates that facilitate the translocation of archival VSG genes to the transcribed telomere [19]. It has been proposed that this transcribed 70-bp repeat tract is also fragile, such that the DNA breaks that trigger antigenic variation originate here [10].
The dominant mechanism of chromosomal double-strand break (DSB) repair in T. brucei is homologous recombination [20]. RAD51-independent, microhomology-mediated end-joining (MMEJ) also operates, while non-homologous end-joining has not been detected [21]. Studies on strains lacking the RAD51 homologous strand-exchange protein [22], the RAD51-3 paralogue [23] or the RAD51-interacting protein, BRCA2 [24], indicate that each of these factors promotes VSG switching. In contrast, TOPO3α, a type 1A topoisomerase, functions with RMI1 as an anti-recombinase, suppressing BES crossovers but promoting duplicative VSG transposition through 70-bp repeat recombination [13], [14].
Despite recent progress, little is known about the subtelomeric DSBs that naturally trigger antigenic variation in T. brucei, the subsequent DNA damage responses, or how these responses determine the mechanism of VSG switching. We show that natural breaks accumulate close to the telomere in both transcribed and non-transcribed BESs. We induced DSBs at different sites within both active and silent BESs and recovered survivors for analysis, those that switch and those that don't. We find that the site of the DSB has a major impact on the DSB response and the probability and mechanism of VSG switching.
Although artificial DNA breaks between the VSG and the 70-bp repeats at the active BES enhance antigenic variation in T. brucei, the presence of natural breaks has only been mapped to the VSG-distal side of these repeats [10]. We, therefore, used ligation-mediated PCR (LM-PCR) to investigate the distribution of natural DSBs in the vicinity of the VSG221 gene, in either the active transcribed or silent state; the VSG221 locus on chromosome 6a is single-copy and hemizygous. LM-PCR involves the ligation of a specific oligonucleotide to sites of DSBs followed by amplification of products using primers specific for the ligated oligonucleotide and for the locus of interest. The PCR products, each representing a distinct DSB, are then separated on a gel and detected using an appropriate probe. LM-PCR, therefore, provides a ‘snap-shot’ of DSBs in a population of cells. We used three specific VSG221 BES primer-probe combinations to assay breaks across three distinct regions (FIG. 1A; see maps to the left-hand side of the blots in FIG. 1B). A chromosome-internal primer-probe combination was used as a control. LM-PCR assays revealed DSBs in all three subtelomeric regions and, in contrast to a previous report [10], transcription status had little impact on the number of DSBs, which were detected at a similar frequency regardless of whether the VSG was transcribed or silent (FIG. 1B). Thus, we suggest that DNA replication rather than transcription generates natural breaks.
Following a comparison of the subtelomeric regions examined, we tentatively suggest that breaks could be more frequent closer to the telomere. We detected several VSG221-flanking breaks when only 4,000 cells were sampled (FIG. 1B), meaning that the frequency of these potential antigenic variation triggers exceeds the frequency of antigenic variation by two orders of magnitude; variants arise at a rate of approximately 1×10−5 per cell division [13]. We conclude that natural subtelomeric breaks typically fail to trigger antigenic variation.
To examine the consequences of DSBs within BESs, a panel of T. brucei strains were established with a tetracycline-inducible I-SceI meganuclease gene [20] and a single I-SceI cleavage site within the active or silent VSG221 BES; I-SceI cleaves a specific 18-bp sequence and produces a single DSB. The three sites selected for integration of the I-SceI site within the active VSG221 BES (FIG. 2A) were adjacent to the BES promoter, approximately 60-kbp from the VSG (VSGpro); adjacent to the 70-bp repeats, upstream of the VSG (VSGup) or; adjacent to the T2AG3 repeats, downstream of the VSG (VSGdown). Antigenic variation is not expected following recombination and repair at a silent site, but we did want to assess the impact of transcription on DSB repair. For this purpose, we also analyzed equivalent DSBs in VSGpro and VSGdown strains with a silent VSG221 BES. Immunofluorescence analysis confirmed that >99% of cells expressed VSG221 in the ‘active-VSG221’ strains and that <0.1% of cells expressed VSG221 in the ‘silent-VSG221’ strains. We also demonstrated that the latter strains could reactivate the VSG221 BES (data not shown).
Using a combination of Southern blotting (FIG. 2B), PCR and drug-sensitivity assays for loss of expression of the break-adjacent selectable marker (data not shown, see FIG. 2A), we confirmed efficient and tightly regulated DSB-induction at the correct locus in all five strains detailed above; at least two independent assays used for each strain. The Southern blot analysis shown in Figure 2B reveals the terminal restriction fragments and the expected in vivo cleaved fragments in the active transcribed and silent VSGdown strains after 6 h of induction. Cleavage is almost complete after 24 h, as indicated by loss of the terminal restriction fragments, and we obtained similar results for the active VSGup strain (FIG. 2B). In contrast, an I-SceI site embedded within T2AG3-repeats was inaccessible (FIG. S1).
We next used a clonogenic assay to assess survival following DSBs in active and silent BESs. Cells were distributed in multi-well plates under DSB-inducing conditions and, after several days, wells with live cells were counted. Cloning efficiency averaged approximately 85% in cells with DSBs in the silent BES but was strikingly lower following DSBs in the active BES (FIG. 2C); only approximately 5% of VSGup or VSGdown cells survived. The low cloning efficiency indicates that a break at the active BES is typically lethal. This may be because transcription interferes with the DSB response or, since VSG expression is required for cell-cycle progression [25], because the DSB response interferes with VSG transcription; the DSB response does indeed interfere with transcription in mouse cells [26]. Importantly, failure to tolerate a DSB is consistent with our observation that natural DSBs far exceed instances of antigenic variation (see above). We suggest that these natural DSBs at the active BES are also typically lethal.
To explore antigenic variation following DSBs at the active transcribed VSG locus, we generated cloned DSB-survivors from the VSGpro (24 clones), VSGup (22 clones) and VSGdown (32 clones) strains. As above, the VSG221 BES was maintained in the transcribed state prior to DSB-induction, using antibiotic-selection (see FIG. 2A), which was removed immediately prior to limiting dilution cloning under DSB-inducing conditions. This ensured that each cloned survivor represented an independent DSB-repair event and, unlike previous approaches, did not require any selection for cells that had modified expression of the VSG or a BES-reporter.
Using immunofluorescence analysis, we scored for survivors that had undergone antigenic variation (FIG. 3A; example fluorescence images are shown in FIG. 4A). In the VSGpro strain, only two survivors (8%) had inactivated VSG221; in the VSGup strain, all survivors (100%) had inactivated VSG221; and, in the VSGdown strain, nine survivors (28%) had inactivated VSG221 (FIG. 3A). Thus, antigenic variation is efficiently triggered by a DSB adjacent to the 70-bp repeats, is less efficiently triggered by a DSB adjacent to the telomeric repeats and is rarely triggered by a DSB adjacent to the BES promoter. Antigenic variation in every DSB-survivor from the active VSGup strain reflects a massive increase in switch frequency at 5×10−2 switches per DSB-induced cell; this is 5,000-fold higher than the natural rate of antigenic variation, estimated at approximately 1×10−5 switches per cell, per generation [13]. As expected, analysis of 24 silent VSGpro (expressing VSG121) and 25 silent VSGdown (expressing VSGX) DSB-survivors failed to reveal any activation of the silent VSG221 gene triggered by a break within the silent BES (data not shown).
Drug-sensitivity assays confirmed that DSBs were generated in the majority of non-switched survivors from the VSGpro and VSGdown active site strains; 22/22 and 18/23 of these non-switched survivors were drug-sensitive, indicating disruption of RFP:PAC, and NPT expression, respectively (see FIG. 2A). Among non-switched VSGpro survivors, three displayed repair via MMEJ as described previously [21]. Based on a previous analysis [27], we speculated that a T2AG3-like sequence downstream of VSG221 served as a telomere-seed in the majority of non-switched VSGdown survivors, allowing for repair by de novo telomere addition. This was confirmed using PCR assays (FIG. S2A–B) and also explains continued NPT expression in five of these clones. Taken together, our results confirm the generation of DSBs in non-switched survivors and show that these breaks often fail to trigger antigenic variation when adjacent to the BES promoter or the T2AG3-repeats.
We also used a series of PCR assays, as above (see FIG. S2A), to confirm that DSBs had been generated in survivors from the silent VSGpro and VSGdown strains. From the VSGpro strain, eight survivors (33%) lost both the promoter-adjacent RFP:PAC gene and the VSG221 gene and nine (38%) lost only RFP:PAC; the remaining seven (29%) repaired within RFP:PAC (data not shown) via MMEJ [21]. From the VSGdown strain, 24 survivors (96%) retained a promoter-adjacent RFP:PAC gene, eleven (44%) retained VSG221 and only five (20%) retained NPT (data not shown). These results illustrate, consistent with the cloning-efficiency data shown in Figure 2C, how DSBs at either end of a silent BES are well-tolerated, even if they result in loss or replacement of part or all of the BES.
We next used our series of PCR assays (see FIG. S2A) to explore the DNA rearrangements associated with antigenic variation. Following a DSB adjacent to the 70-bp repeats (VSGup strain), we found that VSG221 was lost in all but one of the switched survivors (FIG. 3B, clone 15), while only one of these also lost ESAG1 (FIG. 3B, clone 9; FIG. 3D). Thus, antigenic variation typically occurred through recombination within the 70-bp repeats following a break adjacent to these repeats, as reported previously [10]. The clone that lost ESAG1 may have switched through subtelomere loss or replacement, while the clone that retained VSG221 may have switched through telomere crossover or promoter inactivation.
In striking contrast, following a DSB adjacent to the telomeric repeats (VSGdown strain), eight (89%) of the switched survivors lost ESAG1 (FIG. 3C; FIG. 3D); the only clone that retained ESAG1 had lost VSG221 indicating recombination within the 70-bp repeats (FIG. 3C). We, therefore, asked whether a distal reporter adjacent to the promoter remained intact and active in the ESAG1-negative survivors; we had inserted an RFP:PAC-cassette adjacent to the BES promoter (see FIG. 2A) to monitor BES loss in the active VSGdown strain because we had previously observed BES loss following a DSB at the silent VSGdown site [28]. The analysis revealed that all eight ESAG1-negative survivors were also RFP negative by fluorescence microscopy (see FIG. 4A) and all but one of these had lost the RFP-PAC gene (FIG. 4B, FIG. S2C). We conclude that, when the DSB was adjacent to the telomeric repeats, seven of nine switched clones lost or replaced the BES; one clone underwent recombination within the 70-bp repeats and retained ESAG1 while another clone underwent recombination elsewhere within the BES and inactivated the promoter, thereby retaining RFP-PAC.
In the two survivors that switched following a DSB adjacent to the promoter (VSGpro strain), the RFP:PAC, ESAG1 and VSG221 genes were lost in one while all of these genes were retained in the other (FIG. S2D). This indicated BES loss or replacement in the first clone and promoter inactivation in the second; RFP:PAC sequencing revealed repair by MMEJ [21] in this second clone. Thus, DSBs adjacent to the 70-bp repeats trigger recombination within the 70-bp repeats; DSBs adjacent to the telomeric repeats often fail to do so, resulting in loss or replacement of the entire BES in around 25% of survivors, and DSBs at the promoter only rarely bring about antigenic variation. We also show that a break can occasionally lead to promoter inactivation. Figure 4C shows several examples of switched clones expressing new VSGs.
VSG recombination and antigenic variation in T. brucei can occur via RAD51-dependent or RAD51-independent mechanisms [29]. These are most likely based on homologous strand-exchange and MMEJ, respectively [21]. Although T. brucei RAD51 forms sub-nuclear foci following induction of DSBs at a chromosome-internal locus [20], no significant increase in the proportion of cells with RAD51 foci was observed following induction of DSBs at BESs (FIG. 5A). This may reflect failure to accumulate RAD51 or a reduced dosage of accumulated RAD51. We therefore used a rad51 gene knockout approach in both the active VSGpro and VSGup backgrounds (FIG. 5B). Clonogenic assays, using rad51 null strains, allowed us to quantify the contribution of RAD51 to subtelomeric DSB repair and antigenic variation. The cloning efficiency of rad51-null strains is only approximately 10% prior to I-SceI induction, indicating a major defect in DNA repair in the absence of RAD51 (FIG. 5C). Following I-SceI induction, cloning efficiency was reduced further by approximately 90% (VSGpro:rad51 strain) or 70% (VSGup:rad51 strain). By comparing cloning efficiency in the VSGup-rad51 strain and the VSGup-RAD51 strain (2.3% v 6.2%; compare FIG. 5C and FIG. 2C), we see that approximately 40% of VSGup survivors are RAD51-independent. Based on significantly higher DSB-survival in the VSGup:rad51 strain compared to the VSGpro:rad51 strain (FIG. 5C), we tentatively suggest more efficient RAD51-independent repair in the VSGup strain. Among a panel of VSGup:rad51 survivors, twenty (91%) had undergone VSG switching, as determined by VSG221 immunofluorescence assay and, similar to the results in a RAD51 background, all of these had lost VSG221 and only two had lost ESAG1 (FIG. 5D). These results indicated RAD51-independent recombination within the 70-bp repeats. Thus, RAD51-independent (likely MMEJ-based) recombination makes an important contribution to antigenic variation and we suggest that it is more efficient within 70-bp repeat sequences than within non-repetitive sequences.
A common DSB response is local DNA resection, involving degradation of the 5′ strand of dsDNA to generate ssDNA with a 3′ end. The resulting ssDNA serves as a substrate for the assembly of DNA repair and recombination factors [30]. We used a series of slot-blot assays (FIG. 6A) to monitor DNA resection following induced DSBs. In these assays, specific probes are used to detect signals on native DNA and denatured DNA in parallel, revealing the presence of single-stranded regions or the sum of both single-stranded and double-stranded regions, respectively. In all strains analyzed, with breaks at active (FIG. 6B) and silent BESs (FIG. 6C), we detected local resection, typically peaking 12 h after meganuclease induction. The signal is reduced for the active VSGdown strain, but this may be due to the greater distance between the DSB and the regions probed for ssDNA, and also complete loss of the VSG221 and NPT genes in some cells (see reduced signals in the ‘d’ columns). Thus, DNA resection is a common response to DSBs within a BES. We did note, however, failure to detect resection on the DSB-distal side of the 70-bp repeats in the active VSGup strain (FIG. 6B; compare Ψ and VSG221 probes). This suggested inefficient resection through the 70-bp repeats, either due to the rapid formation of recombination intermediates or some other property of the repeat-sequence itself. This is consistent with a role for the 70-bp repeats in facilitating VSG diversification by increasing the efficiency of recombination and also in serving as a ‘buffer’ that helps to protect the rest of the BES and the chromosome from the fragile end.
We previously reported continued cell cycle progression following T. brucei telomere deletion [28] and, in contrast, activation of a G2/M checkpoint in response to a DSB at a chromosome-internal locus [20]. We speculated that a severed DSB response [31] could explain failure to use the 70-bp repeats for recombination in the VSGdown strain. We used DAPI-stained nuclear and mitochondrial (kinetoplast) DNA as cytological markers to define position in the nuclear cell-cycle [32] and to examine cell cycle checkpoint responses; specifically, cells with a single nucleus and two separated kinetoplasts (1N2K) correspond to nuclear G2. A comparison of cells following DSBs in the silent VSGdown strain or in the active VSGdown or VSGup strains, revealed an increased proportion of G2 cells only in the VSGup strain (FIG. 7A). Thus, T2AG3 repeat-adjacent DSBs, in either silent or active BESs, fail to trigger the G2/M checkpoint. This may be analogous to the anticheckpoint mediated by telomere-repeat sequences in yeast [33]. This analysis also revealed a later accumulation of post-mitotic (2N2K) cells, between 24 and 48 h after I-SceI induction, in all three strains with DBSs at the active BES (data not shown). Since VSG expression is required for progression to cytokinesis [25], later accumulation of post-mitotic cells supports the view that DSB responses interfere with local transcription [26] rather than transcription interfering with the DSB response.
Previously, it has not been possible to observe DNA damage and repair foci associated with BESs (see FIG. 5A). We recently described T. brucei γH2A, a phosphorylated form of histone H2A that accumulates at DNA repair foci in response to DNA damage [34]. Immunofluorescence microscopy was used to explore the subnuclear accumulation of γH2A foci in response to DSBs in the strains described above. Although telomere-adjacent breaks failed to trigger the G2/M checkpoint, we observed robust γH2A responses in all strains examined (FIG. 7B); I-SceI induction increased the proportion of cells with γH2A foci from approximately 20%, representing naturally occurring DNA-damage, to >50%, representing additional BES-associated breaks. We next assessed the appearance of these γH2A foci during the cell-cycle. In all cases, foci were predominantly associated with the S- and G2-phases (FIG. 7C), as described previously for natural breaks and for chromosome-internal breaks [34]. Representative images are shown in Figure 7D and reveal indistinguishable foci in the three strains presented. Thus, we conclude that γH2A foci that form in response to telomere repeat-adjacent breaks fail to signal the G2/M checkpoint but are still efficiently disassembled prior to progression to mitosis. These results are consistent with a telomere-adjacent DNA damage response that is severed after DNA resection and γH2A focus assembly but prior to the G2/M checkpoint.
We have shown that the subtelomere, within a VSG expression site in T. brucei, is fragile, displaying more breaks than seen at a chromosome-internal locus and also some evidence of increased fragility closer to the telomeric repeats. We also show that the location of a subtelomeric break has a major impact on probability and mechanism of antigenic variation. We demonstrate subtelomeric DSB responses that include DNA resection, histone modification and checkpoint activation. Notably, breaks immediately adjacent to the telomere fail to trigger a checkpoint, possibly promoting BES loss or replacement. The consequences in terms of antigenic variation, following DSBs at three distinct sites within an active VSG BES, are summarized in Figure 8A. In Figure 8B, we present a model, based on our findings, to explain how repetitive sequences flanking VSG genes cooperate to drive antigenic variation and host immune evasion.
While DSBs were estimated in ∼1% of cells in the only other report of meganuclease-induced breaks at the active BES [10], we report induction of DSBs in close to 100% of cells. The efficiency, specificity and temporal constraint of meganuclease cleavage achieved here allowed us to apply a quantitative approach to dissecting subtelomeric DSB responses and the consequences for antigenic variation. The ability to induce a defined break, in almost every cell in the population, also facilitated genetic dissection of DSB repair, and allowed for analysis both microscopically and using physical monitoring techniques. Accordingly, we assessed the contribution of RAD51 and monitored DNA-damage responses, including assembly of subnuclear repair foci and DNA resection. Importantly, we have been able to study all DSB-survivors, those that undergo antigenic variation, and those that repair the subtelomere without switching VSG expression; as far as we are aware, the first time this has been achieved. Previous studies typically relied upon positive or negative selection protocols, involving activation or inactivation of a VSG-linked drug selectable marker or the VSG itself. These approaches yielded only cells that had undergone antigenic variation, made it difficult to define individual members of a panel of switched clones as independent and potentially introduced bias in terms of the relative contribution of each switching mechanism.
Our analyses provide quantitative insights into the relationship between DSBs, subtelomeric recombination mechanisms and antigenic variation mechanisms in T. brucei. We propose a model whereby both sets of VSG-flanking repeats, telomeric and 70-bp, cooperate to bring about antigenic variation (FIG. 8B); fragility within the subtelomeric region increases the frequency of DSBs, the triggers for antigenic variation, while the 70-bp repeats, in association with archival VSG-associated repeats, facilitate recombination and replacement of the active VSG.
Our survey of the VSG221 locus suggests that natural DSBs could be more frequent closer to the telomeric T2AG3-repeats. Indeed, the subtelomeric regions of a number of cell types have been shown to be fragile and prone to frequent breakage [35]. For example, human subtelomeres are recombination hot-spots [36] and mammalian telomeres are fragile sites [37]. Subtelomeres are also unstable in the malaria parasite, P. falciparum, and undergo frequent breakage and repair [38]. Our findings now indicate that subtelomeres are also fragile in African trypanosomes.
So why are subtelomeres prone to breaks? Our results indicate fragility independent of transcription, implicating DNA replication as the source of these breaks. Indeed, replication stress and fork collapse during S-phase is likely a major source of DSBs in all eukaryotes [39]. Subtelomeric DNA, due to secondary structure or local chromatin structure, could be particularly prone to replication stress, making replication forks more likely to stall and collapse. In this regard, it is notable that an I-SceI site embedded within telomeric repeats at the active BES was not cleaved following I-SceI induction in vivo (FIG. S1), suggesting inaccessible chromatin associated with tracts of T2AG3-repeats. The apparent transition from (I-SceI) accessible to inaccessible chromatin at the T2AG3-repeat junction could present a challenge for the replication machinery to negotiate.
It has been proposed that short telomeres at the active BES are prone to breaks that increase the rate of antigenic variation [40], [41]. This cannot explain high numbers of breaks detected in our LM-PCR assays, however, since the active VSG221-associated T2AG3-tracts are in excess of 5-kbp in all of the strains used here [27, also see FIG. 2B and ]FIG. S1. The 70-bp repeats have also been proposed to be the source of frequent breaks that trigger antigenic variation [10]. Deletion of the 70-bp repeat tract at the active BES demonstrated a role for these tracts in duplicative transposition [10], [19], but these studies did not distinguish between roles in triggering breaks or in subsequent recombination. We suggest that breaks within the 70-bp repeats, or between two blocks of 70-bp repeats [10], would generate effective substrates for single-strand annealing [42], a recombination pathway which would generate a ‘repeat’ deletion, rather than lead to VSG replacement. Breaks on the VSG- and telomere-proximal side of the 70-bp repeats, on the other hand, clearly do trigger antigenic variation [10; this study].
We show that the probability of antigenic variation is highly dependent upon the site of the subtelomeric DSB at the active BES. These breaks are not well-tolerated, however, and cell death is a common outcome. Even successful repair within the active BES commonly fails to bring about antigenic variation following breaks at certain sites. These findings are consistent with the high rate of natural DSBs that we observe at the active BES, relative to antigenic variation, and suggest that cells often die or fail to switch following these natural DSBs. Lesions at the active BES are probably typically lethal because VSG expression is compromised, while genes within silent BESs are dispensable and loss of these genes is tolerated.
Our results also show that the site of a subtelomeric break has a major impact on the mechanism of antigenic variation. Subtelomeric breaks on either side of the active VSG can trigger antigenic variation but a DSB adjacent to the telomeric repeats is substantially less efficient in this regard. It is notable that a DSB within the BES can also trigger promoter inactivation. One switched survivor from the VSGpro strain underwent MMEJ and inactivated the promoter and another from the VSGdown strain inactivated the promoter and lost part of the BES. These are similar to in-situ switching events and may explain RAD51-dependent in-situ switching as reported previously [22]. Thus, in situ switching can be triggered by DSB-repair that does not substantially alter the sequence of the BES.
T. brucei TOPO3α suppresses RAD51-dependent crossovers and recombination beyond the 70-bp repeats within the BES, thereby favoring recombination within these repeats [13]. We find, consistent with previous studies [13], [22], that antigenic variation associated with 70-bp repeat-recombination involves both RAD51-dependent and independent pathways. Notably, however, our results suggest a higher rate of RAD51-independent recombination within the 70-bp repeats than observed in the BES promoter region. MMEJ is RAD51-independent and we suggest that this repair mechanism is more efficient within 70-bp repeat sequences, due to the relative abundance of potential ‘microhomologies’. Thus, recombination followed by Break-Induced Replication to the chromosome end and replacement of the active VSG could be initiated by microhomology.
Our data do not reveal differences in the DNA damage response due to BES transcription in T. brucei. Rather, they reveal a different response due to telomere-repeat proximity. We show that subtelomeric breaks trigger γH2A focus formation and DNA resection. The increase in γH2A foci in response to DSBs allowed us, for the first time, to visualize repair sites associated with VSG recombination. Notably, γH2A focus formation is associated with a G2/M cell-cycle checkpoint following DSBs upstream of the active VSG but not following breaks immediately adjacent to the telomeric repeats. These latter cells also failed to use the 70-bp repeats for recombination and, instead, underwent antigenic variation via BES loss or replacement. Failure to trigger this checkpoint following telomere-repeat-adjacent breaks was independent of the transcription status of the BES.
Telomere-associated proteins are known to repress the DNA damage response [43]. In Schizosaccharomyces pombe, a telomeric DSB-response is severed due to the absence of epigenetic marks required for cell-cycle arrest [31], and telomeric repeats also suppress the checkpoint response in Saccharomyces cerevisiae [33]. This anticheckpoint effect is thought to prevent the fusion of linear chromosomes. We propose the operation of a similar anticheckpoint in T. brucei. Our results suggest a checkpoint bypass mechanism when the break is adjacent to the telomeric repeats and the G2/M checkpoint may be required for efficient participation of the 70-bp repeats in recombination. Natural breaks adjacent to the telomeric repeats may similarly explain previous reports of BES loss or replacement [12], [13], [14], [15].
DNA DSBs are triggers for antigenic variation. Here, we probe DSB responses, BES recombination pathways and mechanisms of antigenic variation. First, we show that subtelomeres are fragile; thereby generating the DNA breaks that trigger antigenic variation. We then demonstrate VSG replacement and BES loss in response to distinct subtelomeric breaks, and also provide evidence for in situ switching as a response to subtelomeric DSBs. It is 70-bp repeat recombination that makes the major contribution to antigenic variation because most archival VSGs are flanked by these repeats and use them for gene-conversion. We suggest that breaks between the telomeric and 70-bp repeats trigger this pathway. What follows is a DNA damage response that includes DNA resection, histone modification and, depending upon the site of the break, a G2/M checkpoint. Formation of 70-bp repeat ssDNA then promotes interaction with similar templates elsewhere in the genome; these repeats may be favored substrates for recombination simply because they are highly repetitive. Recombination is then either RAD51-dependent or RAD51-independent; most probably MMEJ-based in this latter case. In conclusion, we provide novel insight into the triggers, associated DNA damage responses and mechanisms of antigenic variation in African trypanosomes. Our findings may also be relevant to subtelomeric gene rearrangements in human cells and to immune evasion mechanisms in other pathogenic protists, fungi and bacteria, such as Plasmodium sp., Pneumocystis sp. and Borrelia sp., respectively [44].
T. brucei Lister 427 cells were grown and genetically manipulated as described [28]. The strain referred to here as VSGdown-silent was described previously [28]. Puromycin or G418 selection (2 µg/ml) were used to ensure that the VSG221 BES remained active prior to I-SceI induction. I-SceI was induced using tetracycline (Tet) at 1 µg/ml (Sigma). For clonogenic assays, a mean of 0.3 to 50 cells per well were seeded in 96-well plates with or without Tet. Survivors were assessed microscopically after 5–7 days. All clones analyzed were from plates with <30% positive wells. Repaired survivors were scored for puromycin sensitivity at 1 µg/ml. DSB-survivors that displayed >99% VSG221 positive cells, as determined by immunofluorescence analysis, were scored as non-switched, while survivors that displayed >98% VSG221 negative cells were scored as switched. Proportion of 1N2K cells and cells with γH2A repair foci were counted by two of us to generate mean values ± SD.
The BES promoter-targeting constructs, pESP-RFP:PAC, pESP-RSP and pESPi-RSP were derived from pESPiRFP:PAC [28]. Briefly, the tetracycline-operator was removed from pESPiRFP:PAC to derive pESP-RFP:PAC and an I-SceI site was added to derive pESP-RSP. To insert an I-SceI site at the NotI site between the RFP and PAC genes, ‘I-SceI’ primers were annealed and ligated to give pESP-RSP. The RSP cassette replaced RFP-PAC in pESPiRFP-PAC to give pESPi-RSP. Transfections with SacI-KpnI digests of pESP-RSP or piESP-RSP were used to generate VSGpro active and silent strains, respectively. The ES-70 cassette was assembled using primers containing the I-SceI site and targeting fragments to amplify the PAC resistance cassette. The PCR product was transfected to generate VSGup strains. pTMF-Sce [28] was digested with SmaI and transfected to generate VSGdown active strains. To generate the pTMFEm construct SceHexU/SceHexL primers were annealed and ligated to SpeI/PstI digested pTelo1 (pBluescript with sixteen T2AG3-repeats at the MCS). pTMFEm was digested with SmaI and transfected to generate VSGtelo strains. RAD51 gene disruption targets were amplified by PCR from T. brucei genomic DNA, using Phusion high-fidelity DNA polymerase (New England Biolabs). The targets were assembled such that they flanked BSD or NPT selectable markers. Both constructs were digested with Acc651 and NotI prior to transfection. Details of primers/oligonucleotides are available on request.
Ligation-mediated PCR (LM-PCR) was carried out as described [10]. Briefly, DNA DSBs were detected by in-gel blunt-end linker ligation and PCR. The BES locus-specific primers were: LMPCRi (tagcagaatgcaacgtcga), LMPCRii (ttggcgactataacggctg) and LMPCRiii (ggcgttaccaagcttgttga). Slot blots for the detection of ssDNA were carried out as described [20]. Southern blotting and sequencing were carried out according to standard protocols [45]. RFP, PAC, ESAG1 [13], VSG221 and telomere-repeat-specific primers were used for the PCR assays. Other details of oligonucleotides are available on request.
Extracts of total cell protein were separated on SDS-polyacrylamide gels and stained with Coomassie-blue or subjected to western blotting using standard protocols [45]. We used rabbit anti-VSG221, rabbit anti-RAD51 [23] and an ECL+ kit (GE Healthcare). For immunofluorescence microscopy, cells were labeled using a standard protocol with rabbit anti-VSG221 rabbit anti-γH2A [34] or mouse anti-Myc (Source Bioscience), and fluorescein or rhodamine-conjugated goat anti-rabbit or anti-mouse secondary antibodies (Thermo Scientific Pierce Antibodies). RFP was detected directly. Cells were mounted in VectaShield (Vector Laboratories) containing 4, 6-diamidino-2-phenylindole (DAPI). Images were captured on an Eclipse E600 microscope (Nikon) using a Coolsnap FX (Photometrics) charged coupled device camera and processed in Metamorph 5.4 (Photometrics).
|
10.1371/journal.pcbi.1005446 | Navigating in foldonia: Using accelerated molecular dynamics to explore stability, unfolding and self-healing of the β-solenoid structure formed by a silk-like polypeptide | The β roll molecules with sequence (GAGAGAGQ)10 stack via hydrogen bonding to form fibrils which have been themselves been used to make viral capsids of DNA strands, supramolecular nanotapes and pH-responsive gels. Accelerated molecular dynamics (aMD) simulations are used to investigate the unfolding of a stack of two β roll molecules, (GAGAGAGQ)10, to shed light on the folding mechanism by which silk-inspired polypeptides form fibrils and to identify the dominant forces that keep the silk-inspired polypeptide in a β roll configuration. Our study shows that a molecule in a stack of two β roll molecules unfolds in a step-wise fashion mainly from the C terminal. The bottom template is found to play an important role in stabilizing the β roll structure of the molecule on top by strengthening the hydrogen bonds in the layer that it contacts. Vertical hydrogen bonds within the β roll structure are considerably weaker than lateral hydrogen bonds, signifying the importance of lateral hydrogen bonds in stabilizing the β roll structure. Finally, an intermediate structure was found containing a β hairpin and an anti-parallel β sheet consisting of strands from the top and bottom molecules, revealing the self-healing ability of the β roll stack.
| Silk-inspired repeated sequences, variants of the sequence from Bombyx Mori silk, have been used to make supramolecular nanotapes, pH-responsive gels, and most importantly self-assembled coat for artificial viruses. Silk-inspired repeated sequences have shown great potential as promising delivery vehicles in targeted delivery of nucleic acids for gene therapy. However, the mechanisms regarding the folding and docking of silk-inspired polypeptides remain elusive. Elucidation of the folding and docking mechanism might help us create sequences with desired self-assembly properties for many biomedical applications. An enhance sampling method, accelerated molecular dynamics (aMD) simulation, is used in this study to investigate the unfolding of a stack of two β roll molecules to shed light on the folding mechanism by which silk-inspired polypeptides form fibrils and to identify the dominant forces that keep the silk-inspired polypeptide in a β roll configuration. A template-unfolding mechanism and a neat step-wise unfolding fashion are found which agree well with experimental observations.
| An increasing number of functional proteins are reported to have β solenoid structure, such as antifreeze protein [1–3], curli [4], and carbonic anhydrase enzyme [5]. Formed by winding the peptide chain in a left-handed or right-handed fashion, a β solenoid structure usually has a repeat unit consisting of 2, 3, or 4 β strands that are connected by turns [6]. Each of these strands form parallel β sheets with their neighboring strands. As a result, a β solenoid structure usually has two parallel β sheets with the strands facing in different directions. The interior of β solenoid structures contains apolar amino acid side chains that are tightly packed [6], sometimes even interdigitated [7], resulting in a predominately hydrophobic core structure.
Polypeptides in β solenoid structures have been used as building blocks of fibrils via controlled self-assembly in biomaterials applications. They have been used to create viral capsids of DNA strands[8], supramolecular nanotapes [9] and pH-responsive gels [10], which find biomedical application as matrices for human cells [11], Beta solenoids (for short: beta rolls) can form fibrillar structures via two different mechanisms: end-to-end assembly where the terminals are covalently attached to each other, creating a very long β solenoid, or sheet-to-sheet assembly where the sheets associate physically resulting in a stack. Peralta et al. [12] used the former type of self-assembly to generate micron length amyloid fibrils from spruce budworm antifreeze protein, a modified β solenoid protein. Beun et al. used the latter type of self-assembly to produced fibrils made of stacks of pH-responsive silk-collagen-like triblocks [13].
According to experimental reports, the dimensions of fibrils consisting of stacks of Bombyx Mori silk-inspired polypeptides with a sequence of (GAGAGAGX)n, where A and G stand for alanine and glycine, respectively, X is a polar residue and n is the number of repeating units [13] are consistent with β-solenoid structures stacked on top of each other [14]. To obtain a better understanding of the detailed structure of these fibrils, we recently investigated the β-solenoid structure formed by (GAGAGAGQ)10 via conventional molecular dynamics (cMD) simulations and found that the most probable structure formed by this sequence is a β-roll structure, with all the hydrophobic alanine side chains pointing inward, as shown in Fig 1 [7]. This structure was found to be more stable than a structure reported earlier by Schor et al. where all the hydrophobic alanine side chains pointed outwards [14]. Now that the ‘ground state’ structure of the building block in the filament has been determined, we wish to learn more about how initially disordered polypeptides fold into the β-roll structure and assemble to form the filament.
Two different mechanisms have been proposed regarding the folding and docking of silk-inspired polypeptides. The first mechanism, the “template folding” (TF) mechanism, was deduced from the temporal evolution of CD spectra when the polypeptide (GAGAGAGE)n formed fibrils [13]; according to TF the peptide starts to fold into a β roll structure once it attaches to the growing end of a pre-existing filament. The experimental growth rate of filaments is very low (one molecule-per-second) and the fibril formation is irreversible. The second mechanism, “solution folding” (SF), proposed by Schor et al. [15], is based on atomistic simulations, also for (GAGAGAGE); it claims that a polypeptide first folds in solution into a β-roll structure before it docks to the growing end of a fibril via Glu-Glu side chain interactions. Both mechanisms showed up in replica exchange Monte Carlo simulations carried out by Ni et al. [16], of the folding pathways of silk-inspired polypeptides with sequence [EIAIAIAR]12 (I is isoleucin and R is Arginine). They found that at low temperature the polypeptide folds into a β-roll configuration before docking to another molecule, but at high temperature it folds after docking to another molecule. It is important to point out here that the folding pathways proposed by Schor et al. [15] and by Ni et al [16] are based on the β-roll structure predicted by Schor et al. [14], which has all the alanine side chains pointing outwards from the β-roll. However, as mentioned earlier, our recent study [7] found that the β-roll structure formed by the silk-inspired polypeptide (GAGAGAGQ)10 has a higher probability to have a hydrophobic core than a hydrophobic shell, i.e. all the alanine side chains should point inwards rather than outwards. Therefore, the folding pathway of the more probable hydrophobic core structure still remains elusive.
We use an enhanced sampling method, accelerated molecular dynamics (aMD), to study the unfolding behavior of a two-molecule β roll stack. We choose this method because conventional molecular dynamics (cMD) simulations of the folding of a peptide into a β roll are too slow. This is understandable: the experimental time scale, corresponding to the addition of a single molecule to the growing end of a fibrils is thought to take about a second [13] which is orders of magnitude away from time scales reached in MD simulations. Accelerated molecular dynamics (aMD) has been used on many different oligomers and polypeptides, including alanine dipeptide [17], bovine pancreatic trypsin inhibitor (BPT1) [18], G-protein coupled receptors [19], and streptavidin-biotin complex [20]. One of the advantages of aMD compared to other enhanced sampling methods is that it does not require pre-defined reaction coordinates. This allows simulations to explore a broader range of hypothetical kinetic pathways than would otherwise be possible. In aMD simulations of proteins, boost potentials are added to the potential energy; for proteins one can choose to boost either the total potential energy of the system, or the dihedral energy, or both. The boost potential is only added to potentials below a pre-defined threshold energy; for energies above that level the original shape is retained. aMD makes barrier crossing between low energy states easier and therefore provides access to regions of conformational space that are unreachable in cMD simulations.
The long term goal of our study is to elucidate the folding mechanism of silk-inspired polypeptide, as well to identify the dominant forces that maintain the β roll configuration. As there is no unambiguous way to choose a representative starting configuration, we decided to focus on unfolding rather than folding, which should give us clues as to the likeliness of hypothetical folding pathways. To this end, we simulate a stack of two molecules having sequence (GAGAGAGQ)10, each folded into a β roll with a hydrophobic core (Fig 1) and in explicit solvent. aMD is used to study the unfolding behavior of one of the β roll structures at systematically increased values for the threshold. We first study the system with a stack of two β roll molecules, using the lowest threshold. Then we simulate a system with a partially-fixed bottom template and a relatively high threshold; this enables us to observe the unfolding behavior of a single molecule on top of the template, and also gives us information about which forces are most important in maintaining the β roll structure. Finally, we challenge the system with partially-fixed bottom template by a further increase of the threshold to gain a better picture of the intermediate structures that form along the entire unfolding pathway.
The major findings in this paper are the following. First, we find that in the system without fixed atoms the molecules unfold in a step-wise fashion, and both molecules unfold completely by the end of the simulation. The unfolding is initiated mainly from the C terminal. Second, in the stack with a partially-fixed bottom template the top molecule has more difficulty unfolding than the molecules in the stack without any fixed atoms, indicating the importance of the template in stabilizing the folded structure of a β roll molecule in a stack. Third, there is a hierarchy of hydrogen bond strengths. Lateral hydrogen bonds formed between β strands within the β sheets in a β roll structure are stronger than the vertical hydrogen bonds within the β turns; H bonds are weaker when they are closer to the sides of the β turns. The lateral hydrogen bonds within the bottom layer in the top β roll molecule are stronger than those in the top layer, which suggests that the bottom template strengthens the hydrogen bonds within the β sheet that it contacts. As the aMD threshold on a system with partially-fixed bottom template is further increased, we find that the β roll on top tends to form hydrogen bonds with the bottom template to resist leaving it, revealing a “self-healing” property, which helps explain the toughness of the fibrils formed in the experiment.
We begin by describing our aMD simulation results on the stack of two β roll molecules with sequence, (GAGAGAGQ)10, without fixed atoms, and at the lowest threshold as characterized by n = 2 in Eq 7. (Recall that n is an integer in Eq 7 that determines the magnitude of the threshold as a multiple of the acceleration factor.)
In Fig 2, the order parameter of the top molecule, Ω, which measures the departure of the β roll from its ideal structure, is plotted against the simulation time, revealing the striking result that molecules in the β roll structure unfold in a step-wise fashion. In cMD simulations, Ω for the top molecule remains roughly constant around 62±2, indicating that the molecule stays in the β roll structure throughout the entire simulation. In contrast, in aMD it decreases in a step-wise fashion and eventually reaches zero. Roughly, five different plateaus can be discerned in the plot: 1–28 ns (Ω = 52), 32–48 ns (Ω = 38), 52–66 ns (Ω = 30), 70–74 ns (Ω = 24), and 78–90 ns (Ω = 2). The representative structures corresponding to the different states shown in Fig 2 were calculated via clustering analysis. The step-wise unfolding of the β roll structure agrees well with single-molecule force spectroscopy (SMFS) measurements by Sapra et al. on the unfolding pathways of β-barrel-forming membrane proteins, OmpG [21]. By mechanically pulling on a single atom at one end of OmpG they found that each β hairpin of the OmpG β barrel unfolded either individually, or cooperatively with an adjacent β hairpin, causing the OmpG protein to unfold in a step-wise fashion.
The unfolding pathway of the top molecule in the stack as revealed by aMD simulation can be described as follows. The molecule starts from a perfect β roll structure (see Fig 2(A) with a Ω of ~ 62 and then reaches its first plateau which lasts from 1 ns to around 28 ns. The representative structure generated from clustering analysis is shown in Fig 2(B) with one strand lifted off from the N terminal. By 28 ns, another strand from the C terminal starts to come off the β roll structure as shown in Fig 2(C). This is a transient state because the order parameter quickly moves to the second plateau that lasts from around 32 ns to 58 ns. The representative structures that occur during this plateau are shown in Fig 2(D), 2(E) and 2(F). The majority of structures that occur during this plateau have one strand off of the C terminal and two strands off of the N terminal as in Fig 2(D) and 2(F). The second strand from the N terminal goes back to the β roll structure for a short period of time within this plateau, from around 41 ns to 44 ns, as shown in Fig 2(E). After 52 ns, the order parameter reaches another plateau, with Ω ~ 30, during which the third strand peels off the β roll from the C terminal as shown in Fig 2(G). This strand goes back to the roll structure for a short period of time as shown in Fig 1(H), and then quickly comes off the β roll together with the fourth strand from the C terminal at around 70 ns, as shown in Fig 2(I). After that, there are only five strands left in the β roll structure, which is not enough to maintain the configuration. Starting at 73 ns, the molecule collapses quickly and becomes a random coil structure. Some refolding events occur during the unfolding process, mainly when one loose strand goes back to its original neighbor. For example, as shown in Fig 2, the second strand comes off the C terminal at stage d, then goes back to the C terminal at stage e, and finally comes off the C terminal again at stage f. Moreover, the third strand from the C terminal comes off at stage g, then goes back at stage h, and eventually comes off of the C terminal with the fourth strand at stage i.
The β roll molecules thus appear to unfold in an asymmetric fashion, namely mainly from the C terminal, as evidenced by the unfolding pathway just described. This finding agrees well with a report by Alsteens et al. based on a steered molecular dynamics (sMD) simulation study in which a prototypic TpsA protein, FHA [22] unfolds mainly from the C terminal. In addition, our simulation suggests that a β roll configuration needs to have a nucleus of a certain size to maintain its structure: the sequence (GAGAGAGQ)10 needs to have at least half of its strands, 5 strands, in a β roll structure, in order to maintain the β roll configuration. With less folded strands, it collapses and forms an amorphous configuration.
A second simulation of the two-molecule stack without fixed atoms having the same aMD boost parameter (n = 2) was performed in order to check for reproducibility. This second simulation was performed for 200 ns longer than the first simulation as the chain took longer to completely unfold. Both simulations exhibit step-wise unfolding behaviors as can be seen in Fig 3, which plots the order parameter, Ω, versus time for Simulations 1 and 2. The two simulations go through the same stages as the unwrapping occurs, each stage is outlined in blue in Fig 3. This similarity helps to support the reproducibility of our simulations of the unfolding process.
aMD simulations are then performed on systems containing a stack of (GAGAGAGQ)10 β roll molecules with a partially-fixed bottom template at the threshold potential energy with n = 2 in Eq 7. By having a partially-fixed bottom template (as defined above), molecules in the stack do not unfold simultaneously and their unfolded strands do not entangle with each other. Thus we observe the unfolding behavior of just the molecule on top.
Fig 4 shows the final structures in the three different types of simulations that we ran. The first type of simulation uses the threshold with n = 2 in Eq 7 and does not have any atoms fixed. As a result, both molecules in the stack in simulation 1 unfold completely after 80 ns; the snapshot in Fig 4 (A) is taken at the point at which Ω = 0 in Fig 3. A similar completely unfolded state occurs after 140 ns for simulation 2 in Fig 3. The second type of simulation is performed on a system that contains a partially-fixed bottom template and uses an intermediate threshold with n = 2 in Eq 7. As shown in Fig 4(B), the final structure of the top molecule in the stack has three unfolded strands: one off at the N terminus, and two off at the C terminus. The third type of simulation is again for a system with partially-fixed bottom template, and uses the highest threshold with n = 2.5. Now, five strands unfold from the top β roll molecule, as seen in Fig 4(C). This molecule does not unfold completely even after 300 ns of aMD simulations with an increased threshold, in stark contrast to the behavior observed in the system without partially-fixed template, where the chain collapses quickly when there are only 5 strands left in the β roll. These observations once again underline the importance of having a partially-fixed bottom template to stabilize the top β roll structure.
The unfolding process for the top molecule in the two molecule stack with partially-fixed bottom template and boosts n = 2.0 or n = 2.5 resembles that of the molecule in the stack with no atoms fixed. Fig 5 plots the order parameters of the top molecule in the simulations with boost n = 2.5 and n = 2 against simulation time. The top molecule in the simulation with boost n = 2.5 (Fig 5A) shows that it goes through 5 stages to reach the final configuration, which has 3 strands coming off the N terminal and 2 strands coming off the C terminal. This stepwise unfolding is similar to the unfolding behavior of the top molecule in the stack without fixed atoms and n = 2 shown in Fig 2. The sequence of steps is: one strand comes off the N terminal, one strand comes off the C terminal, the second strand comes off the C terminal, the second strand comes off the N terminal, and finally the third strand comes off the N terminal. The only difference between the unfolding process for n = 2.5 with fixed atoms and n = 2 without fixed atoms is that the strands from N terminal come off earlier when n = 2.5 than when n = 2. The top molecule in the simulation with n = 2 and partially-fixed bottom template also unfolds in a step-wise fashion as shown in Fig 5B. Therefore, the unfolding behavior seems to be independent of the value of the boost potential.
To identify the dominant forces in the β roll structure, we plot, for the selected hydrogen bonding atom pairs indicated in Fig 6, hydrogen bond potentials of mean force (PMF) versus distance in Figs 7 and 8. These are based on the trajectories generated by the aMD simulations with boost potential n = 2 and partially-fixed bottom template. Note that here we present the unweighted PMF versus the distance of hydrogen bonded pairs of only one of the two simulations-performed using boost potential n = 2 and partially fixed bottom template (recall that we performed two simulations for each set of parameters as shown in Table 1 in the method section). The unweighted PMF versus the distance of hydrogen bonded pairs of the other simulation is provided in the supporting information in S2 Fig and S3 Fig. Hydrogen bonds in a β roll configuration are categorized as being either lateral or vertical (see Fig 6). Lateral hydrogen bonds refer to the ones formed between the neighboring β strands in a single β sheet, or between neighboring β turns, and vertical hydrogen bonds refer to the ones between atoms in the top and bottom of a single β turn, or between atoms in the β turns of top and bottom molecules. All the unweighted potential of mean force profiles in Fig 7 (lateral H bonds) and 8 (vertical H bonds) are calculated with three different bin sizes, resulting in three curves for each plot. These curves match well with each other, indicating that we have enough samples for the calculation.
The unweighted PMFs associated with the hydrogen bonded atom pairs in Figs 7 and 8 have global minima at ~ 1.9 angstroms, indicating that these atoms prefer to stay within the hydrogen bonding distance. This reveals that the original β roll structure, in which all these atoms can form hydrogen bonds, is more stable than the unfolded structure, where only a few H bonds are possible. The hydrogen bonding strengths are taken to be the values of the PMF at the first peak in the PMF versus distance curves.
The average hydrogen bond strengths in the two simulations with boost potential n = 2 and partially-fixed bottom template are given in Fig 9. The figure shows the hydrogen bonding strengths for the lateral hydrogen bonds between β strands (green) and between β turns (pink), and the vertical hydrogen bonds within β turns (blue) and between the turns in the top and bottom molecules (yellow). The first 3 columns represent the strengths of the lateral hydrogen bonds between the neighboring β strands in the top layer of the β roll structure as shown in Fig 6. The strengths of the lateral hydrogen bonds along the β strands in both the top and bottom layers of the β roll molecule are weaker when the hydrogen bonds are closer to the β turns: e.g. the hydrogen bonds between residues 51–68 and between residues 53–70 have a lower strength than the hydrogen bonds between residues 53–68. The strength of the lateral hydrogen bond formed between residues 56–73, indicated by the height of the 4th bar, is weaker than the lateral hydrogen bonds within the bottom layer but stronger than the lateral hydrogen bonds within the top layer of the top β roll molecule. Something similar is observed for the hydrogen bonds between neighboring β strands in the bottom layer of the β roll structure; see the 5th, 6th and 7th columns in Fig 9.
The lateral hydrogen bonds in the lower layer of the β roll structure are clearly stronger than those in the upper layer of the β roll structure. As seen in Fig 9, the 5th, 6th and 7th columns representing the hydrogen bonding strength in the bottom layer of the β roll, are higher than the first three columns representing the hydrogen bonding strength in the top layer of the β roll. This is likely a consequence of the bottom layer in the top molecule being in direct contact with the bottom template. The effect of stacking on the stability of the roll was investigated in our previous study [7]; there we found that stacking helps stabilize the β roll structure by increasing the number of intra-molecular hydrogen bonds in each β roll molecule. Here we see that in terms of bond energies that conclusion is confirmed.
The vertical hydrogen bonds are usually weaker than the lateral hydrogen bonds. This can be observed by comparing the heights in Fig 9 of the first 7 columns, which represent the strengths of the lateral hydrogen bonds, with the heights of the last 3 columns, which represent the strengths of the vertical hydrogen bonds. The heights of the two blue columns in Fig 9, which represent hydrogen bonds within the β turns, are smaller than those of the first 7 columns, representing the lateral hydrogen bonds. This signifies that lateral hydrogen bonds play a more important roll than vertical hydrogen bonds in keeping the molecule in a β roll configuration. The height of the yellow bar, which represents the average strength of the hydrogen bond between the top and bottom molecules (there is one such bond per strand), is slightly lower than that of the first three columns, indicating that the hydrogen bonds between the two molecules are almost as strong as the lateral hydrogen bonds between the β strands in the upper layer of the top molecule. This suggests that the hydrogen bonds between the two molecules also play a significant role in maintaining the β roll structure of the top molecule.
In the simulation with the highest threshold energy, where n = 2.5 in Eq 7, and a partially-fixed bottom template, a new intermediate structure shows up. It contains a β hairpin structure and an anti-parallel β sheet formed by strands from the top and bottom molecules. Fig 10(A) shows the unweighted PMF versus the distance between the hydrogen on GLN (residue 24) and the oxygen on ALA (residue 26). Two minima are identified in the plot, a local minimum at short distance (a) and a global minimum further out (b). The intermediate structure associated with the local minimum is shown in Fig 10(B) and its side view is shown in Fig 10(C). The structure associated with the global minimum is a distorted β roll structure as shown in Fig 10(D). The potential well of the intermediate structure is located at ~ 2 angstroms, indicating that a hydrogen bond forms between the hydrogen on GLN (residue 24) and the oxygen on ALA (residue 26), i.e., residues 24, 25 and 26 have formed a three-amino-acid turn. Strands 3 and 4 form an antiparallel β sheet structure. Taken together, the turn and the antiparallel structures are essentially a typical β hairpin structure. Another anti-parallel β sheet is formed between strand 2 in the top molecule and the silver strand in the bottom molecule. A side view of this structure is seen in Fig 10(C) which shows how strands 2 and 3 traverse the interface between the two molecules. The reason this structure forms is that the first strand from the N terminal in the bottom template is not fixed and breaks loose. This provides enough room for the second and third strands of the top molecule to reach down one layer, forming β sheets with the strand in the bottom template.
The anti-parallel β sheet formed by strands from the top and bottom molecules in the intermediate structure is of particular interest to us as this configuration could potentially inhibit the unfolding process. In a long fibril with many molecules, the molecules in a β roll structure will probably not always stack as perfectly as in our starting configuration, meaning that strands from some molecules could potentially form β sheets with their folded neighbors, thus preventing the unfolding process. We call this a self-healing ability because it seems to hinder the β roll molecule from completely unwrapping; it might be one reason why the fibrils observed in the experiments are very strong.
Considering the results obtained here with respect to unfolding, we tentatively propose a hypothetical folding pathway; we emphasize that this is highly speculative and should not be considered as a conclusion supported by the simulation data obtained here, but rather as a direction for further studies. The folding process most likely starts with docking of a disordered silk-like (GAGAGAGX)n domain on a pre-folded molecule acting as template. This consistent with the observation that the template provides stability to the folded roll, and with the experimental fact that secondary structure develops in parallel with fibril growth. The disordered domain has to remain long enough in the docked state to allow for nucleation of a minimal folded part, e.g., a 5-stranded β solenoid. This step has a very low probability and is therefore likely to be rate-determining, accounting for the very low growth rates observed experimentally11. Moreover, it also would explain why silk-like domains of higher number of repeating units, which are likely to have longer residence times and a higher nucleation probability, tend to give faster growth [23]. Once nucleation has occurred, the remainder of the silk-like domain can fold to form the complete β solenoid.
We used accelerated molecular dynamics (aMD) simulations to investigate the unfolding of a stack of two β roll molecules, (GAGAGAGQ)10. Although much is known of about the structure of the β solenoid, very little is known about the partially folded conformation of the silk-like polypeptide or the details of the folding/unfolding process. Unfolding simulations can help us understand biological processes and, when well sampled, can provide us with partially-folded structures. aMD is able to maintain the original shape of the energy landscape and let the molecule sample conformational space fairly naturally.
Our goal was to identify the dominant forces that keep the silk-inspired polypeptide in a β roll configuration, to investigate the unfolding mechanism of silk-inspired polypeptides. The β roll structure that we use in this study was obtained from our previous investigation of the stable configuration of the β roll using the same sequence (GAGAGAGA)10. Unlike the structure proposed by Schor et al. [14] where all the alanine residues pointed out, forming a hydrophobic shell, the structure used in this study possesses a hydrophobic core which we have shown to be more stable than that with a hydrophobic shell [7].
The size of the boost potential was chosen carefully. It should not be so small that unfolding is unlikely to occur, as this would essentially be the same as a conventional MD simulation and it should not be too strong, because this might induce an unrealistic unfolding process. The number of boost potentials added to the original potential, n, was therefore chosen to reveal both the unfolding and the any spontaneous refolding of the polypeptide that occurred during the simulations. The unfolding process of the molecule on top without any fixed atoms showed rejoining of the strands as well as unfolding. Moreover, we saw that the unfolding process can be reproduced by additional simulations with the same parameters and even by simulations with different boost potentials.
To justify the convergence of our simulations, the relaxation time of the peptide backbone vectors is estimated from the time autocorrelation function profile. S1 Fig show a plot of the time correlation function of the out-of-plane vectors (the vector that is perpendicular to the plane formed by 3 consecutive carbon atoms) of the polypeptide versus the simulation time. The time at which this reaches zero gives a measure of the relaxation time of the peptide [7]. The relaxation time is less than 10 ns for simulations without fixed atoms, indicating that our 100ns simulation is long enough to reach equilibrium. The 300 ns simulations with partially-fixed bottom template have relaxation times less than 100 ns, which indicates that the molecules in these systems have reached equilibrium.
By comparing the unfolding order parameter, Ω, versus simulation time between cMD and aMD, we found that a molecule in a stack of two β roll molecules unfolds in a step-wise fashion, i.e. one β strand in the β roll molecule at a time, which agrees well with the experimental study on transmembrane β-barrel protein OmpG by Sapra et al. [21]. We also found that it unfolds mainly from the C terminal, which matches with the simulation study on a prototypic TpsA protein, FHA by Alsteens et. al [22]. Through observing the unfolding and spontaneous refolding of single strand in the β roll structure, we get a better idea of the possible intermediates that might occur during the folding process. Schor et al. [15] hypothesize that the molecule folds into a β roll structure with a hydrophobic shell by itself, then docks onto another preformed β roll molecule, a “roll n’ dock” process.
The bottom template is found to play an important role in stabilizing the β roll structure of the molecule on top. This was concluded by comparing the final structure in three sets of simulations with systematically increased threshold energies. At the lowest threshold energy, both molecules unfold and have a random coil structure by the end of the simulation for systems without any fixed atoms. When the bottom template is partially fixed, the top molecule is unable to unfold completely, even by the end of 300 ns simulations, indicating the significance of the bottom template in stabilizing the molecule on top of it.
We further elucidate how the bottom template stabilizes the top β role molecule by quantifying the strengths of the various intra and inter molecular hydrogen bonds. The lateral hydrogen bonds in the lower layer of the top molecule are stronger than those in its upper layer, indicating that the bottom template strengthens the hydrogen bonds in the lower layer of the top molecule. We further confirm the stabilizing effect of the bottom template reported in our previous investigation, in which we concluded that the bottom template induces more intramolecular hydrogen bonds in the top molecule when it docks on to the bottom template[7]. We also found that the lateral hydrogen bonds between the β strands in a β roll configuration become weaker as they get close to the β turns. This is due to the fact that the β turn structure is more flexible than the β sheet structure in the β roll molecule. Vertical hydrogen bonds within the β roll structure are considerably weaker than lateral hydrogen bonds, signifying the importance of lateral hydrogen bonds in stabilizing the β roll structure.
Finally, an intermediate structure was found containing a β hairpin and an anti-parallel β sheet formed by strands from the top and bottom molecules, revealing the self-healing ability of the β roll stack. The β hairpin structures can form fibrils by themselves as reported by other studies [22, 23] in which β hairpins first stack by hydrophobic interactions and then assemble via hydrogen bonds. Here we found β hairpins in the stack of two silk-inspired molecules, indicating that these β hairpins may also play a role in stabilizing silk-inspired fibrils with many molecules. Such β sheets formed between pairs of molecules, an inter-protein β sheet structure, were also reported by Razzokov et al., who studied a sequence similar to ours, (GAGAGAGE)5, using replica exchange molecular dynamics[24]. Overall, the strength of the fibril of β roll molecules comes not only from the stability of each individual molecule, but also from the cooperative effect provided by the anti-parallel β sheet structures formed by strands from the top and bottom molecules. Our results led us to tentatively propose a hypothetical folding pathway that is consistent with the experimental results.
In our previous paper7, we performed conventional explicit-solvent atomistic molecular dynamics simulations on a stack of two β roll molecules with a sequence (GAGAGAGQ)10 using Amber 12 and the ff12SB force field. The simulation details can be found in our previous paper[7). The last 50 ns of those trajectories were used to calculate the time averages of the total potential energy of the system as well as the dihedral energy of the peptides.
The general principles of aMD are as follows. A boost potential ΔV(r) is added to the original potential energy surface of the system when the system’s potential energy is lower than a predefined threshold energy, E [25], as shown in Fig 11.
V*(r)=V(r)+ΔV(r),V(r)<E,V*(r)=V(r),V(r)≥E,
(1)
where V*(r) is the modified (boosted) potential energy, V(r) is the original potential energy (which could be the total potential energy of the system or the dihedral energy of the polypeptide) and r is a positional degree of freedom or a torsional degree of freedom, etc. The general form of the boost potential, ΔV(r) is given by the equation below:
ΔV(r)=(E−V(r))2α+E−V(r)
(2)
where α is the acceleration factor. The acceleration factor is a parameter that governs the size of the boost. As it gets smaller, the energy surface becomes flatter, thus improving the likelihood of transitions between low energy states. The gist of the method therefore is that the global pattern of the potential energy is maintained, but barriers become smaller, allowing easier passage.
Dual-boost aMD simulations are performed in our investigation, which means boost potential energies are added to both the total potential energy of the system and to the dihedral energy of the polypeptides. The total potential energy of the system Vtotal (r) consists of the dihedral energy Vdihedral(r) and the non-dihedral energy Vnon-dihedral(r) as shown below,
Vtotal(r)=Vnon−dihedral(r)+Vdihedral(r).
(3)
In a dual boost aMD simulation, a boost potential ΔVdihedral(r) is first added to the dihedral energy of the peptide as in Eq 4 below,
Vdihedral*(r)=Vdihedral(r)+ΔVdihedral(r)
(4)
where Vdihedral*(r) is the modified dihedral energy of the peptide. Then a boost potential ΔVtotal(r) is added to the total potential energy of the system as shown in Eq 5 below,
Vtotal*(r)={Vnon−dihedral(r)+Vdihedral*(r)}+ΔVtotal(r)
(5)
where Vtotal*(r) is the modified total potential energy and ΔVtotal(r) is the boost potential energy added to the total potential energy of the system, Vtotal(r).
The equations used to calculate the boost potential of the total potential energy of the system Vtotal (r) and the boost potential of the dihedral energy Vdihedral(r) are
ΔVtotal(r)=(Etotal−Vtotal(r))2αtotal+Etotal−Vtotal(r)ΔVdihedral(r)=(Edihedral−Vdihedral(r))2αdihedral+Edihedral−Vdihedral(r)
(6)
where Etotal and Edihedral are the thresholds for the total energy and dihedral energy, and αtotal and αdihedral are the acceleration factors for the total potential energy and dihedral energy. Note that these equations are of the same form as Eq 2.
The pre-defined thresholds, Etotal and Edihedral, and the acceleration factors, αtotal and αdihedral, for the two types of boosts are calculated as below,
Edihedral=Vavg_dihedral+3.5×Nres,αdihedral=3.5×Nres/5Etotal=Vavg_total+n×αtotal,αtotal=0.2×Natoms(n=1,2,3…)
(7)
where Vavg_dihed and Vavg_total are time averages of the dihedral energy and total potential energy obtained from conventional molecular dynamics (cMD). These are calculated only once, before running the aMD simulations, to generate the value of thresholds. The parameters Nres and Natoms are the number of polypeptide residues and the number of atoms in the system, respectively; n in Eq 7 is an integer that determines the magnitude of the threshold as a multiple of the acceleration factor.
Consequently, there are only four input parameters in an aMD simulation: Edihedral, αdihedral, Etotal and αtotal. All the other parameters are calculated based on these four values. Sometimes we increase the threshold of the total potential, Etotal, by making the value of n in Eq 7 larger, to let the system access more conformational space. Increasing the threshold of the total potential energy enables more energy minima to lie below the threshold and have boost potentials added to them. As shown in Fig 11, the third free energy minimum from the left does not get an added boost potential when the threshold is E. When the threshold to is increased to E’, the third minimum falls below the threshold, so a boost potential is added to it. This facilitates the transition between the second and the third minima. We used different threshold energies in different sets of simulations in our investigation, so the value of n in Eq 7 varies from simulation to simulation.
We ran three types of aMD simulations on the stack of two β roll molecules (GAGAGAGQ)1210 shown in Fig 1 and two simulations for each set of aMD boost parameter as shown in Table 1. The first type of simulation uses the lowest threshold potential energy with n = 2 in Eq 7. This type of simulation is performed for 100 ns. The second and the third type of simulations use increased threshold potential energies with n = 2 and 2.5, respectively, and are performed for 300 ns each. During the first type of simulation, no atoms are fixed. However, during the second and the third types of simulations, 10 Cα atoms in the glycine (G) residues at the bottom of the β turns on both sides of the β roll molecule are spatially fixed to maintain the bottom template’s β roll configuration, as shown in Fig 12. These atoms are restrained with a force of 10 kcal/mol, and were chosen to mimic the presence of a substrate. In experiment, a substrate is used to grow fibrils; the fibrils are found to grow perpendicular to the substrate [16,26]. Fixing specific atoms in the bottom β roll molecule helps maintain the bottom molecule in a β roll configuration, while allowing some flexibility to the chain. The simulation times, as well as the number of acceleration factors added to the average system’s total potential energy for different types of simulations are summarized in Table 1.
Hydrogen bonds play a very important role in keeping the folded and stacked structure together. We therefore pay specific attention to the strength of these bonds, by calculating potential of mean force (PMF) profiles. Potential mean force (PMF), used synonymously in the literature to indicate a free energy profile, examines the change of a system’s free energy as a function of some specific reaction coordinate, such as the lateral distance between the hydrogen bonding sites on two neighboring β strands or the vertical distance between hydrogen bonding sites within a β turn structure or between the two stacking molecules. In this work we take as our reaction coordinates the distances between the important hydrogen bonded atoms pairs. The strength of the hydrogen bonds can be gleaned from the height of the first peak in these PMF profiles. The dominant forces controlling docking and folding of the β roll can then be determined by comparing the strengths of these hydrogen bonds. Usually, the free energy profiles generated from the aMD simulations need to be reweighted using the Boltzmann factor for the boost potential [17]. However, given the large size of our system, there is a large noise associated with reweighting the free energy profile. Therefore, and since we use the profiles only for comparative purposes, unweighted PMF profiles are presented in this paper. A similar approach was used in a study of the free energy landscape of large systems of G-protein coupled receptors [19] by Miao et al.
The unweighted potential of mean force or F(Aj) is calculated as a function of the distance Aj between hydrogen-bonded partners, for different atoms pairs. The relevant equation is F(Aj) = -kBTln p(Aj), where p (Aj) signifies the probability of finding the atoms pairs within Aj (distances are divided into a number of equally distributed bins with index j), kB is the Boltzmann constant, and T is the temperature of the system.
The hydrogen bonded atom pairs used to calculate the potential mean force (PMF) are constantly forming and breaking with time. S1 Table gives the hydrogen-bonding average lifetimes which were calculated in the following way. In the hydrogen-bonding data set, we define a hydrogen bonding indicator for each pair of atoms to be 1.0 when they are hydrogen bonded and 0.0 when they are not. The average lifetime is determined by averaging the length of time that a particular hydrogen bond is present continuously. As shown in S1 Table, the number of times that the hydrogen bonds form and break varies from 2007 times to 6808 time. The maximum lifetime for hydrogen bonded atom pairs in PMF calculations is between 50 ps and 223 ps. The average lifetime for atom pairs in PMF calculations is from 6.123 ps to 24.0842 ps. It is apparent that the hydrogen bonds are constantly breaking and forming, over our simulations which of to 300,000 ps.
In order to characterize the state of molecules with respect to folding, an unfolding order parameter, Ω, is needed which measures departures of the β roll from its ideal structure. We here define Ω as the number of amino acids that are at an appropriate distance to the neighbor they would have in an ideal β roll structure, e.g., by counting pairs of residues i and i+16, that are within a distance of 4.0 to 6.0 angstroms of each other. This typical range has been inferred from cMD simulations during which the molecules retain their β roll structure. When the (GAGAGAGQ)10 peptide (which has a total of 80 residues) is in a perfect β roll structure there are 64 pairs (i, i + 16), so that the maximum value of Ω is 64.
|
10.1371/journal.pcbi.1006364 | Comparative structural dynamic analysis of GTPases | GTPases regulate a multitude of essential cellular processes ranging from movement and division to differentiation and neuronal activity. These ubiquitous enzymes operate by hydrolyzing GTP to GDP with associated conformational changes that modulate affinity for family-specific binding partners. There are three major GTPase superfamilies: Ras-like GTPases, heterotrimeric G proteins and protein-synthesizing GTPases. Although they contain similar nucleotide-binding sites, the detailed mechanisms by which these structurally and functionally diverse superfamilies operate remain unclear. Here we compare and contrast the structural dynamic mechanisms of each superfamily using extensive molecular dynamics (MD) simulations and subsequent network analysis approaches. In particular, dissection of the cross-correlations of atomic displacements in both the GTP and GDP-bound states of Ras, transducin and elongation factor EF-Tu reveals analogous dynamic features. This includes similar dynamic communities and subdomain structures (termed lobes). For all three proteins the GTP-bound state has stronger couplings between equivalent lobes. Network analysis further identifies common and family-specific residues mediating the state-specific coupling of distal functional sites. Mutational simulations demonstrate how disrupting these couplings leads to distal dynamic effects at the nucleotide-binding site of each family. Collectively our studies extend current understanding of GTPase allosteric mechanisms and highlight previously unappreciated similarities across functionally diverse families.
| GTPases are a large superfamily of essential enzymes that regulate a variety of cellular processes. They share a common core structure supporting nucleotide binding and hydrolysis, and are potentially descended from the same ancestor. Yet their biological functions diverge dramatically, ranging from cell division and movement to signal transduction and translation. It has been shown that conformational changes through binding to different substrates underlie the regulation of their activities. Here we investigate the conformational dynamics of three typical GTPases by in silico simulation. We find that these three GTPases possess overall similar substrate-associated dynamic features, beyond their distinct functions. Further identification of key common and family-specific elements in these three families helps us understand how enzymes are adapted to acquire distinct functions from a common core structure. Our results provide unprecedented insights into the functional mechanism of GTPases in general, which potentially facilitates novel protein design in the future.
| Guanosine Triphosphate Phosphohydrolases (GTPases) are ubiquitous molecular machines mediating a variety of essential cellular processes [1]. Harnessing the GTP hydrolysis to modulate the affinity of partner molecule binding, GTPases transduce intracellular signals, control cell division and differentiation, and direct protein synthesis and translocation [2–5]. In general, GTP-bound GTPases in the active state are able to interact with partner effectors and regulate effector-mediated processes. GTP hydrolysis leads to the dissociation of GTPases from effectors, whereas exchange of GDP for GTP activates GTPases and restarts the signaling or protein synthesis cycle [6,7]. Two classes of accessory proteins are involved in regulating this reaction cycle. GTPase-activating proteins (GAPs) accelerate the GTPase activity and the inactivation of GTPases, whereas guanine nucleotide exchange factors (GEFs) promote GDP dissociation and subsequent GTP binding, activating GTPases [8–10].
There are three major GTPase superfamilies: small Ras-like GTPase, heterotrimeric G protein α subunit (Gα) and protein-synthesizing GTPase. Both small and heterotrimeric G proteins participate in signal transduction. As the primary coupling molecule to membrane receptors, Gα together with its partner βγ subunits (Gβγ) mediate the very early stage signal transduction initiated by extracellular stimuli. In contrast, small GTPase does not interact with receptors directly and regulates more downstream events in the cascade. Finally, the protein-synthesizing proteins participate in initiation, elongation and termination of mRNA translation. Underlying this functional difference are the low sequence identity (<20%) and overall different molecular shapes among these three types of GTPases. In particular, whereas small G protein consists of a single canonical Ras-like catalytic domain (RasD), Gα has an extra α-helical domain (HD) inserted and elongation factor EF-Tu has two extra β-barrel domains (D2 and D3) subsequent to the C-terminus (Fig 1). In addition, Gα can form a complex with Gβγ and undergoes a cycle of altered oligomeric states during function.
In contrast to the functional and structural diversity, GTPases display significant conservation in the core structure of the catalytic domain. Small GTPase, Gα and EF-Tu contain a RasD consisting of six β strands (β1-β6) and five α helices (α1-α5) flanking on both sides of the β sheet (Fig 1). Three highly conserved loops named P-loop (PL), switch I (SI), and switch II (SII) constitute the primary sites coordinating the nucleotide phosphates. This structural similarity suggests that at a fundamental level small GTPase, Gα and EF-Tu may utilize the same mode of structural dynamics for their allosteric regulation, which is likely inherited from their common evolutionary ancestor [11,12]. However, it is currently unclear what are the general atomistic mechanisms underlying GTPase allostery and how these common mechanisms can be adapted to have specific function.
Recent computational and experimental studies have gained much insight into the allosteric mechanisms of individual small and heterotrimeric G protein systems. Principal component analysis (PCA) of crystallographic structures and molecular dynamics (MD) simulations characterized the structural dynamics of small GTPase Ras and revealed an intriguing dynamical partitioning of Ras structure into two lobes: the N-terminal nucleotide binding lobe (lobe1) and the C-terminal membrane anchoring lobe (lobe2) [13,14]. Several allosteric sites were identified in lobe 2 or between lobes, including L3 (the loop between β2 and β3), L7 (the loop between α3 and β5), and α5. Importantly, α5 is the major membrane-binding site and has been related to the nucleotide modulated Ras/membrane association [15]. In addition, binding of small molecules at L7 has been reported to affect the ordering of SI and SII [16]. Intriguingly, recent studies of Gα have revealed nucleotide associated conformational change and bilobal substructures in the catalytic domain largely resembling those in Ras [17,18]. The allosteric role of lobe 2, which contains the major binding interface to receptors, has also been well established for Gα [18–27]. Furthermore, the comparison between G proteins and translational factors via sequence and structural analysis indicates a conserved molecular mechanism of GTP hydrolysis and nucleotide exchange, and cognate mutations of key residues in the nucleotide-binding regions showed similar functional effects among these systems [2,6,7,12]. Collectively, these consistent findings from separate studies support the common allosteric mechanism hypothesis of GTPases and underscore a currently missing detailed residue-wise comparison of the structural dynamics among different GTPase superfamilies.
In this study, we compare and contrast the nucleotide-associated conformational dynamics between H-Ras (H isoform of Ras), Gαt (transducin α subunit) and EF-Tu (elongation factor thermo unstable), and describe how this dynamics can be altered by single point mutations in both common and family-specific ways. This entails the application of an updated PCA of crystallographic structures, multiple long time (80-ns) MD simulations, and recently developed network analysis approach of residue cross-correlations [18]. In particular, we identify highly conserved nucleotide dependent correlation patterns across GTPase families: the active GTP-bound state displays stronger correlations both within lobe1 and between lobes, exhibiting an overall “dynamical tightening” consistent with the previous study in Gα alone [18]. Detailed inspection of the residue level correlation networks along with mutational MD simulations reveal several common key residues that are potentially important for mediating the inter-lobe communications. Point mutations of these residues substantially disrupt the couplings around the nucleotide binding regions in Ras, Gαt and EF-Tu. In addition, with the same network comparison analysis, we identify Gαt and EF-Tu specific key residues. Mutations of these residues significantly disrupt the couplings in Gαt and EF-Tu but have no or little effect in Ras. Our results are largely consistent with findings from experimental mutagenesis, with a number of dynamical disrupting mutants have been shown to have altered activities in either Ras or Gα. Our new predictions can be promising targets for future experimental testing.
Previous PCA of 41 Ras crystallographic structures revealed distinct GDP, GTP and intermediate mutant conformations [13]. Updating this analysis to include the 121 currently available crystallographic structures (S1 Table) reveals consistent results but with two additional conformations now evident (Fig 2A). In addition to GDP (green in Fig 2A), GTP (red), and mutant forms, GEF-bound nucleotide free (purple) and so-called ‘state 1’ forms (orange) are now also apparent. In the GEF-bound form, the SI region is displaced in a distinct manner– 12Å away from the nucleotide-binding site coincident with the insertion of a helix of GEF into the PL-SI cleft. The state 1 GTP-bound form was first observed via NMR and later high-resolution crystal structures were solved [28–30]. In contrast to the canonical GTP-bound conformation (red), the state 1 form (orange) lacks interaction between the two switches and the γ-phosphate of GTP, resulting in a moderate 7Å displacement of SI away from its more closed GTP conformation.
The first two PCs capture more than 75% of the total mean-square displacement of all 121 Ras structures. Residue contributions from SI and SII dominate PC1 and PC2 (Fig 2D). The height of each bar in Fig 2D displays the relative contribution of each residue to a given PC. PC1 mainly describes the opening and closing of SI–more open in GEF-bound and state 1 forms, and more closed in nucleotide bound structures. PC1 also captures smaller scale displacement of L8 (the loop between β5 and α4), which resides 5Å closer to the nucleotide-binding pocket in the GEF-bound structures than the GTP-bound structure set. PC2 depicts SII displacements and clearly separates GTP from GDP bound forms (red and green, respectively). As we expect, the lack of γ-phosphate in the GDP releases SII from the nucleotide, whereas in the GTP form SII is fixed by the hydrogen bond of the backbone amide of G60 with the γ-phosphate oxygen atom. This is also shown in the state 1 form where the hydrogen bond is disrupted with SII moderately displaced from the nucleotide (4Å on average from the canonical GTP group structures).
PCA of 53 available Gαt/i structures described recently (S2 Table) revealed three major conformational groups: GTP (red in Fig 2B), GDP (green) and GDI (GDP dissociation inhibitor; blue) bound forms [18]. The first two PCs capture over 65% of the total variance of Cα atom positions in all structures. The dominant motions along PC1 and PC2 are the concerted displacements of SI, SII and SIII in the nucleotide-binding region as well as a relatively small-scale rotation of the helical domain with respect to RasD (Fig 2E).
PC1 separates GDI-bound from non-GDI bound forms. In GDI-bound structures the GDI interacts with both the HD and the cleft between SII and SIII of the Ras-like domain, increasing the distance between SII and SIII. Similar to Ras, PC2 of Gαt/i clearly distinguishes the GTP and GDP-bound forms, where again the unique γ-phosphate (or equivalent atom in GTP analogs) coordinates SI and SII. In addition, the SIII is displaced closer to the nucleotide, effectively closing the nucleotide-binding pocket.
PCA of 23 available full-length EF-Tu structures reveals distinct GTP and GDP conformations (S3 Table). PC1 dominantly captures nearly 95% of the total structural variance of Cα atom positions (Fig 2C). It mainly describes the dramatic conformational transition in SI as well as the large rotation of two β-barrel domains D2 and D3 (Fig 2F). In the GTP-bound form, the C-terminal SI is coordinated to the γ-phosphate and Mg2+ ion, forming a small helix near SII. Meanwhile, D2 and D3 are close to RasD and create a narrow cleft with SI, serving as the binding site for tRNA [31]. In the GDP-bound form, the C-terminal helix in SI unwinds and forms a β-hairpin, protruding towards D2 and D3 [32]. The highly conserved residue T62 (T35 in Ras) of EF-Tu moves more than 10Å away from its position in the GTP form and loses interaction with the Mg2+ ion. In addition, D3 rotates towards SI and D2 moves far away from the Ras-like domain. In contrast to PC1, PC2 only captures a very small portion (3.59%) of the structural variance in EF-Tu (Fig 2F). The major conformational change along PC2 is a small-scale rotation of D2 and D3 with respect to RasD in the GTP form.
PCA of Ras, Gαt/i and EF-Tu demonstrates that the binding of different nucleotides and protein partners can lead to a rearrangement of global conformations in a consistent manner. In particular, within RasD, these three families display conserved nucleotide-dependent conformational distributions with major contributions from the switch regions. In the GTP-bound form of these proteins, SI and SII are associated with the nucleotide through interacting with γ-phosphate. Despite these similarities, critical questions about their functional dynamics remain unanswered: How does nucleotide turnover lead to allosteric regulation of distinct partner protein-binding events? To what extent are the structural dynamics of these proteins similar beyond the switch region displacements evident in accumulated crystal structures? How do distal disease-associated mutations affect the functional dynamics for each family and are there commonalities across families? In the next section, we report MD simulations that address these questions, which are not answered by accumulated static experimental structures.
MD simulations reveal distinct nucleotide-associated flexibility at known functional regions. Representatives of the distinct GTP and GDP-bound conformations of Ras, Gαt and EF-Tu were selected as starting points for MD simulation. Five replicated 80-ns MD simulations of these three proteins for each state (GTP and GDP totaling 2.4μs; see Materials and Methods) exhibit high flexibility in the SI, SII, SIII/α3 and loop L3, L7, L8 and L9 regions (Fig 3A–3C). The Cα atom root-mean-square fluctuation (RMSF) in Gαt shows that SI is significantly more flexible in the GDP-bound state (Fig 3B). The C-terminal SI of Ras and EF-Tu, corresponding to the shorter SI in Gαt, is also more flexible with GDP bound (Fig 3A & 3C). Interestingly, the middle part of SI in Ras and EF-Tu show higher fluctuations in the GTP-bound state. Moreover, SII is more flexible in the GTP-bound state in Ras. Detailed inspection reveals that SII always stays away from the nucleotide during the GDP-bound state MD simulations, whereas SII sometimes moves close to and interacts with the unique γ-phosphate of GTP, leading to higher flexibility in the GTP-bound state. In contrast, the flexibility of SII in Gαt has no significant difference between states, whereas SII in EF-Tu is less flexible with GTP bound. This is due to the relatively compact interactions between SII and the unique D2 and D3 in the GTP-bound EF-Tu. In fact, D2 and D3 show extremely higher flexibility in the GDP state (Fig 3C). Overall, the nucleotide-dependent flexibility of RasD in Ras, Gαt and EF-Tu are quite similar except for SII.
The cross-correlations of atomic displacements derived from MD simulations also manifest conserved nucleotide-associated coupling in these three systems (Fig 3D–3F). In both Ras and Gαt, significantly stronger couplings within the catalytic lobe 1 between PL, SI and SII can be found only in the GTP-bound state (red rectangles in Fig 3D & 3E). Interestingly, a unique inter-lobe coupling between SII and SIII/α3 also characterizes the GTP-bound state in both systems (blue rectangles in Fig 3D & 3E). In EF-Tu, the intra-lobe 1 and inter-lobe couplings are similar between states (red and blue rectangles in Fig 3F). Intriguingly, a lot of negative correlations between D2 and RasD of EF-Tu are found in the GDP-bound state, indicating the swing motion of D2 with respect to RasD during MD simulations (lower triangle in Fig 3F).
Consensus correlation networks for each nucleotide state were constructed from the corresponding replicate MD simulations. In these initial networks, each node is a residue linked by edges whose weights represent their respective correlation values averaged across simulations (see Materials and Methods). These residue level correlation networks underwent hierarchical clustering to identify groups of residues (termed communities) that are highly coupled to each other but loosely coupled to other residue groups. Nine communities were identified for Ras and eleven for Gαt and EF-Tu (Fig 4). The two additional family specific communities not present in Ras correspond to two regions of HD in Gαt and D2 and D3 in EF-Tu.
In the resulting community networks the width of an edge connecting two communities is the sum of all the underlying residue correlation values between them. Interestingly, Ras, Gαt and EF-Tu community networks can be partitioned into two major groups (dashed lines in Fig 4) corresponding to the previously identified lobes for Ras and the RasD in Gαt [13,18]. The boundary between lobes is located at the loop between α2 and β4. In these proteins, lobe1 includes the nucleotide-binding communities (PL, SI and SII) as well as the N-terminal β1-β3 and α1 structural elements. Lobe2 includes α3-α5, L8 and the C-terminal β4-β6 strands.
Comparing the GTP and GDP community networks of these three proteins reveals common nucleotide-dependent coupling features. In particular, for Ras and Gαt, comparing the relative strength of inter-community couplings in GTP and GDP networks using a nonparametric Wilcoxon test across simulation replicates reveals common significantly distinct coupling patterns (colored edges in Fig 4A & 4B). Within lobe1 stronger couplings between PL, SI and SII are observed for the GTP state of both families. This indicates that the γ-phosphate of GTP leads to enhanced coupling of these proximal regions. This is consistent with our PCA results above, where PC2 clearly depicts the more closed conformation of SI and SII in the GTP bound structures (Fig 2D & 2E). In addition, a significantly stronger inter-lobe correlation between SII and α3 is evident for the GTP state of both families, which is not available from analysis of the static experimental ensemble alone. This indicates that nucleotide turnover can lead to distinct structural dynamics not only at the immediate nucleotide-binding site in lobe 1 but also at the distal lobe 2 region.
Intriguingly, similar patterns of intra and inter-lobe dynamic correlations are observed in EF-Tu (Fig 4C). Within lobe1, significantly stronger correlations between PL-SI and PL-SII are evident in the GTP state, although SI-SII coupling becomes weaker in this state. In fact, the C-terminal β-hairpin of SI moves towards and interacts extensively with SII and D3 in the GDP bound state, leaving the nucleotide-binding site widely open. Moreover, our results reveal that SII and SIII/α3 of EF-Tu are more tightly coupled in the GTP state, resembling the strong inter-lobe couplings in the GTP bound Ras and Gαt. It is worth noting that this conserved structural dynamic coupling is evident only from the comparative network analysis and is not accessible from PCA of crystal structures.
Comparative network analysis highlights the common residue-wise determinants of nucleotide-dependent structural dynamics. Besides correlations within lobe1, inter-lobe couplings are also significantly stronger in the GTP state networks of Ras, Gαt and EF-Tu. Inspection of the residue-wise correlations between communities reveals common major contributors to the SII–α3 couplings in the three proteins (red residues in S4 Table). In particular, M72Ras in SII and V103Ras in α3 act as primary contributors to inter-lobe correlations in Ras. Interestingly, the equivalent residues in the other two systems, F211Gαt or I93EF-Tu in SII and F255Gαt or V126EF-Tu in α3/SIII also contribute to the inter-lobe couplings. We further examined the importance of these residues by MD simulations of mutant GTP-bound systems. Results indicate that each single mutation M72ARas and V103ARas can significantly reduce the couplings between SI and PL, indicating that these mutations disturb couplings at distal sites of known functional relevance (Fig 5A & 5D). Moreover, the cognate mutations F211AGαt and F255Gαt in Gαt not only decouple SI and PL but also SI and SII (Fig 5B & 5E). Similarly, the analogous mutation I93AEF-Tu decreases the correlations between PL and SI, whereas V126AEF-Tu decouples PL and SII (Fig 5C & 5F). The simulation results indicate that single alanine mutation of residues contributing to SII-α3 couplings diminishes the couplings of the nucleotide binding regions, and this allosteric effect is common in all the three proteins.
Inter-lobe couplings that are distal from the nucleotide binding regions are also shown to be critical for the nucleotide dependent dynamics in Ras, Gαt and EF-Tu. By inspecting the residue level couplings between L3 and α5, we identified common distal inter-lobe couplings in the three proteins. Mutational simulations indicate that the substitutions K188AGαt and D337AGαt significantly decouple SI from the PL and SII regions (Fig 6B & 6E). Interestingly, the mutations K188AGαt and D337AGαt have been reported to cause a 6-fold and 2-fold increase in nucleotide exchange, respectively, but no direct structural dynamic mechanism was established [19]. We further tested mutations of analogous residues in Ras. We considered both D47Ras and E49Ras as the equivalent residues to K188Gαt (due to the longer L3 region of Ras), and R164Ras as the equivalent residue to D337Gαt. Both double mutation D47A/E49ARas and single mutation R164ARas significantly reduce the correlations between PL and SI (Fig 6A & 6D). We note that the functional consequences of mutating these residues in Ras has been highlighted in a previous study, in which the salt bridges between D47/E49Ras in L3 and R161/R164Ras in α5 were shown to be involved in the reorientation of Ras with respect to the plasma membrane, and enhanced activation of MAPK pathway [15]. Moreover, substitutions of analogous residues R75AEF-Tu (L3) and D207AEF-Tu (α5) also significantly reduce the couplings between PL and SI (Fig 6C & 6F). Our results indicate that the conserved interactions between L3 and α5 are important for maintaining the close coordination of the distal SI, SII and PL around the nucleotide, and this is common to these three proteins.
Comparison of the GTP-bound residue-wise networks of Ras, Gαt and EF-Tu reveals that the N-terminus of α3 strongly couples SII only in Gαt and EF-Tu. In particular, we identified residues R201Gαt or A86EF-Tu (SII) and E241Gαt or Q115EF-Tu (α3) as underlying these strong couplings (blue residues in S4 Table). These residues are specific to Gαt and EF-Tu because the corresponding residues E62Ras in SII and K88Ras in α3 have no contribution in Ras (green residues in S4 Table). Mutational MD simulations indicate that substitutions E241AGαt and Q115AEF-Tu have a similar drastic effect on the coupling of nucleotide binding regions (S1 Fig). In particular, the couplings between PL, SII and PL are all significantly reduced (S1B & S1C Fig). We note that E241AGαt in Gαs (the α subunit of the stimulatory G protein for adenylyl cyclase) was previously reported to impair GTP binding but the structural basis for this allosteric effect has been unknown [33,34]. Our results indicate that weakened correlations of the nucleotide-binding regions in E241AGαt as a consequence of allosteric mutations in SIII/α3 and SII likely underlie the reported impaired GTP binding. Moreover, we identified residue E232Gαt as a Gαt-specific primary contributor to the inter-lobe couplings in SIII, which has no direct counterparts in Ras or EF-Tu due to the absence of SIII (purple residues in S4 Table). The simulation of mutation E232AGαt shows diminished couplings between PL, SI and SII, as well (S2A Fig). Similar effects of mutations R201AGαt and D234AGαt are also observed (S2B & S2C Fig).
Mutations of the counterpart residues E62ARas and K88ARas result in no significant change in the coupling of nucleotide binding loops in Ras (S1A Fig). Collectively these findings indicate that in Gαt and EF-Tu both N- and C-terminal α3 positions dynamically couple with SII, whereas in Ras the communication between α3 and SII is mainly through the C-terminus of α3. In addition, our results suggest that SIII plays a unique role in Gαt not only mediating the couplings between the two lobes but also allosterically maintaining the tight correlations between SI, SII and PL.
In this work, our updated PCA of Ras structures captures two new conformational clusters representing the GEF-bound state and “state 1”, respectively, in addition to the canonical GTP and GDP forms. By comparing the Ras PCA to PCA of Gαt/i and EF-Tu, we reveal common nucleotide dependent collective deformations of SI and SII across G protein families. Our extensive MD simulations and network analyses reveal common nucleotide-associated conformational dynamics in Ras, Gαt and EF-Tu. Specifically, these three systems have stronger intra-lobe1 (PL–SI and PL–SII) and inter-lobe (SII–SIII/α3) couplings in the GTP-bound state. Meanwhile, with the network comparison approach we further identify residue-wise determinants of commonalities and specificities across families. Residues M72Ras (SII), V103Ras (α3), D47/E49Ras (L3) and R164Ras (α5) are predicted to be crucial for inter-lobe communications in Ras. Mutations of these distal residues display decreased coupling strength in SI–PL. Interestingly, the analogous residues in the other two proteins, F211Gαt/I93EF-Tu (SII), F255Gαt/V126EF-Tu (α3), K188Gαt/R75EF-Tu (L3) and D337Gαt/D207EF-Tu (α5) also have important inter-lobe couplings and show similar decoupling effects upon alanine mutations. Besides the key residues that are common in the three systems, residues mediating inter-lobe couplings only in Gαt and EF-Tu are identified. These include R201Gαt/A86EF-Tu and E241Gαt/Q115EF-Tu, whose cognates in Ras do not have significant effect on the nucleotide-binding regions upon mutation. In addition, Gαt specific residue E232Gαt in SIII (which is missing in Ras and EF-Tu) is identified to be important to the couplings of the nucleotide-binding regions. Importantly, some of our highlighted mutants (D47A/E49ARas, K188AGαt, D207AGαt and R241AGαt) have been reported to have functional effects by in vitro experiments. Our analysis provides insights into the atomistic mechanisms of these altered protein functions.
Using differential contact map analysis of crystallographic structures, Babu and colleagues recently suggested a universal activation mechanism of Gα [27]. In their model, structural contacts between α1 and α5 act as a ‘hub’ mediating the communications between α5 and the nucleotide. These contacts are broken upon the binding of receptor at α5, leading to a more flexible α1 and the destabilization of nucleotide binding. According to their studies, however, these critical α1/α5 contacts do not exist in Ras structures. Thus, they concluded that, unlike Gα, α5 in Ras does not have allosteric regulation of the nucleotide. It is worth noting that Babu’s work is purely based on the comparison of structures without considering protein dynamics. In fact, our study indicates that functionally important communications may not be directly observed from static structures. For example, the inter-lobe couplings between SII and SIII/α3 are not captured by PCA of structure ensemble, but they are clearly shown in our network analysis of structural dynamics. By inspecting structural dynamics, we find that α5 in Ras actually plays an allosteric role, in which point mutation (R164A) substantially disrupts the couplings in the nucleotide binding regions. The potential salt bridges between D47/E49 in L3 and R161/R164 in α5 are shown in S3 Fig.
A previous study of Ras GTPases via an elastic network model–normal mode analysis (ENM-NMA) revealed similar bilobal substructures and found that functionally conserved modes are localized in the catalytic lobe1, whereas family-specific deformations are mainly found in the allosteric lobe2 [35]. The subsequent study via MD, in constrast, indicated that the conformational dynamics of Ras and Gαt are distinct, especially in the GDP state [36]. We note that in that study only a single MD simulation trajectory was analyzed, which is insufficient to assess the significance of the observed difference. Moreover, few atomistic details were given in that work. In our study, we make improvements by building ensemble-averaged networks based on multiple MD simulations instead of a single trajectory. This increases the robustness of the networks and largely reduces statistical errors. In addition, our correlation analysis provides residue wise predictions of potential important positions that mediate communications between functional regions. Overall, separation of functionally conserved and specific residues in conformational dynamics provides us unprecedented insights into protein evolution and engineering.
Atomic coordinates for all available Ras, Gαt/i and EF-Tu crystal structures were obtained from the RCSB Protein Data Bank [37] via sequence search utilities in the Bio3D package version 2.2 [38,39]. Structures with missing residues in the switch regions were not considered in this study, resulting in a total of 143 chains extracted from 121 unique structures for Ras, 53 chains from 36 unique structures for Gαt/i and 34 chains from 23 unique structures for EF-Tu (detailed in S1–S3 Tables). Prior to analyzing the variability of the conformational ensemble, all structures were superposed iteratively to identify the most structurally invariable region. This procedure excludes residues with the largest positional differences (measured as an ellipsoid of variance determined from the Cartesian coordinate for equivalent Cα atoms) before each round of superposition, until only invariant “core” residues remained [40]. The identified “core” residues were used as the reference frame for the superposition of both crystal structures and subsequent MD trajectories.
PCA was employed to characterize inter-conformer relationships of both Ras and Gαt/i. PCA is based on the diagonalization of the variance-covariance matrix, Σ, with element Σij built from the Cartesian coordinates of Cα atoms, r, of the superposed structures:
Σij=<(ri‑<ri>)>·<(rj‑<rj>)>,
where i and j enumerate all 3N Cartesian coordinates (N is the number of atoms being considered), and <·> denotes the average value. The eigenvectors, or principal components, of Σ correspond to a linear basis set of the distribution of structures, whereas each eigenvalue describes the variance of the distribution along the corresponding eigenvector. Projection of the conformational ensemble onto the subspace defined by the top two largest PCs provides a low-dimensional display of structures, highlighting the major differences between conformers.
Similar MD simulation protocols as those used in [18] were employed. Briefly, the AMBER12 [41] and corresponding force field ff99SB [42] were exploited in all simulations. Additional parameters for guanine nucleotides were taken from Meagher et al. [43]. The Mg2+·GDP-bound Ras crystal structure (PDB ID: 4Q21), Gαt structure (PDB ID: 1TAG) and EF-Tu structure (PDB ID: 1TUI) were used as the starting point for GDP-bound simulations. The Mg2+·GNP (PDB ID: 5P21), the Mg2+·GSP (PDB ID: 1TND) and the Mg2+·GNP (PDB ID: 1TTT) bound structures were used as the starting point for GTP-bound simulations of Ras, Gαt and EF-Tu, respectively. These structures were identified as cluster representatives from PCA of the crystallographic structures. Prior to MD simulations, the sulfur (S1γ)/nitrogen (N3β) atom in the GTP-analogue was replaced with the corresponding oxygen (O1γ) / oxygen (O3β) of GTP. All Asp and Glu were deprotonated whereas Arg and Lys were protonated. The protonation state of each His was determined by its local environment via the PROPKA method [44]. Each protein system was solvated in a cubic pre-equilibrated TIP3P water box, where the distance was at least 12Å from the surface of the protein to any side of the box. Then sodium ions (Na+) were added to neutralize the system. Each MD simulation started with a four-stage energy minimization, and each stage employed 500 steps of steepest descent followed by 1500 steps of conjugate gradient. First, the atomic positions of ligands and protein were fixed and only solvent was relaxed. Second, ligands and protein side chains were relaxed with fixed protein backbone. Third, the full atoms of ligands and protein were relaxed with fixed solvent. Fourth, all atoms were free to relax with no constraint. Subsequent to energy minimization, 1ps of MD simulation was performed to increase the temperature of the system from 0K to 300K. Then 1ns of simulations at constant temperature (T = 300K) and pressure (P = 1bar) was further performed to equilibrate the system. Finally, 80ns of production MD was performed under the same condition as the equilibration. For long-range electrostatic interactions, particle mesh Ewald summation method was used, while for short-range non-bonded Van der Waals’ interactions, an 8Å cutoff was used. In addition, a 2-fs time step was use. The center-of-mass motion was removed every 1000 steps and the non-bonded neighbor list was updated every 25 steps.
We performed a total of 1,920 ns of MD simulation and analyzed results from multiple production phase 80ns simulations for each of our 3 systems, including the wild type in two nucleotide states along with 5 mutant ras, 8 mutant Gαt and 5 mutant EF-Tu systems (see full listing in S5 Table). The RMSD time courses for the above systems is shown in S4 Fig.
Consensus correlation networks were built from MD simulations to depict dynamic couplings among functional protein segments. A weighted network graph was constructed where each node represents an individual residue and the weight of edge between nodes, i and j, represents their Pearson’s inner product cross-correlation value cij [45] during MD trajectories. The approach is similar to the dynamical network analysis method introduced by Luthey-Schulten and colleagues [46]. However, instead of using a 4.5Å contact map of non-neighboring residues to define network edges, which were further weighted by a single correlation matrix, we constructed consensus networks based on five replicate simulations in the same way as described before [18].
Hierarchical clustering was employed to identify residue groups, or communities, that are highly coupled to each other but loosely coupled to other residue groups. We used a betweenness clustering algorithm similar to that introduced by Girvan and Newman [47]. However, instead of partitioning according to the maximum modularity score, which is usually used in unweighted networks, we selected the partition closest to the maximum score but with the smallest number of communities (i.e. the earliest high scoring partition). This approach avoided the common cases that many small communities were generated with equally high partition scores. The resulting networks under different nucleotide-bound states showed largely consistent community partition in Ras, Gαt and EF-Tu, with differences mainly localized at the nucleotide binding PL, SI, SII and α1 regions. To facilitate comparison between states and families, the boundary of these regions was re-defined based on known conserved functional motifs. Re-analysis of the original residue cross-correlation matrices with the definition of communities was then performed. Only inter-community correlations were of interest, which were calculated as the sum of all underlying residue correlation values between two given communities satisfying that the smallest atom-atom distance between corresponding residue pairs was less than 4.5Å (for Gαt and EF-Tu) or 6 Å (for Ras) for more than 75% of total simulation frames. A larger cutoff was selected for Ras because the overall residue level correlations are weaker in Ras. A standard nonparametric Wilocox test was performed to evaluate the significance of the differences of inter-community correlations between distinct states.
|
10.1371/journal.ppat.1006636 | Nanobodies targeting norovirus capsid reveal functional epitopes and potential mechanisms of neutralization | Norovirus is the leading cause of gastroenteritis worldwide. Despite recent developments in norovirus propagation in cell culture, these viruses are still challenging to grow routinely. Moreover, little is known on how norovirus infects the host cells, except that histo-blood group antigens (HBGAs) are important binding factors for infection and cell entry. Antibodies that bind at the HBGA pocket and block attachment to HBGAs are believed to neutralize the virus. However, additional neutralization epitopes elsewhere on the capsid likely exist and impeding the intrinsic structural dynamics of the capsid could be equally important. In the current study, we investigated a panel of Nanobodies in order to probe functional epitopes that could trigger capsid rearrangement and/ or interfere with HBGA binding interactions. The precise binding sites of six Nanobodies (Nano-4, Nano-14, Nano-26, Nano-27, Nano-32, and Nano-42) were identified using X-ray crystallography. We showed that these Nanobodies bound on the top, side, and bottom of the norovirus protruding domain. The impact of Nanobody binding on norovirus capsid morphology was analyzed using electron microscopy and dynamic light scattering. We discovered that distinct Nanobody epitopes were associated with varied changes in particle structural integrity and assembly. Interestingly, certain Nanobody-induced capsid morphological changes lead to the capsid protein degradation and viral RNA exposure. Moreover, Nanobodies employed multiple inhibition mechanisms to prevent norovirus attachment to HBGAs, which included steric obstruction (Nano-14), allosteric interference (Nano-32), and violation of normal capsid morphology (Nano-26 and Nano-85). Finally, we showed that two Nanobodies (Nano-26 and Nano-85) not only compromised capsid integrity and inhibited VLPs attachment to HBGAs, but also recognized a broad panel of norovirus genotypes with high affinities. Consequently, Nano-26 and Nano-85 have a great potential to function as novel therapeutic agents against human noroviruses.
| We determined the binding sites of six novel human norovirus specific Nanobodies (Nano-4, Nano-14, Nano-26, Nano-27, Nano-32, and Nano-42) using X-ray crystallography. The unique Nanobody recognition epitopes were correlated with their potential neutralizing capacities. We showed that one Nanobody (Nano-26) bound numerous genogroup II genotypes and interacted with highly conserved capsid residues. Four Nanobodies (Nano-4, Nano-26, Nano-27, and Nano-42) bound to occluded regions on the intact particles and impaired normal capsid morphology and particle integrity. One Nanobody (Nano-14) bound contiguous to the HBGA pocket and interacted with several residues involved in binding HBGAs. We found that the Nanobodies delivered multiple inhibition mechanisms, which included steric obstruction, allosteric interference, and disruption of the capsid stability. Our data suggested that the HBGA pocket might not be an ideal target for drug development, since the surrounding region is highly variable and inherently suffers from lack of conservation among the genetically diverse genotypes. Instead, we showed that the capsid contained other highly susceptible regions that could be targeted for virus inhibition.
| Human norovirus is recognized as the most important cause of outbreaks of acute gastroenteritis [1]. The virus is a non-enveloped single-stranded RNA virus within the Caliciviridae family. The human norovirus genome contains three open reading frames (ORFs), where ORF1 encodes non-structural proteins, ORF2 encodes the capsid protein (VP1), and ORF3 encodes the minor capsid protein (VP2). The virion comprises of 90 VP1 dimers that form an icosahedral particle (T = 3) 35–40 nm in diameter [2,3]. The VP1 can be expressed in insect cells and self-assembles into virus-like particles (VLPs) morphologically similar to native virions [4]. Smaller icosahedral particles (15–25 nm, T = 1), presumably composed of 30 VP1 dimers, can also self-assemble in insect cells and were found in patient stool specimens [5,6]. The X-ray crystal structure of norovirus native-size VLPs showed that the VP1 can be divided into shell (S) and protruding (P) domains that are connected via a flexible hinge [3]. The S domain forms the scaffold of the capsid, while the surface exposed P domains contain the main determinants of antigenicity and host binding epitopes.
Noroviruses are genetically diverse and can be divided into seven genogroups (GI-GVII) that are further subdivided into numerous genotypes [7]. The GII genotype 4 (GII.4) includes most epidemic and pandemic strains, while GII.17 was recently attributed with major outbreaks in East Asia [8]. Norovirus illness is typically self-limiting and usually subsides in several days. However, chronic infections in vulnerable individuals, such as the young and elderly, can lead to additional complications and even death [9–11]. Currently there are no available vaccines or antiviral treatments for human noroviruses, despite their discovery over four decades ago [2].
Recently, two cell culture systems have shown that human norovirus can replicate in B-cells or stem cell-derived human enteriods [12,13]. However, norovirus pathogenesis is still poorly understood and the interaction with the host receptor(s) is unclear. Nevertheless, histo-blood group antigens (HBGAs) have been shown to be important binding factors for human norovirus infections [14–17]. HBGAs are found as soluble antigens in saliva and are expressed on epithelial cells, which suggest that noroviruses may encounter HBGAs several times during the course of the infection. Soluble HBGAs may interact with virion particles prior attachment to cells [13] or function as binding factors on cell surfaces [12]. Until recently, the norovirus capsid was presumed to bind two HBGA molecules per VP1 dimer, however two additional HBGA binding sites were identified on the VP1 dimer, indicating that the interaction with HBGAs is rather complex [18,19].
Interestingly, the presence of serum antibodies that block norovirus binding to HBGAs has been associated with a decreased risk of infection and illness [12,20,21]. Moreover, in a recent enteroid norovirus replication system inhibition in the blocking assay was correlated with neutralization in cell culture. A recent study suggested that antibodies targeting the HBGA pocket could inhibit norovirus replication by steric interference with the GI.1 HBGA pocket [22]. A number of other studies have identified norovirus-specific monoclonal antibodies (mAbs) and single chain variable domains (VHH or Nanobodies) that could block norovirus VLP binding to HBGAs [20,23–27]. However, most of these antibodies and Nanobodies are genotype specific, which limits their therapeutic potential [28].
Apart from the HBGA binding site, other neutralizing epitopes likely exist. For example, upon binding to cell receptors, picornaviruses, which are structurally similar to noroviruses, initiate multiple structural rearrangements from the mature capsid to expanded intermediate forms, leading to externalization of the internal polypeptide, membrane fusion and release of viral RNA [29]. Neutralizing Nanobodies that interfere with the conformational rearrangement of the capsid were recently reported for poliovirus [30]. In that study, Nanobodies were used to trap transitional conformations of the viral capsid, which occur during cell entry and are required for the receptor binding.
There is still limited information on norovirus particle attachment to cell surfaces and rearrangements during cell entry. Defining the structural dynamics of the norovirus particles during an infection could show transient conformations related to specific functions in the virus life cycle. These snapshots of the particle dynamics could be obtained from the reconstruction of the capsid protein complexes with antibody fragments or Nanobodies. Moreover, the structural analysis could offer insights into vulnerable regions on the capsid that could be targeted by inhibitors. Indeed, we recently discovered that a human norovirus specific Nanobody (termed Nano-85) bound to intact norovirus VLPs and the Nanobody binding interaction caused the VLPs to disassemble [31]. Our results suggested that the Nano-85 binding epitope might represent a vulnerable region on the capsid that is important for the structural integrity.
In the current study, we analyzed a novel panel of norovirus-specific Nanobodies in order to identify other vulnerable regions. The Nanobody binding epitopes were determined using X-ray crystallography and the specific binding interactions were correlated with a surrogate neutralization assay. We found that Nanobody binding could trigger capsid deformation and increase proteolytic degradation of capsid protein, ultimately exposing viral RNA. Our new findings showed that norovirus particles have vulnerable epitopes that were indispensable for capsid assembly, structural integrity and HBGA attachment.
We analyzed six new Nanobodies (Nano-4, Nano-14, Nano-26, Nano-27, Nano-32, and Nano-42), which belonged to six groups based on the sequence similarity. The amino acid sequence identity ranged from 65 to 80%, with most sequence variations located in CDR regions. Nano-32 and Nano-4 had exceptionally long CDR3 loops, with 29 and 18 residues, respectively. Nano-32 had an additional disulfide bridge connecting CDR2 and CDR3. Nanobodies with variable CDR loops were expected to bind to distinct epitopes on the capsid.
The Nanobody binding specificities were initially confirmed with the immunization antigen (i.e., GII.10 VLPs) and the corresponding GII.10 P domain (S1 Fig). All six Nanobodies bound to the GII.10 VLPs and P domain, which indicated that the S domain did not contain any Nanobody binding epitopes. Nano-42, Nano-14, Nano-26, and Nano-4 showed the strongest binding capacities (0.02–0.2 μg/ml), whereas Nano-27 and Nano-32 had lower binding ability (1.5 and 0.2 μg/ml, respectively). Following these results, the cross-reactivities were analyzed with a panel of VLPs and P domains from various GI (GI.1 and GI.11) and GII (GII.1, GII.2, GII.4 2006 and 2012, GII.10, GII.12, and GII.17) genotypes (Fig 1). Nano-85 exhibited the broadest recognition range, detecting GI.11 VLPs and numerous GII P domains. Nano-26 also showed broad reactivity, detecting all GII genotypes. Nano-4 and Nano-42 showed limited cross-reactivity, while Nano-27, Nano-32, and Nano-14 were GII.10 specific (S1 Fig).
In order to determine the HBGA blocking potential of the Nanobodies, a surrogate neutralization assays were performed using GII.10 and GII.4 VLPs. Three Nanobodies (Nano-14, Nano-32, and Nano-26) inhibited the binding of GII.10 VLPs to PGM in a dose-dependent manner (IC50 = 1.7 to 6.6 μg/ml) (Fig 1E). Similarly, Nano-14, Nano-26, and Nano-32 inhibited binding to A-type saliva (IC50 = 0.3 to 3.1 μg/ml) and B-type saliva (IC50 = 1.1 to 4.3 μg/ml) (S2A and S2B Fig). Nano-85 was relatively ineffective in blocking the GII.10 VLPs to PGM or B-type saliva (IC50 > 70 μg/ml) and weakly blocked GII.10 VLPs to A-type saliva (IC50 = 12 μg/ml). Nano-4, Nano-25, Nano-27, and Nano-42 demonstrated no inhibition of GII.10 VLPs. Additionally, both Nano-26 and Nano-85 blocked GII.4 VLPs from binding to PGM (Fig 1F) (IC50 2.4 μg/ml and 3.1 μg/ml, respectively) and B-type saliva (IC50 0.7 μg/ml and 1.2 μg/ml, respectively) (S2C Fig). To demonstrate that Nano-26 and Nano-85 specifically inhibit VLP binding to HBGAs present in PGM and saliva, a blocking assay using synthetic HBGAs was performed (S2D Fig). Nano-26 and Nano-85 blocked GII.4 VLPs from binding to synthetic B-tri saccharide with IC50 ranging between 1 μg/ml to 10 μg/ml. Nano-4 and Nano-42 did not inhibit GII.4 VLPs from binding to PGM.
The thermodynamic properties of Nanobodies binding to GII.10 P domains were analyzed using ITC (Table 1). Most of the Nanobodies (Nano-4, Nano-14, Nano-26, Nano-27, and Nano-42) exhibited exothermic binding with nanomolar affinities (S3 Fig). The binding reaction was driven by a large enthalpy change and was characterized with unfavorable entropy of the binding. This suggested that the net formation of non-covalent bonds between the Nanobody and the P domain was a major contributor to the binding. The stoichiometry indicated the binding of one Nanobody molecule per P domain monomer for all Nanobodies, except Nano-14, where the ratio of P domain:Nanobody was 2:1. Nano-32 binding was characterized by a positive enthalpy change associated with endothermic type of reaction (S3 Fig). Instead, a large positive entropy was the main contributing factor to the ΔG. These different thermodynamic parameters were likely associated with the distinct binding epitope of Nano-32, as presented below.
The structures of GII.10 P domain in complex with Nano-14, Nano-26, Nano-27, Nano-32, and Nano-42 were solved using X-ray crystallography (Table 2). Additionally, the X-ray crystal structure of GII.17 P domain with Nano-4 was determined in order to explain Nanobody cross-reactivity binding interactions at the atomic level. We also solved a double complex structure of GII.10 P domain with Nano-26/Nano-85, which permitted a higher resolution than the GII.10 P domain and Nano-26 complex alone, and explained how two distinct Nanobodies bound simultaneously to one P dimer. The overall structure of the P domains in all complex structures was reminiscent of unbound P domain with limited structural changes observed upon binding of the Nanobodies. All Nanobodies had typical immunoglobulin fold and interacted with the P domain primarily with CDR loops. The electron densities for Nano-4, Nano-32, Nano-26, Nano-27, and Nano-42 were well resolved, whereas for Nano-14, the distant part of a Nanobody close to the two-fold crystallographic symmetry axis was partially disordered. Overall, we could separate six Nanobodies into three distinct binding regions on the P domain: termed top, side, and bottom. Nano-4, Nano-26, Nano-27, and Nano-42 bound to the bottom; Nano-32 bound to the side; and Nano-14 bound on the top of the P domain (Fig 2 and Table 3). The binding sites of Nano-4 and Nano-27 partially overlapped the Nano-85 binding site. Moreover, Nano-42 bound with almost identical orientation as Nano-85.
To support our structural data and exclude the possibility of less probable orientations derived from the crystal packing we performed competitive ITC measurements. Nano-85 showed no binding to the P domain pre-incubated with Nano-4, Nano-27 and Nano-42, indicating that these Nanobodies competed for the same binding region on the P domain (S4 Fig). On the contrary, when Nano-85 titrations were performed to the P domain pre-mixed with Nano-14 or Nano-26 the binding isotherm was reminiscent of the Nano-85 P domain measurement. These data implied that Nano-14 and Nano-26 bound to sites distinct from Nano-85, whereas other Nanobodies competed with the Nano-85 epitope. Therefore, these P domain Nanobody complex structures clearly represented the precise Nanobody binding epitopes.
The structure of GII.10 P domain Nano-14 complex was solved to 1.8Å resolution. Nano-14 bound on the top of the P domain in the grove located between the two P domain monomers (Fig 3A and 3B). A vast network of hydrogen bonds was formed between Nano-14 and both P domain monomers. The majority of interactions were built between one P domain monomer (chain A) and CDR3 of the Nano-14 (Fig 3C). Six P domain residues (chain A: Arg299, Trp381, Lys449, Asp403, and Glu333; chain B: Gln384) formed eleven direct hydrogen bonds with Nano-14. Four electrostatic interactions were observed between Nano-14 and the P domain residues Arg299 and Glu382 (chain A). The numerous hydrogen bonds and electrostatic interactions corresponded well with the large negative binding enthalpy (S3 Fig). Five P domain residues (His298, Val361, Ala363, Arg299, and Trp381) were involved in eight hydrophobic interactions with Nano-14. Two additional interactions were observed: P domain His358 (chain B) formed a π-sulfur interaction with Nanobody Met106, whereas P domain Glu333 (chain A) participated in a π donor hydrogen bond with Phe102.
The nine P domain residues involved in Nano-14 binding were predominantly variable (see Fig 4). This finding corresponded nicely with the ELISA data that showed Nano-14 was GII.10 specific. Remarkably, the Nano-14 binding site, which is largely formed by CDR3 loop, extended between two HBGA binding pockets (Fig 2). Such strategic positioning of Nano-14 resulted in steric interference with the two conventional HBGA binding sites and the two newly identified HBGA binding pockets [18]. Moreover, three P domain residues (Trp381, Glu382 and Lys449) were directly involved in binding HBGAs [32] and Nano-14, indicating a direct competition for the HBGA pocket. Importantly, analysis of the Nano-14 binding site with the ELISA blocking data provided a novel structural basis of GII HBGA binding interference (see Fig 1).
The Nano-32 binding site was located on the side of the GII.10 P domain in a cleft between two P domain monomers (Fig 5A and 5B). In the P domain Nano-32 complex, several P domain loops were slightly shifted compared to the unliganded P domain (S5 Fig) (chain A: residues 487–491 and 517–522; chain B: residues 309–314, 287–300, and 418–421). Moreover, a P domain loop (residues 343–352) was shifted ~4.3Å from the loop in the unliganded structure. Several residues within this loop were also disordered, suggesting a certain degree of P domain flexibility. The loop containing residues 295–300 was positioned identically in both monomers in contrast to the usual asymmetric orientation in unliganded structure [18]. These conformational rearrangements likely correlated with the major entropy change observed in ITC measurements (Table 1). Nano-32 was essentially held equally with two P domain monomers (Fig 5C). Four P domain residues from chain A (Arg287, Asn344, Trp343, and Asp316) and two residues from chain B (Arg492 and Thr519) formed seven direct hydrogen bounds with Nano-32. Several P domain residues were also involved in electrostatic interactions (chain A: Arg287 and Asp247; chain B: Glu236) and hydrophobic interactions (chain A: Pro314; chain B: Val248 and Pro518). Six P domain residues involved in Nano-32 binding were highly variable and five residues were conserved in GII.4 and GII.10 noroviruses (Fig 4). Although Nano-32 strongly inhibited binding of GII.10 VLPs to HBGAs, none of the residues were shared between the HBGA pockets and the Nano-32 binding site. This result suggested that Nano-32 indirectly interfered with the HBGA pockets or utilized another mechanism to inhibit HBGA binding.
We solved the structure of GII.17 P domain Nano-4 complex, since the GII.17 norovirus was of recent clinical concern; and we wanted to analyze the cross-reactive epitopes at the atomic level. According to the ELISA data, Nano-4 bound strongly to the GII.17 VLPs. X-ray data for GII.17 P domain Nano-4 complex was processed to 1.7Å resolution in C121 space group. Nano-4 bound to the bottom of the P domain in close proximity to the previously identified Nano-85 binding site (Fig 2 and Fig 6A and 6B) [31]. An extensive network of direct hydrogen bonds was formed between P domain residues (Thr483, Glu486, Asp516, Asn520, Tyr523, and Ser524) and Nano-4 (Fig 6B). Two P domain residues were involved in hydrophobic interactions (Tyr523 and Ala526) and five electrostatic interactions (Arg482, Glu486, and Asp516) contributed to Nano-4 binding. Only three of nine P domain residues interacting with Nano-4 were variable (Fig 4). The six conserved residues provided a possible explanation for the broad cross-reactivity exhibited with Nano-4 (Fig 1). The Nano-4 binding epitope was located on the opposite side of the HBGA pocket, an observation that is supported by the lack of blocking potential in the surrogate neutralization assay.
Nano-42 bound on the bottom of the P domain and closely overlapped with Nano-85 binding site (Fig 6C). Five direct hydrogen bonds involved P domain residues (Asp526, Trp528, Asn530, and Thr534) and Nano-42 residues (Fig 6D). Two hydrophobic interactions were formed between P domain residues Val529 and Ala536 and Nano-42 residues Tyr100 and Val54, respectively. Interestingly, the Nanobody was also held by three additional hydrogen bonds mediated by an ethylene glycol molecule. Ethylene glycol interacted with P domain residues Arg484 and Asp526 on one side and Nano-42 residues Thr31 and Ser53 on the other side. Moreover, six water mediated bonds provided additional stabilization of the bound Nano-42. Although Nano-42 binding residues were mainly conserved in GII noroviruses and were identical between GII.4 2006 and 2012 strains, Nano-42 apparently distinguished these two strains in the ELISA cross-reactivity study (Fig 1). In addition, although the binding epitope of Nano-42 was rather similar to that of Nano-85, Nano-42 did not inhibit VLP binding to HBGAs.
Nanobodies were previously shown to aid the crystallization process by increasing protein stability and stabilizing flexible regions [33]. We have already utilized Nano-85 to obtain high-resolution complex structures with three different norovirus P domains [31]. Herein, we used Nano-85 to improve the resolution of the GII.10 P domain Nano-26 complex structure and describe the synchronized binding of two Nanobodies. The initial structure of GII.10 P domain Nano-26 complex was solved to ~3Å resolution. A single crystal of GII.10 P domain Nano-85/Nano-26 double complex diffracted to 2.3Å in C121 space group. Binding epitopes and interactions of both Nanobodies were identical to those in the individual complexes [31]. Nano-26 bound at the bottom of the P domain, perpendicular to Nano-85 binding site (Fig 7A).
Nano-26 binding site comprised of residues from both P domain monomers, although the majority of the P domain interactions involved only one chain (chain B). Nano-26 formed seven direct hydrogen bonds with one P domain monomer (chain B: Asp269, Leu272, Gly274, Gln471, Glu472, and Thr276) (Fig 7B). Both P domain monomers were involved in hydrophobic interactions (chain A: Ile231, Pro488; and chain B: Tyr470 and Pro475) with Nano-26. In addition, two electrostatic interactions contributed to the tight binding. Nano-26 binding residues were mainly conserved between GII genotypes, which correlated well with the broad recognition shown with ELISA (Figs 1 and 4). Although the binding site was distant from the HBGA binding pocket, Nano-26 had a high inhibition capacity in the blocking assay, which also suggested indirect HBGA interference.
The Nano-27 binding epitope was located on the bottom region of the P domain (Fig 8). Interestingly, the binding site partially overlapped the Nano-4 binding site. Six P domain residues (Arg484, Gly491, Arg492, Thr493, Glu496, and Thr534) were involved in ten direct hydrogen bonds and two electrostatic interactions. Three residues (Arg484, Ala536, and Pro537) were involved in four hydrophobic interactions with Nano-27. The Nano-27 binding site comprised six conserved residues and two variable residues. The ELISA data showed that Nano-27 was strain specific, which indicated that certain variable residues likely play a crucial role in cross-reactivity (Fig 4). Similarly to Nano-4, Nano-27 also failed to block VLP binding to HBGAs.
We previously showed that Nano-85 was able to disassemble norovirus VLPs [31]. To explore if these six newly identified Nanobodies had a similar ability, we treated native-size VLPs with Nanobodies and examined the treated-particle morphology using EM. Overall, three distinct VLP structural modifications were observed with Nanobody treatment (Fig 9). In the first case, Nano-85 and Nano-26 treatment partially disassembled and deformed the native-size VLPs. Nano-85 treatment also produced a minor fraction of small-size VLPs (20–23 nm). In the second case, Nano-4, and Nano-27 treatment induced a conformational transition from native-size VLPs (35–38 nm) to the small-size VLPs. In case of Nano-42, small and disassembled particles were equally present after treatment. In the third case, Nano-32 treatment produced large aggregates of apparently intact native-size VLPs. None of these effects were observed with Nano-14 treatment.
To investigate a temperature dependence of the Nanobody treatment, we mixed GII.10 VLPs with Nano-85 and Nano-26 at 4°C, room temperature, and 37°C for 30 minutes (S6 Fig). Nano-85 treated VLPs showed a continuous degradation of native-size particles, producing small and/or partially broken particles as major intermediate forms. Nano-26 was more effective across the temperature range and almost completely altered the VLP integrity. The combination of Nano-85 and Nano-26 appeared to cause a more intense degradation of VLPs. The temperature dependence of Nano-85 induced morphological changes indicated the involvement of capsid “breathing” in the disassembly process.
We also performed DLS measurements to quantitatively evaluate GII.10 VLP heterogeneity after Nanobody exposure. Nano-14 treated VLPs had almost identical diameters to native-size particles (37 nm and 35 nm, respectively) (Fig 10A and 10B). Nano-32 treated VLPs displayed 10,000 times increased diameters, confirming the formation of the large aggregates observed using EM. Nano-26 and Nano-85 treated VLPs mainly formed VP1 protein aggregates, although a small peak corresponding to native-size particles remained. Nano-4, Nano-27, and Nano-42 treated VLPs showed peaks corresponding to small-size VLPs (21–23 nm). Overall, the DLS analysis corresponded well with the EM results and provided additional evidence that Nanobody treatment altered the capsid structural integrity.
Two Nanobodies, Nano-26 and Nano-85, exhibited broad cross-reactivities coupled with adverse effects on capsid integrity. To understand if these effects were relevant for clinically important norovirus strains, GII.4 (Sydney 2012) and GII.17 VLPs were treated with Nano-26 and Nano-85 (Fig 11). Both Nanobodies lead to malformed and aggregated GII.4 VLPs and produced only a few small-size particles. In the case of GII.17 VLPs, Nano-26 treatment caused the formation of small-size VLPs, whereas Nano-85 seemed to have no notable effect on the particle size. These EM results were supported with the DLS measurements (Fig 11). Overall, these results suggested that effects of the Nanobody treatment might vary among different genotypes.
To further evaluate the binding effects of Nano-26 and Nano-85 on GII.4 (2012) VLPs, we performed time-, temperature-, and concentration-dependent DLS measurements (S7 and S8 Figs). Nano-26 induced changes in the VLP size distribution after 30 seconds, whereas for Nano-85 15 minutes were required to observe the first noticeable effects (S7 Fig). Fluctuations in VLP sizes were more evident at 37°C for both Nano-26 and Nano-85 after 15 minutes incubation (S7A and S7B Fig). Nanobody effects were also concentration dependent, with minimum concentrations of 12.5 μM and 50 μM required for Nano-26 and Nano-85, respectively (S7C and S7D Fig). These results suggested that one Nano-26 molecule per VP1 dimer was sufficient to cause morphological changes, whereas Nano-85 required >2 times molar excess.
In order to examine the Nanobody effects on norovirus virions, we implemented a modified RNA exposure assay and viral loads were quantified using real-time RT-PCR. Concentrated GII.4 positive stool samples were treated with the broadly reactive Nano-26 and Nano-85, while Nano-14 was used as a negative control and 250 mM citric buffer was used as a positive control. Treated samples were then subjected to RNAse digestion. Nano-26, Nano-85, and citrate treated stool samples showed reduced genome copy numbers compared to the Nano-14 control (approx. 30 times for Nano-26 and Nano-85 and 250 times for citrate) (Fig 10C). These results suggested that the Nano-26 and Nano-85 opened the virions and released the viral RNA, which was degraded by RNAse. To evaluate the Nanobody effects on norovirus virions more directly, we used a stool sample where RNA degradation was not detected and performed RNA extraction with incomplete lysis step (Fig 10D). Additional degradation caused by Nanobodies or citrate lead to an increased number of genome copies compared to untreated samples. Indeed, Nano-26, Nano-85, and citrate treated samples had higher RNA levels than in the control samples (PBS or Nano-14) (Fig 10D). Although, the fold increase was relatively small (5–7 times), the difference was significant.
To further investigate if Nanobody treatment could render norovirus VLPs vulnerable to proteolytic cleavage, we subjected GII.10, GII.4, and GII.17 VLPs to a 30-minute trypsin digestion after Nanobody exposure and observed the protein bands using SDS-PAGE (S10 Fig). Nano-14 treated VLPs produced similar bands as the untreated VLPs. Nano-26 and Nano-85 treatment resulted in multiple cleavage products for GII.10 and GII.4 VLPs. In the case of GII.17 VLPs, only Nano-26 treatment showed additional cleavage of the capsid protein. Overall, these results suggested that Nano-85 and Nano-26 caused the particles to become structurally unstable, more vulnerable to proteolytic cleavage, and viral RNA exposure.
Structural information of antibody and Nanobody binding sites can be instrumental for understanding the neutralizing and immuno-dominant epitopes as well as motion dynamics of the viral capsid. Numerous neutralizing mAbs have been identified in recent years with diverse neutralization mechanisms [34]. One of the most direct mechanisms is blocking the receptor binding sites. Such neutralizing mAbs and Nanobodies were previously identified for influenza virus, HIV, herpes simplex virus, rhinovirus, and others [35–41]. For example, in the case of HIV, with the aid of an extra long CDR3 loop, the neutralizing Nanobody D7 effectively competed for the CD4 binding site on gp120 protein [42]. Previously described Nanobodies and mAbs with therapeutic potential against human norovirus were also proposed to interfere with the HBGA binding site [20,22–27]. MAb termed NV8812 bound to a conformational epitope on the GI.1 P domain and blocked the binding of norovirus VLPs to human and animal cell lines [24]. Four α-GI mAbs isolated from chimpanzees challenged with norovirus blocked VLP binding to carbohydrates and inhibited hemagglutination, although their precise binding sites were not described [20]. Recently, a GI.1 specific mAb was discovered that sterically hindered the HBGA pocket [22]. In our study, we showed that Nano-14 overlapped with the GII.10 HBGA binding sites and inhibited HBGA binding by steric interference and competition for the pocket. The blocking abilities of Nano-14 were also comparable to previously reported blocking Nanobodies (IC50 = 0.34–2.0 μg/ml) [23], scFv fragments (IC50 = 0.3–1.5 μg/ml) [43], and mAbs (IC50 = 0.12–0.74) μg/ml [20,28]. Although exhibiting high inhibition capacity, these mAbs and Nanobodies tend to be strain specific.
The use of mAbs or Nanobodies directed to the HBGA pocket may inherently suffer from the variations and constantly changing amino acids in this region. Therefore, there is a need to identify additional neutralization epitopes, which are less susceptible to sequence variations. Indeed, Nano-32 recognition epitope was distant from the HBGA binding pocket and blocked VLP binding to HBGAs. A similar phenomenon was previously discussed with the norovirus specific blockade mAb NVB71.4, where neither particle disassembly nor steric hindrance could explain NVB71.4 blockade activity [25]. However, it was suggested that the NERK motif (residues 310, 316, 484, and 493 according to GII.4 numbering) could function as a conformational regulator through an allosteric effect [25]. Interestingly, two of these residues were directly involved in Nano-32 binding, suggesting a similar blockage mechanism as observed with mAb NVB71.4. Nano-32 induced conformational rearrangement of several P domain loops, which in turn altered the hydrophobic landscape of the P domain surface. This rearrangement likely caused the particle aggregation leading to interference at the HBGA binding pocket. An inhibition mechanism by allosteric interference was previously described for highly neutralizing mAbs against HIV and dengue virus [44,45]. Also, a recent study showed that the PGT121 mAbs against HIV gp121 protein inhibited CD4 binding, although the binding epitope was remote from the CD4 binding site. Moreover, dengue virus neutralizing mAb 1A1D-2 bound to a partially occluded epitope on envelope glycoprotein E and promoted particle reorganization [45]. These changes in viral surface were likely responsible for the inhibitory properties by this mAb.
To allow structural rearrangement to occur during viral entry and uncoating, the viral capsid proteins need to be exceptionally dynamic. Internal plasticity and motions of the capsid proteins can allow access to buried regions, which often play an important role in the viral life cycle [46]. Indeed, multiple antibodies against picornaviruses and flaviviruses that bind to normally inaccessible sites on the viral capsid were shown to be highly neutralizing [47–52]. For rhinoviruses and polioviruses, buried regions of internal VP4 protein are transiently exposed due to the capsid “breathing” and are targeted by neutralizing mAbs. These cryptic epitopes are often very conserved and therefore provide cross-serotypic neutralization. We previously identified a broadly reactive norovirus mAb [53] and Nano-85 that bound to a conserved region that was occluded in the context of native-size particles [31]. Here, we identified four novel Nanobodies (Nano-4, Nano-26, Nano-42, and Nano-27) that bound to the similar internal and poorly accessible epitopes as Nano-85. In comparison with Nano-85, the binding sites of Nano-27 and Nano-4 were located closer to the P domain crown. In context of the complete particle, this position had fewer steric clashes with neighboring P domains. The Nano-26 epitope was located at the bottom of the P domain, albeit perpendicular to Nano-85 binding site. Although completely different, Nano-26 recognition epitope was also conserved and poorly accessible (Fig 7). The time- and temperature-dependence of the Nanobody-induced degradation suggested an important role of conformational mobility and capsid “breathing” in Nano-85 and Nano-26 binding to these hidden epitopes [46].
Due to their small size and high affinities, the rapid binding of the Nanobodies provided a means to trap transiently exposed regions, otherwise buried in the native state of the particles. Trapping the particles in a particular conformation or otherwise inhibiting capsid “breathing” is a common antiviral strategy shared by many neutralizing mAbs, Nanobodies, and drugs against HIV, flaviviruses, picornaviruses, influenza, and others [54–60]. For example, several neutralizing Nanobodies against poliovirus and respiratory syncytial virus were shown to specifically stabilize either the native or expanded conformation of capsid, preventing it from further rearrangement necessary for the infection process [30,61]. It is plausible that in the case of norovirus Nanobodies described here, binding resulted in a stabilization of the particular P domain conformation, thus reducing the mobility and influencing the position on the S domain. The interaction between S and P domains was previously shown to control the size and stability of the GI.1 norovirus capsid [62]. Superposition of P domain Nano-26 complex on the cryo-EM VLP structure revealed an extensive clash with the S domain (Fig 12). Nano-26 binding likely disrupted normal S-P domain orientations, which consequently resulted in particle disassembly. Nano-26 required less time and concentration to achieve particle disassembly than Nano-85. This observation suggested that the restriction of a normal S-P domain relationship had a more destabilizing effect than interference with P-P domain interactions. Of note, only Nano-26 was able to influence the morphology of GII.17 VLPs, whereas GII.17 VLPs tolerated Nano-85 binding. Apparently, Nano-26 binding stabilized the S-P domain conformation that was incompatible with the morphology of native-size GII.17 particles, but supported the formation of small-size VLPs. Furthermore, three other Nanobodies, Nano-27, Nano-4 and Nano-42, drove a shift from native-size GII.10 VLPs to a smaller-size form. Likely, these Nanobodies could selectively stabilize the A/B conformation of the P dimer. The inability of the A/B dimer to reassemble into C/C dimers could lead to the formation of small particles, where all dimers are identical and resemble A/B dimer for T = 3 capsids.
Interference with the capsid motions and integrity provides one possible explanation for the blocking properties of both Nano-26 and Nano-85 in the surrogate neutralization assay. Nanobody binding caused the loss of normal VLP morphology and the treated VLPs showed a reduced signal in the blocking assays. Indeed, chemically disassembled VLPs showed no binding in a PGM assay (S10 Fig). These observations support the assumption that Nano-85 and Nano-26 inhibited the binding of norovirus VLPs to HBGAs by compromising capsid morphology instead of directly competing for the HBGA pocket. Interestingly, Nano-42, Nano-27, and Nano-4, which stimulate the formation of small-size particles, did not interfere with the attachment to the HBGAs. It was previously shown that small-size VLPs effectively bound to the surface of CaCo2 cells and competed with the native-size VLPs [24]. Apparently, the small-size VLPs that resulted from Nanobody exposure were equally able to bind HBGAs.
Intriguingly, our structural data indicated that closely overlapping epitopes are responsible for distinct functions. A striking example is Nano-42 and Nano-85, which despite having almost identical binding footprints, showed distinct binding and blocking properties. Nano-42 seemed to be less effective in disassembling the VLPs compared to Nano-85. Similar observations were previously reported for 80S poliovirus specific Nanobodies, where despite identical binding sites, the structures of the expanded virus differed in each complex [48]. Likewise, although Nano-4 and Nano-27 shared five of eight binding residues, Nano-27 was strain specific, whereas Nano-4 was cross-reactive. Even though the epitopes closely overlap with Nano-85 binding site, these Nanobodies did not exhibit blocking properties. Analysis of GII.10 P domain residues involved in Nano-27, Nano-4, and Nano-42 binding suggested that residues 484, 491–493, and 496 might constitute the molecular switch responsible for preferential assembly of small particles. Thus, additional high-resolution structural information could be instrumental in understanding epitope-function relationships by providing the exact location and interactions of the binding partners. This information might remain elusive when more general epitope mapping methods are used.
In addition to identification of functional epitopes on the norovirus capsid, our data provided insights of Nanobody potential neutralization properties in context of infectious norovirus virions. Recently, it was shown that silver dihydrogen citrate exposure compromises GII.4 VLPs integrity and facilitates viral RNA degradation [63]. Similarly, we showed that Nanobody-induced morphological changes of norovirus capsid resulted in exposure of viral RNA from the norovirus virions in clinical samples. The naked RNA was especially vulnerable to RNAse digestion and a similar RNA degradation assay was shown to greatly reduce the infectivity of murine norovirus [64]. In addition to exposing the viral RNA, Nanobodies increased the susceptibility of capsid protein to proteases, which are abundant in the gut. Although the exact role of proteolytic cleavage in the norovirus life cycle is largely unknown, cleaved capsid protein was shown to lose the ability to bind HBGA and maintain capsid assembly [65].
In summary, we identified several Nanobodies that impaired normal capsid motions, assembly, and integrity with subsequent release of viral RNA. Four Nanobodies blocked norovirus binding to cell attachment factors (HBGAs), utilizing three distinct inhibition mechanisms: steric occlusion of the HBGA binding site, allosteric interference, and violation of normal capsid morphology. Therefore, Nanobodies could act as broad inhibitors in multiple stages of the norovirus life cycle. The Nanobody capacity to inhibit human norovirus infections in the recently developed cell culture needs to be further evaluated. Nevertheless, the extensive evidence that interference with viral capsid dynamics could impair normal functioning suggested that Nanobodies could become effective norovirus therapeutics in future.
The norovirus P domains, GI.1 (Norwalk virus, Genbank accession number M87661), GI.11 (Akabane, EF547396), GII.1 (Hawaii, U07611), GII.2 (Snow Mountain, AY134748), GII.4 (Sydney-2012, JX459908 and Saga4 2006, AB447457), GII.10 (Vietnam026, AF504671), GII.12 (Hiro, AB044366), and GII.17 (Kawasaki308, LC037415 were expressed in E.coli, purified and stored in GFB (25mM Tris-HCl pH7.6, 0.3M NaCl) [66]. The full-length capsid genes, GI.1 (AY502016.1), GI.11, GII.1, GII.2, GII.4, GII.10, GII.12, and GII.17, were expressed in insect cells using the baculovirus expression system and stored in PBS [67,68].
Norovirus specific Nanobodies were produced at VIB Nanobody service facility, Belgium as previously described [31]. Briefly, a single alpaca was injected with GII.10 VLPs. A VHH library was constructed from isolated peripheral blood lymphocytes and screened for the presence of antigen-specific Nanobodies using phage display. Thirty-five Nanobodies were isolated and allocated to 17 distinct groups based on a sequence alignment. Six Nanobodies (Nano-4, Nano-14, Nano-26, Nano-32, Nano-42, Nano-27, and Nano-8) that represented different groups were analyzed in this study. The Nanobody genes were cloned to pHEN6C vector, expressed in WK6 E.coli cells, purified and stored in PBS or GFB.
Nanobody titers to norovirus P domains or VLPs were quantified with direct ELISA (17). Briefly, microtiter plates were coated with 7 μg/ml of GII.10 P domains or 2 μg/ml of GII.10 VLPs. For cross-reaction experiments, 15 μg/ml P domain and 4 μg/ml VLPs were coated on ELISA plates. The VLPs or P domain were detected with serially diluted Nanobodies and HRP-conjugated mouse α-His-tag monoclonal antibody. Absorbance was measured at 490 nm (OD490) and all experiments were performed in triplicate.
Pig gastric mucin (PGM) and saliva blocking assays were performed as previously described [69]. Briefly, ELISA plates were coated with 10 μg/ml PGM (Sigma, Germany) or with saliva type A or B diluted in PBS 1:2000. Nanobodies were two-fold serially diluted in PBS containing 2.5 μg/ml GII.10 VLPs (for PGM assay), 0.5 μg/ml GII.10 VLPs (for saliva assay) or 0.5 μg/ml GII.4 2006 VLPs (both PGM and saliva assay) and incubated for 1 h at RT. The VLPs-Nanobodies mixture was added to the plates and bound VLPs were detected with a α-GII.10 or α-GII.4 VLPs rabbit polyclonal antibody. For synthetic HBGA blocking assay, 10 μg/ml synthetic blood type B trisaccharide amine derivative (Dextra, UK) was coated on Pierce maleic anhydride activated plates (Thermo Fisher Scientific) overnight at 4C. Serially diluted Nanobodies were pre-incubated with 5 μg/ml GII.4 VLPs for 1h at RT. Following steps were performed as above. The binding of VLPs-only was set as a reference value corresponding to a 100% binding. The half maximal inhibitory concentrations (IC50) values for Nanobody inhibition were calculated using GraphPad Prism (6.0a).
Isothermal Calorimetry (ITC) experiments were performed using an ITC-200 (Malvern, UK). Samples were dialyzed into the identical buffer (GFB or PBS) and filtered prior titration experiments. Titrations were performed at 25°C by injecting consecutive (1–2 μl) aliquots of Nanobodies (100–150 μM) into P domain (10–20 μM) with 150 second intervals. The binding data was corrected for the heat of dilution and fit to a one-site binding model to calculate the equilibrium binding constant, KA, and the binding parameters, N and ΔH. Binding sites were assumed to be identical. For the competitive ITC measurements, the P domain was mixed with Nano-4, Nano-42, and Nano-27 in a 1:1 molar ratio. Titrations with Nano-85 were then performed as above.
P domain and Nanobody complexes were purified by size exclusion chromatography (39). The P domain and Nanobody complexes were crystallized using the following conditions: GII.10 P domain Nano-26/Nano-85 [0.1 M sodium citrate, 40% (w/v) PEG600]; GII.17 P domain Nano-4 [0.2 M calcium acetate, 10% (w/v) PEG8000, 0.1 M imidazole (pH 6.5)]; GII.10 P domain Nano-42 [0.2 M potassium iodide, 20% (w/v) PEG3350]; GII.10 P domain Nano-14 [0.1 M sodium citrate (pH 5.5), 20% (w/v) PEG3000]; GII.10 P domain Nano-32 [0.2 M magnesium formate]; and GII.10 P domain Nano-27 [2 M sodium chloride, 0.1 M sodium acetate]. Crystals were grown in a 1:1 mixture of the protein sample and mother liquor at 18°C. Prior to data collection, crystals were transferred to a cryoprotectant containing the mother liquor in 30% ethylene glycol, followed by flash freezing in liquid nitrogen.
X-ray diffraction data were collected at the European Synchrotron Radiation Facility, France at beamline BM30, ID30A, ID23-1 A and processed with XDS [70]. Structures were solved by molecular replacement in PHASER Phaser-MR [71] using GII.10 P domain (PDB ID 3ONU) or GII.17 P domain (5F4M) and a Nano-85 (4X7D) as search models. Structures were refined in multiple rounds of manual model building in COOT [72] and refined with PHENIX [73]. Alternative binding interfaces derived from the crystal packing were analyzed using an online server PDBePISA. The orientation of the Nanobody with the highest interface surface area and contact with CDRs was selected as the biologically relevant interface. Atomic coordinates were deposited to the Protein Data Bank (PDB).
The norovirus VLP morphology was analyzed using negative stain electron microscopy (EM) as previously described [31]. Nanobodies (1 mg/ml) and VLPs (1 mg/ml) were mixed in 1:1 ratio and incubated for 1 h at room temperature. Prior to loading on carbon coated EM grids, all samples were diluted 30 times with distilled water. Grids were washed two times with distilled water and stained with 1% uranyl acetate. The grids were examined on a Zeiss 910 electron microscope (Zeiss, Oberhofen, Germany) at 50,000-fold magnification. VLP diameter was measured with ImageJ software using calibrated pixel/nm scale bar. The hydrodynamic diameters of treated and untreated norovirus VLPs were measured using dynamic light scattering (DLS) on ZetaSizer Nano (Malvern Instruments, UK). Samples were diluted 1:50 with PBS up to a final volume of 1 ml. Three × 12 measurement runs were performed with standard settings (Refractive Index 1.331, viscosity 0.89, temperature 25°C). The average result was created with ZetaSizer software.
In order to determine the effects of the Nanobodies on native virions, we collected GII.4 positive stool samples from two individuals with acute norovirus infection [74]. A 10% (w/v) stool suspension was prepared in PBS and clarified by centrifugation at 10,000 × g for 10 min. First stool sample was concentrated by ultracentrifugation at 285,000 × g for 3 h at 4°C. Then, 70 μl of the supernatant were treated with 150 μl of each Nanobody (1 mg/ml) for 30 min at room temperature. Samples were digested with 50 U of RNAse One (Promega, Germany) for 30 min at 37°C. After treatment total RNA was extracted with QIAamp Viral RNA extraction kit (Qiagen, Hilden, Germany). One step RT-qPCR was performed with previously published GII.4 primers NKP2F (5’-ATGTTYAGRTGGATGAGATTCTC-3’), NK2R (5’-TCGACGCCATCTTCATTCAC-3’) and probe RING2-TP (5’-FAM-TGGGAG GGCGATCGCAATCT-TAMRA-3’) using qScript XLT One-Step RT-qPCR ToughMix (Quantabio, USA). For incomplete lysis, samples were diluted twice with PBS prior to RNA extraction with shortened incubation time. cDNA was synthesized using High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, USA). qPCR with melt curve analysis was performed using SYBR Green Master Mix (Bio-Rad, Hercules, USA). GII.4 specific primers, sense JJV2F (5’-CAAGAGTCAATGTTTAGGTGGATGAG-3’) and antisense COG2R (5’- TCGACGCCATCTTCATTCACA-3’) were used for norovirus detection as previously described [63]. Viral load was quantified by comparison to a standard curve of GII.4 norovirus RNA transcripts of a known concentration. Average values for two independent experiments for concentrated virus and three independent experiments for RNAse free stool are presented. Statistical analysis was performed using one-way ANOVA test. Differences were considered significant when P≤0.05.
To evaluate the impact of Nanobody binding on capsid susceptibility to proteolytic digestion norovirus VLPs (1 mg/ml) were incubated with Nanobodies (1 mg/ml) in 1:1 ratio for 30 min at 37°C. Then, trypsin-EDTA was added to final concentration of 10 μg/ml for 30 min at 37°C. The concentration of trypsin was chosen to yield only partial cleavage with visible intermediate products. After digestion, samples were loaded on the SDS-12% polyacrylamide gel and stained with coomassie stain.
|
10.1371/journal.pntd.0002860 | Comparative Assessment of ELISAs Using Recombinant Saposin-Like Protein 2 and recombinant Cathepsin L-1 from Fasciola hepatica for the Serodiagnosis of Human Fasciolosis | Two recombinant Fasciola hepatica antigens, saposin-like protein-2 (recSAP2) and cathepsin L-1 (recCL1), were assessed individually and in combination in enzyme-linked immunosorbent assays (ELISA) for the specific serodiagnosis of human fasciolosis in areas of low endemicity as encountered in Central Europe. Antibody detection was conducted using ProteinA/ProteinG (PAG) conjugated to alkaline phosphatase. Test characteristics as well as agreement with results from an ELISA using excretory–secretory products (FhES) from adult stage liver flukes was assessed by receiver operator characteristic (ROC) analysis, specificity, sensitivity, Youdens J and overall accuracy. Cross-reactivity was assessed using three different groups of serum samples from healthy individuals (n = 20), patients with other parasitic infections (n = 87) and patients with malignancies (n = 121). The best combined diagnostic results for recombinant antigens were obtained using the recSAP2-ELISA (87% sensitivity, 99% specificity and 97% overall accuracy) employing the threshold (cut-off) to discriminate between positive and negative reactions that maximized Youdens J. The findings showed that recSAP2-ELISA can be used for the routine serodiagnosis of chronic fasciolosis in clinical laboratories; the use of the PAG-conjugate offers the opportunity to employ, for example, rabbit hyperimmune serum for the standardization of positive controls.
| To improve the serodiagnosis of human fasciolosis caused by Fasciola hepatica, we comparatively evaluated the accuracy of two different enzyme-linked immunosorbent assays (ELISAs) based on the use of two published recombinant antigens. The best performance was achieved with the recombinant F. hepatica saposin-like protein-2 antigen (recSAP2). Although the F. hepatica E/S antigen exhibited a slightly higher diagnostic sensitivity, the higher specificity performance of recSAP2 renders this antigen very suitable for application in low endemic areas, especially when coupled to an easy and standardized production facility as compared to the relatively complex production procedure for an E/S antigen. Conclusively, the recSAP2-ELISA can be used as a routine individual serodiagnostic test for human fasciolosis, especially when backed up by a compatible clinical history together with other serodiagnostic technique for other helminth infections of the liver, e.g. alveolar or cystic echinococcosis.
| In Central Europe, the most frequently encountered autochthonous helminthic infections that require appropriate immunodiagnostic support include both forms of echinococcosis (Echinococcus multilocularis and Echinococcus granulosus), toxocarosis (Toxocara spp.), trichinellosis (Trichinella spp.), ascariosis (Ascaris lumbricoides, A. suum) and fasciolosis (Fasciola hepatica). Other helminthoses are diseases encountered in the context of travel medicine and sojourn in tropical or subtropical areas. Generally, the immunodiagnosis of helminthic infections is challenged particularly by the problem of high serological cross-reactivity when using crude or inadequately purified antigens. Another serodiagnostic problem relates also to cancer patients who raise antibodies against predominantly carbohydrate epitopes that might be common to helminth antigens [1], [2], [3], as exemplified e.g. by cross-reactive anti-P1 antibodies that can be elevated in some cancer patients as well as in echinococcosis and fasciolosis patients [4], [5].
Thus far, immunodiagnostic tools/methods for echinococcosis [6], [7], toxocarosis [8], trichinellosis [9] and ascarosis [10] that achieve measures of specificity and sensitivity permissible for routine use or commercialization have been developed. However, the immunodiagnosis of fasciolosis, in Central European regions of low endemicity, has remained a major challenge, and routine diagnostic laboratories are struggling with the selection of a suitable and reliable test. Nevertheless, recent improvements have been published, mainly by Latin and North American groups on the use of purified antigens, such as Fas2 [11], CL1 [12] or FhSAP2 [13], [14]. To date, these antigens have not yet been (i) validated according to the standard/s required of routine diagnostic laboratories operating under Central European infectiological conditions and ISO 17025 norms, (ii) assessed in relation to specificity (e.g., considering cancer patients) or (iii) directly compared with each other for diagnostic performance.
Based on a review of the literature, we selected two promising but different recombinant Fasciola antigens, the F. hepatica saposin-like protein-2 antigen (SAP2) [15] and the cathepsin L1 cysteine proteinase (CL1) [16] to establish and subsequently assess an optimized ELISA for the serodiagnosis of human fasciolosis. In this assessment, an emphasis was placed on the immunodiagnostic discrimination from other (hepatic) parasitological problems encountered in Central Europe, such as alveolar echinococcosis, toxocarosis and ascariosis, but also other parasitic diseases acquired during overseas travel. In addition, one of the most frequently encountered differential diagnostic problems in hepatic and other organ disorders are tumors, which even upon use of various imaging procedures, may not be readily discriminated from particular parasitoses. Moreover, sera from cancer patients are also known sometimes to cause serological cross-reactivity, as has been documented, e.g. for echinococcosis serology [1], [2], [3], [17], [18]. Therefore, one of the crucial considerations for the present study was the inclusion of sera from 121 cancer patients that had already been previously investigated for their putative cross- or non-specific reactivity with Echinococcus antigens [2], [3].
The working hypothesis of the present study was that, if both recombinant antigens exhibit a similarly high specificity, then their direct combination might yield a higher diagnostic sensitivity than when employed as single antigens. Therefore, we compared the ELISAs using recSAP2, recCL1 and recSAP2 plus recCL1 with the conventional ELISA (ISO-17025) using excretory-secretory products from adult F. hepatica (Fh_E/S). In preliminary experiments with the conventional FhES-ELISA, we had shown that a conventionally used anti-huIgG-alkaline phosphatase conjugate exhibited the same diagnostic performance as a ProteinA-ProteinG-AP-conjugate [PAG-AP] (Gottstein et al., unpublished). Based on these findings and the fact that for PAG-AP a positive control serum of animal origin can be used, we elected to conduct the present study using PAG-AP.
All serum samples from humans were collected as part of public health and clinical diagnostic activities, were available prior to the commencement of this study and were treated anonymously, Samples from blood donors were obtained under informed written consent and provided by the Swiss Blood Transfusion Center (SRK). This study was approved by the IPA Review Board of the Vetsuisse Faculty of Bern, Switzerland.
Excretory-secretory products (FhES) from F. hepatica were prepared as described elsewhere [20]. Briefly, adult flukes were collected from the bile ducts from sheep livers obtained from a slaughterhouse and were washed several times in 0.01 mol/L phosphate-buffered saline (PBS), pH 7.4 at room temperature. The flukes were incubated under sterile conditions at 37°C for 24 h in serum-free RPMI-1640 medium supplemented with 25 mmol/L HEPES buffer, 7.5% sodium bicarbonate, containing 100 µL penicillin and 100 µg/mL streptomycin. The medium was then sedimented (5,000× g for 10 min at 4°C) to remove any remaining particles. The supernatants were collected and then concentrated using an YM-10 membrane filter system (Amicon Corp., Lexington, MA). Protein concentrations were assessed with a Bradford based protein assay (BioRad Laboratories, Cressier, Switzerland).
A fresh, morphologically intact and viable adult F. hepatica was isolated from an ovine bile duct and immediately put into RNA later (Invitrogen) for storage. Using peqGold RNAPure and peqGOLD OptiPure (both PeqLab), RNA was isolated according to the manufacturer's manual and by using a poly-T primer cDNA was prepared with the Omniscript RT kit (Qiagen). The coding sequence of the saposin-like protein-2 antigen (recSAP2) was amplified by PCR (initial denaturation: 98°C - 3 min, amplification: 25×98°C–20 sec, 58°C–20 sec, 72°C–30 sec, and a final 72°C step for 5 min) using the primers FhSAP-forward (5′-CACCAACCCACTGTTCGTGTTAATG) and FhSAP-reverse (5′-CTAGCACAGCTTGATTAAACG). Primer FhSAP-dw contained a N-terminal CACC stretch needed for the directional in-frame cloning of the amplicon (306 bp) into the Champion pET Directional Topo Expression Kit (Invitrogen). Insertion was verified by sequencing, and clones containing a perfect matching sequence were used for pilot experiments of expression. The clone expressing the highest level of recSAP2 was then used for large scale expression: 10 ml of overnight culture were diluted in 1 l Luria Bertani (LB) medium containing 100 µg/ml ampicillin (Sigma) and shaken at 37°C until the OD600 reached 0.5. The protein expression was then induced by adding 1 mg IPTG. After shaking for 3.5 h at 37°C, the cells were pelleted by sedimentation (15 min, 4,000×g) and the recSAP2 was isolated under denaturating conditions using 2 Protino Ni-IDA 1000 packed columns (Machery-Nagel) according to the manufacturer's instructions, with the following exception. After washing, under denaturating conditions, the columns were washed with 10 ml non-denaturating buffer (50 mM NaH2PO4, 300 mM NaCl), and the recombinant protein was eluted three times with 1 ml non-denaturating elution buffer (50 mM NaH2PO4, 300 mM NaCl, 250 mM imidazole, pH 8.0). To reach ELISA-stage, the recSAP2 was precipitated with saturated ammonium sulfate solution, and the precipitate dissolved in ELISA coating buffer (100 mM sodium carbonate, pH 9.6). Storage prior to use for ELISA was at −80°C.
The purity and antigenicity of the recSAP2 were assessed by silver-staining of SDS-PAGE gels [21] and Western blot analyses, as described previously for recP29, a recombinant antigen of E. granulosus [22].
The complete cDNA sequence encoding F. hepatica secreted CL1 was retrieved from GenBank. Forward (5′- GTACCCGACAAAATTGACTGG-3′) and reverse (5′- TCACGGAAATCGTGCCACCAT-3′) primers were designed to amplify the appropriate region of the protein (220 amino acid), without the C-terminal propeptide (55 amino acid). A CACC-tag was added to the 5′ end of the forward primer for further cloning into the Champion pET Directional Expression kit (Invitrogen).
The cDNA encoding the CL1antigen (24.2 kDa) was amplified by PCR of 250 ng of F. hepatica cDNA as a template (the same as used to amplify recSAP2), 200 µM dNTPs, 0.5 µM of each forward and reverse primer, in a total volume of 50 µl with 1 U of Phusion High-Fidelity DNA polymerase (New England Biolabs). The amplification was carried out using an initial denaturation of 98°C for 1 min, followed by 25 cycles of denaturation at 98°C for 30 s, annealing at 58°C for 30 s and an extension at 72°C for 30 s. The final polymerization was carried out at 72°C for 5 min. The 658 bp PCR product was purified using High Pure PCR Product Purification Kit (Roche) and then cloned into Champion pET expression vector. Competent E. coli (TOP10) cells were transformed using the manufacturer's instructions (Invitrogen). The transformed bacteria were incubated on LB plates containing 100 µg/ml of ampicillin at 37°C overnight, and colonies containing the insert were identified by colony PCR. Five positive clones were grown overnight in LB medium containing ampicillin, and then the plasmids were isolated using a QIAprep Spin Miniprep Kit (Qiagen) according to the manufacturer's protocol. Each vector construct was sequenced to ensure an open reading frame. Recombinant CL-1 was expressed as a fusion protein with His-Tag in E. coli BL21 as described above for the recSAP2. RecCL1 from E. coli was purified under denaturing conditions (8M Urea) using packed columns (Protino Ni-IDA 150, Macherey & Nagel) according to the instructions of the manufacturer. The eluate was passed through PD10 desalting columns (GE Healthcare) and then was dialyzed against PBS. Purified protein samples were examined in silver-stained SDS-PAGE gels [21] and by Western blot [22].
FhES-, recSAP2- and ecCL1-ELISAs were carried out essentially as described for Echinococcus antigens [23]. Briefly, sera were diluted 1∶100 and tested using the following antigens (at optimized coating concentrations): FhES-antigen (10 µg protein per ml); recSAP2-antigen (0.1 µg protein per ml); recCL1-antigen (0.1 µg protein per ml). A fourth ELISA included a double-coating with a mix of 0.1 µg protein of recSAP2 and recCL1 per ml. As a conjugate, an anti-human-IgG-alkaline phosphatase [αhuIgG-AP] conjugate (Sigma; 1∶1'000 dilution) or a ProteinA-ProteinG-AP-conjugate [PAG-AP] (Thermo Scientific no. 32391; 1∶10'000 dilution) was used. The four Fasciola-antigen-ELISAs validated here were first calibrated, in order to determine the optimal threshold (cut-off value) for the discrimination between positive and negative findings. The individual cut-off value was thus determined by testing blood donor sera and tumor patients' sera and potentially cross-reactive sera (from patients with other parasitic diseases) as a one group together, thus reaching a representative average number of the “negative samples” encountered in a routine laboratory. Inter-test and intra-test variations in test results were calculated as coefficients of variation for reference negative and positive sera, all tested in triplicate on each test plate; variation of ≤15% was recorded, which is considered acceptable for serodiagnostic assays [2].
For statistical analyses, the samples from the 30 patients with confirmed F. hepatica infection represented the positive status (1), whereas the 20 healthy individuals, the 121 cancer patients and the 87 patients infected with potentially cross-reacting or diagnostically relevant other parasitic infections, all represented the negative status (0). The comparative evaluation of the four assays (i.e. FhES-ELISA; recSAP2-ELISA; recCL1-ELISA; recSAP2-recCL1-ELISA) was carried out using this classification. To quantify the linear numerical correlation between the raw data measurements of the three assays, Spearman rank correlation coefficients were derived using all 248 samples. The distribution of OD405 nm-values for the samples in the three assays was displayed using dot plots and box plots. In a Receiver-Operator-Characteristic (ROC) analysis, threshold (cut-off) values for the following four conditions were derived according to:
In addition, modified two-graph ROC curves were drawn for the different assays, and the areas under the ROC curves (AUC) were statistically compared for significant differences. Descriptive statistics, plots and ROC analyses were done Microsoft Excel 2010 (www.microsoft.com) and the statistical software package NCCS 8 (www.ncss.com).
The highest potential for a false positive interpretation of test results relates to sera from patients suffering from other parasitic diseases. In Table 1, we present the different parasitic diseases and their rate of cross-reactivity determined in the different ELISA-types, yielding a relative specificity index linked to cross-reactivity. To determine the threshold between positive and negative results, the cut-off value was arbitrarily set at the Youdens J maximum value, calculated as described above (MedCalc software version 12.7.5.0; http://www.medcalc.org).
GenBank accession number for the used cDNA-SAP2 stretch: AF286903.1.
GenBank accession number for the complete cDNA sequence encoding F. hepatica secreted CL1: U62288.2.
Five batches of recSAP2 and five batches of recCL1 were both independently produced and analysed by silver staining (data not shown). As all batches yielded identical purities, they were pooled to obtain two single working batches, respectively. These batches were each assessed with known fasciolosis-sera by Western blot to verify the antigenicity of the two single bands of expected relative mobilities of Mr 15000 and Mr 26000 for recSAP2 and recCL1, respectively (data shown for recSAP2 only, Figure 1). Fasciolosis-sera were selected, such as to cover the whole range of antibody levels measured by the conventional FhES-ELISA. The relationship between “banding intensity” and FhES-ELISA antibody level was relative. Serum a1 has the highest levels in FhES-ELISA (>100 AU, relative antibody units: The quantification of these ELISA-results, expressed in relative antibody units [AU], arises from the routine serology carried out at the Institute of Parasitology in Bern, and is not further specified in this article), and exhibited also the strongest staining intensity by Western blot. Sera nos. a2–a5 yielded medium FhES-ELISA antibody levels (70, 40, 51 and 66 AU), while staining intensity in Western blot varied considerably and appeared not to be directly linked to the recSAP2-ELISA findings. Serum a6 was very weak in FhES-ELISA, and was not detectable by Western blot analysis. However, the same serum (a6) was, nevertheless, also weakly positive in recSAP2-ELISA (Figure 1).
The distribution of absorbance values varied considerably between test positive and negative samples, as to be expected, but also between different assays (Figure 2), and different optimal cut-off values were established for the four tests (see section of ROC analyses).
The highest overall accuracy (agreement between references status and test result) reached was 0.984, while the highest combined sensitivity and specificity (Youdens J) was 0.905 (Table 2, Figure 3). The recSAP2-ELISA cut-off 0.084 showed the best combination of sensitivity (0.867) and specificity (0.989) of the three recombinant-antigen assays (Table 2). For this assay, the empirical values for sensitivity, specificity, overall accuracy and Youdens J, as a function of the cut-off value, were plotted in a multi-line ROC graph (Figure 4) to illustrate the pattern of these test characteristics over a range of cut-off values. When comparing the AUC values of the four tests, test results of the recCL1 assay were significantly lower than for all other assays (p<0.001), while results of the FhES assay were significantly higher, even when compared with the second best assay (recSAP2; p<0.047).
Cross-reactions, as a results of a positive-negative discrimination based on the cut-off value selected, are presented according to the different parasitic disease groups investigated (Table 1). One serum (a case of alveolar echinococcosis) consistently cross-reacted in all four ELISAs, whereas all other cross-reactions were individually scattered among individual ELISAs. The best score in specificity (99%) was achieved by recSAP2-ELISA, with a single instance of cross-reactivity (described above).
Using the selected cut-offs, FhES exhibited the best level of diagnostic sensitivity (93%) (28 positive sera of 30 Fasciola-cases), followed by recSAP2 (26 of 30 cases; 87%), while the sensitivities achieved using recCL1 and the combination of recCL1 and recSAP2 were all below 77% for all elaborated cut-offs (Table 2).
Coprological diagnosis, based on the identification of F. hepatica eggs found in stools, duodenal contents or bile analysis is still commonly employed as a “gold standard” to detect human fasciolosis. This is the case, despite the consensus that this method is not entirely reliable [24] for reasons such as: (i) eggs are not detected until the patent period of infection, when much of the liver damage has already occurred by the migration of juvenile flukes in the liver parenchyma, (ii) eggs are released sporadically from the bile ducts and, hence, stool samples from infected patients may not necessarily contain eggs [25]. Therefore, serological techniques play an important complementary role in the diagnosis of clinical cases of fasciolosis. In the situation of low endemicity, such as encountered in many countries of Central Europe, serological methods require not only a good diagnostic sensitivity, but more importantly also a high specificity. The reason is that potentially cross- or false-positively reacting sera will be much more frequently found in routine diagnosis than actual true cases of fasciolosis. This has to be considered particularly in the context of a differential diagnosis, predominantly related to any other hepatic disorders resembling, symptomatically, those of fasciolosis.
Serodiagnosis of fasciolosis of humans has been successfully performed by employing several antigens (antigen fractions) of F. hepatica, where, to date, ES products have become the most commonly used antigen in-house ELISAs [24]. Nevertheless, the use of FhES is associated with several problems when used for routine diagnostic laboratory conditions, including (1) a dependence on the availability of living flukes and (2) representing an antigen mixture that is subjected to variations due to natural and artificial conditions (e.g. time between slaughter and cultivation). This makes antigen standardization between diagnostic laboratories difficult, whereas a recombinant single component antigen exhibits a constant composition. With regard to recombinant antigens, cathepsin L1 (CL1) and saposin-like protein 2 (SAP2) from F. hepatica are of the most frequently referenced candidates being used for detecting anti-Fasciola antibodies in different epidemiological situations [12], [15], [16], [26], [27]. For the present study, we selected both SAP2 and CL1 as key candidates to be compared, alone or in combination, in the form of bacterially-expressed recombinant antigens (recSAP2 and recCL1). As a serological standard, we used a conventionally employed F. hepatica ES antigen (FhES). In order to improve routine applicability of any of these tests, we carried out a preliminary study comparing the efficacy of anti-human-IgG-alkaline phosphatase and a ProteinA-ProteinG-alkaline phosphates conjugate (PAG-AP) to detect anti-Fasciola-antibodies (data not shown). We documented a comparable performance of both conjugates (even slightly improved for PAG, although statistically not significant). PAG-AP offers the considerable advantage that it binds also to IgG of several animal species, including, for example, rabbit IgG. The availability of sufficient positive control serum for routine diagnostic application under ISO 17025 accreditation conditions is one of the big problems in routine diagnostic laboratories, as the procurement of such serum (in larger quantities) from fasciolosis patients is difficult or even impossible in countries of low endemicity. Alternatively, hyperimmune serum from rabbits or other appropriately immunized animals can assume the role of positive control reagents, thus considerably facilitating the establishment of standardized operating procedures (SOPs) of Fasciola-serology.
The results of the present, comparative study demonstrated similar performances of the FhES and the recSAP2 antigen with regard to diagnostic sensitivity and specificity, whereas the assay using antigen recCL1 did not reach the expected performance. Our initial working hypothesis had been based on the assumption that a combination of recSAP1 and recCL1 would increase diagnostic sensitivity, provided specificity could be maintained. This hypothesis was justified by an advanced appraisal of previous publications on the topic. Carnevale et al. [26], who used a recombinant CL1 containing the proregion of the protein, reported 100% sensitivity and 100% specificity. In this study, however, the selection criteria, from both clinical and epidemiological perspectives, have not been described, such that one cannot elucidate whether 100% sensitivity represents a true diagnostic sensitivity, as encountered in a routine diagnostic laboratory. Similarly, Tantrawatpan et al. [28], who used a peptide-based form of CL1 deduced from F. gigantica, reported 100% sensitivity and 99.7% specificity. However, in the latter publication, the pre-selection criteria for the fasciolosis sera were not documented as to allow an assessment of the actual diagnostic sensitivity. Such tests should also include sera from acute cases, for which infection had been proven but eggs could not be detected repeatedly using a coprological sedimentation technique. For example, sera from acute cases were not included in the study by O'Neill et al. [16]; these authors reported 100% sensitivity for an ELISA using CL1 expressed in Saccharomyces cerevisiae (yeast) and an anti-IgG4-detection systems following the testing of sera from 26 cases with egg-excretion. However, these authors did not evaluate the true sensitivity by testing sera from cases representing various forms of infection and disease stages. A relatively recent study on recombinant CL1 was carried out by Gonzales Santana et al. [27] upon use of Pichia pastoris for expressing recFhCL1. Conversely to our study, the antigen demonstrated not only an excellent diagnostic sensitivity, but also an optimal specificity. The reason for the difference encountered between these study findings and our study may be the found by the fact that expression of a metazoan gene such as cl1 in P. pastoris may much better lead to carboxylation of the antigen, and thus to an improved formation of relevant epitopes within this proteinic antigen. We will address this important feature in our next studies. Another reason why the diagnostic sensitivity of recCL1 was higher in other studies [16] may have been the use of a subclass-specific anti-IgG4-conjugate. The same approach (anti-IgG4-conjugate) was chosen by Tantrawatpan et al. [28], who furthermore employed a peptide-based synthetic FhCL1-antigen. Although we know that the protein A and protein G used in our study, both principally bind to human IgG4 (http://www.amsbio.com/brochures/Protein-A-G%20-Affinity-for-IgG-subclasses.pdf), it will be nevertheless interesting to compare, in future studies, both conjugate types directly with regard to the diagnostic sensitivity yield.
In our study, bacterially expressed recCL1 exhibited two related problems, which finally rendered this antigen not useable for our purpose. First, in comparison to recSAP2, recCL1 displayed relatively high background reactivity with both, sera from cancer patients and sera from patients suffering from other parasitoses (see Figure 2). This translated into a relatively high cut-off level for the discrimination between seropositivity and seronegativity and, thus, resulted in a relatively low diagnostic sensitivity. Nevertheless, we raised the question whether, among the four recSAP2-seronegative fasciolosis patients, one or more of them would be recCL1-positive, thus justifying a possible combination of the two antigens. However, this was not the case. In this context, it is also important to mention that the two FhES-negative fasciolosis patients were also seronegative against the recSAP2 and recCL1 antigens. Consequently, an overall appraisal of the results did not suggest a routine application of recCL1 antigen for the serodiagnosis of human fasciolosis. Nevertheless, as other previous reports clearly documented a good diagnostic performance of recCL1 if produced by a different expression system [16] or as a synthetic polypeptide [28], we will, in future studies, switch from bacterial expression to the other expression/synthesis systems.
A detailed comparison between the diagnostic operating characteristics of the recSAP2-ELISA and the other Fasciola-ELISAs included in our study demonstrated clearly an excellent performance of recSAP2 in relation to both specificity and cross-reactivity (specificity 99%, see Tab. 1). Regarding diagnostic sensitivities, among the 30 fasciolosis patients available for our study, only one patient with a coprologically confirmed fasciolosis remained consistently negative in all four tests, including the recSAP2-ELISA. Due to the lack of clinical data from the respective patient, the reason for this false-negative result could not be established. Here, the lack of clinical and radiological evidence might indicate a chronic infection status with a low infection intensity, accompanied by a decline or even disappearance of parasite-specific antibody levels. In this respect, it is possible that especially the hepatic parenchymal migration stage of juvenile flukes early during infection induces the strongest immune response, whereas a few adult worms remaining in the bile ducts during the chronic phase of infection might not be sufficient to sustain an antigen stimulus to maintain a detectable serum antibody level. Overall, sera from four fasciolosis patients were test-negative in both recombinant antigen-based assays, while three of them were clearly seropositive in the conventional FhES-ELISA. However, this slight diagnostic inferiority of recSAP2 as compared with FhES was largely compensated by other parameters that favored SAP2 as a routine serodiagnostic tool, particularly when applied in a low endemicity area. In such a situation, the comparatively higher specificity of recSAP2 (99% versus 95%, see Table 1) might be superior to the comparatively higher sensitivity of FhES (87% versus 93%). Importantly, in contrast to FhES, recSAP2 did not exhibit occasional cross-reactions with sera from neuro-cysticercosis and filariosis patients (see Table 1). Such cross-reactions might hamper the diagnostic performance in cases where other clinical data are inconclusive, and where serology becomes a crucially important diagnostic tool.
In conclusion, we consider the recSAP2-PAG-AP-ELISA as serological test system for routine diagnosis of human fasciolosis, particularly if test results are supported by clinical history and the use of other serological tests controlling for possible cross-reactions due to antibodies induced by other helminths. In addition, this test system might serve as an excellent serodiagnostic tool for epidemiological studies of human fasciolosis, particularly in the context of outbreaks, or accumulated case numbers, for example, as observed recently in Switzerland [19]. Our conclusions are in perfect agreement with a previous report from Figueroa-Santiago et al. [15], who were the first authors to document the excellent diagnostic performance of the recSAP2-ELISA.
|
10.1371/journal.pntd.0006112 | Tracking the career development of scientists in low- and middle-income countries trained through TDR’s research capacity strengthening programmes: Learning from monitoring and impact evaluation | The Special Programme for Research and Training in Tropical Diseases (TDR) co-sponsored by UNICEF, UNDP, World Bank and WHO has been supporting research capacity strengthening in low- and middle-income countries for over 40 years. In order to assess and continuously optimize its capacity strengthening approaches, an evaluation of the influence of TDR training grants on research career development was undertaken. The assessment was part of a larger evaluation conducted by the European Science Foundation. A comprehensive survey questionnaire was developed and sent to a group of 117 trainees supported by TDR who had completed their degree (masters or PhD) between 2000 and 2012; of these, seventy seven (77) responded. Most of the respondents (80%) rated TDR support as a very important factor that influenced their professional career achievements. The “brain drain” phenomenon towards high-income countries was particularly low amongst TDR grantees: the rate of return to their region of origin upon completion of their degree was 96%. A vast majority of respondents are still working in research (89%), with 81% of respondents having participated in multidisciplinary research activities; women engaged in multidisciplinary collaboration to a higher extent than men. However, only a minority of all have engaged in intersectoral collaboration, an aspect that would require further study. The post-degree career choices made by the respondents were strongly influenced by academic considerations. At the time of the survey, 92% of all respondents hold full-time positions, mainly in the public sector. Almost 25% of the respondents reported that they had influenced policy and practice changes. Some of the challenges and opportunities faced by trainees at various stages of their research career have been identified. Modalities to overcome these will require further investigation. The survey evidenced how TDR’s research capacity grant programmes made a difference on researchers’ career development and on south-south collaborations, by strengthening and localizing research capacity in lower income regions, and also showed there is more that needs to be done. The factors involved, challenges and lessons learnt may help donors and policy makers improve their future interventions with regard to designing capacity strengthening programmes and setting funding priorities.
| The Special Programme for Research and Training in Tropical Diseases (TDR) co-sponsored by UNICEF, UNDP, World Bank and WHO has been providing training grants to strengthen research capacity in low- and middle-income countries for over 40 years. In order to assess to what extent TDR’s grants made a difference on the career development of these grantees, a survey tool was developed and implemented in collaboration with the European Science Foundation. The survey was answered by 77 individual trainees who completed their degree (masters or PhD) with support from TDR between 2000 and 2012. The study provided valuable information on factors involved in the career development of the trainees and influencing the local retention of the capacity that has been built, to prevent “brain drain”. Encouraging aspects, such as a 96% of the capacity being retained locally, a 92% full-time employment rate at the time of the survey, or 89% of the respondents still working in research showed the positive influence of TDR’s capacity strengthening grants on researchers’ career development. This was in line with 80% of the respondents rating TDR’s support as “very important”. The challenges, lessons learnt and further opportunities identified may be helpful to donors and policy-makers when designing research capacity programmes, fostering south-south collaboration, and setting funding priorities.
| The Special Programme for Research and Training in Tropical Diseases (TDR), co-sponsored by UNICEF, UNDP, the World Bank and WHO, has a long track record in research capacity strengthening. Created in 1975 to support research and research capacity strengthening in the fight against tropical diseases, TDR’s goal is to improve health and reduce the burden of infectious diseases in low- and middle-income countries (LMICs). For more than 40 years, TDR has strengthened health research capacities in these countries by: i) supporting individuals’ education and training through fellowships, scholarships and learning-by-doing programmes for specific skills, particularly on good practice for health research and fostering mentorships; ii) supporting institutional capacity by establishing national and international training and research centres; and iii) developing networks and collaborative research projects [1].
Regular external reviews of its research capacity strengthening programmes have helped TDR to evolve its strategy in light of the global environment so as to remain a fit-for-purpose programme. The latest evaluation of TDR’s contribution to career development of a selected group of individuals and institutional capacity development grantees was conducted in 2010. The main objective was to identify factors that positively influenced and improved the research capacity and career development of TDR trainees and that are of broader relevance to the objectives and goals of international development and aid agencies [2, 3]. One of the recommendations was to better track the career development of grantees to help evaluate the influence of these early learning supports.
To respond to these recommendations a career tracking survey tool was developed to study the potential links between the grants received by TDR trainees and their career development. The survey is conducted every 2–3 years to provide quantitative and qualitative data to better understand TDR’s grants impact on grantees’ careers. It provides an instant view of a trainee’s career, with performance indicators to allow monitoring and evaluation of career development. This survey tool was developed and implemented in collaboration with the European Science Foundation in France, a European structure that generates evidence to support the decision-making of countries or organizations. It was implemented in 2014 to study the contribution of TDR support on TDR training grantees’ careers between 2000 and 2012. The survey responses have highlighted the challenges, bottlenecks and opportunities of different research career stages, which are being used to identify intervention points or specific actions needed to achieve desirable career progression.
TDR was invited to respond to a call for research support and funding organizations to join a doctorate career tracking project. The survey was launched in late 2014 by the European Science Foundation in Strasbourg, France. The aim of this call was to develop a methodology to design and implement a career tracking survey tool.
Five organizations joined the study: the AXA Research Fund, Paris, France (AXA); the Fonds National de la Recherche, Luxembourg (FNR); the Goethe Graduate Academy (GRADE), Frankfurt, Germany; the Paul Scherrer Institute (PSI), Villingen, Switzerland and TDR. All data were disaggregated by organizations. Six hundred and thirty eight (638) trainees from the five partners responded to the survey with the following breakdown: 110 from the AXA fund, 84 from FNR, 105 from GRADE; 133 from PSI and 77 from TDR. The aggregated results from the 638 participants have been published [4].
Data were then disaggregated in 2015 and results specific to the TDR trainees are presented in this paper.
A total of 304 TDR trainees who completed their doctorate or master’s degree between 2000 and 2012 with a TDR grant were identified in the TDR information and management system. These included recipients of any of the following scheme of grants: research training grants (RTG); re-entry grants (REG); the Multilateral Initiative on Malaria (MIM); research grants and institution strengthening grants (ISG). RTGs were awarded to individuals in LMICs to pursue studies leading to a postgraduate degree (MSc or PhD) at their home country institution, in another LMIC or in a high income country. REGs were intended to facilitate the career development of young scientists returning to their home institution within 12 to 24 months, following completion of a graduate degree (MSc or PhD) or a post-doctoral fellowship. ISGs were designed to provide up to three years of support to an institution or research group to enhance infrastructure and the research environment. MIM grants [3] were used to provide support to core African research groups for the development of malaria control tools (Box 1).
Information on all individuals and institutions that received grants between 2000 and 2012 was extracted from the TDR information management system and tabulated for range and scope of research topics.
Trainees were contacted individually, through e-mail, to ascertain their willingness to participate in the career tracking survey and to update their personal information. From a total of 304 trainees identified, 117 trainees (39%) responded positively while 187 did not respond, either due to out of date e-mail addresses or possible lack of interest.
The questionnaire design was based on existing surveys of doctorate graduates conducted by the Organization for Economic Cooperation and Development (OECD), Eurostat, the European Commission Marie Sklodowska-Curie actions, Wellcome Trust, UNESCO and the US National Science Foundation. The range of topics covered by the survey included demographics, mobility (virtual, physical and sectoral), research outcomes, roles and responsibilities, competence development and skills utilization. Several drafts of the questionnaire were reviewed by the five participating organizations and pre-tested in-house by ESF staff members, with the final questionnaire peer-reviewed by two independent international experts. The resulting questionnaire contained 52 questions, written in English.
Participants were informed about the detailed data protection and confidentiality arrangements that were in place for the survey such as the anonymization of replies before analysis. This included destroying all contact details before conducting any survey analysis and avoidance of any questions likely to collect sensitive or identifying information of any kind (date of birth, thesis title, disciplinary field, institution name, etc.). Written assurance was also given that contact details would only be used for the purpose of contacting the trainees during the data collection phase. Since ESF is located in Strasbourg, France, the modalities of the survey were declared to the Commission Nationale de l’Informatique et des Libertés (CNIL), the independent French authority protecting privacy and personal data.
The list of TDR trainees and their contact details were shared with ESF and names and e-mail addresses entered into an online survey database. The survey was launched with an explanatory cover note from ESF in September 2014. The questionnaire and an introductory message were sent to each of the 117 participants.
Any queries received by the ESF team from participants were dealt with on an individual basis, including practical questions regarding completion of the questionnaire. The number of respondents was logged on a daily basis and the percentage of responses on a weekly basis. A total of five reminders to participate in the survey were sent. The survey was closed in November 2014 and all respondents were thanked for their participation.
The survey data were imported into the Statistical Package for the Social Sciences (SPSS) for analysis by ESF.
Among the 304 TDR trainees identified, 117 trainees expressed availability to participate and were included in the survey. These included 54 RTG, 29 REG, 18 MIM and 16 ISG grants. Ultimately, 77 trainees responded to the survey (66% of those included). Unfortunately it was not possible to break down the analysis by grant as 68% responded to the question “Do you know the type of grant you received from TDR?” with “don’t know” and the survey was anonymous.
WHO Member States are grouped into six regions: Africa (AFR), the Americas (AMR), South-East Asia (SEAR), Europe (EUR), Eastern Mediterranean (EMR), and Western Pacific (WPR). Profiles of the TDR trainees who responded to the survey are shown in Fig 1A. The majority of respondents came from AFR (53%); 21% originated from AMR, mainly from Brazil (59%) and Argentina (25%); 11% were from SEAR, 11% from EMR and 4% from WPR.
Of the 77 respondents, 58% were men and 42% were women (Fig 1B). Representation of women was slightly higher than in the group of 304 initially contacted (62% men, 38% women). As shown in Fig 1C, women are well represented in all WHO regions except AFR where men are more represented (77%) than women (23%). In EMR, women are more represented (70%) than men (30%).
In terms of age, the majority of respondents were between 35 and 45 (51%). Women were slightly older than men: 34% of women were above 50 years of age as compared to 18% of men. In all WHO regions, except AMR, the vast majority of women have children (92%) but only 29% of women had children in AMR. Further investigation would be needed to understand any potential barrier for AMR women with children to access TDR training grants. Fifty-eight percent of men and 42% of women had other caring responsibilities such as care of an elderly person or an adult with a disability.
In AFR, the majority of respondents were from English speaking countries (64%) followed by French (33%) and Portuguese (3%) (Fig 1D). The response rate (number of trainees who responded to the questionnaire / number of trainees who received the questionnaire) was higher in Francophone (87%) than in Anglophone (55%) trainee sub-groups. When aiming at enhancing support to Francophone and Lusophone countries, further study may be needed to better understand the factors involved.
The 77 respondents were supported by TDR to obtain either a MSc (14 respondents), a medical doctorate (MD) (5 respondents) or a PhD (58 respondents).
All trainees studying for a master’s degree obtained their degree through structured means, involving a combination of defined courses and independent research. For trainees studying for a PhD, the majority of respondents (86%) achieved their degree through the traditional means of an independent research study under the guidance of a supervisor and only 14% through structured means. There was no relationship between the time taken to complete the degree and the structure followed. There was also no difference in the time taken and the structure followed to complete their degree between men and women.
The median time taken by respondents to complete their PhD was four years. Support provided by TDR did not always cover the full duration of the degree and ranged from one year (28%) or less (6%) to two years (20%), three years (25%), or more (21%). More men were supported for three years (30%) as compared to women (19%), and more women were supported for one year or less (39%) as compared to men (30%). The reasons for the variation of length of support are not clear. It would be important to better understand why duration of support was shorter for women than for men and what the implications were. Indeed early career support to acquire a degree is known to be a key factor for career development and a potential future leadership role [5].
Eight respondents (seven men from AFR and one woman from AMR) took a career break for one year or more. Of the seven men, only one found it very difficult to return to their previous position. The only woman who took a study break found it relatively easy to return to her position. However, the reasons for having taken a study break were not clearly explained; a more explicit question will be added in the next survey.
A proportion of respondents (35% overall) moved outside of their country of origin to complete their degree, the majority from AFR. Forty-one percent of AFR respondents who moved abroad went to high-income countries, mainly in North America and Europe.
Overall, 65% of respondents completed their degree in their region of origin. Seventy-nine percent of TDR grantees from AFR who completed their degree in their region of origin were trained in three countries: Kenya, Nigeria and South Africa. Sixty-nine percent of TDR grantees from AMR who completed their degree in their region of origin were trained in Argentina and Brazil. All of the countries where training took place have a relatively high national gross domestic income, with a well-developed health research structure and capabilities [6]. Thus, the survey showed the great benefit of TDR and other agencies supporting capacity strengthening programmes to promote collaboration between scientists in countries with more advanced health research capacities and countries with lower health research capacities within the same region.
The gain of south-south collaboration as compared to north-south collaboration in term of career development was analysed based on the 41 trainees from AFR. Fifteen (15) studied in high-income countries (north-south collaboration), 20 studied in three other African countries (Kenya, Nigeria and South Africa) (south-south collaboration) while six studied in their own country. There was no difference in response to the different questions between trainees who studied in high-income countries and those who studied regionally. This may suggest that south-south collaborations are as effective and at a lower cost than north-south collaborations.
All of the TDR trainees who responded to this survey were employed, with 95% holding a position at a university or research institution and 89% working as academic researchers.
Fig 1E provides details of TDR trainees’ current employment. Most of the respondents held a full-time position with more than 30 hours per week (92%) in either a permanent (83%) or temporary position (9%). Women were more often in permanent full time positions (91%) than men (78%). While the number is small, only men were self-employed (2%).
The vast majority of respondents worked in the public sector (83%) in non-profit (79%) or for-profit (4%) institutions, followed by the private sector (12%) and others, including public-private partnerships (2%). Twenty eight (28%) were directly funded by their employer, while 72% were employed on grants funded by some other external party.
Table 1 presents TDR respondents working as academic researchers (89%) by career stages as per the Frascati definition [7, 8]. The only difference between men and women was that a higher proportion of women described themselves as R1 researchers (first stage) and more often they held positions as junior researchers. This was the case for all WHO regions.
The minority not working as researchers (11%) were asked to indicate the reason(s) for this. The most common reasons cited were the difficulty of obtaining a suitable academic research position (100%), the difficulty to secure a tenured post (100%), the lack of research career opportunities (80%) and the low remuneration in research positions (75%).
In terms of occupational areas, the highest proportion of respondents worked in life sciences (47%), followed by education (34%), training (31%), healthcare (31%) that included healthcare practitioners and healthcare support occupations, social sciences (5%) and administrative support.
Table 2 clearly shows that a higher proportion of men were involved in management. However, similar proportions of women and men worked in life sciences, education, healthcare, social sciences and administrative support. There was no difference based on country of origin, country of study, country of work and their career stages.
In general, it is quite difficult to compare salaries across the geographical spread of the various WHO regions. However, some gender differences in salary levels were evident, since a higher proportion of women earned less than €20 000 per year (55% women versus 44% men) regardless of the region of origin. This is perhaps due to the fact that women more often held positions of junior researchers (Table 1). It could also reflect the worldwide issue of the gender pay gap.
The survey asked respondents to indicate in how many different countries they had physically studied or worked for a period of more than three months during and after TDR support (physical mobility). The majority of respondents had studied and worked solely in their own country (72%) while 28% percent had studied or worked in other countries. It is worth noting that international physical mobility was higher for AFR respondents (48%). In general the survey showed a strong mobility to countries with more advanced health research capacities. The highest international physical mobility was to Europe (60%) and North America (38%) then Argentina and Brazil (11% each) and Australia (7%).
Virtual mobility, or collaboration via information and communication technology platforms, was also considered. The majority of respondents (75%) acknowledged virtual mobility had taken place solely within their own countries. From the 25% remaining respondents, the highest international virtual mobility was to countries in Europe (46%) and North America (31%). As was the case with physical mobility, virtual mobility was higher for AFR respondents (42%). Interestingly, international physical mobility to Europe was higher than virtual mobility.
The survey showed that 58% of respondents conducted research in collaboration with researchers based in another country, mainly in Africa, Europe and North America, through a joint publication (55%) and/or a joint project (52%), in line with the mobility trends described above.
There was a considerable proportion of respondents who reported having engaged in multidisciplinary research activities (81%). Multidisciplinary collaboration was reflected through either joint publications (81%), collaborating at distance with occasional face-to-face (67%) or through web-based technologies (52%). Women seemed to engage in multidisciplinary approaches to a higher extent than men. Indeed, a higher proportion of women worked with researchers from a different field of expertise, either through joint publications (89% for women versus 74% for men) or virtual collaborations (63% for women versus 44% for men). This could be due to the fact that more women work in the field of social sciences than men. In a previous study analysing TDR support of 116 research training grants, 11/36 (30.50%) women and 11/80 (13.75%) men worked in the domain of social sciences. However in a recent study analazing gender differences in scientific collaborations, it is clear that women in the natural sciences domain have more collaboration in other fields than men [9]. Further research would be needed to better understand why TDR women trainees tend to engage in multidisciplinary approaches more often than men.
Intersectoral collaboration, in terms of joint activities between research institutions, industry or commercial ventures, was limited: 23% worked on a joint publication and 19% collaborated on a joint research project with industry. Men collaborated only slightly more frequently with industry (33%) than women (26%). There is a clear need to encourage intersectoral collaboration. This could be done though promoting and fostering mobility between research institutions, government and nongovernmental agencies, and the public and private sectors. It would help to make the career perspective after graduation more attractive and to reduce existing barriers to collaborative work between these sectors. Some of these barriers to career development are attitudinal, reflecting a lack of knowledge and sometimes a negative perception that academic staff may have about a career outside the university’s walls. The quality of career mentorship provided at the doctoral level could be an essential element to help overcome these concerns. Other barriers could be structural and institutional, bringing into question the reliance on publication output as the sole or main criterion for scientific recognition and career development.
Respondents, regardless of the region they came from, reported that they regularly used their doctoral skills in their current position (92%). They most often used these skills in managing research activities (74% of respondents dedicated more than 20% of their time to these activities). This was followed by staff management activities (47%), which included supervising students either at undergraduate and master levels (82%) and/or PhD level (65%) or supervising their peers’ work (75%); teaching activities (46% dedicated more than 20% of their time) and administrative activities (37% dedicated more than 20% of their time). Some dedicated time to transferring technology to industry (21%). There was no significant difference between genders in any of these activities.
Respondents reported having made presentations at national (73%) and international conferences (72%) and women were more active than men in international presentations (75% for women versus 66% for men).
Over 70% of respondents had been either lead authors (65%) or co-authors (70%) on peer-reviewed publications in the last 12 months. Similar proportions of men and women had been lead authors (69% for women versus 62% for men) but a higher proportion of men were co-authors (55% for men versus 45% for women) on peer-reviewed publications.
In terms of research and development, 20% of respondents had produced new research software resources and 9% of them had filled a patent. None of them had registered or licenced a product in the last 12 months.
Almost 25% of respondents claimed that their work had made a significant impact on influencing changes in policy and practice. This relatively low percentage could be due to the fact that respondents come from a largely academia-based group (98% held a position in university or in research institutions and 89% worked as academic researchers) which usually report impact more through publications, conference presentations and research awards. All trainee contact details or other identifying information of any kind (date of birth, thesis title, disciplinary field, institutional name, etc.) were destroyed before conducting any analysis. As a consequence, there was no possibility to verify if the work of the trainee had an impact on policy and practice. However, this percentage suggests the need to maintain and enhance efforts to bridge the gap between health research and policy-making and practice, as well as the need to capture such evidence in a systematic way. Indeed, the lack of evidence on translating research results into health policies, interventions or new tools has been identified for decades as a weakness in the evaluation of research capacity strengthening organizations [10].
Activities to communicate results to the public had been undertaken by 30% of respondents and media coverage was achieved by 22% of respondents. Men were more likely than women to claim impact on policy and practice changes (29% versus 19%), to communicate to the public (40% versus 16%) and to receive media coverage (27% versus 16%).
Respondents were asked to rate the importance of TDR support on achieving their professional career goals. Eighty (80) percent rated TDR support as very important, and 91% of respondents rated the TDR support as very or fairly important; no difference was observed between genders. This outcome is substantially higher when compared to the other four organizations involved in this survey, which scored an average of 54% of importance with the support received. These results confirm the need for research capacity strengthening in low-and middle-income countries and the catalytic role that TDR has played in research career development. Two additional elements, the first post-doctorate employer and the academic advisor, were rated as important for career progression by 64% and 63% of the respondents, respectively.
The post-degree career choices made by the respondents were strongly influenced by academic considerations. The most important reason influencing the decision to accept a post-doctorate position was the willingness to get additional training in the same area of their degree (70%). This was seen as a necessary step towards the employment they aspired to (67%). This is an important result to be taken into consideration when implementing future research capacity strengthening programmes for development. Increasingly, countries have identified the need for building capacity in research for implementation in order to enhance health care delivery and reach vulnerable populations. Research for implementation helps solve implementation bottlenecks, identify optimal approaches for real life settings and speed up the bench-to-bedside translation. TDR has recently shifted its strategic focus toward research for implementation, and is building upon capacity already developed with previous trainees.
The survey generated valuable information that highlighted the positive impact of TDR training grants on the research career development of its trainees. The response rate (68% of all TDR trainees contacted) was high in comparison to average online surveys (30%) [11].
However, this study presents two main limitations. Fist the population of TDR grantees who responded is small, i.e. 77 respondents from the 117 TDR trainees who had a valid e-mail (66%) and from the total of 304 TDR trainees (25%) who had initially been contacted. This illustrates the challenges to maintain contact with past trainees, as identified in previous evaluations of TDR’s capacity building activities [2]. In order to help keep track of former trainees, TDR launched the TDR Global initiative in 2016. The TDR Global platform, is an efficient and flexible web-based platform based on an existing open access “research networking tool”. It builds profiles of researchers affiliated to TDR and maps their expertise, their research activity and academic networks based mainly on their publications and co-authorship. The platform also helps track their career and professional achievements based on data they provide. This platform was launched publicly in November 2016 and data on its use and utility are being collected [12].
In addition, influence of the trainee’s selection on the training intervention outcome is difficult to assess. The current survey was not designed to analyse this element. Heads of institutions supported by TDR expressed in a previous survey [2] that TDR supported training had a high impact on the ability to develop research project. This, suggests that at least TDR supported training made a difference in some research skills.
The results presented in this paper highlight the important link perceived by respondents between TDR support and their career advancement. Most of the trainee respondents (80%) rated TDR support as a very important factor that influenced their professional career achievements. In order to address the potential social desirability bias (i.e. respondent giving a positive answer to please the questioner) a multiple choice questionnaire was included asking the importance of: (1) sponsoring organization; (2) the PhD supervisor/ mentor; and (3) the employer.
A high proportion of respondents (89%) remained in the field of research. The return rate to their region of origin (96%) is high with a very limited ‘brain drain’ rate to high-income countries (4%). These results do not take into account the 75% of trainees who could not be followed up. The TDR Global platform should potentially allow for a more comprehensive analysis. In the meantime, in order to assess the level of trainees who remained in the field of research, TDR developed a short survey on 212 trainees supported by TDR in Brazil. Brazil has a national research information system called Lattes which is coordinated by the Brazilian National Council for Scientific and Technological Development (CNPq). It is mandatory for researchers to fill in their profile on Lattes in order to apply for grants, faculty positions or staff appraisal. A search in the public interface of Lattes (http://lattes.cnpq.br/) showed that 86% of the 212 Brazilian TDR trainees had updated their profile in Lattes in the past two years and were still involved in research. Although Brazil is merely an illustrative example, this result reinforces the role of TDR on developing research capacity in low- and middle-income countries.
For decades, research capacity strengthening programmes targeting scientists in LMICs focused on north-south collaboration. According to the UNESCO Science Report 2015, from 2008 to 2014, the top three partners for the Economic Community of West African States (ECOWAS) came from France, the United States of America, and the United Kingdom, in that order [13, 14]. During this period, efforts increased research productivity in LMICs to a small extent. For example, in Sub-Saharan Africa the number of researchers rose from 0.9% to 1.1% (58 800 to 82 000) while in South Africa the number of researchers remained stable (0.3%). According to the same report, between 2008 and 2014, the percentage of worldwide scientific articles from Sub-Saharan Africa rose from 1.2 to 1.4 and from 0.5 to 0.7 in South Africa.
Some programmes have promoted a south-south collaboration approach to effectively address local health research problems and needs. An example is the Consortium for Advanced Research Training in Africa (CARTA) which is part of the African Institutions Initiative supported by the Wellcome Trust. CARTA aims to make a difference by rebuilding and strengthening the capacity of African universities to train locally skilled researchers [15]. A real time evaluation of the first four years of the CARTA programme [16] shows that although a critical mass of PhD and MSc graduates has been created, the long term impact, as for all the capacity building programmes, is still to be demonstrated. Indeed, although south-south collaboration should offer the possibility of facilitating the transfer of knowledge and best practices across the institutions [15], the effectiveness of this approach has to be carefully analysed [17]. The results presented in this article do not show any difference for a respondent from AFR, whether they studied in an LMIC or a HIC. This highlights the potential cost effectiveness of south-south collaboration. Collaboration across regions encourages mobility which is an important factor to develop independence following a post-doctoral position and gain leadership skills. Interestingly, most of the TDR trainee respondents worked in their own region during the period following their TDR grant. It would be important in a future study to analyse the factors involved in this low level of mobility and the level of south-south collaboration as well as north-south-south collaboration.
The survey also identified the challenges, bottlenecks and opportunities that trainees faced at various stages of their research careers. Although women are well represented in most WHO regions (except AFR), they do not always reach the same level and salary as men do. As a result of this survey, TDR initiated a new Women in Science programme to explore how to help more women enter and stay in science careers. Factors influencing access to TDR training grants from non-English speaking countries have not been identified properly and would need to be studied in future surveys.
The lessons learnt from this study are summarized below:
The results of this study help highlight some factors influencing the effectiveness of TDR’s capacity strengthening programmes from 2000 to 2012. Lessons learnt could also help donors and policy-makers when setting programmes and funding priorities.
|
10.1371/journal.ppat.1003193 | Phylodynamic Analysis of the Emergence and Epidemiological Impact of Transmissible Defective Dengue Viruses | Intra-host sequence data from RNA viruses have revealed the ubiquity of defective viruses in natural viral populations, sometimes at surprisingly high frequency. Although defective viruses have long been known to laboratory virologists, their relevance in clinical and epidemiological settings has not been established. The discovery of long-term transmission of a defective lineage of dengue virus type 1 (DENV-1) in Myanmar, first seen in 2001, raised important questions about the emergence of transmissible defective viruses and their role in viral epidemiology. By combining phylogenetic analyses and dynamical modeling, we investigate how evolutionary and ecological processes at the intra-host and inter-host scales shaped the emergence and spread of the defective DENV-1 lineage. We show that this lineage of defective viruses emerged between June 1998 and February 2001, and that the defective virus was transmitted primarily through co-transmission with the functional virus to uninfected individuals. We provide evidence that, surprisingly, this co-transmission route has a higher transmission potential than transmission of functional dengue viruses alone. Consequently, we predict that the defective lineage should increase overall incidence of dengue infection, which could account for the historically high dengue incidence reported in Myanmar in 2001–2002. Our results show the unappreciated potential for defective viruses to impact the epidemiology of human pathogens, possibly by modifying the virulence-transmissibility trade-off, or to emerge as circulating infections in their own right. They also demonstrate that interactions between viral variants, such as complementation, can open new pathways to viral emergence.
| Defective viruses are viral particles with genetic mutations or deletions that eliminate essential functions, so that they cannot complete their life cycles independently. They can reproduce only by co-infecting host cells with functional viruses and ‘borrowing’ their functional elements. Defective viruses have been observed for many human pathogens, but they have not been thought to impact epidemiological outcomes. Recently it was reported that a lineage of defective dengue virus spread through humans and mosquitoes in Myanmar for at least 18 months in 2001–2002. In this study, we investigate the emergence and epidemiological impact of this defective lineage by combining genetic sequence analyses with mathematical models. We show that the defective lineage emerged from circulating dengue viruses between June 1998 and February 2001, and that it spreads because—surprisingly—its presence causes functional dengue viruses to transmit more efficiently. Our model shows that this would cause a substantial rise in total dengue infections, consistent with historically high levels of dengue cases reported in Myanmar during 2001 and 2002. Our study yields new insights into the biology of dengue virus, and demonstrates a previously unappreciated potential for defective viruses to impact the epidemiology of infectious diseases.
| Although the high deleterious mutation rate of RNA viruses ensures that many genomes are defective [1], the long-term evolutionary and epidemiological consequences of the presence and transmission of defective viruses rarely have been discussed. To date, most work has focused on laboratory studies of defective-interfering (DI) viruses (or particles); these are characterized by major deletion mutations which give them a replication advantage over full-length viral genomes [2]–[3]. DI particles interfere with the functional virus by competition for materials essential for replication and transmission, such as polymerase enzymes or capsid proteins [4]. It has been hypothesized that, by limiting the production of the functional virus, DI particles may play an important role in persistent infections [5] and could even serve as therapies for viral infections [6]–[8]. Recently, defective viruses with full-length genomes also have been recognized as a general phenomenon in many major human pathogens, including human immunodeficiency virus, hepatitis B virus, hepatitis C virus, West Nile virus, and dengue virus [2], [9]–[16]. However, the role they play in natural viral populations is unclear [17] and an epidemiological impact has not been demonstrated.
It was shown recently that a defective dengue virus (DENV) lineage, characterized by a full-length genome and truncated E protein, was able to able to pass between individuals in a natural transmission cycle involving humans and mosquitoes during 2001 and 2002 in Myanmar [14]. Dengue virus is a vector-borne RNA virus that is transmitted between humans and mosquitoes, and infects 50–100 million people globally each year. There are four serotypes of dengue virus (DENV-1 to 4), all of which are endemic in Myanmar [18]. This lineage of defective viruses arose from a point mutation that introduced a stop codon in the viral envelope (E) glycoprotein gene of DENV-1. It transmitted persistently for at least 18 months, increased markedly in frequency from 2001 to 2002, and also spread thousands of kilometers to other geographic areas such as New Caledonia and Singapore [14]. Although defective viruses are frequently found in dengue patients [3], [14], [19], their long-term transmission was surprising since they were not thought to transmit naturally between hosts.
The most likely explanation for the persistence of the defective DENV genomes in nature is complementation with a fully competent strain of the same virus (here termed the ‘functional virus’) in dually infected cells. The strict requirement for complementation by the functional virus means that sustained transmission of defective particles requires frequent infection of host individuals (and host cells) by both types of particles. Given that the phenotypic effect most commonly associated with defective particles is reduced production of the functional virus due to interference, the mechanisms responsible for the emergence and spread of this defective lineage of DENV-1 are mysterious. Intriguingly, Myanmar saw historically high levels of reported dengue cases during the period when the transmitted defective particle (tDP) was reported [18].
In this study, we address the questions of how the tDP emerged and spread and evaluate its relationship with dengue transmission using a phylodynamic approach [20]–[21]. This approach unifies phylogenetic and dynamic modeling techniques to analyze genetic and epidemiological data, and thus is an important tool for the study of emerging viruses. The emergence of the tDP lineage also highlights several general research themes regarding viral emergence. Successful sustained transmission of a newly emerged viral strain depends on a complicated interplay between evolutionary and ecological processes [22]. On one hand, the high mutation rate and short generation time of RNA viruses mean that evolutionary processes occur rapidly, and evolutionary outcomes depend on the fitness of viral mutants at different stages of viral life-cycles and, sometimes, on the interactions between viral variants [23]–[27]. On the other hand, ecological and epidemiological factors, such as host contact patterns [28]–[29], transmission routes [30] and host movements [31]–[32], also determine the success of viral transmission in a population. The unexpected finding of dengue tDPs raises questions about how these factors interacted to shape the emergence of the tDP, and what impact the defective lineage may have had on the epidemiology of competent dengue strains. These questions highlight gaps in our understanding of the possible roles played by defective particles in the transmission biology of all viruses, and the potential for defective particles to emerge as circulating infections in their own right, i.e. always super-infecting over a functional virus, in the manner of satellite viruses [17]. More broadly, the example of tDPs offers opportunities to examine the roles that factors such as complementation and interaction between viral variants play in the process of viral emergence.
To examine the evolution of the tDP lineage with its associated (functional) DENV-1 lineages, we first categorized the sequences reported by Aaskov et al. [14] into the three distinct lineages identified in their study. We denote the three lineages as the stop-codon lineage (the lineage consisting of the tDP isolates), the wt-1 lineage (the functional DENV-1 lineage associated with the stop-codon lineage) and the wt-2 lineage. Within each host, most values of the ratio of nonsynonymous to synonymous substitutions per site (dN/dS) for E genes from the stop-codon lineage are close to 1, indicating neutral evolution of the defective lineage; in contrast, all the dN/dS values calculated from the competent lineages are below 1, indicating purifying selection acting on the competent lineages (Table S1). This is consistent with typical characteristics of defective particles and the conclusion that the E gene of tDPs does not encode functional protein [14]. We further examined the sequences of 24 isolates from a patient in New Caledonia sampled in February 2003 [13], and found that, among the 24 sequences, seven belong to the stop-codon lineage and four belong to the wt-1 lineage, confirming that the defective isolates and some functional isolates were derived from the lineages circulating in Myanmar.
To estimate the time of tDP emergence, we first derived consensus sequences for the wt-1 viral lineage and the stop-codon lineage in each individual where more than one viral isolate was available (Fig. 1A). To cover a broader time period in our estimation, we randomly selected eight additional DENV-1 sequences isolated in Myanmar during 1998 and 2001 [18]. A Maximum Clade Credibility (MCC) tree estimated using BEAST [33] showed that, as expected, all stop-codon sequences fall in the same clade of the tree, and that both the functional viruses and the tDP detected in New Caledonia were transmitted from Myanmar (Fig. 1B). The estimated time of divergence of the stop-codon lineage from the wt-1 lineage is February 2000 (95% Highest Probability Density: June 1998 to February 2001).
The MCC tree (Fig. 1B) revealed two additional interesting features. First, the consensus sequences of wt-1 viruses isolated from tDP-infected individuals are closely related to the stop-codon lineage (red branches in Fig. 1B). In fact, all those sequences except individual 47662 (47662_1, where ‘1’ denotes the wt-1 lineage) form a monophyletic group with sequences of the stop-codon lineage (green box in Fig. 1B), suggesting they are derived from a common ancestor. (A tree topology test showed that alternative trees in which sequence 47662_1 belongs to this monophyletic group cannot be excluded statistically; see Fig. S1 and Table S2). This clustering pattern suggests that the DENV-1 lineage isolated from dually infected individuals shares the same transmission history as the stop-codon lineage. This would be expected if the tDP co-transmits with functional DENV-1, i.e. if both DENV-1 viruses and tDPs are transmitted simultaneously to new hosts in the same contact event. The second interesting feature is that the stop-codon lineage rose from being quite rare in 2001 to being found in all sampled patients infected by DENV-1 in 2002 (Fig. 1B), indicating a possible transmission advantage for this stop-codon lineage. Below, using dynamical models, we provide evidence that tDP is primarily co-transmitted with DENV-1 and that this co-transmission increases transmission fitness, thereby allowing tDP to rise to a high frequency in the population.
To identify potential mechanisms that allow for sustained transmission of the tDP, we constructed a seasonally forced dynamical model for the transmission dynamics of the tDP and DENV-1 by combining aspects of established models for dengue [34] and defective particles [7] (see Methods and Text S1). We first focus on the possible mechanisms of transmission of tDPs. The donor host (either human or mosquito) must be dually infected with DENV-1 and tDP, and we consider two types of contacts that may lead to tDP transmission (Fig. 2A and B): contact either with uninfected susceptible individuals (possibly leading to infection of the susceptible with one or both viruses) or with individuals infected with DENV-1 only (possibly leading to super-infection with tDP). Three types of transmission events are possible: transmission of tDP only (which matters only if the host is already infected with DENV-1), transmission of DENV-1 only, or transmission of both tDP and DENV-1. The rates at which these three alternatives occur, relative to the rate of transmission of the functional virus from DENV-1-infected hosts, are modeled using dimensionless scaling parameters P, Q and W, respectively (Fig. 2C). These three parameters incorporate the changes in viral transmission rates from dually infected human and mosquito individuals as a result of all relevant factors including changes in viral titers and host movement or behavior patterns.
To investigate the key mechanisms contributing to tDP emergence and transmission, we simulated the model with different values of P, Q and W while holding other parameters constant. Note that the values of P and Q, i.e. transmission of tDP only and DENV-1 only, are probably small because of the high number of viruses thought to be transmitted between human and mosquito [35]. Nonetheless, we allow them to vary in a wide range (0 to 1) to be comprehensive. We found that the essential determinant of long-term transmission of the tDP is the value of parameter W, i.e. the efficiency of co-transmission of both the tDP and functional DENV-1. Continuous transmission of the tDP over multiple years requires that co-transmission of tDP and DENV-1 is more efficient than transmission of wild-type DENV-1 in the absence of tDP (i.e. W>1.0, irrespective of the values of P and Q, as shown in Fig. 3A). For the abundance of dually infected individuals (D) to rise to a level comparable to DENV-1 infected individuals (I) within 3 years, as observed in the data from Myanmar, the co-transmission of tDP and DENV-1 must be 15% more efficient than the wild-type transmission (W>1.15, red dots above the horizontal dashed line in Fig. 3A), averaging over humans and mosquitoes. An alternative explanation of the observed rise in frequency of dually infected individuals is genetic drift without any transmission advantage. To test the validity of our deterministic modeling approach, we conducted a stochastic analysis based on a Wright-Fisher model and found that the probability that the observed rise in frequency occurred due to purely neutral evolution is extremely small (see Text S1 for details). Therefore, from epidemiological arguments, transmission of tDP is driven primarily by co-transmission of tDP and DENV-1, which is more efficient than transmission of DENV-1 by singly-infected hosts. However, in simulations with considerably higher values of W, both DENV-1 and the tDP go extinct due to depletion of susceptible individuals during the post-epidemic refractory period.
To better understand the transmission biology of the defective virus, we evaluated the relative importance of the two mechanisms of transmission in driving tDP spread. By comparing the numbers of dually infected human individuals arising from each type of contact, we found that super-infection accounts for <8% of dually-infected cases (Fig. 3B), indicating that co-transmission of the tDP and DENV-1 to uninfected individuals is by far the dominant transmission route (Fig. 3B). The reason is that the number of DENV-1 infected individuals is much smaller than the number of susceptible individuals for both humans and mosquitoes in dengue-endemic areas [34], [36], and therefore the rate of contact between dually infected and DENV-1 infected individuals is too low to maintain sustained transmission by super-infection. Furthermore, the fraction of super-infected individuals in the simulation (Fig. 3B) is likely an overestimate, since other factors not considered in the model, such as super-infection exclusion, a process whereby an infected cell cannot be infected with the same or a closely related virus [37], may further restrict the frequency of super-infection events.
To characterize the epidemiological conditions that allow the tDP to emerge and rise to high frequency in the infected population, we calculated the effective reproduction number, Reff,co (see Eqn.2 in Methods), for the co-transmission route in a simplified model that ignores seasonal forcing and super-infection, i.e. transmission route denoted by dashed lines in Fig. 2C. We found that successful invasion of tDP (Reff,co>1) depends on the values of four parameters that characterize the dually infected individuals: the infectious period of dually infected humans (1/γH,D), the incubation period of dually infected mosquitoes (1/σV,D), and the relative efficiencies of co-transmission by dually infected humans (WH) and mosquitoes (WV). The overall co-transmission parameter W, analyzed in Fig. 3, is the geometric mean of WH and WV. Apart from parameters WH and WV, the dependence of Reff,co on the parameters γH,D and σV,D arises from the altered durations of the infectious periods of dually infected humans and mosquitoes, respectively. Note that since dengue infections of mosquitoes are life-long, a shorter incubation period increases the time spent in the infectious state and therefore increases transmission potential.
To assess the possible epidemiological impact of tDP emergence, we simulated the full model from the emergence of tDP (assumed here to occur in year 2000) through the period for which we have data (to the end of 2002), for a range of biologically plausible parameter values (see Methods). We randomly sampled the four parameters that determine the value of Reff,co, and the two additional transmission parameters P and Q, and computed the value of Reff,co for each simulation. When Reff,co<1, the fraction of human cases that were dually infected in year 2002 was negligible, and the total number of dengue cases during the three years after tDP emergence did not change appreciably from the number in the absence of tDP. In contrast, values of Reff,co>1 led to increases in both the fraction of dengue-infected humans who were dually infected and in the total number of dengue cases (Fig. 4). Interestingly, the model reveals a lower bound of the fold increase in total dengue cases for a given observed fraction of dually infected individuals. This is because increases in the fraction of dually infected individuals result from more efficient co-transmission, which also increases the total number of infected individuals. Aaskov et al. reported that 5 out of 5 human patients sampled in 2002 were dually infected [14]. With reference to the results in Fig. 4, the observation that all cases were dually infected in 2002 predicts a 2.5–4 fold increase in total dengue cases during 2001 and 2002, though of course the sample size is small so uncertainties are large. (For 5/5, the lower bound of the 95% C.I. for the proportion is 0.48, which corresponds to a lower bound of a 1.3 fold increase in total cases.)
To derive more robust estimates of the possible impact of tDP transmission on overall dengue transmission, we used a likelihood framework to estimate the value of Reff,co based on previously reported data [14] and on our finding that tDP emerged between June 1998 and February 2001 (see Text S1). Since Reff,co is influenced by the four parameters describing dual infections, as shown above, and the realized changes in these four parameters are unknown, we explored four scenarios where the changes in Reff,co arise entirely from changes in each parameter. Maximum likelihood (ML) estimation was used to infer parameter values for each scenario, yielding an estimate of Reff,co, and the time of tDP emergence (temg) was estimated simultaneously (see Fig. S2 for a comparison between data and simulation using ML parameter values). The same qualitative picture emerges for all four scenarios (Table 1): the ML estimates of Reff,co fall in the range 1.24–1.28 (95% C.I.: 1.13–1.89), giving rise to a 2.3–3.2 (95% C.I.: 1.4–4.1) fold increase in overall DENV-1 cases during 2001 and 2002 as a result of tDP transmission. Sensitivity analysis (see ) showed that our results are robust to the assumed magnitude of seasonality in mosquito populations (Table S3 and S4) and phase of multi-annual cycles in dengue incidence (Table S5 and Fig. S3). For all scenarios analyzed, the ML estimate of Reff,co falls between 1.2 and 1.3. Because of limited data and inherent challenges in fitting the non-stationary dynamics of a complex system, we interpret these results not as precise estimates but as confirmation of our qualitative conclusion that co-transmission of tDP and DENV-1 has a substantially increased transmission potential, which in turn is expected to lead to elevated incidence compared with DENV-1 alone.
Our results reveal a significant impact of transmissible defective particles (tDPs) on the epidemiological dynamics of dengue virus, a phenomenon that has not been reported previously for any human pathogen. We first showed that co-transmission of tDP and the functional virus to uninfected hosts is the primary mechanism of tDP transmission, and, unexpectedly, this co-transmission route has a higher transmission potential than the transmission of the functional virus only. This qualitative conclusion is robust to assumptions about parameter values and underlying epidemiology of dengue virus. Based on this higher transmission potential, our model predicts a substantial increase in the total DENV-1 incidence during 2001 and 2002, which is consistent with the historically large outbreaks reported in Myanmar during this period [18].
The finding that co-infection of previously uninfected individuals constitutes the primary transmission route sheds light on the biology of dengue infection and transmission. Successful establishment of defective particles in a newly infected host requires that dual infections of host cells occur frequently throughout the full course of infection, including in the initial stage. This indicates that a large number of virions must be transmitted, consistent with the idea of a relatively wide transmission bottleneck for dengue [9], [13]–[14]. It also suggests that the process of viruses entering host cells near the site of infection is highly constrained spatially, such that the infecting dose of virions is restricted to a relatively small number of host cells available for infection. To maintain the transmission chain, these conditions must hold for co-infections in both humans and mosquitoes, although the relevant mechanisms of infection are completely different.
Several different mechanisms could account for the higher transmission potential of dually infected hosts relative to singly infected hosts. It is possible that the higher transmission arises from intrinsic properties of the functional DENV-1 genotypes in dually infected individuals, and the defective lineage has no effect. In this case the unprecedented finding of a co-transmitted defective lineage (and 100% frequency of dual infection in 2002) is strictly coincidental, and has no causal relationship with the increased fitness of its associated DENV-1 lineage. We think this is unlikely. A more parsimonious explanation is that the tDP increases the transmission potential by modulating the within-host replication of DENV-1 from a non-optimal level. Previous work on the theory of virulence evolution suggests that there exists an optimal viral load that maximizes transmission potential [38]–[39]. Transmission increases with viral load when viral loads are low, but once viral loads exceed the optimal value, the negative impact of viruses on the host (virulence) removes the host from being infectious, e.g. via host death or hospitalization, thereby decreasing the virus's transmissibility. In light of this, we postulate two potential mechanisms by which the tDP could modulate transmission. The first is that the tDP reduces viral loads through interference, as is known for some other defective particles [3]. Lower viral loads lead to milder disease [40], which allows dually infected humans to be more mobile than humans infected with wild-type DENV-1 only. Because the spatial spread of dengue is driven chiefly by human movements [32], dually infected humans can facilitate greater disease dissemination. This scenario is plausible if the virulence of the functional virus in humans exceeds the optimal value for transmission. The second possible mechanism is that the tDP increases production of the functional virus, by circumventing constraints in viral gene expression within a cell. Differential gene expression is a major challenge for (+)ssRNA viruses such as dengue, because of constraints arising from their genomic architecture and particularly the necessity to translate individual protein products from a single polyprotein precursor [41]. The presence of tDP in either infected human cells or mosquito cells could increase the abundance of gene products that are otherwise limiting, thereby increasing virus fitness. This scenario is plausible if the current dengue viral loads in either humans or mosquitoes are below the optimal value for transmission. While we cannot discriminate between these competing hypotheses with current data, they could be tested by measuring the relative viral load or clinical severity of dually infected versus singly infected hosts, and would yield interesting insights about the virulence of DENV-1.
The potential for the presence of tDPs to increase the transmission potential of DENV-1 suggests that tDPs may emerge and spread often, raising the question of why tDPs have not been reported more frequently and in more geographic regions. This could be explained by study designs that have focused almost exclusively on consensus sequences, thereby avoiding any dissection of intra-host genetic variation. In addition, our simulations (Fig. 3) suggest that higher transmission potential of the co-transmission route may cause the tDP to go extinct due to depletion of the susceptible population following epidemics. Hence, tDPs may have emerged and died out multiple times in history. Indeed, defective DENV-1 lineages harboring the same stop-codon mutation have been identified elsewhere on at least one occasion [19]. Finally, the conditions that favor tDP emergence may depend on local ecological or epidemiological factors, such as human movement patterns, vector species or strains, and immunological interactions between the four serotypes of dengue. More intensive sampling, and sequencing efforts focusing on intra-host dengue diversity, would help to characterize the true frequency of tDP emergence and spread in populations worldwide.
Our model predicts that the emergence of tDPs should lead to a substantial increase in DENV-1 incidence. This prediction arises wholly from our finding that co-transmission of tDP and DENV-1 is more efficient than wild-type transmission, which is derived only from the rise in relative frequency of dual infection and is robust across epidemiological backgrounds. Although limited data prevent a precise assessment of this prediction, it is consistent with the observation that the number of reported dengue cases reached historically high levels during the 2001 and 2002 seasons [18]. Of course, many other factors can influence dengue epidemiology, such as immunological interactions arising from switches in dominant serotypes [42] or changes of fitness resulting from mutations elsewhere in the viral genome [43]. However, these outbreaks do not share patterns typically associated with serotype switches, since all four dengue serotypes were circulating in Myanmar leading up to the large outbreak in 2001, and almost half of the DENV-1 infections in 2001 were primary infections [18]. It is also possible that an increase in reported incidence could be explained by improved surveillance, although this is unlikely given that a comprehensive clinical and laboratory surveillance program has been established in Myanmar since 1984 and did not change in the years when higher numbers of cases were reported.
The existence of tDPs for dengue virus raises the possibility that sustained transmission of defective particles may be a more general phenomenon for other viruses. Our analyses highlight several conditions that facilitate long-term spread of defective particles: 1) relatively wide transmission bottlenecks, 2) frequent dual infection at the level of hosts and the level of cells, and 3) potential to increase the transmissibility of the functional virus by modulating the viral load within hosts. Interestingly, a recent study provided evidence that a lineage of defective particles of canine influenza transmitted for at least 4-months in a high-density dog population [11]. Similarly, the transmission of defective particles (characterized by stop codon mutations) in experimental transmission studies of swine influenza among pigs has also been reported [44]. As sampling efforts focusing on within-host genetic diversity become more common, it is likely that tDPs will be observed more frequently than currently appreciated. This would narrow the functional distinction between defective particles and satellite viruses, another well-known class of transmissible sub-viral agents, which also require complementation but are not immediately derived from their helper viruses [17].
Finally, our work reveals some general principles concerning viral emergence. First, complementation can be a powerful factor in determining the evolutionary dynamics of natural viral populations. The extreme case reported here, with long-term spread of a totally defective viral lineage, has implications for emergence pathways that need to cross fitness valleys [45]. If co-infection is common, a lineage could easily cross a wide fitness valley by pairing with a competent strain. Second, interactions among strains that lead to modified virulence (or other host-level phenotypes) can lead to increased transmission fitness [23], [46], and hence emergence. Altogether, this case study expands the range of mechanisms that may be pertinent to the study of viral emergence, and re-emphasizes the need to use appropriate models, at intra- and inter-host scales, to understand the processes giving rise to epidemiological patterns and associated pathogen sequence data.
The 290 sequences and the eight additional sequences from Myanmar were extracted from Genbank (accession numbers DQ264868 to DQ265157, AY588273, AY606062, AY618877, AY618878, AY618880, AY620948, AY620950 and AY726555). The 24 additional sequences from the patient in New Caledonia were obtained from Ref. [13]. The relationship of these 24 sequences with the sequences from Myanmar was evaluated by constructing a phylogenetic tree using the maximum likelihood method in Garli 0.951 [47], employing the GTR+I+Γ4 model of nucleotide substitution. The ratio of nonsynonymous to synonymous substitutions per site (dN/dS) was estimated using the Datamonkey webserver [48] employing the SLAC method [49]. The time of tDP emergence was estimated by constructing a Maximum Clade Credibility (MCC) tree using BEAST, again employing the GTR+I+Γ4 model [33]. The time to common ancestry was estimated using the Bayesian skyline coalescent model and an uncorrelated lognormal relaxed clock model [50], with a total of 5,000,000 states collected from the MCMC chain and the first 500,000 states excluded as burn-in. The effective sample size for each parameter in the estimation was checked using Tracer v1.5 to ensure convergence [51], with statistical uncertainty reflected in values of the 95% Highest Probability Density (HPD). Note that the exact months when the additional eight sequences [18] were isolated is unknown. We assumed that they were isolated in June of the appropriate year, and confirmed that estimation of the time of tDP emergence was robust to the choice of month of isolation. Nucleotide sequences and relevant parameters estimated in the software are available from authors upon request. Tree topology tests were performed using TREE-PUZZLE 5.2 with HKY+Γ4 [52].
We constructed a human-vector SEIR compartmental model considering the dynamics of DENV-1 and tDP (schematic shown in Fig. 2C). This model considers the demographic changes of human and mosquito populations, with the mosquito birth rate seasonally forced in accordance with monthly data [53]. It keeps track of the infection dynamics of one dengue serotype (DENV-1) and its associated defective particles (tDP) at the population scale. The full model is shown in Eqn. 1. The human population size (NH) is assumed to be constant, with individuals born into the susceptible compartment (SH) at per capita rate μH, and all human individuals subject to per capita death rate μH. The rate constant for transmission of DENV-1, encompassing the contact rate and probability of transmission, is β. When susceptible humans have contact with dually infected mosquitoes, three types of transmission events can potentially occur (Fig. 2A,B): tDP transmission, DENV-1 transmission and dual transmission. The three scaling parameters, P, Q and W, are used to model the efficiency of these three types of transmission, respectively, relative to the transmission rate (β) from individuals infected only with DENV-1. We assume that dually infected individuals may have different infection characteristics from DENV-1 infected individuals, and hence may differ in the latent period, infectious period and recovery rate. The mean latent periods of DENV-1 infected individuals and dually infected individuals are 1/σH and 1/σH,D, respectively. The DENV-1 infected latent (E) or infectious (I) individual can move to the dually infected latent (G) or infectious (D) compartment if they are super-infected by tDP. The infectious DENV-1 infected individuals (I) and dually infected individuals (D) recover to become recovered (and immune) individuals (R) at rates γ0 and γ1, respectively.
For the mosquito population, we do not consider vertical transmission of dengue virus, since it has been shown that vertical transmission at the rates reported in the literature does not have a strong impact on transmission dynamics [54]. We explicitly consider the seasonal forcing (a*cos(2*π*t+b)) of the mosquito birth rate due to changes in rainfall and temperature. The infection dynamics for the mosquito are modeled in a similar manner to the human infection dynamics, except that mosquitoes do not recover from dengue infection. State variables for the mosquito population match those for the human population, with an additional subscript ‘V’.
The resulting ordinary differential equation model is shown below:(1)
The description and initial values of the state variables are shown in Table S6, and the description and the values of the parameters in the equation are shown in Table S7.
We do not consider other dengue serotypes and their immunological effects on DENV-1 transmission in our model. This is because DENV-1 was the major circulating strain from 2000 to 2002 in Myanmar, and almost half of the DENV-1 infections in 2001 were primary infections [18], indicating that serotype interactions were not the dominant driver of observed dynamics. Importantly, tDP has been associated with DENV-1 only, and thus it experiences the same competitive interactions with other serotypes as wild-type DENV-1. Therefore, the increased frequency of dually infected individuals among DENV-1 infected individuals (which is the primary basis for our model conclusions) should be independent of the interactions with other serotypes. Hence, this simplification of the model will not alter the major findings of our study.
The model was first simulated without tDPs to establish the baseline endemic dynamics of dengue in the population. Based on the observation that DENV-1 incidence peaked in 2001 and 2002 in Myanmar, we defined two consecutive years with peak incidences of dengue to be 2001 and 2002 in the simulation. Using this simulation scenario as a baseline, tDP was introduced into the system at the beginning of year 2000, and the model was simulated another 3 years after tDP introduction to generate the results for Figs. 3 and 4. The qualitative conclusion that the presence of tDP increases DENV-1 transmission is robust to the choice of this mapping between simulation and calendar years, as long as dengue is endemic in the model (Fig. S3 and Table S5), but quantitative predictions of the magnitude of the rise in DENV-1 incidence differ. The details of the sensitivity analyses and the likelihood-based procedure for estimation of Reff,co and temg are presented in Text S1.
We performed a next-generation matrix analysis [55] on the simplified model, to calculate the effective reproduction number, Reff,co, for the co-transmission route. Reff,co is defined as the average number of secondary dually infected cases infected through the co-transmission route by the first dually infected individual, when it is introduced into the system where the number of DENV-1-only infected cases is at non-zero equilibrium. Then, the condition for tDP emergence is Reff,co>1. Given our emphasis on the phenomenon of co-transmission, we further distinguish co-transmission from human-to-mosquito and mosquito-to-human using parameters WH and WV, respectively. We also allow for the possibility that dually infected human and mosquito individuals might have incubation periods and infectious periods that differ from singly infected individuals. Then, Reff,co can be approximated as:(2)
Therefore, the emergence of the tDP is determined by four parameters characterizing the dually infected individuals: the scaling parameters for force of infection, WH, WV, the infectious period of dually infected humans, 1/γH,D and the incubation period of dually infected mosquitoes,1/σV,D.
|
10.1371/journal.ppat.1006777 | Two classes of protective antibodies against Pseudorabies virus variant glycoprotein B: Implications for vaccine design | Pseudorabies virus (PRV) belongs to the Herpesviridae family, and is an important veterinary pathogen. Highly pathogenic PRV variants have caused severe epidemics in China since 2011, causing huge economic losses. To tackle the epidemics, we identified a panel of mouse monoclonal antibodies (mAbs) against PRV glycoprotein B (gB) that effectively block PRV infection. Among these 15 mAbs, fourteen of them block PRV entry in a complement-dependent manner. The remaining one, 1H1 mAb, however can directly neutralize the virus independent of complement and displays broad-spectrum neutralizing activities. We further determined the crystal structure of PRV gB and mapped the epitopes of these antibodies on the structure. Interestingly, all the complement-dependent neutralizing antibodies bind gB at the crown region (domain IV). In contrast, the epitope of 1H1 mAb is located at the bottom of domain I, which includes the fusion loops, indicating 1H1 mAb might neutralize the virus by interfering with the membrane fusion process. Our studies demonstrate that gB contains multiple B-cell epitopes in its crown and base regions and that antibodies targeting different epitopes block virus infection through different mechanisms. These findings would provide important clues for antiviral drug design and vaccine development.
| Pseudorabies virus (PRV) is an emerging veterinary pathogen that infects many domestic animals. Since 2011, highly pathogenic PRV variants have emerged in many farms in China and posed great economic burdens to the animal industry. However, the current marketed vaccines cannot provide effective protection against these emerging strains. The envelope glycoprotein B (gB) is a major viral antigen known to play a crucial role in PRV entry. In order to control PRV epidemics and treat associated diseases, we combined structural and immunological approaches to generate potential neutralizing antibodies targeting PRV gB and investigate their working mechanisms. A total of 15 monoclonal antibodies (mAbs) were identified with good neutralizing activity. Among them, fourteen mAbs target the domain IV of PRV gB and block virus entry though complement effect. In contrast, the remaining 1H1 mAb recognizes domain I of PRV gB, which can neutralize virus entry independent of complement and probably by interfering with the membrane fusion process. Our work reveals the structural details and immunogenic properties of PRV gB and may offer important guidance for developing antiviral therapeutics and vaccines against PRV infections.
| Pseudorabies virus (PRV) belongs to the family Herpesviridae, subfamily Alphaherpesvirinae, and genus Varicellovirus [1]. It is an important nervous system tropic pathogen in livestock and infects a variety of mammalian species, including ruminants, carnivores, and rodents [2]. Pigs, the natural host of PRV, are a unique animal species that can survive a productive PRV infection and suffer life-long latent infections in the peripheral nervous system [3]. For other susceptible animals, PRV infection is usually fatal [4]. In pigs, the clinical signs of infection vary with the age of the infected individuals. Newborn piglets infected with PRV may develop nervous system disorders and even deaths. For adult pigs however, PRV infection often leads to respiratory diseases. In addition, the virus can even cross the placental barrier of pregnant sows to infect the fetuses and cause abortion [3–5].
Although several countries have eradicated PRV, such as the USA, New Zealand and many members of the European Union, it is still circulating sporadically in many regions all around the globe [6]. Highly pathogenic PRV variants have emerged in numbers of pig farms in China since late 2011 [5, 7–10]. In many large-scale farms, around 50% of pigs were infected, resulting in 3–5% mortality [7, 9]. The driving force behind the high virulence of emerging PRV variants remains unknown. The marketed attenuated live vaccine Bartha-K61 is widely used in the pig industry of China in recent years. Unfortunately, Bartha-K61 cannot confer effective protection against the emerging PRV variants [8, 10]. Thus, the epidemics have the potential to spread outside China to reach the surrounding countries, posing great threats to the pig industry of Asia-Pacific and south-east Asia. Further research on the emerging PRV variant strains is urgently required to control the epidemic situation and further eradicate the disease [11–13].
PRV is a double-stranded DNA virus with a 143-Kb genome containing at least 72 genes [2, 14]. A total of 11 different glycoproteins named gB, gC, gD, gE, gG, gH, gI, gK gL, gM, and gN are distributed in the viral envelope [2]. As the viral fusogen, gB is essential for both viral entry and cell-to-cell spread [15–18]. To induce membrane fusion, the gH-gL heterodimer is required to cooperate with the fusogenic gB [17, 19–21]. Most alphaherpesviruses also require gD to bind receptors and further activate gB to become fusion competent [19, 21, 22].
Similar to other herpesviruses such as herpes simplex virus (HSV) and human cytomegalovirus (HCMV), PRV gB induces protective humoral immunity against viral infection [23–26]. Though PRV has not been reported to infect humans, an antibody raised against PRV gB were observed to cross-react with HSV gB [27], indicating the gBs of different herpesviruses may share some common epitopes. Previously, 26 gB-specific mAbs were identified that cannot directly neutralize PRV in vitro but effectively blocked the virus entry in the presence of complement [28]. Further biochemical studies mapped the epitopes of these antibodies to be within three main regions of gB, residues 59–126, 216–279, and 540–734, respectively [29]. Most antibodies target epitopes within residues 540–734 [29]. In order to develop better therapeutics and vaccines, more efforts should be made to characterize the structural and immunogenic properties of gBs of these emerging highly pathogenic PRV variants.
In this work, we immunized mice with soluble PRV gB to generate gB specific mAbs and identified a total of 15 neutralizing antibodies that effectively block PRV entry with either complement dependent or independent mechanisms. To locate their epitopes, we also determined the crystal structure of PRV gB and verified the binding footprints by mutagenesis. These findings would enormously advance our understanding of PRV gB immunogenicity and provide important guidance for antiviral drug design and vaccine development.
We expressed the ectodomain of PRV gB with the Bac-to-Bac expression system. The purified soluble gB protein was then applied to immunize mice. A total of 43 gB-specific mAbs were originally selected by indirect ELISAs from hundreds of hybridoma cells. Virus entry inhibition assay was then conducted to assess the neutralizing activities of these mAbs. The tests were performed with addition of exogenous rabbit complement or without in parallel to identify the potential complement dependent neutralizing activities (See details in Materials and Methods). Fifteen neutralizing mAbs in total were identified, which effectively blocked PRV entry into pig kidney cells (PK-15) (Table 1). Among them, fourteen mAbs blocked the virus entry only in the presence of complement, indicating they exerted the neutralizing activity by complement effect. The remaining one, 1H1 mAb, however directly neutralized the virus without addition of complement (Table 1). Thus, 1H1 mAb might block PRV entry by interfering with either the receptor binding or membrane fusion processes.
As the only complement-independent neutralizing antibody, we further analyzed the neutralizing efficacies of 1H1 mAb against eight different PRV strains, including vaccine strains (Bartha and HB98), classical virulent strains (RA and SU), and the current emerging variant strains (HN1201, 188–5, 072–1, and BH1). Interestingly, the 1H1 mAb displayed a broad-spectrum neutralizing activity against all PRV strains tested though with varied efficacies (Fig 1). It effectively neutralized HN1201, BH1, 188–5, RA, and Bartha strains, with IC50 values ranging from 15.2 to 31.6 μg/mL. In comparison, it is less effective to block the entry of the SU, 072–1, and HB98 strains, for which the IC50 values varied from 68.31 to 92.73 μg/mL. As 1H1 mAb targets PRV gB and the gBs of different PRV strains share more than 95% sequence identities (S1 Fig), it is quite conceivable that 1H1 mAb neutralized a broad panel of PRV strains.
Previous studies have indicated residue 540–734 of PRV gB as a hotspot to elicit complement-dependent neutralizing antibodies [29]. As the gBs of HSV and PRV share a sequence identity of ~52% (S2 Fig), we assume that PRV gB would adopt a similar fold as that of HSV gB. In analogy to the structure of HSV gB, residue 540–734 is supposed to be at the upper portion of domain III central helix and the entire domain IV. Thus, we recombinantly expressed a truncated PRV gB protein, denoted as PRV gB-D_IV, which includes the domain IV and a portion of the adjacent domain III central helix (Fig 2A; S3 Fig). With both the soluble PRV gB and PRV gB-D_IV, we tested the binding of all 15 neutralizing antibodies to these two soluble proteins to locate the binding sites. Based on the ELISA experiments, all the 14 complement-dependent neutralizing antibodies can bind to both gB and gB-D_IV (Table 1; S4 Fig), indicating the epitopes of these antibodies are very probably within the domain IV of gB. With these observations and previous reports [29], this region is therefore a suitable immunogen to elicit complement-dependent neutralizing antibodies against PRV infection. In contrast, the 1H1 mAb only reacted with gB but not gB-D_IV (Table 1; S4 Fig), which implied its binding site at other portions of gB beyond the region of gB-D_IV.
To further characterize these neutralizing antibodies, we sequenced the variable regions of all these mAbs. Three pairs of mAbs were found to share the same genes for the variable regions (6D2/9B10, 3E1/7B11, and 1H9/9A10) (Table 2). Thus, we actually obtained 12 unique antibodies, among which 11 of them exert complement-dependent neutralizing activities.
In order to further characterize the immunogenic properties of PRV gB and identify the epitopes of these neutralizing antibodies, we determined the crystal structures of PRV gB and gB-D_IV at 3.1 and 2.7 Å resolution, respectively (Fig 2). The overall structures of PRV-gB and HSV gB are highly similar, with a sequence identity of 52% (S2 Fig). The two structures could be well superimposed with an overall rmsd of 0.962 Å (Fig 2C). Similar to other class III viral fusion proteins, PRV gB exists as homotrimers and each protomer could be divided into five domains. Each domain could be superimposed better with their counterpart in HSV gB than the entire gB molecule, as slight domain movement was observed for domain IV relative to the other portion of gB molecule (Fig 2C). In the middle of each gB protomer, a potential furin cleavage site was identified and thus the protein would intend to be processed into two fragments during expression in the cell (Fig 2A). Consistent with this feature, the soluble gB displayed three-band SDS-PAGE profile though eluted as a monodispersed peak in size-exclusion chromatography (S3B Fig). Similar cleavage processing was also observed in HSV gB [15], while it is unclear whether the cleavage is required for the fusogenic activity of gB.
The trimeric gB is mainly stabilized by the central helix bundle formed by domain III in the membrane-distal portion. Domain IV wraps around the top of the central helix bundle to form a crown in the bottle-shaped gB trimer (Fig 2B). The trimerization interfaces are highly stable, such that the truncated gB-D_IV could also assemble into trimers (S3C Fig; Fig 2D). The structure of gB-D_IV trimer could be ideally superimposed with the corresponding portion in the context of the entire gB ectodomain, with a rmsd of 0.598 Å (Fig 2E). The highly stable structure of gB-D_IV further supported the conclusion from our ELISA based assays that these complement-dependent neutralizing antibodies target the crown region of PRV gB and very probably within domain IV.
As the only complement-independent neutralizing antibody identified, the 1H1 mAb recognize PRV gB in regions different from all the other 14 mAbs. To further characterize the antigen recognition properties of this mAb, we solved the crystal structure of 1H1 Fab at a resolution of 2.5 Å (Fig 3A). In this structure, we observed an unusual N-linked glycosylation modification at residue N103 of the HCDR3 loop (S5 Fig). To testify whether the glycans participate in antigen recognition of 1H1 mAb, we introduced site mutations to disrupt the “NXS” motif to eliminate the glycosylation modification (Fig 3B). The mutants were then subjected to bio-layer interferometry (BLI) assay to test their binding affinities to PRV gB. Compared with the wild type 1H1 mAb, the affinities of both the NL and SL mutants decreased almost 1000 times and the binding kinetics also displayed significant differences. The intact 1H1 mAb binds PRV gB with a quite slow kinetics but seems not to dissociate, while the two mutants are much faster in both binding and dissociation processes (Fig 3C–3E). These observations demonstrated that the glycans in the HCDR3 loop play a key but not indispensible role for antigen recognition of 1H1 mAb.
To precisely identify the epitope of 1H1 mAb, we made great efforts to determine the structure of PRV gB-1H1_Fab complex. Unfortunately, we failed to obtain high quality diffractive crystals of the complex. As an alternative approach, we conducted 3-dimensional (3D) reconstruction by negative stain electron microscopy (EM) method. A 35 Å resolution EM map of PRV gB in complex with 1H1 Fab was obtained (Fig 4; S6 Fig). In this complex, there are three copies of 1H1 Fab binding to the three protomers in the bottle-shaped gB trimer, which follows the rule of 3-fold symmetry (Fig 4A–4C). The atomic structures of PRV gB and 1H1 Fab were perfectly fitted into the density. With the 3-fold symmetry and the hinged structure of 1H1 Fab, we could correctly identify the orientations of heavy chain and light chain (Fig 4A–4C). Thus, a pseudo-atomic model of PRV gB in complex with 1H1 Fab was built so that the interaction details could be inferred.
According to the structure, the epitope of 1H1 was unambiguously mapped to the bottom of domain I in PRV gB (Fig 4A and 4C). The binding of 1H1 Fab to PRV gB is likely mediated by the HCDR3, LCD1 and LCDR3 loops, among which the HCDR3 probably interacts with the 220-strand region and the LCDR1 and LCDR3 mainly recognize the 210-helix (Fig 4D and 4E). We then analyzed all the residues within 5 Å distance in the binding interface to identify the key residues governing the interactions. Among them, four residues in the 210-helix (Q206, D210, R214 and R215) and two in the 220-strand (K221 and E223) seemed to contribute the most interactions, which form a two-portion discrete footprint on the surface of PRV gB (Fig 4D and 4E).
Though only one glycan residue was observed in the density map of 1H1 Fab (Fig 3A), there is probably a long glycan chain attached to N103 in the HCDR3 loop as the glycosylation modification in eukaryotic cells often involves multiple glycan residues. Therefore, the space between the 210-helix and 220-strand could possibly accommodate the glycan chain and the 220-strand is probably involved in the interactions with glycans (Fig 4D and 4E). As glycans play a minor role in the interaction as shown by previous biochemical studies, the main binding footprint would thus fall into the 210-helix region. To further verify the location of 1H1 epitope, we performed mutagenesis on the four candidate residues in the 210-helix to test their effects on the interactions.
The 293T cells were transiently transfected with plasmids encoding full-length wild type or mutant PRV gB and the 1H1 mAb was applied to stain the transfected cells. The binding was visualized and quantified by flow cytometry (Fig 5). As expected, a PRV gB mutant with all the four residues replaced by alanine (Mut4) completely abolished the binding (Fig 5F). Interestingly, Q206A and D210A single mutations displayed no obvious effect on the binding, and mutant R215A only slightly reduced the binding affinity (Fig 5B, 5C and 5E). The R214A single mutation, however, significantly impaired the reactivity of PRV gB to 1H1 Fab (Fig 5D), with the same effect as the quadruple mutant (Mut4). To exclude the possibility that gB mutants failed to be displayed on cell surface due to misfolding, we also stained the cells transfected with gB mutants encoding plasmids by 5G12 mAb which targets the crown region of gB. As shown by the flow cytometry-based assays, all five mutants could be detected on the surface of transfected cells with similar expression levels as the wild type gB (S7 Fig), which demonstrated that the inability of 1H1 mAb to bind gB mutants expressing cells solely results from the substitution of key interacting residues within the epitope. Collectively, these observations implied that the main epitope of 1H1 neutralizing antibody is probably located in the 210-helix region of PRV gB, among which the residue R214 plays a critical role in the interactions.
Outbreaks of newly emerging highly pathogenic PRV variants in Chinese pig farms have caused serious public concerns [5, 8–10]. Poor protective efficacy of marketed vaccines and lack of effective therapeutic drugs further raised the threat that the epidemics might cross the border to affect surrounding countries. Better understanding the structures and antigenic properties of PRV proteins is therefore urgently required for developing effective vaccines and therapeutic drugs.
In common with other members in the Herpesviridae family, PRV harbors a pool of glycoproteins embedded in its envelope to form a huge machinery for virus entry, among which gB is the main fusogen responsible for inducing membrane fusion [2, 17, 27]. Previous studies have indicated gB as an effective immunogen to elicit complement-dependent neutralizing antibodies against PRV infections [26, 28, 29]. Our studies further supported the conclusion and identified other antigenic sites to elicit direct neutralizing antibodies as well. Combining our findings with previous reports, we can conclude that domain IV is the immunodominant region of PRV gB. Based on the gB structure we present in this study, domain IV is located on the apexes of the gB trimer to form a "crown", making it fully accessible for both antibody and potential receptor recognition. This domain is quite conceivable to become a hotspot for antibody targeting and thus a suitable candidate to develop subunit vaccines.
It has been established that the entry of herpesviruses involves multiple viral glycoproteins and possibly multiple receptors as well, and gB was also shown to play a role in receptor binding [17]. However, all the antibodies targeting domain IV of gB reported to date do not directly block PRV entry but dependent on the complement effect instead, as shown by cell-based assays in vitro [26, 28, 29]. This phenomenon implies that domain IV is probably not a receptor binding site for PRV gB, and that it might retain the same fold before and after membrane fusion with only domain rearrangement in the process of gB conformational changes to mediate membrane fusion. These findings would thus provide important clues to understand the entry mechanisms of PRV and other herpesviruses.
Besides, we also identified a direct neutralizing antibody that effectively blocked the entry of PRV in the absence of complement, 1H1, which targets the domain I of gB. This is the first complement-independent neutralizing epitope of PRV gB reported to date to our knowledge. The 1H1 mAb binds gB at the bottom of domain I, which is very close to the fusion loops. Although the atomic interaction details cannot be elucidated by the low-resolution EM structure, the binding region of 1H1 mAb can be definitely determined, which is also confirmed by the mutagenesis work. Three copies of Fab fragment surround the peripheral of fusion loops, making it probably unable to reach the membrane of host cells. In the context of full-length IgG, the other Fab arm might render extra steric hindrance, which further blocks the interactions between gB and cell membrane. This hypothesis is strongly supported by the observation that gB fusion loops directly interact with lipid bilayer captured by cryo-EM imaging [30]. In analogy to other class III viral fusion proteins, e.g. the glycoprotein of vesicular stomatitis virus (VSV GP), the domain I of gB might adopt similar fold before and after membrane fusion [31, 32]. Therefore, the 1H1 mAb could probably recognize gB in different conformations, including pre-fusion, post-fusion and the intermediates in between, which is possibly the reason contributing to its high neutralizing efficacy. In addition, these observations also indicate that domain I of gB could possibly serve as an ideal immunogen to elicit direct neutralizing antibodies against herpesviruses.
In summary, we combined both immunological and structural approaches to systematically characterize the envelope protein gB of an emerging highly pathogenic PRV variant. We identified two classes of neutralizing antibodies that effectively block PRV infection in vitro, which utilize different mechanisms with complement dependence or without respectively. These two classes of antibodies recognize gB with epitopes in two separate domains and thus indicate these domains as potential subunit vaccines to prevent PRV infections. These findings would intensify our understanding of the immunogenic properties of PRV glycoproteins and provide important guidance for antiviral drug design and vaccine development.
The protocol in this study was approved by the Committee on the Ethics of Animal Care and Use of National Research Center for Veterinary Medicine (Permit 20160313088). The study was conducted following the Guide for the Care and Use of Animals in Research of the People's Republic of China.
Pig kidney (PK-15) cells (CL33, obtained from ATCC) and African green monkey kidney (Vero) cells (GNO10, obtained from cell resource center of Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences) were cultured in Dulbecco's modified Eagle's medium (DMEM, Gibco) supplemented with 10% fetal bovine serum (Gibco) in a humidified chamber containing 5% CO2 at 37°C. PRV strain HN1201 was isolated in 2012 from an affected pig farm in China [5], and was propagated in Vero cells. PRV strains Bartha-K61, HB98, RA, SU, 188–5, 072–1, and BH1 were provided by the National Research Center for Veterinary Medicine, China. To make the current epidemic status and the major molecular changes of PRV clear, a phylogenetic analysis based on all PRV genomes available in the GenBank database was performed using the distance-based neighbor-joining method in MEGA4 software (S8 Fig). The Genbank accession numbers were included in the names of PRV strains.
Both proteins (HN1201 strain [5]) were expressed with the Bac-to-Bac baculovirus expression system (Invitrogen). The PRV gB construct covers nearly the full-length ectodomain, including residues A59 to N752, followed by a C-terminal 6×His tag. The PRV-gB-D_IV includes two continuous regions, residues A59 to S148 and N546 to T700, linked by a GGSG polypeptide, and a 6×His tag is fused at the C terminus. The GP67 signal sequence was engineered at the N-terminus of each protein to facilitate secretion during protein production. To produce recombinant proteins, Hi5 cells were infected with high-titer recombinant baculovirus and grown for 48 h. The cell culture supernatant was harvested by centrifugation applied to metal affinity chromatography with a HisTrap HP column (GE Healthcare). The eluted product was further purified by size-exclusion chromatography using a Superdex 200 16/600 pg column (GE Healthcare) equilibrated with a buffer containing 20 mM Tris–HCl and 150 mM NaCl (pH 8.0). The final product reached a purity of ~95% as shown by SDS-PAGE.
To generate gB-specific mAbs, purified recombinant PRV-gB (100 μg/mouse) was blended with Freund’s complete adjuvant and used to immunize 6-week-old female BALB/c mice. Booster immunizations were performed twice with 3-week intervals. Then the spleen cells were isolated and fused with SP2/0 myeloma cells. Hybridoma culture supernatants were screened for reactivity to purified PRV-gB by standard indirect ELISAs. Positive hybridoma clones were subcloned by limited dilution at least three times. The mAbs were initially captured from hybridoma cell culture supernatants by affinity chromatography with a Protein G HP column (GE Healthcare) and further purified by size-exclusion chromatography.
The coding sequences for 1H1 variable fragments (VH and VL) were fused to the sequence encoding constant regions of a mouse IgG1 (CH, CL and Fc) to generate chimeric IgG1 expression vector using the backbone of pCAGGS plasmid. Two mutant protein-expressing plasmids (1H1_NL and 1H1_SL) were constructed by site-directed mutagenesis to replace N103 or S105 in HCDR3 with alanine (Fig 3B). The plasmids were transiently transfected into human embryonic kidney 293T (HEK293T) cells for protein expression. After three to seven days post transfection, the supernatants containing secreted IgG1s were harvested and subjected to HiTrap ProteinG chromatography (GE Healthcare). Target proteins were eluted with 0.1 M glycine (pH 3.0) and further purified by size-exclusion chromatography using a Superdex 200 16/600 pg column (GE Healthcare). Fab fragments were generated by ficin digestion and purified using the pierce mouse IgG1 Fab preparation kit (Thermo Scientific) following the manufacturer’s instructions. The products were buffer-exchanged into a buffer containing 20 mM Tris-HCl (pH 8.0), 50 mM NaCl by an additional round of size-exclusion chromatography for crystallization.
Briefly, 96-well microtiter plates were coated with purified PRV-gB and PRV-gB-D_IV at 200 ng/well in a carbonate-bicarbonate coating buffer (pH 9.6) at 4°C overnight. Plates were blocked at 37°C for 1 h with PBST containing 5% skimmed milk. Antibodies were then added in the well and incubated for 1 h at 37°C. After three times of washing, the wells were incubated with goat anti-mouse IgG-HRP (Santa Cruz) for 40 min at 37°C. The plates were washed again for five times before the reaction substrate TMB was added. The reaction was conducted in dark at room temperature for 5 min and was stopped with 2 M H2SO4. The optical density at 450 nm (OD450) of each well was read using a microplate reader (Thermo). Statistical presentations were generated with GraphPad Prism 5 (San Diego, CA).
PK-15 cells were seeded in 96-well plates. Pseudorabies viruses (200 times of TCID50) were incubated with serial two-fold dilutions of the 1H1 mAb at 37°C for 1 h prior infecting cells. Then, the mixture was added to PK-15 monolayers in 96-well plates and incubated for 1 h. Each concentration was conducted with eight replicates. The supernatant was removed after incubation and replaced by fresh DMEM medium. The cells were cultivated for another 72 h at 37°C before analysis. The cellular pathology was directly observed using microscopy. All experiments were conducted in three independent trials. The half maximal inhibitory concentration (IC50) was measured to describe the neutralization titer of each antibody. To identify complement-dependent neutralization activities, the experiment was conducted following the same protocol as above except that fresh rabbit serum (working concentration: 5%) as an exogenous complement was added into the virus-antibody mixture before infecting the cells. The data was graphed using GraphPad Prism 5 for presentation (San Diego, CA).
The V gene sequences of each mAb clone were amplified as previously described [33]. Briefly, 106 hybridoma cells were collected by centrifugation. Total RNA was extracted using TRIzol regent (Takara) according to the manufacturer’s protocol. Reverse transcription and PCR amplification were performed using a set of primers [33]. PCR products were identified by agarose gel electrophoresis and purified using a commercial kit (Tiangen). The DNA fragments were cloned into the pMD 18-T vector (Takara) and sequenced individually.
The protein samples were concentrated to 10 mg/mL for crystallization using the sitting drop vapor diffusion method at 18°C. PRV gB was crystallized with a reservoir solution of 34% PEG200 and 0.1 M citric acid, pH 6.5. The crystals of PRV-gB-D_IV were obtained in reservoir solution containing 0.05 M calcium chloride dihydrate and 0.1 M MES, pH 6.0. The 1H1 Fab was crystallized with a reservoir solution of 0.2 M potassium sulfate and 20% PEG3350, pH 6.8. X-ray diffraction data was collected at the Shanghai Synchrotron Radiation Facility (SSRF) BL17U at a wavelength of 0.97915 Å [34]. The datasets were processed with HKL2000 software [35]. Structures were determined by the molecular replacement method using the Phaser program [36] in the CCP4 suite [37]. The PRV gB structure was solved using the HSV-1 gB structure (PDB ID: 2GUM) as the search model. The Fab structure (PDB ID: 1SY6) was used as the search input for 1H1 Fab structure determination. Initial restrained rigid-body refinement was performed using PHENIX [38], which was followed by manual rebuilding and adjustment in COOT [39]. Further refinement was performed using PHENIX [38]. The stereochemical qualities of the final models were assessed using MOLPROBITY [40]. All the data collection and refinement statistics are summarized in S1 Table.
The binding affinities of wild type 1H1 Fab or mutants to PRV gB were measured by BLI at room temperature (298K) with the Octet RED96 biosensor method (ForteBio, Inc.). The runnning buffer is composed of 20 mM Hepes (pH 7.4), 150 mM NaCl and 0.005% (vol/vol) Tween 20. Soluble gB was immobilized on an Ni-NTA-coated biosensor surface and then exposed to a series of analytes at different concentrations (6.25–100 nM for 1H1_WT, 62.5–1000 nM for 1H1_NL or 1H1_SL). Background subtraction was used to correct the errors of sensor drifting. The data was processed by the ForteBio’s data analysis software and plotted with Origin 8.0 program.
To prepare the PRV gB-1H1 Fab complex, soluble PRV gB and 1H1 Fab samples were mixed with a molar ratio of 1:1.5 and incubated at 4°C for 2 h. The mixture was then separated by size-exclusion chromatography using a Superose 6 10/300 GL column (GE Healthcare). The complex sample at a concentration of 0.02 mg/mL was applied to glow-discharged copper grids coated with continuous carbon films and stained with 2% uranyl acetate. The excessive stain liquor was blotted with a filter paper and let the grid to air-dry. The specimen was then loaded onto a Tecnai F20 transmission electron microscope (FEI) equipped with a field emission gun for data collection, which was operated at 200 kV acceleration voltage and with a defocus range of—(1–3) μm. Images were recorded with a 4k×4k BM-Eagle CCD camera with a calibrated pixel size of 1.36 Å.
A total of ~8000 particles were semi-automatically picked from 200 micrographs using e2boxer.py in EMAN2 [41] package. The contrast transfer function (CTF) parameters were estimated by e2ctffit.py [41] and applied to correct the images by phase-flipping [42] method. All the subsequent classification and reconstruction processes were conducted with relion-2.0 [43] using the phase-flipped particles without further CTF corrections. After several rounds of iterative 2D and 3D classifications, a stack of ~3000 particles was selected with 3 copies of 1H1 Fab bound to a gB trimer. The stack was subjected to 3D refinement with 3-fold symmetry applied, which resulted in a reconstruction of 35 Å resolution as determined by the gold-standard fourier shell correlation (FSC) 0.5 cut-off value (S6 Fig).
Though with low resolution, the reconstructed map clearly shows the feature of gB trimer in the center and three copies of 1H1 Fab density branching out at one end. We first fitted the crystal structure of PRV gB into the density map using CHIMERA [44], which showed a high degree of matching. The atomic structure of 1H1 Fab was further fitted into the remaining density using SITUS [45] with several rounds of orientation search. The special hinged structure of the Fab density allowed us to distinguish the relative orientation of heavy chain and light chain of Fab molecules despite the low resolution of the reconstruction. All the fitting processes were performed following the rigid-docking protocols without local adjustments. No obvious clash or close contact was observed in the final pseudo-atomic model, which was used for further structural analysis. All the EM density related figures were rendered using CHIMERA [44].
The gene sequence encoding the full length wild type PRV gB (amino acids 1–914) was cloned into the vector pEGFP-N1 to generate a gB expression vector with an EGFP tag fused at the C-terminus. A QuickChange site-directed mutagensis kit was used to obtain the mutants with the indicated mutations (Q206A, D210A, R214A, R215A or Mut4). Mut4 denotes the quadruple mutant with all four amino acids replaced by Alanine. Protein expression was verified by fluorescence microscopy.
The binding between the 1H1 mAb and gB/mutants was analyzed by flow cytometry. Briefly, Human Embryonic Kidney 293 cells with large T antigen (293T cells, obtained from cell resource center of Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences) were transfected with these plasmids above separately. After 24 h, the transfected cells were incubated with 1H1 mAb at room temperature for 30 min. The cells were then washed 3 times with 1×PBS to remove the unbound antibodies. Subsequently, the cells were further incubated with APC-linked goat anti-mouse IgG (minimal x-reactivity) (Biolegend, U.S.A) secondary antibody for 30 min at room temperature (avoiding light). Again, discard the liquid and wash 3 times with 1×PBS. Finally, the cells were loaded onto the flow cytometry (BD FACSCalibur) to detect the APC fluorescence signals.
|
10.1371/journal.pntd.0001783 | Re-Visiting Trichuris trichiura Intensity Thresholds Based on Anemia during Pregnancy | The intensity categories, or thresholds, currently used for Trichuris trichiura (ie. epg intensities of 1–999 (light); 1,000–9,999 epg (moderate), and ≥10,000 epg (heavy)) were developed in the 1980s, when there were little epidemiological data available on dose-response relationships. This study was undertaken to determine a threshold for T. trichiura-associated anemia in pregnant women and to describe the implications of this threshold in terms of the need for primary prevention and chemotherapeutic interventions.
In Iquitos, Peru, 935 pregnant women were tested for T. trichiura infection in their second trimester of pregnancy; were given daily iron supplements throughout their pregnancy; and had their blood hemoglobin levels measured in their third trimester of pregnancy. Women in the highest two T. trichiura intensity quintiles (601–1632 epg and ≥1633 epg) had significantly lower mean hemoglobin concentrations than the lowest quintile (0–24 epg). They also had a statistically significantly higher risk of anemia, with adjusted odds ratios of 1.67 (95% CI: 1.02, 2.62) and 1.73 (95% CI: 1.09, 2.74), respectively.
This analysis provides support for categorizing a T. trichiura infection ≥1,000 epg as ‘moderate’, as currently defined by the World Health Organization. Because this ‘moderate’ level of T. trichiura infection was found to be a significant risk factor for anemia in pregnant women, the intensity of Trichuris infection deemed to cause or aggravate anemia should no longer be restricted to the ‘heavy’ intensity category. It should now include both ‘heavy’ and ‘moderate’ intensities of Trichuris infection. Evidence-based deworming strategies targeting pregnant women or populations where anemia is of concern should be updated accordingly.
| Infection by the soil-transmitted helminth Trichuris trichiura is defined as ‘light’, ‘moderate’ and ‘heavy’ depending on its intensity. However, these intensity categories were developed in the 1980s, before any epidemiological data were available on the association between specific T. trichiura infection intensities and adverse health outcomes. Here, we re-analyzed data from a study of T. trichiura infection and anemia in pregnant women to determine the threshold (i.e. the lowest infection intensity) associated with an increased risk of anemia. Women with T. trichiura infections of intensities ranging from 601 to 1632 eggs per gram of feces (epg) (ie. a ‘moderate’ level of intensity) had a significantly higher prevalence of anemia and a significantly lower hemoglobin level than the reference group (i.e. women with T. trichiura infections of intensities ranging between 0 and 24 epg). This finding contrasts with the common belief that only ‘heavy’ T. trichiura infection (10,000 epg and above) can cause anemia.
| The most recent comprehensive estimation of the prevalences of the soil-transmitted helminthiases (STH) documents a global prevalence of 17% for Trichuris trichiura infection, with approximately 800 million persons infected at any one time [1], [2]. Community-wide prevalences are frequently over 30–40% and it is not uncommon to observe prevalences exceeding 80% in community sub-groups like school-age children and preschool-age children [3]–[7]. T. trichiura infections contribute to the STH-attributable burden of disease by adversely affecting the growth and cognitive development of children and the health and productivity of adults [8], [9]. Because of its co-occurrence with other infections, malnutrition and poverty, it also diminishes the economic potential, not only of the infected individual, but also of the family and community as well [10].
In 1987, an expert committee convened by the World Health organization (WHO) established infection intensity categories for STH, including T. trichiura, in order to inform the management of large-scale deworming programs [11]. T. trichiura infection was defined as light (1–999 epg) or heavy (>10,000 epg) [11]. These categories were based primarily on expert opinion and little dose-response data from the field, and were described as “arbitrary” by this committee. [11]. A further category of ‘moderate’ (i.e. for epg counts between 1,000 and 9,999 epg) was subsequently added by WHO [12]. The original 1987 report had also mentioned that anemia attributable to T. trichiura infection reflected a ‘very heavy worm burden’ [11].
Since then, the association between T. trichiura (prevalence and intensity) and hemoglobin (Hb) levels or anemia, has been assessed in several epidemiologic studies mostly conducted in Africa and in Asia and of which the majority found no significant association [13]–[18]. However, four studies conducted in the Americas (Jamaica, Panama, Mexico and Peru) reported statistically significant associations [19]–[22]. In addition, T. trichiura infection has been associated with a lower increase of Hb in iron-supplemented pregnant women [22]. Mechanisms by which T. trichiura infection may cause anemia include ingestion of blood by the parasite, blood loss from parasite-induced lesions in the intestinal mucosa, and inflammatory responses such as tumor necrosis factor α (TNFα) leading to decreased appetite; the relative contributions of these factors being unknown [9].
Anemia is a major public health problem because it impairs the growth and cognitive development in children and because severe anemia increases the risk of maternal mortality. Its worldwide prevalence is estimated at 48.8% [23]. The importance of the cluster of STH to the global risk of anemia is relatively well known, but among helminth species, T. trichiura has received much less attention than hookworms.
The objectives of this study were to determine a threshold for T. trichiura-associated anemia in pregnant women, and to describe the implications of this threshold in terms of the need for primary prevention and chemotherapeutic interventions.
Ethics approval was obtained for the original RCT from the following review committees: Research Institute of the McGill University Health Centre (Canada), The “Comite Institucional de Etica de la Universidad Peruana Cayetano Heredia” (Peru); and the “Comite Etica de la Direccion General de Salud de las personas del Ministerio de Salud de Peru” (Peru). The research procedures followed were in accordance with the ethical standards of these three ethics committees and with the Helsinki Declaration. Written informed consent was obtained from all women.
The data source for this study originated from a randomized controlled trial on mebendazole during pregnancy and its effect on birth weight which had been conducted in the highly STH-endemic Amazon area of Peru whose methods have been described elsewhere [24]. Briefly, 1,042 pregnant women were recruited in their second trimester and randomly assigned to receive either a single dose of 500 mg mebendazole or a placebo. Women in both groups received daily iron supplements throughout their pregnancy. At enrolment (second trimester) and again in the third trimester, blood and stool specimens were collected from participants for hemoglobin (Hb) ascertainment by HemoCue and for STH determination by the Kato-Katz method. There was no statistically significant difference between intervention groups in the prevalence of anemia or in mean hemoglobin levels in the third trimester. However, women having Trichuris trichiura infection in the second trimester were at a higher risk of anemia in their third trimester [22].
To determine a threshold for the effect of T. trichiura infection intensity on hemoglobin and anemia, the 935 mothers for whom complete information was available (i.e. on helminth infection and hemoglobin level in both the 2nd and 3rd trimester, plus covariates) were divided into quintiles based on T. trichiura infection intensity in the second trimester. Mean hemoglobin concentrations and anemia prevalence in the third trimester were calculated for each group. Mean hemoglobin concentrations in the third trimester of each T. trichiura quintile were compared to the lowest quintile using generalized linear model (GLM) analysis. The prevalence of anemia, defined as hemoglobin <11 g/dL [23], in the third trimester in each quintile was compared to that of the lowest quintile by logistic regression. Covariates found to be statistically significantly associated with the outcome were included in regression models: the model predicting hemoglobin levels included hookworm intensity and the model predicting anemia included hookworm intensity and the time interval between assessments for hemoglobin levels [22].
Among the 935 pregnant women included in the analysis, 82% were infected with Trichuris trichiura, and 43% were co-infected with T. trichiura and hookworms. The highest T. trichiura infection intensity was 25,200 epg. Participants' characteristics are described in more detail elsewhere [22].
Women in the lowest three T. trichiura intensity quintiles had similar hemoglobin concentrations, with arithmetic mean levels of 11.53, 11.55 and 11.58 g/dL, respectively. In contrast, the fourth and fifth quintiles had significantly lower mean hemoglobin concentrations than the reference group (i.e. 11.24 and 11.05 g/dL, respectively) (Table 1). The fourth and fifth quintiles also had a statistically significantly higher risk of anemia, with adjusted odds ratios of 1.67 (95% CI 1.02, 2.62) and 1.73 (95% CI 1.09, 2.74), respectively (Table 2).
The fact that a statistically significant association between T. trichiura infection and anemia was found in this study, but not in any other study of pregnant women, can be explained, in part, by the fact that this time the association between T. trichiura and anemia was determined in a population of women who had received daily iron supplements. Therefore, the fraction of anemia attributable to an insufficient dietary intake may have been reduced in the study population, resulting in an increased fraction attributable to T. trichiura. This likely strengthened the association between T. trichiura and anemia in our study population, a finding that may not have been easily observable in other populations.
The 601–1632 epg T. trichiura infection intensity category was the lowest epg category where a statistically significant association between hemoglobin and anemia was found. This indicates that the threshold for the T. trichiura effect on hemoglobin and the risk of anemia in iron-supplemented pregnant women appears to be somewhere between 601 and 1632 epg. In other words, iron-supplemented pregnant women with “light” or “moderate” T. trichiura infection intensities, based on the current classifications, may indeed be at an increased risk of morbidity from anemia as a result of the infection.
This finding has implications for STH control programs, in particular, those programs targeting pregnant women, because the efficacy of the commonly used deworming regimens of single-dose albendazole or mebendazole against T. trichiura is not optimal [25].
This analysis also provides support for categorizing a T. trichiura infection ≥1,000 epg as “moderate”, as currently defined by WHO. In addition, for pregnant populations, even if they are receiving iron supplements during pregnancy, it may be that 601 epg should be considered a lower limit for this ‘moderate’ category.
The most important implication of these analyses is that moderate T. trichiura infection in pregnant women is a significant risk factor for anemia which, in turn, increases the risk of adverse maternal and infant health outcomes. Therefore, in a pregnant population where there is a high prevalence of T. trichiura infection and where intensity levels exceed 600 epg, it may be that additional care options beyond the commonly used single-dose albendazole or mebendazole should be considered.
|
10.1371/journal.ppat.1006020 | Merkel Cell Polyomavirus Small T Antigen Promotes Pro-Glycolytic Metabolic Perturbations Required for Transformation | Merkel cell polyomavirus (MCPyV) is an etiological agent of Merkel cell carcinoma (MCC), a highly aggressive skin cancer. The MCPyV small tumor antigen (ST) is required for maintenance of MCC and can transform normal cells. To gain insight into cellular perturbations induced by MCPyV ST, we performed transcriptome analysis of normal human fibroblasts with inducible expression of ST. MCPyV ST dynamically alters the cellular transcriptome with increased levels of glycolytic genes, including the monocarboxylate lactate transporter SLC16A1 (MCT1). Extracellular flux analysis revealed increased lactate export reflecting elevated aerobic glycolysis in ST expressing cells. Inhibition of MCT1 activity suppressed the growth of MCC cell lines and impaired MCPyV-dependent transformation of IMR90 cells. Both NF-κB and MYC have been shown to regulate MCT1 expression. While MYC was required for MCT1 induction, MCPyV-induced MCT1 levels decreased following knockdown of the NF-κB subunit RelA, supporting a synergistic activity between MCPyV and MYC in regulating MCT1 levels. Several MCC lines had high levels of MYCL and MYCN but not MYC. Increased levels of MYCL was more effective than MYC or MYCN in increasing extracellular acidification in MCC cells. Our results demonstrate the effects of MCPyV ST on the cellular transcriptome and reveal that transformation is dependent, at least in part, on elevated aerobic glycolysis.
| In 2008, Merkel cell polyomavirus (MCPyV) was identified as clonally integrated in a majority of Merkel cell carcinomas (MCC), a rare but highly aggressive neuroendocrine carcinoma of the skin. Since then, studies have highlighted the roles of the MCPyV T antigens in promoting and sustaining MCC oncogenesis. In particular, MCPyV small T antigen (ST) has oncogenic activity in vivo and in vitro. We performed transcriptome analysis of normal human fibroblasts with inducible expression of MCPyV ST and observed significant alterations in levels of metabolic pathway genes, particularly those involved in glycolysis. MCT1, a major monocarboxylate transporter, was rapidly induced following ST expression and inhibition of MCT1 activity reduced the ST growth promoting and transforming activities. The metabolic perturbations induced by this oncogenic human polyomavirus reflect a potent transforming mechanism of MCPyV ST.
| Human polyomaviruses are a diverse family of small DNA tumor viruses that typically cause asymptomatic, lifelong infections in healthy individuals [1, 2]. However, immune deficiencies enable more severe polyomavirus induced diseases including Merkel cell carcinoma (MCC). MCC is a rare and aggressive skin cancer that primarily affects the elderly and immunocompromised [3, 4]. Transcriptome sequencing of MCC led to the discovery of Merkel cell polyomavirus (MCPyV) and demonstration that viral DNA was clonally integrated in approximately 80% of MCC tumors [5]. The integrated MCPyV early-region (ER) expresses wild-type small T antigen (ST) and a truncated form of large T antigen (LT). The truncated LT retains the LXCXE motif that binds the retinoblastoma protein (RB1) but is unable to support viral replication due to loss of the DNA binding and helicase domains [6, 7]. In some MCC tumors, ST can be detected in the absence of LT, suggesting that ST is required for tumorigenesis [7].
The precise mechanisms for how MCPyV ST promotes cellular transformation are still unresolved. The contribution of the related polyomavirus simian virus 40 (SV40) ST to transformation has been shown to be dependent on its ability to bind and inhibit protein phosphatase 2A (PP2A) activity that, in turn, perturbs a wide range of signaling pathways [8, 9]. However, PP2A binding by MCPyV ST may not be necessary for transformation [7]. Compared to ST from other polyomaviruses, MCPyV ST has unique properties including inhibition of the E3 ubiquitin ligase FBXW7 and an ability to increase cap-dependent translation through hyperphosphorylation of 4E-BP1 [10–12], activities which likely contribute to the initiation and maintenance of an oncogenic state.
Under normal physiological conditions, cells can convert one glucose molecule into two pyruvate molecules followed by pyruvate oxidation in mitochondria resulting in the synthesis of 38 ATP molecules per molecule of glucose [13]. In hypoxic conditions, oxidative phosphorylation is inhibited and anaerobic glycolysis is activated, leading to the production of only 2 ATP molecules and secretion of lactate into the extracellular space [14]. Cancer cells may convert pyruvate to lactate under normoxic conditions resulting in aerobic glycolysis, known as the Warburg effect. Increased aerobic glycolysis has been recognized as a hallmark of cancer due to the requirement for large quantities of biosynthetic intermediates to be generated for sustained tumorigenesis [15, 16]. Excess lactate production increases the acidity of the tumor cell microenvironment that favors two additional hallmarks, tumor cell invasion and metastasis [17].
Viruses typically do not express their own metabolic enzymes and instead rely on manipulation of host signaling and metabolic pathways to establish productive infections [18]. For example, RNA viruses such as hepatitis C and dengue virus promote efficient replication by altering host lipid metabolism [19]. Adenovirus can perturb the function of the transcription factor MYC to promote glucose and glutamine metabolism [20], and Kaposi’s Sarcoma-associated Herpesvirus (KSHV) has also been shown to induce MYC expression leading to increased glutaminolysis [21].
To gain insight into the impact of MCPyV ST on gene expression, we generated normal human cells with inducible expression of MCPyV ST or GFP and performed transcriptome profiling. The resulting analysis revealed that ST significantly altered metabolism-related pathways with upregulation of glycolysis and metabolite transporter genes. Correspondingly, we found that increased ST levels led to increased aerobic glycolysis. Furthermore, inhibition of the major monocarboxylate transporter MCT1 suppressed ST induced cellular proliferation, transformation and MCC viability.
Given the unique properties of MCPyV ST compared to ST from other human polyomaviruses [1, 11], we sought to obtain a global view of MCPyV ST induced transcriptional perturbations in normal human cells. IMR90 human diploid lung fibroblasts were selected because of their wide use in transcriptome and genomic analyses, including viral oncoprotein perturbations [22]. Cells were transduced with lentiviral vectors capable of doxycycline (dox) inducible MCPyV ST or GFP expression. Cells were treated with dox for 96 hours and harvested every 8 hours followed by processing for RNAseq in triplicate (Fig 1A). ST transcript levels were rapidly induced with peak levels between 24 to 40 hours after dox addition, followed by a gradual decrease in levels (Fig 1B). ST protein levels followed the transcript profile, peaking between 48 to 64 hours before declining (Fig 1C). Similarly, GFP protein levels were detectable within 8 hours and decreased by 64 hours after dox addition.
Genes that were most differentially expressed between ST and GFP were ranked by LIMMA (see Methods) with 2854 genes passing p-value and fold-change cutoffs [23]. Model-based clustering grouped the differentially expressed genes into 50 clusters (Fig 1D and S1 Fig and S1 Table). Clusters were evaluated for significant enrichment in biological processes, including Gene Ontology (GO) terms (S2 Table) and the Cancer Hallmark gene sets in the Molecular Signatures Database (MSigDB) (S3 Table) [24]. ST expressing cells were significantly enriched for gene clusters with the GO terms for energy coupled proton transport, isoprenoid and L-serine biosynthetic processes, and glutamine, lysine and arginine transport. ST expressing cells were also significantly enriched for the Cancer Hallmarks including epithelial to mesenchymal transition (EMT), TNFA signaling via NF-κB, hypoxia, mTORC1, oxidative phosphorylation, glycolysis, MYC, and cell cycle including E2F targets, G2/M checkpoint and mitotic spindle.
Elevated aerobic glycolysis is a hallmark present in many cancers and represents a potential vulnerability for targeting cancer cell proliferation [15, 16, 25]. Expression profiling of MCPyV ST cells revealed a high prevalence of significantly altered metabolism-related genes, specifically those involved in glycolysis such as Hexokinase 2 (HK2) in cluster 6 (Fig 2A) and metabolite transport. Transport of metabolites is mediated in large part by members of the SLC gene family [26]. A large proportion of all SLC transporter genes were significantly upregulated following ST expression (S2 Fig). A consequence of increased aerobic glycolysis is elevated acidification of the surrounding microenvironment. Notably, expression of SLC16A1 (MCT1), the major monocarboxylate transporter for lactate and pyruvate [27, 28], was significantly increased after ST induction (Fig 2B and 2C). Levels of MCT1 were increased in IMR90 cells expressing ST compared to GFP following dox addition.
Activation of glycolysis is normally accompanied by an increase in the rate of glucose import. Consistent with this, we found that MCPyV ST increases the expression of two glucose transporters, GLUT1 (SLC2A1) and GLUT3 (SLC2A3). We validated the increase in the expression of GLUT1 in ST expressing using RT-qPCR (Fig 2E). Among the larger family of hexose transporters (SLC2A1-14) we also observed upregulation of SLC2A8, 13, and 14 (Fig 2F). Additionally, we found that ST cells have higher expression of the carbohydrate response element binding proteins (ChREBPs) MLX and MLXIP, which can bind and activate the promoters of genes encoding glycolytic enzymes, thus increasing the rate of glycolysis (S3A Fig).
Glycolysis is a multistep biochemical process. HK2 serves as an upstream regulator that irreversibly commits glucose to enter the pathway. A byproduct of aerobic glycolysis is lactate. Production of lactate from pyruvate is mediated by lactate dehydrogenase (LDH), comprised of homo- or heterotetramers of two subunits, LDHA and LDHB. A major function of MCTs, including MCT1, is to prevent the toxic buildup of lactate in the intracellular milieu by co-exporting lactate together with protons out of the cell [27]. We observed that levels of glucose were depleted from the media and lactate levels increased at a significantly higher rate following ST expression compared to GFP cells (Fig 3A).
Given the significant increase in lactate levels in the media as early as 2 days after ST induction, we measured the extracellular acidification rate (ECAR) of IMR90 cells inducibly expressing ST before and after dox addition for 48 hours. We found a significant increase in the ECAR of these cells following ST induction, consistent with an increase in aerobic glycolysis (Fig 3B and 3C). Inhibition of ATP synthase by oligomycin treatment led to an increase in the glycolytic rate of cells in response to the lack of ATP production from oxidative phosphorylation (Fig 3C) [29]. As expected, we found the ECAR to be significantly decreased in both ST and GFP expressing cells treated with the MCT inhibitor α-cyano-4-hydroxycinnamate (CHC) (Fig 3C).
There was no significant difference in the oxygen consumption rate (OCR) of ST cells compared to GFP cells (Fig 3D). This suggests that the level of oxidative phosphorylation is maintained in ST cells despite the increased rate of glucose being converted to lactate. ST cells may use alternative carbon sources, like glutamine, to fuel the TCA cycle. Consistent with this hypothesis, we found that ST cells upregulate the expression of the glutamine transporter SLC1A5, as well as the enzymes glutaminase (GLS) and glutamate dehydrogenase (GLUD1), which are necessary to convert glutamine to the TCA cycle intermediate α-ketoglutarate through the metabolic pathway called glutaminolysis (S3B Fig).
To determine the importance of MCT1 activity on the fitness of ST expressing cells, we measured the growth of IMR90 cells expressing either ST or GFP following CHC or DMSO treatment (Fig 3E). CHC significantly suppressed the growth rate of ST expressing cells, while GFP cells were largely unaffected.
We assessed the glycolytic pathway in three MCPyV-positive MCC cell lines. ECAR measurements revealed that MKL-1 and MKL-2 cells had similar rates of glycolysis, while WaGa cells had a lower glycolytic rate (Fig 4A). Conversely, following oligomycin treatment, MKL-1 and MKL-2 cells increased their ECAR levels higher than WaGa cells consistent with their higher level of glycolysis. Inhibition of MCT1 in a highly glycolytic cell can lead to intracellular acidification through the accumulation of monocarboxylates and protons. MCT1 inhibition has previously been shown to be toxic to certain tumors with high MCT1 expression [27, 30]. A number of MCT1 inhibitors are current in clinical trials for treating advanced solid tumors, with promising results in cancers with elevated MCT1 expression [31, 32]. SR13800 and SR13801 are pyrole pyrimidine-based molecules with high specificity for MCT1 [33]. We tested the effects of the MCT1 inhibitors CHC, SR13800 and SR13801 on the viability of MKL-1, MKL-2 and WaGa MCC cell lines over 7 days (Fig 4B, 4C and 4D, respectively). All three MCC cell lines showed high sensitivity to CHC treatment, while only MKL-1 and MKL-2 cells were affected by SR13800 and SR13801. MKL-1 and MKL-2 cells were more dependent than WaGa cells on MCT1 activity and glycolysis for continued proliferation reflecting their higher ECAR levels.
The expression of HK2, LDH, MCT1 and other glycolytic genes are regulated, at least in part, by MYC [25, 34]. Given the pronounced effects of ST on glycolysis, we sought to determine how sensitive MCPyV-positive MCC cell lines were to perturbations in this pathway. Since the MCC cell lines MKL1, MKL-2, WaGa and BroLi were previously uncharacterized with regards to MYC, we assessed the levels of the different MYC isoforms in these cells. By immunoblotting, these MCC cell lines had no detectable MYC, but did have detectable levels of MYCN and MYCL (S4A Fig).
We next determined the ability of each MYC family member to regulate glycolytic gene expression and aerobic glycolysis in MCC cell lines. MKL-1 and WaGa cells were transduced with dox inducible vectors expressing MYC, MYCN or MYCL. Following selection, cells were treated with dox for 72 hours, and then lysates were harvested for immunoblotting (Fig 5A). We observed that MYC and MYCN but not MYCL led to increased levels of HK2 and LDHA in MKL-1 and WaGa cells. MYC and MYCL led to increased levels of MCT1 while MYCN led to decreased MCT1 levels in WaGa cells.
Given how the various MYC isoforms differentially affected HK2, LDHA and MCT1 levels, we compared the effects of MYC expression in MKL-1 cells on aerobic glycolysis by measuring the ECAR following 72 hours of dox treatment (Fig 5B). GFP and MYCN expressing cells had similar basal ECAR, while MYC and MYCL expressing cells had higher ECAR levels. Notably, the MYCL expressing MKL-1 cells had significantly higher ECAR following oligomycin treatment, indicating that MYCL was more effective in facilitating the glycolytic capacity of these cells compared to the other MYC isoforms.
We sought to identify specific signaling pathways that ST utilized to increase glycolytic and MCT1 gene expression and if these pathways contributed to cellular transformation. Given the significant enrichment of genes from both the MYC and NF-κB Cancer Hallmark gene sets across several ST induced gene clusters (Fig 1D), we examined the MCT1 promoter region [-1000, +100] for MYC and NF-κB binding sites (Fig 6A). Chromatin accessibility patterns in the parental IMR90 cells were assessed in the ENCODE DNase I hypersensitivity experiments (GEO: GSM468801, GSM530665, GSM530666, GSM468792). A MYC binding site was located within a relatively open region in the MCT1 promoter, while a NF-κB binding site was located in an area of reduced hypersensitivity.
To investigate the regulation of glycolysis gene expression in the context of a MCPyV-transformed cell line, IMR90 cells were serially transduced with retroviral constructs expressing a dominant-negative form of p53 (p53DD), telomerase reverse transcriptase (hTERT), and a tumor-derived form of MCPyV early-region (ER) that expresses ST plus truncated LT to generate p53DD-hTERT-ER (PHE) cells. Unlike SV40 large T antigen (LT), MCPyV LT cannot bind and inhibit p53; therefore p53DD is required to bypass senescence and apoptotic checkpoints in these IMR90 cells [35, 36].
To determine if MYC family proteins could cooperate with MCPyV IMR90 cells, we treated cells stably expressing ST, GFP, or p53DD + hTERT (PH) and MCPyV tumor-derived early-region (PHE) with inducible expression of MYC, MYCN or MYCL with dox for 48 hours and immunoblotted for HK2, MCT1 and LDHA (Fig 6B). ST alone could induce MCT1 levels and both PH and PHE cells had higher levels of MCT1 expression than ST alone. Consistent with the effects seen in MCC cell lines (Fig 5A), MYC and MYCN induction but not MYCL led to increased levels of HK2 and MCT1 in PH and PHE cells. The presence of MCPyV ER did not appear to affect induction of HK2 or MCT1 by MYC or MYCN. In contrast, PHE cells with MYCL had very high levels of MCT1 protein in the uninduced state, likely due to leakiness of the vector (S4B and S4C Fig). These results indicate that MYCL but not MYC or MYCN was able to cooperate with MCPyV ER to induce MCT1.
In addition to MYC, NF-κB is a prominent inducer of metabolic and growth-promoting genes and has been shown to independently regulate MCT1 [33, 37, 38]. RelA, also known as p65, is a key subunit of canonical NF-κB signaling and forms a homo- or heterodimer with other NF-κB subunits to activate target genes including IκBα [39]. To assess the role of NF-κB in the regulation of MCT1 in the context of MCPyV, IMR90 PH and PHE cells with inducible MYC or MYCL were transfected with siRNA targeting RelA (siRelA) or a non-targeting control (siCtrl). Cells were re-fed with or without dox-containing media 24 hours after transfection, then lysed 48 hours later for immunoblotting (Fig 6C). Following siRelA but not siCtrl treatment, reduced levels of RelA and the downstream target IκBα were observed in both MYC and MYCL expressing PH and PHE lines. Knockdown of RelA did not affect the ability of MYC to induce HK2 in the PH and PHE cells. In contrast, MCT1 levels were reduced following knockdown of RelA across all conditions, indicating that canonical NF-κB signaling likely has a complementary role in regulating MCT1 levels.
We assessed whether MCPyV-mediated transformation could be attenuated by MCT1 inhibition. We found that overexpression of MYCL in PHE (PHEL) IMR90 cells led to robust IMR90 anchorage-independent growth in soft agar, while PH, PHL and PHE cells lacked significant colony formation (Fig 7A). We chose MYCL in this context as it was previously shown that MYCL is amplified in MCC tumors, and may therefore have oncogenic potential in the presence of MCPyV [40]. We measured basal ECAR of IMR90 cells stably expressing p53DD, PH, PHE and PHEL (Fig 7B) and found that PH, PHE and PHEL cells had significantly higher ECAR than p53DD cells, with PHE cells maintaining the highest rate while PHEL cells had a significantly lower ECAR than PHE cells.
To determine if MCT1 activity was necessary to support cellular transformation of IMR90 cells by MCPyV, we assessed anchorage-independent growth by culturing PHE and PHEL cells in soft agar in the presence of the MCT1 inhibitors or DMSO. Proliferation of PHEL cells was highly attenuated by CHC treatment, while SR13800 and SR13801 inhibitors had modest but significant effects on growth compared to DMSO treatment (Fig 7C). PHEL cell growth in soft agar was significantly inhibited by CHC, SR13800 and SR13801 compared to DMSO treated cells (Fig 7D). These results indicate that MCT1 activity was required for transformation of IMR90 cells by MCPyV and MYCL.
Metabolic perturbations represent a key hallmark in many cancers, as the energetic and biosynthetic demands of tumor cells increase to sustain proliferation. MCPyV ST is a potent oncoprotein with transforming potential in vitro and in vivo and contributes to MCC. By performing temporal transcriptional profiling and metabolic analysis of ST expressing cells, we determined that ST significantly increases aerobic glycolysis and that inhibition of this pathway can suppress MCPyV-induced transformation as well as MCC growth. Cancers with viral etiology are particularly likely to undergo metabolic alterations due to the fundamental need for viruses to create a pro-replicative environment. Many viruses, including adenovirus, hepatitis C virus and HIV, induce aerobic glycolysis in infected cells to support viral replication [18]. Our results indicate that MCPyV ST can specifically alter the metabolic state of a cell.
We designed a time-series RNA-sequencing experiment to characterize the dynamics of gene expression in cells after expression of MCPyV ST. Comparing with statistically distinct behavior in the ST-expressing cells relative to GFP-expressing cells, we found that most of the differential expression trends appeared already at 16 hours post-induction, with down-regulated genes first reaching a minimum at around 32 hours and up-regulated genes building more gradually to peak at the 48 hour mark. Most genes exhibited only down- or up-regulation throughout the time course of 96 hours. We grouped differentially expressed genes into clusters to build a global picture of how ST remodels the transcriptional landscape. Among the 50 resulting clusters and their GO term and pathway enrichment, we observed a strong signature of metabolism-related changes (Fig 1D and S1 Fig). Many of the up-regulated clusters were enriched for the glycolysis pathway, rRNA processing, amino acid transport and response to glucose starvation. Among down-regulated clusters, there was enrichment in fatty acid oxidation, purine and pyrimidine metabolic processes, lipid metabolism, and mitochondrial respiration and ATP synthesis genes. The transcriptional signature of ST-expressing cells exhibited many of the characteristics associated with activation of aerobic glycolysis. In particular, we found that ST upregulated glucose import, lactate export and ChREBPs, transcription factors that specifically activate glycolytic enzymes. In addition, we found evidence that ST cells maintain normal levels of oxidative phosphorylation through anaplerosis, through increased levels of glutamine transporter and GLS and GLUD1, critical for glutaminolysis.
MCPyV ST also induced changes in many genes that were not annotated to be involved in metabolic processes. There were 14 out of the 50 clusters that showed enrichment for GO terms not involved in metabolism (S2 Table). The up-regulated genes were enriched for the mitotic cell cycle (clusters 46, 12, 31, 39), SMAD and BMP pathways (clusters 23 and 15), vascular permeability (3), keratinocyte migration (9) and pinocytosis (14). The down-regulated genes were involved in synapse assembly (19), SMAD protein import into the nucleus (43), extracellular matrix organization (38), cell adhesion (24), and negative regulation of viral genome replication (40). Therefore, ST appears to have a major impact on genes in metabolic and cell cycle pathways as well as several additional transcriptional programs, including the SMAD/BMP differentiation pathway and cell adhesion.
Given the functional enrichment for glycolysis-related genes in ST expressing cells (Fig 1D and S3A Fig and S3 Table), we focused on characterizing the role of MCT1 (SLC16A1), a major monocarboxylate transporter that most closely followed the ST expression pattern. Genetic and pharmacological inhibition of MCT1 has been an effective strategy for targeting highly glycolytic tumors, inhibiting tumor growth through a combination of effects including accumulation of intracellular lactate, altering the production of glycolytic intermediates, reducing glucose transport and ATP levels, and reducing glutathione levels [33, 37, 41–43]. Cancer cells with elevated MCT1 expression have also been found to be exquisitely sensitive to the glycolysis inhibitor 3-bromopyruvate [44].
Through extracellular flux experiments, we found a significant increase in the ECAR of ST expressing IMR90 cells (Fig 3C), accompanied by increased sensitivity to MCT1 inhibition (Fig 3E). We observed that MCPyV-transformed IMR90 cells exhibited significantly higher ECAR levels compared to untransformed IMR90 cells (Fig 7B). Furthermore, MCPyV-transformed IMR90 cells (Fig 7D) and MCPyV-positive MCC cell lines (Fig 4) were sensitive to MCT1 inhibition. The sensitivity of these MCC cell lines to MCT1 inhibition corresponded to their relative ECAR levels. These results indicate that ST, together with LT, can manipulate cellular energy states to meet the demands of tumorigenesis.
We have demonstrated that ST plays a significant role in altering the metabolic state of the host cell, but our results do not rule out the possibility that further modulation by LT is required for its effect on transformation. The presence of MCPyV LT in the transformed cell experiments using PHE and PHEL cells could have an effect on the ability of ST to induce glycolysis genes. Previous transcriptional profiling of LT did not indicate any significant perturbation of the metabolic pathways being investigated here [22]. As MCPyV LT can still bind pRB, there could be perturbations of metabolic genes that cooperate with ST [45], as seen in PHE cells where HK2 and LDHA levels were increased compared to PH and ST-only cells (Fig 6B).
The regulation of MCT1 expression has been proposed to involve numerous factors, including MYC [33] and NF-κB [37]. Intriguingly, one study found that a significant number of MCC tumors contained genomic amplification of MYCL [40], a close relative of MYC that is also amplified in small cell lung cancer [46]. MYCN, another MYC isoform, is amplified in pediatric neuroblastoma and has been shown to regulate metabolic pathways in a similar way as MYC [47]. Given these findings, we investigated whether MYC, MYCN and MYCL could cooperate with MCPyV to regulate glycolysis gene expression.
By generating a series of dox inducible constructs expressing the MYC, MYCN and MYCL isoforms in both MCC cells and IMR90 lines, we found that MYC and MYCN could robustly affect the expression of MCT1 and the critical glycolysis enzyme HK2 (Figs 5A and 6B). We found that unlike IMR90 cells, MCC cell lines lacked endogenous MYC expression but did express MYCL and MYCN (S4A Fig). MYCL overexpression led to increased levels of MCT1 in MCC cells, but decreased levels in IMR90 PHE cells. This suggests that the observed lack of MYC expression in these MCC cells may alter the transcriptional activity of MYCL.
MCPyV ER expression independently upregulated LDHA expression in IMR90 cells in a manner not dependent on MYC signaling, whereas MYC affected LDHA expression in MKL-1 cells, suggesting that metabolic regulation by MCPyV may involve cell type-specific factors. Interestingly, PHEL cells had significantly lower ECAR than PHE cells (Fig 7B), although PHEL cells are fully transformed and are sensitive to specific MCT1 inhibitors (Fig 7C and 7D). This may be due to the observation that expression of MYCL in PHE cells decreased MCT1 expression (Fig 6B and 6C). Taken together, these results suggest that MYC and MYCN behave quite differently from MYCL, with MYCL appearing to have a particular synergy with MCPyV ST that influences both gene expression and transformation.
Soft agar experiments, besides testing for anchorage-independent growth, also place cells in an environment that likely has a lower diffusion rate for extracellular metabolites compared to the typical environment encountered on a plastic dish containing liquid growth media. While the effects of SR13800 and SR13801 are modest in the standard cell culture plate setting used in Fig 7C, the significant decrease in transformation shown in Fig 7D by these inhibitors is likely due to the fact that as a colony of cells grows in size in soft agar, it is forced to become more glycolytic as hypoxia increases. This increased glycolytic load, compounded with the inhibition of MCT1, may lead to toxic intracellular acidification that results in the significant colony formation defect.
Previous work from our lab and others has suggested that MCPyV ST has specific effects on NF-κB signaling [22, 48]. We found that MCT1 expression could be efficiently suppressed through RNAi knockdown of the canonical NF-κB subunit RelA in IMR90 PH and PHE cells (Fig 6C). Overexpression of MYC could still induce MCT1 in PHE cells following RelA depletion, although to a lower degree, suggesting that MYC is the primary regulator of MCT1 expression in these cells with NF-κB potentially having a supplemental role. These results agree with our promoter analysis (Fig 6A) and earlier studies indicating that MCT1 is regulated by multiple transcription factors [33, 37, 38]. Other transcription factors for genes encoding glycolytic enzyme, PPP enzymes and glucose transporters, such as the carbohydrate responsive element binding proteins (CHREBPs) [49], would be interesting candidates to pursue in future studies. We focused on the MYC family in this study as they are a major driver of glycolysis gene expression, have been implicated in the regulation of MCT1 and in particular, MYCL has been found amplified in MCC.
MCT1 primarily mediates intracellular or extracellular acidification depending on the cell and tumor type [27, 50]. Our data indicates that MCT1 inhibition in MCPyV-expressing IMR90 fibroblasts dramatically alters extracellular acidification. The effect of CHC compared to the more specific MCT1/2 inhibitors suggests that broad MCT inhibition may be more potent across MCC cell lines. This is supported by observations that MCT4 can also export lactate in a redundant fashion [51]. WaGa cells may have higher levels of MCT4 that could contribute to monocarboxylate transport and thereby limit viability defects from specific MCT1 inhibition, or could uniquely utilize lactate for energy production.
Our results here represent the first temporal transcriptome analysis of a DNA tumor virus protein, leading to the identification of key metabolic perturbations that contribute to the proliferative effects of MCPyV ST. We have also shown the differential effects of MYC, MYCN, MYCL and NF-κB on aerobic glycolysis and MCT1 regulation. Recent genetic analysis of patients with severe ketoacidosis identified several inactivating mutations in MCT1, correlating with disease severity, MCT1 protein levels and transport capacity, highlighting yet another critical role of this transporter in human health [52]. The key role of MCT1 in MCC viability should be considered in future treatment regimens, perhaps in combination with metformin or other metabolic agents that have previously shown promise when combined with MCT1 inhibition [51].
293T and IMR90 cells were obtained from ATCC. The MKL-1 and MKL-2 cell lines were kind gifts from Masahiro Shuda and Yuan Chang (University of Pittsburgh, PA), BroLi cells from Roland Houben (University of Wuerzburg, Germany) and WaGa cells were from Jürgen Becker (Medical University Graz, Austria). 293T cells were cultured in Dulbecco’s modified Eagle medium (DMEM) (Cellgro) supplemented with 1% Pen Strep (GIBCO), 1% Glutamax (GIBCO), and 10% fetal bovine serum (FBS) (Sigma). IMR90 cells were cultured with a similar media composition as the 293T cells with the exception of 15% FBS and addition of 1% non-essential amino acids (GIBCO). MCC cell lines were cultured in RPMI 1640 media (GIBCO) supplemented with 1% Pen Strep, 1% Glutamax, and 10% FBS.
Packaging and envelope plasmids were co-transfected with lentiviral or retroviral expression vectors into 293T cells using Lipofectamine 2000 (Life Technologies). Two days after transfection, 293T cell supernatant was clarified with a 0.45 μm filter and supplemented with 4 μg/mL polybrene (Santa Cruz) before transducing recipient cells. Stable cell lines were generated after selection with 2 μg/mL puromycin (Sigma), 5 μg/mL blasticidin (Invivogen), 500 μg/mL neomycin (Sigma) and 50 μg/mL hygromycin (Santa Cruz) as required by each vector. For inducible cell line experiments, doxycycline (Clontech) was used at 1 μg/mL. For MCT1 transport inhibitor experiments, dimethyl sulfoxide (DMSO) (Sigma), α-cyano-4-hydroxycinnamate (CHC) (Sigma) (5 mM), SR13800 (Calbiochem) (100nM), and SR13801 (Tocris) (100 nM) were used at the indicated concentrations.
MCPyV ST, MYC-T58A, MYCN, MYCL and GFP Gateway-compatible cDNA entry clones were transferred from pDONR221 donor vectors to the pLIX_402 doxycycline inducible lentiviral Gateway destination vector (a gift from David Root; Addgene plasmid # 41394) via Gateway cloning (Life Technologies). pBabe-neo-p53DD was a gift from William C. Hahn (Dana-Farber Cancer Institute). pBabe-hygro-hTERT was a gift from Bob Weinberg (Addgene plasmid # 1773) [53]. Tumor-derived (MCCL21) MCPyV ER cDNA was generated as previously described [36] and cloned into the pLenti CMV Blast DEST (706–1) vector (a gift from Eric Campeau; Addgene plasmid # 17451) [54]. Lentiviral packaging plasmid psPAX2 and envelope plasmid pMD2.G were gifts from Didier Trono (Addgene #12260, #12259). Retroviral packaging plasmid pUMVC3 was a gift from Bob Weinberg (Addgene # 8449) [55] and envelope plasmid pHCMV-AmphoEnv from Miguel Sena-Esteves (Addgene # 15799) [56].
Dox inducible IMR90 lines expressing ST or GFP were seeded in 6 cm dishes 24 hours before initiation of time course with dox-containing DMEM. Cells were harvested every 8 hours for 96 hours and total RNA was purified using RNeasy Plus Mini Kit (Qiagen) and mRNA was isolated with NEBNext Poly(A) mRNA Magnetic Isolation Module (New England BioLabs). IMR90 cells were refed after 48 hours with dox-containing media. Sequencing libraries were prepared with NEBNext mRNA library Prep Master Mix Set for Illumina (New England BioLabs), passed Qubit, Bioanalyzer and qPCR QC analyses and sequenced on HiSeq 2000 system (Illumina). The complete set of RNAseq data can be accessed from the Gene Expression Omnibus (GEO) repository GSE79968.
The following antibodies were used: MCPyV Ab5 [36, 57]; GFP (D5.1, Cell Signaling); vinculin (H-10, Santa Cruz); actin (D6A8, Cell Signaling); HK2 (C64G5, Cell Signaling); LDHA (EP1565Y, Abcam); MCT1 (A1512, NeoBiolab); LAT1 (5347, Cell Signaling); RelA (D14E12, Cell Signaling); IκBα (L35A5, Cell Signaling); MYC (9E10, Santa Cruz); MYCN (9405, Cell Signaling); MYCL (AF4050, R&D Systems).
Whole cell lysates were prepared using RIPA buffer (Boston BioProducts) supplemented with protease inhibitor cocktail set I (Calbiochem) and phosphatase inhibitor cocktail set I (Calbiochem). Clarified protein extracts were boiled in SDS sample buffer (Boston BioProducts), resolved by SDS-PAGE (Criterion TGX precast gels; Bio-Rad), transferred to nitrocellulose membranes (Bio-Rad), blocked and incubated with the appropriate primary antibody in TBS-T overnight at 4°C. Detection of proteins was performed with horseradish peroxidase-conjugated secondary antibodies (Rockland), developed using Clarity Western ECL substrate (Bio-Rad), and imaged with a G:BOX Chemi detection system (Syngene).
Total RNA was purified using RNeasy Plus Mini Kit (Qiagen). cDNA was synthesized from the RNA using a High-Capacity cDNA Reverse Transcription kit (Thermo Fisher). qPCR was performed using Brilliant III SYBR Master Mix (Agilent Genomics) following the manufacturer’s instructions. RPLP0 was used as an internal loading control to normalize RNA levels.
IMR90 PH and PHE cells were seeded in 6 cm dishes (5 x 105 cells/dish) and were transfected with 100 nM siRNA ON-TARGETplus SMARTpool siRNA against human RelA (L-003533-00-0005) or ON-TARGETplus Non-Targeting siRNA#1 (D0018100105) from GE Healthcare Dharmacon using Lipofectamine RNAiMAX (Life Technologies). After 24 hours, cells were refed with media with or without dox. Cells were then harvested for subsequent immunoblotting after 48 hours.
Reads were mapped to a transcriptome index generated from the hg19 human reference genome and the Merkel cell polyomavirus sequence, using Tophat 2.0.4 and Bowtie1 with default parameters. Novel junctions were not allowed. MCPyV-aligned reads were transformed into gene-level counts using HTSeq. Human-aligned reads were quantified at the gene level using Cufflinks 1.3.0 along with its ancillary algorithms that correct for biases and multi-mapped reads. Log-transformed FPKM (fragments per kilobase of transcript per million mapped reads) values were input into further statistical analysis of human transcript levels. All genes that had a zero FPKM value in any sample were deemed to have low expression and removed from the analysis.
The R package limma was used to rank genes by their differential expression in the ST cell line between every time point and the zero time point, relative to the same comparison in the GFP cell line [23]. We selected all genes with P < 0.01 and with total absolute fold change across all time points (relative to GFP) above a cutoff of 4. This resulted in a list of 2854 genes that were differentially perturbed over time by ST. We used the R package mclust to cluster the genes, based on their expression across all time points in the ST inducible cell line. Expression values were mean-centered and scaled by the standard deviation across the ST samples. The 50 clusters were tested for GO term enrichment using the R package GOstats with p-values adjusted for multiple testing by the Benjamini-Hochberg method. Similarly, the clusters were also tested for enrichment in pathways representing the hallmarks of cancer downloaded from the Molecular Signatures Database (MSigDB) [58]. The significance of each overlap was evaluated using a hypergeometric test and adjusted for multiple testing using Benjamini-Hochberg.
Position weight matrices (PWMs) for NFKB1 and MYC::MAX were extracted from the R package MotifDB. The SLC16A1 promoter from -1000 to +100 relative to TSS was mapped to the binding motifs using a cutoff of 0.8 for the estimated probability of a match between the promoter sequence and the PWM [59]. DNAse I hypersensitivity data for IMR90s was downloaded from the Gene Expression Omnibus (GEO), accession numbers GSM468801, GSM530665, GSM530666, and GSM468792.
Extracellular acidification rate (ECAR) was measured using the Seahorse Bioscience XFe24 Extracellular Flux Analyzer according to the manufacturer’s protocol. For IMR90 extracellular flux analysis, cells were seeded into assay culture plates (2 x 104 cells/well) 24 hours prior to the assay. For MCC cell line analysis, assay culture plates were coated with Cell-Tak (Corning) following protocol from Seahorse Bioscience. Cells (1 x 105 cells/well) were adhered to coated assay plate wells via centrifugation.
Cells were rinsed and cultured in XF Base Medium (Seahorse Bioscience) supplemented with 10 mM glucose (GIBCO), 1 mM sodium pyruvate (Sigma), 1% Glutamax, and pH was adjusted to 7.4 prior to performing the assay. Where described, DMSO and CHC were added to complete XF media before the start of the assay. Real-time OCR and ECAR data are representative of two biological replicates, with values representing the means and error bars representing standard deviation of five technical replicates at each time point.
For media glucose and lactate measurements, IMR90 cells inducibly expressing either ST or GFP were seeded in duplicate (1 x 105 cells/well) in a 24-well plate using standard IMR90 culture media supplemented with 5 mM lactate (Sigma). Dox was added and media was collected daily for 5 days. Day 0 media corresponds to a sample of fresh growth medium. Glucose and lactate was measured using a YSI Biochemistry Analyzer.
IMR90 anchorage-independent growth was performed as described [36] using 6-well dishes with SeaPlaque Agarose (Lonza) at concentrations of 0.3% top and 0.6% bottom layers. Agarose was diluted with 2X MEM (Gibco) supplemented with 2X Glutamax, 2X Pen Strep, and 30% FBS. IMR90 cells (104) were seeded in triplicate in the top agarose layer. Wells were fed with top agarose twice per week. After 4 weeks, cells were stained with 0.005% crystal violet (Sigma) in PBS and colonies were counted. MCT1 inhibitors were included into soft agar layers at the concentrations described above.
IMR90 proliferation assays were performed as previously described [60]. Briefly, IMR90 cell lines were seeded in triplicate in 24-well plates (day 0; 5 x 103 cells per well). Cell density was measured by crystal violet staining at intervals after plating as previously described [22]. MCC cell line proliferation assays were performed in triplicate in 48-well plates using XTT assay (Roche) following the manufacturer’s protocol.
|
10.1371/journal.pgen.0030093 | Selection for Robustness in Mutagenized RNA Viruses | Mutational robustness is defined as the constancy of a phenotype in the face of deleterious mutations. Whether robustness can be directly favored by natural selection remains controversial. Theory and in silico experiments predict that, at high mutation rates, slow-replicating genotypes can potentially outcompete faster counterparts if they benefit from a higher robustness. Here, we experimentally validate this hypothesis, dubbed the “survival of the flattest,” using two populations of the vesicular stomatitis RNA virus. Characterization of fitness distributions and genetic variability indicated that one population showed a higher replication rate, whereas the other was more robust to mutation. The faster replicator outgrew its robust counterpart in standard competition assays, but the outcome was reversed in the presence of chemical mutagens. These results show that selection can directly favor mutational robustness and reveal a novel viral resistance mechanism against treatment by lethal mutagenesis.
| Understanding the conditions that favor the constancy of phenotypes in the face of deleterious mutation pressure—mutational robustness—is an outstanding question in evolutionary biology. Theoretical and in silico studies utilizing digital organisms predict that slow-replicating populations can outcompete those with higher individual fitness if the former show greater robustness. This “survival of the flattest” hypothesis sits in contrast to most models of natural selection based on individual fitness, and hence challenges the “survival of the fittest” paradigm. In this work, the authors use experimental populations of the rapidly evolving vesicular stomatitis RNA virus to provide the first evidence of natural selection for mutational robustness. Based on the analysis of fitness distributions, genetic variability, and the ability to tolerate mutation accumulation, two populations with different levels of robustness were characterized. At artificially enhanced mutation rates following the application of mutagens, the more robust viral population outcompeted the other population despite having a lower replication rate. This study has important implications for lethal mutagenesis—an antiviral strategy that consists of increasing viral mutation rates through the use of mutagenic drugs—since selectively favored mutational robustness may allow RNA viruses to evolve resistance to this form of treatment.
| Lethal mutagenesis consists of overwhelming viral populations with an excessive number of deleterious mutations and has been proposed as a candidate therapeutic strategy against RNA viruses [1–4]. Several mutagens have been used to artificially increase error rates in RNA viruses such as vesicular stomatitis virus (VSV), poliovirus type 1, foot-and-mouth disease virus, lymphocytic choriomeningitis virus, hepatitis C virus, and the human immunodeficiency virus type 1 [1]. These experiments show that while mutagens efficiently reduce viral fitness, they also impose a strong selective pressure for the evolution of resistance mechanisms. One mechanism of resistance is increased replication fidelity, which has been demonstrated in experimental populations of poliovirus type 1 subjected to ribavirin treatment [5]. Resistance mechanisms involving changes in viral polymerases that specifically reduce the incorporation efficiency of the mutagen have also been reported [6]. Another potential mechanism of resistance is increased mutational robustness, although the later remains unexplored experimentally.
Robustness to mutation determines the phenotypic expression of genetic variation and thus should be central to many evolutionary processes [7]. Insofar as it is heritable, exhibits variability among individuals, and affects the probability of survival, robustness is a potential target for selection and evolutionary optimization. However, whether robustness can be directly favored by natural selection remains controversial [8–10]. Population genetics theory predicts that robustness can only be efficiently selected for if mutation is highly frequent [7,11,12] . Similarly, quasispecies models predict that, in small replicons, robustness can significantly influence the mean fitness of the population at high mutation rates or low population sizes [13,14].
The conditions for the evolution of robustness have been experimentally explored in silico [15–17]. Such experiments have produced digital organisms with increased mutational robustness, although they typically pay the cost of reduced replication rates, an evolutionary trade-off that is expected from both theoretical and molecular considerations [13,18]. Specifically, following Wright's adaptive landscape model, organisms with high robustness are located in a low and flat peak, as opposed to faster but less robust replicators. Although the flatter populations should be readily outcompeted by their faster counterparts at low mutation rates, they can have a selective advantage at high mutation rates, a phenomenon dubbed the “survival of the flattest” (Figure 1) [17]. Recent experiments with viroids (plant pathogens constituted of small noncoding RNA) have revealed that a slow-replicating but highly variable species can improve its fitness relative to a faster but less variable replicator in UVC-irradiated plants, compatible with the survival of the flattest hypothesis [19].
Due to their high spontaneous mutation rates [20], RNA viruses are obvious candidates for studying the beneficial effects of mutational robustness. However, this task has to date remained elusive. Early work with the RNA phage φ6 showed that the evolution of different genotypes depended on the topology of the neighboring adaptive landscape [21]. More recent experiments with φ6 have provided some indirect evidence that robustness is an evolvable character [22]. However, direct proof of the survival of the flattest requires at least the following conditions: (i) that the outcome of competition experiments between genotypes should depend on the mutation rate and, (ii) that average mutational effects should be estimated for each competitor.
Herein, we provide evidence for the survival of the flattest using experimental populations of VSV, a lytic negative-stranded RNA virus of the Rhabdoviridae family. We characterized two populations adapted to a common environment but with different evolutionary histories. We studied the genetic variability and the individual fitness distribution of each population near the mutation-selection balance, and we estimated mutation rates and selection coefficients for deleterious mutations. This allowed us to conclude that one population was replicating near a high fitness peak, but was also less robust to mutation than the other. Whereas at spontaneous mutation rates, the fitter population was the best competitor, the addition of mutagens provided an advantage to the more robust, flatter, population.
Two VSV populations with different evolutionary histories were chosen to start the experiments. Population A came from transfection of a full-length infectious cDNA into standard baby hamster kidney cells (BHK21) [23]. This cDNA was an artificial assemblage of different viral isolates and had previously experienced no propagation in natural or laboratory conditions. In contrast, population B had a complex laboratory passage history, since it originally came from a natural isolate of the Indiana serotype and, after a few passages in BHK21 cells, it had been replicating in human cervical cancer HeLa cells [24]. The consensus sequence of these two populations differed at 54 nucleotide positions spread throughout the genome.
A single clone from each of populations A and B was picked and evolved separately by serial passages for approximately 100 generations in the same, constant, environment, after which the two evolved lineages were assayed for fitness by standard growth assays. In accordance with previous studies [25], population A readily adapted to the new environment, showing a 53.9% fitness increase between generations 5 and 100 (Mann-Whitney test, p < 0.001). In contrast, population B showed a non-significant 7.0% increase in fitness (p = 0.075) during the same time interval. Comparison of ancestral and evolved genomic consensus sequences revealed a single fixed nucleotide substitution for population A and no changes for population B. EMBL accession numbers of the genomic consensus sequences for A and B evolved populations can be found in the Accession Numbers list of the Supporting Information section of this paper. The two adapted populations were used for all subsequent experiments. In both cases, a mutation-selection balance had probably been reached at different local peaks of the adaptive landscape, as suggested by the fact that evolution for an additional 50 generations produced no changes in mean fitness.
To gain some insights into the topology of each local adaptive peak, we first measured the fitness of 12 randomly chosen clones from both A and B. Although there were no significant differences in average fitness (Mann-Whitney test, p = 0.143), it was striking that fitness variance among clones was 30 times higher for population A than for population B. We therefore investigated each population more thoroughly. Because of the large samples required, we used lysis plaque size as a proxy to measure fitness rather than standard growth assays. Preliminary experiments with 12 clones from each population showed that fitness and log-plaque size were highly correlated (Pearson r = 0.944 for A, r = 0.959 for B, p < 0.001 in both cases). Estimation of fitness values of 1,000 random individual clones from each population showed that population A contained the fittest variants but also a large tail of unfit clones, whereas the fitness distribution of population B fitness was more tightly clustered around intermediate values (Figure 2). Reduced fitness variance in population B could be due to either a lower mutation rate or a higher mutational robustness.
We further characterized each population by studying their genetic heterogeneity at two variable regions of the genome. Region P covered half of the gene encoding the viral phosphoprotein, a short intergenic region, and a small fraction of the gene encoding the matrix protein, whereas region G was entirely located within the viral envelope glycoprotein gene. After obtaining cDNAs by reverse transcription of purified RNAs, these regions were amplified by PCR, cloned, and sequenced. In population A, six different mutations were found in 87 clones, giving a mutation frequency of 1.35 × 10−4, whereas in population B, this frequency was elevated to 2.91× 10−4 (Table 1). The observation that population B was roughly twice as variable as A can be explained by a higher mutation rate, but also by increased robustness. Recalling that the characterization of individual fitness distributions (Figure 2) indicated that population B should have either a lower mutation rate or increased robustness, that B has increased robustness seems to be the only hypothesis compatible with the observations.
If B was indeed more robust to mutation than A, we expected that their relative fitness should depend on the mutation rate. To test this prediction, we performed competition experiments taking advantage of the fact that the two populations were easily distinguishable by a monoclonal antibody resistance marker situated on genotype B. Standard assays with a 1:1 input ratio indicated that population A had an advantage over B (log relative fitness, logWB/A = −0.114 ± 0.012, Mann-Whitney test: p = 0.002), confirming that A was located at a higher fitness peak. We then carried out the same experiments in the presence of the mutagen 5-fluorouracyl (5-FU). As expected, the outcome of the competition steadily changed for increasing concentrations of 5-FU (Spearman ρ = 0.933, p < 0.001) and B prevailed beyond a dose of ~30 μg/mL (Figure 3). To assess whether this pattern was drug specific, we repeated the competition assays using 5-azacytidine (5-AzC) instead of 5-FU. Similarly, we observed that B improved its performance relative to A as the dose of 5-AzC increased (Spearman ρ = 0.854, p < 0.001; Figure 3).
If selection was insensitive to mutation and hence targeted individual clones, we expected that the faster replicators of population A would always outcompete those of population B. A thousand simulations in which the empirical distributions of individual fitness values were used to predict the outcome of the competition confirmed that, if selection acted on individual clones, we should expect A to be the winner of the competition. Whereas the results from real competition experiments in the absence of mutagens were consistent with this prediction, competitions with mutagens were the opposite, suggesting that selection was favoring the population with the higher average growth rate, even though it did not harbor the fittest individuals. The most likely scenario is thus that population B was located at a lower but flatter peak of the fitness landscape, in which mutations would tend to be less deleterious.
To exclude the possibility that the RNA polymerase encoded by genotype B had an increased ability to selectively purge base analogs from its active site, we obtained molecular clone sequences of the P and G regions after three passages at 80 μg/mL 5-FU. As expected, the presence of 5-FU altered mutational patterns and increased mutation frequencies. Whereas 44% of all spontaneous nucleotide substitutions were transitions, after 5-FU mutagenesis, this percentage increased to 90%. Mutation frequencies increased to 7.09 × 10−4 for population A and to 9.55 × 10−4 for population B (Table 2). Compared to the spontaneous mutation rate (Table 1), the relative increase in mutation frequency was higher for population A, whereas in absolute terms, it was higher for B. However, population B remained more variable than A after 5-FU mutagenesis, demonstrating that even if there were differences in the ability to exclude 5-FU, these could not account for the observations. Similarly, it is in principle possible that genotype B might have a better ability to replicate in cells stressed by mutagenesis, although this explanation is similarly unable to account for all observations.
The conclusion that B is more robust to mutation than A relies on the assumption that no beneficial mutations were sweeping through either population at the time of competition assays. Several precautions were taken to ensure that this requisite was met. First, as mentioned above, experimental evolution for additional 50 generations produced no changes in mean fitness, suggesting that beneficial mutations were not frequent. Second, competition assays were seeded with only ~100 plaque-forming units (pfu) of each population to minimize the probability that a rare beneficial mutation distorted the results. Using binomial probabilities, it can be proven that high fitness variants were very unlikely to be present in the inoculum of the competitions experiments and at the same time, go unnoticed during the plaque size fitness screening. Third, we repeated the above competition experiments using individual clones instead of populations. To do that, we randomly selected four clones from each population and we did one-to-one competitions against each of four clones drawn from the other population. We observed a significant correlation between the 5-FU dose and the log-fitness of B clones relative to A clones (Spearman ρ = 0.406, p = 0.004), making it apparent that, on average, competitions between single clones reproduced the same trend observed with populations (Figure 4).
Serial passages were done at the lowest possible population size to minimize the action of natural selection and hence favor the fixation of deleterious mutations via genetic drift. If it is true that genotype B is more robust than A, we expected this passage regime to produce a lower fitness decline in the former. For each population, starting from a single clone, we seeded 24 independent lineages that were propagated plaque to plaque for approximately 25 generations. We observed that the average fitness of these lineages decreased and fitness variance increased in both cases. However, as expected, the fitness loss was more marked for genotype A (Figure 5; Mann-Whitney test: p < 0.001). Similarly, the observed increase in genetic variance was much more evident for genotype A (ΔlogWA = 0.876, ΔlogWB = 0.169).
The expected change in average log fitness equals the product of the deleterious mutation rate (Ud) and the average selection coefficient against deleterious mutations (s) [26,27]. Therefore, mutational robustness can be achieved by increasing the fraction of neutral mutations or by decreasing deleterious mutational effects. Although distinguishing between these two alternatives is not straightforward, mutation accumulation data can be used to estimate Ud and s separately using the Bateman-Mukai method [26,27]. No statistical differences were observed between the two Ud estimates (Ud A = 0.062, Ud B = 0.112, bootstrap test p = 0.247), whereas the s value was significantly higher for A (sA = 0.443, sB = 0.061, p = 0.003), suggesting that B was more robust to mutation than A because deleterious mutational effects were lower on average. We also applied a maximum likelihood method to estimate Ud and to characterize the distribution of mutational effects [28], but this approach yielded nonconvergent estimates.
In evolutionary theory, adaptation is often assumed to be a necessarily uphill path towards the highest-fitness genotype, a paradigm known as the “survival of the fittest” [29]. However, the fittest genotype cannot be always defined in terms of individual fitness. Previous work with the RNA phage Φ6 showed that the long-term fitness of a genotype can be determined by its mutational neighborhood [21]. Here, we have characterized two VSV populations with different levels of mutational robustness and shown that this difference has a selective value at high mutation rates. Our conclusion that one population is more robust than the other is supported by the observation of three expected consequences of robustness: reduced dispersion of the individual fitness distribution, increased genetic heterogeneity, and reduced sensitivity to mutation accumulation.
In contrast to the survival of the fittest, the survival of the flattest hypothesis states that at high mutation rates, fast replicators can be outcompeted by slower replicators provided the latter benefit from increased mutational robustness. The survival of the flattest takes place when the average mutational effects influences the fitness of the population, implying that the fitness of a given genotype is determined by the occurrence of subsequent mutations or by the influx of mutations regenerating the genotype from neighbors in sequence space. These second-order effects, also termed quasispecies effects [30], are negligible for low mutation rates, but become increasingly important for higher mutation rates. Similarly, these effects become more relevant at low population sizes, because this regime favors the accumulation of mutations in the population [13]. Quasispecies effects are not fully accounted for in some classical population genetics models, which, despite incorporating the negative impact of mutation, assume that it is rare enough to allow for the neglect of multiple and back mutations. A classical formulation for asexual replicators is
W̄=
, where
W̄is the average fitness of the population and W0 is the fitness of the fittest genotype [31]. As a consequence, the success of a given genotype can be influenced by the rate of deleterious mutation, but not by the average selection coefficient against deleterious mutations. In our experiments, W0A > W0B and
, but at artificially increased mutation rates,
W̄A <
W̄B, which contradicts this model and clarifies the need to consider mutational robustness to predict the fitness of viral populations subjected to mutagenesis.
RNA viruses are highly sensitive to mutation when compared to more complex microorganisms [32], a lack of robustness that is generally expected in small, compact genomes with little redundancy, no repair systems, and strong pleiotropy [13]. These species typically exist as very large populations, thus making selection efficient at purging deleterious mutations and promoting the preservation of the unmutated genotype. Genetic hypersensitivity seems to be the rule among RNA viruses but, under some conditions, the evolution of mutational robustness can be favored. Here, at the highest tested doses, 5-FU reduced the maximum viral titers by three orders of magnitude and produced more than a 3-fold increase in mutation frequencies, which should provide a favorable scenario for the evolution of robustness. Similar scenarios are likely to apply in nature following antiviral treatments [1], transmission bottlenecks [33], or host-induced mutagenesis [34].
Lethal mutagenesis has been proposed as a therapeutic strategy against RNA viruses [1–4]. For example, mutagenesis is one of the mechanisms of action of ribavirin, a drug that is currently used in combination with interferon to combat hepatitis C [35]. However, RNA viruses are known to have a remarkable potential for escaping antiviral strategies and evolving resistances. Previous work has shown that poliovirus-1 and foot-and-mouth disease virus can increase their replication fidelity or substrate specificity in response to ribavirin treatment [5,6]. By showing that robustness is a selectable trait at high mutation rates, our results reveal a novel resistance mechanism against lethal mutagenesis and open several research avenues. For example, the genetic basis of increased robustness in genotype B still needs to be identified. Here, 53 substitutions separate the two genotypes, but thermodynamic analyses predict that one or few changes suffice to modify the robustness of a given protein [36]. Related to this issue, it remains unclear whether the transition from a non-robust to a robust state can occur as a direct consequence of the selective pressure exerted or, in contrast, an episode of genetic drift is required to produce a jump in the adaptive landscape. Finally, adaptation to mutagenesis also has consequences for the evolvability of RNA viruses. Whereas the evolution of increased replication fidelity has been shown to restrict RNA virus evolvability [37], increased robustness may favor the generation of inconspicuous genetic variation and foster long-term evolvability [7,38].
BHK21 (American Type Culture Collection, http://www.atcc.org) were cultivated at 37 °C, 5% CO2, in 100 cm2 plates under 12 mL of Dulbecco modified Eagle's minimum essential medium (DMEM) supplemented with 10% fetal calf serum, and passaged upon confluence. The VSV infectious cDNA used to obtain population A was kindly provided by G. T. W. Wertz, University of Alabama, United States of America [23].
A 100-μL volume of an appropriate dilution was added to 60 mm culture plates and cells were topped with DMEM medium containing 0.4% agarose. Monolayers were stained at 18–22 h post-inoculation (hpi) with a solution of 2% crystal violet (Sigma, http://www.sigmaaldrich.com) in 10% formaldehyde (Panreac, http://www.panreac.com/new/ing/menu.htm).
Approximately N0 = 5 × 103 pfu were inoculated to C = 106 BHK21 cells in 25 cm2 flasks, incubated for 24 h, and the supernatant was used to seed a new infection passage under the same conditions, up to 25 passages. Viral population sizes at the end of each infection passages were approximately Nf = 5 × 109 pfu. The number of viral infectious cycles (generations) per passage was estimated from N0, Nf, and C as previously described [39], giving ~4 generations per passage.
Viral RNA was extracted from the supernatant of the infected cultures using the High Pure Viral Nucleic Acid Kit following the manufacturer instructions (Roche, http://www.roche.com). The anti-genomic VSV cDNA was synthesized using the reverse transcriptase of the Moloney murine leukemia virus (Promega, http://www.promega.com) plus a battery of random hexamers (Promega). The genomic region containing the original mutations was amplified by PCR using Taq polymerase (Amersham, http://www.gehealthcare.com) and specific primers. Sequencing was carried out using ABI PRISM BigDye Terminator v3.0 Ready Reaction Cycle Sequencing KIT (Applied Biosystems, http://www.appliedbiosystems.com) on an ABI 3700 automated sequencer. Sequences were visualized and edited with the Staden software package (http://staden.sourceforge.net).
RNA viral extraction, RT-PCRs, sequencing, and editing of sequences were performed as described above. A first round of PCR was carried out using Pfu polymerase (Amersham) and specific primers [40]. Amplified DNA products of each region were purified with High Pure PCR product Purification Kit (Roche) and directly cloned into EcoRV-digested pBluescript II SK (+) phagemid (Stratagene, http://www.stratagene.com). A second round of PCR was then carried out by adding a single transformed bacterial colony in each PCR tube and using Taq polymerase. Vector-based primers KS and SK (Stratagene) were used in this second round of PCR, as well as for the subsequent sequencing. Before sequencing, amplified DNA was purified using a PCR clean-up kit (Macherey-Nagel, https://www.macherey-nagel.com). Plasmid DNA was purified with High Pure Plasmid Isolation Kit (Roche) and clones were sequenced using vector-based primers KS and SK (Stratagene). EMBL accession numbers of the obtained sequences for nonmutagenized populations and for 5-FU mutagenized populations are in the Accession Numbers list in the Supporting Information section of this paper.
Following protocols established in previous work [41–43], we used growth rate assays to estimate fitness. We seeded ~5 × 103 pfu of each population into ~105 BHK21 cells and incubated the culture until the population grew up to a titer of ~107 pfu/mL (i.e., 7–8 hpi). From final and initial titers, the growth rate (r) was calculated as the slope of log-titer regression against time (hpi). We defined fitness (W) as the number of descendants per individual per hour, i.e., W = er − 1. Fitness assays were done in triplicate.
Viruses were plated as detailed above, paying care to put less than 50 pfu per plate to avoid plaque overlapping. All platings were done in a single block and the same overlay medium batch was used for both populations and staining was done at 24 hpi. Pictures of the plates were taken with a 5 megapixel Canon PowerShot G5 digital camera (http://www.canon.com) and image analysis was done with AnalySIS v.3.2 software (Soft Imaging System, http://www.soft-imaging.net). The vast majority of plaques were automatically identified, whereas the rest were manually delineated in a zoomed image. After defining plaques as single objects, we automatically obtained their surface area in pixels, S.
To calibrate the relationship between plaque size and fitness, we selected 12 clones from each population that widely varied in fitness and, for each, we determined the average plaque size from four independent plates. Since we had no evidence that fitness and plaque area were linearly related, we performed a log–log regression to obtain a calibration line of the form logW = p + mlogS. For population A, we obtained m = 1.907 ± 0.318 and p = −13.785 ± 2.246 (r = 0.884), whereas for population B, we obtained m = 0.927 ± 0.096 and p = −5.874 ± 0.678 (r = 0.950). The calibration line was used to transform each observed plaque size into a predicted fitness.
Although the calibrations described above seemed relatively accurate, it is unavoidable that there was some degree of uncertainty and consequently no guarantee that no bias was introduced. We therefore carried out a second analysis that bypassed the calibration step. To achieve this, we simply assumed that the number of lysed cells was proportional to the number of viruses produced. First, we estimated the number of lysed cells on a plaque, D, by dividing its pixel surface area per the total pixel surface area of the 60 mm plate and multiplying per the total number of cells on a plate, which was obtained by counting cells in an hemocytometer (Neubauer). Second, we estimated the average number of viruses produced per cell (K) by titrating a fully lysed plate. This was done separately for populations A and B, yielding K = 2,204 ± 32 and K = 2,712 ± 82 respectively. Using these estimations, plaque sizes were transformed into predicted fitness as Wpred = (KD)1/24 − 1. The results obtained with this latter fitness estimation method (unpublished data) were consistent with those obtained using the calibration line (population A showed higher maximum fitness and higher variance).
Standard growth rate assays were performed as described above except that in competitions between populations (Figure 3), only 100 pfu from each competitor were inoculated to minimize the effect of rare beneficial mutations (see below). Viruses from the two genotypes were inoculated into the same well. Genotype B carried a monoclonal antibody resistance marker that allowed us to estimate the titer of each population in the mixture by plating in presence and absence of antibody. Samples were taken at 0, 7, 8, 9, 10, 11, 12, 15, 20, 25, and 30 hpi for competition with no mutagen. For competitions in the presence of 5-FU or 5-AzC, cells had been pretreated with the indicated 5-FU concentration 12 h prior infection. Samples were taken at 0, 15, 20, 25, 30, 35, 39, and 49 hpi for 20 and 40 μg/mL 5-FU and at 0, 20, 25, 30, 35, 39, and 49 hpi for 80 μg/mL 5-FU. Samples were taken at 0, 10, 12, 15, 24, 28, 32, 38, and 48 hpi for 5 and 10μg/mL 5-AzC, and at 0, 24, 28, 32, 38, and 48 hpi for 20 μg/mL 5-AzC. Growth rates for each population were calculated as the slope of the log-titer against hpi during the exponential growth phase. For competitions between populations, 12 replicates of the competition assay were done for each different mutagen dose, whereas for competitions between clones three replicates were done for each dose and each of the four clone pairs.
We reduced the initial inoculum to ~100 pfu of each population to minimize the effect of rare beneficial mutations in the outcome of the competition. Since 1,000 clones were sampled during the plaque size fitness screening, the probability that beneficial mutations at a frequency f in the population went unnoticed during this screening is given by the binomial distribution function, Bi(0, 1,000, f). For each competition assay, the probability that these beneficial mutations were present in the inoculum is 1 − Bi(0, 100, f). Therefore, for each replicate of the competition assay, the joint probability that beneficial mutations that had been missed in the plaque size screening and were present in the competition assays is p = Bi(0, 1,000, f) (1 − Bi(0, 100, f)). For all values of f, p ≤ 0.035 for each replica of the competition assay, thus making it very unlikely that rare beneficial mutations had affected the results of the competition after 12 replicates.
The expected outcome of a competition between population A and B, based on plaque sizes, was obtained as follows. First, to mimic the inoculum size of the real competition assays, we randomly sampled a subset of 100 individuals from each population and we got their expected progeny per hour, which is equal to the predicted fitness. We then let the simulated growth proceed until the total population size was similar to that observed at the last time point of real competitions. The expected winner of the competition was the one with the larger progeny number at the final time point. The random sampling and the simulation were repeated 1,000 times.
Two random clones from populations A and B were picked from a 60 mm plate and stored at −80 °C as ancestors. Twenty-four mutation accumulation lines were founded from each ancestor by randomly picking 24 lysis plaques. Following previous plaque-to-plaque experimental designs [44], for each lineage, viruses were plated and at 24 hpi, a lysis plaque was randomly sampled, resuspended in DMEM medium, and directly plated onto a fresh monolayer. This protocol was pursued on a daily basis for12 passages. The number of generations elapsed during each plaque-to-plaque passage was estimated to be approximately two. To obtain enough viruses for fitness assays, the ancestors and the final derived clones were given a single passage in a 96-well plate containing ~104 BHK21. Fitness assays were performed as detailed above simultaneously for the ancestors and the derived lines in three experimental blocks. In each block, the 24 derived clones of each population were assayed once, whereas the ancestors were assayed six times.
The expected change in log-fitness after t generations (t = 24) is ΔlogW = logWt − logW0 = UdtE[log(1 − s)]. The expected genetic variance for log-fitness among lineages is σG2(logWt) ≈ UdtE[(log(1 − s))2], and the total variance is σT2(logW) = σE2(logW) + σG2(logW) where sub-index E refers to environmental variance. Since genetic variance is null for the ancestor and all fitness assays were performed in the same environmental conditions, σE2(logWt) = σE2(logW0) = σT2(logW0), and therefore σG2(logWt) = σT2(logWt) − σT2(logW0) = ΔσT2(logW). It follows that
and that
where θ is the coefficient of variation (standard deviation to mean ratio) of log W associated to single deleterious mutations. (1 + θ2)E[log(1 − s)] and Ud/(1 + θ2) were directly estimated from the data.
For an exponential distribution (a simple and relatively accurate model for describing fitness effects associated to single mutations), θ = 1. Nonetheless, the actual distribution of mutational effects might have a heavier tail and thus a θ > 1. A previous fitness dataset of 28 nonlethal random mutants of VSV [42] gave the empirical estimation θ = 1.598. Using this figure, we obtained estimates of Ud and E[log(1 − s)] and then, of s. For A, θ = 1.598 should be an accurate estimation of the true θ-value, because it was obtained for the above-mentioned full-length infectious cDNA clone. It is possible, however, that, due to its higher mutational robustness, B showed a lower θ-value. This potential bias could account for the slightly higher Ud estimate in B, but it could not account for the nearly one order of magnitude difference in s-values (in the extreme case θB = 0, the true Ud B would be 2.598 times lower than our estimate and the true sB 2.598 times higher).
To address whether differences between Ud and s between A and B were statistically significant, we generated 1,000 bootstrap pseudo-replicates from both the 24 averaged fitness values of the derived clones and the six averaged fitness values of the ancestors. After obtaining the corresponding 1,000 pseudo-replicates of Ud, we counted the number of times Ud was larger for A versus B, and the same for s.
Parameters Ud and s were also estimated using the maximum likelihood approach implemented in the program MLGENOMEU (http://homepages.ed.ac.uk/eang33/mlgenomeu/mlginstructions.html), in which a Gamma distribution of the form Ga(s) = αβsβ − 1e − αs/Γ(β) is used to describe the distribution of mutational effects (s =β/α) [28]. Following the author's recommendations, we ran the program to estimate Ud and α at fixed β-values varying from 99 to 0.01. The log likelihood monotonically increased as β decreased, yielding increasingly higher α-values (s → ∞ and Ud → 0).
Statistics were done with MS Excel (http://www.microsoft.com) and the SPSS 12.0 package (http://www.spss.com). Resamplings and simulations were done using a Perl script.
The European Molecular Biology Laboratory (EMBL) (http://www.ebi.ac.uk/cgi-bin/emblfetch) database accession numbers for the A and B evolved populations are AM690336 and AM690337, respectively. For nonmutagenized populations, accession numbers are AM689332–AM689519 and for 5-FU mutagenized populations, accession numbers are AM689705–AM689876. |
10.1371/journal.pcbi.1005231 | Synchronization and Random Triggering of Lymphatic Vessel Contractions | The lymphatic system is responsible for transporting interstitial fluid back to the bloodstream, but unlike the cardiovascular system, lacks a centralized pump-the heart–to drive flow. Instead, each collecting lymphatic vessel can individually contract and dilate producing unidirectional flow enforced by intraluminal check valves. Due to the large number and spatial distribution of such pumps, high-level coordination would be unwieldy. This leads to the question of how each segment of lymphatic vessel responds to local signals that can contribute to the coordination of pumping on a network basis. Beginning with elementary fluid mechanics and known cellular behaviors, we show that two complementary oscillators emerge from i) mechanical stretch with calcium ion transport and ii) fluid shear stress induced nitric oxide production (NO). Using numerical simulation and linear stability analysis we show that the newly identified shear-NO oscillator shares similarities with the well-known Van der Pol oscillator, but has unique characteristics. Depending on the operating conditions, the shear-NO process may i) be inherently stable, ii) oscillate spontaneously in response to random disturbances or iii) synchronize with weak periodic stimuli. When the complementary shear-driven and stretch-driven oscillators interact, either may dominate, producing a rich family of behaviors similar to those observed in vivo.
| For decades, cardiovascular physiology has been an area of intense research, and we have a fundamental understanding of the mechanisms the heart uses to drive blood flow through the distributed network of vessels in the body. The lymphatic system is now receiving similar attention as more is learned about its functional role in disease processes. The importance of the lymphatic system in collecting excess fluid from tissues and returning it to the blood is well known, but how the lymph flow is regulated without a central pump is poorly understood. Each segment of collecting lymphatic vessel can independently contract yielding a network of distributed pump/conduits. This paper shows how the lymphatic muscle cells that squeeze fluid along the lymphatic vessels can be effectively regulated using only chemical and mechanical signals that they receive from their immediate microenvironment. Using stability theory and the tools of nonlinear dynamics we identify two complementary oscillators that respond to stretch of the vessel wall and shear of fluid flowing over the vessel wall. Numerical simulations of the combined oscillators show that they have characteristics well suited to the regulation of distributed systems in general and may have application in other biological and physical contexts.
| To maintain fluid homeostasis, interstitial fluid drains into the lymphatic system through initial lymphatic vessels that carry it to the collecting lymphatic vessels. The collecting lymphatic vessels transport the fluid (known as lymph) both passively and actively to lymph nodes and back to the systemic blood circulation. Collecting lymphatic vessels are surrounded by specialized lymphatic muscle cells (LMCs) [1] and sub-divided by valve structures that define individual segments called lymphangions (Fig 1) [2]. Lymphangions serve as both pumps and conduits. In contrast to the blood circulation, where a single pump drives flow through relatively passive conduits, each lymphangion has the ability to pump lymph through the converging network to lymph nodes and eventually to the thoracic duct. Pumping occurs when expansions in radius draw fluid into the upstream end of the lymphangion and then expel it downstream during a contraction. Directional flow is enforced by intraluminal valves, which favor flow toward the thoracic duct. Lymphatic vessel contractions are triggered when cytosolic Ca2+ entering from intravascular stores and outside the cell surpasses a threshold concentration in the cytoplasm of the LMC, resulting in actin and myosin cross-bridging within the LMCs [3]. The contraction phase ends as transmembrane pumps restore cytoplasmic Ca2+ concentration to equilibrium allowing actin-myosin binding to relax, and the trans-wall pressure and the passive elastic properties of the wall to reopen the vessel. The effects of Ca2+ on LMC contraction are moderated by endothelial-derived relaxation factors (EDRFs) that act as potent dilators of lymphatic and blood vessels when produced by the vessel-lining lymphatic endothelial cells (LECs) in response to dynamic fluid shear stresses. The best known EDRF is nitric oxide although others such as histamine have been shown to be important [4, 5]. For notational simplicity we represent the entire class of EDRFs herein as NO. The NO and Ca2+ levels are both subject to mechanical regulation; Ca2+ can enter the cell through stretch-activated ion channels [6, 7], and NO is produced by LECs when they are exposed to increased fluid shear stress [8]. Although rhythmic contractions can be produced by purely chemical oscillations in Ca2+ within the LMCs [9, 10], it is likely that feedback regulation is necessary for robust homeostasis. Indeed, we previously used a relatively complex numerical simulation of lymphatic pumping to demonstrate that a wide spectrum of oscillatory behaviors is possible, and that the behavior is very sensitive to local levels of stretch and stress [11].
Our present aim is to reduce the complexity of our previous model to examine how the observed oscillations arise from the integration of simple mechanical and chemical processes within a physiological system. Once simplified, we employ tools such as linear stability analysis to identify key parameter groups that determine the qualitative dynamic behaviors of such systems. Linear stability analysis seeks to determine which parameter values cause a small disturbance to grow rapidly from an initial state, or alternatively decay back to equilibrium. In addition, our approach allows us to show how oscillations can arise from the interactions between mechanical and chemical processes that lack intrinsic oscillators when considered separately. Here we develop generic formulations of the mechano-chemical processes in the muscular lymphatic vessel wall based on Ca2+ and NO signaling, then explore the dynamics of each chemical species while holding the effects of the other constant. Finally we examine the behavior of the fully coupled system. Linear stability analysis reveals a new class of oscillator arising from the dynamics of shear and NO that can act alone or in concert with the better recognized Ca2+ dynamics. The most remarkable feature of the shear-NO mechanism is its ability to offer distributed control of the pumping process, which is essential for managing a decentralized network of pumps and conduits.
Our model is based on a single lymphangion (Fig 1) bounded at each end by one-way valves. The radius of the lymphangion is governed by the radial forces which are determined by the contractile (Ca2+) and dilatory (NO) signaling molecules. Neglecting inertial effects, the radial forces on the vessel wall balance as
DdRdt=−E(R)−F(R,CCa,CNO)+AR(t)
(1)
where the left hand side represents rate dependent effects with D incorporating the visco-elastic material properties of the vessel wall as well as viscous losses in the flow and lags in the transduction of concentrations into force, E is a restoring force-including elastic forces and vessel “tone” -imparted by the material properties of the vessel wall and F is a dynamic inward acting contractile force produced by muscle cells that surround the vessel. To a first approximation, we assume that the concentration of Ca2+ is transduced into a contractile force as F=FCaCCa/(1+αCNO) where α scales the possible desensitizing effect of NO [12]. The activation term AR can include a steady component from the mean transmural pressure difference pm that influences the baseline radius as well as extrinsic disturbances to the radius from the surrounding tissue and adjacent lymphangions.
The restoring force is typically highly nonlinear[13–15]. Here we adopt the form E(R)=AeaR−poffset with stiffening coefficient a, scaling coefficient A, and offset pressure poffset selected to give a good fit of Shirasawa and Benoit (see the third figure of reference [15]) at typical operating pressures. In our numerical simulations we retain the full nonlinear form of E(R), but for the stability analysis that follows we linearize the elastic force near an equilibrium radius R1 in the absence of dynamic increments to Ca or NO as E(R)≈E0+E1(R−R1) where E0 is the elastic force at equilibrium and a Taylor series expansion near equilibrium yields E1=AaeaR1. We find the equilibrium radius by solving 0=DdRdt=−(AeaR1−poffset)−FCaSCa0/KCa+pm for R1 where pm is the mean transmural pressure. Given the stiffening behavior of the wall (E1∝eaR1), we expect that appropriate values of E1 will be larger at higher mean transmural pressures where the equilibrium radius will be somewhat larger.
The concentrations of the signaling molecules (i ∈ {Ca, NO}) are governed by the generic conservation law
dCidt=−Ki(CCa,CNO)+Si(R,R˙,CCa,CNO)+Ai(t)
(2)
where all concentrations are taken to be dimensionless ratios relative to a suitable reference, Ki is clearance of the signaling species through chemical reaction, transmembrane ion pumps and advective-diffusive transport, Si is a dynamic source term for the signaling molecule and Ai is an additional source term that can include the effects of imposed flow from upstream fluid pressure, inflammation, pace-making signals from adjacent cells, neural signaling, random disturbances, etc.
Since our focus is on the interactions between Ca2+ and NO, we introduce a minimal representation of the Ca2+ dynamics rather than a fully detailed model of Ca2+ oscillations as may be found in the literature [10, 16]. We retain the following features: i) at rest, Ca2+ is at a low concentration in the cytoplasm of LMCs; ii) a contraction is initiated when Ca2+ is rapidly admitted to the cytoplasm through ion-selective channels thereby triggering cross bridge formation between actin and myosin chains creating a contractile force [17]; iii) relaxation of LMCs coincides with a drop in cytoplasmic Ca2+ concentration due to a drop in the rate of influx and the restoration of baseline conditions by ion pumps in the cell membrane and sarcoplasmic reticulum; and iv) the LMC is refractory to a new contraction cycle until Ca2+ levels have returned to near equilibrium. As the Ca2+ levels approach their threshold level, we hypothesize that the membrane acquires sensitivity to small perturbations. Furthermore, the sensitivity is enhanced when the membrane is stretched to a larger radius. This models stretch-sensitive ion channels found in LMCs [6, 18]. Each step in the process has the potential for modulation by NO. Alternatively, each form of modulation by NO can be disabled to demonstrate behaviors that have been observed in experimental preparations, for example after removal of LECs (which produce NO) or the genetic or pharmacological suppressions of NO [19–21].
We mathematically express the release of Ca2+ into the cytoplasm from intracellular stores and the extracellular fluid as the sum of a steady source Sca0 needed to maintain the baseline Ca2+ concentration and a transient component that is sufficiently rapid to be modeled as an impulse function δ(t) where t is the time since the Ca2+ concentration most recently passed the threshold necessary to trigger another contraction CCaThresh. This is expressed as,
SCa=SCa0+SCa1δ(t)(1+γCNO)
(3)
where SCa1/(1 + γCNO) is the magnitude of a bolus of Ca2+ with a possible reduction due to NO that is scaled by γ. We model the clearance of Ca2+ from the cytoplasm with
KCa=KCa(1+βCNO)(CCa−CCaThresh)
(4)
where KCa is a rate constant and β scales the possible enhancement of Ca2+ clearance attributed to NO [12, 16]. At high concentrations of Ca2+ the clearance rate may be limited by the membrane pump capacity, but near the threshold required to trigger a contraction we assume clearance rates proportional to the concentration increment. The threshold itself may include a random component that we incorporate into the activation term.
The linearized form of the model for a constant level of NO can now be written as
DdRdt=−E1R−FCaCCa+AR(t)
(5)
and
dCCadt=−KCaCCa+SCa1δ(t)+ACa(t)
(6)
where the constants have been absorbed into the activation terms so that the radius and concentration now represent the increments from baseline values. The parameters SCa1 and KCa now include the adjustments due to NO introduced in Eqs 3 and 4.
In the previous section we developed the model so that it reproduces Ca2+ induced contractions. We next considered how NO, created when LECs experience increased shear stress, can modulate contractions when it diffuses rapidly into adjacent LMCs. A suitable form for the NO source term can be obtained by considering steady laminar flow in a circular tube with negligible inertia [22]. Conservation of mass in a tube of time-varying radius requires that ∂Q∂z=−2πRdRdt which when integrated with respect to z along a vessel yields Q(z)=−2πRdRdtz+C1 where the constant of integration depends on the end conditions for the lymphangion. When the segment is contracting dRdt<0 and the upstream valve is closed [Q(0) = 0] we have Q(z)=−2πRdRdtz. Alternatively, if the segment is expanding dRdt>0 and the downstream valve is closed (Q(L) = 0), we obtain Q(z)=2πRdRdt(L−z). We can express the mean flow along the length more compactly as Q¯=πR|dRdt|L. When additional flow Q0 is imposed on the segment by an axial pressure gradient we have Q¯=πR|dRdt|L+Q0. Approximating the velocity profile with that of steady laminar flow with negligible inertia [22] we relate the mean shear stress τ to the flow rate by τ=4μQ¯πR3. The mean shear stress along the segment due to dynamic changes during contraction or expansion is therefore approximately τ=4μLR2|dRdt|+4μQ0πR3. Recent studies show that valves in collecting lymphatic vessels are biased toward the open condition [23], but the simplification employed here allows us to study the basic stability of the system, at the possible expense of some accuracy in the predictions of pumping efficiency. There may be levels of shear stress below which NO production is negligible and above which NO production saturates at a maximum, but here we linearize the transduction of shear stress into the production of NO in an intermediate range to yield
SNO=SNOR2|dRdt|+SNO0
(7)
where SNO has absorbed the remaining constants in the shear stress expression and SNO0 represents NO released due to the through-flow term Q0 or chronic sources of NO such as might arise during inflammation. The source term SNO0 can be time varying, but arises from the local environment of the lymphangion and mathematically acts as an input to our model of a single lymphangion rather than as an interaction within the system itself. SNO0 therefore can serve as an external trigger to the system or as a steady offset, but does not directly impact the dynamics of an individual lymphangion, except by parametrically (rather than dynamically) changing the equilibrium radius.
We can examine the effects of fluid viscosity on the pressure by using the same set of assumptions. The pressure will vary due to viscous flow effects according to ∂p∂z=−8μQπR4. When contracting we have ∂p∂z=16μR3dRdtz, which when integrated along the length gives p(z)=16μR3dRdtz2+p(0). Averaging over the length of the segment yields p¯=16μL23R3dRdt+p(0) where the first term gives the magnitude of the pressure decrement (or increment for vessel expansion) due to flow induced by the contraction of a single lymphangion. We see that the pressure increment due to flow induced by the single lymphangion also multiplies dRdt, so it can be absorbed into the overall damping term D. For typical vessel sizes, we find that the lag due to viscosity is orders of magnitude smaller than that from chemical and mechanical lags which are on the order of one second.
NO does not produce a true outward force. However, it is conceptually equivalent to consider an effective force produced by NO that has the effect of countering FCa and the elastic effects. Mathematically, for small αCNO, we can write this as
F=FCaCCa01+αCNO≈FCaCCa0(1−αCNO)
(8)
By defining FNO ≡ FCaCCa0α, we can write
FNO=−FNOCNO
(9)
And the net force from Eq 8 becomes
F=FCaCCa0−FNOCNO
(10)
where the net contractile force is decomposed into a positive term set by the baseline Ca2+ levels and a negative term that represents how the Ca2+ levels are modulated by NO. Thus, the NO-dependent term is not a true outward force, but arises mathematically from a reduction in the Ca2+-dependent contractile forces.
The parameter values used in the simulations that follow are given in Table 1. The parameter values were based on experimental data where possible, but were chosen to demonstrate a wide range of mathematical behaviors for the system rather than to mimic a particular experimental data set in detail. As a representative example, we show simulations based on measurements in rats [24] which offer data relating Ca2+ concentrations to lymph vessel diameter and contractile tension. Our own experiments discussed later [20] were done on mice which have smaller collecting lymphatic vessels than rats.
Specific parameters governing the effects of NO are difficult to estimate, but fortunately may not be necessary here. As will be shown in the results section, we require only estimates of combinations of parameters such as SNO and FCaCCa0α, rather than values for each parameter individually. Unlike the geometrically-detailed continuum model of lymphatic NO transport in Wilson et al [25] that includes shear-induced production and clearance by diffusion, convection and reaction, our present model employs averages over a single lymphangion and combines the sensitivity to shear stress with the rate of production of NO. To that end, we employ parameter values that yield diameter changes due to NO on the order of 10% as observed in [20, 21]. Moreover, we expect the effects of NO to be rapid. NO is released by endothelial cells about 2 seconds or less after increases in shear stress as observed previously [26–29]. Lymphatic vessels can be expected to dilate faster than blood vessels [30] because their muscle cells contain more rapid-acting contractile proteins than those of blood vessels [31]. We take the clearance of NO to be similar to, but somewhat faster than, that of Ca2+ [32]. Parameters such as the contractility, NO production and the mechanical stiffness appear to depend on anatomical location, species and age [13, 28, 33–36], suggesting that the full range of possibilities realizable in vivo awaits further investigation.
Here we first present analytical and numerical results for cases in which the stretch-Ca2+ dynamics are active, but with constant NO levels. We will then present results of our model for when shear-NO dynamics are active, but Ca2+ levels do not spike, but remain near baseline. Finally, we will present results of our model for fully coupled stretch-Ca2+ and shear-NO processes.
In the absence of dynamic activation, Eq 6 implies the Ca2+ concentration during each contraction cycle will decay as CCa(t)=SCa1e−KCat. Using this as an input to Eq 5, we find that the radius varies from its baseline value during each contraction cycle as
ΔR(t)=FCaSCa1E1KCa(e−t/tCa−e−t/tmech)(tmech−tCa)
(11)
where we see that the return to equilibrium depends on two characteristic times, one set by the rate of Ca2+ clearance tCa = 1/KCa and the other by the mechanical lag tmech = D/E1 which can include the lag between the concentration increase and force production. As a reference time scale we have selected tCa = 1/KCa = 1 s which is in the range observed by Shirasawa and Benoit [15]. They observed a similar lag between the rise in Ca2+ concentration and the peak force generation. Here we use this lag as an estimate of tmech which we take to incorporate the visco-elastic and chemical-mechanical transduction lags.
While both time constants contribute to the overall response, the slower of the two characteristic times gives the dominant time constant tc that determines the return to equilibrium. Experimental observations of the magnitude of radius change as a function of pressure show that it decreases with increased internal pressure [21]. This phenomenon is reproduced by our model as the vessel wall stiffens (larger E1 in the denominator) at greater radius. The amplitude may be further modified if the myosin cross bridging is length dependent as seen in skeletal muscle [36].
The frequency of contractions at constant NO levels is set by the interplay between the characteristic time for calcium tCa and the magnitude of the random activation term. In addition to the steady source of Ca2+ that establishes the vascular tone, we include a random component ACaRand(R,t) with zero mean and a standard deviation σ that can be applied to either the concentration itself or to the threshold level at which a new release of Ca2+ is triggered. A higher radius leads to more stretch in the LMC membrane and therefore greater sensitivity of ion channels. This can be modeled by increasing the noise level (for example let σ ∝ pm). This will lead to a higher frequency at larger radii as found experimentally [21]. Exponential decay of Ca2+ near equilibrium leads to a latency period between contractions that varies as T = tc log(Cmax/σ) where Cmax is the magnitude of the Ca2+ increment from baseline.
We further investigated random activation with the aid of numerical simulations implemented with the Euler-Maruyama method [37], which properly scales the computational time step with the standard deviation of the noise (Fig 2). The simulation presented in Fig 2d–2f results from a higher mean pressure than Fig 2a–2c. Thus, the baseline radius is larger in Fig 2d–2f, which in turn yields a stiffer wall (larger E1) leading to a smaller mechanical time constant (smaller tmech) and reduced amplitude for change in the radius.
We note that our simple model of stretch-Ca2+ dynamics replicates important features of the contraction cycles observed in vivo [20] where contractions are generally similar to one another in magnitude and duration, but may be separated by inconsistent periods of latency. In Fig 2, we see that the simulated contractions are nearly identical to each other (that is, the trajectories nearly retrace one another in the phase portraits shown Fig 2c and 2f), but occur on inconsistent intervals. Even though the intervals between contractions are not perfectly uniform, they are well estimated by tc log(Cmax/σ). We also see that increased transmural pressure can reduce the interval between contractions by stiffening the wall (pm ↑⇒ R1 ↑⇒ E1 ↑⇒ tmech ↓⇒ tc ↓⇒ T↓) and also by increasing the sensitivity of the Ca2+ channels by stretching the vessel wall (pm ↑⇒ σ ↑⇒ T ↓) yielding higher frequency contractions (compare Fig 2b and 2e).
We also find that our model of the stretch-Ca2+ process readily synchronizes when we impose extrinsic rhythmic pace-making since only small variations in the Ca2+ concentration relative to the threshold level are needed to initiate the next contraction cycle (Fig 3). Such small variations in Ca2+ concentration can be readily introduced by diffusion or voltage signals from adjacent LMCs. Alternatively, the vessel may be locally stretched by lymph arriving from upstream, which can also trigger a local contraction. In this way, neighboring LMCs can synchronize contractions to coordinate flow along a series of lymphangions throughout a connected network of collecting lymphatic vessels [38, 39].
The conditions for oscillations in radius to arise near baseline Ca2+ levels in the absence of sharp spikes in Ca2+ as considered in the previous section are available from linear stability analysis of the shear-NO process near a point R1 which yields
[R˙CNO.]=[−E1DFNOD−SNOE1DR12 sgn(R˙)SNOFNODR12−KNO][RCNO]+[inputs]
(12)
where the inputs include all extrinsic disturbances from the adjacent lymphangions and surrounding tissue. We treat small variations in R parametrically so that the dynamics of the system may be characterized by the eigenvalues of the Jacobian matrix [40] which are roots of the characteristic polynomial:
λ2+(E1D+KNO−sgn(R˙)SNOFNODR12)λ+E1KNOD=0
(13)
Since all of the coefficients are positive, stability requires only that the second term be positive. Thus the system is always stable during contraction (R˙<0). However during dilation (R˙>0) the second term can be positive or negative which allows the system to switch between stability and instability (Fig 4).
Herein is the key feature from which NO can induce spontaneous oscillations in the radius without the sharp spikes in Ca2+ concentration described in Eq 6. If the radius is large enough so that E1/D + KNO > SNOFNO/DR2, then the fixed point is inherently stable. If instead, the radius is small enough that E1/D + KNO < SNOFNO/DR2 then the radius will unstably increase when perturbed. The instability arises because a slight increase in radius (from point 0 on Fig 5) pulls fluid into the lymphangion, increasing shear and temporarily creating a runaway effect wherein more NO is released from the LEC further increasing the radius and drawing in still more fluid (upper branch from point 0 to point 2 on Fig 5b). The instability persists until the radius becomes large enough at point 2 that the shear stresses begin to drop because more cross-sectional area is available for lymph flow. Thereafter the release of NO occurs more slowly than its degradation so that the system can return stably to equilibrium along the lower branch of the trajectory from point 2 to 0. Mathematically, the unstable increase in radius persists until the sign of R˙ changes at point 2. A change in the sign of R˙ does not require that the eigenvalues move to the left half of the complex plane at point B on Fig 4 as required for inherent stability, but rather requires only that the radius increase sufficiently to move the eigenvalues off of the real axis beyond point A, thus permitting at least a partial cycle of oscillation that includes a time at which R˙=0. As the vessel begins to contract, the sign of SNOFNO/DR2 changes at the R-nullcline where R˙=0, leading to an unconditionally stable return to the original radius. The time scale for contraction is approximated by tc ≈ tmech + tNO + tFNO where the three contributions arise from mechanical lag tmech as before, the clearance of NO tNO = 1/KNO, and the rate of force modulation by NO tFNO = SNOFNO/E1KNOR2. To a similar degree of approximation, the instability of the NO-shear dynamics requires tmech + tNO ≤ tFNO. In other words, the change in force elicited by shear stress must persist longer than processes that tend to dissipate its effects. Exact algebraic expressions for the eigenvalues may be employed if desired, but this approximation captures the key dependencies. See Table 2.
The NO cycle can be generalized into a controllable and synchronizable oscillator. Fig 6 shows the behavior of the NO cycle in response to small random disturbances. During the stable contraction process, the vessel remains refractory to disturbances until close enough to equilibrium for a random disturbance to trigger another cycle, much as we found with the stretch-Ca process. Here the period of the NO-induced oscillations varies as tc log(Cmax/σ) as before but tc and Cmax now refer to NO rather than Ca2+. As with Ca2+, a relatively quiet environment or reduced sensitivity to disturbance will elicit longer latency periods between cycles but will not significantly change the shape of the shear-NO cycle.
The NO dynamics can also readily synchronize with externally-imposed, small-amplitude sinusoids (Figs 7 and 8). Fig 7b and 7c show how the radius oscillates at precisely the input frequency for frequencies reasonably close to the response when noise triggered (Fig 6). However, when the input frequency is too high (Fig 7a) or too low (Fig 7d) synchronization occurs, but at half or double the input frequency, respectively. Fig 8 shows a parametric study of synchronization over a wide range of input frequencies and amplitudes where we find that synchronization can include a variety of integer ratios between input and output frequencies as explained further in the Discussion.
Having explored the system dynamics when Ca2+ and NO are taken to be constant relative to each other we now consider their combined, dynamic effects. Our model includes three possible interactions reported in the literature: NO may i) desensitize the LMCs to Ca2+ as modeled by Eq 10, ii) modify the availability of Ca2+ by 1/(1 + γCNO) or iii) speed clearance of Ca2+ by 1 + βCNO [41, 42].
Our simulations (Fig 9) show that the dynamic effects of NO are most pronounced when the shear-NO dynamics are unstable. When the shear-NO dynamics are unstable, the radius can overshoot the nominal radius before or after a Ca2+-induced contraction, yielding oscillations in radius that are more symmetrical about equilibrium than when shear-NO is stable. At marginal stability (Fig 9d), the NO concentration rings at a frequency determined by the point where the eigenvalues of the shear-NO oscillator cross the imaginary axis f = (E1KNO/D)1/2/2π. At larger radii, the shear-NO mechanism is inherently stable, but can still reduce the magnitude of the oscillations driven by the stretch-Ca2+ process. This process is important in the presence of an assisting pressure gradient because the dilation induced by the forced flow can put the vessel into the range of radii where the shear-NO mechanism can inhibit contractions that would otherwise tend to restrict free flow through the vessel.
The overall frequency is set by a complex interplay of stretch-Ca2+ and shear-NO mechanisms, but will typically be dominated by the faster of the two processes. Long latency intervals between Ca2+-induced contractions can permit NO to produce an unstable dilation, whereas, short intervals due to Ca2+ can suppress the autonomous oscillations possible through the NO mechanism. Interestingly, the published clearance rates for Ca2+ and NO cover a wide enough range that either possibility exists in vivo [15, 32].
Experimental observations of diameter in vivo show cycles consistent with the model predictions (Fig 10) (data from [20]). In the absence of direct measurements of concentrations, we employ an alternative phase portrait of diameter plotted against the rate of change of diameter. Fig 10a and 10b are from a wild-type mouse in which Ca2+ and NO effects can operate normally. Here, we observe complex oscillations that include both rapid contractions and occasional strong dilations above the baseline diameter as expected from the shear-NO mechanism. In contrast, when the NO effects have been genetically deleted in eNOS-/- mice in Fig 10c and 10d, we see wave forms that are nearly identical to each other but dominated by contraction with the dilatory effects of NO appearing to be substantially weakened. In all cases, we see cycles occurring on irregular intervals as we expect from noise-triggered oscillators.
While the correspondence between the experiments and the model is encouraging, we should not expect a model of an isolated lymphangion to reproduce all features of a vessel in an intact, in vivo vascular network. For example, the effects of flow introduced from upstream or disturbances from surrounding tissue are inputs present in the animal, but are not included in the modeled dynamics. We have also not yet included effects due to nonlinear valve efficiencies or the bias of the check valves toward the open position [43]. Nonetheless, phase portraits, such as those newly employed here, promise to assist further study of the nonlinear dynamics that govern vascular oscillations.
While our results await further experimental validation and improved estimates of key parameter values, it is interesting to consider the newly identified shear-NO oscillator more generically from a nonlinear dynamics perspective. The shear-NO oscillator has some important similarities and differences from the well-known Van der Pol oscillator [44] which has the form
x¨−A(1−x2)x˙+x=g(t)
(14)
The form of the characteristic polynomial in Eq 13 implies that the shear-NO oscillator may be written as
x¨+A(1−sgn(x˙)B/x2)x˙+(x−x1)=g(t)
(15)
where the generic variable x fills the role of the radius in the shear-NO model, x1 is the nominal operating point and g(t) is a forcing function that can include steady, random or periodic components.
The stability-determining second term in both oscillators can change sign based on the magnitude of the variable with a positive second term implying stability. The Van der Pol oscillator is known to self-sustain oscillations about the origin in phase space (x,x˙)=(0,0) when g(t) = 0, as its second term changes sign during different phases of each cycle. In contrast, the shear-NO oscillator operates near (x,x˙)=(x1,0), but with x > x1. Therefore, the second term in Eq 15 can be either i) always be positive (x12>B) regardless of the sign on x˙ implying inherent stability or ii) can be conditionally positive depending on both the magnitude of B and the sign of x˙. As a result, the shear-NO oscillator cannot produce self-sustained oscillations for large radius (x12>B). Furthermore, even when the radius is sufficiently small (x12<B), the radius will return unequivocally to equilibrium as long as x˙<0 unless a non-zero forcing function is present to change the sign of x˙. However, we find that when x12<B the magnitude of the forcing needed to start a new cycle can be arbitrarily small and in the form of either random noise or a periodic stimulus provided that enough time has passed for the system to approach its equilibrium point.
In the context of the shear-NO dynamics, the key to oscillations is the inverse dependence on radius for the NO source due to shear stress in Eq 7. As long as the exponent on R−2 remains negative (increasing radius leading to lower shear stress and less NO production), then the NO-shear mechanism will be capable of a mathematical transition from unstable to stable as seen above in the generic oscillator in Eq 15. The physiological impact of this result then depends on the relative magnitude of the time scales identified herein, not on any single parameter value. For example, the stability of the NO-shear mechanism depends on groups of parameters such as tFNO = SNOFNO/E1KNOR2, which combines the sensitivity of the vessel to shear stress, the contractile force, the wall stiffness and the NO clearance rates.
Inputs of constant magnitude have the effect of adjusting the equilibrium point. Using the shear NO oscillator as an example, an increase in transmural pressure will dilate the vessel, as will a pressure gradient that assists flow by inducing NO production via a steady shear stress. Likewise, a steady source of NO from local inflammation will chronically dilate the vessel [20]. If the vessel becomes sufficiently large, the stability criterion found above suggests that the shear-NO process will not support self-sustaining oscillations, in part due to the direct effect of radius on the stability criterion, but also due to greater stiffness of the wall at larger radius (higher E1). Nonlinearities in the force production and chemical source/elimination terms may also alter the stability in similar ways.
Numerical simulation and examination in the phase plane reveal that the stretch-Ca2+ and shear-NO processes possess numerous symmetries that offer intriguing possibilities when the processes act together (Figs 2 and 6, Table 3). Most notably, we see that the shear-NO process produces rapid and unstable dilation toward a larger radius, followed by stable contraction, while the stretch-Ca2+ process causes the vessel to contract rapidly and unstably toward a smaller radius and then to dilate stably. An essential feature of both the Ca2+ and NO mechanisms is that taken separately they do not produce traditional, self-sustaining limit cycles, but instead have a one-sided stability near equilibrium from which a new cycle begins only with a perturbation from the local environment. Interestingly, a suitable trigger for the stretch-Ca2+ oscillator can be an increase in radius produced by the shear-NO mechanism. And conversely, the shear-NO oscillatory can be triggered by a contraction arising from the stretch-Ca2+ mechanism. Balanov et al [45] reviews a variety of similar, so-called “noise-induced” oscillators in contexts outside of lymphatic physiology such as neurons and electrical monovibrators, but to our knowledge, the coupling of symmetric, noise-induced oscillators described in the present study has not been previously investigated.
Balanov et al [45] also review how nonlinear oscillators can synchronize with small-amplitude sinusoidal inputs. Here we found that synchronization of either oscillator can occur over a wide range of frequencies (shear-NO shown in Fig 8, similar behaviors for stretch-Ca2+ acting alone and in combination with shear-NO can be observed). The synchronization behavior seen here is similar to that for the forced Van der Pol oscillator in its ability to produce so-called Arnold tongues which are broad domains within which the input and output frequencies are locked in ratios of m:n where m and n are small integers [44, 46].
Kornuta et al [47] recently showed that lymphatic vessels studied ex vivo synchronize their contractions in a 1:1 fashion with imposed oscillatory variations in shear stress when the amplitude of the stimulus is sufficient large and the frequency of the input is relatively close to the autonomous frequency. Interestingly, they also observed that small amplitude variation in transmural pressure did not yield 1:1 frequency locking. However, our examination of their results (Fig 8 in [47]) suggests that 2:3 locking may have occurred. In the absence of imposed flow, they also found that the vessel continued to contract, but with a lower and more erratic frequency consistent with our simulated noise-triggered oscillator (Figs 2 and 3) in the absence of the shear-NO mechanism. Ohhashi et al [48] also examined sinusoidal variations in transmural pressure at frequencies well away from the spontaneous frequency. Here too, 1:1 frequency locking did not arise, but the frequency of the contractions responded strongly to the input waveform. Given the subtlety of identifying non-1:1 synchronization, further examination of the experimental record may be warranted.
In conclusion, we have presented a model of a vascular oscillator. The present analysis is sufficiently general to point toward several features that are likely found in other systems. The linear stability analysis shows: (i) complementary mechanisms for dilation and contraction of collecting lymphatic vessels, (ii) a fast, unstable process that recovers slowly and stably to a one-sided equilibrium, (iii) disturbance-based triggering that facilitates either synchronization with a cyclic pacemaker or spontaneous oscillations from random disturbances and (iv) the capability for reciprocal modulation between contractile and relaxation effects. Those features are not only limited to the presented example of Ca2+ and EDRFs but can be extended into other fields. The ability of the Ca2+ and NO based oscillators to respond to each other and external stimuli explains how lymphatic pumping can be coordinated along extended lengths of collecting lymphatic vessels without the need for higher order coordination. This new class of coupled, noise-driven oscillator can help to explain the diverse pumping behavior of lymphatic vessels.
|
10.1371/journal.pntd.0005764 | Unraveling the genetic diversity and phylogeny of Leishmania RNA virus 1 strains of infected Leishmania isolates circulating in French Guiana | Leishmania RNA virus type 1 (LRV1) is an endosymbiont of some Leishmania (Vianna) species in South America. Presence of LRV1 in parasites exacerbates disease severity in animal models and humans, related to a disproportioned innate immune response, and is correlated with drug treatment failures in humans. Although the virus was identified decades ago, its genomic diversity has been overlooked until now.
We subjected LRV1 strains from 19 L. (V.) guyanensis and one L. (V.) braziliensis isolates obtained from cutaneous leishmaniasis samples identified throughout French Guiana with next-generation sequencing and de novo sequence assembly. We generated and analyzed 24 unique LRV1 sequences over their full-length coding regions. Multiple alignment of these new sequences revealed variability (0.5%–23.5%) across the entire sequence except for highly conserved motifs within the 5’ untranslated region. Phylogenetic analyses showed that viral genomes of L. (V.) guyanensis grouped into five distinct clusters. They further showed a species-dependent clustering between viral genomes of L. (V.) guyanensis and L. (V.) braziliensis, confirming a long-term co-evolutionary history. Noteworthy, we identified cases of multiple LRV1 infections in three of the 20 Leishmania isolates.
Here, we present the first-ever estimate of LRV1 genomic diversity that exists in Leishmania (V.) guyanensis parasites. Genetic characterization and phylogenetic analyses of these viruses has shed light on their evolutionary relationships. To our knowledge, this study is also the first to report cases of multiple LRV1 infections in some parasites. Finally, this work has made it possible to develop molecular tools for adequate identification and genotyping of LRV1 strains for diagnostic purposes. Given the suspected worsening role of LRV1 infection in the pathogenesis of human leishmaniasis, these data have a major impact from a clinical viewpoint and for the management of Leishmania-infected patients.
| Leishmaniasis is a well-known parasitosis due to an infection by the protozoan Leishmania parasites firmly established in South America. In French Guiana, where leishmaniasis is a public health problem, having an annual incidence of 5.6 cases/10,000 inhabitants, 80% of Leishmania spp. parasites are infected by an endosymbiotic virus, Leishmania RNA virus 1 (LRV1). The purpose of this study was to gain insights on the genetic variability and evolution of LRV1 at the genomic level. We subjected 20 Leishmania isolates obtained from cutaneous lesions with next generation sequencing and de novo sequence assembly. This allowed generating 24 LRV1 full-length coding sequences presenting among themselves a significant genetic diversity. The inferred phylogenetic relationships enabled to identify distinct well-supported genotypes and support the hypothesis of co-evolution between LRV1 strains and their parasite hosts. In addition, we identified multiple LRV1 infections in three parasite isolates. Based on these data, future characterization of new strains from other geographic areas and other parasite species should extend knowledge of LRV1 diversification processes. Finally, these results allowed us to identify genomic regions where best to allocate resources for subsequent analyses of LRV1 viral diversity and genotyping that could serve for accurate routine molecular diagnostic applications.
| Protozoan parasites of the genus Leishmania are unicellular eukaryotes with a complex digenetic cycle. In the vertebrate host, they are obligatory intracellular parasites. They cause a broad spectrum of diseases, collectively known as leishmaniases, that occur predominantly in tropical and subtropical regions [1]. Given their frequency and the severity of certain clinical forms, these diseases represent a major public health problem in endemic countries. Depending on their behavior in the vector’s gut, parasites of the Leishmania genus are divided into two subgenera: Leishmania and Viannia. The Leishmania subgenus Viannia, endemic in Latin America, is the etiological agent of cutaneous (CL) and mucocutaneous leishmaniasis (MCL) [2]. Within this subgenus, L. (V.) guyanensis and L. (V.) braziliensis are hosts of a double-stranded RNA virus called Leishmania RNA Virus (LRV).
Leishmania RNA viruses are small, nonenveloped double-stranded RNA viruses [3]. LRV (genus Leishmaniavirus) are members of the Totiviridae family, along with several other groups of protozoal or fungal viruses, including Giardia lamblia virus (GLV), Trichomonas vaginalis viruses and Eimeria spp. viruses [3–9]. They have a nonsegmented genome approximately 5 kb (4.9–5.3 kb) in length with two long, partially overlapping, open reading frames on the positive strand encoding the capsid/coat protein (CP; orf2) and the RNA-dependent RNA polymerase (RdRp; orf3) and a small 5’-proximal potential orf1 [10]. LRV orf1 presents similarities with orf1 of ScV-L-A virus, a totivirus of the yeast Saccharomyces cerevisiae that encodes a toxin, but its function remains unsolved. LRV genomic RNA is flanked at its extremities by 5’ and 3’ untranslated regions (UTR). It has been shown that the 5’UTR promotes internal initiation of translation with the presence of an internal ribosomal entry site (IRES) [11]. The 5’UTR also possesses five predicted stem-loop structures (I to V) and a consensus cleavage sequence [12–14]. Stem-loop IV and the consensus cleavage site represent the minimal essential components for the viral capsid-dependent RNA cleavage that might play a regulatory role for maintaining persistent infection of host cells [13, 15].
The presence of virus-like particles in Leishmania parasites was first reported in the early 1970s [16]. Since then, two types of LRV have been identified: LRV1 carried by some New World Leishmania species and LRV2 by Old World species [3, 8, 17]. To date, LRV1 was detected in Leishmania parasites from Colombia, Brazil, Peru, Bolivia, Suriname and French Guiana, all originating in the Amazon basin [18–22]. LRV2 was first isolated from L. hertigi, a non-human parasite species [16]. It was then found in a single isolate of L. major and, more recently, in strains of L. aethiopica isolated from biopsies of cutaneous leishmaniasis patients in the Ethiopian highlands as well as in two different Leishmania isolates from Iran belonging to the L. infantum and L. major species [17, 23, 24]. At the time of their discovery, arbitrary identifiers were given to type 1 Leishmania RNA viruses showing a certain degree of sequence conservation on the basis of hybridization analysis, namely LRV1-1 to LRV1-12 [20]. Two additional types, LRV1-13 and LRV1-14, were then described based on sequence comparison of fragments of two genomic regions, i.e., the 5’UTR and orf3 regions [25]. Several subtypes were isolated from L. (V.) guyanensis, LRV1-(1, 4, 5, 6, 7, 8 and 9), others from L. (V.) braziliensis, LRV1-(2, 3, 13 and 14) and untyped Leishmania spp., LRV1-(10, 11, 12). In addition to L. (V.) guyanensis and L. (V.) braziliensis, Cantanhêde et al. recently reported the identification of LRV1 in L. (V.) lainsoni and L. (Leishmania) amazonensis parasites [26]. So far, no LRV1 has been detected in other L. (V.) species and none in L. (L.) species [18, 21]. These results, proving the wide distribution of LRV1 in the different New World parasite species, combined with the lack of an infectious phase for these viruses, suggested that LRV1 arose prior to the divergence of New World parasites and further supported the theory that LRVs are ancient viruses [25]. Nevertheless, most of these viruses have only been characterized on short stretches of sequences a few hundred base-pairs long and, depending on the studies, different regions of the genome have been amplified. Complete cDNA sequences are only available for two New World virus isolates (LRV1-1 and LRV1-4), one LRV2-1 isolated from L. major and three more LRV2-1 strains from L. aethiopica [10, 12, 23, 27]. The LRV1-1 and LRV1-4 entire genomic sequences share 77% overall nucleotide identity.
Little is known on the biology of LRV, which persistently infects Leishmania parasites and is unable to produce extracellular infectious particles. Nevertheless, LRV1 infection of parasites seems to affect their virulence. Indeed, Ives et al. showed that metastasizing parasites have a high LRV1 burden that is recognized by host Toll-like receptor 3 (TLR3) to induce proinflammatory cytokines and chemokines in murine models [28]. They also showed that LRV1 in the metastasizing parasites subverted the host immune response to Leishmania, promoting parasite persistence and affecting treatment efficiency. We and others recently showed that LRV1 status in L. (V.) guyanensis- and L. (V.) braziliensis-infected patients was significantly predictive of first-line treatment failure and symptomatic relapse, and might stand to guide therapeutic choices in acute cutaneous leishmaniasis (ACL) [29, 30]. However, other studies reported contradictory results with respect to the association between the severity of leishmaniasis (cutaneous vs. mucocutaneous) and the presence of LRV1 [19, 22, 26, 31, 32].
In French Guiana, where CL is a public health problem with an average of 127 diagnosed cases per year between 2006 and 2013, oscillating between 96 and 160 cases per year, five coexisting Leishmania parasite species are known to infect humans: L. (V.) guyanensis, L. (V.) braziliensis, L. (L.) amazonensis, L. (V.) lainsoni and L. (V.) naiffi [33, 34]. Among them, L. (V.) guyanensis and L. (V.) braziliensis are the two predominant species, accounting for more than 85% and about 10% of CL cases, respectively [33]. The three other species are only occasionally diagnosed [33, 35]. For L. (V.) guyanensis and L. (V.) braziliensis, patients present a broad spectrum of clinical manifestations (nodular, ulcerative, disseminated and mucous forms). In addition, symptomatic relapses and treatment failures are frequently seen in L. (V.) guyanensis infections during which patients variously respond to first-line anti-leishmanial treatments and are more prone to developing chronic CL. Factors underlying disease evolution are yet to be fully understood. We recently reported that 74% of Leishmania spp. isolates collected between 2011 and 2014 were LRV1-positive [21]. LRV1 was detected in 80% (90/112) of L. (V.) guyanensis isolates, 55% (6/11) of L. (V.) braziliensis and in none of the (0/6) L. (L.) amazonensis isolates. Considering the paucity of sequence data available for LRV1, because of the many questions that arise about its origin and distribution, we were interested in gaining insight into the genetic diversity, at the genomic level, and the evolution of this virus in the circulating isolates of Leishmania spp. in French Guiana. To this end, using Illumina deep-sequencing technology and de novo sequence assembly, we generated and analyzed LRV1 sequences derived from Leishmania parasite cultures obtained from skin lesions of 20 patients suffering of ACL, i.e., with lesions present for less than 6 months. Polymorphism and phylogenetic analyses were then performed to measure LRV1 genome-wide diversity. Here we report the full-length coding sequences of 24 LRV1 strains and use them to make the first-ever estimate of LRV1 viral diversity that exist in Leishmania (V.) guyanensis. These results represent the largest number of LRV1 full-length coding sequences published to date and the first time that in some parasite strains multiple viral infections have been identified.
The Leishmania isolates used in this study were obtained from a previously published study [21]. Briefly, these isolates had been successfully cultured from biopsies collected between 2011 and 2014 from 20 adult patients (16 males and four females; 19–66 years old) diagnosed with ACL and enrolled in the study before treatment with pentamidine isethionate (Pentacarinat; Rhone Poulenc) (Table 1). All patients, in consultation for a suspicion of leishmaniasis at the dermatology unit of the Cayenne hospital (Centre Hospitalier Andrée Rosemon, CHAR), Cayenne, French Guiana, were informed during consultation by the clinician that case records and biological data might be further used in research and that they had the right to refuse. Biological samples were taken for diagnostic purposes and oral informed consent was documented in the case records by the clinician during consultation. Patient data was anonymized at the Laboratory of Parasitology and Mycology of the Cayenne hospital, which carried out diagnostic analyzes and initial parasite culture. The project did not raise any concerns and was approved by the Ethical Committee at Cayenne Hospital. Ethical approval was granted based on the human experimentation guidelines of the “Comité consultatif sur le traitement de l'information en matière de recherche” (CCTIRS: 2012–42). The monocentric audit of retrospective anonymized case record data was permitted by the “Commission nationale de l’informatique et des libertés” (CNIL: DR.2014-091) as well as the regulations of Cayenne Hospital (http://www.ch-cayenne.net/Droits-et-Devoirs.html).
All 20 isolates of Leishmania spp. had formerly been tested for their LRV1 status by RT-PCR and Leishmania spp. were identified using PCR-RFLP, as previously described (Table 1) [21, 36].
Parasites were cultured in Schneider’s medium (Sigma-Aldrich) complemented with 10% heat-inactivated fetal calf serum (FCS) (Gibco), 0.6 mg/L Biopterin (Sigma-Aldrich) and 5 mg/L folate hemin (Sigma-Aldrich) without antibiotics. Dry pellets were constituted by centrifugation 5 min at 2500 rpm and removal of supernatant. The pellets were stored at −80°C until RNA extraction.
For parasite cloning, 0.5 mL of culture medium containing 2–4.107 parasites was spread onto freshly prepared Schneider plates (Schneider’s medium, 10% FCS, Biopterin 0.6 mg/L and folate hemin 5 mg/L, 2% low-melting-point agarose). After 7 days, the colonies were harvested and cultured separately in tubes containing complemented Schneider’s medium for an additional 5–7 days. Separate cultures were then processed for RNA isolation and conventional RT-PCR using primer pairs specific to each virus (see below).
Total RNA was extracted from dry pellets of parasites (108 parasites) using the RNeasy mini kit (QIAGEN) as recommended by the manufacturer, treated with Turbo DNase (Life Technologies) for 1 h at 37°C and then measured on a Biophotometer (Eppendorf). Then the generation of double-stranded cDNA and SPIA amplification was performed from 50 ng of total RNA using the Ovation RNA-Seq system V2 (NuGEN Technologies, Inc.) as specified by the manufacturer’s protocol.
The libraries were generated using the NextFlex PCR-free DNA Seq Kit (Bioo Scientific, Austin, TX, USA) with (15 strains, with a two-letter and two-number code) or without (five strains, with a 20XX code) a 10-cycle PCR enrichment before quantification and validation. They were then sequenced on an Illumina MiSeq instrument in 250-base paired-end reads (Illumina, San Diego, CA, USA). Sequence files were generated using Illumina Analysis Pipeline version 1.8 (CASAVA).
Raw sequences were filtered and trimmed using fastq-mcf with the following parameters: duplicates were removed when 50 bases were identical between reads (-D parameter) while minimum read length was set to 50 [37]. Adapter sequences were removed using -t (% of occurrence) set to zero. The other parameters were left to their default settings. Then Deconseq was used to eliminate phiX174 phage sequences [38]. Fastq clean files were assembled using SPAdes (version 3.0.0) with the following range of k-mers: 21, 55, 77, 99 and 127 [39]. We used blastn for sequence comparison between assembled contigs and reference LRV1 genomes (LRV1-1 and LRV1-4, accession numbers M92355 and U01899, respectively) [6]. The LRV1-positive contigs were then extracted and used as references for mapping using the bwa mem algorithm (version 0.7.10-r789) with a kmer set to 55 and a minimum alignment score of 40, using a minimum fastq quality of 30 [40]. Some assembly abnormalities were manually corrected. The contigs and known LRV genomes were then multiple-aligned with MAFFT alignment software (version 7.037b) [41].
Conventional methods were used to extend sequences when needed. Three sets of degenerate consensus primers were designed according to the NGS sequences obtained. Primers were degenerated so as to amplify products from all samples regardless of the sequence. Full information on the primers is available in Table 2. These sets of primers were used in RT-PCR reactions carried out using the Transcriptor one-step RT-PCR kit (Roche), as recommended by the manufacturer, using 500 ng of input total RNA. RT-PCR products at the expected size were sent to Beckman Coulter Genomics (http://www.beckmangenomics.com/) for direct sequencing. Furthermore, to specifically amplify the different viruses in case of multiple-infected Leishmania isolates, one set of nondegenerate specific primers per virus was designed and used for RT-PCR (Table 2).
Contigs from HTS as well as raw sequences downloaded from the Beckman Coulter Genomics website were analyzed and edited in MEGA 5.05 [42]. Multiple sequence alignments were constructed using ClustalW with all published complete and partial LRV1 sequences. Alignments were checked manually. Sequences were translated into amino acids and both nucleotide and amino acid sequences were checked for irregularities. For phylogenetic trees inferred from the aligned nucleotide sequences, the MrModeltest2.3 program was used to determine the optimal model of nucleotide evolution [43]. The GTR model, with a gamma distribution shape parameter (G) and invariable sites (I), was identified and used for the Bayesian approach, which was performed with Mr. Bayes 3.2.2 to infer phylogenetic relationships [44]. Markov Chain Monte Carlo (MCMC) simulations were run for 10,000,000 generations, with four simultaneous chains, using a sample frequency of 100 and a burn-in of 25,000. Majority rule consensus trees were obtained from the output. Validation of the inference was assessed based on the standard deviation of split frequencies, which was less than the expected threshold value of 0.01. For phylogenetic trees inferred from the aligned amino acid sequences, the ProtTest3 program was used to determine the optimal model of amino acid evolution. The JTT model, with gamma (G) distribution, was identified and used for the Bayesian approach [45], which was performed with Mr. Bayes 3.2.2 [44]. MCMC simulations were run for 10,000,000 generations, with four simultaneous chains, using a sample frequency of 100 and a burn-in of 25,000. Majority rule consensus trees were obtained from the output. Validation of the inference was assessed based on the standard deviation of split frequencies, which was less than the expected threshold value of 0.01.
LRV1 strains were named according to a recent proposal to the International Committee on Taxonomy of Viruses, i.e., “LRV1” followed by two letters corresponding to the parasite species to which the strain designation was then assigned [46]. For example: LRV1-Lg-LF94 for LRV1 from L. (V.) guyanensis strain LF94 (Table 1).
Sequences reported in this paper were deposited in the GenBank nucleotide database under accession numbers KY750607 to KY750630. The raw sequencing data are available in the Sequence Read Archive under accession number PRJNA371487 (www.ncbi.nlm.nih.gov/bioproject/371487).
In the context of epidemiological screening, we tested 129 isolates of Leishmania sp. Collected in French Guiana between 2011 and 2014 for the presence of LRV1 [21]. Here, we analyzed a subset of previously identified LRV1-positive isolates obtained from patients diagnosed with ACL. A total of 20 parasite cultures, obtained from human biopsies isolated from 16 men and four women, aged 19–66 years with a median age of 35 years, were selected for high-throughput sequencing. The data (gender and age) collected for each patient are listed in Table 1. In all, we sequenced 19 LRV1-positive L. (V.) guyanensis isolates and one LRV1-positive L. (V.) braziliensis isolate.
Using Illumina MiSeq sequencing technology, 24 full-length or almost full-length genomic sequences were de novo assembled into individual contigs with a mean 49.8 depth of coverage. Sequences were 4.8–5.3 kb long. Alignment of the full-length genome sequences obtained with those available in the database made it possible to design three sets of consensus-degenerate primers (Table 2). These primer pairs were used for conventional RT-PCR and Sanger sequencing to extend the sequences to the 5’ and 3’ ends that were missing for some strains. High-throughput sequencing combined with conventional RT-PCR applied to the 20 LRV1-positive isolates generated almost complete genomic sequences (from nucleotide 60 to 5248 of LRV1-1, accession number NC_002063) from each viral strain. This covers the full-length coding sequence of the virus. In addition, Sanger sequencing of the RT-PCR products also allowed verifying the correctness of the sequences obtained. A total of 24 different viral sequences were obtained from the 20 LRV1-positive parasite isolates tested. All sequences were 5.2 kb long with an average G+C content ranging from 44.7% to 46.6%. Detailed genome statistics are outlined in Table 1. Unexpectedly, co-infections by two or three LRV1 viruses were identified in three parasite isolates, 2028, WF69 and XJ93. LRV1 full-length coding sequences obtained from the same isolate were named 2028G1, 2028G2 and 2028G3, WF69G1 and WF69G2, XJ93G1 and XJ93G2.
The 24 sequences obtained had the same structural organization as partially overlapping open reading frames surrounded by untranslated regions similar to the sequence already described [10, 12]. In addition, they showed an identical genome size. All but one sequence (strain LF94) showed a double-nucleotide insertion in the 5’UTR region at positions 201–202 relative to LRV1-1 (numbers given here and below refer to the nucleotide positions in the published sequence of LRV1-1, acc. # NC_002063). At this exception of this shared insertion, only a few other nucleotide insertions and deletions were seen in the 5’UTR region of four viral sequences relative to LRV1-1. Thus, strain 2014 exhibited one nucleotide deletion at position 363, strain 2001 possessed two nucleotide insertions at positions 142 and 198 (these two insertions were also present in LRV1-2, LRV1-8 and LRV1-9 5’UTR sequences), strain XJ93G2 showed a double-nucleotide insertion at position 198–199 and one nucleotide deletion at position 446, while strain YA70 presented one nucleotide insertion at position 198 and one deletion at position 393. In addition to these indels in the 5’UTR, of the 24 sequences, only the coding sequence of the RNA-dependent RNA polymerase gene of the YA70 strain presented one codon insertion at positions 73–75 (D25), one codon deletion at positions 169–171 and a double codon deletion at positions 493–498. These codon indels did not alter the reading frame. YA70 shared the first amino acid insertion at position 25 (aspartic acid or lysine, respectively) and the first codon deletion with strain LRV1-Lb2169 (international code MHOM/BO/2011/2169, acc. # KC862308). These two strains were identified from L. (V.) braziliensis parasites. The YA70 strain also possessed a double stop codon at the end of the coat and RNA-dependent RNA polymerase genes. Three other strains also possessed a double stop codon at the end of the RNA-dependent RNA polymerase gene: LRV1-1, LF94 and 2001. Except for YA70, no deletions or insertions were detected in the coding sequences of the other strains. According to these results, orf2 was 2229 bp in length and orf3 was 2631–2637 bp, encoding a capsid protein of 742 aa and a RNA-dependent RNA polymerase of 876–878 aa, respectively.
The almost complete genome sequences obtained for each strain as well as the nucleotide and amino acid sequences of the capsid and RNA-dependent RNA polymerase were compared to one another and to those published in the databases. All sequences were unique. On the basis of the common ≈ 5260 bp sequences obtained for each strain, pairwise comparisons showed that nucleotide sequence identities ranged from 76.5% (2001 vs. LRV1-4) to 99.5% (PD46 vs. VL91). CP coding sequences exhibited among themselves from 76.3% (Lg1398 vs. LRV1-Lb2169) to 99.7% (2008 vs. LF98) nucleotide identity and from 89.5% (LRV1-4 vs. LRV1-Lb2169) to 100% (PD46 vs. VL91 and 2008 vs. LF98 and 2028G1 vs. LL28/LV11) amino acid identity. For RdRp coding sequences, the nucleotide identities ranged from 72.4% (2001 vs. LRV1-Lb2169) to 99.4% (PD46 vs. VL91) and the amino acid identities ranged from 79.9% (2001 vs. LRV1-4 vs. LRV1-Lb2169) to 99.5% (2008 vs. LF98). Based on the genetic distance (Table 3) and phylogenetic relationship results (see below), six clades were clearly identifiable. We tentatively called them clades A–F. With the exception of clade A sequences for which the nucleotide divergence ranged from 0.3 to 10.4%, the intra-clade variability of all clades was below 10% at both the nucleotide and amino acid levels. Moreover, except for the A-B inter-clade variability, which ranged from 95.5 to 98.2% on CP and from 92.3 to 94.9% on RdRp on amino acid sequences, the inter-clade variability was over 10% at the nucleotide and amino acid levels. It is noteworthy that the two LRV1 sequences obtained from isolate WF69, WF69G1 and WF69G2, had a 91.8% nucleotide identity. Strains 2028G2 vs. 2028G3 gave the same percentage of nucleotide identity (91.8%), while 2028G1 vs. 2028G2 and 2028G1 vs. 2028G3 showed 86.9 and 87.4% nucleotide identity, respectively. Finally, strains XJ93G1 vs. XJ93G2 exhibited 83.4% nucleotide identity. Nucleotide substitutions occurred throughout the sequences.
5’UTR sequences (from nucleotide 60 to 450) exhibited 86.2 (YA70 vs. 2001) to 100% (VL91 vs. YZ58 and LF98 vs. 2008) nucleotide identities between them and with the LRV1-1 and LRV1-4 5’UTR sequences. Of interest, it should be noted that YA70, which corresponded to the only sequence obtained from a L. (V.) braziliensis isolate, was the most divergent sequence. If we exclude it from the data set, the nucleotide sequence conservation between LRV1 5’UTR sequences from L. (V.) guyanensis isolates ranged from 88.8 to 100%. The same analysis restricted to the LRV1 strains from L. (V.) guyanensis from French Guiana (excluding strain 2001, for which the geographic location of contamination was reported as the region of Manaus, Amazonas state, Brazil) demonstrated a nucleotide sequence conservation ranging from 91.1% to 100%. In addition, taking into account all the sequences analyzed, 74.7% of the 5’UTR nucleotide positions were identical. By excluding YA70, this percentage increased to 79.1%. Again, by restricting our analysis to LRV1 5’UTR sequences obtained from L. (V.) guyanensis strains from French Guiana, 82.1% of the nucleotide positions were identical. These 100% conserved nucleotide positions were distributed into blocks of different sizes. The longest one, a 50-bp motif, covering nucleotides 313–362, encompassed the previously identified pyrimidine-rich region (333-UUUCUUGUUACUAUU-347), whose first nine nucleotides are complementary to a purine-rich region of the Leishmania 18S rRNA [47], as well as the nucleocapsid endoribonuclease cleavage site 314-GAUCˇCGAA-321 [13, 14]. In addition, analysis of the stem-loop IV sequences located upstream, from nucleotide 266 to 281, showed that all but two sequences were 100% conserved and identical to the published one [15]. The only two divergent sequences corresponded to strains 2001 and XJ93G2. Strain XJ93G2 presented two complementary substitutions at positions U269C and A278G of the stem, while strain 2001 presented one substitution at position A278G able to form a wobble base pair with the uracil at position 269 (all positions are given relative to LRV1-1 sequence, accession number NC_002063). This shows that LRV1 5’UTR sequences are strongly conserved in both nucleotide sequences and predicted RNA secondary structures.
Phylogenetic analyses were done on the almost-complete genome sequences (nucleotides 60–5248 relative to the LRV1-1 sequence) (S1 Fig), as well as on either nucleotide or amino acid sequences of the capsid (Fig 1A) and RNA-dependent RNA polymerase (Fig 1B). They all gave strictly identical tree topologies. As mentioned above, we identified six monophyletic “clades”, A–F, all supported by high posterior probabilities. Most of our strains clustered in clades A and B. Indeed, nine of the strains, with LRV1-4 and LRV1-LgM5313, belonged to the monophyletic clade A while clade B was composed of 11 of our strains. All LRV1 strains belonging to clades A and B were from L. (V.) guyanensis parasites. Clades A and B belonged to a monophyletic lineage supported by a posterior probability value of 1. In addition, in clade A, LRV1-LgM5313 and LRV1-4 sequences formed a monophyletic well-supported branch in clade A (Fig 1A). These two sequences correspond to Brazilian L. (V.) guyanensis isolates, whereas the others were from French Guiana. Clade C contained only a new divergent strain, XJ93G2, which is well supported phylogenetically in all analyses, while clade D was composed of strains 2001 and LRV1-Lg1398. In addition, a phylogenetic analysis restricted to a common 219-bp fragment of the 5’UTR region (S2 Fig), which allowed including sequences from most of the LRV1-1 to LRV1-13 strains (except LRV1-3, -5, -6 and -12 for which no sequence is available), demonstrated that strain 2001 was close to LRV1-8 and LRV1-9 strains. These four strains were identified from L. (V.) guyanensis isolates from Brazil. Clade E comprised two strains, one of our sequences, LF94, and LRV1-1. Finally, YA70 and LRV1-Lb2169, identified from L. (V.) braziliensis strains, belonged to a separate clade, F, distinct from the other clades composed of LRV1 sequences of L. (V.) guyanensis (Fig 1). Interestingly enough, the two viral sequences identified from WF69 belonged to two distinct groups of sequences within clade A, two of the three strains from 2028 (2028G2 and 2028G3) were also part of two distinct groups of sequences within clade A, while the third sequence, 2028G1, belonged to clade B. Finally, XJ93G1 belonged to clade B and XJ93G2 was the only representative of clade C.
The last phylogenetic analysis, restricted to partial capsid sequences (299bp/99aa), allowed including sequences recently published by Adaui et al. obtained from L. (V.) braziliensis isolates from Peru and Bolivia [29]. All of the LRV1 sequences from L. (V.) braziliensis isolates formed a distinct monophyletic clade, highly supported by a posterior probability of 0.91 (Fig 2). LRV1 sequences from L. (V.) braziliensis isolates were subdivided into distinct clades as observed for those from L. (V.) guyanensis. The two sequences previously identified as clade F (YA70 and LRV1-Lb2169) grouped with the other sequences from L. (V.) braziliensis isolates. Interestingly, YA70, which corresponded to the unique strain obtained from a L. (V.) braziliensis isolate from French Guiana, possessed a basal position to the clade. Nevertheless, phylogenetic relationships of LRV1 isolates from L. (V.) guyanensis strains were less robust given that clades D and E clustered with clade F of L (V.) braziliensis viruses.
To determine if the multiple LRV1 strains identified in three Leishmania isolates were real viral co-infections or due to the simultaneous presence of distinct parasite populations in our cultures, we proceeded to the biological cloning of the parasites. Two of the three parasites, XJ93 and WF69, were tested. The initial parasite cultures were divided for either RNA isolation or biological cloning. Seven Leishmania clones and three subclones of one of the clones were obtained from XJ93. Five clones were obtained from WF69. RNA extracted from the initial cultures and from the different clones and subclones was then submitted to RT-PCR using different combinations of primers, each one specific to a virus (Table 2). RT-PCR proved the presence of co-occurring viruses in the initial cultures as well as in all the different clones and subclones.
Herein we present the full-length coding sequences of 24 LRV1 strains obtained from 18 L. (V.) guyanensis and one L. (V.) braziliensis isolates from across French Guiana as well as one L. (V.) guyanensis isolate from the Manaus area, Amazonas state, Brazil. This sequence data set is the largest analysis of LRV1 sequences undertaken so far. Analysis of these sequence data reveals that they have the same genomic organization. With the exception of a few indels detected in five strains, the 5’UTR sequences show, as previously reported, a higher degree of nucleotide sequence conservation than the coding sequences [47]. This high level of sequence conservation, with long– 100% conserved–motifs (up to 50 bp in length), emphasizes the biological importance of this region in the viral life cycle and for maintaining persistent infection [13, 15, 47]. Moreover, only two sequences possessed indels in the coding sequence of the RNA-dependent RNA polymerase. Two of the three indels observed were located at the same position between the two strains. Given that both strains derived from the only two L. (V.) braziliensis isolates for which complete cds of the RdRp are available, these indels could correspond to a molecular signature of LRV1 strains from L. (V.) braziliensis. Further studies are necessary to confirm this point.
From a phylogenetic perspective, it appears that differential clustering exists between viral genomes of L. (V.) guyanensis and L. (V.) braziliensis isolates and among viral genomes of L. (V.) guyanensis isolates. These results support the hypothesis that LRV1 is an ancient virus that has co-evolved with its parasite [25]. In addition, the intra- and inter-clade variability is, overall, below and above 10%, respectively, at both the nucleotide and amino acid levels. Finally, geographical clustering is observed with some clades related to the supposed geographical origin of the parasites. Nevertheless, insufficient geographical sampling, with many areas/countries lacking sufficient data, only based on very short sequences, currently limits phylogeographic interpretations. These results emphasize the crucial need for a better assessment of LRV1 distribution and genetic variation in the different parasite species at a wider geographical scale. Analysis of other L. (V.) guyanensis isolates from different geographical regions as well as of other New World Leishmania parasite species should thus expand our understanding of the diversification processes and the evolutionary history of LRV1. These results are also of major importance from a taxonomic viewpoint. They should help define analytical methods and criteria to be used for virus classification and to set up an informative naming system.
A unique viral genome sequence was recovered from 17 of the 20 Leishmania isolates tested. Unexpectedly, the three other isolates studied contained more than one viral sequence, indicating that they were co- or super-infected with two or more viruses. The different strains identified from the same isolate showed from about 8% to almost 17% nucleotide divergence among them on the almost complete genome sequences generated. A closer look at the 5’UTR motifs identified as minimal essential elements for site-specific targeting of the capsid endoribonuclease, i.e. the stem-loop IV structure and single-site specific cleavage site, of the different strains showed their perfect conservation, confirming that they had exact endoribonuclease sites [15]. Furthermore, biological cloning of these parasite isolates followed by RT-PCR using primers specific to each virus confirmed the presence of multiple viruses in each clone. Taken together, these data sustain that the multiple viruses identified in these parasite isolates correspond to true persistent viral co-infections. These results suggest that co- or super-infections with divergent LRV1 strains is common given that three of 20 (15%) isolates were multiply infected. The frequency of co-infections will nevertheless have to be confirmed on a larger series. Furthermore, the relevance of viral co-infection for parasite pathogenesis is unknown. Additional surveys on their biological significance will have to be implemented.
These results provide, moreover, a unique opportunity for implementing reliable diagnostic testing methods. Indeed, knowledge of the genomic sequence of many LRV1 strains was essential for oligonucleotide design ensuring adequate detection of the virus (100% detection rate for LRV1s-LRV2as and 5LRVs1-LRV11as/LRV14as). It is noteworthy that the primers and probes that have previously been described, whether used for quantification or identification of LRV1 in clinical samples, possessed mismatches relative to our sequence data set [22, 26, 29, 32, 48–50]. This could have led to underestimation of the viral load or under-detection of divergent unknown strains. Furthermore, phylogenetic analyses conducted on sequences obtained with our primer pairs (S3 Fig and S4 Fig) gave an identical tree topology to those based on complete coding sequences of CP or RdRp. These couples of consensus-degenerate primers are thus robust diagnostic tools intended to be used for future screening of large panels of clinical isolates as well as for molecular phylogenetic studies. Finally, characterization of highly conserved sequence motifs, especially in the 5’UTR region, should help, by defining primers and probes with no mismatch relative to the available sequence data set, implement a pan-LRV1 real-time quantitative reverse-transcription (qRT)-PCR assay for unbiased quantitation of viral RNA. Therefore, given previous results by Ives et al. showing in murine models that a high LRV1 burden, eliciting a high pro-inflammatory profile, was associated with metastasizing parasites, it will be of particular interest to analyze various types of clinical samples to further define the role of LRV1 viral load in parasite virulence, disease progression and response to treatment [28]. These tools should contribute to drawing a more definite causal relationship of LRV1 in disease manifestation.
|
10.1371/journal.pntd.0000555 | Seroprevalence to the Antigens of Taenia solium Cysticercosis among Residents of Three Villages in Burkina Faso: A Cross-Sectional Study | There is limited published information on the prevalence of human cysticercosis in West Africa. The aim of this pilot study was to estimate the prevalence of Taenia solium cysticercosis antigens in residents of three villages in Burkina Faso.
Three villages were selected: The village of Batondo, selected to represent villages where pigs are allowed to roam freely; the village of Pabré, selected to represent villages where pigs are usually confined; and the village of Nyonyogo, selected because of a high proportion of Muslims and limited pig farming. Clustered random sampling was used to select the participants. All participants were asked to answer an interview questionnaire on socio-demographic characteristics and to provide a blood sample. The sera were analysed using an AgELISA. The prevalence of “strong” seropositive results to the presence of antigens of the larval stages of T. solium was estimated as 10.3% (95%CI: 7.1%–14.3%), 1.4% (0.4%–3.5%) and 0.0% (0.0%–2.1%) in the 763 participants who provided a blood sample in Batondo, Pabré and Nyonyogo, respectively. The prevalence of “weak” seropositive test results to the presence of antigens of the larval stages of T. solium was 1.3% (0.3%–3.2%), 0.3% (0.0%–1.9%) and 4.5% (2.0%–8.8%) in Batondo, Pabré and Nyonyogo, respectively. The multivariate logistic regression, which included only Batondo and Pabré, showed that village, gender, and pork consumption history were associated with AgELISA seroprevalence.
This study illustrates two major points: 1) there can be large variation in the prevalence of human seropositivity to the presence of the larval stages of T. solium cysticercosis among rural areas of the same country, and 2) the serological level of the antigen, not just whether it is positive or negative, must be considered when assessing prevalence of human cysticercosis antigens.
| Taenia solium cysticercosis is a neglected tropical zoonosis transmitted between humans and pigs. This infection is particularly prevalent in areas where sanitation, hygiene and pig management practices are poor. There is very little information about the importance of this infection in West Africa, even though pork meat is widely consumed in many areas. This pilot study, conducted in three villages of Burkina Faso, demonstrated that people living in areas where pigs are raised were more likely to be infected with cysticercosis than people living in a Muslim village in which there were very few pigs. It also demonstrated variation in the level of infection between the two villages where pigs were raised. Finally, the results suggest that the source of infection in these three villages may differ. These results are significant because they show that there is clustering of infection within villages, even if they are geographically very close to one another. This should encourage future researchers not to combine data from several villages into one summary value. In addition, more work is needed to better describe different potential sources of infection among villages.
| Taenia solium is a tapeworm transmitted among humans and between humans and pigs. Taeniasis is acquired by humans when eating raw or undercooked pork contaminated with cysticerci, the larval stage of T. solium. When ingested, the cysticerci migrate to the intestine of humans where they establish and become adults. These adult worms shed eggs in human feces that can infect other humans and pigs by direct contact or by indirect contamination of water or food. This can be especially problematic in developing countries where pigs are often allowed to roam freely and to eat human feces and where levels of sanitation and hygiene are poor. Ingested eggs result in larval worms which migrate to different parts of the pig or human body and form cysts. A principle site of establishment of the larvae in humans is the central nervous system. Human neurocysticercosis (NCC) occurs when the cysts develop in the brain or spinal cord. Seizures are believed to be the most common presentation of NCC, affecting from 66% to 90% at some stage of their disease [1],[2].
There is limited published information on the prevalence of human cysticercosis in West Africa. In Burkina Faso, no prevalence study has ever been conducted, although NCC has been reported. In a retrospective review of the medical records of 532 persons with seizure disorder seen either as inpatients or outpatients at Yalgado Ouédraogo Teaching Hospital in Ouagadougou, 6.3% of the 158 cases in whom a presumed cause was identified were attributed to NCC based on clinical evidence [3]. No imaging was used to confirm the diagnosis which could have lead to an underestimation of the proportion of seizure cases attributable to NCC. In addition, no definition of “clinical evidence of cysticercosis” was provided, whichmakes this estimate very difficult to interpret. Case reports of human cysticercosis in Burkina Faso have also been published [4]. A third study reviewed 3410 histopathological samples from any location (surgical and biopsy) collected between 1991 and 1995 at the two reference hospitals in Bukina Faso and found 18 with evidence of current infection with T. solium larvae [5].
In neighboring pig-raising countries, community-based seroprevalence estimates of cysticercosis in humans range from 1.3% to 3.95% [6]–[9]. There have also been case reports of human cysticercosis in Ivory Coast, Ghana and Senegal [10].
The main objective of the present study is to estimate the prevalence to the antigens of T. solium cysticercosis as an indicator of current infection, in three villages in Burkina Faso. A secondary aim is to measure the association between potential risk factors and the prevalence of seropositivity to the antigens of T. solium larval stages.
Informed consents for the interviews of participants and the provision of blood samples were obtained separately. The consent process was done orally because a very large proportion of the population had never been to school (62.5%). Oral consent was documented on the individual consent forms by the research staff. The study protocol was reviewed and approved by the ethical committee of the Center MURAZ (Ref. 02-2006/CE-CM) and by the Institutional Review Board of the University of Oklahoma Health Sciences Center (IRB# 12694) in regard to both human and porcine participants. Both IRBs approved the use of oral consents. The sampling of blood from pigs was approved by the OUHSC IACUC committee (approval #06-018).
The pilot study was conducted in the villages of Batondo, Pabré and Nyonyogo, located close to the Capital City of Ouagadougou (Figure 1). The three villages were conveniently selected to represent three types of pig managements. The village of Batondo, located in the commune of Ténado (province of Sanguié) 140 km west of Ouagadougou, was selected to represent villages where pigs are owned and raised by women and are allowed to roam freely. The village of Pabré, in the commune of Pabré (province of Kadiogo), located 25 km north of Ouagadougou, was selected to represent villages where pigs are raised and are usually confined for some period of time during the year. The village of Nyonyogo, located in the commune of Dapelogo (province of Oubritenga) was selected due to a high proportion of Muslims and hence limited pig farming.
A census of all concessions (a grouping of several households usually members of the same family) and households in each village was first conducted. In Batondo and Nyonyogo, all concessions were included. In Pabré, 50% of the concessions were selected at random. Within each concession, all households were included and one person was randomly sampled from each household for participation in the interview and venipuncture for collection of blood samples for serological testing. The random selection was done by placing the names of each household member in a bowl and by asking a child to pick one name from the bowl.
This cross-sectional study was conducted between May and October 2007.
The head of each household was first interviewed to collect information about each member of his family. In households where pigs were raised, the caretaker of the pigs was interviewed regarding pig management practices. A cooking practices interview was conducted with the wife of monogamous households and with the “senior” woman in polygamous households. The individual sampled at random in each household was asked to answer an epilepsy screening questionnaire which also included socio-demographic information and was administered by a trained member of the study staff. All questionnaires were translated from French to the local languages and back-translated to French. The questionnaires were also pilot tested among a small group of people with and without epilepsy prior to the start of the field study.
Blood samples were left to decant at the end of each sampling day and the sera were put in freezers (−20°C) until the samples were brought to the IRSS (Institut de Recherche en Sciences de la Santé) in Bobo-Dioulasso where they were centrifuged and the sera kept at −20°C. The serum samples were tested for circulating antigens of the metacestode of T. solium using the enzyme-linked immunosorbent assay (ELISA) [11],[12]. This test is designed to measure the presence of current infection with the larval stages of T. solium and not the history of past or present exposure. A seropositive result is indicative of current infection and may or may not be associated with symptoms. The cut-off value was calculated as described by Dorny et al., 2004 [13]. A ratio for each test was calculated dividing the optical density of the sample by the cut-off value. The ratios were used to classify the results as negative (ratio between 0 and 1.0), “weak” positive (ratio between 1.0 and 1.35) and “strong” positive (ratio>1.35). Samples with a coefficient of variation of more than 50% were considered as missing values (n = 3). The sensitivity and specificity of the AgELISA for current infection with cysticercosis has only been reported from a preliminary study conducted in Vietnam. There, the study indicated a sensitivity of 94.4% and a specificity of 100% for the diagnosis of current infection with cysticercosis [14]. The sensitivity and specificity of Ag-ELISA for current cysticercosis infection has been determined in pigs, with a sensitivity and specificity of 86.7% and 96.7%, respectively in Zambia [13], and 76.3% and 84.1%, respectively in South Africa [15].
The prevalence of seropositivity to the larval stages of T. solium cysticercosis was estimated separately for “weak” and “strong” seropositivity as the number with “weak” and “strong” positive AgELISA results, respectively, divided by the number of people who provided a blood sample. We then fitted two random-effect models (xtlogit) with the “weak” results assumed either positive or negative to estimate the proportion of the total variance contributed by the village-level clustering (statistic rho).
The results from Nyonyogo were analysed separately and included only in the univariate analyses due to the small number of seropositives and the very small number of pigs raised in that village, which made this village very different from the two others.
For risk factor analyses in Pabré and Batondo, only those with “strong” responses were considered as positive. Univariate associations between being positive to AgELISA and socio-demographic and pork consumption variables at the individual level, as well as pork preparation and pig management variables at the household level were first assessed. Comparisons were made by calculating a prevalence proportion ratio (PPR) with 95% confidence intervals (95%CI). Variables with significant or borderline significant associations with seropositivity in the univariate analyses were then included in a multivariate logistic model adjusting for the effect of village. The results are reported as prevalence odds ratios (POR) with 95% confidence intervals (95%CI). A random-effect logistic regression model with clustering at the concession level was also fitted to take into consideration the clustered nature of the sampling. The results of this model were identical to those from the simple model, however, and thus, only the latter are presented. All analyses were conducted in Stata 10 SE.
A total of 888 individual interviews were conducted with participants in the three villages. All sampled individuals agreed to answer the interview questionnaire. Of these, 766 (86.3%) provided a blood sample. Table 1 shows the proportion of participants providing a blood sample by selected socio-demographic characteristics. Briefly, the proportion of people providing a blood sample varied somewhat from village to village due to a variety of factors such as the presence of visible veins, the difficulty of obtaining a blood sample, and refusal to provide a sample following the interview. Among those who provided a blood sample, 763 had a valid AgELISA test result.
The prevalence of “strong” seropositive test results was estimated as 10.3% (95%CI: 7.1%–14.3%), 1.4% (0.4%–3.5%) and 0.0% (0.0%–2.1%) in Batondo, Pabré and Nyonyogo, respectively. The prevalence of “weak” seropositive test results was 1.3% (0.3%–3.2%), 0.3% (0.0%–1.9%) and 4.5% (2.0%–8.8%) in Batondo, Pabré and Nyonyogo, respectively. The random-effect models showed that the variance due to the village contributed to a large proportion of the overall variance. The rho statistic was estimated to 0.49 (95%CI: 0.08–0.92) when the “weak” seropositives were assumed negative and to 0.16 (95%CI: 0.03–0.55) when the “weak” seropositives were assumed positive.
Table 2 shows the prevalence of AgELISA seropositivity within categories of several potential risk factors and stratified by village. In Nyonyogo, univariate analyses showed males to have an increased prevalence of presenting a “weak” serological results as compared to females (PPR = 8.02 (95%CI: 1.01, 63.86)). None of the 8 cases with “weak” results had gone to school and none reported using the toilet to defecate. However, only 18.2% and 11.3% of the population of the village had ever attended school and reported using a toilet to defecate, respectively. Nyonyongo children aged less than 16 tended to have a higher prevalence proportion, based on “weak” results, than adults with a PPR = 3.63 (95%CI: 0.95–13.95), and there was a tendency for a higher prevalence in concessions of larger sizes. This latter trend by concession size was observed to a lesser extend in Batondo, but not in Pabré. Except for the association with gender, none of the variables noted in relation to seroprevalence (“weak” only) in Nyonyongo were observed in the other two villages (with “strong” seropositive tests) (Table 2). Because there were so few pigs being raised in Nyonyogo, “weak” seropositive tests were not found to be associated with pork consumption or pig raising in this village.
The multivariate logistic regression, which included only Batondo and Pabré, showed that village, gender, and pork consumption habits were associated with “strong” AgELISA seropositivity (Table 3). The odds of being seropostive were considerably higher in Batondo than in Pabré (POR = 8.86; 95%CI = 3.01, 26.14), in men compared to women (POR = 2.34; 95%CI = 1.10, 4.97) and in those eating pork either in the past (POR = 19.62; 95%CI = 1.91, 2010.95) or currently (POR = 8.75; 95%CI = 1.11, 68.88) compared to those who never ate pork. The ownership of pigs by one household member confounded the association between seropositivity and pork consumption such that when this variable was included in the model the association between eating pork and seroprevalence became significant. Although not itself statistically significant, because of this interaction the variable “pig ownership” was retained in the multivariate model.
This study is the first to estimate the seroprevalence to the presence of antigens to T. solium cysticercosis in Burkina Faso. The strengths of this study are that the results are based on a clustered-random sample of residents of three rural villages, the participation proportion for the interview was excellent (100%) and very good (86.3%) for the serology, the participants answered almost all questions in the interviews (few missing values), and the majority of serological tests conducted had valid results.
Our results show considerable variation in seroprevalence in the three study villages. The prevalence to the presence of antigens of T. solium cysts was nearly 8 times higher in Batondo than in Pabré. Several reasons may explain the difference between Batondo and Pabré. For example, the proportion of participants who had gone to school was much higher in Pabré (55.6%) than in Batondo (30.0%), and a larger proportion of the participants used the toilet to defecate in Pabré (37.6%) than in Batondo (7.9%). Pigs were also more often penned in Pabré (54.8%) during the rainy season as compared to Batondo (10.9%) where pigs were more often tethered (94.0%). Tethered pigs are more likely than penned pigs to have access to human feces as they are often moved to be able to feed, and hence, are more likely to be exposed to a contaminated site. It was also observed during the field study that pigs were living in closer contact with humans in Batondo than in Pabré. These factors, possibly in addition to other village-level variables that were not measured, may contribute to the lower prevalence of the presence of antigens of T. solium cysts in Pabré compared to Batondo.
All seropositive results in Nyonyogo were “weak”. One could speculate, as an hypothesis yet to be tested, that the force of infection from the environment is lower in Nyonyogo due to the presence of very few pigs and therefore, to a lower prevalence of taeniasis among the population. It is also possible that the source of infection is different in Nyonyogo compared to that in the other villages as discussed below. Another hypothesis is that most of the infections in Nyonyogo were either very recent or old resulting in a lower density of antigens in the blood.
Although based on a small number of “weak” cases, the finding that children tended to have a higher seroprevalence than adults only in Nyonyogo deserves further exploration. One hypothetical explanation for this observation is that children in Nyonyogo acquire infection through playing in the contaminated environment since few adults consume pork and thus are at lower risk for taeniasis. Sanitation was very poor in Nyonyogo with only 14.4% of the household having a toilet and 11.2% of the people using the latrine to defecate. In the other two villages, more people consume pork and the prevalence of taeniasis is probably higher, which could increase the risk of auto-infections or infection through the contamination of food and water. This possibility could also explain the difference in the strength of AgELISA optical densities between the villages. These interpretations are hypotheses which would need to be verified in a larger cohort study.
To our knowledge, there has been only one other community-based sero-survey done in Sub-Saharan Africa on the presence of antigens to the larval stages of T. solium cysticercosis [9], which reported a seroprevalence varying between 0.4% and 3.0% in the three rural communities in Menoua district, Cameroon, between 1999 and 2000. One important limitation of this study is that sampling was done among volunteers. Nonetheless, the study's results indicate some variation in prevalence by community, but less variation than what was observed in the present study in which participants were selected according to a clustered random sampling strategy. Also, we had specifically sampled a village where the majority of the population was Muslim and where there were very few pigs being raised.
In a recent hospital-based case-control study from the Kiremba area of Burundi, the prevalence of seropositivity using AgELISA in controls (persons without epilepsy) was estimated as 20% [16]. In this study, 80% of the controls were people being vaccinated at the Kiremba area hospital and age-matched to a group of patients with epilepsy who were being seen at that hospital. This design makes it difficult to assess the source of the controls and therefore assess to what extent they represent the general population since controls were age matched to epilepsy patients. In a study in Cameroon of people with epilepsy receiving care in rural clinics, the seroprevalence by AgELISA was estimated as 1.2% [17]. The estimated prevalence of seropositivity to the antigens of Taenia solium cysts among people in our sample with confirmed epilepsy was considerably higher at 15.2%.
In the Burundi case-control study, the POR of seropositivity to the antigens of Taenia solium cysts was 2.5 (95%CI = 1.8, 3.4) when comparing males to females and 1.7 (95%CI: 1.1–2.5) when people eating pork were compared to those not eating pork [16], results similar to those of the present study, although the association with pork consumption in our cross-sectional study was much larger.
In Batondo and Pabré, the presence of pigs within the household tended to reduce the seroprevalence and it confounded the association between pork consumption and seroprevalence. This is because there was a higher proportion of participants from households where pigs were raised who consumed pork than in household where pigs were not raised. This uneven distribution of pork consumption according to the presence of pig raising leads to an underestimation of the effect of pork consumption on the seroprevalence to the antigens of Taenia solium cysts if not adjusted. The proportion of participants aware of the link between consumption of undercooked pork and taeniasis was similar among people who raised pigs (15.1%) and those who did not (14.2%). It may be that members of households where pigs are not raised are more likely to consume pork at the market than are those from household where pigs are raised, which may expose them to greater risk than those eating their meals at home. Unfortunately, our questionnaire did not include a question on the location of consumption of pork meat, which would have helped in explaining this confounding factor.
This study has some limitations. First, 13.7% of the 888 interviewed participants did not provide a blood sample. The reasons for not providing a sample may have been linked to the difficulty in obtaining blood at the time of sampling or refusal to provide a sample following the interview. Refusals were a minority of those who did not provide a sample. Due to this, we believe that important selection bias is unlikely. Second, it would have been very interesting to re-test participants with “weak” results a few months later. This would have indicated whether those “weak” cases were new or old infections. Unfortunately, this was not feasible in the context of this project. Third, it would have also been interesting to obtain results from a valid antibody serological test. We did test the samples for the presence of antibodies according to the method of Arruda et al. 2005 [18]. However, this test is now thought to be invalid due to several cross reactions with other helminthes and protozoa such as Echinococcus, Filaria , Fasciola , Strongyloides, Schistosoma, Toxocara, Amoeba ,and Plasmodium [19]. Another alternative would have been to test the samples using the well validated Western Blot test (EITB) for the detection of antibodies to T. solium [20]. This was not feasible due to the complexity and cost of the test. However, as a test for the presence of current infection with the larval stages of T. solium, preliminary studies indicated a sensitivity of 94.4% and a specificity of 100% of the Ag ELISA test in Vietnam [14].
We have shown that the prevalence to the antigens of the larval stages of T. solium can be very high in some villages of Burkina Faso and virtually nonexistent in other villages in the same region. This study illustrates two major points; first, there can be large variation in the prevalence of antigens to human cysticercosis among rural areas of the same country, and second, the serological level of the antigen, not just whether it is positive or negative, must be considered when analyzing data in order to arrive at more valid conclusions. The first point is especially relevant to studies of the burden of diseases. If all studies are concentrated in areas where pigs are roaming, the overall burden of the infection for the country would be over-estimated; the converse would be true as well.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.