1. Neuroscience
  2. Structural Biology and Molecular Biophysics
Download icon

Mutational analysis to explore long-range allosteric couplings involved in a pentameric channel receptor pre-activation and activation

  1. Solène N Lefebvre
  2. Antoine Taly  Is a corresponding author
  3. Anaïs Menny
  4. Karima Medjebeur
  5. Pierre-Jean Corringer  Is a corresponding author
  1. Institut Pasteur, Université de Paris, CNRS UMR 3571,Channel-Receptors Unit, France
  2. Sorbonne Université, Collège doctoral, France
  3. Institut de Biologie Physico-chimique, Fondation Edmond de Rothschild, PSL Research University, France
  4. Laboratoire de Biochimie Théorique, CNRS, Université de Paris, UPR 9080, France
Research Article
  • Cited 0
  • Views 454
  • Annotations
Cite this article as: eLife 2021;10:e60682 doi: 10.7554/eLife.60682

Abstract

Pentameric ligand-gated ion channels (pLGICs) mediate chemical signaling through a succession of allosteric transitions that are yet not completely understood as intermediate states remain poorly characterized by structural approaches. In a previous study on the prototypic bacterial proton-gated channel GLIC, we generated several fluorescent sensors of the protein conformation that report a fast transition to a pre-active state, which precedes the slower process of activation with pore opening. Here, we explored the phenotype of a series of allosteric mutations, using simultaneous steady-state fluorescence and electrophysiological measurements over a broad pH range. Our data, fitted to a three-state Monod-Wyman-Changeux model, show that mutations at the subunit interface in the extracellular domain (ECD) principally alter pre-activation, while mutations in the lower ECD and in the transmembrane domain principally alter activation. We also show that propofol alters both transitions. Data are discussed in the framework of transition pathways generated by normal mode analysis (iModFit). It further supports that pre-activation involves major quaternary compaction of the ECD, and suggests that activation involves principally a reorganization of a ‘central gating region’ involving a contraction of the ECD β-sandwich and the tilt of the channel lining M2 helix.

Introduction

Pentameric ligand-gated ion channels (pLGICs) mediate fast synaptic communication in the brain. In mammals, this family includes the excitatory nicotinic acetylcholine (ACh) and serotonin receptors (nAChRs and 5-HT3Rs) as well as the inhibitory γ-aminobutyric acid (GABA) and glycine receptors (GABAARs and GlyRs) (Jaiteh et al., 2016). pLGICs are also present in bacteria, notably with the pH-gated channels GLIC (Bocquet et al., 2007) and sTeLIC (Hu et al., 2018), the GABA-gated channel ELIC (Zimmermann and Dutzler, 2011), and the calcium-modulated DeCLIC (Hu et al., 2020).

pLGICs physiological function is mediated by alternating between different allosteric conformations in response to neurotransmitter binding. Initially, a minimal four-state model could describe the main allosteric properties of the muscle-type nAChR (Heidmann and Changeux, 1980; Sakmann et al., 1980). In this model, the ability of ACh binding to activate the nAChR involves a resting- to active-state transition, and prolonged ACh occupancy promotes a biphasic desensitization process. Subsequently, kinetic analysis of the close-to-open transitions recorded by single-channel electrophysiology revealed multiple additional states that are required to account for the observed kinetic patterns. For activation, short-lived intermediate ‘pre-active’ states named ‘flipped’ (Lape et al., 2008) and ‘primed’ (Mukhtasimova et al., 2009) were included in the kinetic schemes of the GlyRs and nAChRs, while rate-equilibrium free-energy relationship analysis of numerous mutants of the nAChR suggested passage through four brief intermediate states (Gupta et al., 2017). Likewise, analysis of single-channel shut intervals during desensitization is described by the sum of four or five exponential components, suggesting again additional intermediate states (Elenes and Auerbach, 2002). Kinetics data thus show that pLGICs go through complex structural reorganizations during both activation and desensitization. These events are at the heart of the protein’s function, allowing coupling between the neurotransmitter site and the ion channel gate which are separated by a distance of more than 50 Å.

The past decade has seen great structural biology efforts to increase our understanding of the molecular mechanisms involved in gating (Nemecz et al., 2016). At least one structure of each major member of prokaryotic (Hilf and Dutzler, 2008; Bocquet et al., 2009; Hu et al., 2018; Hu et al., 2020) and eukaryotic pLGICs (Althoff et al., 2014; Du et al., 2015; Polovinkin et al., 2018; Gharpure et al., 2019; Masiulis et al., 2019) have been resolved by X-ray crystallography or cryo-electron microscopy (cryoEM). They highlight a highly conserved 3D architecture within the family. Each subunit contains a large extracellular domain (ECD) folded in a β-sandwich and a transmembrane domain (TMD) containing four α-helices, with the second M2-helix lining the pore. However, the physiological relevance of these structures or their assignment to particular intermediates or end-states in putative gating pathways remains ambiguous and poorly studied. Conversely, it is possible that intermediate conformations, unfavored by crystal packing lattice or under-represented in receptor populations on cryoEM grids, are missing in the current structural galleries.

Understanding the allosteric transitions underlying gating thus requires complementary techniques, where the protein conformation can be followed in near-physiological conditions, that is at non-cryogenic temperature on freely moving protein, and over a broad range of ligand concentrations. To this aim, we previously developed the tryptophan/tyrosine induced quenching technique (TrIQ) on GLIC (Menny et al., 2017), a proton-gated channel (Parikh et al., 2011; Bocquet et al., 2007; Laha et al., 2013; Gonzalez-Gutierrez et al., 2017). In this technique, the protein is labeled with a small fluorophore(bimane, and collisional quenching by a neighboring indole (tryptophan) or phenol (tyrosine) moiety is used to report on changes in distance between two residues within the protein over a short distance range of 5–15 Å (Mansoor et al., 2002; Mansoor et al., 2010; Jones Brunette and Farrens, 2014). Bimane-quencher pairs on GLIC combined with kinetic analysis allowed us to characterize pre-activation motions occurring early in the conformational pathway of activation (Figure 1A). We found that they occur at lower proton concentrations than pore opening, and are complete in less than a millisecond, much faster than the rise time of the active population that occurs in the 30–150 ms range in electrophysiology recordings (Laha et al., 2013).

Figure 1 with 2 supplements see all
Electrophysiological and fluorescence characterization of the quenching pairs of GLIC.

(A) Scheme for GLIC activation, showing first a pre-activation step involving full compaction of the ECD and motion of the M2-M3 loop as monitored by fluorescence, followed by a pore opening step. Blue spheres indicate the location of sensors Bim136-Q101W and Bim250-Y197 used thereafter in this study. (B) GLIC-pH 4 (pdb code 4HFI) structure side view, the light blue rectangle represents the position of the membrane. Quenching pairs generated in our previous study (Menny et al., 2017) are highlighted: blue spheres show the Cα of the residues that were mutated into cysteines and bimane labeled (Bim33, Bim133, Bim135, Bim136, and Bim250), black spheres show the Cα of the quenchers (W160, L103W, W72, Q101W, and Y197). (C) pH-dependent response curves of Bim136-Q101W and Bim250-Y197 sensors, by electrophysiology after labeling (top panel) and with bimane fluorescence quenching (lower two panels). Fluorescence data are shown normalized to the fluorescence of the denatured protein (FSDS), bimane fluorescence is shown without quencher (○) and in presence of the quencher (●). ECD, extracellular domain.

Figure 1—source data 1

Fluorescence quenching and electrophysiological current measurements of the different mutants tested.

https://cdn.elifesciences.org/articles/60682/elife-60682-fig1-data1-v2.xlsx

Here, to explore the conformational landscape of GLIC during pH-gating, we further exploited the TrIQ approach. We performed electrophysiological and fluorescence quenching experiments on a series of allosteric mutants of GLIC, as well as in the presence of the general anesthetic propofol. We modeled the whole data set with a three-state allosteric model comprising a resting state, a pre-active state, and an active state. To help the interpretation of the fluorescence quenching data into structural terms, we built atomistic models of the various bimane-labeled proteins, and computed their gating transition pathways using iMODfit. Our results indicate that mutations alter the function via distinct mechanisms and differentially displace the allosteric equilibria involved in fluorescence quenching and electrophysiology recordings. This supports that pre-activation involves a major quaternary compaction of the ECD, and suggests that activation involves principally a reorganization of a ‘central gating region’ involving a contraction of the ECD β-sandwich and the tilt of the channel lining M2 helix.

Results

Fluorescence and electrophysiological measurements

Quenching pairs used in the study

In our previous fluorescence quenching experiments, a bimane fluorophore was introduced on GLIC by covalent labeling on an engineered cysteine, after mutation of the single endogenous cysteine C27S. A Trp or Tyr quenching residue was incorporated when necessary to generate a quenching pair. We created five quenching pairs (Figure 1B): three are located across the ECD interface and report on a quaternary compaction following pH drop (Bim136-Q101W, Bim133-L103W, and Bim33-W160), one reports on a tertiary reorganization at the top of the ECD (Bim135-W72), and one reports on the outward movement of the M2-M3 loop at the ECD-TMD interface (Bim250-Y197). In the present study, we used the Bim136-Q101W as sensor of the ECD compaction, along with Bim250-Y197 as a sensor of the M2-M3 loop motion. We also investigated in detail Bim135-W72, but the complex results for this pair precluded clear conclusions. The related data are thus presented and discussed in Figure 1—figure supplement 1.

To accurately compare mutants, we first measured detailed pH-dependent fluorescence and electrophysiological curves (Figure 1C). Fluorescence was measured in steady-state conditions on detergent (DDM)-purified protein, and normalized to the fluorescence intensity under denaturing conditions (1% SDS), as previously described (Menny et al., 2017). GLIC allosteric transitions are particularly robust in different lipid/detergent conditions (Sauguet et al., 2014; Carswell et al., 2015) and DDM-purified protein yielded similar results to that of azolectin-reconstituted protein (Menny et al., 2017), while allowing better reproducibility. For both sensors, we confirmed that the pH-dependent fluorescence changes are essentially abolished when mutating the quenching partner to phenylalanine, which does not quench bimane fluorescence (Mansoor et al., 2002). We also confirmed that pH-dependent quenching curves for Bim136-Q101W and Bim250-Y197 display higher sensitivity (especially for Bim250-Y197) and lower apparent cooperativity than the pH-dependent activation curves recorded by electrophysiology (Figure 1—figure supplement 2).

The quaternary compaction at the ECD top is strongly allosterically coupled with the lower part of the ECD interface

Using the Bim136-Q101W conformational sensor, we first investigated allosteric mutants located at the inter-subunit interface in the lower part of the ECD (Figure 2A). We previously showed that E26Q produces a decrease in pH50 for activation (Nemecz et al., 2017), a phenotype that is conserved here on the Bim136-Q101W background (Figure 2B and C). The fluorescence quenching curve of Bim136-Q101W-E26Q also shows a decrease in pH50 (Figure 2D), and the ∆pH50 between Bim136-Q101W and Bim136-Q101W-E26Q are nearly identical in electrophysiology and fluorescence (−0.59 and −0.57, respectively; Table 1). Interestingly, the Bim136-Q101W-E26Q fluorescence quenching curve has a remarkable feature as compared to most mutants investigated thereafter: in the pH 7–8 range, where the pH-dependent fluorescence quenching is not yet observed, the fluorescence (F/FSDS) is significantly lower (F0=0.53) as compared to the Bim136-Q101W alone (F0=0.71). This suggests that substantial quenching is present at neutral pH and that E26Q not only alters the allosteric transition, but also modifies the conformation of the resting state itself which appears to be more compact when the E26Q mutation is present.

Allosteric coupling within the ECD.

(A) Structure of two monomers of GLIC pH 4 (4HFI) showing positions of the fluorescence sensor (Bim136-Q101W) and the two mutated residues at the bottom of the ECD resulting in a partial loss of function. The lower panel shows a zoom on the interface with E26 and Y28 residues and their interactions with surrounding residues and a network of water molecules (blue spheres). (B) Electrophysiological recordings in oocytes of the mutants labeled with bimane showing shifted responses to higher proton concentrations in comparison with GLIC Bim136-Q101W. pH applications are shown above each trace and the horizontal scale represents 1 min of recording. Graphs represent pH-dependent curves showing a shift to higher proton concentrations in electrophysiological responses (C) and fluorescence quenching responses (D) for both mutants. ECD, extracellular domain.

Table 1
pH-dependence of electrophysiological and fluorescence quenching responses.

pH50 and Hill coefficient nH average and standard deviation values are shown after individual fitting of each measurement. n corresponds to the number of oocytes for electrophysiology and the number of fluorescence measurements, each measurement including values for a full pH range. F0 corresponds to the initial fluorescence value at pH 7/8 and ∆Fmax is the maximum variation in fluorescence amplitude within the pH range (absolute values). To reasonably fit Bim136-Q101W + propofol current and Bim136-Q101W-Y28F fluorescence, Hill coefficients have been constrained to 2.5 and below 3 respectively. ∆pH50s are calculated between mutants and their parent construct Bim136-Q101W or Bim250-Y197 (labeled Ref). Their significance was calculated with a one-way ANOVA test using a Dunnett’s multiple comparisons test. The p-value is significantly different with p-value≤0.0001 (****), ≤ 0.001 (***), ≤0.01(**), ≤0.05 (*) or not significantly different when p-value>0.05 (ns). NF stands for non-functional and ND for not determined. To compare electrophysiological pH50 and fluorescence pH50 for each mutant (right column), unpaired t-tests were done with two-tailed p-value and 95% confidence intervals.


Mutant
Electrophysiological response bimane labeledFluorescence quenching responseFluorescence/
in detergent solutionelectrophysiology
pH50nHn∆pH50pH50F0∆FMAXnHn∆pH50∆pH50
Bim136-Q101W C27S5.42±0.082.68±0.3310Ref5.85±0.210.71±0.030.45±0.060.77±0.1817Ref0.43***
+ E26Q4.83±0.122.98±0.646−0.59****5.28±0.340.53±0.020.22±0.021.13±0.274−0.57**0.45*
+ Y28 F3.88±0.082.63±0.683−1.54****3.68±0.340.70±0.010.38±0.09<33−2.17****−0.2***
+ Y28 F & C275.34±0.112.03±0.126−0.08nsNDNDNDND
+ D32E4.65±0.122.62±1.516−0.77****5.52±0.050.68±0.020.38±0.010.75±0.053−0.33ns0.87***
+ E222Q4.68±0.092.51±0.335−0.74****5.36±0.140.66±0.030.40±0.040.74±0.183−0.49*0.68***
+ H235Q4.04±0.211.19±0.316−1.38****5.00±0.090.70±0.010.42±0.010.78±0.103−0.85****0.96***
+ Propofol5.16±0.13=2.53−0.26*5.33±0.060.67±0.020.38±0.011.21±0.214−0.52**0.17ns
+ H235 Q & propofol4.71±0.151.53±0.486−0.71****5.67±0.140.70±0.010.43±0.011.25±0.043−0.18ns0.96***
+ H235 FNFNF35.25±0.080.70±0.050.38±0.041.06±0.173−0.60**
+ L157 ANFNF35.42±0.650.67±0.010.19±0.090.71±0.464−0.43*
+ L246 ANFNF34.87±0.140.65±0.010.24±0.010.79±0.123−0.98****
Bim250-Y1974.66±0.182.20±0.5412−0.76****5.83±0.170.59±0.040.33±0.091.19±0.288−0.02nsRef1.17***
+ H235 FNFNF35.40±0.130.54±0.010.20±0.011.16± 0.154−0.43**
+ L157 ANFNF35.81±0.190.49±0.060.45±0.070.64±0.193−0.02ns
+ L246 ANFNF35.53±0.030.64±0.010.30±0.011.69±0.293−0.3*

Another mutation, Y28F, was reported to produce a moderate gain of function on a wild-type background (Nemecz et al., 2017). Surprisingly, mutating Y28F in the Bim136-Q101W (C27S) background yields a drastic loss of function characterized by a slow activating receptor and a marked decrease in pH50 (Figure 2B and C; Table 1). However, mutating back the C27 endogenous cysteine (Bim136-Q101W-Y28F (C27)) reverses the phenotype to that of Bim136-Q101W (C27S) (Figure 2C and Table 1), demonstrating that this loss of function is due to the combination of the C27S and Y28F mutations. In fluorescence, the quenching curve of Bim136-Q101W-Y28F (C27S) also shows a large decrease in pH50, associated with an apparent higher cooperativity. Again, the ∆pH50 is in the same range in fluorescence quenching (–2.2) and in electrophysiology (more than −1.5, the plateau could not be reached with this mutant preventing accurate measurement of the pH50).

In conclusion, the quaternary compaction of the top of the ECD, monitored with the Bim136-Q101W sensor, is strongly coupled in an allosteric manner with the lower part of the ECD interface.

Long-range allosteric coupling between the TMD and the top of the ECD

To investigate whether allosteric coupling occurs with more distant regions of the protein, we selected three loss of function mutations further away from the Bim136-Q101W pair: D32E near the ECD-TMD interface; H235Q in the middle of the TMD and E222Q, at the bottom of the TMD and lining the pore (Figure 3A; Sauguet et al., 2014; Nemecz et al., 2017).

Allosteric coupling between the top of the ECD and the TMD.

(A) Structure of two monomers of GLIC pH 4 (4HFI) showing positions of the fluorescence sensor (Bim136-Q101W) at the top of the ECD and three mutations distributed along the protein. Right panels show zooms on important interactions with the mutated residues. (B) Electrophysiological recordings of the three mutants in oocytes, labeled with bimane. Recording of GLIC Bim136-Q101W is shown for comparison. pH applications are shown above each trace and the horizontal scale represents 1 min of recording. pH-dependent curves for electrophysiological response (C) and fluorescence quenching (D) for the three mutants in comparison with Bim136-Q101W showing a shift to higher proton concentrations of the response for all three mutants. ECD, extracellular domain; TMD, transmembrane domain.

Performing these mutations on the Bim136-Q101W-C27S background shows overall conservation of their previously published phenotype, with a 0.7 unit (D32E and E222Q) and 1.3 unit (H235Q) decreases in the pH50 of activation as compared to Bim136-Q101W-C27S (Figure 3B and C; Table 1). The fluorescence quenching curves are also shifted to lower pH50s, with ∆pH50s of 0.3–0.5 (D32E and E222Q) and 0.85 (H235Q) (Figure 3D).

The quenching data thus reveal an allosteric coupling between both ends of the protein, since the structural perturbations performed around the TMD are transmitted to the top of the ECD, impairing its compaction. However, as opposed to the ECD mutations E26Q and Y28F/C27S, these mutations have a stronger effect on the pH50 of the electrophysiological response as compared to fluorescence quenching. It thus suggests that both processes are not fully coupled for mutations further away from the sensor site.

Total loss of function mutations differentially alter ECD and TMD allosteric motions

To further explore the allosteric coupling within GLIC, we extended the analysis to mutations known to strongly or completely prohibit channel opening (Figure 4A and B). We selected three mutants: H235F, L157A, and L246A which show robust surface expression and no substantial current in oocytes (Figure 4C and Figure 4—figure supplement 1). For those mutants, in addition to electrophysiological recordings and fluorescence quenching measurements on Bim136-Q101W and Bim135-W72 (Figure 4D and Figure 1—figure supplement 1), we also monitored the motion of the M2-M3 loop with Bim250-Y197 (Figure 4E).

Figure 4 with 1 supplement see all
Non-functional mutants differentially alter ECD and TMD motions.

(A) Structure of two monomers of GLIC pH 4 showing the position of the fluorescence sensors (Bim136-Q101W and Bim250-Y197) and three mutations causing a total loss of function. (B) Zooms on important re-organizations of the mutated residues between structures at pH 4 (4HFI-gray) and pH 7 (4NPQ-black). (C) Electrophysiological recordings in oocytes of the three mutants labeled with bimane showing no current in comparison with GLIC presenting sensor mutations only. pH-dependent curves in fluorescence for the three mutants with the sensor Bim136-Q101W (D) and Bim250-Y197 (E). ECD, extracellular domain; TMD, transmembrane domain.

Mutants L157A and L246A reveal unique quenching phenotypes. Combined with Bim136-Q101W, they both show a pH-dependent quenching of fluorescence with a decreased amplitude (∆Fmax) associated with a significant decrease in pH50 as compared to the Bim136-Q101W background (Table 1). In contrast, they only weakly alter the motions at Bim250, which occur with a complete amplitude and small changes in pH50. The mutation H235F leads to a phenotype opposite to that of L157A or L246A. Its Bim136-Q101W pH-dependent curve shows a nearly full quenching amplitude together with a decrease in pH50, while it impairs the motion of the M2-M3 loop, with only a partial pH-dependent de-quenching at Bim250-Y197.

Thus, while those mutants do not have a measurable access to the active state, they still show allosteric motions as revealed by fluorescence. Unlike the moderate loss of function mutants investigated above, these mutations alter the amplitude of the fluorescence curves, revealing profound changes of either the protein conformations and/or allosteric equilibria.

Long-range allosteric coupling between the ECD top and propofol

We further used the TrIQ technique to study the mechanism of action of the general anesthetic propofol, an allosteric modulator of GLIC. Propofol binds to at least three main sites within the TMD: one site in the pore itself near the middle of the TMD, and two sites in the upper part of the TMD at intra- or inter-subunit locations (Figure 5A). Propofol is an inhibitor of GLIC, but it has been shown to be a potentiator of the H235Q mutant (Fourati et al., 2018). We verified that these effects are conserved in the Bim136-Q101W background, with propofol decreasing the pH50 of activation of Bim136-Q101W while increasing the pH50 of Bim136-Q101W-H235Q (Figure 5B and C; Table 1). Fluorescence quenching experiments essentially parallel the electrophysiological data. Addition of 100 μM propofol on Bim136-Q101W decreases the fluorescence pH50 by half a unit, while it increases that of Bim136-Q101W-H235Q by more than half a unit (Figure 5D). Interestingly, a similar pattern is seen on sensor Bim135-W72 (Figure 1—figure supplement 1). Our data thus shows that propofol does act on the global allosteric transitions by displacing the equilibria of both pre-activation and activation. It is noteworthy that propofol is also likely to generate local effects upon binding to modulate the function, which are not investigated here. For instance, its binding into the pore may sterically block ion translocation to produce inhibition (Fourati et al., 2018).

Allosteric coupling between the top of the ECD and propofol binding.

(A) Structure of two monomers of GLIC pH 4 showing positions of the fluorescence sensor Bim136-Q101W at the top of the ECD and three propofol binding sites intra, inter-subunit, and in the pore identified by X-ray crystallography (Fourati et al., 2018). (B) Example of electrophysiological response to 100 μM propofol during a low pH application (scale bars represent 100 nA and 30 s). (C) Electrophysiological pH-dependent curves of Bim136-Q101W with (◆) and without (●) the H235Q mutation showing inhibition and potentiation, respectively. (D) Effect of 100 μM propofol on fluorescence quenching without (top panel) and with H235Q mutation (lower panel) for the Bim136-Q101W sensor. ECD, extracellular domain.

Fit of the data with a three-state MWC model

To characterize the effect of mutations in a more quantitative manner, we fitted the whole data set with a Monod-Wyman-Changeux (MWC) model. Since the fluorescence and electrophysiological pH-dependent curves presented here underlie two major allosteric steps, pre-activation (a fast process causing the changes in fluorescence as previously identified in stopped flow experiments; Menny et al., 2017) and activation (a slower process responsible for channel opening), we used a three-state model where the protein is in equilibrium between a resting R state, a pre-active pA state, and an active A state. Changes in fluorescence, although measured here in equilibrium conditions, actually occur with fast kinetics and are likely not related to desensitization. For the measure of activation, we used peak currents recorded in oocytes, assuming that desensitization would be negligible in these conditions. The putative slow-desensitized state of GLIC was thus not included in this model.

In our allosteric model, we first defined a single proton binding site present in five copies with intrinsic affinities for each state named KR, KpA, and KA. The equilibria between the states at pH 7 are governed by isomerization constants LpA=R/pA and LA=pA/A (see Materials and methods for detailed equations). For each fluorescent sensor (Bim136-Q101W and Bim250-Y197), each allosteric state has a defined fluorescence intensity FR, FpA, and FA. As the model involves numerous parameters that cannot be fitted simultaneously given the available data, we adopted a stepwise strategy (summarized in Figure 6—figure supplement 1). We formulated several reasonable hypotheses to fit some parameters to the experimental data which were kept fixed while others were constrained to change together:

  1. Since the changes in fluorescence occur mainly during pre-activation (Menny et al., 2017), we infer that both sensors display identical fluorescence intensity in the pA and A states (FpA=FA).

  2. A majority of the allosteric mutants followed with the Bim136-Q101W sensor have almost identical fluorescence values at both ends of the pH curves. This suggests that allosteric transitions at low and high pHs are complete and that at pH 7 a majority of proteins are in the R state (R≈1, R representing the fraction of proteins in the R state) while conversely at pH 4 pA + A ≈1. Consequently, for the Bim136-Q101W sensor, FpH7=FR and FpH4=FpA=FA.

  3. The mutations alter the allosteric isomerization constant between states but not the intrinsic affinities for protons.

Setting the pre-activation parameters using total loss of function mutants

We started the fitting procedure with the total loss of function mutants, which do not have access to the A state, and simplify the model to one with two states (R and pA). We note that these mutants producing drastic phenotypes, could profoundly alter the protein conformations, possibly including that of the R and pA states and their intrinsic fluorescence. However, the H235F mutant has been shown by X-ray crystallography to adopt a well-folded conformation, captured in the crystal in a ‘Locally closed’ conformation corresponding to an active-like ECD and resting-like TMD conformation (Prevost et al., 2012; Prevost et al., 2013) from which we infer that the fluorescence from this mutant reports on WT-like motions.

Fitting of the Bim136-Q101W-H235F curves is constrained by two experimental values (pH50 and Fmax) with three variable parameters, KR, KpA, and LpA. In consequence, for each value of LpA, the two other parameters are fully constrained by the experimental data. As an illustration, we fixed LpA=100. After manual fitting of the curves, we were able to extract KR=3.6×10–6 and KpA=1.0×10–6. We then used these KR and KpA values to fit the fluorescence quenching curves of Bim250-Y197-H235F, and sensors Bim136-Q101W and Bim250-Y197, only adjusting LpA (Figure 6—figure supplement 2). The model thus provides a minimal set of parameters accounting for the pH50s and absolute fluorescence changes of these four constructs. It notably suggests that H235F causes a marked stabilization of the R state over the pA state (increase in LPA).

Setting the activation parameters

In a second step, we added the activation state in the MWC model and sought to fit the pH-dependent electrophysiological response curves. Keeping the pre-activation parameters defined above, we found that a three-state model comprising a single proton site could not account for the separation between the fluorescence and electrophysiological curves. With a unique proton site, the model does not allow for more than a fivefold difference, between the curves, when in our experimental data Bim250-Y197, pH50s are, respectively, 5.83 and 4.66, more than 1 order of magnitude difference. To fit the activation curves, we thus added a second proton site (named primed, present in five copies), that specifically drives the activation step (KR’=KpA’, KpA’>KA’), while the first proton site specifically drives the pre-activation step (KR>KpA, KpA=KA). This model is reasonable since it is established that several proton sites are contributing to GLIC activation (Nemecz et al., 2017). Using this three-state two-site model, we found a set of parameters accounting for the pH-dependent curves of the sensors (Figure 6A). With the Bim136-Q101W sensor (Figure 6C), variations in the first half of the fluorescence curve result from the apparition of the pA state which is maximally populated (pA around 0.5) near the pH50 of the fluorescence curve. At lower pHs, the equilibria are further displaced toward the A state, contributing to the decrease in fluorescence in the second half of the curve, and to the parallel apparition of current. With the Bim250-Y197 sensor (Figure 7), the mutation introduced causes a destabilization of the A state over the pA state (increase in LA), displacing the pH-dependent activation curve to lower pHs. The pH-dependent fluorescence curve is consequently mainly caused by the apparition of the pA state with a maximal pA value reaching more than 0.8.

Figure 6 with 2 supplements see all
The three-state MWC model fits experimental data for Bim136Q101W mutants.

(A) Scheme showing the three states and parameters of the model. (B) Table with multiplication factors of isomerization constants for pre-activation and activation as compared to Bim136-Q101W. For the Bim136-Q101W-H235Q in presence of propofol, the multiplication factors are given in comparison with Bim136-Q101W-H235Q. Isomerization constants of Bim136-Q101W and Bim136-Q101W-H235Q are shown above the table (see full table in Figure 6—figure supplement 2—source data 1). (C) Superposition of experimental data points and theoretical curves. Data points shown as spheres correspond to fluorescence intensities normalized on FSDS (blue, ◆), and to electrophysiological response normalized to the maximal current in (black, ●) except H235Q without propofol for which values were normalized to the values in the presence of propofol. Theoretical curves: the population of A state is shown in black lines and the fluorescence curve (blue line) is calculated from the sum of the three states’ fractional populations weighted by their intrinsic fluorescence intensity (see formula in Materials and methods section). For each mutant, the fit from Bim136-Q101W (Bim136-Q101W-H235Q for the last panel) is shown in dotted blue and black lines for a visual comparison and arrows are illustrating the shift in pH50. MWC, Monod-Wyman-Changeux.

The three-state MWC model fits experimental data for total loss of function mutants.

Data are presented as in Figure 6. (A) Multiplication factors of isomerization constants for pre-activation and activation of total loss of function mutants as compared to sensors Bim136-Q101W and Bim250-Y197. (B) Superposition of experimental data points and theoretical curves. Data points (in spheres) correspond to fluorescence intensities normalized on FSDS (blue ◆), and to electrophysiological response normalized to the maximal current (black, ●). Theoretical curves: the population of A state is shown in black lines for sensors and the fluorescence curve (blue line) is calculated from the sum of the three states’ fractional populations weighted by their intrinsic fluorescence intensity. For each mutant, the fit from the associated sensors (Bim136-Q101W or Bim250 Y197) is shown in dotted lines for a visual comparison and arrows are illustrating the shift in pH50. MWC, Monod-Wyman-Changeux.

It should be noted that while the pre-activation parameters are substantially constrained by the experimental data (relying only on the assumption that LpA=100 for the H235F mutant), the activation parameters were chosen arbitrarily and other combinations of affinity and isomerization constants could also fit the data. In addition, the data set itself is heterogeneous since fluorescence experiments were performed on purified receptors, while electrophysiology was done on Xenopus oocytes. Therefore, the activation parameters used here are only meant to evaluate, as a proxy, the relative effect of each mutation on the activation transition.

Differential effects of GLIC mutants on pre-activation versus activation

Based on the sensors’ parameters, we fitted the various mutants by adjusting the isomerization constants of pre-activation and activation (Figures 6 and 7). Overall, reasonable fits can be achieved in most cases. Variations in isomerization constants between mutant and parent sensors were calculated as multiplication factors for pre-activation and activation (Lmutant/Lsensor, Figures 6B and 7A). In parallel, we performed the whole set of fits with different starting values of the H235F LpA constants (1000 and 100,000), yielding different sets of isomerization constants but in each case similar effects of the mutants (Figure 6—figure supplement 2—source data 1, all values presented below were taken from the LpA=100 fit unless indicated otherwise). A discrepancy between data points and fits is however consistently observed concerning the apparent cooperativity of most pH-dependent fluorescence curves. The pH-dependent decreases in fluorescence observed experimentally arise over a relatively large range of pHs, while theoretical curves display sharper shapes. It is possible that the pre-activation transition, modeled here by a single allosteric step, might actually involve multiple steps that are not implemented here. Despite this limitation, the model allows us to highlight clear-cut effects.

Experimentally, the most phenotypically striking mutant is Y28F, which produces a 2 orders of magnitude shift of the pH-dependent curves, and a near equalization of fluorescence and electrophysiological pH50s. Y28F is readily fitted by the simple assumption that the R-pA equilibrium is strongly displaced toward the R state, that is, the R state is thermodynamically stabilized over both the pA and A states, with no changes in the pA to A equilibrium. In this condition, the fluorescent changes are entirely caused by the apparition of the active state, and the fraction of receptors in the pA state, in these equilibrium conditions, remains below 0.1% at every pH. This does not mean that this mutant does not populate the pA state during activation, since the pA state may be kinetically favored and actually appear in a transient manner. Mutant E26Q has a milder phenotype, but the fitting also suggests it has a stronger effect on pre-activation than on activation (multiplication factors of isomerization constants of 15 for pre-activation and 10 for activation, Figure 6B).

In contrast, mutants in the lower ECD (D32E), or in the TMD (H235Q and E222Q), are found to preferentially alter the activation transition, destabilizing the A state over the pA state, with small effects on pre-activation (multiplication factor of isomerization constants for pre-activation: 2, 1.5, and 7 and for activation: 80, 40, and 700, respectively). In consequence, they show a large displacement of the activation curve than that of the fluorescence curve when compared to the parent sensor (Figure 6). On the Bim136-Q101W sensor and H235Q mutant, propofol acts respectively as a negative and positive allosteric modulator (Fourati et al., 2018). For both constructs, propofol is found to have a dual effect, altering principally the activation transition but also the pre-activation transition.

Finally, for total loss of function mutants, while H235F fluorescence quenching could be fitted reasonably well with a two-state R-pA model (Figure 7, Figure 6—figure supplement 2—source data 1) the best fits of L157A and L246A were of lower quality. In particular, pH-dependent curves of Bim136-Q101W-L157A and Bim136-Q101W-L246A are rather flat, the former being better represented by a straight line. The tentative fits are thus not satisfactory, suggesting that these mutants display complex phenotypes, plausibly driving the conformations into states that are not implemented in our model.

Investigation of quenching pairs reorganizations using iMODfit and bimane docking

In our previous study, the various quenching pairs were designed on the basis of the comparison of the X-ray structures of GLIC solved at pH 7 and pH 4, selecting pairs of residues that undergo large changes in backbone Cα distances. GLIC-pH 7 is in a non-conductive conformation with a closed hydrophobic gate in the upper part of the pore, consistent with a resting-like state. The GLIC-pH 4 structure shows in contrast an open gate compatible with a conductive conformation (Cheng and Coalson, 2010; Fritsch et al., 2011; Sauguet et al., 2013; Gonzalez-Gutierrez et al., 2017) consistent with an active-like structure.

However, the orientation of bimane fluorophore and the surrounding residues including the main quencher is not known for Bim136-Q101W and B250-Y197. Their distances should be taken into consideration to propose a more faithful picture of the underlying molecular reorganizations. In addition, the comparison of crystallographic structures alone does not inform us on the time course of the quenching process during the movement. For instance, at Bim136-Q101W and Bim250-Y197, relatively large changes in Cα distance (2–5 Å) are observed between GLIC-pH 7 and GLIC-pH 4, but one can ask how the distances (and quenching) evolve during these movements.

To investigate these issues, we computed approximate trajectories between the two states using iMODfit, and then modeled on them the bimane/quencher pair using a simple docking approach. iMODfit has been originally designed to fit structures inside electron-microscopy envelopes, notably from a very different starting conformation (Lopéz-Blanco and Chacón, 2013). This flexible fitting is made via the deformation of the structure using normal mode analysis (NMA) (see Materials and methods section). NMA approximates the surface of the conformational landscape and decomposes the movements into discrete modes. It takes advantage of a simplified but physically meaningful representation of the interaction between the atoms, based on simple springs connecting close pairs of atoms in the native structure. This method provides a time-independent equation and allows the study of slow (biologically relevant) and collective conformational transitions. NMA has been shown previously to allow the study of pLGIC gating mechanisms (Taly et al., 2005; Bahar et al., 2010). In addition, we have shown on NMDA receptors that iMODfit’s NMA-based fitting process can actually visit biologically relevant intermediate structures (Esmenjaud et al., 2019). The aim of this study is therefore not to capture the fine details of the transition pathway, but to generate plausible trajectories capturing the main features of the conformational reorganization.

Generation of two distinct conformational pathways using iMODfit

Two independent trajectories were computed. Trajectory A (12 frames) starts from the closed GLIC-pH 7 structure to reach the open GLIC-pH 4 structure, and trajectory B (11 frames) starts from the GLIC-pH 4 structure to reach the GLIC-pH 7 structure. Both trajectories are fully reversible and are equally relevant to describe either activation or deactivation, since normal modes deformation can be applied in the two directions. RMSD analysis between each frame and the reference structure indicates gradual reorganization of GLIC across the length of both simulations (Figure 8A). Both trajectories, when visualized from the resting to active state, show three major reorganizations components: a quaternary twist of the pentamer, a ‘central gating reorganization’ comprising opening/closure of the pore, and a quaternary compaction of the ECD.

Two distinct trajectories for GLIC activation computed using iMODfit.

(A) RMSD evolution throughout the frames of trajectories A and B against GLIC structures at pH 4 (●); pH 7 (▲), pdb codes are 4HFI and 4NPQ, respectively. Both trajectories are shown with frame one being the closest to GLIC-pH 7. (B) Twist angle measured throughout the frames on both trajectories. The twist angle is measured by the angle formed between vectors from the centers of mass of the ECD and TMD as defined in Calimet et al., 2013. Each trace corresponds to the trajectory of a single subunit within the pentamer.

In trajectory A, the twist motion occurs in the first half of the trajectory (Figure 8B). This motion describes opposite rotations between ECD and TMD domains, as measured by the twist angle defined by center of mass vectors of the ECD and TMD (Taly et al., 2005; Calimet et al., 2013). The pore reorganization happens in the second half of the trajectory and leads to the opening of its upper part which contains the activation gate, as measured at Ile233 Cα (also named I9’; Figure 9A and B). This motion is associated with a central gating reorganization of the GLIC structure involving: 1/ a tilt of M2 toward M3 (as measured by a decrease in distance between His235 nitrogen and the carbonyl backbone of Ile259; Figure 9A and C), 2/ an outward motion of the M2-M3 loop (measured as an increase in distance between Pro250 Cα and the phenolic oxygen of Tyr197; Figure 9A and D), and 3/ a contraction of the β-sandwich at the bottom of the ECD (measured as a decrease in distance between Cα of residues Asp32 and Gly159; Figure 10). In addition to these two consecutive global motions, the progressive quaternary compaction of the ECD, another crucial landmark of GLIC reorganization, occurs throughout the trajectory. This compaction is quantified through measurement of inter-subunit distances at the top ECD (between Cβ Asp136/Gln101 and Arg133/Leu103; Figure 10A, B and C and Figure 10—figure supplement 1), and the bottom ECD (measured by a decrease in the inter-subunit distance of Cβ Lys33/Trp160, Figure 10—figure supplement 2), indicating a progressive decrease in the distance throughout the frames. It is noteworthy that these inter-subunit distances are highly variable, due to the asymmetric nature of the ECDs of the GLIC-pH 7 structure, where each subunit β-sandwich presents a unique orientation as well as relatively high B-factors (Sauguet et al., 2014). This variability decreases over the frames to reach the structure of GLIC-pH 4 which is compact and essentially symmetric.

Key TMD motions in trajectories A and B.

(A) Snapshots of GLIC TMD top view in the first and last frame of the trajectory A with a Bim250-Y197 quenching pair modeled at one interface. Bimane is shown in blue and Y197 in black spheres. One subunit is shown in gray, the others are in white, the M2-M3 loop is shown in blue, and the loop 2 from ECD is shown in purple for one subunit. Atoms used for measurements are shown in pale blue spheres and distances are indicated in angstroms. (B) Pore radius measured at the Ile233 level. (C) Intra-subunit separation of M2 and M3 helices measured between atoms indicated. Points at positions 0 and 13 are the distances measured in pH 4 and pH 7 X-ray structures. (D) Inter-subunit distances showing M2-M3 loop outward motion at the Pro250-Tyr197 level (top panel) and between bimane and Tyr197 centroids (bottom panel) in both trajectories A and B. ECD, extracellular domain; TMD, transmembrane domain.

Figure 10 with 3 supplements see all
Key ECD motions in trajectories A and B.

(A) Snapshots of two subunits of GLIC ECD in the first and last frame of the trajectory A with a Bim136-Q101W quenching pair modeled at the interface. One subunit is shown in gray, the other in white with sheets of the β-sandwich shown in dark and light purple; bimane is shown in blue and Trp101 in black spheres; Cα and Cβ atoms used for measurements are shown in pale blue spheres and distances are indicated in angstroms. Inter-subunit distances showing ECD compaction measured at the Asp136-Gln101 level (B) and between bimane and Q101W centroids (C) in both trajectories A and B. Points at frames 0 and 13 are the distances in pH 4 and pH 7 X-ray structures. (D) Intra-subunit distance showing contraction at the bottom of the β-sandwich measured by Cα distances between Asp32 and Gly159. ECD, extracellular domain.

Trajectory B shows substantially the same components but with an inverted sequence of events. The central gating reorganization starts first and is associated with an increase in pore radius at Ile233, followed by the twist motion in the second half of the trajectory, the latter being associated with further fluctuations of the pore radius. The ECD compaction is also spread over the whole trajectory. In conclusion, using iMODfit we could generate two distinct trajectories that are in principle equally plausible to describe a gating transition of GLIC activation or deactivation.

Visualization of quenching pairs on iMODfit trajectories

To relate the conformational reorganizations of GLIC to our fluorescence quenching data, we modeled the fluorophore/quenching pairs in both trajectories. To this aim, the cysteine and quencher mutations were modeled and the bimane moiety was docked into each frame while keeping it at a covalent-bond compatible distance to the sulfur atom of the cysteine. The distance between bimane centers of mass and their quenching indole/phenol moieties was then measured in each frame to follow its evolution throughout the trajectories.

For Bim136-Q101W, the procedure shows that Bim136 and the Trp101 indole ring are separated in the resting-like state (first frame of both trajectories), and are in close contact in the active-like state (last frames of both trajectories) (Figure 10A). These observations are in good agreement with fluorescence data that show a decrease in fluorescence intensity upon pH drop reporting a decreased distance within the pair. The trajectory A shows a progressive decrease in distance that parallels the ECD quaternary compaction movement (Cβ Asp136/Gln101). The trajectory B shows a different pattern, characterized by important fluctuations followed by a sharper distance decrease only in the last frames (Figure 10C).

For the ECD-TMD interface quenching pair Bim250-Y197, the procedure shows that Bim250 is in close contact with the Tyr197 phenol ring in the resting-like state, while both moieties are separated in the active-like state, the bimane moiety moving on the other side of loop 2 (Figure 9A). This is also in agreement with the fluorescence data that showed an increase in fluorescence upon pH-drop indicating that the Bim250 is moving away from its quencher Tyr197. Interestingly, both trajectories A and B show an abrupt change in Bim250-Y197 distances, corresponding respectively to a late versus early separation, and these changes occur during the outward motion of the M2-M3 loop (P250/Y197O distance; Figure 9D).

In conclusion, visualizing the quenching pairs using a simple docking procedure shows good agreement with fluorescence. At position P250, data also show a clear switch of bimane from one side of loop 2 to the other during the quenching/dequenching process.

Discussion

Long-range allosteric coupling associated with pre-activation and pore-opening processes

In this study, we revisited several fluorescent sensors by performing detailed pH-dependent quenching curves and parallel iMODfit/docking calculations. Our data clearly support that Bim136-Q101W and Bim250-Y197 sensors are bona fide reporters of the ECD compaction and the outward M2-M3 motion, respectively. In contrast, data related to the Bim135-W72 sensor (presented and discussed in Figure 1—figure supplement 1 and Figure 10—figure supplement 3) show complex patterns of quenching in both in silico and fluorescence experiments. We infer that, because of the buried location of Bim135 within the protein, it is sensitive to subtle structural reorganizations, the complexity of which precludes clear conclusions. This emphasizes that the fluorescence quenching approach requires screening of multiple positions to select the ones reporting on well-defined local motions.

Using these appropriate sensors, we found that a series of five loss-of-function mutations, which shift the pH-dependent electrophysiological curves to higher concentrations, also shift the pH-dependent fluorescence quenching curve of ECD-compaction at the extracellular top of the protein. The ECD-compaction is thus sensitive to mutations scattered along the protein structure down to the opposite cytoplasmic end, indicating substantial allosteric coupling. Since the conformational motions followed by fluorescence occur early in the pathway of activation, it is expected that a shift in the fluorescence curve will be reflected by a parallel shift in the electrophysiological curve. Mutations in the ECD E26Q and Y28F/C27S both present such a phenotype with similar ∆pH50 in electrophysiology and fluorescence, suggesting that those mutations would mainly impact the pre-activation transition. In contrast, D32E, E222Q, and H235Q lead to a stronger pH50 shift in electrophysiology than in fluorescence suggesting that these mutations would alter not only the pre-activation, but also the downstream pore-opening transitions leading to an additive effect on the pH50.

Discriminating pre-activation versus activation phenotypes through allosteric modeling

To interpret the mutant phenotypes in a more quantitative manner, we fitted the whole series of data using a three-state two-site model. We had to implement two proton binding sites to account for the separation of the fluorescence and electrophysiological curves of most constructs. This idea is supported by a mutational analysis that showed that several proton activation sites, located at multiple loci, contribute to activation (Nemecz et al., 2017). In addition, chimeric receptors made up of the GLICECD fused to the TMDs of various pLGICs (Duret et al., 2011; Ghosh et al., 2017; Laverty et al., 2017) or of the ELICECD fused to the GLICTMD (Schmandt et al., 2015) all preserve a proton-gated ion channel function, with the GLICECD-GABAρTMD chimera showing a markedly biphasic pH-dependent activation curve (Ghosh et al., 2017). This suggests that the proton activation sites, whose loci are not known, are scattered throughout the GLIC structure, in both the ECD and the TMD. In our model, we arbitrarily tuned the affinity constants of site 1 to drive the pre-activation transition, and of site 2 to drive the activation transition, to minimize the number of parameters involved.

We also postulated that the various mutants only alter the isomerization constants between states. However, the data set does not allow for the discrimination between effect on binding affinity versus isomerization constants. The effects of mutations on the isomerization constants are thus used here to evaluate the global effect of the mutations on pre-activation versus activation, but it is possible that they actually report on alteration of isomerization constants, affinity constants, or both. Among the various mutations investigated here, E26Q, E222Q, and H235F/Q neutralize the charge of titratable amino acids. It is thus possible that in these cases the mutation eliminates a proton binding site. However, a local impact of a mutation on a proton binding site, or on a set of inter-residues interactions altering the allosteric equilibria, will be equally valid in assigning local structural alterations to pre-active/active phenotypes.

The pattern of effect on LpA versus LA among the various mutants allows us to dissect their allosteric impact. As anticipated from measured ∆pH50, the fits illustrate that Y28F and E26Q principally alter the pre-activation transition and that Bim250, D32E, H235Q, and E222Q principally alter the activation transition, while propofol alters similarly both processes (Figure 11). Concerning the total loss of function mutants, we found that they do preserve pre-activation-like allosteric motions, although with an impaired sensitivity and amplitude of the fluorescence curves. H235F is acceptably fitted according to an R-pA model, suggesting that this mutant isomerizes to a pre-active-like state but cannot isomerize further to the active state. L257A and L246A show a more complex phenotype, but fluorescence data show at least partial pre-active-like motions.

Effect of mutations on pre-activation and activation.

(A) One subunit of GLIC showing the positions of tested mutations in red spheres, with in gray the region involved in the central gating pathway identified by iModFit. (B) Multiplication factor shown on a log scale for each mutant to visualize how isomerization constants LpA and LA were modified in comparison with Bim136-Q101W, or for propofol (H235Q) in comparison with the same mutant without propofol.

Structural reorganizations associated with pre-activation versus activation

Comparison of the GLIC-pH 7 and GLIC-pH 4 X-ray structure highlighted key reorganizations involved in gating (Sauguet et al., 2014), notably a quaternary compaction of the ECD, a tertiary compaction of the β-sandwich in the lower part of the ECD, an outward motion of the M2-M3 loop, and a tilt of the M2 helix toward the M3 helix. Our combined electrophysiological and fluorescence study untangles evaluating the contribution of these specific motions to the pre-activation versus activation transitions.

The ECD quenching pairs at Bim136, Bim133, and Bim33 already showed that pre-activation involves a major quaternary compaction of the whole ECD. We strengthen this idea further by showing that E26Q and Y28F/C27S, that are also located at the subunit interface in the lower part of the ECD, strongly impair the pre-activation process with weaker effects on activation. In addition, the quenching pair at Bim250 showed that the pre-activation involved a key outward movement of the M2-M3 loop. Our data indicate that pre-activation also includes motions of the TMD, since mutation H235Q, as well as propofol binding, are shown to significantly alter pre-activation.

For the activation, our mutational analysis points to a key role of the lower inner part of the ECD β-sandwich (D32E), the M2-M3 loop (Bim250), and the TMD (H235F/Q and E222Q). Interestingly, D32E is involved in strong interactions between sheets in the lower part of the β-sandwich through a salt bridge with R192 (Figure 3A). Mutation, D32E, elongating the side chain by one carbon atom is thus predicted to disfavor the β-sandwich compaction. In addition, at the middle of the TMD, H235 from M2 interacts with the main-chain carbonyl of I259 from M3 through an H-bond favoring the interaction between both helices (Prevost et al., 2012; Rienzo et al., 2014; Figure 3A). Its mutation into Q and F is predicted to weaken or abolish this interaction and disfavor the tilt of M2 toward M3. This assumption is consistent with the X-ray structure of the H235F and H235Q mutants which shows a ‘locally closed conformation’ where M2 and M3 are separated (Prevost et al., 2012; Fourati et al., 2018). Our data thus provide evidence that the compaction of the β-sandwich and the tilt of M2 are principally involved in the activation process.

The mutational analysis also shows for most mutations mixed effects on the isomerization constants of activation and pre-activation, suggesting that both processes involve overlapping regions. The Bim250 position is noteworthy in this respect, since the bimane, reporting an outward motion of the M2-M3 loop, monitors pre-activation, while the modification itself (P250C mutation plus reaction with bimane) principally alters the activation process. It is thus plausible that the M2-M3 loop could move in two successive steps, a first one during pre-activation conditioning dequenching and a second one during activation. In either case, our data further highlight a central role for this loop in ECD-TMD coupling.

Speculative interpretation of the mutant phenotypes in the context of computational trajectories

The transition pathway of GLIC has been previously studied by atomic-level molecular dynamics simulations in an explicit membrane environment. While the timescale of the transition greatly exceeds that of even the longest possible simulations, two studies addressed this issue. The first one started from the GLIC-pH 4 structure and instantly set it to neutral pH, followed by a 1 μs simulation (Nury et al., 2010), yielding concomitant closing of the pore and twist of the whole structure. The second one is based on the string method, using the ‘swarms of trajectories’ approach, computing a trajectory between GLIC-pH 7 and GLIC-pH 4 (Lev et al., 2017). The trajectory shows a sequence of events starting from the closed to the open conformation. A first major reorganization involves the opening of the pore, its hydration, and the compaction of the lower part of the ECD β-sandwich. This is followed by a major reorganization of the ECD, notably its twist and its quaternary compaction. This sequence of events appears hardly compatible with our quenching data, although a comprehensive integration of both sets of data would require extensive in silico investigations of the bimane-labeled mutants to analyze the reorganizations of the quenching pairs. Of note, an important limitation of the method is that it implicitly postulates the occurrence of a single trajectory. However, a coarse-grained simulation (hybrid elastic-network Brownian dynamics) predicted two possible pathways for GLIC gating, that are characterized by different compactions of the ECD (Orellana et al., 2016).

In this study, we performed iMODfit/bimane docking calculations and generated two distinct trajectories with an inverted sequence of events. While these trajectories are coarse and do not implement fine atomistic interactions, they allow the visualization of plausible collective motions in relation to the reorganization of the quenching pairs.

Remarkably, both trajectories show complex quaternary asymmetric reorganizations of ECD compaction. As stated above, ECD compaction is critically involved in pre-activation, a feature consistent with recent work by electron paramagnetic resonance (EPR) spectroscopy (Tiwari et al., 2020) showing a proton-induced inward tilting motion of the ECDs, and recent cryoEM work showing marked structural flexibility of the ECD in the closed-channel state at pH 7 (Rovsnik et al., 2021). Interestingly, fluorescence curves of pre-activation, especially the one of Bim136-Q101W, are endowed with markedly low cooperativities. We thus speculate that this low cooperativity might arise from the contribution of multiple asymmetric intermediate states to the transition, in a manner reminiscent of the asymmetric motions recently described for the desensitization of the GABAA receptor (Gielen et al., 2020).

In addition, both trajectories show a ‘central gating motion’ involving several key concerted reorganizations: a compaction of the lower part of the β-sandwich, an outward movement of the M2-M3 loop, as well as a tilt of the M2 helix toward M3, that involves a marked increase in the opening of the pore. Similar structural couplings are also observed in string simulations (Lev et al., 2017). This observation nicely parallels our finding that these motions are principally involved in activation. We can thus speculate that the central gating motion constitutes the heart of the activation transition.

Concerning the order in which reorganizations are observed, trajectory A is a better fit to the fluorescence data. It suggests a scenario involving, during pre-activation, progressive ECD compaction and beginning of the M2-M3 loop motion, generating the fluorescence variations. Then the M2-M3 loop completes its movement in concert with β-sandwich compaction and pore opening. Future computational studies are needed to explore this possibility.

Consequences on the gating mechanism within the pLGIC family

The conservation of the general gating mechanism between bacterial and eukaryotic pLGICs is well documented by the available structures with the common allosteric regulatory sites for ligands and mutations (Sauguet et al., 2015; Bertozzi et al., 2016; Rienzo et al., 2016), together with the allosteric compatibility between eukaryotic and prokaryotic ECD/TMD domains to form functional chimeras (Duret et al., 2011; Moraga-Cid et al., 2015; Laverty et al., 2017). It is therefore tempting to speculate that the pre-activation transition of GLIC that we characterize here might have counterparts in human neurotransmitter-gated receptors. In this line, some recent structures of eukaryotic receptors including the 5-HT3R (Polovinkin et al., 2018), the GABAAR (Masiulis et al., 2019) and the GlyR (Yu et al., 2021) show pre-active-like conformations characterized by marked agonist-elicited reorganization of the ECD but a closed channel at the TMD. Additionally, the flipped or primed states, where the conformational change of the orthosteric site is predicted to be complete, but where the channel is closed, would fit the functional requirement of a pre-active state (Lape et al., 2008; Plested, 2014).

Our work also investigates the mechanism of action of allosteric mutations by measuring their effects at different levels of the protein, dissecting their phenotype along the gating pathway (Galzi et al., 1996). Allosteric mutations of neurotransmitter-gated receptors, causing congenital pathologies including myasthenia and hyperekplexia have been extensively studied (Taly and Changeux, 2008; Bode and Lynch, 2014; Hernandez and Macdonald, 2019). Most of the hot spots mutated here on GLIC were found associated with pathologies on human receptors. In particular, the lower part of the ECD-ECD interface is the site of a de novo S76R mutation in GABAA α1 (homologous to Glu26) causing epilepsy (Johannesen, 2016) and the mutation L42P in the nAChR δ (homologous to Cys27) causing myasthenia (Shen et al., 2008). This latter mutation (as well as mutation of N41, homologous to E26) decreases activation kinetics and this residue was shown to be energetically coupled to Y127 on the other side of the interface. Interestingly, equivalent residues in GLIC (C27, E26, and Y111) are part of a water network at the bottom of the ECD (Figure 2A). Another noteworthy example is the mutation P250T in GlyR α1 that causes hyperekplexia (Saul et al., 1999) and which is homologous to E222 in GLIC. Interestingly, on the glycine receptor α1, other mutations have been studied by single-channel recordings and are described to affect principally a flip pre-activation-like step for A52S in the loop 2 at the ECD-ECD interface (Plested et al., 2007) or gating for K276E on the M2-M3 loop (Lape et al., 2012). These data suggest that mutations produce similar allosteric perturbations on GLIC and GlyR in those regions.

Our work on GLIC provides general mechanisms of how mutations affect pLGICs transitions and further documents conformational changes, beyond information provided by structures. Further work, for example by voltage-clamp fluorometry, would be required to challenge such mechanisms in the context of congenital pathologies on neurotransmitter receptors.

Materials and methods

Key resources table
Reagent type (species) or resourceDesignationSource or referenceIdentifiersAdditional information
Gene (Gloeobacter violaceus)glvI, GLICUniProtQ7NDN8
Strain, strain background (Escherichia coli)BL21(DE3) C43Sigma-AldrichCMC0019Chemically competent cells
Biological sample (Xenopus laevis)Xenopus oocytesCentre de Ressources Biologiques Xénopes (Rennes- France) and Ecocyte Bioscience (Dortmund-Germany)
AntibodyAnti-HA Tag (rabbit)EuromedexHA-1A1-20 µL(1:200)
AntibodyAnti-rabbit – Alexa Fluor 645 (goat)Molecular probesA21246(1:1000)
Recombinant DNA reagentPet20b-MBP-GLICBocquet et al., 2007
Recombinant DNA reagentpMT3-GLIC-HAtagNury et al., 2011
Recombinant DNA reagentPmt3-GFPNury et al., 2011
Chemical compound, drugMonobromo-BimaneThermo Fisher ScientificM1378
Chemical compound, drugBunte salt BimaneMenny et al., 2017
Chemical compound, drugPropofolSigma-AldrichY0000016
Software algorithmiMODfitLopéz-Blanco and Chacón, 2013
Software algorithmMOLEonlinePravda et al., 2018
Software algorithmClampfitMolecular devices
Software algorithmAxoGraph Xhttps://axograph.com/

Mutagenesis

All GLIC mutants were obtained using site-directed mutagenesis on the C27S background of GLIC, except Bim136-Q101W-Y28F (C27) for which the endogenous cysteine was introduced back. Similarly to previous studies (Sauguet et al., 2014; Menny et al., 2017; Nemecz et al., 2017), two different vectors were used: a pet20b vector with GLIC fused to MBP by a linker containing a thrombin cleavage site under a T7 promoter for expression in Escherichia coli BL21; a pmt3 vector for expression in oocytes with GLIC containing a Cter HA tag and in Nter the peptide signal from α7-nAChR. Incorporation of the mutations in both vectors was verified by sequencing.

GLIC mutants production and purification

Request a detailed protocol

Protein production of MBP-GLIC and labeling was done as previously described (Menny et al., 2017) with a few modifications. In brief, MBP-GLIC was expressed in BL21 E. coli cells overnight at 20°C after induction by 100 µM IPTG. Cells were collected and resuspended in buffer A containing 20 mM Tris; 300 mM NaCl at pH 7.4 and subsequently disrupted by sonication. After membrane separation by ultracentrifugation, membrane proteins were extracted overnight in buffer A supplemented with 2% DDM. After ultracentrifugation, supernatant was incubated with amylose resin and MBP-GLIC was eluted using buffer A supplemented with DDM 0.02% and a saturating concentration of maltose. To remove the endogenous maltoporin contaminant, a first size exclusion chromatography was performed on superose 6 10/300 GL in buffer A with 0.02% DDM. GLIC-MBP concentration was measured and the protein was incubated overnight at 4°C with thrombin to cleave off MBP and with monobromobimane (mBBr) at a 1:5 (GLIC monomer:fluorophore) ratio, to label the protein. The mBBr dye being solubilized in DMSO, the sample volume was adjusted to remain below 1% DMSO final concentration. After labeling, a second gel filtration was done to get rid of the MBP and unbound dye molecules. GLIC-Bimane samples were flash-frozen in liquid nitrogen and stored at –80°C prior to fluorescence measurements.

Steady-state fluorescence measurements

Request a detailed protocol

Fluorescence measurements were done as previously described (Menny et al., 2017). Samples were equilibrated to room temperature and diluted with buffer A with 0.02% DDM to reach a concentration around 40 µg.ml–1. Fluorescence recording buffers consisting of 300 mM NaCl, 2.7 mM KCl, 5.3 mM Na2HPO4, and 1.5 mM KH2PO4 were prepared beforehand and their pH was adjusted either to 7.4 or to different pH in order to reach the desired pH value (from pH 8 to 3) after mixing equal volumes with buffer A 0.02% DDM. Measurements were done at 20°C in 1 ml disposable UV transparent 2.5 ml cuvettes in a Jasco 8200 fluorimeter with 385 nm excitation wavelength and the emission spectra were recorded through 2.5 nm slits from 420 to 530 nm. Parameters were kept constant throughout the study. On the sample at pH 7.4, an addition of SDS to reach 1% final concentration was done to obtain the FSDS value and a tryptophan emission spectrum was done before and after SDS addition in order to monitor denaturation.

Fitting of fluorescence measurements was done on each fluorescence series (values from 1 pH range) with at least three series per mutant using the following Hill equation:

y(x)=ΔFmax+xnHxnH+EC50nH+F0

where ∆Fmax represents the maximal change in fluorescence amplitude, F0 represents the initial fluorescence at pH 7.8; nH represents the Hill number, and EC50 represents the proton concentration for which half of the maximal fluorescence change is measured. For Bim136-Q101W and Bim250-Y197 and in some other mutants, we excluded from the fit the data point below pH 3.5 that show a small but significant change in fluorescence intensity in the opposite direction to the quenching curves. We did not fit the Bim135-W72 mutant that shows a bell-shaped curve.

Electrophysiological recordings

Request a detailed protocol

Electrophysiological recordings of GLIC were made on Xenopus oocytes provided either by the Centre de Ressources Biologiques Xénopes (Rennes-France) or by Ecocyte Bioscience (Dortmund-Germany). Recordings were made as previously described (Nury et al., 2011) with oocytes 48–96 hr post nucleus injection with a mix containing 80 ng.µl–1 of GLIC cDNA and 25 ng.µl–1 of GFP cDNA. Recordings were done in MES buffer containing 100 mM NaCl, 3 mM KCl, 1 mM CaCl2, 1 mM MgCl2, and 10 mM MES with pH adjusted by addition of 2 M HCl. The perfusion chamber contained two compartments and only a portion of the oocyte was perfused with low pH solution. Bunte salt bimane labeling was performed prior to recording by incubation for 1 hr at room temperature with the dye concentrated at 1 mM in MES buffer. To correct data for rundown, a solution with a pH value in the middle of the pH range (usually pH 5) was used as a reference at the beginning and the end of the recording and every 3/4 applications. To limit the effect of propofol that can stay in the membrane in-between applications (Heusser et al., 2018), only a limited number of pH solutions were tested per oocyte.

Electrophysiological recordings were analyzed using AxoGraph X and Prism was used to fit individual pH-dependent recording using the Hill equation:

yx=Imax+xnHxnH+EC50nH

where Imax represents the maximal current in percentage of the response from the reference solution. nH represents the hill number and EC50 represents the proton concentration for which half of the maximal electrophysiological response is recorded.

Xenopus oocytes immunolabeling

Request a detailed protocol

Mutants generating currents smaller than 500 nA at high proton concentrations (pH 3.5) were categorized as non-functional. For these non-functional mutants, expression tests were performed by immunolabeling of oocytes as previously described (Prevost et al., 2012; Sauguet et al., 2014). 34 days postinjection, GFP-positive oocytes were fixed overnight in paraformaldehyde (PFA) 4% at 4°C. Immunolabeling was performed after 30 min saturation by 10% horse serum in phosphate-buffered saline. Rabbit anti-HA-tag primary antibody was incubated for 90 min in 2% horse serum and the secondary antibody anti-rabbit coupled to Alexa Fluor 645 was incubated for 30 min. After a second PFA fixation overnight, oocytes were included in warm 3% low-melting agarose and 40 µm slices were made using a vibratome on a portion of the oocyte. Several slices per oocyte were mounted on a slide and analyzed in an epifluorescence microscope using constant exposure time between non-functional mutant and functional mutants used as positive controls.

Molecular modeling

Request a detailed protocol

The iMODfit flexible fitting method (Lopéz-Blanco and Chacón, 2013) searches the conformational space using the lowest normal modes for the best cross-correlation fit of a starting conformation atomic model into a target conformation density map. Two trajectories were generated here. In trajectory A structure, 4NPQ (GLIC-pH 7) is fitted to the density of 4HFI (GLIC-pH 4), and in trajectory B structure, 4HFI is fitted to the density of 4NPQ.

The detailed procedure is performed as follows, taking as an example trajectory A: 1/ A computed EM density map was generated for the X-ray structure of the target 4NPQ using the pdb2vol tool (called 4NPQ map). The EM density map resolution was set to 5 Å and the grid size to 0.5 Å, that is, the resolution was set at a relatively large value to avoid being locked in local minima during the iMODfit procedure.

2 / 4HFI was represented with the detailed all heavy-atoms force field (all atoms are considered except hydrogens), called the 4HFI model.

3/ The lowest-frequency NMA-modes of the 4HFI model were computed. For the subsequent steps, the range of modes considered (−n option) was set to 0.5, that is, half of the modes, corresponding to the lower frequency modes, are considered for computing the conformational changes.

4/ During the iMODfit procedure, starting from the 4HFI model, 10% of the modes are randomly selected and used to generate a very small conformational change. The new conformation is used to compute a simulated density map, and the new conformation is accepted only if the cross-correlation between the targeted 4NPQ and simulated maps improves. This process is repeated iteratively until the conformation deviates by an RMSD of 0.5 Å from the starting/previous model, in which case an intermediate structure is generated and stored. The entire process is then repeated iteratively to generate a series of intermediate states that progressively converge to the targeted structure.

The geometry of the ion channel has been computed with MOLEonline webserver (mole.upol.cz), with the ‘pore’ mode (Sehnal et al., 2013). We used the FreeRadius value computed at the level of the I9’ residue.

For the Bimane docking procedure on each intermediate structure, the position of side chains was first optimized with the software Scwrl4 (Krivov et al., 2009) while keeping the main chain rigid. This step also allowed the introduction of point mutations. The structure of the protein and bimane was converted to pdbqt files with the software open babel 2.4.1. Covalent docking was then performed with the software smina (Koes et al., 2013). The box for docking has been defined around the mutated cysteine residue, with a size of 30 Å in each direction. Covalent docking forced the bimane to be in appropriate distance with the sulfur atom of the introduced cysteine. Only the first pose was kept for further analysis.

MWC model building

Request a detailed protocol

To build a three-state MWC model, the following equations were used to obtain the population of each state resting, pre-active and active:

A¯=(1+α)5×(1+α)5(1+α)5×(1+α)5+LpALA(1+CpACAα)5×(1+CpACAα)5+LA(1+CAα)5×(1+CAα)5
pA¯=LA(1+CAα)5×(1+CAα)5(1+α)5×(1+α)5+LpALA(1+CpACAα)5×(1+CpACAα)5+LA(1+CAα)5×(1+CAα)5
R¯=LpALA(1+CpACAα)5×(1+CpACAα)5(1+α)5×(1+α)5+LpALA(1+CpACAα)5×(1+CpACAα)5+LA(1+CAα)5×(1+CAα)5

With constants defined below:

LpA=RpH8¯pApH8¯CpA=KpAKRCpA=KpAKRα=[H+]KA
LA=pApH8¯ApH8¯CA=KAKpACA=KAKpAα=[H+]KA

The weighted fluorescence value was calculated as followed:

F=R×FR+pA×FpA+A×FA

With fluorescence values set at:

FRBim136Q101W=0.70FRBim250Y197=0.56
FpABim136Q101W=0.30FpABim250Y197=0.92
FABim136Q101W=0.30FABim250Y197=0.92

Isomerization constants were manually adjusted to fit theoretical and experimental fluorescence quenching curves and normalized electrophysiological curves. Of note, the fluorescence variations of E26Q mutant were normalized to that of the Bim136-Q101W, to correct for its effect on the fluorescence at pH 7 which likely reflects an alteration of the structure of the resting state, independently of the allosteric transitions. Additionally, the A´ population for the Y28F mutant does not reach 1, so it was normalized in order to compare the values with the normalized experimental data.

Data availability

Table 1 included in the manuscript correspond to a summary table for figures 4 to 8.

References

Decision letter

  1. Cynthia M Czajkowski
    Reviewing Editor; University of Wisconsin, Madison, United States
  2. Olga Boudker
    Senior Editor; Weill Cornell Medicine, United States
  3. Andrew JR Plested
    Reviewer; Humboldt Universität zu Berlin, Germany
  4. Grace Brannigan
    Reviewer; Rutgers, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

How structural motions in pentameric ligand gated ion channels(pLGICs) lead to functional channel gating transitions remains poorly understood. Using the prototypic bacterial proton-gated channel GLIC, fluorescent reporters of protein conformation and kinetic modeling, the authors found a set of mutations that mostly alter pre-gating transitions and others that mainly alter gating (pore opening). Using structural trajectories identified by normal mode analysis to interpret their data in structural terms suggests that pre-activation transitions involve quaternary compaction of the extracellular domain and that activation involves a re-organization of a central gating region. This paper adds new mechanistic information about pLGIC activation.

Decision letter after peer review:

Thank you for submitting your article "Mutational analysis to explore long-range allosteric coupling and decoupling in a pentameric channel receptor" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Olga Boudker as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Andrew J R Plested (Reviewer #1); Grace Brannigan (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

How structural motions in pLGICs lead to functional channel gating transitions is not well understood. In a previous study (eLife 2017), this group used time-resolved fluorescence quenching experiments in a prokaryotic pLGIC, GLIC, and identified motions in the extracellular domain (ECD) that occur prior to channel opening. In this manuscript, the authors extend these studies. To interpret their previous fluorescence quenching data in structural terms, the authors used normal mode analysis to identify structural trajectories underlying GLIC closed to open channel gating transitions. The simulations yielded two pathways leading from closed to open channel end states. To monitor whether motions in the ECD are coupled to motions in the TMD (pore opening), the authors used mutations and the allosteric drug modulator propofol to perturb GLIC gating transitions and compared the effects of these perturbations on proton-induced current responses and steady-state fluorescence quenching data. Based on their data, the authors conclude that they have identified new structural conformations and possible new allosteric pathways during gating, and that GLIC and its mutants have access to a large repertoire of conformational states.

While the data are interesting, the overall concern is with interpretation of the data. The logic of the arguments is not clear enough to support the conclusions. Comparisons between theories are not precise nor quantitative enough to support claim of a new gating model. Simulation description and limitations of this approach are not well described. As written, it is difficult to follow the authors line of reasoning and authors do not discuss alternative mechanisms that can fit their data. Significant re-writing and new analyses to make the arguments clear and focused on what novel contributions the data in this paper are providing are needed. The manuscript would be significantly strengthened by addressing the following major concerns.

Essential revisions:

1) A 1-D, steady state signal (quenching) is used to discern multiple states (meaning, functional conformations, division in space) because it takes multiple values. But the signal is the infinite time average over all states, in different conditions. How do we know that the signal doesn't represent different balances of occupancies of the same small number of states (such as 2 states). There are no apparent reasons why results could not be explained from mutants that change open-closed equilibrium as from new conformational states. The authors need to fully discuss and address this alternative mechanism for explaining the data.

2) What experiments provide direct evidence for intermediates? Can the intermediates states represent abstract transitional milestones, not concrete conformations? That interpretation would be consistent with what has been long established in protein folding, but would entail a major shift in how structures are interpreted, because there would not be a meaningful way to connect conformations to kinetic models.

3) Description of the iMOD fit simulations is unclear and does not provide enough detail. How many simulations were run to yield the two trajectories? It is not clear whether iMod-Fit returns two trajectories from a single calculation or whether multiple simulations were run and the trajectories were clustered. In the methods section (page 25, lines 25-26), trajectory A conformation change is from closed to open whereas trajectory B conformation change is from open to closed. However, in the figures, for both trajectories, frame 1starts at GLIC-pH7 (closed).

4) Since their previous time-resolved fluorescence quenching data (eLife 2017) demonstrated that motions monitored at Bimane-136 (ECD β sandwich compaction) and Bim-250 (M2-M3 loop motion) mainly occur prior to channel opening (activation scheme in Figure 4B), it is unclear why the authors conclude that their Bim-250 data support pathway A identified from their simulations. In pathway A, Bim-250-Y197 unquenching happens in same time frame as the increase in pore radius at the 9' position, whereas in pathway B the unquenching of Bim250-Y97 happens before complete pore radius dilation (Figure 2B, 2D). The experimental data at this position seems compatible with the B trajectory. Moreover, the unquenching of Bim-136-W101 appears consistent with either trajectory (Figure 3B) and occurs before the simulated pore dilation. It is unclear what new information the simulations are providing except that the steady-state fluorescence quenching data show good agreement with the simulated end states.

5) To monitor how motions in the ECD are coupled to motions in the TMD (pore opening), the authors used mutations and the allosteric drug modulator propofol to perturb GLIC gating transitions and compared the effects of these perturbations on proton-induced current responses and steady-state fluorescence quenching data. Since the fluorescence data are steady-state, whether a mutation causes an effect on the timing of the structural change in the gating pathway or an effect on the percentages of different conformational states in the ensemble at steady-state is not known and confounds data interpretation. The authors need to discuss this point.

5) Data supporting the statement that mutations of H235, L157 and L246 lead to new global conformations are limited. One could argue that these mutations have dramatic functional effects (eliminate current) and thus, it is not all that surprising that alternative conformations might be adopted that normally would not be visited.

6) Data to support the claim that propofol specifically affects the pre-activation step are limited. Propofol could affect open-closed equilibrium.

7) It is striking that the fluorescence quenching profile for the Q235 mutant with propofol closely mirrors that of H235 without propofol (Black vs light/dashed green lines in Figure 8D). The addition of propofol reverses the effect of the H235Q mutation on structure, not just qualitatively, but close-to-quantitatively, for both the Bim136-Q101W and Bim135-W72 sensors. This is remarkable, especially for the latter sensor where the curve is complex. Yet the discussion treated the two sensors differently despite a similar reversal of the effect of the H235Q mutation, and the authors say the data for Bim135-W72-H235Q cannot be interpreted in structural terms. This explanation is confusing, especially since the corresponding figure is given equal weight in the paper. The authors need to revisit this section to improve clarity or move the mutant results and discussion of explicit technical limitations to supplementary information.

8) The authors claim that their results "challenges the conventional concept that receptor activation involves a single conformational pathway." Do the authors believe their results are inconsistent with the 4 state Heidemann and Changeux models from the 1980s?

9) TMD motions were not measured. Monitoring motion of the M2-M3 loop does not monitor changes in the TMD or pore opening. In previous work, authors used Bimane243 to monitor M2 motions. Additional reporters of TMD motions would be helpful.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Mutational analysis to explore long-range allosteric coupling and decoupling in a pentameric channel receptor" for further consideration by eLife. Your revised article has been reviewed by 2 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Olga Boudker as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Andrew J R Plested (Reviewer #1); Grace Brannigan (Reviewer #2).

We appreciate the authors efforts in responding to the previous critiques and in revising the original paper. The revised paper is improved and the reviewers feel that no additional experiments are required. The paper has changed considerably. The biggest concern is that the revised manuscript, as written, is still difficult to follow, even for experts in the field, making it difficult to evaluate and appreciate its impact. The manuscript requires considerable editorial work to improve clarity. The specific points are:

1. The interpretation of the data with propofol is still a weak point. For instance, the manuscript says that "Our data thus shows that propofol does not act locally by altering the conformation of the TMD, but rather acts on the global allosteric transitions by displacing the equilibria and preserving ECD-TMD coupling" However, the preceding discussion is minimal. I spent a while trying to piece together how the data show that propofol's mechanism of action is not local. It is clear that the binding of propofol does affect the conformation far from the TMD, but this sentence implies that the authors have shown that TMD conformational change is insufficient for action. In other words, I could not figure out how the authors showed that ECD conformational shifts were necessary for propofol to act. If I am missing something, a few more sentences spelling out the logic here would be helpful or rewording of conclusions is needed.

2. The new Figure 9 is so full of important information that it gets hard to follow. I request that all the isomerization constants be compiled in a table for ease of comparison. The caption for that table can then refer to the equations and methods from which they were derived and differentiate between normalization schemes. While it may be useful to keep them in Figure 9 as well, Figure 9 is already overwhelmingly complex. I suggest either having fewer plots or fewer elements per plot. The arrows from one plot to the next were also not intuitive. Mainly, I request that the authors look at this figure with fresh eyes and consider breaking it up or streamlining it somehow.

3. Line 37 "Seminal work in the 80s showed that a minimal four-state model describes the main allosteric properties of the muscle-type nAChR (Heidmann and Changeux, 1980; Sakmann et al., 1980)."

First, the word seminal is best avoided; it is rather outdated. I would not take these similarly out-of-date works as a benchmark, rather use them as a counterpoint – perhaps you can say: "although xxx work in the 1980s, …" and then introduce the updates?

4. Line 66 "However, the physiological relevance of these structures or their assignment to particular intermediates or end-states in putative gating pathways remains ambiguous and poorly studied."

This is a very important point and underlines the importance of the work at hand.

5. Line 69 "Conversely, it is likely that key conformations, unfavored by crystal packing lattice or under-represented in receptor populations on cryo-EM grids, are missing in the current structural galleries."

On the other hand, this is overstated. I would say "possible", not likely. Intermediates might be missing, but are they "key"?

6. Line 86 "much faster than ionic current measurement that occurs in the 30-150 millisecond range "

Much faster than the rise time of population or ensemble currents.

7. Line 116 "Two independent trajectories, A and B, were computed starting from each of the two end-state structures and divided into 12 and 11 frames respectively. "

I appreciate that the authors tried to explain better now, but this is still somewhat opaque. Please just say one trajectory goes from rest to active, and the other from active to rest. The use of "end-state structures" is confusing. They are both end and start, depends on which trajectory it is. Or are there really four structures – two crystals and two end states? Some of the figures suggest that each trajectory does not conclude in really the right place. I can't say which one because I have no idea what figure is which (figures not numbered and some do not correspond to figure legend order).

8. With iModFit, I think it is important to discuss how plausible it is that the transitions are not ergodic. This is mentioned in the discussion. In one way, we should not take these trajectories too seriously. But it is also important to consider the possibility that they are pulling out important information from the structures. Are the authors suggesting that the isomerisations are, preferentially, not reversible? I mention below that it would be a great future insight to have non-equilibrium data that could report the non-reversible motion at some of these sites.

9. Later in the paper, the text again makes me feel like I don't understand what iModFit does.

Line 176 "For the ECD quenching pair Bim136-Q101W, the simulations show that Bim136 and the Trp101 indole ring are separated in the resting-like state, and are in close contact in the active-like state (Figure 3A). "

The simulations? Are you referring to the docking results? The resting and active-like states are from structures, aren't they, not from iModfit? If the states used are from structures, there is little predictive power from the docking to these structures that couldn't be deduced by eye, is there? Or are you comparing the state at the end of the iModFit run, which isn't the other crystal state?

Surely iModFit (simulations?) only tells you about the trajectories of the fluorophores? This time-order of transitions between distances is interesting. But why is it mixed up with end state information (surely known from PDB)?

At the very least, a better description is needed. Overall, I still do not understand how the iModFit trajectories help to understand the steady state fluorescence.

10. Line 155 "In conclusion, using iMODfit we could generate two distinct trajectories that are in principle equally plausible to describe a gating transition of GLIC activation. "

But a really key point that doesn't really come up, but I think it should, is that the different trajectories really consist of at least two steps, Twist and the central gating motion, but they occur in different orders. This is a clear appeal to the intermediate states like flip and prime, and motivates the rest of the paper. The role of the compaction is less clear. If there were not distinct movements, the hysteresis in the motions would be much harder to understand. Still, the connection of the iModFit to the steady-state data is less convincing than any non-equilibrium data would be. This is the distinction between a plausible model (as the authors present) and evidence. The change of fluorescence at given sites should have different orders for activation and deactivation, shouldn't it? This would be worth mentioning. It is something for the future of course. And this is not to diminish the insight from the steady-state measurements.

11. Additionally, the flipped state, where the conformational change of the orthosteric site is predicted to be rather complete, but where the channel is closed, would fit the functional requirement of a pre-active state (Lape et al., 2008).

This is trivial because the flip state is just the name of a non-open agonist bound state. Also, flip is not the only type of state that fits, they might all be the same, from different perspectives. I wrote a comment about this once: "Don't flip out: AChRs are primed to catch and hold your attention"

12 The quantitative details of the fitting and the agreement or otherwise seem reasonable but I cannot claim to check in detail, I'm afraid. The individual conformations have various proton bound states and equilibria, so the 3-state model is quite a bit more complicated than at first sight. It might be nice to include the full model (to indicate the assumptions) in a supplementary figure. If, as the authors say, a relatively complicated proton binding scheme is needed to describe even equilibrium data, this is something of a find and needn't be buried. I don't think we have many ideas about how may protons are needed to gate.

13. I did notice that the Y28F mutant has the biggest change in the Pre-open constant. This selective effect was the case for the nearby A52S mutant in the glycine receptor – a big change in flip, no change in the main gating constants (Plested et al., 2007). Quite different at the K276E below that (Lape et al., 2012). But there are tons of mutants on these positions, maybe there are better ones to compare.

14. Table 1 statistics. Multiple mutants are being compared to the same reference value. Unpaired t test is not the correct test to use for these data. Investigators should use an ANOVA with a posthoc test such as a Dunnett or Bonferroni.

15. MWC fitting of the fluorescent and current data is used to conclude that mutations at the ECD alter the pre-activation step while those at the ECD-TMD interface and TMD alter the activation step (gating). Due to the assumption that the mutations do not effect proton affinity to the sites, the authors need to be careful about overinterpreting the data. The modeling provides support but is not conclusive.

16. How the fluorescence quenching data relate to motions identified by iMODfit is not obvious. On page 10, lines 315-318, the authors state "the fluorescence and electrophysiological pH-dependent curves presented in this paper underlie two major allosteric steps, pre-activation (a fast process causing the changes in fluorescence as previously identified in stopped flow experiments and activation (a slower phase). Based on their 2017 eLife paper, the bim136 fluorescence reports early pre-gating motions, and bim250 reports early pre-gating motions and some later motions. In the revised manuscript, based on iMOD fit/normal mode analyses, the authors state that bim136 is monitoring a quaternary compaction of the ECD that is occurring throughout the gating cycle and that bim250 is monitoring motion of m2-m3 loop which is part of the 'central gating reorganization' including opening of pore (see page 14, lines 452-455). Later in the paper (page 16, lines 528-529) they state 'ECD compaction is critically involved in pre-activation". This is confusing and requires additional explanation and discussion. If the fluorescent reporters at these positions are monitoring fast, early pre-gating motions then why is the quaternary compaction and m2-m3 loop motion part of the central gating reorganization? Am I missing something?

17. In the abstract, the authors state that 'preactivation involves major asymmetric quaternary motions of the extracellular domain'. It is unclear to me what experimental data support this conclusion. Is this based on the starting pH7.0 crystal structure? The authors need to clarify if the asymmetry that they are describing is at the subunit level or is based on two different motions in the ECD (twisting and compaction). Without strong experimental evidence for asymmetric motions, this conclusion should be removed from the abstract.

18. They use iModFit and NMA as synonyms in some parts of the paper, which causes confusion. iMODfit/Normal mode analysis treats the protein like a 3D elastic network. It doesn't capture interactions with solvent or specific residue-residue interactions. It superimposes multiple local low-energy fluctuations to find likely larger scale conformational fluctuations. You would get asymmetry when it costs less total energy to move a few chains by a lot, than to move all of them by a little. The more chains the protein has, the more likely it is that imodFit will find asymmetry. I'm not sure the imodfit simulations add much regarding asymmetry, but the expected behavior of these macromolecules at room temperature makes it an uncontroversial claim, albeit one without significant new evidence.

19. In revised manuscript (page 5, lines 155-157), authors state 'using iMODfit we could generate two distinct trajectories that are in principle equally plausible to describe a gating transition of GLIC activation'. Additional discussion spelling out the logic here is essential.

It is important for the reader to understand the limitations of iModFit, and for a non-computational reader to know what iModFit is not. The authors need to add further discussion in methods or result sections. It is not a physics-based simulation technique like molecular dynamics – I'd call it a numerical approach for generating hypothetical pathways, and then experiments or simulations need to distinguish between them. They have used it here as a conceptual framework. The software itself is not designed to generate trajectories, but to generate structures. Motions or trajectories generated by Normal Mode Analysis are always reversible. Then imodfit applies a bias on top of that, based on the structure, to get a directional trajectory. They applied two different biases (based on two different structures) so they ended up with two different hypothetical and reversible trajectories.

In their response letter, the authors state "one simulation is starting from the closed conformation to reach the open conformation, and the other from the closed to the open. Both trajectories represent plausible pathways for activation and deactivation." It is unclear whether the authors think that simulation A describes activation (closed channel to open channel) pathway and simulation B describes deactivation pathway (open channel to closed channel) or if they think that both trajectories can describe activation (closed channel to open channel)? Please clarify.

20. The authors should discuss and compare their results from iMODfit/NMA analyses to results from Toby Allen lab (PNAS 2017) using all-atom molecular dynamics with a string method to solve for GLIC gating pathways. What new information has been gained from the iMODfit/NMA?

21. Figures 3 supplementary 1 and 2 and 3 are in in different order compared to figure legends and text on page 6 lines 174-175. Authors need to check the order of the supplementary figures. It would be helpful if figures were labeled for review purposes.

22. Abstract should state which experimental results support their conclusions and describe the novel contributions that the data are providing.

https://doi.org/10.7554/eLife.60682.sa1

Author response

While the data are interesting, the overall concern is with interpretation of the data. The logic of the arguments is not clear enough to support the conclusions. Comparisons between theories are not precise nor quantitative enough to support claim of a new gating model. Simulation description and limitations of this approach are not well described. As written, it is difficult to follow the authors line of reasoning and authors do not discuss alternative mechanisms that can fit their data. Significant re-writing and new analyses to make the arguments clear and focused on what novel contributions the data in this paper are providing are needed. The manuscript would be significantly strengthened by addressing the following major concerns.

We thank the referees for their evaluation and the insightful general comments. As you will see bellow, we generally agree with most comments, and thus revisited profoundly the interpretation of the data in a more quantitative manner.

To do so, we performed a modeling of the whole set of pH-dependent curves using a 3-states Monod-Wyman-Changeux Model (including resting, pre-active and active states). MWC-type models, calculating the fractional population of the different states in equilibrium conditions, is indeed well suited to simulate the steady state data of fluorescence quenching and electrophysiology (recorded at the current plateau). This modeling procedure is now extensively described at the end of the result section, including two new figures 9 and 10. The resulting set of parameters provides a reasonable and unifying description of most mutant phenotypes. It exemplifies in a quantitative manner the major conclusion that mutations at the ECD principally alter the pre-activation step, while those at the ECD-TMD interface and TMD principally alter the activation step. However, the model also accounts for the phenotype of non-functional mutant (especially H235F), challenging our initial interpretation that those mutants are stabilized in “unique intermediate conformations where motions are decoupled”. Consequently, we do not any more focus the article on “the possibility of multiple conformational pathways during gating”, but rather on the input of our experimental data on the understanding of pre-activation and activation mechanisms. In this context, we further revisited the iModfit data. We agree that the available data do not allow for the discrimination between the two generated trajectories. However, it allows for the assignment of mutant phenotypes to specific protein motions, giving overall a very coherent picture of the key reorganizations involved in pre-activation and activation. Most of the Discussion section was thus rewritten.

Finally, to simplify the flow of the text, we removed from the main text all data related to the Bim135 mutant, since the complexity of the fluorescence data at this level precludes clear conclusions. We however kept this material in the Figure 4 – Supplementary1 to further document effects of mutations and the robustness of the effect of propofol.

Overall, thanks to the referee’s comments, we think we provide now a clearer and much less speculative interpretation of the data. The revised version combines a unique and solid set of fluorescence/electrophysiological data, which are interpreted in the framework of NMA and MWC modeling and which highlight key proteins motions involved in GLIC gating.

Essential revisions:

1) A 1-D, steady state signal (quenching) is used to discern multiple states (meaning, functional conformations, division in space) because it takes multiple values. But the signal is the infinite time average over all states, in different conditions. How do we know that the signal doesn't represent different balances of occupancies of the same small number of states (such as 2 states). There are no apparent reasons why results could not be explained from mutants that change open-closed equilibrium as from new conformational states. The authors need to fully discuss and address this alternative mechanism for explaining the data.

In the first version of the article, we interpreted the phenotypes of non-functional mutants proposing that they are stabilized at low pH in new unorthodox conformations. We agree with the general comments of the referee that this idea, while plausible, is one possibility among others, the current data not allowing to draw a firm conclusion. We thus completely removed this idea in the new version.

Following the referee’s advice, we thus tried to fit the entire set of data with a three-state allosteric model, that represents the minimal number of states required to account for the separation of the fluorescence and electrophysiological curves. We show that this model reasonably accounts for the majority of mutants, except for two non-functional ones, L157A and L246A. This may suggest that indeed these mutants adopt unorthodox conformations. However, this idea, still very speculative given the current data, is no more emphasized in the new version.

2) What experiments provide direct evidence for intermediates? Can the intermediates states represent abstract transitional milestones, not concrete conformations? That interpretation would be consistent with what has been long established in protein folding, but would entail a major shift in how structures are interpreted, because there would not be a meaningful way to connect conformations to kinetic models.

The experiments providing direct evidence for intermediate protein “motions” are published in our eLife paper of 2017. The two major arguments being the separation of the pH-dependent fluorescence and electrophysiological curves, as well as the kinetic studies, for which a two states model can clearly not account for. The present work further expends this idea, for instance since non-functional mutants still undergo pH-dependent fluorescence changes, another direct demonstration of protein movement not directly related to activation.

The referees ask the question whether such intermediate motion underlie passage through a well-defined single allosteric state, or a multiplicity of conformational pathways. This is an interesting and complex question, that actually is relevant not only to these “intermediate motions”, but also to the end states themselves. In particular, it is more and more admitted that the resting state of GLIC displays high structural flexibility (notably from X-ray, Cryo-EM and MD simulation work cited in the paper). In other word, some of the currently called “allosteric states” actually correspond to a family of conformations. Nonetheless, allosteric models involving discrete states remain useful. While they represent a simplification of the highly complex molecular reorganizations, they are essential to interpret functional data into structural term.

In the new version of the article, we now interpret the data with a 3-state allosteric model. In addition, we discuss that the model does not account for the low cooperativity of the fluorescence curves and we propose, based on NMA analysis, that the implemented intermediate “pre-active” state could correspond to a wider family of states, for instance involving asymmetric movement of individual subunits, in a manner reminiscent of the desensitization of GABAA receptors (Gielen et al., 2020). We also attempt to emphasize in the discussion that the modelling is not meant to demonstrate the occurrence of discrete states, but to refine the interpretation of the mutant phenotypes in term of their relative effect on pre-activation versus activation.

3) Description of the iMOD fit simulations is unclear and does not provide enough detail. How many simulations were run to yield the two trajectories? It is not clear whether iMod-Fit returns two trajectories from a single calculation or whether multiple simulations were run and the trajectories were clustered. In the methods section (page 25, lines 25-26), trajectory A conformation change is from closed to open whereas trajectory B conformation change is from open to closed. However, in the figures, for both trajectories, frame 1starts at GLIC-pH7 (closed).

We thank the reviewer for pointing that the description of the Molecular Modeling methods was not detailed enough, and we tried to fix this issue. The iMOD fit simulations are now detailed in the method section Two simulations were run and the frame numbers have been inverted for one of them in the figures which, we agree, causes confusion. One simulation is starting from the closed conformation to reach the open conformation, and the other from the closed to the open. Both trajectories represent plausible pathways for activation and deactivation. In the result section, we have chosen for clarity to represent both trajectories in the closed to open direction. We propose to clarify and add the missing information as follows:

“Each structure (4NPQ and 4HFI) was fitted, using iMODfit (Lopéz-Blanco and Chacón, 2013), to the simulated electron-microscopy envelope of the other structure. The EM density map resolution was set to 5 Å and the grid size to 0.5 Å. iMODfit was then used to compute two trajectories, fitting each structure to the density of the other structure. In trajectory A structure 4NPQ is fitted to the density of 4HFI, and in trajectory B structure 4HFI is fitted to the density of 4NPQ. The range of modes considered (-n option) was set to 0.5, i.e. half of the modes are considered for the conformational change. The geometry of the ion channel has been computed with MOLEonline webserver (mole.upol.cz), with the ’pore’ mode (Sehnal et al., 2013) We used the FreeRadius value computed at the level of the 9’ residue.”

4) Since their previous time-resolved fluorescence quenching data (eLife 2017) demonstrated that motions monitored at Bimane-136 (ECD β sandwich compaction) and Bim-250 (M2-M3 loop motion) mainly occur prior to channel opening (activation scheme in Figure 4B), it is unclear why the authors conclude that their Bim-250 data support pathway A identified from their simulations. In pathway A, Bim-250-Y197 unquenching happens in same time frame as the increase in pore radius at the 9' position, whereas in pathway B the unquenching of Bim250-Y97 happens before complete pore radius dilation (Figure 2B, 2D). The experimental data at this position seems compatible with the B trajectory. Moreover, the unquenching of Bim-136-W101 appears consistent with either trajectory (Figure 3B) and occurs before the simulated pore dilation. It is unclear what new information the simulations are providing except that the steady-state fluorescence quenching data show good agreement with the simulated end states.

We agree with the referee that discriminating both trajectories on the basis of the current fluorescence data was rather speculative and not clear-cut. We thus removed these considerations in the new version of the paper.

However, iMOD fit simulations still give important information. Regardless of the order in which specific protein motions are appearing in the trajectories, iMOD fit reveals two distinct conformational reorganizations, an asymmetric quaternary reorganization of the ECD, as well as what we call a “central gating motion”, which involves a compaction of the lower part of the β-sandwich, a tilt of M2 toward M3 and in marked increase in the pore radius. Notably, several of the mutations investigated in the paper are predicted to specifically alter some of the key motions involved. iMOD fit simulations thus provides a unique framework to interpret the mutant phenotypes and propose specific protein motions differentially involved in pre-activation versus activation. This analysis is now included in the discussion in the section “Structural interpretation of the mutant phenotypes in the context of NMA trajectories”.

5) To monitor how motions in the ECD are coupled to motions in the TMD (pore opening), the authors used mutations and the allosteric drug modulator propofol to perturb GLIC gating transitions and compared the effects of these perturbations on proton-induced current responses and steady-state fluorescence quenching data. Since the fluorescence data are steady-state, whether a mutation causes an effect on the timing of the structural change in the gating pathway or an effect on the percentages of different conformational states in the ensemble at steady-state is not known and confounds data interpretation. The authors need to discuss this point.

The referees are perfectly correct. In the new version of the manuscript, we interpret the whole series of data using an equilibrium 3-state MWC model, which is adequate to describe the steady state data of fluorescence and electrophysiology (at the plateau of the current). We kept the consideration about the timing of the structural changes at the end of the discussion in a single section (Structural interpretation of the mutant phenotypes in the context of NMA trajectories), that gives overall a coherent and plausible picture of key events involved in gating.

5) Data supporting the statement that mutations of H235, L157 and L246 lead to new global conformations are limited. One could argue that these mutations have dramatic functional effects (eliminate current) and thus, it is not all that surprising that alternative conformations might be adopted that normally would not be visited.

We agree with the referee that data concerning these mutants are limited to conclude that they adopt new conformations. We thus reconsidered their phenotypes, and aimed at integrating them into a single MWC model accounting for all the mutant. Among non-functional mutants, H235F is the best characterized and its X-ray structure shows a well folded conformation, the so-called locally-closed, that is adopted by numerous other GLIC mutants including the WT GLIC (Sauguet et al., 2014). In the course of the fitting procedure, we found actually that we could readily account for the phenotype of this mutant (including shifts in pH50s and partial amplitude of the fluorescence curves) by increasing the LpA constant, i.e., by stabilizing the resting state over the pre-active state. Therefore, there is no need to assume a new global conformation in this case. Concerning L157 and L246, as stated in the new version of the article, fits are of lower quality, notably because some of their pH-dependent curves are rather flat and better represented by a straight line. This suggests that those mutants undergo more complex reorganizations than simulated here by a 3-state model, but the data do not allow to conclude on the mechanisms involved. Those two mutants were thus discussed mainly in their ability to preclude the activation transition to occur.

6) Data to support the claim that propofol specifically affects the pre-activation step are limited. Propofol could affect open-closed equilibrium.

To interpret the propofol data, we used the three-state allosteric model, which suggests that propofol affects both the pre-activation and activation equilibria. It is noteworthy that several propofol binding sites have been identified on GLIC by crystallography, and plausibly other unknown sites are present. It is thus difficult to drawn more precise conclusion regarding this matter. In particular our fluorescence sensor reports allosteric effect but would not sense a potential pore blocker effect of propofol suggested in other studies (Fourati et al., 2018).

7) It is striking that the fluorescence quenching profile for the Q235 mutant with propofol closely mirrors that of H235 without propofol (Black vs light/dashed green lines in Figure 8D). The addition of propofol reverses the effect of the H235Q mutation on structure, not just qualitatively, but close-to-quantitatively, for both the Bim136-Q101W and Bim135-W72 sensors. This is remarkable, especially for the latter sensor where the curve is complex. Yet the discussion treated the two sensors differently despite a similar reversal of the effect of the H235Q mutation, and the authors say the data for Bim135-W72-H235Q cannot be interpreted in structural terms. This explanation is confusing, especially since the corresponding figure is given equal weight in the paper. The authors need to revisit this section to improve clarity or move the mutant results and discussion of explicit technical limitations to supplementary information.

In the original version of the manuscript, we have chosen to show data related to the Bim135 sensor to provide a complete description of the quenching data that we have collected. However, as discussed, data related to this position are complex, especially because of the biphasic nature of the fluorescence curve that mix two components. We agree that data related to H235Q is remarkable, but the various mutants tested at this position did not allow us to propose a reasonable interpretation, even a speculative one. Clearly, interesting reorganization are occurring at this level, but they remain mysterious, probably because Bim135 monitor subtle local changes in structure, as illustrated in the docking simulation. To clarify the text, we thus chose to move all the data related to this sensor in the supplementary figures.

8) The authors claim that their results "challenges the conventional concept that receptor activation involves a single conformational pathway." Do the authors believe their results are inconsistent with the 4 state Heidemann and Changeux models from the 1980s?

This claim was removed from the new version. However, not directly linked to this work, a recent paper from the group shows that for the GABAA receptor, asymmetric motions of individual subunit accounts for the biphasic desensitization kinetics (Gielen et al., 2020). It is possible that such a mechanism also applies to nAChRs and other pLGICs and those observations are not necessarily incompatible with a 4 states MWC model.

9) TMD motions were not measured. Monitoring motion of the M2-M3 loop does not monitor changes in the TMD or pore opening. In previous work, authors used Bimane243 to monitor M2 motions. Additional reporters of TMD motions would be helpful.

In the 2017 eLife article, we performed mutational work to generate a fluorescent sensor of pore opening. To this aim, based on the Bim243 mutant, we introduced the quenching residue tryptophan at various locations in the upper part of the TMD, nearby the channel gate (figure 3F). Unfortunately, most pairs did not show pH dependent changes in fluorescence, and the only mutant which did so was not functional by electrophysiology. It is clear that mutating this key gating region, especially introducing the bulky tryptophan residue, is detrimental to the gating process. Overall, our work shows that the TriQ technique is powerful to monitor reorganization at the level of solvent-accessible regions of the protein, but not suited to monitor motions in buried locations such as the ion channel.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

1. The interpretation of the data with propofol is still a weak point. For instance, the manuscript says that "Our data thus shows that propofol does not act locally by altering the conformation of the TMD, but rather acts on the global allosteric transitions by displacing the equilibria and preserving ECD-TMD coupling" However, the preceding discussion is minimal. I spent a while trying to piece together how the data show that propofol's mechanism of action is not local. It is clear that the binding of propofol does affect the conformation far from the TMD, but this sentence implies that the authors have shown that TMD conformational change is insufficient for action. In other words, I could not figure out how the authors showed that ECD conformational shifts were necessary for propofol to act. If I am missing something, a few more sentences spelling out the logic here would be helpful or rewording of conclusions is needed.

The referee is right in pointing out that our conclusion about propofol is somewhat overstated. Indeed, the whole study does not demonstrate that the reorganizations of the ECD are a necessary condition for TMD reorganization and channel opening, although all functional mutants tested do show almost full fluorescence changes at Bim136. We thus reword the conclusion:

“Our data thus shows that propofol does acts on the global allosteric transitions by displacing the equilibria of both pre-activation and activation. It is noteworthy that propofol is also likely to generate local effects upon binding to modulate the function, that are nor investigated here. For instance, its binding into the pore may sterically block ion translocation to produce inhibition”.

2. The new Figure 9 is so full of important information that it gets hard to follow. I request that all the isomerization constants be compiled in a table for ease of comparison. The caption for that table can then refer to the equations and methods from which they were derived and differentiate between normalization schemes. While it may be useful to keep them in Figure 9 as well, Figure 9 is already overwhelmingly complex. I suggest either having fewer plots or fewer elements per plot. The arrows from one plot to the next were also not intuitive. Mainly, I request that the authors look at this figure with fresh eyes and consider breaking it up or streamlining it somehow.

The figure 9 (figure 6 is the new version) has been modified. For clarity the isomerization constants have been summarized in a table and population of the pA state is not shown anymore. To visualize the shifts in isomerization constants, the fits of the reference (Bim136-Q101W or Bim136-Q101W-H235Q in case of Bim136-Q101W-H235Q + propofol) are shown in dashed lines for both current and fluorescence quenching.

3. Line 37 "Seminal work in the 80s showed that a minimal four-state model describes the main allosteric properties of the muscle-type nAChR (Heidmann and Changeux, 1980; Sakmann et al., 1980)."

First, the word seminal is best avoided; it is rather outdated. I would not take these similarly out-of-date works as a benchmark, rather use them as a counterpoint – perhaps you can say: "although xxx work in the 1980s, …" and then introduce the updates?

That sentence has been re-phrased “Initially, a minimal 4-states model could describe the main allosteric properties of the muscle-type nAChR”.

4. Line 66 "However, the physiological relevance of these structures or their assignment to particular intermediates or end-states in putative gating pathways remains ambiguous and poorly studied."

This is a very important point and underlines the importance of the work at hand.

This point has been repeated in the discussion and abstract.

5. Line 69 "Conversely, it is likely that key conformations, unfavored by crystal packing lattice or under-represented in receptor populations on cryo-EM grids, are missing in the current structural galleries."

On the other hand, this is overstated. I would say "possible", not likely. Intermediates might be missing, but are they "key"?

This sentence has been rephrased “Conversely, it is possible that intermediate conformations, unfavored by crystal packing lattice or under-represented in receptor populations on cryoEM grids, are missing in the current structural galleries.”.

6. Line 86 "much faster than ionic current measurement that occurs in the 30-150 millisecond range "

Much faster than the rise time of population or ensemble currents.

This sentence has been rephrased “much faster than the rise time of the active population that occurs in the 30-150 millisecond range in electrophysiology recordings”.

7. Line 116 "Two independent trajectories, A and B, were computed starting from each of the two end-state structures and divided into 12 and 11 frames respectively. "

I appreciate that the authors tried to explain better now, but this is still somewhat opaque. Please just say one trajectory goes from rest to active, and the other from active to rest. The use of "end-state structures" is confusing. They are both end and start, depends on which trajectory it is. Or are there really four structures – two crystals and two end states? Some of the figures suggest that each trajectory does not conclude in really the right place. I can't say which one because I have no idea what figure is which (figures not numbered and some do not correspond to figure legend order).

To clarify the presentation of the trajectories, we now state explicitly that:

“Two independent trajectories were computed. Trajectory A (12 frames) starts from the closed GLIC-pH7 structure to reach the open GLIC-pH4 structure, and trajectory B (11 frames) starts from the GLIC-pH4 structure to reach the GLIC-pH7 structure. Of note, both trajectories are fully reversible and are equally relevant to describe either activation or deactivation, since normal modes deformation can be applied in the two directions.”

Figures have been re-numbered.

8. With iModFit, I think it is important to discuss how plausible it is that the transitions are not ergodic. This is mentioned in the discussion. In one way, we should not take these trajectories too seriously. But it is also important to consider the possibility that they are pulling out important information from the structures. Are the authors suggesting that the isomerisations are, preferentially, not reversible? I mention below that it would be a great future insight to have non-equilibrium data that could report the non-reversible motion at some of these sites.

We perfectly agree with the referee. This is why we moved the iMODFit to the end of the result section, and presented them more carefully, emphasizing the limitation of the “rough” trajectories and “simple” docking approach. We also state that trajectories, as normal mode deformations, are fully reversible (see answer to point 7). We also agree that these simple computations are pulling out important information from the structure, for instance the “central gating region”, or more locally the reorganization of bimane when attached at position 250.

9. Later in the paper, the text again makes me feel like I don't understand what iModFit does.

Line 176 "For the ECD quenching pair Bim136-Q101W, the simulations show that Bim136 and the Trp101 indole ring are separated in the resting-like state, and are in close contact in the active-like state (Figure 3A). "

The simulations? Are you referring to the docking results? The resting and active-like states are from structures, aren't they, not from iModfit? If the states used are from structures, there is little predictive power from the docking to these structures that couldn't be deduced by eye, is there? Or are you comparing the state at the end of the iModFit run, which isn't the other crystal state?

Surely iModFit (simulations?) only tells you about the trajectories of the fluorophores? This time-order of transitions between distances is interesting. But why is it mixed up with end state information (surely known from PDB)?

At the very least, a better description is needed. Overall, I still do not understand how the iModFit trajectories help to understand the steady state fluorescence.

The improper term “simulation” has been removed concerning iMODFit. For this analysis, we considered the extreme intermediates of the trajectories (first and last frames), not the crystallographic structures. This is now stated. Of note, crystallographic structures would show identical results. We do not agree that iMODFit is useless in understanding quenching. For Bim136, the center of mass distances along the trajectories is indeed poorly informative since the bimane moiety, that is exposed to the solvent and rather mobile, adopts multiples poses in the docking procedure generating large fluctuation in center of mass distance. In contrast, the docking procedure is quite interesting for Bim250, which flips from both sides of loop 2, a feature that could not be identified by visual inspection.

10. Line 155 "In conclusion, using iMODfit we could generate two distinct trajectories that are in principle equally plausible to describe a gating transition of GLIC activation. "

But a really key point that doesn't really come up, but I think it should, is that the different trajectories really consist of at least two steps, Twist and the central gating motion, but they occur in different orders. This is a clear appeal to the intermediate states like flip and prime, and motivates the rest of the paper. The role of the compaction is less clear. If there were not distinct movements, the hysteresis in the motions would be much harder to understand. Still, the connection of the iModFit to the steady-state data is less convincing than any non-equilibrium data would be. This is the distinction between a plausible model (as the authors present) and evidence. The change of fluorescence at given sites should have different orders for activation and deactivation, shouldn't it? This would be worth mentioning. It is something for the future of course. And this is not to diminish the insight from the steady-state measurements.

It is true that the twist and central gating motions appear in distinctive order in both trajectories. However, we do not have sensors of the twist, so we cannot investigate this global motion in fluorescence experiments. This is why we focused the analysis and discussion on the ECD compaction, M2-M3 movement, and, indirectly from structural considerations, on β-sandwich compaction and M2 movement toward M3. It is also true that the role of the ECD compaction is unclear from iMODFit trajectories. However, the role of the ECD compaction is clear from fluorescence quenching data, that are more solid. This is another reason to present the more speculative iMODFit data at the end of the Results section, and use them only for the interpretation of the fluorescence data. Finally, concerning the hysteresis, both trajectories are theoretically fully reversible, so we think that it is too speculative to discuss the pathways of activation and deactivation in this paper.

11. Additionally, the flipped state, where the conformational change of the orthosteric site is predicted to be rather complete, but where the channel is closed, would fit the functional requirement of a pre-active state (Lape et al., 2008).

This is trivial because the flip state is just the name of a non-open agonist bound state. Also, flip is not the only type of state that fits, they might all be the same, from different perspectives. I wrote a comment about this once: "Don't flip out: AChRs are primed to catch and hold your attention"

The mentioned reference has been added.

12. The quantitative details of the fitting and the agreement or otherwise seem reasonable but I cannot claim to check in detail, I'm afraid. The individual conformations have various proton bound states and equilibria, so the 3-state model is quite a bit more complicated than at first sight. It might be nice to include the full model (to indicate the assumptions) in a supplementary figure. If, as the authors say, a relatively complicated proton binding scheme is needed to describe even equilibrium data, this is something of a find and needn't be buried. I don't think we have many ideas about how may protons are needed to gate.

A supplementary figure has been added with a workflow scheme illustrating the steps and assumptions used to build the MWC model. Indeed, the requirement of a second binding site is a new finding that is consistent with our previous study on GLIC proton site where we could not find a single proton site that would abolish activation when mutated (Nemecz et al., 2017).

13. I did notice that the Y28F mutant has the biggest change in the Pre-open constant. This selective effect was the case for the nearby A52S mutant in the glycine receptor – a big change in flip, no change in the main gating constants (Plested et al., 2007). Quite different at the K276E below that (Lape et al., 2012). But there are tons of mutants on these positions, maybe there are better ones to compare.

In the examples of pathological mutations used in the discussion we focused on the mutation homologous to the ones used on GLIC. The papers mentioned here are studying other residues but their conclusions are in line with the residues we studied in the same regions. We added them at the end of the discussion:

“Interestingly, on the glycine receptor α1, other mutations have been studied by single channel recordings and are described to affect principally a flip pre-activation-like step for A52S in the loop 2 at the ECD-ECD interface (Plested et al., 2007) or gating for K276E on the M2-M3 loop (Lape et al., 2012). These data suggest that mutations produce similar allosteric perturbations on GLIC and GlyR in those regions.”

14. Table 1 statistics. Multiple mutants are being compared to the same reference value. Unpaired t test is not the correct test to use for these data. Investigators should use an ANOVA with a posthoc test such as a Dunnett or Bonferroni.

A one-way ANOVA followed by a Dunnett test has been performed on data where mutants were compared to a common reference (Bim136-Q101 of Bim250-Y197). To compare individually fluorescence to electrophysiological pH50 for each mutant, we kept the t tests.

15. MWC fitting of the fluorescent and current data is used to conclude that mutations at the ECD alter the pre-activation step while those at the ECD-TMD interface and TMD alter the activation step (gating). Due to the assumption that the mutations do not effect proton affinity to the sites, the authors need to be careful about overinterpreting the data. The modeling provides support but is not conclusive.

Indeed, choosing to modify isomerization constants and not affinities is a choice made in this model and conclusions were already mitigated in the discussion:

“We also postulated that the various mutants only alter the isomerization constants between states. However, the dataset does not allow for the discrimination between effect on binding affinity versus isomerization constants. The effects of mutations on the isomerization constants are thus meant here to evaluate the global effect of the mutations on pre-activation versus activation, but it is possible that they actually underlie alteration of isomerization constant, affinity constants, or both.”

We now develop further the idea:

“Among the various mutations investigated here, E26Q, E222Q, H235F/Q neutralize the charge of titratable amino acids. It is thus possible that in these cases the mutation eliminates a proton binding site. However, a local impact of a mutation on a proton binding site, or on a set of inter-residues interactions altering the allosteric equilibria, will be equally valid in assigning local structural alterations to pre-active/active phenotypes.”

16. How the fluorescence quenching data relate to motions identified by iMODfit is not obvious. On page 10, lines 315-318, the authors state "the fluorescence and electrophysiological pH-dependent curves presented in this paper underlie two major allosteric steps, pre-activation (a fast process causing the changes in fluorescence as previously identified in stopped flow experiments and activation (a slower phase). Based on their 2017 eLife paper, the bim136 fluorescence reports early pre-gating motions, and bim250 reports early pre-gating motions and some later motions. In the revised manuscript, based on iMOD fit/normal mode analyses, the authors state that bim136 is monitoring a quaternary compaction of the ECD that is occurring throughout the gating cycle and that bim250 is monitoring motion of m2-m3 loop which is part of the 'central gating reorganization' including opening of pore (see page 14, lines 452-455). Later in the paper (page 16, lines 528-529) they state 'ECD compaction is critically involved in pre-activation". This is confusing and requires additional explanation and discussion. If the fluorescent reporters at these positions are monitoring fast, early pre-gating motions then why is the quaternary compaction and m2-m3 loop motion part of the central gating reorganization? Am I missing something?

In our eLife 2017 paper, we showed for Bim136-Q101W and Bim250-Y197 that the majority of the changes in fluorescence occur with very fast kinetics. Bim250 indeed shows some fluorescence changes appearing with slower kinetics, but they contribute by less than 20% to the total ∆F, and thus were neglected in the present steady-state study. Still, Bim250 monitors a pre-activation motion by fluorescence, but the modifications themselves alter principally the activation transition (stronger effect on the pH50 of ∆I as compared to ∆F). We now emphasize this point in the discussion by stating:

“The mutational analysis also shows for most mutations mixed effects on the isomerization constants of activation and pre-activation, suggesting that both processes involve overlapping regions. The Bim250 position is noteworthy in this respect, since the bimane, reporting an outward motion of the M2-M3 loop, monitors pre-activation, while the modification itself (P250C mutation plus reaction with bimane) principally alters the activation process. It is thus plausible that the M2-M3 loop could move in two successive steps, a first one during pre-activation conditioning dequenching and a second one during activation. In either case, our data further highlight a central role for this loop in ECD-TMD coupling.”

We then later in the text speculate about one possible succession of reorganizations:

“Concerning the order in which reorganizations are observed, Trajectory A fits much better the fluorescence data. It suggests a scenario involving, during pre-activation, progressive ECD compaction and beginning of the M2-M3 loop motion, generating the fluorescence variations. Then the M2-M3 loop completes its movement in concert with β-sandwich compaction and pore opening. Future computational studies are needed to investigate this possibility.”

17. In the abstract, the authors state that 'preactivation involves major asymmetric quaternary motions of the extracellular domain'. It is unclear to me what experimental data support this conclusion. Is this based on the starting pH7.0 crystal structure? The authors need to clarify if the asymmetry that they are describing is at the subunit level or is based on two different motions in the ECD (twisting and compaction). Without strong experimental evidence for asymmetric motions, this conclusion should be removed from the abstract.

The abstract has been re-written leaving out the notion of asymmetry, which we agree is not addressed in the fluorescence experiments.

18. They use iModFit and NMA as synonyms in some parts of the paper, which causes confusion. iMODfit/Normal mode analysis treats the protein like a 3D elastic network. It doesn't capture interactions with solvent or specific residue-residue interactions. It superimposes multiple local low-energy fluctuations to find likely larger scale conformational fluctuations. You would get asymmetry when it costs less total energy to move a few chains by a lot, than to move all of them by a little. The more chains the protein has, the more likely it is that imodFit will find asymmetry. I'm not sure the imodfit simulations add much regarding asymmetry, but the expected behavior of these macromolecules at room temperature makes it an uncontroversial claim, albeit one without significant new evidence.

The asymmetric nature of the transition has been removed from the abstract and just discussed at the end of the discussion in the section “Speculative interpretation of the mutant phenotypes in the context of computational trajectories”.

19. In revised manuscript (page 5, lines 155-157), authors state 'using iMODfit we could generate two distinct trajectories that are in principle equally plausible to describe a gating transition of GLIC activation'. Additional discussion spelling out the logic here is essential.

It is important for the reader to understand the limitations of iModFit, and for a non-computational reader to know what iModFit is not. The authors need to add further discussion in methods or result sections. It is not a physics-based simulation technique like molecular dynamics – I'd call it a numerical approach for generating hypothetical pathways, and then experiments or simulations need to distinguish between them. They have used it here as a conceptual framework. The software itself is not designed to generate trajectories, but to generate structures. Motions or trajectories generated by Normal Mode Analysis are always reversible. Then imodfit applies a bias on top of that, based on the structure, to get a directional trajectory. They applied two different biases (based on two different structures) so they ended up with two different hypothetical and reversible trajectories.

In their response letter, the authors state "one simulation is starting from the closed conformation to reach the open conformation, and the other from the closed to the open. Both trajectories represent plausible pathways for activation and deactivation." It is unclear whether the authors think that simulation A describes activation (closed channel to open channel) pathway and simulation B describes deactivation pathway (open channel to closed channel) or if they think that both trajectories can describe activation (closed channel to open channel)? Please clarify.

The iMODfit procedure, and its limitations, have been presented in detail in the results and method section to clarify the various concerns raised here and in previous questions.

20. The authors should discuss and compare their results from iMODfit/NMA analyses to results from Toby Allen lab (PNAS 2017) using all-atom molecular dynamics with a string method to solve for GLIC gating pathways. What new information has been gained from the iMODfit/NMA?

The discussion now includes an overview of the gating trajectories generated in silico, in the section “Speculative interpretation of the mutant phenotypes in the context of computational trajectories”. We started with all-atom MDs, notably the work from Allen and co-workers, and then move to iMODfit trajectories explicitly stating that:

“While these trajectories are rough and do not implement fine atomistic interactions, they allow to visualize plausible collective motions in relation with the reorganization of the quenching pairs”.

21. Figures 3 supplementary 1 and 2 and 3 are in in different order compared to figure legends and text on page 6 lines 174-175. Authors need to check the order of the supplementary figures. It would be helpful if figures were labeled for review purposes.

Figures have been re-labeled.

22. Abstract should state which experimental results support their conclusions and describe the novel contributions that the data are providing.

The abstract has been re-written.

https://doi.org/10.7554/eLife.60682.sa2

Article and author information

Author details

  1. Solène N Lefebvre

    1. Institut Pasteur, Université de Paris, CNRS UMR 3571,Channel-Receptors Unit, Paris, France
    2. Sorbonne Université, Collège doctoral, Paris, France
    Contribution
    Conceptualization, Formal analysis, Investigation, Methodology, Validation, Visualization, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1333-2042
  2. Antoine Taly

    1. Institut de Biologie Physico-chimique, Fondation Edmond de Rothschild, PSL Research University, Paris, France
    2. Laboratoire de Biochimie Théorique, CNRS, Université de Paris, UPR 9080, Paris, France
    Contribution
    Conceptualization, Investigation, Methodology, Resources, Validation, Writing – review and editing
    For correspondence
    antoine.taly@ibpc.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5109-0091
  3. Anaïs Menny

    1. Institut Pasteur, Université de Paris, CNRS UMR 3571,Channel-Receptors Unit, Paris, France
    2. Sorbonne Université, Collège doctoral, Paris, France
    Present address
    Institut de Génomique Fonctionnelle, Université de Montpellier, CNRS, INSERM, Montpellier, France
    Contribution
    Investigation, Methodology, Validation, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6044-4119
  4. Karima Medjebeur

    Institut Pasteur, Université de Paris, CNRS UMR 3571,Channel-Receptors Unit, Paris, France
    Contribution
    Investigation, Validation
    Competing interests
    No competing interests declared
  5. Pierre-Jean Corringer

    Institut Pasteur, Université de Paris, CNRS UMR 3571,Channel-Receptors Unit, Paris, France
    Contribution
    Conceptualization, Funding acquisition, Project administration, Supervision, Writing – original draft, Writing – review and editing
    For correspondence
    pjcorrin@pasteur.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-4770-430X

Funding

Agence Nationale de la Recherche (ANR-13-BSV-0020)

  • Solène N Lefebvre
  • Anaïs Menny
  • Karima Medjebeur
  • Pierre-Jean Corringer

Agence Nationale de la Recherche (ANR-11-LABX-0011)

  • Antoine Taly

European Research Council (grant No. 788974)

  • Pierre-Jean Corringer

Sorbonne University - Doctoral school ED3C (PhD fellowship)

  • Solène N Lefebvre

Fondation pour la Recherche Médicale (PhD fellowship complement)

  • Solène N Lefebvre

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The work was supported by the ‘Agence Nationale de la Recherche’ (Grant ANR-13-BSV8-0020, Pentagate), the doctoral school ED3C and the ‘Foundation pour la Recherche Médicale’ (PhD funding to SNL), the ‘Initiative d'Excellence’ (cluster of excellence LABEX Dynamo, ANR-11-LABX-0011 to AT) and the ERC (Grant no. 788974, Dynacotine). The authors would like to thank Stuart Edelstein for helping with MWC equations, Marc Gielen, Akos Nemecz, and Marie Prévost for critical reading of the manuscript.

Senior Editor

  1. Olga Boudker, Weill Cornell Medicine, United States

Reviewing Editor

  1. Cynthia M Czajkowski, University of Wisconsin, Madison, United States

Reviewers

  1. Andrew JR Plested, Humboldt Universität zu Berlin, Germany
  2. Grace Brannigan, Rutgers, United States

Publication history

  1. Received: July 2, 2020
  2. Preprint posted: July 4, 2020 (view preprint)
  3. Accepted: September 29, 2021
  4. Accepted Manuscript published: September 30, 2021 (version 1)
  5. Version of Record published: October 11, 2021 (version 2)

Copyright

© 2021, Lefebvre et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 454
    Page views
  • 95
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Neuroscience
    Zhengchao Xu et al.
    Tools and Resources Updated

    The dorsal raphe nucleus (DR) and median raphe nucleus (MR) contain populations of glutamatergic and GABAergic neurons that regulate diverse behavioral functions. However, their whole-brain input-output circuits remain incompletely elucidated. We used viral tracing combined with fluorescence micro-optical sectioning tomography to generate a comprehensive whole-brain atlas of inputs and outputs of glutamatergic and GABAergic neurons in the DR and MR. We found that these neurons received inputs from similar upstream brain regions. The glutamatergic and GABAergic neurons in the same raphe nucleus had divergent projection patterns with differences in critical brain regions. Specifically, MR glutamatergic neurons projected to the lateral habenula through multiple pathways. Correlation and cluster analysis revealed that glutamatergic and GABAergic neurons in the same raphe nucleus received heterogeneous inputs and sent different collateral projections. This connectivity atlas further elucidates the anatomical architecture of the raphe nuclei, which could facilitate better understanding of their behavioral functions.

    1. Neuroscience
    Shankar Ramachandran et al.
    Research Article Updated

    Neuromodulators promote adaptive behaviors that are often complex and involve concerted activity changes across circuits that are often not physically connected. It is not well understood how neuromodulatory systems accomplish these tasks. Here, we show that the Caenorhabditis elegans NLP-12 neuropeptide system shapes responses to food availability by modulating the activity of head and body wall motor neurons through alternate G-protein coupled receptor (GPCR) targets, CKR-1 and CKR-2. We show ckr-2 deletion reduces body bend depth during movement under basal conditions. We demonstrate CKR-1 is a functional NLP-12 receptor and define its expression in the nervous system. In contrast to basal locomotion, biased CKR-1 GPCR stimulation of head motor neurons promotes turning during local searching. Deletion of ckr-1 reduces head neuron activity and diminishes turning while specific ckr-1 overexpression or head neuron activation promote turning. Thus, our studies suggest locomotor responses to changing food availability are regulated through conditional NLP-12 stimulation of head or body wall motor circuits.