Using molecular dynamics to elucidate the structural basis for function in pLGICs

Ligand-gated ion channels incorporate neurotransmitter binding sites and a transmembrane ion channel into a single protein complex. At fast chemical synapses, they detect the presence of extracellular neurotransmitters and open and alter the cellular membrane potential. The pentameric ligand-gated ion channels (pLGICs), also known as the Cys-loop receptors, are a major family of neurotransmitter receptors (1). Neuroscientists and biophysicists have sought to elucidate the structural basis for the functions of these channels in ever greater detail. The recent proliferation of pLGIC crystal structures has made the goal of understanding the atomic basis for their functional properties feasible. However, the crystal structures only provide static pictures. It is their dynamic properties, the conformational changes they undergo from closed to open to desensitized states, that are of fundamental interest. Computational biophysics and molecular dynamics (MD) provide tools to bridge the gaps between the static images and to create dynamic movies, at an atomic level, of these channels as they transition between functional states. In the paper by Calimet et al. in PNAS (2), the authors sought to accomplish this goal for pLGIC members using MD based on X-ray crystal structures.

The pLGIC superfamily includes receptors for acetylcholine (nACh), serotonin type 3 (5-HT3), γ-amino butyric acid (GABA_A), glycine (Gly), and, in invertebrates, receptors for histamine, serotonin, and glutamate (distinct from the excitatory glutamate receptors) (1). Prokaryotic family members also exist (3). pLGICs contain five identical or homologous subunits arranged quasisymmetrically around the central channel axis. All subunits have a similar transmembrane topology, with an ∼200-aa extracellular N-terminal domain and a similarly sized C-terminal domain with four transmembrane segments (M1, M2, M3, and M4). The ligand-binding sites are located in the extracellular domain (ECD) at the subunit–subunit interface about 30 Å above the surface of the membrane (4). The channel is lined by residues from the M2 membrane-spanning segment. Several locations have been suggested for the position of the channel gate that blocks ion conduction in the closed state. One is in the middle of the channel (4, 5), and the other is at the cytoplasmic end (6). These are not necessarily mutually exclusive, there may be multiple narrow regions in the closed channel. Furthermore, the closed channel gates may be distinct from

Despite the concerns about which state the structures represent, much can be learned from the MD simulations reported by Calimet et al.

the gate(s) that prevent conduction in the desensitized states. Channel gating involves conformational changes propagating from the binding site to the gate.

pLGICs have at least three functional states, but probably more (7): a closed, resting nonconducting state that predominates in the absence of agonist; and, in the presence of agonist, two activated states (i.e., an open, conducting state and a desensitized, nonconducting state). The open state is a transient, metastable state that exists for tens of milliseconds to seconds before transitioning to the desensitized state(s). In the continuous presence of agonist, virtually all of the channels will desensitize.

Over the past decade, medium- to high-resolution pLGIC structures have been solved. The Torpedo nACh receptor structure was solved in the absence of agonist by cryo-electron microscopy (4). Subsequently, X-ray crystal structures of two prokaryotic pLGICs were solved: the Erwinia chrysanthemi ion channel (ELIC) and the Gloeobacter violaceus ion channel (GLIC) (8–10). At the time it was crystallized, ELIC’s agonist was unknown, and it was assumed to be in a closed state, although the relevance of ELIC’s structure for eukaryotic pLGICs has been questioned (11). GLIC is activated by low pH and was crystallized at pH 4–4.6, presumably in an activated state. Most recently, the Caenorhabditis elegans glutamate-gated Cl⁻ channel (GluCl) α subunit structure was solved (12). GluClα was crystallized in the presence of its agonist glutamate and an allosteric activator, ivermectin. In the structure, all five binding sites for each molecule are occupied. Because glutamate and ivermectin activate the GluClα channel, Calimet et al. explicitly assume that the crystal structure represents an open state (2). The ELIC, GLIC, and GluCl structures serve as the starting points for the MD studies. The authors explicitly assume that ELIC and GLIC represent closed and open states, respectively, and use 100-ns MD simulations of these two to define “structural observables” for the closed and open states.

Do the Crystal Structures Represent an Open or Desensitized State?

This raises a fundamental issue: do the GLIC and GluCl structures represent an open state or a desensitized state? This is critical because, if they represent the open state, then the MD trajectories may describe the process of channel gating (i.e., channel opening and closing). However, if they represent the desensitized state(s), then the structural observables and the GluCl MD trajectory are describing recovery from desensitization. Similar issues have vexed MD studies of voltage-dependent channels (13). Therefore, what is the evidence that the GLIC structure represents the open state? Mainly that crystallization was performed at a pH that activates GLIC and the structure is different from ELIC (2). However, GLIC desensitizes on a time scale of seconds to minutes (14–16), and crystallization takes days to weeks. In the continued presence of agonist, the desensitized state is the most energetically stable state (14, 15). Consistent with GLIC being desensitized, the narrowest point of the channel lumen at the cytoplasmic end is 5–6 Å in diameter, which is too small to allow passage of hydrated monovalent cations (9, 10). Thus, it seems more likely that the GLIC structure represents a desensitized state, not the open state. The GLIC crystal structure with propofol provides further support for it being a desensitized state. Propofol, an i.v. general anesthetic, inhibits GLIC currents (17). It could do this either by stabilizing the closed state or the desensitized state. Propofol binds in the middle of the subunit four-helix bundle. Strikingly, the GLIC structure is identical with or without propofol bound (17). The fact that propofol binding did not alter the GLIC structure (14) suggests that it represents a desensitized state. The final issue, which is in a sense unknowable, is that detergent solubilization may alter the relative stability of the three functional states. Similar issues arise with GluCl, but less functional information is available. The native GluCl channel is a heteropentamer of α and β subunits and is activated by glutamate. Heterologous expression of GluClα alone yields ivermectin but not glutamate-activated currents (18). Given the structural similarity between GluCl and GLIC (12), it is likely that they represent a similar functional state, quite possibly a desensitized state.

Despite the concerns about which state the structures represent, much can be learned from the MD simulations reported by Calimet et al. (2). Over the 100-ns simulations, the ELIC, GLIC, and GluCl with ivermectin bound structures remain relatively stable. In contrast, the 200-ns GluCl simulation following ivermectin removal shows significant changes. Most of the structural rearrangements observed in the MD simulation following ivermectin removal appear to occur independently in each subunit. The subunits do not undergo significant concerted, coupled conformational changes. Ivermectin binds in the subunit interface near the extracellular end of the transmembrane segments. It interacts with M3 in one subunit and M1 in the adjacent subunit. The ivermectin binding site is separated by ∼40 Å from the ECD glutamate binding site. Among the first structural changes that occur following initiation of the ivermectin-free MD simulation is the opening of the C-loop, an anti-parallel β strand that forms a trapdoor over the glutamate binding site. The C-loop is formed by the β9-β10 strands. β10 is connected to the extracellular end of the M1 segment, providing a direct connection between the ivermectin binding site and the C-loop. The C-loop opens in two subunits within the first 15 ns and, in a third subunit, by 40 ns. Curiously, in the other two subunits the C-loop remains closed throughout the simulation. C-loop opening is followed by loosing of the bound glutamates, but during the 200 ns simulation, none of the glutamates exits its binding site. The other early structural change is the initiation of an ECD twist of each subunit relative to its own transmembrane domain. The twisting is complete within 100 ns and reorients the ECD, so that it is more similar to the presumably closed state ELIC ECD structure.

Conformational Coupling Between Extracellular and Transmembrane Domains

Conformational changes also occur at the interface between the ECD and transmembrane domain. The tip of the ECD β1-β2 loop interacts with the M2-M3 loop that connects the M2 and M3 segments. Unwin had noted this interaction and described it as a “pin in socket” interaction (4). Disulfide cross-linking studies have shown that these loops are in close proximity in both GABA_A and 5-HT3 receptors (19, 20). Interestingly, in the 5-HT3A receptor, mutation to Cys of the residue at the tip of the β1-β2 loop, K81, resulted in desensitization of the channels (20). Covalent modification of the Cys with small sulfhydryl reagents partially restored channel function, whereas modification with larger sulfhydryl reagents completely restored function. It suggested that the interaction between the β1-β2 loop tip and the M2-M3 loop was important for recovery from fast desensitization (20). In the MD simulation, following ivermectin removal, the residue at the β1-β2 loop tip, V45, lifts upward to allow an absolutely conserved M2-M3 loop proline, P268, to move inward toward the central channel axis (2). This allows the M2 and M3 segments to move as a rigid body inward toward the central axis. This inward movement of M2 and M3 occurs in the same two subunits that undergo rapid C-loop opening. Coupled movement of M2 and M3 was previously inferred to occur during GABA_A receptor gating (21). During the simulation, the channel cross-sectional area does vary, but it does not change significantly by the end of the simulation (2). Presumably, the simulation’s 200-ns duration is not sufficient to allow the channel to relax to a closed, resting-state conformation. Given the kinetic time constants for conformational transitions in other pLGICs, these conformational changes occur on a microseconds timescale. Thus, to see the full relaxation would require longer MD simulations, which would likely require faster supercomputers. Future studies combining functional, structural, and computational approaches will continue to elucidate the gating processes and the structural basis for function in pLGICs.