The Challenge of Interpreting Glutamate-Receptor Ion-Channel Structures

Abstract

Ion channels activated by glutamate mediate excitatory synaptic transmission in the central nervous system. Similar to other ligand-gated ion channels, their gating cycle begins with transitions from a ligand-free closed state to glutamate-bound active and desensitized states. In an attempt to reveal the molecular mechanisms underlying gating, numerous structures for glutamate receptors have been solved in complexes with agonists, antagonists, allosteric modulators, and auxiliary proteins. The embarrassingly rich library of structures emerging from this work reveals very dynamic molecules with a more complex conformational spectrum than anticipated from functional studies. Unanticipated conformations solved for complexes with competitive antagonists and a lack of understanding of the structural basis for ion channel subconductance states further highlight challenges that have yet to be addressed.

Main Text

Glutamate-receptor ion channels (iGluRs), the major mediators of excitatory synaptic transmission in the brains of vertebrates, couple neurotransmitter release triggered by presynaptic voltage-gated Na⁺ and Ca²⁺ channels to depolarization of postsynaptic neurons in the central nervous system. Experiments using subtype selective ligands, and sequence analysis of iGluR cDNA clones, established that there are four iGluR families in vertebrates, named α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA), kainate, N-methyl-D-aspartate (NMDA), and δ-receptors, and that these share a conserved architecture despite substantial differences in their maximum open probability and rates and extent of activation and desensitization (1). The simplest conceivable kinetic scheme for iGluR gating, defined as conformational changes triggered by the binding of agonists, would include transitions between a resting state and ligand-bound closed, open, and desensitized states (Fig. 1 A), which for NMDA receptors requires binding of both glutamate and glycine (Fig. 1 B). However, from whole-cell recording, single-channel recording, and structural studies we know that iGluR gating is much more complicated and involves multiple conformations and ligand-bound states (Fig. 1 C).

Figure 1.

Domain organization and gating in iGluR tetramers. (A) Simplified state diagram for iGluR gating shows transitions between apo (R), agonist-bound (RA), open (O), and desensitized (D) states. (B) Minimal NMDA receptor state diagram showing that binding of both glutamate and glycine is required to open the channel; desensitized states were omitted from this model. (C) State diagram for an iGluR tetramer during sequential binding of four molecules of glutamate with transitions between apo (R), glutamate-bound (R1–R4), open (O1–O4), and desensitized (D1–D4) states. (D) Atomic coordinates from a GluK2 antagonist-bound resting-state structure, for which three subunits are drawn in transparent gray shading, with the fourth subunit colored using rainbow shading from the N-terminus (blue) to C-terminus (red), highlights assembly of both the ATD and LBD as a dimer or dimers, and connection of the ATD to LBD and LBD to TMD by linker peptides. To see this figure in color, go online.

The central issue that confounds structural analysis of iGluR gating is that these receptors have a unique architecture consisting of a large multidomain extracellular assembly, in which the individual domains are coupled to each other and to the membrane-embedded ion channel via flexible polypeptide linkers. This architecture is both a blessing and a curse. The blessing is that individual iGluR domains can be excised and studied in isolation using high resolution x-ray diffraction, NMR, and other spectroscopic and biophysical techniques such as analytical ultracentrifugation, approaches that are challenging when applied to membrane proteins. The curse is that in intact receptors the extracellular domains are loosely packed, surprisingly mobile, and in cryo-electron microscopy (cryo-EM) experiments adopt an unusually large spectrum of conformations. The field is mature in the sense that structural studies on iGluRs have been pursued for nearly as long as those on voltage-gated ion channels, but with decades of data obtained using patch-clamp techniques, it is proving extremely challenging to correlate kinetic states identified in electrophysiological experiments with an ever-growing library of “full-length” iGluR structures resolved by x-ray diffraction and cryo-EM. In this review, I discuss recent advances in the field and challenges for the future.

iGluR architecture

iGluRs assemble as tetramers, frequently as heteromeric assemblies. The structural core of iGluRs consists of two extracellular domains, a 380 residue amino-terminal domain (ATD), also referred to as the NTD, and a 250 residue ligand-binding domain (LBD) for glutamate (or glycine for vertebrate NMDA receptor GluN1/GluN3 subunits), as well as a transmembrane ion channel domain (TMD). In combination, these three domains form a layered assembly (Fig. 1 D). The ATD is connected via linkers of ∼20 residues to the LBD, which in turn is connected to the TMD via a set of three linkers each ∼10 residues in length. The ion channel is formed by three membrane-spanning α-helices (M1, M3, and M4), with a short pore helix (M2) followed by a pore loop, but with an inverted orientation compared to the architecture of voltage-gated ion channels. Currently, the Protein DataBank and EMDataBank have entries for 64 “full-length” iGluR structures, 21 solved by x-ray diffraction at resolutions of 3.2–4.8 Å and 43 solved by cryo-EM at resolutions of 3.8–26 Å. With the exception of cryo-EM structures for full-length AMPA and kainate receptors, the majority of iGluR structures solved to date have been for genetically engineered constructs in which the cytoplasmic domain (CTD), which varies in length from 10 to 150 residues in AMPA and kainate receptors, and from 100 to 630 residues in NMDA receptors, was deleted. Secondary-structure prediction algorithms and circular dichroism spectroscopy for an isolated GluN2 CTD reveal this structure to be largely disordered (2), and consistent with this, for kainate receptor GluK2 cryo-EM structures, density was weak for this region (3).

The majority of iGluR structures trapped in what is almost certainly a resting state have in common the following features: the ATD and LBD layers are both formed by a dimer-of-dimers assembly, with dimer pairs related by quasi-twofold molecular symmetry, but with subunit crossover between the ATD and LBD layers, such that the AB and CD subunits form ATD dimers while the AC and BD subunits form LBD dimers (4, 5, 6, 7, 8). By contrast, the TMD is fourfold symmetric. Symmetry in the ATD and LBD layers is broken, to different extents, by off-axis rotations and tilts of the dimer pairs, particularly for NMDA receptor structures (9). Contacts between the ATD and LBD layers are loose in AMPA and kainate receptors (8), whereas for NMDA receptors these domains interact via tightly packed molecular surfaces (6, 7, 10, 11). Compared to iGluR crystal structures, cryo-EM structures typically show greater symmetry break down, likely reflecting the greater conformational freedom of single molecules trapped in vitreous ice compared to the environment in a crystal lattice. However, it is relevant to note that in the unit cell for many iGluR crystal structures, the receptor is in an asymmetric environment; this asymmetry is often imparted to the receptor, giving rise to structures that deviate from overall twofold symmetry. Thus, although crystal lattice contacts may restrain conformational mobility, they can also impart asymmetry, which may or may not be representative of native conformations.

Structures of isolated ligand-binding and amino-terminal domains

Hundreds of crystal structures have been solved for AMPA, kainate, NMDA, and δ-receptor LBDs expressed as soluble proteins, some at resolutions as high as 1.2 Å (12, 13, 14). Such work represents an unprecedented triumph of neurotransmitter receptor structural biology that continues to yield important new information (15, 16, 17). These structures provide extraordinary insight into the mechanisms underlying binding of subtype selective ligands and the action of allosteric modulators and ions that augment or inhibit iGluR activity. Comparing iGluR LBD structures for agonist and competitive antagonist complexes gave insights into the initial conformational changes that underlie activation, produced by closure of bilobed LBDs in dimer assemblies, and desensitization, produced by LBD dimer dissociation (18, 19). Similar advances for isolated ATD crystal structures revealed the architecture of this domain (20, 21), identified binding sites for NMDA-receptor-selective allosteric modulators (22, 23), and, combined with analytical ultracentrifugation, gave insights into the mechanisms underlying subtype-selective receptor assembly (23, 24, 25, 26). Such studies revealed that the ATDs assemble as dimers with nanomolar to low micromolar dissociation constants, whereas, by contrast, the dimer pairs interact weakly, and thus, ATD tetramer assemblies are much less stable. Similar studies revealed that although the LBDs frequently crystallize as dimers, and occasionally even as tetramer assemblies (27), these interactions are too weak to reliably measure, with dissociation constants in the millimolar range (28), indicating that the LBD layer of the receptor is a highly dynamic assembly, which would fall apart if it were not held in place by linkers connecting it to the ATD and ion channel. As a result, iGluRs are exceptionally mobile proteins that can sample a large conformational space, rendering analysis of their gating much more challenging than initially anticipated.

Competitive antagonists

Competitive antagonists are drugs that bind to the LBDs of iGluRs but do not trigger closure of the LBD clam-shell assembly and subsequent ion-channel gating; their binding is displaced by agonists according to the principles of mass action. Structures for numerous competitive antagonist complexes have been solved for isolated iGluR LBDs, and they reveal that many of the same residues that bind glutamate or glycine also form binding sites for competitive antagonists. In a simple model for iGluR gating, competitive antagonists stabilize the resting state, in which the open-cleft LBD structure has a solvent-exposed crevice between two lobes that is capable of binding a wide range of glutamate and glycine analogs. A more realistic description allows for the fact that the energy landscape of the apo state of the LBD is quite shallow and thus samples a spectrum of low-energy conformations in which the angle between the lobes fluctuates, but with closure to the same extent as occurs when agonists bind energetically unfavorably (29, 30, 31). Thus, the definition of a competitive antagonist can be changed to that of a ligand that binds to the LBD, but does not trigger a sufficiently large conformational change to activate ion-channel gating. Consistent with this, crystal structures for iGluR LBD antagonist complexes reveal a substantial range of domain closure (32, 33), with both hyperextension compared to apo-state crystal structures and modest cleft closure, but never to the same extent as that produced by agonists. In some cases, this yields a delicate balance that can be biased when iGluRs form complexes with auxiliary proteins like transmembrane AMPA receptor regulatory proteins (TARPs), which lower the energy required to activate ion-channel gating, converting the action of competitive antagonists to that of weak agonists (34). Consistent with this, structural studies show that when the LBD closed-cleft conformation is stabilized by engineered disulfide bonds, some competitive antagonists can indeed bind to the closed-cleft active conformation, but presumably with very low probability in native receptor assemblies (35).

Antagonist complexes of full-length iGluRs

Crystal structures of full-length (CTD deleted) AMPA receptors bound to competitive antagonists follow the principles discussed above for the isolated LBDs (Fig. 2 A). Overall, the receptor structure contains a closed ion channel, with intact ATD and LBD dimer assemblies. These antagonist-bound structures closely resemble, but are not identical to, crystal structures of the apo state, possibly due to perturbations imposed by the crystal lattice, or possibly reflecting sampling of one of a range of low-energy conformations for the apo state that differs slightly from that stabilized by individual competitive antagonists (4, 36). The majority of cryo-EM structures for AMPA receptors and full-length kainate receptors bound to competitive antagonists are essentially identical to those for AMPA receptor crystal structures, suggesting that both techniques capture highly populated low-energy states (5, 8, 37).

Figure 2.

Cryo-EM structures of agonist and antagonist bound iGluRs. (A) Side views of LBD dimer assemblies from full-length (CTD deleted) AMPA-receptor GluA2 cryo-EM structures for the glutamate complex (left; PDB: 4UQ6) and the competitive antagonist ZK 200775 complex (right; PDB: 4UQJ). (B) The upper structures show corresponding views for the NMDA-receptor GluN1/GluN2B LBD heterodimer assembly with glycine and glutamate (left; PDB: 5IOU) and the competitive antagonist DCKA/AP5 complex (right; PDB: 5IPS); the lower structures show the GluN1/GluN2 LBD dimer-of-dimers assembly viewed from the top, after rotation by 90°, illustrating dissociation LBD dimers for the DCKA/AP5 complex. In all structures, one subunit in the dimer assembly is colored blue (GluN1 and GluA2) and the second subunit orange (GluN2B and GluA2), with α-helices D and J shown in darker shading; dashed lines connect helices D and J within and between dimer assemblies. To see this figure in color, go online.

For NMDA receptor competitive antagonist complexes, an ensemble of cryo-EM structures, the resolution of which varied from 10–15 Å, revealed unexpected reorganization of the ATD and LBD layers compared to cryo-EM structures of the glutamate- and glycine-bound complexes, and substantial differences from crystal structures of antagonist-bound NMDA-receptor isolated LBD dimer assemblies (10, 38). In these antagonist-bound cryo-EM structures, several of which superficially resemble those for the desensitized state of AMPA and kainate receptors, the ATD dimer pairs separate by up to 140 Å, and the LBDs rotate to adopt quasi-fourfold symmetric structures (Fig. 2 B). Unfortunately, no apo state structures have been reported for full-length NMDA receptors, and thus, we do not know if these conformations are representative of the apo state or are induced by the binding of competitive antagonists. If the apo state of NMDA receptors differs from that of AMPA and kainate receptors, this requires that the activation mechanisms are also substantially different, with the binding of glutamate and glycine to NMDA receptors first stabilizing a compact conformation of the ATD and LBD assemblies that resembles that of AMPA and kainate receptor resting states, then activating ion-channel gating, followed by desensitization. Double electron-electron resonance spectroscopy for detergent-solubilized NMDA receptors yielded results consistent with the cryo-EM structures (10), but it does not address the issue of whether these represent native conformations or are in unknown ways induced by experimental manipulation such as detergent solubilization or cryo-EM imaging conditions, and thus, this remains a major area for future investigation. Functional experiments to measure the affinity of competitive antagonists in NMDA receptors locked in a conformation resembling the resting state of AMPA and kainate receptors with Cys mutant cross-links could shed light on this.

In addition to competitive antagonists that bind to the LBD, NMDA receptors are also inhibited by negative allosteric modulators that bind to the ATD, of which ifenprodil and Ro25-6981 are the most widely used experimental ligands. A series of crystal structures for isolated ATD heterodimers, and lower-resolution cryo-EM and crystal structures for full-length (CTD deleted) NMDA receptors, convincingly establish that these drugs bind in a cleft formed at the dimer interface of the GluN1/GluN2B ATD dimer assemblies to stabilize a conformation in which the GluN2B ATD closes 20° compared to the ATD apo state structure (6, 7, 9, 10, 11, 23). This conformational change induces an increase in separation of the lower lobes of the ATD, which, different from AMPA and kainate receptors, form numerous contacts with the underlying LBD. In these structures, which were solved with agonists bound to the LBD, the ion channel is closed, and thus, the structures likely represent either an inhibitor-modulated closed state, or a desensitized state triggered by agonist and inhibitor binding. Although precise details of the transduction mechanism by which these drugs inhibit NMDA receptor activity remain to be elucidated, these structures mark a major advance in our understanding of the role of the ATD in modulating NMDA receptor gating. Likewise, for allosteric regulators of AMPA receptor activity, we have clues about the location of binding sites (39), and some understanding of the conformations they stabilize, but a real biophysical understanding of precisely how they regulate ion-channel activity is only just beginning to emerge (40).

Auxiliary protein complexes

It is likely that most native iGluRs exist as complexes with auxiliary membrane proteins, of which several structurally diverse families have been identified. Of these, the first identified and best studied are the claudin family of tetraspan membrane proteins, which includes AMPA receptor regulatory proteins (TARPs) of which the γ2 subunit named stargazin is the founding member (41), whereas kainate receptors form complexes with the CUB domain NETO family (42). TARPs, NETO, and multiple families of other structurally unrelated modulatory proteins regulate both trafficking of iGluRs from the endoplasmic reticulum and Golgi apparatus to the plasma membrane and also iGluR gating. Both TARPS and NETO augment iGluR activity, typically slowing deactivation, increasing sensitivity to agonists, and increasing the efficacy of partial agonists (41, 42). In addition, TARPs decrease the biphasic rectification of Ca²⁺-permeable AMPA receptors that results from channel block by cytoplasmic polyamines (43, 44). Conversely, GSG1L, another member of the claudin family, also increases the efficacy of AMPA-receptor partial agonists, but in addition, GSG1L reduces weighted mean single-channel conductance and Ca²⁺ permeability, stabilizes the desensitized state, and increases biphasic rectification mediated by cytoplasmic polyamines (45, 46, 47, 48).

How these diverse effects are produced is in part revealed by cryo-EM structures of AMPA-receptor TARP and GSG1L complexes (37, 49, 50, 51), augmented by functional analysis of mutant and chimeric receptors and by spectroscopic techniques (37, 52, 53, 54). The GluA2 cryo-EM structures for auxiliary protein complexes, which reveal assemblies with either one, two, or four copies of γ2, and one or two copies of GSG1L poised to interact with the LBD, illustrate how the asymmetric arrangement of the LBD assembly likely impacts modulation by claudin family proteins. In addition, the γ2 TARP and GSG1L transmembrane α-helices interdigitate with the M1 and M4 helices of the AMPA receptor TMD, providing a second substrate for modulation of gating, independent of the LBD, although it remains to be established how TARPs augment and GSG1L decreases single-channel conductance (48, 55). These structures reveal that conserved residues in lower lobes of the LBD, which are absent in kainate receptors, interact with an extracellular loop in the γ2 TARP, likely stabilizing the glutamate-bound closed-cleft conformation. Conversely, in GSG1L, extension of the β1–β2 extracellular loop, compared to that in γ2, facilitates interactions with a desensitized conformation of the LBD layer, in which the upper lobes of the dimer pairs separate, accompanied by rotations of the A and C subunits (37). Unexpectedly, for both the apo state and the complex with the competitive antagonist ZK200775, the GluA2 GSG1L complex solubilized in digitonin adopts a second conformation in which the LBD dimer pairs twist, creating a gap between the proximal A and C subunits in the LBD tetramer assembly; this gap contains a detergent molecule (37). At present, it is unknown whether this is a native conformation that could potentially bind endogenous lipids, or drugs, or represents a conformation that occurs only when receptors are solubilized in digitonin.

An abundance of desensitized-state conformations

The end result of conformational changes underlying desensitization is to allow the ion channel to remain closed, even though individual LBDs adopt closed-cleft active conformations, uncoupling binding of glutamate from ion-channel activation. Given that AMPA and kainate receptors desensitize to similar extents and have similar resting-state structures, it might be expected that their desensitized state structures would also be similar, but this is not so. Not only do the desensitized-state structures of AMPA and kainate receptors differ, but also, for AMPA receptors, coassembly with the γ2 TARP or GSG1L yields additional desensitized-state structures (37, 51). For homomeric AMPA receptors, the ATD assembly undergoes a profound reorganization in the desensitized state, with the dimer pairs adopting a large ensemble of conformations with different angles between, and different extents of separation of the ATD dimer assemblies (8, 36, 56). By contrast, kainate receptors do not show any reorganization of the ATD assembly in the desensitized state (3). This striking difference is almost certainly due to differences in the Kd for ATD dimer-tetramer assembly, which is much weaker for AMPA than for kainate receptors (24, 57, 58), coupled with the result that the ATD plays essentially no role in desensitization, which proceeds normally in ATD-deletion constructs (59, 60). Adding to this complexity, recent cryo-EM structures for the desensitized state of GluA2 complexes with the auxiliary proteins GSG1L and stargazin reveal much smaller rearrangements of the ATD layer (37, 51), suggesting that the diverse conformational ensemble observed for isolated AMPA receptors could result from either single-molecule isolation, cryo-EM imaging conditions, or a combination thereof.

Within the LBD layer of full-length AMPA and kainate receptor structures, the dimer assemblies are ruptured in the desensitized state, and here also, there are substantial structural differences between AMPA and kainate receptors and between isolated AMPA receptors and AMPA receptor complexes with TARPs and GSG1L. The LBDs of isolated AMPA receptors adopt multiple conformations (8, 36), whereas kainate receptors adopt a quasi-fourfold symmetric arrangement in which the B and D subunits undergo dramatic 125° rotations compared to the resting state (8). In the GluA2 TARP and GSG1L complexes, LBD dimer assemblies are also disrupted, but the rotation of LBD is more subtle and is strikingly similar to that found in the crystal structure of the isolated LBD for the GluA2 G725C mutant (61).

Crystal structures for the desensitized state of isolated AMPA receptors suggest that it is possible for desensitization to occur when only one of the pair of LBD dimer assemblies is reorganized (36), but it cannot be excluded that a fourfold symmetric arrangement is also possible, as found in low-resolution cryo-EM structures (8); likewise, for kainate receptors, it is plausible that there exist intermediates with a mix of both intact and reorganized LBD dimer assemblies. For NMDA receptors, multiple conformations have been reported for glutamate- and glycine-bound complexes in which the ion channel is closed (6, 7, 9, 10, 11), perhaps corresponding to desensitized states, but these could equally well correspond to activation intermediates.

The long-sought open state

Open-state cryo-EM structures solved at overall resolutions of 4.2–4.9 Å for GluA2 complexes with the γ2 TARP, full agonists, and allosteric modulators that block desensitization, reveal an iris-like expansion of the M3 helix bundle crossing, accompanied by unwinding by one turn of the M3 α-helix in the A and C subunits (47, 51). Expansion of the LBD gating ring in these structures, measured by vectors connecting CA atoms at the N-terminus of α-helix E for the AC- and BD-subunit LBDs (62), reveals a much larger movement of the BD subunits than in any of the prior active-state GluA2 structures in which the ion channel was either closed or not resolved (Fig. 3). Atomic models of the pore loop and ion-channel domain reveal structures sufficiently open to support ion permeation in all-atom MD simulations (51), as also seen in a model for the open state built before these structures were solved, but for which the pore is less open than the experimentally determined structures (40). Surprisingly, the pore loop has nearly fourfold rotational symmetry in one structure (47), whereas in the other structure the M2-pore loop region is twofold symmetric (51), mirroring the two-fold symmetry of the LBD layer in the active state. Comparison of the open-state γ2 TARP complex structure (51) with crystal structures for isolated AMPA receptor active states containing closed ion channels (36, 63) reveals that TARPs might facilitate receptor activation by preventing downward movement of the LBD layer that was also observed in an active-state cryo-EM structure for isolated GluA2 (8), perhaps giving insight into the conformational ensemble that AMPA receptors likely sample after binding glutamate, before they open.

Figure 3.

Vectors between CA atoms at the N-terminus of α-helix E for the AC- and BD-subunit LBDs illustrate how expansion that occurs in the open state is due to much larger movements of the BD subunits compared to prior x-ray and cryo-EM active-state structures of GluA2 agonist complexes stabilized by allosteric modulators (Mod) that block desensitization.

In prior work, multiple structural studies for both AMPA and NMDA receptors were performed using conditions which, based on functional experiments, would have been expected to yield an open state, but which instead revealed a closed channel, with the LBDs in an agonist-bound active conformation. Many factors likely contributed to this, but central is the fact that the open state(s) of iGluRs are likely to be energetically unfavorable, especially for AMPA and kainate receptors, which desensitize within milliseconds after binding of glutamate. In many cases, the structures solved were for partial agonist complexes with proteins that were engineered to increase stability, yielding constructs with impaired ion-channel activation compared to wild-type receptors and full agonists, further reducing the chance of capturing an open-state structure. Another confounding, often-overlooked consideration, is that the maximum open probability of iGluRs rarely exceeds a value of 0.6, and is often much lower; as a result, an equilibrium ensemble of agonist-bound receptors, even under conditions where desensitization is blocked, will contain a high population of closed states. For AMPA and kainate receptors, it is well established that “the open state” is actually an ensemble of subconductance states, and that the probability for entering the highest conductance state is much less than one, thus further expanding the ensemble of conformations that must be resolved to obtain high-resolution open-state structures. Indeed, the structural basis for substates is completely unknown, but it is related somehow to occupancy of the LBDs by agonists (64, 65, 66).

Missing partners and cryo-EM imaging conditions

The trans-synaptic cleft is a crowded structure in which receptors likely form densely packed arrays in the postsynaptic membrane. In addition, iGluRs form molecular contacts with synaptic organizer proteins projecting from the presynaptic membrane (67, 68) and with numerous proteins in the postsynaptic membrane (69). Both types of contact would likely act as conformational restraints, perhaps eliminating some of the states seen in anisotropic network models (70), and in studies of isolated single-receptor assemblies, as was found recently for the desensitized state of GluA2/TARP and GluA2/GSG1L complexes compared to the conformational variability observed for isolated GluA2. A second issue is that in the thin vitreous ice required for collecting high-resolution cryo-EM data, soluble macromolecular complexes are sometimes torn apart into their constituent components due to shear stress imposed by the high surface tension of the thin aqueous layer before vitrification. It seems likely that such forces could strongly impact the conformational ensemble of highly flexible molecules like iGluRs, perhaps pulling them into rare conformations that are energetically less favorable in a native cellular environment. This is the structural biologist's analog of Heisenberg's uncertainty principle, where imaging molecules perturbs their structure. Additional, largely unknown complications arise from the replacement of membrane lipids by detergent molecules, since, based on the observation of tunnels and crevices in TMD structures for closed-state iGluR structures, lipid protein interactions likely play a major role in iGluR ion-channel regulation (71), and lipid molecules may be important but as yet ill-defined structural elements of the ion channel itself. Although there is increasing excitement in the cryo-EM community for using nanodiscs to solve membrane protein structures in a membrane mimetic environment, caution should be applied to the interpretation of these structures, because the narrow lipid belt in nanodisc structures imparts conformational restraints that are absent in a lipid bilayer.

Concluding remarks

Brian Matthews gave a keynote lecture at the 2001 Proteins Gordon Research Conference entitled “Structure isn't everything but it sure helps”. The literature cited here indicates that although we now have a large library of iGluR structures, their interpretation has become unexpectedly challenging, and there are many areas where progress is needed. As revealed by kinetic analysis of single-channel activity, iGluRs cycle through many short-lived intermediate states that are connected by low-energy barriers. As a result, trapping the full ensemble of underlying conformations, and linking the structures to kinetically identified states, is a daunting task that will require multiple approaches pushing current technical boundaries. Although structures alone are unlikely to solve this conundrum, many of the cryo-EM structures for full-length iGluRs do give tantalizing hints about some of the underlying conformations, but due to their relatively low resolution, they lack the detail necessary for building sufficiently accurate atomic models that give information about mechanisms controlling transitions between states. Currently, we have no clue about the structural basis of subconductance states in AMPA and kainate receptors and thus no detailed understanding of how the binding of glutamate gates ion-channel activation, nor do we understand precisely how polyamines permeate and block these channels. Molecular dynamic simulations, performed using a computationally-generated model of the closed state of an NMDA receptor TMD (72), provide important clues about the coordination of Mg²⁺ and Ca²⁺, but further work using an open state structure will be required to understand permeation and block in NMDA receptors. Addressing these issues requires accurate atomic models, which are challenging to build even with cryo-EM electron-density maps at resolutions of 4 Å; the large majority of iGluR structures are of still lower resolution, which precludes precise placement of side chains. It is important to keep in mind that nearly all of the higher-resolution iGluR cryo-EM structures were solved using focused refinement, which hides the fact that in these structures the domains excluded from refinement adopt a range of conformations, as expected for a multidomain protein with highly mobile segments. As noted previously by Zhou and Wollmuth (73), further progress will require the combined application of multiple experimental approaches and very likely will include important contributions from long-duration molecular dynamics calculations and time-resolved spectroscopic approaches, both augmented by higher-resolution structures that ongoing improvements in processing of cryo-EM data and sample preparation are likely to yield, perhaps coupled with tomographic analysis of synaptic membranes with intact receptor assemblies in their native lipid environment embedded in a network of interacting proteins.

Editor: Brian Salzberg.