A Crucial Role for Side-Chain Conformation in the Versatile Charge Selectivity of Cys-Loop Receptors

Abstract

Whether synaptic transmission is excitatory or inhibitory depends, to a large extent, on whether the ion channels that open upon binding the released neurotransmitter conduct cations or anions. The mechanistic basis of the opposite charge selectivities of Cys-loop receptors has only recently begun to emerge. It is now clear that ionized side chains—whether pore-facing or buried—in the first α-helical turn of the second transmembrane segments underlie this phenomenon and that the electrostatics of backbone atoms are not critically involved. Moreover, on the basis of electrophysiological observations, it has recently been suggested that not only the sign of charged side chains but also their conformation are crucial determinants of cation-anion selectivity. To challenge these ideas with the chemical and structural rigor that electrophysiological observations naturally lack, we performed molecular dynamics, Brownian dynamics, and electrostatics calculations of ion permeation. To this end, we used structural models of the open-channel conformation of the α1 glutamate-gated Cl⁻ channel and the α1 glycine receptor. Our results provided full support to the notion that the conformation of charged sides chains matters for charge selectivity. Indeed, whereas some rotamers of the buried arginines at position 0′ conferred high selectivity for anions, others supported the permeation of cations and anions at similar rates or even allowed the faster permeation of cations. Furthermore, we found that modeling glutamates at position −1′ of the anion-selective α1 glycine receptor open-state structure—instead of the five native alanines—switches charge selectivity also in a conformation-dependent manner, with some glutamate rotamers being much more effective at conferring selectivity for cations than others. Regarding pore size, we found that the mere expansion of the pore has only a minimal impact on cation-anion selectivity. Overall, these results bring to light the previously unappreciated impact of side-chain conformation on charge selectivity in Cys-loop receptors.

Introduction

Whereas the excitatory glutamate receptors, the ATP receptor channels, and the proton-gated acid-sensing ion channels are exclusively selective for cations, the members of the superfamily of pentameric ligand-gated ion channels (pLGICs, also known as “Cys-loop receptors”) are either highly selective for cations or highly selective for anions while retaining the same overall structure. Because all cation-selective neurotransmitter-gated ion channels discriminate poorly between Na⁺ and K⁺, fast inhibitory synaptic transmission can only be mediated by the anion-selective pLGICs. Certainly, no K⁺-selective ionotropic receptor seems to be expressed in animal cells. Thus, it is tempting to speculate that the versatility of the pLGIC fold may have played a crucial role during the evolution of the nervous system.

Charge selectivity in pLGICs has been conclusively shown to be governed by the intracellular end of the transmembrane pore, more precisely, by the first turn of the pore-lining second transmembrane-segment (M2) α-helices (1). An alignment of amino-acid sequences reveals that this short stretch of amino acids is highly conserved in vertebrate animals. Most cation-selective pLGIC-forming subunits (and, certainly, all of those that can form homomeric pentamers) have either a GEK or a GER motif—occupying positions −2′, −1′, and 0′, respectively—and a four-residue M1-M2 linker (Fig. 1). On the other hand, all anion-selective pLGIC-forming subunits have either a PAR or an AAR motif in the first turn of M2 and a five-residue M1-M2 linker (Fig. 1). Both functional (2) and structural data (3 and references therein) on the open-channel conformation of pLGICs indicate that positions −2′ and −1′ of the first turn of M2 face the aqueous lumen of the pore, whereas position 0′ is oriented toward the back, away from the pore, in both the cation-selective and the anion-selective members of the superfamily (Fig. 1 C).

Figure 1.

The charge-selectivity filter of wild-type pLGICs. (A) A schematic representation of the membrane-threading pattern of pLGICs is shown. The C-terminus of M1, the M1-M2 linker, and the N-terminus of M2 are highlighted. The approximate location of some residues is indicated with the prime-numbering notation used throughout this work. (B) Alignment of residues in and flanking the M1-M2 linker is shown. The classification of these residues as belonging to the linker or to one of the flanking α-helical termini is tentative and was made on the basis of the structural models of the anion-selective β₃-GABA_AR (PDB: 4COF; (45)) and the cation-selective GLIC (PDB: 4HFI; (46)). The broken horizontal line separates the sequences of subunits that form cation-selective channels (top) from those that form anion-selective ones (bottom). The invertebrate organisms in this list are Schistosoma mansoni (a parasitic flatworm), C. elegans (a nematode), Drosophila melanogaster (an arthropod), and Lymnaea stagnalis (a mollusc), and the bacteria are Gloeobacter violaceus and Dickeya dadantii (formerly known as Erwinia chrysanthemi). The color code identifying the M1-M2 linker and flanking regions is the same as in (A). (C) Structural alignment of the cation-selective GLIC and the anion-selective α1-GluCl performed with STAMP (47) as implemented within the MultiSeq plugin (48) in Visual Molecular Dynamics (VMD; (31)) is shown. Only the M2 α-helix and the preceding M1-M2 linker of each subunit are shown. The approximate locations of the Cα atoms at positions −2′, −1′, and 0′ are indicated with spheres for the five subunits and the two structural models; these locations are essentially the same regardless of whether the channel is selective for cations or anions. (D) Alignment of M2 α-helix sequences is shown. Identical residues are indicated with a purple background. To see this figure in color, go online.

Decades of site-directed mutagenesis and electrophysiology experiments have revealed that only a few mutations in the first turn of M2 and the preceding M1-M2 linker are required to change the sign of the charge selectivity of pLGICs (4, 5, 6, 7, 8, 9). However, the mechanism behind the effect of these mutations—and therefore, the molecular basis of charge selectivity in these channels—has remained controversial for many years. In fact, a careful analysis of the literature reveals that the proposed mechanisms cover nearly the whole range of possibilities. Whereas some authors have suggested that the charge selectivity of these channels is governed solely by interactions between the passing ions and backbone atoms (4, 6, 9), others have favored the notion that only pore-facing charged side chains underlie this process in both the cation-selective and the anion-selective pLGICs (8).

The mechanistic basis of the opposite charge selectivities of pLGICs has only recently begun to emerge. As a result of a mutagenesis and electrophysiology study of different cation-selective and anion-selective pLGICs from vertebrate, invertebrate, and bacterial origin, we presented compelling evidence for the critical role played by charged side chains—rather than backbone atoms—in this phenomenon, whether pore-facing or buried (1). Furthermore, our data suggested that not only the charge sign but also the conformation of these side chains are critical determinants of charge selectivity in pLGICs. For example, we found that charge selectivity in these ion channels is highly sensitive to insertions and deletions in the M1-M2 linker and to residue-to-residue mutations at position −2′, both involving nonionizable amino acids. Because these mutations seemed unlikely to change the secondary structure of the protein drastically (so as to, for example, switch the location of critical residues between the front side and the backside of the M2 α-helix), we inferred that the different charge selectivities of these mutants must reflect the different conformations adopted by the charged side chains at positions −1′ and 0′ (1). It seemed inescapable to suggest that the torsional free-energy landscape of the acidic and basic side chains of the selectivity filter must be sensitive to the details of the local structure such as the identity of the residue at position −2′ and the length of the M1-M2 linker.

In a single-channel electrophysiological study of the (heteromeric) muscle AChR, in which we tested the effect of engineering different combinations of the native glutamate and mutant neutral residues at position −1′, we described the occurrence of distinct patterns of subconductance levels (10). Having ruled out protonation-deprotonation events as the underlying cause, we proposed that this multiplicity of currents levels reflects the occupancy of different rotamers of the glutamates at position −1′ while the channel is open; we could not explain our observations in any other way. Remarkably, a subsequent computational study—a combination of all-atom molecular dynamics (MD) and Brownian dynamics (BD)—provided ample support for the notion of a robust relationship between single-channel conductance and the conformation of glutamate side chains at this position of the muscle AChR (11). However, although single-channel conductance and charge selectivity are certainly related, by no means can conclusions about the latter be drawn solely on the basis of our knowledge of the former. Here, we are specifically concerned with charge selectivity and the potential role played by the conformation of charged side chains at the selectivity filter in this phenomenon.

Admittedly, no matter how plentiful electrophysiological data might be, experimentally obtained reversal potentials and calculated permeability-coefficient ratios cannot be invoked to identify side-chain conformation as one of the important structural variables in charge selectivity. Therefore, we decided to test this hypothesis using a computational approach. We performed molecular simulations of ion permeation on two structural models of anion-selective pLGICs (12, 13) that we have recently deemed to represent—or, at least, to closely approximate—the open-channel conformation (3). It seemed to us that only this type of approach was able to provide the chemical and structural rigor that our electrophysiological observations, naturally, lacked.

Methods

MD simulation setup

MD simulations were set up using the M1-M3 stretch of each subunit of the x-ray crystal structure model of the Caenorhabditis elegans α1-GluCl bound to ivermectin (residues 213–300; Protein Data Bank (PDB): 3RHW; (12)) or the cryo-EM structural model of the zebrafish α1-GlyR bound to glycine and ivermectin (residues 234–324; PDB: 3JAF; (13)). System setup and input files for minimization and equilibration were created using the CHARMM graphical user interface (CHARMM-GUI; (14)), and simulations were run using NAMD 2.10 (15). The CHARMM36 force field was used for lipids, proteins, and ions (16). Structures were placed in a pre-equilibrated membrane of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine solvated in TIP3P water containing a 150 mM KCl solution. The dimensions of the simulation box were, approximately, 106 × 108 × 97 Å³. Simulations were run at 300 K with a constant pressure of 1 atm. All simulations used a 2-fs timestep with nonbonded interactions calculated every 2 fs and full electrostatics calculated every 4 fs. A 10-Å switching distance and a 12-Å cutoff for nonbonded interactions were used. Long-range electrostatic interactions were calculated using the particle mesh Ewald method with a 1-Å grid spacing and with periodic boundary conditions being applied in all dimensions. A constant temperature was maintained using Langevin dynamics with a damping coefficient of 1 ps⁻¹. All simulations used harmonic restraints of 0.5 kcal ⋅ mol⁻¹ ⋅ Å⁻² on protein-backbone atoms so as to maintain a stable pore geometry.

Umbrella-sampling MD simulations

To compute the ion-permeation PMF, umbrella-sampling simulations were set up, starting with the last frame of the initial equilibration period described above. Umbrella windows were created along the ion-permeation axis using a spacing of 0.5 Å and spanning a distance of 51 Å. Using the collective-variables module in NAMD, a harmonic restraint of 5 kcal ⋅ mol⁻¹ ⋅ Å⁻² was placed on an individual ion at every window for the entire length of the transmembrane region. Each window was simulated for 6 ns, and the first 500 ps were removed from the analysis to account for equilibration. The umbrella windows were then combined using the weighted histogram analysis method (17, 18) to generate the corresponding single-ion PMF profile.

BD simulation setup

The simulated reduced models of α1-GluCl and the α1-GlyR were used as inputs for BD simulations using the Grand Canonical Monte Carlo/BD simulation software (19). Input files for the Grand Canonical Monte Carlo/BD simulations were generated using the CHARMM-GUI (20). Simulations were carried out with a grid spacing of 0.5 Å and a simulation box with dimensions of 75 × 75 × 91 Å³ for the GlyR (when the pore was expanded by 2 Å in diameter, the dimensions were 77 × 77 × 91 Å³; when expanded by 4 Å in diameter, the dimensions were 79 × 79 × 91 Å³) and 74 × 74 × 91 Å³ for GluCl, with a 5-Å buffer region. The diffusion coefficients of K⁺ and Cl⁻ in bulk solution were taken as 0.196 and 0.203 Å² ps⁻¹, respectively (21), whereas their values inside the channel’s pore were assumed to be uniform and equal to one-half their values in bulk (in an attempt to account for the effect of confinement; (20)). A gradual transition between bulk and inside-the-pore values was achieved using a 10-Å switching function on both ends of the channel (20). The dielectric constants of both the membrane and protein regions were set to 2, and that of all aqueous regions was set to 80. Both the pore and the membrane were taken to be 40 Å in length. All simulations were run for 2 μs, and each simulation was replicated six times with different velocity seeds at all tested voltages. BD simulations were also used to calculate multi-ion PMF profiles using the expression −RT ln(C_i/C_o), where C_o is the charge density of the bulk solution (for 150 mM KCl, C_o was taken as 9.03 ⋅ 10⁻⁵ e ⋅ Å⁻³) and C_i is the mean charge density at each point along the pore’s central axis over the entire simulation. Pore expansions were performed by increasing the distance between the center of mass of each of the five subunits and the center of the pore along the two dimensions perpendicular to the pore’s central axis.

Electrostatic calculations

Electrostatic calculations were performed using APBSmem 2.0.1 (22), an interface that adds additional functions to the APBS program (23). To perform the calculations, each PDB file was converted to a pqr file in APBSmem using the SWANSON force field (24, 25). Ion-solvation calculations using the nonlinear Poisson-Boltzmann equation and zero-boundary conditions used a grid with dimensions 161 × 161 × 161 Å³ that uses a 320 × 320 × 320 Å³ coarse, 160 × 160 × 160 Å³ medium, and 64 × 64 × 64 Å³ fine grid size with 0.1 M counterions of 2.0-Å radius and a temperature of 300 K. A dielectric constant of 2 was used for both the protein and the 40-Å thick membrane, whereas a dielectric constant of 80 was used for the solvent. Spl2 and Spl4 were used for the charge and surface model, respectively, with a surface spline width of 0.3 Å. K⁺ and Cl⁻ were modeled with their respective formal charges with ion radii of 1.33 and 1.81 Å, respectively. Ion-solvation electrostatics were calculated along the pore’s central axis from −45 to 45 Å every 0.5 Å.

Results

Open-state models and the basic side chain at position 0′

The main objective of this study was to test the idea that the conformation adopted by the charged side chains in the first α-helical turn of M2 is a major determinant of charge selectivity in pLGICs. To do this within a framework of molecular detail and stereochemical rigor, we decided to perform molecular simulations, and of course, for simulations of ion permeation to have any realistic meaning, an appropriate model of the open-channel structure must be used. In recent work, we presented evidence supporting the notion that the structural models of the C. elegans α1 glutamate-gated Cl⁻ channel (α1-GluCl) bound to glutamate and ivermectin (PDB: 3RIF; 3.35 Å; x-ray crystallography; (12)) or to ivermectin alone (PDB: 3RHW; 3.26 Å; x-ray crystallography; (12)), as well as the model of the zebrafish α1 glycine receptor (α1-GlyR) bound to glycine and ivermectin (PDB: 3JAF; 3.8 Å; single-particle electron cryo-microscopy (cryo-EM); (13)), are likely to represent the open-channel conformation of animal pLGICs. Because the two models of GluCl are very similar to each other at the level of the transmembrane pore, we set out to perform simulations using only one of them (PDB: 3RHW) and the GlyR model.

In the anion-selective pLGICs, the only ionizable side chain at the intracellular end of M2 is that of the buried arginine (or lysine, in some invertebrates) at position 0′ (Figs. 1 and S1). Although this position of the α-helix faces away from the pore, we have previously found that proton binding to basic side chains engineered on the backside of M2 or the front side of the (non-pore-lining) M1 and M3 α-helices decreases the single-channel conductance of the muscle AChR markedly (2, 26). Furthermore, we have recently shown that mutation of the arginine at position 0′ of the highly anion-selective chimera between the extracellular domain of the chick α7 AChR and the M1-M4 domain of the C. elegans β-GluCl receptor (P_Cl⁻/P_K⁺ = 410) to asparagine renders the channel nearly ideally nonselective (P_Cl⁻/P_K⁺ = 0.89; (1)). Mutations of this arginine to nonionizable residues in the rat α1-GlyR were found to reduce the channel’s functional expression below practical levels (1, 8), and thus, the effect of this arginine on the charge selectivity of the GlyR could not be assessed electrophysiologically. As for α1-GluCl, we are not aware of attempts to mutate the 0′ arginine to probe its role in selectivity. However, given the marked similarities between the primary sequences of the α1-GlyR and the α1/β-GluCl subunits at the level of the M2 α-helices (Fig. 1 D), it seems prudent to generalize the crucial role of the arginine at position 0′ (which we have recently identified in the β-GluCl pore) to the entire subset of anion-selective pLGICs.

In what follows, we refer to the different conformations of amino-acid side chains using the “mpt” nomenclature of the Penultimate Rotamer Library (27). Briefly, for each sp³-sp³ bond, the torsional free-energy landscape has three minima corresponding (in ideal cases) to the well-known −60° (m, for “minus”), +60° (p, for “plus”), and 180° (t, for “trans”) staggered-angle arrangements (Fig. S2). On the other hand, for bonds involving planar functional groups, the location of minima is more difficult to anticipate, and thus, the corresponding dihedral angles (such as the χ₄ dihedral angle of arginine) are designated using numerical values. Here, we simulated six rotameric conformations of arginine—out of the 34 most frequently observed in high-resolution x-ray crystal structures of soluble proteins (27)—designated as (χ₁, χ₂, χ₃, χ₄): mtt180°, mtm105°, ptt85°, ttp85°, mmm180°, and tpt85°, with m-, p-, and t-specific values for each rotamer given by Coot (version 0.7, revision 4459; (28)). These particular six rotamers provide a wide sampling of arginine’s side-chain torsional space and include its most frequent (9%) conformation in the library (that is, the mtt180° rotamer; (27)).

Molecular simulations of ion permeation

Homomers of the α1-GluCl subunit from C. elegans form anion-selective channels (29). As is the case for all other anion-selective pLGICs, a basic residue occupies position 0′ in the first α-helical turn of M2, and this position faces away from the pore and toward the M1 and M3 transmembrane α-helices from the same subunit (12). We set out to probe the effect of different rotamers of the buried arginine at position 0′ on the anion selectivity of this channel using molecular simulations. To reduce the complexity of the system, we simulated only the stretch between the beginning of M1 and the end of M3 of each subunit in the pentamer (the bound molecules of ivermectin were removed); it is this stretch that has the highest impact on ion permeation (2, 26, 30).

We started by computing the ion-permeation potential-of-mean-force (PMF) profile of this reduced model using umbrella-sampling MD. The solution bathing both ends of the model was 150 mM KCl, and the electrical potential was zero. Quite unexpectedly for an anion-selective channel, this profile showed a larger energetic barrier to the permeation of Cl⁻ (∼15 kcal ⋅ mol⁻¹; Fig. 2) than to the permeation of K⁺ (∼8 kcal ⋅ mol⁻¹; Fig. 2), and even the latter seemed uncharacteristically large for an open-channel model. Inspection of the structural model (PDB: 3RHW) indicated that the five arginine side chains at position 0′ occupy the mtt180° rotamer (Figs. 3 and S3), a side-chain conformation that is amply justified by the corresponding electron-density map. Inspection of the MD trajectories indicated that the five arginines remained within this “slice” of torsional space within each umbrella window.

Figure 2.

One-ion-permeation PMF profile computed using umbrella-sampling MD. The profile was computed at zero voltage and under symmetrical (150 mM KCl on both sides) ion-concentration conditions for a reduced structural model of α1-GluCl (a truncated version of PDB: 3RHW; (12)). The barrier to Cl⁻ permeation was higher than that to K⁺. Errors were estimated using 1000 iterations of bootstrapping error analysis (18), and they were smaller than the thickness of the lines for the entire length of the pore. Arrowheads indicate the approximate locations of the Cζ of the arginine at position 0′, the Cγ of the leucine at position 9′, and the Cβ of the serine at position 20′ along the axis of ion permeation. To see this figure in color, go online.

Figure 3.

Six rotamers of arginine: view from the intracellular side. (A–F) The different arginine rotamers studied in this work were modeled at position 0′ of an α1-GluCl structure (PDB: 3RHW; (12)) using Coot (28). Only the five M2 α-helices are shown. The mtt180° rotamer is the conformation adopted by the arginine at position 0′ in the original crystal structure model. Side chains are displayed in stick representation with carbon atoms in yellow, nitrogens in blue, and oxygens in red. Molecular images were prepared with VMD (31). To see this figure in color, go online.

We then proceeded to simulate current-voltage (I-V) relationships. To isolate the effect of specific side-chain rotamers from the effects of all other variables that could also change during an MD simulation in which ion crossings are counted, we decided to use BD simulations. In BD, the protein, the phospholipid membrane, and the water outside and inside the transmembrane pore are treated implicitly as static homogeneous media with defined dielectric constants. Only the mobile ions are treated explicitly and are allowed to move, with assumed values for their diffusion coefficients inside the pore (21). As for the structural features of the different models, these are captured by the particular distribution of fixed partial atomic charges contained within the low-dielectric region representing the ion-channel protein. The successful application of this type of coarse-grained simulation to probe the effect of glutamate rotamers on the single-channel conductance of a structural model of the muscle AChR (11) lends credence to the use of this computational approach to the study of charge selectivity in AChR-related channels. In the open-channel conformation, the animal members of this superfamily display narrowest pore constrictions that are wide enough (∼5–6 Å in diameter) for them to be appropriate targets of this computational approach.

We simulated ion conduction through the α1-GluCl reduced model under asymmetrical (150 mM KCl on the intracellular side and 15 mM KCl on the extracellular side) and symmetrical (150 mM KCl both sides) ion-concentration conditions. The asymmetrical situation has the advantage of mimicking the “dilution” conditions used during typical charge-selectivity experiments. As is the case for the corresponding experimental I-V relationships, each simulated “asymmetrical” I-V curve yields a zero-current (“reversal”) potential from which a permeability-coefficient ratio can be calculated. The disadvantage of this simulation approach, however, is that these I-V curves often display marked rectification (much like their experimental counterparts), and thus, the value of the reversal potential may be difficult to identify in some cases. Simulations under symmetrical ion-concentration conditions across the membrane typically relieve the rectification, but the curve’s reversal potential becomes zero irrespective of the channel’s charge selectivity. To address this limitation, we took advantage of the fact that cation crossings can be counted separately from anion crossings in simulations. Thus, under symmetrical conditions, we estimated charge selectivity from the ratio between the simulated cation-carried current and the anion-carried current at different membrane potentials. Although this ratio cannot be compared to any electrophysiological observable, we deemed it to be a valid estimator of charge selectivity.

The effect of arginine side-chain conformation

Fig. 4 shows simulated I-V curves for the α1-GluCl model. Surprisingly for an anion-selective channel but in full agreement with the corresponding umbrella-sampling MD PMF profile (Fig. 2), the reversal potential under asymmetrical conditions was ∼−56.0 mV (Fig. 4 A), nearly the same value as the equilibrium potential for the cation under a 150/15 KCl concentration ratio and a temperature of 300 K (−59.5 mV; −55.0 mV, using ion activities). Consistent with this observation, the I-V curve simulated under symmetrical KCl conditions shows that most of the inward current through this model is carried by K⁺ (∼83% of the net current at −200 mV), with only a minor fraction (the remaining ∼17%) being carried by outward-going Cl⁻ (Fig. 4 B). Because the implicit representation of protein structure used during BD remains fixed, the arginine side chains can be thought of as having remained in the original model mtt180° rotamer throughout these simulations. We proceeded, then, to simulate the effect of different arginine side-chain conformations on charge selectivity using symmetrical ion-concentration conditions. For these simulations, only the conformation of the five arginine side chains at position 0′ was changed (using Coot; (28)); all other structural aspects remained as specified by the x-ray crystal structure model. We tested five different rotamers of arginine (Figs. 3, B–F, 4, C–G, and S3) and found that all of them switch the sign of the charge selectivity of the model, albeit to very different extents. The rank order of this anion-selectivity-conferring effect of the arginine side chain was mtm105° < ptt85° ∼ ttp85° ∼ mmm180° < tpt85°. Whereas in the original model (in which the arginines occupy the mtt180° rotamer), Cl⁻ carried only ∼17% of the net current at −200 mV, changing the five 0′ arginine rotamers to tpt85° increased the fraction of the net current carried by Cl⁻ to ∼95% (Fig. 4 G). Consistent with this observation, the reversal potential of an I-V curve simulated under 150/15 mM KCl conditions for the GluCl model with tpt85° 0′ arginines was ∼55.0 mV (Fig. 4 H), nearly the same value as the equilibrium potential for the anion under a 150/15 KCl concentration ratio and a temperature of 300 K (59.5 mV; 55.0 mV, using ion activities). To establish a “baseline” for the effect of positive charges at position 0′, we also mutated the five arginines to alanines (using psfgen in VMD; (31)) and simulated the I-V relationship under symmetrical conditions (Fig. 4 I). For this mutant, we found that Cl⁻ carries only ∼2.3% of the net current at −200 mV. We note, then, that the crystal structure mtt180° rotamer of arginine increased the fraction of current carried by Cl⁻ relative to an alanine side chain (from ∼2.3 to ∼17%) but not enough to render the channel as anion-selective as it is known to be from experiments (29). We would like to emphasize, here, that by no means do these results imply that that the tpt85° rotamer is the actual conformation of the 0′ arginines in the membrane-embedded α1-GluCl, let alone that all five of them have to adopt the same rotamer or that side chains are fixed in a single conformation. Our results simply suggest that merely changing the conformation of ionized side chains at the charge-selectivity filter—while keeping everything else constant—may have a large impact on the charge selectivity of pLGICs even when these side chains do not face the lumen of the transmembrane pore directly. We do not know the reason why, in this structural model, the arginine side chain adopts a conformation that is not consistent with anion selectivity, but the fact that the x-ray diffraction data were obtained from detergent-solubilized GluCl suggests an explanation. Regarding mechanisms, nothing more complex than electrostatics needs to be invoked here (32, 33): formal charges on ionized side chains interact with the passing ions in a manner that depends on distance, and different rotamers place the charged moiety at different positions. Although ionized groups on the front side of M2 exert a larger effect on ion permeation than do those on the backside, the effect of the latter is still considerable (2).

Figure 4.

The charge selectivity of the structural model of the ivermectin-bound α1-GluCl. (A and B) BD-computed I-V curves for a reduced structural model of the α1-GluCl channel (a truncated version of PDB: 3RHW; (12)) under asymmetrical (150 mM KCl on the intracellular side and 15 mM KCl on the extracellular side) and symmetrical (150 mM KCl on both sides) ion-concentration conditions, respectively, are shown. In the crystal structure model, the arginine at position 0′ adopts the mtt180° rotamer. (C–H) The effect of different rotamers of the 0′ arginine on charge selectivity estimated using BD-computed I-V curves under symmetrical (C–G) and asymmetrical (H) KCl concentration conditions is shown. (I) The effect of arginine neutralization at position 0′ on charge selectivity is shown. For all panels, symbols denote the mean of six BD simulations at each voltage, error bars (omitted when smaller than the symbols) denote the corresponding standard errors, and solid lines are cubic-spline interpolations. To our knowledge, the single-channel I-V curve of α1-GluCl has not been recorded experimentally, and thus, its shape and slope cannot be compared to those computed here using simulations. We only know that, under asymmetrical KCl conditions, the computed reversal potential needs to be consistent with selectivity for anions. To see this figure in color, go online.

Fig. 5 shows the ion-permeation PMF profiles obtained from the BD simulations in the absence of an applied membrane potential in increasing order of Cl⁻ selectivity. In going from one end of this range (the Arg-to-Ala mutant) to the other (the tpt85° rotamer of arginine), the penalty to K⁺ permeation increased and that to Cl⁻ decreased. In the case of K⁺, note that although the main barrier at position 0′ remained at 6.5–8.0 kcal ⋅ mol⁻¹ for the different arginine rotamers, a second barrier at position 9′ became increasingly prominent in going from Fig. 5, B–G. Also, note that despite the quantitative differences between the PMF profile computed using MD (Fig. 2) and that computed using BD for the (crystal structure) mtt180° rotamer (Fig. 5 B), both point to the same conclusion: the ion-permeability properties simulated for this particular structural model are inconsistent with what is known about α1-GluCl from experiments (29). To calculate the electrostatic component of the ion-permeation free-energy landscape, we computed electrostatic profiles for the reduced model of α1-GluCl with 0′ arginines in the mtt180° (Fig. 6 A) or tpt85° (Fig. 6 B) rotamer using APBSmem (22). These two profiles show that changes of the arginine side-chain conformation alone are sufficient to switch the sign of the ion that faces the larger permeation barrier. Decomposing these plots so as to only show the contribution of the 0′ arginines to the entire transmembrane pore’s profile (Fig. 6, C and D) confirms the idea that the tpt85° rotamer places the guanidinium moiety in such a way that its effect on ion permeation through α1-GluCl is larger than that exerted by this group when the side chain adopts the mtt180° rotamer.

Figure 5.

Multi-ion-permeation PMF profiles computed using BD. These profiles were computed at zero voltage and under symmetrical (150 mM KCl on both sides) ion-concentration conditions for a reduced structural model of α1-GluCl (a truncated version of PDB: 3RHW; (12)). (A) The effect of arginine neutralization at position 0′ is shown. (B–G) The effect of different rotamers of the 0′ arginine is shown. The mtt180° rotamer, in (B), is the conformation adopted by the 0′ arginine in the crystal structure model of α1-GluCl. Darker lines denote the mean of six BD simulations, and the lighter shades denote the corresponding standard errors. Arrowheads indicate the approximate locations of the Cζ of the arginine at position 0′, the Cγ of the leucine at position 9′, and the Cβ of the serine at position 20′ along the axis of ion permeation. To see this figure in color, go online.

Figure 6.

The electrostatic component of the ion-permeation free-energy landscape. The electrostatic energy was calculated for a reduced structural model of α1-GluCl (a truncated version of PDB: 3RHW; (12)) using APBSmem (22). (A and B) The effect of different rotamers of the arginine at position 0′ is shown. The mtt180° rotamer, in (A), is the conformation adopted by the 0′ arginine in the crystal structure model of α1-GluCl. The tpt85° rotamer, in (B), is the conformation of the 0′ arginine that conferred the highest selectivity for anions in our BD simulations of the reduced structural model (Figs. 4G and 5G). (C and D) Specific contribution of the arginines at position 0′ (in the mtt180° and tpt85° rotamers, respectively) to the electrostatic profile is shown. Arrowheads in (A) indicate the approximate locations of the Cζ of the arginine at position 0′, the Cγ of the leucine at position 9′, and the Cβ of the serine at position 20′ along the axis of ion permeation. To see this figure in color, go online.

Regarding the possible limitations of using reduced models of the protein for the simulations, it may be asked, for example, whether having included the extracellular domain of α1-GluCl could have imparted the anion selectivity that the simulated M1-M3-only model of the protein (with the arginine at position 0′ in its original rotamer) lacked. In this regard, we note that electrophysiological experiments have clearly shown that a chimeric construct containing the transmembrane domain of the anion-selective β-GluCl (which is highly homologous to α1-GluCl) and the extracellular domain of the cation-selective α7 AChR remains highly selective for anions (1, 9). It follows, then, that the transmembrane domain of GluCl is sufficient to confer anion selectivity; it does not matter whether the extracellular domain corresponds to an anion-selective or a cation-selective pLGIC. Thus, we infer that the lack of the extracellular domain in our modeled structure is unlikely to underlie our observations.

We then turned to the model of the α1-GlyR bound to glycine and ivermectin, which we have recently deemed to closely approximate the receptor’s open-channel conformation (3). As was the case for the α1-GluCl model, we removed the bound molecules of ivermectin and simulated only the stretch between the beginning of M1 and the end of M3 of each subunit in the pentamer. Under symmetrical conditions, we found that ∼96% of the net current is carried by Cl⁻ (Fig. 7 A). Consistent with this finding, the I-V curve simulated under asymmetrical conditions reverted at ∼57.0 mV (Fig. 7 B), nearly the same value as the equilibrium potential for the anion under a 150/15 KCl concentration ratio and a temperature of 300 K. The conformation of the arginines at position 0′ of this (original) GlyR model is mmm180°, a rotamer that, when modeled in the background of the α1-GluCl structure, also supported selectivity for anions (∼82.5% of the net current at −200 mV was carried by Cl⁻; Figs. 4 F and 5 F). The cryo-EM density in this region of the map does not point unambiguously to this particular side-chain conformation; it seems as though other arginine rotamers could also have been modeled.

Figure 7.

The charge selectivity of the structural model of the glycine-and-ivermectin-bound α1-GlyR. (A and B) BD-computed I-V curves for a reduced structural model of the α1-GlyR (a truncated version of PDB: 3JAF; (13)) under symmetrical (150 mM KCl on both sides) and asymmetrical (150 mM KCl on the intracellular side and 15 mM KCl on the extracellular side) ion-concentration conditions, respectively, are shown. (C) The effect of arginine neutralization at position 19′ on charge selectivity estimated using BD-computed I-V curves under symmetrical KCl concentration conditions is shown. (D) The effect of changing the rotamer of the arginine at position 0′ to mtt180° on charge selectivity in the background of a model with an all-neutral position 19′ is shown. The rotamer of the 0′ arginine in the cryo-EM model is mmm180°. (E) The effect of arginine neutralization at positions 0′ and 19′ on charge selectivity is shown. For all panels, symbols denote the mean of six BD simulations at each voltage, error bars (omitted when smaller than the symbols) denote the corresponding standard errors, and solid lines are cubic-spline interpolations. To see this figure in color, go online.

Unlike α1- or β-GluCl, the α1-GlyR contains two rings of basic residues in M2 (Fig. 1 D). In addition to the arginine at position 0′, the α1-GlyR has a second arginine at position 19′ whose mutation to alanine has been found not to compromise the channel’s high selectivity for anions (P_Cl⁻/P_K⁺ = 285; (1)). Thus, to extend our observations to the GlyR and be able to focus on the arginines at position 0′, we mutated the five arginines at position 19′ to alanines (using psfgen in VMD; (31)). We simulated the corresponding I-V relationship under symmetrical conditions, and—in keeping with the experimental evidence (1)—we found that the mutated model retains a high selectivity for anions (Cl⁻ still carried ∼91% of the net current at −200 mV; Fig. 7 C). Using this mutant-GlyR model as the background, we proceeded to probe the effect of different side-chain conformations at position 0′. Changing the original rotamer (mmm180°) to that of the 0′ arginine in the crystal structure model of α1-GluCl (mtt180°) led to a rather nonselective channel; ∼60% of the net current was carried by Cl⁻ (Fig. 7 D). Thus, in the open-channel structural models of both the α1-GluCl channel and the R19′A mutant of the α1-GlyR, the mtt180° rotamer of arginine at position 0′ was unable to confer selectivity for anions. Mutating the five arginines at position 0′ to alanine in the background of the R19′A model (for which there is no experimental counterpart) rendered the channel cation selective (85% of the net current was carried by K⁺; Fig. 7 E). This observation is in full agreement with the role of the 0′ arginines in the charge selectivity of anion-selective pLGICs inferred from experiments on β-GluCl (1).

Having found support for the idea that different conformations of the arginine at position 0′ contribute to charge selectivity to different extents, we proceeded to examine the six simulated rotamers (Fig. 3) in search for structural clues as to why this is the case. We analyzed different geometric variables of the six α1-GluCl models and found that the BD-computed charge selectivity (expressed as the ratio of anion-carried current/cation-carried current at −200 mV under symmetrical KCl conditions) decreases nearly monotonically with the distance between the Cζ atoms of the guanidinium groups (Fig. S2 A) and the center of mass of the five-leucine ring at position 9′ located halfway through the transmembrane pore (Fig. 8). On the other hand, no simple relationship seems to exist between the BD-computed charge selectivity and the distance between the Cζ atoms and the pore’s central axis (Fig. 8), a distance that disregards the position of the guanidinium group along this axis. Remarkably, a similar analysis performed for the effect of six rotamers of glutamate at the adjacent position −1′ of the muscle AChR on the BD-computed single-channel conductance yielded, essentially, the same result (11): the mean distance between the side chains’ charged moieties and the 9′-position five-leucine ring was a better indicator of electrostatic impact on ion permeation than was the “horizontal” distance to the central axis.

Figure 8.

Charge selectivity-distance relationships. The mean distance between the Cζ atoms (Fig. S2A) of the five arginine side chains at position 0′ of a structural model of the α1-GluCl channel (PDB: 3RHW; (12)) and the pore’s central axis is plotted using orange symbols for the six rotamers of arginine studied here (Figs. 3 and S3). The mean distance between all five Cζ atoms and the center of mass of the 9′-leucine ring—located halfway through the transmembrane pore—is plotted in purple for the same six models. Charge selectivity (plotted on the y axis) was inferred from the ratio between the anion-carried current and the cation-carried current obtained from BD simulations performed at −200 mV under symmetrical (150 mM KCl) ion-concentration conditions (Fig. 4, B–G). The relationship between the BD-computed charge selectivity at −200 mV and the distances plotted in purple is the more straightforward of the two; hence, the fit to a straight line (solid line). Symbols denote the mean of six BD simulations at −200 mV, and error bars (omitted when smaller than the symbols) denote the corresponding standard errors. To see this figure in color, go online.

The effect of pore width

It has been known for many years (4, 5, 8) that insertions and deletions in the M1-M2 linker and residue-to-residue mutations at position −2′, both involving nonionizable residues, often have a large effect on charge selectivity. Our recent study of this phenomenon using electrophysiological recordings led us to propose that these mutations exert their effect by changing the rotamer preferences of the native charged side chains in the first turn of the M2 α-helices (1). However, others (e.g., (34)) have suggested that amino-acid insertions and deletions in the M1-M2 linker exert their effect by changing the size of the pore. According to this view, 1) insertions contract, whereas deletions expand, the narrowest constriction of the pore; and 2) narrower pLGIC pores support selectivity for anions, whereas wider pores support selectivity for cations. Although the existence of wild-type cation-selective γ-aminobutyric acid type-A receptors (GABA_ARs) with long M1-M2 linkers (Fig. 1 B; (35)) and anion-selective AChRs with short M1-M2 linkers (Fig. 1 B; (36)) calls these ideas into question, we wanted to test the suggestion of a relationship between charge selectivity and pore size using ion-permeation simulations. To this end, we took the M1-M3 model of the α1-GlyR and translated each of the five subunits as rigid bodies radially, away from the permeation axis, while keeping the fivefold symmetry of the structure. Using this protocol, we generated two models wider than the original one by 2 and 4 Å in diameter (Fig. S4 A). Ion-permeation BD simulations performed under symmetrical KCl conditions revealed that, although the anion selectivity seemed to decrease as the pore was made wider, the channel retained high selectivity for anions. Indeed, whereas ∼96% of the net current at −200 mV was carried by Cl⁻ through the original model (Fig. 7 A), this value was ∼94% through the 2-Å-wider (in diameter) model (Fig. S4 B) and ∼88% through the 4-Å-wider model (Fig. S4 C). Consistent with the expected effect of pore width on single-channel conductance, the computed value of the net current increased with pore size. At −200 mV, for example, the net current was −4.9 pA for the original model, −8.7 pA for the 2-Å-wider model, and −14.9 pA for the 4-Å-wider model. Of course, we cannot rule out the possibility that, in an experimental setting, pore expansion is accompanied by a change in arginine side-chain conformation and therefore that a change in pore size may end up having an effect on selectivity. However, a change in pore diameter alone did not have any dramatic effect on charge selectivity in these simulations.

The effect of glutamate side-chain conformation

All cation-selective pLGICs have an acidic residue near the front side of the first α-helical turn of M2, at position −1′ (in most cases) or −2′ (Fig. 1, B and C). On the basis of electrophysiological measurements, we have recently shown that the effect of mutating the alanine at position −1′ of the (anion-selective) α1-GlyR to glutamate on charge selectivity depends markedly on the amino-acid sequence of the neighboring residues (1). For example, mutating the anion-selective rat α1-GlyR (P_Cl⁻/P_K⁺ = 119 ± 57) so that the sequence of the stretch including the M1-M2 linker and the first turn of M2 becomes INMD-AAER (positions −8′ through 0′, with mutated positions indicated in bold; the wild-type sequence is INMDAAPAR; Fig. 1 B) switched the sign of the channel’s charge selectivity (P_K⁺/P_Cl⁻ = 11 ± 2), but mutating this stretch to INMDAAGER, instead, rendered the α1-GlyR rather nonselective for cations or anions (P_K⁺/P_Cl⁻ = 1.3 ± 0.2). Although the few differences between these two sequences were not expected to lead to drastic rearrangements of the secondary structure, we could not rule out the occurrence of subtler changes that could have affected the rotamer occupancies of the introduced glutamate side chains. Because earlier we had found evidence—both electrophysiological (10) and computational (11)—for the distinct contributions of different glutamate rotamers to the single-channel conductance of the muscle AChR, we proposed that the same could be the case for the charge selectivity of these mutant GlyRs (1). To test this idea using BD, we replaced all five alanines at position −1′ of the open-channel GlyR model with glutamates and performed simulations under symmetrical KCl conditions. We simulated four models—each one having all five glutamates in the same conformation—differing from each other in the rotamer adopted by these glutamates. The simulated rotamers were pt, mt, mm, and tp. When the effect of these glutamate conformations on single-channel conductance was simulated in the background of an open-channel model of the muscle AChR, the computed single-channel conductance grew in the order pt < mt < mm < tp (11).

Fig. 9, A–D show the simulated I-V curves, and Fig. 9, E–H show the graphical representations of the simulated glutamate side-chain conformations. Consistent with their differential contribution to the computed single-channel conductance of the (cation-selective) AChR, the GlyR model with five pt (Fig. 9 A) or mt (Fig. 9 B) glutamates at position −1′ retained selectivity for anions (Cl⁻ still carried ∼76% of the net current at −200 mV for both models). In contrast, the GlyR model with five mm (Fig. 9 C) or tp (Fig. 9 D) glutamates at position −1′ had its charge selectivity switched: in the presence of mm glutamates, K⁺ carried ∼98% of the net current at −200 mV, whereas in the presence of tp glutamates, this value was ∼95%. Remarkably, the cation selectivity-conferring effect of the mm and tp rotamers of glutamate had nothing to do with the size of the pore. Certainly, regardless of whether an alanine or a glutamate occupied position −1′ of the GlyR model and regardless of whether the mutated glutamate was modeled in the pt, mt, mm, or tp rotamer, the narrowest constriction (estimated using Hole; (37)) remained unchanged at ∼5.0 Å in diameter.

Figure 9.

Switching the charge selectivity of the GlyR. Some, but not all, rotamers of glutamate conferred cation selectivity to the Ala-to-Glu mutant at position −1′ of a reduced model of the glycine-and-ivermectin-bound α1-GlyR (a truncated version of PDB: 3JAF; (13)). (A–D) The effect of four different rotamers of the glutamate at position −1′ on charge selectivity estimated using BD-computed I-V curves is shown under symmetrical (150 mM KCl on both sides) ion-concentration conditions. Symbols denote the mean of six BD simulations at each voltage, error bars (omitted when smaller than the symbols) denote the corresponding standard errors, and solid lines are cubic-spline interpolations. (E–H) The four rotamers of glutamate were modeled at position −1′ of the α1-GlyR structural model using Coot (28). Only the five M2 α-helices are shown. The view is from the intracellular side. Side chains are displayed in stick representation with carbon atoms in yellow, nitrogens in blue, and oxygens in red. Molecular images were prepared with VMD (31). To see this figure in color, go online.

The effect of optimizing side-chain conformations

The strong anion selectivity of the α1-GlyR model obtained in simulations (Fig. 7, A and B) is consistent with the electrophysiologically estimated charge selectivity of the GlyR channel. However, the rectifying shape of the I-V relationship simulated under symmetrical KCl conditions (Fig. 7 A) stood out as a clear difference when compared to the experimentally observed linearity of this relationship (e.g., (3)) and thus made us wonder about the limitations of the cryo-EM model. In particular, we wondered about the validity of the side-chain conformations (in general, not only of the 0′ arginines) because they are far from being unambiguously defined by the experimental data (PDB: 3JAF) and because their implicit representation in BD remains fixed during the simulations. To address this concern, we proceeded to optimize the conformations of side chains using SCWRL4—a program that predicts side-chain conformations on the basis of interatomic interaction energies and backbone-dependent rotamer frequencies as observed in high-resolution structures of soluble proteins (38)—and simulated ion permeation through the resulting new model. The main difference between the side-chain-optimized and the original structural models at the level of the transmembrane pore lining was the conformation of the leucine at position 9′, in the middle of M2. This change resulted in the narrowing of the pore in this region (Fig. S5 A), but the narrowest constriction along the ion-permeation pathway remained at the first intracellular turn of M2 with an unchanged value of ∼5.0 Å in diameter. Moreover, the rotamers of the buried arginines at positions 0′ and 19′ also changed (from mmm180° to mtm105°, in the case of position 0′). We simulated this side-chain-conformation-optimized model and found that, under asymmetrical KCl conditions (Fig. S5 B), the reversal potential is 59.5 mV, in remarkable agreement with the equilibrium potential for Cl⁻ under this KCl gradient (59.5 mV; 55.0 mV, using ion activities). Under symmetrical KCl conditions (Fig. S5 C), the model retained high selectivity for Cl⁻ over K⁺, and the I-V curve became more rectilinear, the single-channel conductance in the inward direction (∼93 pS) being in good agreement with experimental estimates in cell-attached patches with ∼150 mM Cl⁻ in the pipette solution (∼95–99 pS; (3)). Furthermore, the effect of pore expansion (Fig. S5 D) and the effect of neutralizing mutations at positions 0′ and 19′ (Fig. S5, E and F) were essentially the same as those computed in the background of the original M1-M3 structure. Thus, although optimizing the conformations of side chains led to a better agreement between simulated and experimentally obtained data, the high selectivity for anions and its absolute dependence on the buried arginine at position 0′ was still well-captured by the original reduced model. Evidently, charge selectivity in pLGICs seems to be much more insensitive to the pore-radius profile than are the value of the single-channel conductance and the conductive versus nonconductive nature of the pore.

Discussion

Using electrophysiological recordings, we have previously shown that acidic residues at pore-facing position −1′ and basic residues at buried position 0′ (Fig. 1 C) are the main determinants of charge selectivity in the pLGICs from animals (1). However, one of the most striking conclusions of these experiments was that the mere presence of charged side chains is not sufficient to impart selectivity for cations or anions. Indeed, we found that the charge selectivity of pLGICs is sensitive to the details of the local amino-acid sequence that have nothing to do with formal charges, such as the number and identity of nonionizable residues in the M1-M2 linker (Fig. 1 A). For example, mutating the cation-selective mouse serotonin type-3A receptor (5-HT_3AR; P_K⁺/P_Cl⁻ = 31 ± 4) so that the sequence of the stretch including the M1-M2 linker and the first turn of M2 becomes LPPDSΑGAR (positions −8′ through 0′, with mutated positions indicated in bold; the wild-type sequence is LPPD-SGER; Fig. 1 B) reduced the selectivity for cations (P_K⁺/P_Cl⁻ = 7.6 ± 0.5), whereas mutating this stretch to LPPD-SAAR (P_K⁺/P_Cl⁻ = 1.4 ± 0.1) or LPPD-SPAR (P_K⁺/P_Cl⁻ = 0.78 ± 0.06) rendered the channel nonselective for cations or anions, and mutating it to LPPDSΑPAR switched the sign of the charge selectivity (P_Cl⁻/P_K⁺ > 9.2). Similarly, the mutant α1-GlyR with an INMD-APAR motif was highly selective for anions (P_Cl⁻/P_K⁺ = 36 ± 6), the mutant with an INMD-AAAR motif was only mildly selective for anions (P_Cl⁻/P_K⁺ = 6.5 ± 0.5), and the mutant with an INMD-AGAR motif was nonselective (P_Cl⁻/P_K⁺ = 1.3 ± 0.1).

In this work, a computational approach provided ample support for the notion that the charge selectivity of pLGICs is not uniquely determined by the mere charge sign of the native ionized side chains. Instead, the conformation of these side chains seems to be critical for the magnitude of their electrostatic effect on the passing ions. Our results lend credence to the notions that 1) different rotamers of ionized side chains at positions −1′ and 0′ contribute to charge selectivity to different extents and 2) the occupancy probabilities of these rotamers depend steeply on the local amino-acid sequence. We envision that, in anion-selective pLGICs, the basic side chains at position 0′ would interconvert swiftly among rotamers that support anion selectivity and that the details of the torsional free-energy landscape would limit the frequency and duration of excursions out of this subset. Clearly, the same would be the case for the acidic side chains at position −1′ of cation-selective pLGICs.

However unrealistic it may seem to have kept the protein (or, more accurately, its implicit representation) fixed during BD ion-permeation simulations, we considered that this was necessary to analyze the specific effect of different side-chain rotamers and pore width on charge selectivity without the interference of other variables that could also have changed during more detailed (and less restrained) computational approaches, such as all-atom MD. In addition, from a purely practical point of view, it seems unlikely that a study of the sort we presented here—which required counting ion crossings at several transmembrane voltages (six replicates per voltage) under symmetric and asymmetric KCl ion-distribution conditions for a variety of side-chain conformations, mutant side chains, and pore widths—could have been completed in a reasonable timeframe using MD even with special-purpose hardware of the kind used in, for example, (39). It should be noted that to be able to make meaningful comparisons with the existing experimental data, we restricted our simulations to voltages in the experimentally accessible range (that is, lower than 200 mV). At these voltages, however, ion permeation is a relatively slow process. For example, it takes ∼320 ns for 10 ions to cross a 100-pS channel (like the α1-GlyR) in a net manner at 50 mV (∼160 ns at 100 mV, ∼110 ns at 150 mV, and ∼80 ns at 200 mV), and it takes five times longer for 10 ions to do so through a 20-pS channel (like, perhaps, α1-GluCl). To ensure robust statistics, each simulation reported in this work was 2-μs long (per construct, per voltage, per replicate). Nevertheless, despite the higher computational cost of MD, we would like to stress that the main reason why we used BD here to count ion crossings was a conceptual one: we needed to have a strict control over the protein structure to dissect the specific role of side-chain conformation and pore width on charge selectivity. Clearly, future work will have to address the question as to how the rest of the protein reacts to different rotamers of the selectivity filter’s charged side chains and whether and how these changes, in turn, affect the ion-permeation free-energy profile. We expect to study these conformational changes using all-atom MD, but given the well-known dependence of pLGIC function on the chemical composition of the lipid bilayer (e.g., (40, 41)), realistic models of cell membranes should ideally be used (e.g., (42)), and a more detailed understanding of the lipid requirements for function of the simulated channels will have to be attained.

It remains to be understood how the length and sequence of the M1-M2 linker affect the rotamer occupancies of side chains in the (adjacent) positions −1′ and 0′ in the first turn of M2. This knowledge would be necessary if we wished to design ion channels with desired charge selectivities and single-channel conductances. Certainly, we cannot predict charge selectivity from the mere inspection of amino-acid sequences of mutant pLGICs, yet. Although extracellular domain-transmembrane domain chimeras have allowed the generation of pLGICs with novel combinations of ligand specificities and charge selectivities, we still do not know how to engineer opposite charge selectivities—with high, wild-type-like permeability-coefficient ratios—without changing entire transmembrane domains and without concomitantly reducing the single-channel conductance. As interest in developing chemogenetic tools with increasingly finely tuned functional properties continues to grow (43, 44), we anticipate that this knowledge will prove highly instrumental.

Conclusion

Although widely accepted as crucial for enzyme catalysis, side-chain conformation has not typically been considered to be an important variable in the permeation of ions through charge-selective channels. Indeed, the idea that stereochemical considerations matter for selectivity and single-channel conductance may seem overly sophisticated for ion channels that discriminate among biologically relevant ions only on the basis of their charge. However, several lines of experimental (1, 10) and computational evidence ((11) and this work) now point strongly to the major impact of side-chain conformation on ion permeation through pLGICs. Higher-resolution structures, and the development of experimental and computational methods aimed at facilitating the identification of the functional states these models represent, will be absolutely needed to advance our understanding of charge-selective ion permeation through this class of ion channel.

Author Contributions

T.J.H. and C.G. designed research, analyzed data, and wrote the manuscript. T.J.H. performed research.

Editor: Andrew Plested.