The ionotropic glutamate receptors (iGluRs) are a family of ligand-gated ion channels known for their involvement in plasticity and signal transmission [1]. This family includes the α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid (AMPA), N-methyl-d-aspartate receptor (NMDA), and kainate receptor, which are known to have isolated glutamate-gated ion-channel activity. The delta receptor, also included in this family, is silent in isolation and is suggested to have glycine-gated activity when in complex with neurexin and cerebellin [1,2]. Among these members, AMPA receptors mediate fast excitatory signaling in the central nervous system by forming cation-specific channels when activated by glutamate. AMPA receptors are made up of four different genes that encode GluA1-4 and further produce more variants through alternative splicing and RNA editing, which express differently throughout the central nervous systems [1,3,4]. AMPA receptors respond quickly to glutamate binding, resulting in fast channel opening (activation) and channel closure in the continued presence of agonist through desensitization (Figure 1) [5,6]. Further functional studies demonstrate that AMPA receptors recover from desensitization in milliseconds [6, 7, 8]. Additionally, AMPA receptors are associated with an intricate network of proteins that allow for fine-tuning of the gating process and provide an additional means of regulation of neuronal signaling [9, 10, 11, 12, 13, 14]. In this review, we will focus on recent structural studies that provide additional insight into receptor activation and desensitization processes and modulation of these by interacting auxiliary proteins.

Initial structural studies on isolated ligand-binding domains of the AMPA receptor [15, 16, 17] and the full-length AMPA receptor in isolation provided insight into the structural changes associated with the fundamental functional process of activation and desensitization [18]. These studies show that the tetrameric receptor is arranged as a dimer of dimers. Each subunit has an extracellular amino-terminal domain and agonist-binding domain, a transmembrane domain, and an intracellular carboxyl-terminal domain (Figure 2). The agonist-binding domain contains the binding pocket for the neurotransmitter glutamate, which is released from the presynaptic terminals. The transmembrane domain is composed of three transmembrane-spanning helices (M1, M3, M4) separated by a reentrant loop (M2) that lines the channel, and the channel pore is formed from the M3 helices and the M2 reentrant loop. The M2 loop contains the Q/R editing site, which is edited in the GluA2 subunit, making the AMPA receptors containing these subunits calcium impermeable [19]. The bi-lobed agonist-binding domain is made up of two lobes, the upper (D1) and lower (D2) lobe, with a cleft in between the two lobes where the agonist binds. The cleft is more open in the resting apo structure and more closed cleft when bound to the agonist glutamate (Figure 1) [15,20,21∗]. This closed-cleft conformation of the agonist-binding domain is thought to drive activation by the pulling of the D2 lobe on the transmembrane segments, leading to the opening of the transmembrane channel (Figure 1) [5,22]. Recent long-time microsecond molecular dynamics (MD) simulations performed using dimeric agonist-binding domains in isolation provide an atomistic-level movie of these conformational changes involved in the activation steps [23∗]. These studies show that the activation process has two steps. First, within 10–100 ns, the agonist glutamate engages with the upper lobe of the agonist-binding domain at residue R485 through interactions at the alpha-carboxylate of the glutamate. This interaction facilitates the subsequent clamshell closure of the agonist-binding domain, which occurs in 0.5–10 μs. The closure of the individual agonist-binding domain leads to the separation of the lower lobes of the domain within the dimer, with the dimer adopting a structure similar to that seen in the full-length activated, open cryogenic electron microscopy (cryo-EM) structure [22]. These computational studies agree with the previous time-resolved vibrational spectroscopic study [24]. The time-resolved vibrational spectroscopic investigations showed a similar sequence of events at the agonist-binding domain; an initial docking at the alpha-carboxylate of the glutamate agonist, followed by the cleft closure induced by agonist bridging the two lobes of the agonist-binding domain. In addition to these computational, spectroscopic, and structural investigations, single-molecule fluorescence resonance energy transfer (smFRET) investigations on the isolated agonist-binding domain [25,26] show that in both apo- and agonist-bound states, the agonist-binding domain exists in a range of cleft closure states and the fraction of receptors in closed-cleft conformation dictates the extent of activation and also higher channel conductance, thus highlighting the dynamic nature of the protein [21].

The mechanism of desensitization has been shown through mutations and structures to be driven by the decoupling of the dimer interface at the agonist-binding domain (Figure 1) [16,22,27, 28, 29]. This was further confirmed with cysteine cross-linking studies, showing that decoupling at the dimer interface is necessary for desensitization [27]. Given that this process occurs in milliseconds, the recent long-time microsecond MD simulations on wild-type protein were not sufficient to see this process in the dimer. To fasten this process and obtain an atomistic view of this process, the simulations were performed using mutations from the kainate receptor, which is known to have faster desensitization kinetics [23∗]. Using this mutant dimer, the MD simulations showed two different conformations, one with a separation/decoupling at the upper lobe at site S741 between the two subunits of 7 Å and a second with a separation of 12 Å. The decoupled state at the upper lobe of the dimeric agonist-binding domain allowed for the shortening of the distance across the lower lobe of the agonist-binding domain. This, in turn, leads to channel closure, as seen in desensitization (Figure 1). The 7-Å separated structure in the MD simulations was similar to that observed in the initial X-ray structures. These two conformations were observed in smFRET investigations of the full-length AMPA receptors, which allows for investigations of the complete conformational landscape for a given distance being probed. The smFRET studies showed a 7-Å more decoupled state relative to the coupled dimer state as the major conformation and a second conformation that was 13-Å more decoupled state relative to the coupled dimer state present to a lesser extent [30]. This contributes to the multiple desensitized conformations that have been seen both structurally and functionally [5,18,31]. However, connecting specific structures to functional states has been largely elusive. Stabilizing the receptor in specific functional states using ligands, mutations, and auxiliary subunits may make this correlation possible.

As discussed earlier, the fundamental structure-dynamics changes have been studied extensively for AMPA receptors in isolation. However, the native AMPA receptor complexes are intricate assemblies composed of AMPA receptors in the complex with various structurally unrelated transmembrane auxiliary subunits, including the transmembrane AMPA-receptor regulatory proteins (TARPs), the germ-cell-specific gene 1–like protein (GSG1L), the cornichon homologs (CNIHs), and the Shisa/cysteine-knot AMPA-receptor modulating protein family, and SynDIG, which all modulate AMPA-receptor-mediated currents (Figure 3) [1,20,22,29,∗32, 33, 34, 35, 36]. Additionally, based on sequence homology and functional differences, TARPs are further divided into two groups: Type I consisting of γ2, γ3, γ4, γ8 and Type II consisting of γ5, γ7 [37].

Most structural and functional studies have focused on GluA2 AMPA receptors with Type I TARPs, γ2 and γ8 [20,38]. These early structural studies [22,28], along with smFRET [30,39] and cross-linking biochemical investigations [40], showed that the extracellular domains of AMPA receptors are more tightly packed in the presence of γ2 and γ8. This suggests that this tight packing could underlie the stabilization of the active state of the receptor and destabilization of the desensitized state of the receptor as observed in the presence of these subunits. Recent work has aimed to bring forth the structures of other AMPA-receptor subunits and auxiliary proteins, particularly focusing on the differences between Type I and Type II TARPs to understand the structural difference underlying their differences in functional modulation. While both types increase ion-channel conductance and attenuate polyamine block of calcium-permeable AMPA receptor (AMPAR)s, the regulation of other functional properties diverges among the different TARPs. The effect on rates of desensitization and deactivation, as well as on the steady-state currents, widely varies for the different TARPs [41]. In addition, Type II TARPs have increased peak current on the edited GluA2(R) receptors with no effect on GluA2(Q) receptors. In contrast, Type I increases peak currents for both [41].

The recent cryo-EM structure of GluA2 in complex with TARP γ5 and the high-resolution structure of TARP γ2 shed light on the structural differences that could underlie these functional differences. The structures of GluA2 in complex with TARP γ5 show 2:1 stoichiometry for GluA2:γ5 in contrast to the 4:1 maximum stoichiometry previously observed in Type I TARPs [34,42]. Additionally, the high-resolution structure of TARP γ2 shows that there is a disulfide bond between C40 and C68 in TARP γ2 but is not observed in Type II TARPs [22,43∗]. This disulfide bond is expected to rigidify the extracellular loop of TARP γ2 in Type-I TARPs and could play a role in Type I TARPs binding to all four subunits, thus leading to a 4:1 ratio at maximal levels. In comparison, Type II TARPs can only interact with two subunits (A and C sites), thus having only two proteins binding to AMPA receptors. The 4:1 versus 2:1 stoichiometry could contribute to differential functional effects between Type I and Type II TARPs, as shown by functional measurements, where the number of Type I TARP binding altered the effect on function with maximal changes observed in the 4:1 stoichiometry [43∗]. However, it is also likely that the additional disulfide bond seen in the extracellular domain of Type I may contribute to differential interactions with the AMPA-receptor subunits, thus contributing to the differential functional effects.

While structurally similar to TARPs, GSG1L lacks the extracellular helix and also lacks the Type I disulfide bond but has a longer TM2 helix [22,31]. As in Type II TARPs, a 2:1 stoichiometry between GSG1L auxiliary subunits and the AMPA receptor is observed (Figure 2). While the stoichiometry of the two complexes is similar, further comparison of the glutamate-bound structures of GluA2-γ5 and GluA2-GSG1L shows that the upper-lobe dimers of the agonist-binding domain undergo a 30-degree rotation relative to each other in GluA2-GSG1L complex in addition to the decoupling. The agonist-binding domain of GluA2-γ5 undergoes decoupling but no rotation and maintains its two-fold symmetry. In addition, there is an overall shortening of the receptor and a rotation of the amino-terminal domain in the GluA2-GSG1L complex. These differences could underlie the functional differences seen between Type II TARPs and GSG1L. While GSG1L slowed desensitization similar to TARPs, the level of steady-state currents was significantly reduced and nearly similar to GluA2 receptors without auxiliary subunits, whereas TARPs had higher levels of steady-state currents [32∗].

On the other hand, CNIHs, transmembrane proteins initially identified as cargo transporters, also modulate the gating of AMPA by slowing desensitization and increasing single-channel conductance [12,13,37,44,45]. Terunaga Nakagawa [33] resolved the interaction between CNIH3 and homomeric GluA2 and demonstrated that CNIH3 interacts similarly to TARPs and GSG1L through interactions at the M1 and M4 of adjacent subunits. However, CNIH3 does not demonstrate a large extracellular domain that could interact with the agonist binding domain of GluA2 as TARPs and GSG1L; therefore, their modulation likely arises from interactions at the pore. It should also be noted that these structures were in the antagonist-bound state. This differs from TARPs as the extracellular regions are thought to be the major contributor to their gating modulation [46]. Additionally, it has been shown that when loops are changed between the auxiliary subunits, the increase in agonist affinity persists and may be preserved through transmembrane segments [29].

TARPs are known for their association with the receptor core early in biogenesis, whereas CNIHs lack a synaptic targeting PDZ-binding site, suggesting a role in AMPA-receptor trafficking toward extrasynaptic pools. Certain et al. [47∗] used blue native polyacrylamide gel electrophoresis and fluorescence imaging to understand the molecular and structural determinants underlying these functional differences. These studies show that CNIH2 plays a crucial role in enhancing the tetramerization of both wild-type and mutant AMPA receptors. However, by making mutations at M4 in the transmembrane domain (TMD) that destabilize the tetramerization of the AMPA receptors, they showed that CNIH2 enhanced the tetramerization through interactions at the transmembrane domain. CNIH2 and CNIH3 exhibited distinct subunit specificity, with CNIH2 facilitating tetramerization of both GluA1 and GluA2, whereas CNIH3 weakly enhanced GluA1 tetramerization. In contrast, TARP γ2, TARP γ-8, and GSG1L exert negligible effects on AMPA-receptor tetramerization. This enhancement of the tetramerization is a proposed mechanism underlying the function of CNIHs as endoplasmic reticulum cargo transporters for AMPA receptors.

The recent high-resolution structure of TARP γ2 also sheds light on differences between this subunit and other members of the claudin family [43∗]. There is a π-stacking interaction between three residues H60, Y32, and W178, which is conserved across TARPs and the TARP-like subunit GSG1L but not in other claudins. This π stacking contributes to the structural rigidity of TARP γ2 by facilitating interactions within the extracellular domain and between the extracellular and transmembrane domains. The π-stacking interaction is thought to play a crucial role in preventing TARP oligomerization, distinguishing them from claudins, and enabling their interaction with AMPARs and other synaptic proteins.

While we have obtained extensive insight into the structural mechanism underlying the AMPA receptor and its modulation in the presence of auxiliary proteins, we do not know much in terms of how interactions of AMPA receptor with trans-synaptic proteins, such as neural pentraxin, would alter the structure and function of these proteins. These transsynaptic proteins are expected to interact with the amino-terminal domain of the receptors and possibly stabilize the domain to alter functional properties [48, 49, 50].