Tuning of AMPA receptor activation by inhibitory auxiliary subunits and epilepsy-associated disease mutations

Thomas P Newton; Maria V Yelshanskaya; Muhammed Aktolun; Shanti Pal Gangwar; Laura Y Yen; Alexey A Alekseev; Ivan A Sobolevsky; Maria G Kurnikova; Alexander I Sobolevsky

doi:10.1038/s41467-026-77416-0

. 2026 Sep 3;17:10472. doi: 10.1038/s41467-026-77416-0

Tuning of AMPA receptor activation by inhibitory auxiliary subunits and epilepsy-associated disease mutations

Thomas P Newton ^1,², Maria V Yelshanskaya ¹, Muhammed Aktolun ^1,³, Shanti Pal Gangwar ¹, Laura Y Yen ^1,⁴, Alexey A Alekseev ¹, Ivan A Sobolevsky ¹, Maria G Kurnikova ³, Alexander I Sobolevsky ^1,^✉

PMCID: PMC13635229 PMID: 42830315

Abstract

AMPA receptors (AMPARs) facilitate excitatory neurotransmission and are implicated in neurological and neuropsychiatric disorders. Binding of the neurotransmitter glutamate triggers AMPAR activation, which is modulated by potentiating and inhibitory auxiliary subunits. While modulation by potentiating auxiliary subunits has been studied extensively, activation involving inhibitory subunits has not been examined. Here, we present structures of the activated GluA2 AMPAR alone and in complex with inhibitory subunits GSG1L or γ5. Our results suggest that gating kinetics strongly depend on auxiliary subunit identity, while pore size and conductance do not. We find that receptor deactivation becomes slower if an auxiliary subunit promotes tightening of the interface between two ligand-binding domain dimers during activation. In addition, we functionally characterize the effects of epilepsy-associated disease mutations on AMPAR activation and desensitization kinetics, displaying strong dependence on the type of bound auxiliary subunit. Modulation by auxiliary subunits therefore has to be considered when designing AMPAR-targeting therapeutics.

Subject terms: Ligand-gated ion channels, Cryoelectron microscopy, Ion channels in the nervous system

AMPA receptors mediate the majority of excitatory neurotransmission. Here, the authors report structural and functional regulation of GluA2 AMPA receptor by inhibitory auxiliary subunits GSG1L and TARP γ5 and by epilepsy-associated disease mutations.

Introduction

The superfamily of ionotropic glutamate receptors (iGluRs) are ligand-gated, cation-selective ion channels within the mammalian central nervous system (CNS). These receptors are subdivided into four major classes, the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs), kainate receptors (KARs), N-methyl-D-aspartate receptors (NMDARs), and delta receptors^1,2. The primary facilitators of fast excitatory neurotransmission, the AMPARs, are distinct from other iGluR subclasses because of their speedy gating kinetics and extensive repertoire of pharmacological and biophysical modulation. The natural agonist glutamate (Glu) is released into the synaptic cleft upon the fusion of Glu-carrying presynaptic vesicles with the presynaptic membrane³. Glu binds to the postsynaptic AMPARs, causing conformational changes associated with its activation and channel opening. Ions rush into the postsynaptic neuron via the open channel, depolarizing the postsynaptic membrane on the timescale of milliseconds⁴. Due to the rapid pace at which AMPARs activate, the corresponding component of the excitatory postsynaptic current increases the likelihood of action potential transmission, and thus the electrical excitation of the postsynaptic cell⁵.

With the pivotal roles of AMPARs in higher-order cognitive functions, such as learning and memory, their dysregulation is subsequently implicated in several neurological, neurodevelopmental, and neuropsychiatric disorders, such as epilepsy, Alzheimer’s disease, schizophrenia, autism spectrum disorder, and chronic pain^6–9. Alleviation of the harmful effects linked to these diseases is of utmost importance, making AMPARs an ardent target for biochemical and mechanistic investigation. Currently, the sole clinically approved therapeutic against AMPAR-related neuropathies is perampanel, an anticonvulsant and non-competitive antagonist capable of treating seizures by blocking Glu-induced activation^10–12. Given the limited availability of effective rectifiers of neurological abnormalities, a deeper mechanistic understanding of receptor function is imperative for guiding the design of pharmacological agents aimed at dysfunctional AMPARs.

AMPARs, ubiquitously expressed throughout the CNS, are localized to developmentally mature postsynaptic densities, rendering them necessary for the determination of synaptic strength and plasticity⁶. AMPARs are able to assemble into homo- or hetero-tetramers composed of subunits GluA1-A4, with the GluA2 subunit being the most widely expressed in the brain^1,13. The core of the receptor complex is conserved across all four subtypes, with a modular four-domain, Y-shaped architecture^1,2,14,15. The amino-terminal domain (ATD), responsible for proper assembly and synaptic localization, is the most extracellular domain, protruding deep into the synaptic cleft. The following domain, the ligand-binding domain (LBD), consisting of polypeptide stretches S1 and S2, adopts a clamshell-like architecture capable of capturing and closing on agonists and is the prime suspect for the initiation of channel gating. The ion channel pore is housed in the transmembrane domain (TMD), which consists of three membrane-spanning helices (M1, M3, and M4), and a re-entrant M2-loop forming the selectivity filter. Finally, the structurally enigmatic carboxy-terminal domain projects into the postsynaptic density, with roles in receptor trafficking and interactions with intracellular scaffolds.

Modulation of AMPAR kinetics, ion permeation, trafficking, and pharmacology is further facilitated by their association with peripheral, transmembrane proteins, or auxiliary subunits^16,17. Among the most abundant auxiliary proteins in the CNS are the transmembrane AMPAR regulatory proteins (TARPs), originally coined as “γ-subunits” due to their high sequence similarity to the voltage-gated calcium channel subunit γ1. TARPs play various roles in the pathophysiology of neurological disorders¹⁸. Of the eight known γ-subunits, six have been classified as AMPAR auxiliary subunits, subdivided into type I (γ2, γ3, γ4, and γ8) and type II (γ5 and γ7) TARPs based on overall modulatory effects on AMPAR function and sequence homology. Generally speaking, type I TARPs demonstrate potentiating effects, including increased surface expression, steady-state current amplitude, and open probability while also displaying slowed rates of deactivation and desensitization^19–23. On the other hand, type II TARPs are deemed inhibitory or suppressive. In particular, TARP-γ5, which has been structurally resolved in AMPAR complexes captured in the non-conducting desensitized and resting states, introduces little changes in the kinetics of desensitization and deactivation, but substantially lowers the channel open probability, steady-state current amplitude, and diminishes the apparent affinity to Glu^11,24,25. The second most abundant class of auxiliary proteins, the cornichon (CNIH) family, including brain-specific CNIH2 and CNIH3, plays activating roles by enhancing AMPAR trafficking, decelerating the kinetics of deactivation and desensitization, and increasing potency for Glu^11,26–28. Lastly, a markedly distinct auxiliary subunit, germline-specific gene 1-like (GSG1L), resembles type I TARPs in its ability to slow down the kinetics of deactivation and desensitization, but serves inhibitory roles by slowing recovery from desensitization and favoring the desensitized state of the receptor^{15,24,29–32}. To date, molecular and structural determinants of AMPAR activation have primarily been investigated utilizing potentiating auxiliary subunits, while the corresponding mechanisms for homomeric AMPARs alone or in complex with inhibitory subunits remain unclear.

Here, we present the activated state of the homomeric GluA2 AMPAR, both in the absence and presence of inhibitory auxiliary subunits GSG1L or TARP-γ5, solved by cryo-electron microscopy (cryo-EM). Coupled with electrophysiological recordings and molecular dynamics (MD) simulations, our structural data reveal the molecular determinants of AMPAR gating regulation by auxiliary subunits. Interestingly, while pore dimensions and unitary conductance appear to be independent of specific auxiliary subunits, activation and desensitization kinetics show strong dependence on the presence of these proteins. We find that auxiliary subunits, which promote tightening of the LBD dimer-dimer interface during activation, slow down receptor deactivation. In addition, epilepsy-associated disease mutations within the gate region of GluA2 produce remarkably distinct effects on AMPAR kinetics depending on the presence of particular auxiliary subunits. Therefore, the specific contributions of different auxiliary subunits to regulation of AMPAR function must be considered in the development of therapeutic interventions targeting AMPARs.

Results

Structure of homomeric GluA2 in the open state

To date, structures of full-length homomeric GluA2 have only been solved in nonconducting states, either apo or in the presence of the antagonist ZK200775 (ZK) or partial agonists^14,22,33. Thus, there are no structures of the naked, homomeric GluA2 core in the open state. Therefore, first, we aimed to capture homomeric GluA2 in the open state, void of any partner auxiliary proteins. Using cryo-EM, we solved the activated state structure of the Ca²⁺-permeable AMPAR composed of modified GluA2 subunits (previously defined as GluA2*)¹⁴. This receptor, with truncations in the linkers between the ATD and LBD layers, as well as glycosylation sites knocked out, has previously been shown to function like wild-type GluA2 and used for structural studies in the presence of competitive antagonists and partial agonists^14,22,33. However, none of the combinations of pharmacological agents has provided an open conducting state of GluA2 alone. Here, we utilized a synthetic biaryl-propylsulfonamide derivative, (R,R)−2b, which acts as a high-potency positive allosteric modulator (PAM) by blocking AMPAR desensitization (Fig. 1a)^22,34. Using (R,R)−2b with high concentrations of Glu, we were able to trap GluA2 in the open state, GluA2_{Glu + RR2b}, without any protein binding partners, and solved the structures of the ATD-LBD and TMD regions at 3.46- and 4.29-Å resolution, respectively (Supplementary Fig. 1 and Supplementary Table 1).

The cryo-EM structure of GluA2_{Glu + RR2b} reveals the classical Y-shaped overall architecture of the receptor (Fig. 1b). The core structure of the receptor strikingly resembles that of the cyclothiazide (CTZ) and Glu bound GluA2-γ2 structure⁴. The ATD layer displays the canonical dimer-of-dimers domain arrangement, followed by a doublet of the domain-swapped back-to-back LBD dimers. Due to their prominent roles in gating, we focused on the LBD-TMD for further analysis. We find that one molecule of PAM (R,R)−2b binds per LBD dimer interface, with Glu molecules bound within all four LBD clamshells (Fig. 1c–f). Similar to CTZ, (R,R)−2b binds at the D1-D1 dimer interface to serve as a stabilizer, preventing the interface rupture that accompanies receptor desensitization^15,34–37. The LBD D2 lobes swing toward the D1 lobes to close around Glu, with a closure angle of 26°. This leads to a 32-Å separation of D2 lobes within each LBD dimer, measured as a distance between Cα’s of S635, compared to an 18-Å separation in the closed state (PDB ID: 3KG2). Meanwhile, the D1-D1 interface remains intact, maintaining an unchanged distance between Cα’s of S741 (~17 Å). The widening of the D2-D2 interface and consequent expansion of the LBD tetramer (Fig. 1g) were previously proposed to trigger pulling forces that are applied to the LBD-TMD linkers, causing the opening of the ion channel^1,2,4,38.

In agreement with the previous findings⁴, AMPAR activation requires the M3 helices of the TMD to undergo conformational changes that are asymmetric between the diagonal subunits B/D and A/C (Fig. 1h, i). While the M3 helices in subunits B/D show kinking at the gating hinge residue A618, the M3 helices in subunits A/C display an unwinding of the α-helix downstream of the SYTANLAAF motif, starting at residue T625. The pore shows a distinct dilation, displayed by the increased cross-pore diameter at the gate region residues T617, A618, A621, and T625. Although supported by weaker density than the rest of the TMD, we were able to model the M2-loop, a segment that was notoriously difficult to resolve in homomeric GluA2 structures without auxiliary subunits. With the Q586 residue at the top, which makes this channel calcium-permeable^39,40, the extended region of the M2-loop, mostly lined by backbone carbonyls of Q587, G588, C589, and D590, creates the second narrowing of the AMPAR channel pore, the selectivity filter, which is critical for ion permeation and channel block^41,42 but much less involved in gating compared to the M3 bundle crossing. Therefore, the structural changes revealed by GluA2_{Glu + RR2b} suggest that we have captured the open state of the GluA2 receptor core in the absence of auxiliary subunits.

Open-state structures of GluA2 in complex with inhibitory subunits GSG1L or γ5

Previously, structures of homomeric GluA2 in complex with GSG1L have been solved in the closed and desensitized states by utilizing a tandem fusion construct with the C-terminus of GluA2 concatenated with the N-terminus of GSG1L^15,24. We used this construct to elucidate the mechanism of AMPAR activation in the presence of the inhibitory subunit GSG1L, Glu and (R,R)−2b. Cryo-EM data processing for GluA2-GSG1L revealed two distinct classes of particles (Supplementary Fig. 2). The anticipated Glu and (R,R)−2b bound activated state, GluA2-GSG1L_{Glu + RR2b}, was represented by ~55% of particles with the LBD-TMD resolved at 3.49 Å (Fig. 2a, b), while the remaining particles yielded a reconstruction in the apo, resting state, GluA2-GSG1L_apo, similar to the previously published structure⁴ but at the improved resolution of 3.35 Å (Supplementary Table 1). Two GSG1L subunits bind per GluA2 tetramer at the B′ and D′ positions, consistent with the previously solved structures of GluA2-GSG1L in the resting and desensitized states^15,24. These positions are characterized by the TMD interfaces between the TM3 and TM4 helices of GSG1L and segments M1 and M2 of AMPAR subunits A or C and M4 segments of AMPAR subunits B or D.

Fig. 2 — a, b 3D cryo-EM reconstruction (a) and molecular model (b) of GluA2-GSG1L_{Glu + RR2b} viewed parallel to membrane (top) or extracellularly down the TMD (bottom), with subunits A and C colored light orange, B and D light blue, GSG1L red brick, Glu light green and (R,R)−2b dark blue. In the model, the protein is shown as ribbons, while Glu and (R,R)−2b as space-filling models. c, d 3D cryo-EM reconstruction (c) and molecular model (d) of GluA2-γ5-CNIH_{Glu + RR2b} viewed parallel to membrane (top) or extracellularly down the TMD (bottom), with subunits A and C colored light orange, B and D light blue, γ5 dark green, CNIH magenta, Glu light green and (R,R)−2b dark blue. In the model, the protein is shown as ribbons, while Glu and (R,R)−2b as space-filling models. **e–h** Molecular models of GSG1L (e), γ5 (f), CNIH (g), and γ2 (PDB ID: 5WEO, h), with the termini and structural elements labeled. i Superposition of GSG1L (red brick), γ5 (dark green), and γ2 (blue).

Next, we aimed to capture the activated state of homomeric GluA2 in complex with TARP-γ5^11,24. Reconstruction of GluA2-γ5 in the presence of Glu and (R,R)−2b yielded the GluA2-γ5-CNIH_{Glu + RR2b} structure at 3.39-Å resolution upon focusing on the gating-relevant LBD and TMD (Fig. 2c, d, Supplementary Figs. 3 and 4, and Supplementary Table 1). Like GluA2-GSG1L, the structure of GluA2-γ5 revealed two molecules of γ5 per GluA2 homotetramer, despite the presence of four tandem fusions. The other two γ5 molecules are likely to be disordered, as otherwise their large and rigid extracellular domains would clash with GluA2 LBDs²⁴. Consistent with the previous study¹¹, GluA2-γ5 expressed in HEK 293 cells in the absence of antagonists was co-purified with endogenous cornichons (CNIHs). Mass spectrometry analysis of the purified protein revealed the presence of CNIH1, CNIH2 and CNIH4 subunits (Supplementary Table 2). Cryo-EM density at the N-terminus of CNIH identified it as CNIH1 or CNIH2, because the corresponding region is missing in CNIH4. Since CNIH1 also showed slightly higher mass spectrometry counts than CNIH2 (Supplementary Table 2), this subunit was used in the structural model. However, given the possible ambiguity in assignment, we further refer to this subunit as “CNIH”. GluA2-γ5-CNIH_{Glu + RR2b} shows the maximal stoichiometry of 1:2:2 of GluA2:γ5:CNIH around the TMD, with γ5 occupying positions B′ and D′ and CNIH positions A′ and C′. Interestingly, CNIH densities are completely missing on the periphery of GSG1L-containing complexes. While puzzling, we suspect that such exclusion happens through some yet unknown mechanism in the endoplasmic reticulum that allows GluA2-GSG1L complexes to be shed of CNIHs before being trafficked to the membrane, while the same mechanism does not work for GluA2-γ5 complexes.

GSG1L and γ5 architectures and interactions with GluA2

In the GluA2-GSG1L_{Glu + RR2b} complex, GSG1L adopts the tetraspanin fold, characterized by four transmembrane helices (TM1–4) and an extracellular domain (ECD) composed of a core five-stranded β-sheet that includes a long β1-β2 hairpin loop with a structurally unresolved flexible tip (Fig. 2e). GSG1L shows extensive contacts between TM3/TM4 and M1 of the GluA2 subunit A/C and M4 of the adjacent GluA2 subunit B/D. Electrostatic interactions occur between the lower D2 lobe of the LBDs in GluA2 subunits B/D and the TM3-β5 loop of GSG1L (Fig. 2b).

Similarly, for GluA2-γ5-CNIH_{Glu + RR2b}, the γ5 fold follows the canonical claudin topology, with four transmembrane helices (TM1-4) and an extracellular domain consisting of a β-sheet with five strands (β1–β5) (Fig. 2f). The receptor-facing helices TM3 and TM4 of γ5 make extensive contacts with M1 of subunits A/C and M4 of the neighboring GluA2 subunits B/D. TM4 makes minor contacts with the M2 re-entrant loop of GluA2 facilitated by hydrophobic interactions. The β1-β2 hairpin loop of γ5 interacts with the lower lobe D2 of GluA2 subunits A/C, while TM3, TM4, and the TM3-β5 loop make extensive hydrophobic and electrostatic interactions with the S1-M1 and S2-M4 linkers of the adjacent GluA2 subunits B or D. CNIH adopts a tetrahelical fold with both N- and C-termini extracellular (Fig. 2g). TM1 and TM2 interact with M1 of GluA2 subunit B/D and M4 and the S2-M4 linker of the neighboring GluA2 subunit A/C, while the peripheral TM3 and TM4 helices do not directly interact with the receptor core.

The CNIH fold is substantially different from GSG1L and γ5, which both contain ECDs. On the other hand, the claudin fold of GSG1L and γ5 is typical for all known TARPs, including the founding member, stargazin or γ2 (Fig. 2h). The biggest difference between the inhibitory auxiliary subunits GSG1L or γ5 and potentiating subunit γ2 is the region preceding the transmembrane helix TM2. In γ2, it forms an extracellular helix (ECH), which kinks towards the membrane, running nearly parallel to the membrane, while the ECH is perpendicular to the membrane in γ5 and is missing in GSG1L (Fig. 2i). Instead, TM2 in GSG1L is longer and its extended N-terminus runs in the same direction as the ECH in γ5.

Interactions between GluA2 LBD and auxiliary subunit ECD

Next, we looked at the interfaces between the ECDs of auxiliary subunits and GluA2 LBDs in the closed versus open states (Fig. 3). The interactions between the ECD of GSG1L and the lower D2 lobe of GluA2 in GluA2-GSG1L_Apo are subtle (Fig. 3c), with K697 of GluA2 jutting outward in the direction of the TM3-β5 loop of GSG1L. This residue is in the middle of the highly conserved KSKGK motif (K695-K699) of the LBD, part of which (named KGK motif) has been previously identified as an important site of AMPAR regulation^21,43,44. In contrast to GluA2-GSG1L_Apo, the K697 residue is pulled further away from the ECD in GluA2-GSG1L_{Glu + RR2b} (Fig. 3d), thus preventing the formation of this stabilizing salt-bridge between the ECD of GSG1L and the D2 lobe of the GluA2 LBD. The loss of this interaction aligns with the ability of GSG1L to drive the receptor to a preferred desensitized state, emphasized by restored interaction of K697 with the GSG1L ECD upon desensitization²⁴. Conversely, the same K697 residue in GluA2-γ5-CNIH_{Glu + RR2b} (Fig. 3b) and GluA2-γ2_{Glu + CTZ} (PDB: 5WEO) (Fig. 3f) maintain a salt-bridge with the TM3-β5 loop of γ5/γ2, stabilizing the LBD-γ5/LBD-γ2 binding interfaces in the open states. Furthermore, the large and rigid ECD of γ5 extends outwards and its β1-β2 loop interacts with K697 in the LBD D2 lobe of the neighboring subunit. Therefore, while the loss of this interaction accompanies the transition of GSG1L-bound GluA2 to the desensitized state, maintaining it in the γ5-bound structure does not cause a similar effect. Nevertheless, the overall inhibitory action of γ5 is likely mediated by all interfaces, which are specific to γ5 and different from γ2. Because of the upright orientation of the ECH in γ5 and horizontal in γ2, all three types of interfaces between GluA2 and γ5 are different from the interfaces between GluA2 and γ2 (Fig. 3).

The precise molecular mechanisms distinguishing the inhibitory behavior of γ5 from the activating behavior of γ2 are likely based on state-dependent changes in these interfaces. These changes are hard to predict as the corresponding interfaces are poorly resolved in the majority of available cryo-EM structures. We used MD to test whether the putative interactions between the auxiliary subunit ECD and AMPAR LBD in GluA2_{Glu + RR2b}, GluA2-GSG1L_{Glu + RR2b}, and GluA2-γ5-CNIH_{Glu + RR2b} were maintained throughout the simulation time (Supplementary Table 3). For all the trajectories, the distances between the centers of mass of the auxiliary subunit ECD and the corresponding LBD D2 lobes remained stable throughout the simulations and consistent with the distances observed in the cryo-EM structures (Supplementary Fig. 5). In some instances, the average distances were even reduced relative to the starting structure, indicating persistent close association of the two domains. Analysis of residue-residue contacts further showed that, although not all individual interactions observed in the cryo-EM structures were maintained continuously, the interface remained populated by a network of hydrophobic and hydrogen-bond interactions throughout the simulations (Supplementary Fig. 6 and Supplementary Table 4). New interactions also formed dynamically, consistent with the conformational flexibility expected at protein-protein interfaces. In addition, representative structures from the MD simulations show exclusion of water molecules from hydrophobic contact regions at the interface, further supporting persistent association between the TARP ECD and the LBD. Nevertheless, it will require additional structural information to deduce the molecular mechanisms underlying distinct functional effects of TARP subunits.

We tested the importance of K697 within the KSKGK motif of GluA2 for functional modulation by GSG1L using whole-cell patch-clamp technique (Fig. 4). The kinetics of deactivation and desensitization, as well as the extent of desensitization for the K697A mutant remained similar to wild-type GluA2-GSG1L (Fig. 4a, c–e). However, the K697A mutation significantly accelerated the recovery from desensitization compared to wild-type, thus diminishing the characteristic slowdown of this recovery caused by GSG1L (Fig. 4b, f). This confirms that K697 in the KSKGK motif plays an integral role in the ability of GSG1L to stabilize the desensitized conformation of GluA2 and to exert its inhibitory action on AMPARs.

Activation of GluA2-GSG1L and GluA2-γ5-CNIH at the LBD level

The receptor cores of GluA2-GSG1L_{Glu + RR2b} and GluA2-γ5-CNIH_{Glu + RR2b} are globally similar to that of GluA2_{Glu + RR2b} and follow a similar mechanism of activation gating. In the ion channel open states, the LBD clamshells are closed around Glu, contrasting the wide-open clamshell geometry in the resting apo state (Fig. 5a). Clamshell closure is coupled to the D2 lobes swinging upward, causing their separation within individual LBD dimers (Fig. 5b), and leading to the dramatic expansion of the entire LBD layer (Fig. 5c). Individual clamshell closures do not change significantly in the absence or presence of different auxiliary subunits (Fig. 5d). Furthermore, because the D1-D1 interface is strongly enforced by the PAM wedged in the middle, the separation of the LBD D1 lobes in all open states remains similar to the closed state (Fig. 5e). With the D1-D1 interfaces intact, the closure of the LBD clamshells causes substantial increases in the D2-D2 distances within individual LBD dimers, roughly to the same extent in the absence and presence of different auxiliary subunits (Fig. 5f). Accordingly, the agonist binding is further translated into expansion of the LBD tetramer, emphasized particularly by an increase in the distance between Cα’s of S635 in subunits B/D (Fig. 5g). Strong separation of the D2 lobes in subunits B/D is translated into a force that pulls on the M3-S2 linkers to open the channel gate. Intriguingly, the extent of the B/D separation is smaller in GluA2-GSG1L_{Glu + RR2b} compared to TARP-bound and naked GluA2 (Fig. 5g).

Fig. 5 — a Superposition of individual LBDs from the GluA2-GSG1L_Apo (blue) and GluA2-GSG1L_{Glu + RR2b} (red) structures, with the glutamate molecule shown in sticks (green) and Cα atoms of S635 and S741 as spheres of the corresponding color. Rotation of the lobe D2 towards D1 in GluA2-GSG1L_{Glu + RR2b} relative to GluA2-GSG1L_Apo is illustrated by red arrow. b Superposition of LBD dimers from GluA2-GSG1L_Apo (blue) and GluA2-GSG1L_{Glu + RR2b}, with the (R,R)−2b molecule at the LBD dimer interface shown in sticks (dark blue). Distances between Cα atoms of S635 are indicated. c Superposition of LBD tetramers from GluA2-GSG1L_Apo (blue) and GluA2-GSG1L_{Glu + RR2b}, with the Cα atoms of S635 connected and black arrows illustrating the separation of the LBD D2 lobes in B/D subunits that generates putative pulling forces applied to the M3-S2 linkers to open the channel gate. **d–g** Angle of LBD clamshell closure, α, (d), D1 (e) and D2 (f) lobe separation within the LBD dimer, and subunits B/D D2 lobe separation (g), measured as the corresponding distances between Cα atoms of S635 for GluA2-GSG1L_Apo, GluA2_{Glu + RR2b-Open}, GluA2-GSG1L_{Glu + RR2b-Open}, GluA2-γ5-CNIH_{Glu + RR2b-Open}, and GluA2-γ2_{Glu + CTZ-Open} (PDB ID: 5WEO). Parameters for the closed and open state structures are represented by blue and red circles, respectively. h, i LBD layer surface in GluA2-GSG1L_Apo (h) and GluA2-GSG1L_{Glu + RR2b} (i), with the surface area of contact interface between two LBD dimers indicated by green ovals. j Surface area of contact interface between two LBD dimers in GluA2-GSG1L_Apo, GluA2_{Glu + RR2b-Open}, GluA2-GSG1L_{Glu + RR2b-Open}, GluA2-γ5-CNIH_{Glu + RR2b-Open}, and GluA2-γ2_{Glu + CTZ-Open}. k Deactivation time constant as a function of the contact interface surface area between two LBD dimers. The straight line through the points is a linear fit with the value of the Pearson’s coefficient indicated. Apart from new data used to calculate the τ_Deact values, previously published data^15,21,24 were used for GluA2-GSG1L, GluA2-γ5, and GluA2-γ2. The data are mean ± SEM. The number of biological replicates, n = 9 for GluA2, n = 20 for GluA2-GSG1L, n = 8 for GluA2-γ5, and n = 6 for GluA2-γ2.

Interestingly, upon receptor activation, the contact surface area between two LBD dimers undergoes variable changes, dependent on the presence of auxiliary subunits (Fig. 5h, i). The naked receptor in the open state GluA2_{Glu + RR2b} has the smallest (105 Å²) interdimer surface area, while GluA2-GSG1L_{Glu + RR2b} has the largest (401 Å²) (Fig. 5j). Tightening of the LBD dimer-dimer interface upon receptor activation appears to stabilize the active state and correspondingly slow down the rate of deactivation. Indeed, the LBD interdimer interface tightening demonstrates strong correlation with the time constant of receptor deactivation, τ_Deact (Fig. 5k)^15,21,24. Given the small number of examples as well as the presence of PAMs in structures used for the interface surface area measurements and lack of such in receptors used for deactivation time measurements, the linear character of this correlation might not be universal and rather requires further testing. Nevertheless, it may represent a general trend of slower deactivation for AMPAR complexes with larger LBD interdimer interfaces. Overall, we can conclude that, although the mechanism of activation at the level of individual LBDs and LBD dimers appears to be similar across all complexes, different extents of LBD tetramer compression may underlie variations in gating kinetics observed in the presence of different auxiliary subunits.

Opening of the ion channel pore

LBD clamshell closure around the agonist causes separation of D2 lobes in subunits B and D (Fig. 5c, g), which generates putative pulling forces applied to the LBD-TMD linkers to open the ion channel pore (Fig. 6a, b). Structures of GluA2 bound to γ2 in the open state suggested that the pore-lining residues in the gate region dictate permeation of ions through the pore and represent markers of receptor activation^4,40. The same pore architecture was observed in the open state of the receptor alone (Fig. 1h, i) and in its complexes with the inhibitory auxiliary subunits GSG1L or γ5. In turn, comparison of GluA2-GSG1L_Apo and GluA2-GSG1L_{Glu + RR2b} illustrates the canonical structural changes in the gate region upon channel opening (Fig. 6a, b). These changes are relatively minor in subunits A and C, largely representing a single α-helical turn unwinding of M3, but much more substantial in subunits B/D, where the M3 helices kink at the gating-hinge residue A618. Subunits B and D therefore play the major role in AMPAR gating, exemplified by dramatic separation of their S635 residues from 55 Å in the closed state to 63 Å in the open state. Kinking of the M3 helices allows opening of the pore for conductance of ions and water. This widening of the ion channel pore was observed in all open-state structures, independent of the presence of auxiliary subunits (Fig. 6c). To infer whether the small differences in the pore diameter predicted by cryo-EM structures have functional implications, we measured ion channel conductance using non-stationary noise analysis of AMPAR-mediated currents for the receptor alone and for the complexes with inhibitory (GSG1L or γ5) or potentiating (γ2) auxiliary subunits.

Fig. 6 — a, b Superposition of the subunits A/C (a) and B/D (b) pore-lining segments in GluA2-GSG1L_Apo (blue) and GluA2-GSG1L_{Glu + RR2b} (red), with residues lining the pore shown as sticks. Distances between Cα atoms of S635 and the gating hinge where M3 segments are kinking during channel opening are indicated. Black arrows illustrate putative pulling forces generated by separation of the LBD D2 lobes in B/D subunits and applied to the M3-S2 linkers to open the channel gate. c Pore radius for GluA2-GSG1L_Apo, GluA2_{Glu + RR2b}, GluA2-GSG1L_{Glu + RR2b}, GluA2-γ5-CNIH_{Glu + RR2b}, and GluA2-γ2_{Glu + CTZ} (PDB ID: 5WEO) calculated using HOLE. The vertical dashed line denotes the radius of a water molecule (1.4 Å). d, e Exemplar current-variance plots for whole-cell recordings from HEK 293 cells transfected with GluA2-γ5 (d) and GluA2-GSG1L (e) in response to 2-ms applications of 3 mM Glu. The parabolic fit gives estimates of single-channel conductance (red). f Average values of the single-channel conductance obtained from different cells using fitting illustrated in (d, e). The data are mean ± SEM. The number of biological replicates, n = 7 for GluA2, n = 14 for GluA2-γ5, n = 6 for GluA2-GSG1L, and n = 9 for GluA2-γ2. Source data are provided. g Integral occupancy of channel pores by water molecules (violet continuum) for MD simulations of GluA2_{Glu + RR2b}, GluA2-γ5-CNIH_{Glu + RR2b}, GluA2-GSG1L_{Glu + RR2b} and GluA2-γ2_{Glu + CTZ}. Pore-forming segments M2 and M3 for GluA2 subunits B and D are shown as ribbons, with residues lining the pore shown as sticks. Continuous water occupancy is observed in all structures. h Average water conductance estimated from MD simulations of GluA2_{Glu + RR2b}, GluA2-γ5-CNIH_{Glu + RR2b}, GluA2-GSG1L_{Glu + RR2b} and GluA2-γ2_{Glu + CTZ} with error bars represent standard deviation of water permeation counts within each simulation (8.5 ± 0.9 for GluA2_{Glu + RR2b}, 9.5 ± 1.0 for GluA2-γ5-CNIH_{Glu + RR2b}, 10.3 ± 0.7 for GluA2-GSG1L_{Glu + RR2b} and 10.7 ± 1.1 for GluA2-γ2_{Glu + CTZ}).

Noise analysis

Non-stationary noise analysis (NSNA) was performed on deactivation phases of currents recorded using whole-cell patch-clamp technique from HEK 293 cells transfected with GluA2, GluA2-GSG1L, GluA2-γ5, and GluA2-γ2 (Fig. 6d, e)⁴⁵. Surprisingly, regardless of the presence of auxiliary subunits, no significant differences were observed between the values of experimentally derived mean unitary conductance (γ) (Fig. 6f), indicating that the subtle differences in the pore diameter observed in cryo-EM structures (Fig. 6c) have little functional consequences and rather represent open states with comparable characteristics of ion channel permeation and conductance. The values of the mean unitary conductance derived from our NSNA (12–17 pS) are somewhat lower than the average values reported from NSNA before^25,46,47 or the values of high conductance states derived from single-channel recordings²⁴, likely because our analysis predominantly characterizes closing transitions from the lowest (O1 and O2) subconductance states. This suggests that our cryo-EM structures are likely to represent the lowest conductance states as well. Accordingly, less prevalent higher conducting states that are not represented in our cryo-EM structures but can be visualized by single-channel recordings can be the reason for the higher average unitary conductance values reported by NSNA before. This also explains why an increase in the channel conductance in the presence of γ5 reported before²⁴ is not so dramatic in our experiments (Fig. 6f). Indeed, while the conductances of individual subconductance levels for GluA2-γ5 are higher than for GluA2, the lower sub-conductance levels might prevail, thus lowering their mean NSNA conductance. To infer whether dynamic properties of our cryo-EM structures are in concert with functional predictions, we subjected our structures to MD simulations and calculated conductance of water through the ion channel pore.

MD simulations of water permeation

We ran MD simulations of GluA2_{Glu + RR2b}, GluA2-GSG1L_{Glu + RR2b} and GluA2-γ5-CNIH_{Glu + RR2b}, and compared their water permeation to that of GluA2-γ2_{Glu + CTZ} (Fig. 6g, h, Supplementary Fig. 7, and Supplementary Table 3)^4,40. Water occupancy provides a useful measure of pore hydration and maintenance of conducting channel architecture because a continuous water-filled pore is a prerequisite for ion conduction. Direct quantification of ion conductance requires sufficient ion permeation events for statistical analysis, which are often not observed in unbiased simulations on accessible timescales. Moreover, ion permeation is governed by additional energetic and electrostatic factors, meaning that water permeation rates cannot be directly translated into permeability or conductance of ions. Therefore, we use water occupancy and permeation analyses to assess pore hydration and solvent accessibility, while recognizing that hydration alone does not establish ionic permeability.

As a control, we ran MD simulations of GluA2 in the apo state, which displays a non-hydrated pore incapable of passing cations through the channel (Supplementary Fig. 8). In contrast, the pores of GluA2_{Glu + RR2b}, GluA2-GSG1L_{Glu + RR2b}, and GluA2-γ5-CNIH_{Glu + RR2b} remained fully hydrated, confirming that water could freely access and traverse the channel, consistent with these structures representing open states. All the MD simulations with or without auxiliary subunits showed continuous permeation of water molecules through the pore, with average cumulative water permeation rates of 8.5 ns⁻¹ for GluA2, 10.3 ns⁻¹ for GluA2-GSG1L, 9.5 ns⁻¹ for GluA2-γ5-CNIH and 10.7 ns⁻¹ for GluA2-γ2. These values were in between the water permeation rates predicted by our previous MD simulations for GluA2-γ2 subconductance states O1 (6.136 ns⁻¹) and O2 (13.406 ns⁻¹)⁴⁰, supporting the conclusion of the NSNA that our cryo-EM structures are likely to represent the lowest conductance states. Importantly, consistent with the independence of the NSNA-derived mean unitary conductance on the presence and type of auxiliary subunits (Fig. 6d–f), there was no significant difference between the water permeation rates in the simulated systems, supporting the idea that in the lower conductance states, auxiliary subunits do not produce substantial disruption of the pore conformation and support water permeation like in the naked receptor.

Epilepsy-associated disease mutations V626L and T625N in GluA2

We further aimed to characterize the role of the conventionally suppressive auxiliary proteins GSG1L and γ5, as well as potentiating γ2, in modulation of AMPAR function by clinically relevant disease-causing mutations. Several mutations within the M3 gate region, specifically of residues just downstream of the most highly conserved in iGluRs SYTANLAAF motif⁴⁸, have been implicated in epileptic disease phenotypes, including developmental epileptic encephalopathy, and reported to cause multiple types of seizures in young patients⁴⁹. We selected two such mutations in GluA2, V626L and T625N (Fig. 7a, b). Using whole-cell patch-clamp recordings, we analyzed the effects of these mutations on receptor function in the absence and presence of potentiating (γ2) or inhibitory (GSG1L or γ5) auxiliary subunits. Since these recordings were made from the same HEK 293 cells as were used for protein purification and structure determination, wild-type and mutant GluA2-γ5 complexes are likely to include CNIH subunits.

Fig. 7 — a, b Extracellular view of the channel gate region in GluA2-GSG1L_Apo (a) and GluA2-GSG1L_{Glu + RR2b} (b). Residues T625 and V626, which are mutated to asparagine and leucine in epilepsy-associated mutants, are shown as space-filling models, while the surrounding residues are shown as sticks. c Whole-cell patch-clamp currents recorded at −60 mV membrane potential from HEK 293 cells transfected with GluA2, GluA2-γ5, GluA2-GSG1L, and GluA2-γ2 carrying mutations V626L (first row) and T625N (second row) in response to applications of 3 mM Glu alone (black and colored traces) or Glu in the presence of 30 μM CTZ (grey traces). d Steady-state fraction of non-desensitized receptors for different AMPAR complexes without (Control) and with epilepsy-associated mutations (V626L and T625N) measured as a ratio of the steady state and maximal current amplitudes. The asterisks indicate a statistically significant difference (two-sided two-sample t-Test, the significance is assumed if P < 0.05). The number of biological replicates (n) and P for comparison between the receptor alone and receptor co-expressed with auxiliary subunits γ5, GSG1L, and γ2 are indicated. The data are mean ± SEM, source data are provided. e Superpositions of currents recorded at −60 mV membrane potential from HEK 293 cells transfected with GluA2, GluA2-γ5, GluA2-GSG1L, and GluA2-γ2 carrying mutations V626L (left) and T625N (right) in response to 2-ms applications of 3 mM Glu. The current decline fitted with one or two exponentials yielded the weighted time constants of deactivation (τ_Deact). f Deactivation time constants for different AMPAR complexes without (Control) and with epilepsy-associated mutations (V626L and T625N). The asterisks indicate a statistically significant difference (two-sided two-sample t-Test, the significance is assumed if P < 0.05). The number of biological replicates (n) and P for comparison between the receptor alone and receptor co-expressed with auxiliary subunits γ5, GSG1L, and γ2 are indicated. The data are mean ± SEM, source data are provided.

First, by performing long, 1-s application of Glu in the absence or presence of PAM CTZ, we tested the effect of auxiliary subunits on the fraction of non-desensitized receptors, measured as the ratio of the steady-state and maximal currents (I_SS/I_Max, Fig. 7c, d). Remarkably, the V626L mutation increased the fraction of non-desensitized receptors in the receptor alone by 2.6 times. Interestingly, γ5 caused similar but weaker (2 times) increase in I_SS/I_Max, while GSG1L had no significant effect (0.032 versus 0.027). The potentiating auxiliary subunit γ2, which dramatically increased (11 times) the steady state current in WT channels, caused a comparable (12 times) increase in the V626L mutant channels. However, the two increases, caused by the presence of γ2 and by the V626L mutation, together yielded a dramatic elevation of I_SS/I_Max, signifying only weak (39%) desensitization of GluA2-V626L-γ2 receptors (Fig. 7c, upper right traces). The T625N mutation had a much stronger effect on the steady-state current in the receptor alone, where it eliminated desensitization nearly completely (I_SS/I_Max = 0.89 ± 0.03, n = 7; Fig. 7c, lower left traces). Here, the auxiliary subunits displayed a different pattern of effects. While TARPs (γ2 and γ5) did not produce significant changes in the desensitization-eliminating effect of T625N, GSG1L caused its slight attenuation (Fig. 7d).

In the presence of epilepsy-associated mutations, the auxiliary subunits also showed distinct effects on the rate of receptor deactivation, which is characterized by the current decline after a short, 2-ms application of Glu (Fig. 7e). In WT channels, GSG1L and γ2 increased the time constant of deactivation (τ_Deact) by 2.4 and 2.1 times, respectively, whereas γ5 did not cause significant alterations (Fig. 7f). In the V626L background, while GSG1L and γ2 produced comparable effects by increasing τ_Deact by 1.9 and 2.4 times, respectively, γ5 caused significant acceleration of deactivation by reducing τ_Deact 1.6 times. The T625N mutation caused a much more dramatic effect on deactivation kinetics compared to V626L by making deactivation much slower both in the absence (17 times) and presence (12 times for γ5 and GSG1L and 15 times for γ2) of auxiliary subunits. The relative effect of auxiliary subunits on the time constant of deactivation in the T625N background was comparable to the WT background, with GSG1L and γ2 increasing τ_Deact by 1.7 and 1.9 times, respectively, and γ5 not causing significant alterations (Fig. 7f).

Auxiliary subunits tune the effects of V626L on receptor desensitization kinetics

Because the T625N mutation eliminates desensitization nearly completely in the absence or presence of auxiliary subunits (Fig. 7c, d), kinetic analysis of receptor desensitization in its background is practically impossible. In contrast, V626L leaves a substantial fraction of receptors to desensitize, allowing us to estimate the effects of different auxiliary subunits on both receptor entry and recovery from desensitization (Fig. 8). The rate of desensitization was characterized by the current decline in the continuous presence of Glu (Fig. 8a). In the WT receptors, the time constant of desensitization (τ_Des) was increased in the presence of all three auxiliary subunits (Fig. 8b), γ5 (1.6 times), GSG1L (1.6 times) and γ2 (2.6 times). In contrast, the V626L mutation made γ5 and GSG1L ineffective, while substantially slowing down desensitization in the presence of γ2 (12-times increase in τ_Des for GluA2-γ2 compared to the naked receptor).

We also studied the impact of auxiliary subunits on the rate of recovery from desensitization estimated using the two-pulse protocol (Fig. 8c). In the WT background, both inhibitory auxiliary subunits slowed down the rate of recovery from desensitization by increasing the time constant of recovery from desensitization (τ_RecDes) by 2.6 (γ5) or 16.5 (GSG1L) times (Fig. 8d). In contrast, the type I TARP γ2 had no effect on τ_RecDes in WT channels. While the recovery from desensitization in the V626L mutant channels remained unaffected by γ2, the effect of inhibitory subunits changed dramatically. Indeed, the τ_RecDes value was 5.4 times reduced in the presence of γ5 but only 8 times increased in the presence of GSG1L (Fig. 8d). Thus, the effects of the epilepsy-associated mutation V626L on the rates of entry and recovery from desensitization appear to be strongly dependent on the presence of particular auxiliary subunits.

Discussion

The mechanisms underlying activation of AMPARs, the main facilitators of excitatory neurotransmission in the mammalian CNS, and their modulation by inhibitory subtypes of auxiliary subunits serve as attractive targets for therapeutic intervention in various neurological diseases. Here, we present the open state structures of the naked GluA2 core AMPAR and in the presence of inhibitory auxiliary subunits GSG1L or TARP γ5. While the ion-conducting states of GluA2 alone (GluA2_{Glu + RR2b}) and in complex with suppressive binding partners (GluA2-GSG1L_{Glu + RR2b} and GluA2-γ5-CNIH_{Glu + RR2b}) closely resemble one another globally, subtleties in the domain architecture represent the structural determinants behind kinetic divergences. The means by which the channel opens remains conserved independent of auxiliary subunit presence. Closure of LBD clamshells around agonist is coupled to the separation of the D2 lobes in LBD dimers, followed by mechanical tension on LBD-TMD linkers, opening the channel pore. However, we note the importance of interactions, or lack thereof, between the lower D2 lobes of GluA2 with the ECDs of auxiliary subunits as defining features of activated state stabilization. Particularly, we show that the desensitization-favoring GSG1L loses integral binding interfaces, specifically the one contributed by K697 in the KSKGK motif of the D2 lobe, during activation, driving the complex towards a non-conducting state to regenerate such interaction. This props the interface between the LBD D2 lobe and the auxiliary subunit ECD as a site of modulation and a potential region to be targeted by small molecule drugs.

The architecture of the LBD tetramer can further be attributed to dictating the kinetics of deactivation. Specifically, we show that the degree of LBD tetramer compression, quantified by the contact surface area between the two LBD dimers, correlates with the rate of deactivation. This suggests that a larger compression of the LBD upon activation, mediated by the extracellular domains of auxiliary proteins, leads to the requirement of a larger energetic threshold to close the channel, slowing the rate of deactivation. Strikingly, this behavior directly contrasts the recently defined phenomenon in the closely related GluK2 KARs in complex with auxiliary subunit Neto2, where the presence of Neto2 prevents tightening of the LBD dimer-dimer interface during activation, differentiating the mechanisms by which AMPARs and KARs are structurally altered by the presence of auxiliary subunits in the activated state⁵⁰. In general, the interactions between two LBD dimers in iGluRs appear to cause different, sometimes opposite, effects on gating. Thus, restraining the interface between two LBD dimers with engineered cysteine crosslinks in AMPARs or KARs appears to have inhibitory^34,51,52, whereas in NMDARs activating⁵³, effects on function. Interestingly, differences in the LBD layer of AMPARs caused by the presence of auxiliary subunits did not uncover any dramatic changes in the ion channel, with the pore regions showing just minor differences. Consequently, despite the minor variations in the pore geometry, our nonstationary noise analysis and MD simulations showed that the corresponding conformational changes do not lead to functional alterations in the pore conductance or permeation.

To further analyze the effects that potentiating or inhibitory subunits have on receptor activation, we functionally characterized the impact of two epilepsy-associated disease mutations attributed to the gate region of GluA2, V626L and T625N, just downstream of the iGluR-conserved SYTANLAAF motif⁴⁹. The introduction of the T625N mutation resulted in a non-desensitizing phenotype irrespective of auxiliary subunit presence or identity, characteristic of epileptic manifestations rooted in channel overactivity. Remarkably, we show that the V626L disease mutant displays deviations based on the presence of specific auxiliary subunits, noting that the potentiating γ2 conveys an extreme, minimally desensitizing complex as opposed to the unchanged desensitization kinetics of the mutant receptors with or without GSG1L or γ5. Conversely, the inhibitory auxiliary subunits GSG1L and γ5 attenuate the slowdown of recovery from desensitization kinetics caused by V626L, while γ2 has no influence. Altogether, we show that clinical intervention for the mediation of AMPAR-related neuropathies requires consideration of not only the AMPAR core subunit makeup, but also the types of auxiliary subunits composing the periphery and conformational specificity.

While we observe the dramatic consequences of the presence of auxiliary subunits for the function of homomeric GluA2-containing AMPARs, we recognize a key limitation of this study being the lack of information about heteromeric AMPAR assemblies. Specifically, within the natural brain environment, AMPARs are predominantly composed of different subtypes, such as the most highly abundant heteromeric GluA1/A2 and GluA1/A2/A3 complexes^54,55. Details on the mechanisms of activation for these di- and tri-heteromeric AMPARs in the presence of a broad range of potentiating and inhibitory subunits will provide insights into the most physiologically relevant AMPAR complexes. Similarly, the use of desensitization-blocking (R,R)−2b diverges from natural AMPAR complex conditions, therefore inviting future studies of AMPAR activation in the absence of PAMs. In addition, there remains a knowledge gap in understanding activation of other receptors in the absence of auxiliary subunits, both for other homomeric AMPARs and various heteromers. Further studies on the structural mechanisms of activation of other AMPAR core complexes will be instrumental for in depth understanding of auxiliary subunit modulation.

Methods

Plasmids and strains

The construct for GluA2 was the previously established GluA2*¹⁴. The construct for GluA2-GSG1L was previously made by fusing GluA2* to mouse GSG1L truncated after F237 via a GTG linker¹⁵. The previously reported construct for GluA2-γ5 was also made by fusing GluA2* to mouse TARP-γ5 truncated after E207 via a GTG linker²⁴. All constructs had a C-terminal TGG-linked thrombin cleavage site (LVPRGS), followed by an enhanced green fluorescent protein (eGFP) and a Strep-II affinity tag (WSHPQFEK). Constructs were introduced into pEG BacMam vector for baculovirus transduced expression in HEK 293S GnTI^– cells⁵⁶. Disease mutants (T625N and V626L) for electrophysiological recordings were introduced into the corresponding fusion constructs in pIRES vector for transient expression in HEK 293 T cells using Gibson assembly cloning³³.

Protein expression and purification

Purification of recombinant proteins followed previously established protocols^11,56. Bacmid and baculovirus for GluA2*, GluA2-GSG1L, and GluA2-γ5 were made using standard methods⁵⁶. P2 viruses produced in Sf9 cells (GIBCO, 12659017) were added to HEK 293S GnTI^– cells (ATCC, CRL-3022) grown in Free style media supplemented with 2% fetal bovine serum and incubated at 37 °C with 7% CO₂. At 12–15 h post-transduction, 10 mM sodium butyrate was added to the cells, and cells were moved to 30 °C at 5% CO₂. Cultured cells were harvested 64–72 h post-transduction through low-speed centrifugation (5500 × g, 10 min), followed by a wash with 1× PBS pH 7.4 and another centrifugation (5500 × g, 15 min). The cell pellet was resuspended in 50 mL ice-cold lysis buffer (150 mM NaCl, 20 mM Tris-HCl pH 8.0, 0.8 μM aprotinin, 2 μg/mL leupeptin, 1 mM phenylmethylsulfonyl fluoride (PMSF) and 2 μM pepstatin A). Buffer was supplemented with 1 mM β-mercaptoethanol (BME) for GluA2-γ5. Cells were lysed by 3 cycles of sonication (3 min total, 15 s ON, 15 s OFF, AMP 8) under constant stirring on ice. The cell lysate was centrifuged (5500 × g, 15 min) to remove debris and unlysed cells. The resulting supernatant was ultracentrifuged (186,000 × g, 40 min) to isolate cell membranes. The cell membrane pellets were manually homogenized and solubilized for 2–3 h at 4 °C in solubilization buffer (150 mM NaCl, 20 mM Tris-HCl pH 8.0, 1% digitonin, 0.8 μM aprotinin, 2 μg/mL leupeptin, 1 mM PMSF and 2 μM pepstatin A). The solubilization buffer for GluA2-γ5 consisted of 150 mM NaCl, 20 mM Tris-HCl pH 8.0, 1 mM BME, 1% N-Dodecyl-B-D-Maltoside, 0.1% cholesterol hemisuccinate (CHS), 0.8 μM aprotinin, 2 μg/mL leupeptin, 1 mM PMSF and 2 μM pepstatin A. Insoluble material was then removed by ultracentrifugation (186,000 × g, 1 h). The supernatant was combined with pre-equilibrated streptavidin-linked resin (2–3 mL) and incubated at 4 °C for 12–16 h with gentle rotation. The protein-bound resin was washed with 25–35 mL wash buffer (150 mM NaCl, 20 mM Tris-HCl pH 8.0, 0.05% digitonin for GluA2* and GluA2-GSG1L; 150 mM NaCl, 20 mM Tris-HCl pH 8.0, 1 mM BME, 0.05% digitonin, 0.005% CHS for GluA2-γ5) and protein was eluted using 12–15 mL of corresponding wash buffer supplemented with 2.5 mM d-desthiobiotin. The eluted protein was concentrated and subjected to thrombin digestion (1:200 w/w) at 22 °C for 90 min to remove eGFP and the streptavidin affinity tag. The digest reaction was centrifuged at 186,000 × g for 10 min and then injected into a Superose 6 10/300 GL size-exclusion chromatography column (GE Healthcare) pre-equilibrated with buffer containing 150 mM NaCl, 20 mM Tris-HCl pH 8.0, and 0.05% digitonin (with added 1 mM BME and 0.005% CHS for GluA2-γ5). Peak fractions corresponding to tetrameric receptor were pooled, concentrated to 5–6 mg/mL, and used for cryo-EM sample preparation. All procedures, unless otherwise specified, were carried out at 4 °C.

Cryo-EM sample preparation and data collection

UltrAuFoil R 1.2/1.3, 300-mesh gold grids (EMS) were used for sample preparation. Before application of the sample, the grids were glow-discharged using a PELCO easyGlow cleaning system (Ted Pella, 25 s, 15 mA) to render the surfaces hydrophilic. Purified protein was supplemented with 250 μM (R,R)−2b on ice for 30 min before plunge freezing. Before plunge freezing, the sample was spiked with 5 mM Glu and then immediately put onto the grid, for a total application time of under 1 min. Grids were made using Vitrobot Mark IV (Thermo Fisher Scientific) set at 4 °C and 100% humidity with a 2 s blot time, blot force of 10, and a wait time of 30 s. Grids were imaged on a 300-kV Titan Krios transmission electron microscope (Thermo Fisher Scientific) equipped with a post-column GIF Quantum energy filter with a slit width set at 15–20 eV and Gatan K3 (Gatan) direct electron detection camera. We collected 9058 movies for GluA2_{Glu + RR2b}, 16,745 movies for GluA2-GSG1L_{Glu + RR2b}/GluA2-GSG1L_Apo, and 12,358 movies for GluA2-γ5-CNIH_{Glu + RR2b}.

Cryo-EM image processing

Data was processed using cryoSPARC v4.4. For GluA2, movies were aligned using patch motion correction algorithm. Contrast transfer function (CTF) estimation was performed using patch CTF estimation. A subset of 1000 micrographs were used for initial blob picking, binned particle extraction, cleanup by 2D classification and heterogeneous refinement, and then to create templates for template-based particle picking. A stack of 2,253,013 particles was extracted at a 440-pixel box size, binned to 110-pixel box size, yielding 1,196,558 particles further cleaned up by heterogeneous refinement. A stack of 245,055 particles was extracted at an unbinned 440-pixel box size, yielding a consensus reconstruction at 3.61-Å resolution. Local refinements using a mask of the TMD and ATD-LBD provided reconstructions at 4.20- and 3.39-Å resolution, respectively.

For GluA2-GSG1L, movies were aligned using patch motion correction algorithm. CTF estimation was performed using patch CTF estimation. A subset of 1000 micrographs was used for blob picking and extraction, followed by multiple rounds of 2D classification, ab-initio reconstructions, and heterogeneous refinement to create templates for template-based particle picking. A stack of 1,299,343 particles was extracted with a 416-pixel box size binned to a 104-pixel box size. 2D classification yielded a stack of 770,061 particles that were then further cleaned up by iterative ab-initio reconstruction and heterogeneous refinement to remove “junk” particles. The best set yielded 337,707 particles that were then extracted at the unbinned 448-pixel box size, producing a non-uniform refinement at 3.97-Å resolution. Particles were further cleaned up via heterogenous refinement and classified into two subsets of particles using a mask around the LBD-TMD, showing two distinct conformations of the complex. Signal subtraction was used to remove the ATD, followed by 3D Variability Analysis, giving one class of 163,358 particles corresponding to GluA2-GSG1L_{Glu + RR2b} and another with 129,053 particles corresponding to GluA2-GSG1L_apo. Local refinements using an LBD-TMD mask produced maps at 3.49-Å and 3.35-Å for GluA2-GSG1L_{Glu + RR2b} and GluA2-GSG1L_apo, respectively.

For GluA2-γ5-CNIH_{Glu + RR2b}, processing steps were similar to that used for GluA2-GSG1L. In short, 2,712,779 particles were picked from template-based picking and extracted at 416-pixel box size down sampled to 104-pixel box size. After several rounds of 2D classification and heterogenous refinement, 620,192 particles were extracted at the unbinned box size. This subset of particles was subjected to heterogenous refinement as an extra step of particle cleanup, leading to a full-length reconstruction at 3.82-Å resolution with 472,357 particles. Further cleanup by heterogenous refinement gave a clean set of 209,640 particles, yielding a non-uniform refinement at 3.56-Å. A local refinement using an LBD-TMD mask gave a map for GluA2-γ5-CNIH_{Glu + RR2b} at 3.39-Å resolution.

Mass spectrometry sample preparation and protein digestion

Purified proteins were resuspended in 100 mM Tris-HCl (pH 8.0), and concentrations were determined using the Pierce BCA Protein Assay Kit. A total of 5 µg protein per sample was reduced with 10 mM DTT at 56 °C for 30 min and alkylated with 10 mM iodoacetamide at room temperature in the dark for 30 min, followed by quenching with 5 mM DTT for 15 min. Protein cleanup was performed using SP3 magnetic beads⁵⁷, and samples were digested overnight at 37 °C with Lys-C/Trypsin (Promega) at a 1:100 enzyme-to-protein ratio. Digestion was stopped by addition of TFA to 1% (v/v). Peptides were clarified by centrifugation (14,000 × g, 10 min), desalted using SDB-RPS StageTips⁵⁸, dried by SpeedVac, and reconstituted in 3% acetonitrile with 0.1% formic acid.

LC–MS/MS analysis

Peptides were separated using a Thermo Scientific UltiMate 3000 RSLCnano system equipped with an EASY-Spray source. Samples were loaded onto an Acclaim PepMap 100 trap column (2 cm × 75 μm) and separated on a C18 analytical column (25 cm × 75 μm, 1.5 μm; IonOpticks) using a 5–35% acetonitrile gradient in 0.1% formic acid over 50 min at a flow rate of 300 nL/min. The column temperature was maintained at 50 °C. Mass spectrometry was performed on an Orbitrap Fusion Tribrid mass spectrometer operated in data-dependent acquisition mode. MS1 scans were acquired over 350–1500 m/z at 120,000 resolution (at 200 m/z) with an AGC target of 3 × 10⁶ and a maximum injection time of 50 ms. Data-dependent MS/MS acquisition was performed in top-speed mode (3 s cycle time), selecting precursor ions with charge states 2–6 and intensities >5 × 10⁴ for HCD fragmentation (30% NCE) following quadrupole isolation (1.6 Th). MS2 spectra were acquired in the ion trap at rapid scan rate with dynamic exclusion set to 22 s (10 ppm tolerance). Monoisotopic precursor selection was enabled.

Mass spectrometry data analysis

Raw mass spectrometry data were analyzed using MSFragger (v4.3) within FragPipe (v23)⁵⁹. Spectra were searched against the UniProt human database with common contaminants, using a reversed decoy strategy for FDR estimation. Trypsin specificity with up to two missed cleavages was applied. Precursor and fragment mass tolerances were set to 20 ppm. Carbamidomethylation (C) was defined as a fixed modification, with oxidation (M) and protein N-terminal acetylation as variable modifications. Peptide-spectrum matches were filtered to 1% FDR at the PSM and protein levels using PeptideProphet and ProteinProphet. Label-free quantification was performed using IonQuant with match-between-runs enabled.

Model building, refinement, and validation

For all structures, molecular models of GluA2, GSG1L, TARP-γ5, and CNIH subunits were built in Coot using cryo-EM density as well as structures 5WEO, 7RZ9, 8SS9, and 8SS7 as guides⁶⁰. The resulting models were real space refined in Phenix and visualized in Chimera/ChimeraX or Pymol (The PyMOL Molecular Graphics System, version 2.0, Schrödinger)^61,62. Domain movements were determined using the DynDom server (http://dyndom.cmp.uea.ac.uk/dyndom/). Pore radii were calculated using HOLE⁶³.

Whole-cell patch-clamp recordings

DNA corresponding to the modified GluA2, fusion constructs, and disease mutants were introduced into a pIRES plasmid for transient expression in HEK 293 cells (ATCC, CRL-1573). Cells were grown on glass coverslips in 35 mm dishes and transiently transfected with 1–5 μg of the plasmid DNA using Lipofectamine 2000 Reagent (Invitrogen). Recordings were made at 24–72 h after transfection, performed at room temperature. Currents from whole cells, typically held at −60 mV potential, were recorded using Axopatch 200B amplifier (Molecular Devices, LLC), filtered at 5 kHz, and digitized at 10 kHz using low-noise data acquisition system Digidata 1440 A and pClamp 10.2 software (Molecular Devices). The external solution contained (in mM): 150 NaCl, 2.4 KCl, 4 CaCl₂, 4 MgCl₂, and 10 HEPES pH 7.3; 7 mM NaCl was added to the extracellular activating solution containing 3 mM Glu to improve visualization of the border between two solutions coming out of a two-barrel theta glass pipette, which allowed its more precise positional adjustment for faster solution exchange. The internal solution contained (in mM): 150 CsF, 10 NaCl, 10 EGTA, 20 HEPES pH 7.3. Rapid solution exchange was achieved with a two-barrel theta glass pipette controlled by a piezoelectric translator. Typical 10–90% rise times were 200–300 μs, as measured from junction potentials at the open tip of the patch pipette after recordings. Data analysis was performed using Clampfit 10.7 and Origin 2025 software (OriginLab Corporation).

Noise analysis

Nonstationary noise analysis was performed according to previously established protocols⁴⁵. Briefly, whole-cell patch-clamp recordings were performed from HEK 293 cells transfected with GluA2, GluA2-GSG1L, GluA2-γ5, or GluA2-γ2, using Glu as the activating ligand. A total of 100−200 sweeps were recorded from each cell at a holding potential of −60 mV. Each sweep consisted of a 2-ms Glu application at a concentration of 3 mM. Following each application, stimulus-free currents were recorded for 1 s. The first 250 ms of the current deactivation phase was used for non-stationary noise analysis. Only sweeps obtained from cells with high-quality, stable seals and negligible current rundown were included. Current variance was calculated as half of the mean squared difference between consecutive sweeps⁶⁴. The relationship between the variance and mean current was fitted with a parabolic function to determine the mean unitary current (i, Eq. (1)) and the background current variance (σ_b² at the current value of 0 pA). Given low open probability and complex subconductance distribution, the parabolic dependence did not allow for proper open probability determination. The unitary conductance values (γ) were determined as quotient of mean unitary currents and membrane voltage.

σ^{2} = P_{o} (1 - P_{o}) N i^{2} + σ_{b}^{2} = {- P}_{o}^{2} N i^{2} + P_{o} N i^{2} + σ_{b}^{2} = - \frac{I^{2}}{N} + I \cdot i + σ_{b}^{2}

System preparation for molecular dynamics simulations

Initial atomic coordinates for MD simulations were obtained from the open cryo-EM structures of the GluA2 AMPA receptor in its naked form and in complex with different auxiliary proteins. Five systems were simulated, including GluA2_Apo, GluA2_{Glu + RR2b}, GluA2-GSG1L_{Glu + RR2b}, GluA2-γ5-CNIH_{Glu + RR2b} and GluA2-γ2_{Glu + CTZ} (5WEO)⁴. All systems were homotetrameric GluA2 receptors, and only the LBD, TMD and auxiliary subunits were included in the simulations. The GluA2_Apo, GluA2_{Glu + RR2b} and GluA2-γ2_{Glu + CTZ} (5WEO) systems were derived from full-length structures and truncated to include residues 392–820, while the GluA2-γ5-CNIH_{Glu + RR2b} and GluA2-GSG1L_{Glu + RR2b} systems were LBD-TMD-only structures comprising residues 394–820.

Missing intracellular M1-M2 loops were remodeled in all systems using MODELLER⁶⁵. In addition, missing LBD-M3 and LBD-M4 linkers were remodeled for the GluA2-GSG1L_{Glu + RR2b} system, and missing LBD-M4 linkers were remodeled for the GluA2_{Glu + RR2b} system.

All systems except GluA2_Apo and GluA2-γ2_{Glu + CTZ} (5WEO) contained four glutamate molecules bound to the LBDs and two (R,R)−2b PAM molecules located at the LBD dimer interfaces. GluA2-γ2_{Glu + CTZ} (5WEO) contained four glutamate molecules bound to the LBDs and four CTZ PAM molecules located at the LBD dimer interfaces. GluA2-γ2_{Glu + CTZ} (5WEO) system contained four γ2 auxiliary subunits. The GluA2-GSG1L_{Glu + RR2b} system contained two GSG1L auxiliary subunits, whereas the GluA2-γ5-CNIH_{Glu + RR2b} system contained two CNIH and two γ5 auxiliary subunits. The GluA2_Apo and GluA2_{Glu + RR2b} systems did not include any auxiliary proteins; although a spermidine molecule was present in the experimental structure of GluA2_{Glu + RR2b}, it was removed prior to simulations.

Each simulation box was constructed using the CHARMM-GUI^66,67 membrane builder by embedding the protein complexes into a POPC bilayer, followed by solvation with TIP3P water molecules and addition of 150 mM KCl. The systems were prepared for MD simulations using the tleap module of AmberTools⁶⁸. Protein and ions were described using the Amber ff99SB-ILDN⁶⁹ force field, water using the TIP3P model, and lipids using the Lipid14⁷⁰ force field. All ligands, including glutamate, CTZ and (R,R)−2b, were parametrized using the general AMBER force field (GAFF2)⁷¹.

The total number of atoms in the final simulation boxes was 270,489 for the GluA2_Apo system, 355,382 for the GluA2_{Glu + RR2b} system, 297,288 for the GluA2-GSG1L_{Glu + RR2b} system, 375,319 for the GluA2-γ2_{Glu + CTZ} system and 293,297 for the GluA2-γ5-CNIH_{Glu + RR2b} system. The GluA2_Apo system contained 59,857 water molecules and 483 lipid molecules, GluA2_{Glu + RR2b} system contained 83,162 water molecules and 585 lipid molecules, GluA2-γ2_{Glu + CTZ} system contained 88,193 water molecules and 524 lipid molecules, GluA2-GSG1L_{Glu + RR2b} system contained 66,966 water molecules and 472 lipid molecules, and the GluA2-γ5-CNIH_{Glu + RR2b} system contained 65,127 water molecules and 444 lipid molecules.

Molecular dynamics simulation protocols

All MD simulations were performed using the pmemd.cuda program of the AMBER MD package. Simulations of the GluA2_Apo, GluA2-γ2_{Glu + CTZ}, GluA2-GSG1L_{Glu + RR2b} and GluA2-γ5-CNIH_{Glu + RR2b} systems were carried out using Amber20⁶⁸, while simulations of the GluA2_{Glu + RR2b} system were carried out using Amber24⁷². The systems were simulated in the NPT ensemble at a temperature of 300 K and a pressure of 1 bar with semi-isotropic pressure scaling. Temperature was maintained using a Langevin thermostat with a collision frequency of 1 ps^–¹, and pressure was controlled using a Berendsen barostat with a relaxation time of 1 ps.

All covalent bonds involving hydrogen atoms were constrained using the SHAKE algorithm⁷³. A time step of 2 fs was used for the GluA2_Apo, GluA2-γ2_{Glu + CTZ}, GluA2-GSG1L_{Glu + RR2b} and GluA2-γ5-CNIH_{Glu + RR2b} systems. For the GluA2_{Glu + RR2b} system, a 4 fs time step was employed using hydrogen mass repartitioning^74,75. Long-range electrostatic interactions were calculated using the particle mesh Ewald method⁷⁶, with a non-bonded interaction cutoff of 10 Å. Periodic boundary conditions were applied in all directions.

Each system was initially energy minimized with harmonic restraints applied to protein Cα atoms and ligand heavy atoms. The systems were then gradually heated from 0 to 300 K under constant volume conditions, with all protein main-chain heavy atoms as well as ligand and lipid heavy atoms restrained to their minimized positions using a force constant of 40 kcal mol^–1Å^–1. This was followed by equilibration under constant pressure for approximately 100 ns, during which the restraints on the protein and ligands were gradually released.

For the GluA2_{Glu + RR2b} system, additional restraints were applied during equilibration due to collapse of the channel gate observed in preliminary simulations. Distance restraints with a force constant of 10 kcal mol⁻¹Å⁻¹ were applied between the Cα atoms of the gate residues A621 across all four subunits and between the Cα atoms of T625 across all four subunits throughout the equilibration phase and an additional 40 ns simulation. All restraints were subsequently removed prior to production simulations.

All the production simulations were carried out without restraints for 500 ns.

Molecular dynamics trajectory analysis

Post-processing and analysis of the MD trajectories were performed using the CPPTRAJ module⁷⁷ of AmberTools24 and VMD 1.9.4⁷⁸. VMD 1.9.4 was used to visualize trajectories and to generate simulation snapshot figures. Gate areas were computed by calculating the area of the polygon formed by the gate residues across the four subunits, separately for T617, A621, and T625, and reported in angstrom (Å) as the square root of actual value.

Water permeation analysis was performed using VMD by counting water molecules crossing the channel pore along the membrane normal. The average water permeation rate per nanosecond was calculated as the cumulative average of total upward (+Z) and downward (−Z) permeation events divided by the total simulation time, where +Z corresponds to the extracellular direction and −Z corresponds to the intracellular direction. A permeation event was counted when a water molecule passed between the boundary points defined by the Cα atoms of residues T617 and T625. Cumulative averages were calculated by integrating the mean permeation count over the duration of the simulation. Rapid water movements were captured using trajectory frames saved every 100 ps. Water permeation data were calculated from continuous production trajectories. Water occupancy maps were generated using the VolMap Occupancy tool in VMD. Water oxygen positions from all trajectory frames were mapped onto a 1 Å grid and averaged over the simulation to obtain frame-averaged water occupancy. Three-dimensional isosurfaces were rendered using an occupancy isovalue of 0.1 to visualize regions that remained persistently hydrated throughout the simulation.

Interfacial interactions were analyzed over the production trajectories using two criteria. Side-chain contacts were defined by a 4-Å distance cutoff between any pair of side-chain heavy atoms of the auxiliary subunit and the receptor, and the occupancies were reported as the fraction of frames in which the criterion was satisfied. Hydrogen bonds were identified using a 3.6-Å donor-acceptor distance cutoff together with a 90° angular criterion and were reported as cumulative occupancies summed over all donor-acceptor pairs formed by a given residue pair.

Statistics and reproducibility

Statistical analysis was performed using Origin 2023. Statistical significance was calculated using two-sample t-Test, with the significance assumed if P < 0.05. In all figure legends, n represents the number of independent biological replicates. All quantitative data were presented as mean ± SEM.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information^{(25.1MB, pdf)}

Reporting Summary^{(78.8KB, pdf)}

Transparent Peer Review file^{(1.7MB, pdf)}

Source data

Source Data file^{(126.8KB, zip)}

Acknowledgements

We thank R. Grassucci and Z. Zhang (Columbia University Cryo-Electron Microscopy Center), and G. Angiulli, J. Miller, and C. Hernandez (Simons Electron Microscopy Center), for help with microscope operation and data collection. Some of this work was performed at the Columbia University Cryo-Electron Microscopy Center. Some of this work was performed at the National Center for CryoEM Access and Training (NCCAT) and the Simons Electron Microscopy Center located at the New York Structural Biology Center, supported by the National Institutes of Health (NIH) Common Fund Transformative High Resolution Cryo-Electron Microscopy program (U24 GM129539) and by grants from the Simons Foundation (SF349247) and NY State Assembly Majority. Some of this work was performed at the Stanford-SLAC Cryo-EM Center (S2C2), supported by the NIH Common Fund Transformative High Resolution Cryo-Electron Microscopy program (U24 GM129541). A portion of this research was supported by National Institutes of Health (NIH) grant U24GM129547 and performed at the Pacific Northwest Center for Cryo-EM (PNCC) at Oregon Health and Science University and accessed through the Environmental Molecular Sciences Laboratory (grid.436923.9), a Department of Energy Office of Science User Facility sponsored by the Office of Biological and Environmental Research. We thank R.K. Soni and C.W. Karanja from the Proteomics and Structural Biology Shares Resource Core Facility for help with mass spectrometry data collection and analysis. We thank R.P. Clausen from the University of Copenhagen for providing us with the positive allosteric modulator (R,R)−2b.

Author contributions

T.P.N. and A.I.S. conceptualized the project. T.P.N., M.V.Y., S.P.G., L.Y.Y., and A.I.S. designed the experiments. T.P.N., S.P.G. and I.A.S. made constructs for protein expression and electrophysiology. T.P.N. and S.P.G. performed protein expression and purification. T.P.N., S.P.G. and L.Y.Y. made the grids and collected cryo-EM data. T.P.N., S.P.G. and L.Y.Y. processed the cryo-EM data. M.V.Y. and A.A.A. performed patch-clamp recordings and electrophysiological data analysis. M.A. and M.G.K designed computational studies. M.A. performed MD simulations and analysis. T.P.N., I.A.S. and A.I.S. built molecular models. T.P.N. and A.I.S. wrote the manuscript, which was then edited by all the authors. A.I.S. supervised the project.

Peer review

Peer review information

Nature Communications thanks Oliver Beckstein, Sushant Kumar, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

T.P.N. was supported by the NIH grant F31 NS147755. S.P.G. was supported by the NIH grant NS139087. L.Y.Y. was supported by the NIH grant F31 NS132554. A.I.S. was supported by the Human Frontier Science Program (HFSP) Award and the NIH grants (NS083660, NS107253).

Data availability

The cryo-EM density maps have been deposited to the Electron Microscopy Data Bank (EMDB) under the accession codes EMD-76060 for the ATD-LBD map of GluA2_{Glu + RR2b}, EMD-76061 for the TMD map of GluA2_{Glu + RR2b}, EMD-76059 for the consensus map of GluA2_{Glu + RR2b}, EMD-76058 for the full-length composite map of GluA2_{Glu + RR2b}, EMD-76057 for the LBD-TMD map of GluA2-GSG1L_{Glu + RR2b}, EMD-76056 for the LBD-TMD map of GluA2-GSG1L_Apo and EMD-76055 for the LBD-TMD map of GluA2-γ5-CNIH_{Glu + RR2b}. The atomic coordinates have been deposited to the Protein Data Bank (PDB) under the accession codes 11UI for GluA2_{Glu + RR2b}, 11UH for the LBD-TMD of GluA2-GSG1L_{Glu + RR2b}, 11UG for the LBD-TMD of GluA2-GSG1L_Apo and 11UF for the LBD-TMD of GluA2-γ5-CNIH_{Glu + RR2b}. The molecular dynamics simulation trajectories and topology files have been deposited in Zenodo under the accession code https://doi.org/10.5281/zenodo.19000885 [https://doi.org/10.5281/zenodo.19000885]. The repository also includes the initial coordinates for the production runs, as well as a table summarizing the corresponding file names. The raw simulation data have not been deposited due to their size. However, access to them can be obtained by contacting the authors. Mass spectrometry results are included in the Source Data file. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77416-0.

References

1.Hansen, K. B. et al. Structure, function, and pharmacology of glutamate receptor ion channels. Pharm. Rev.73, 298–487 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Yelshanskaya, M. V. & Sobolevsky, A. I. Structural insights into function of ionotropic glutamate receptors. Biochem. Mosc. Suppl. Ser. A16, 190–206 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Reiner, A. & Levitz, J. Glutamatergic signaling in the central nervous system: ionotropic and metabotropic receptors in concert. Neuron98, 1080–1098 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Twomey, E. C., Yelshanskaya, M. V., Grassucci, R. A., Frank, J. & Sobolevsky, A. I. Channel opening and gating mechanism in AMPA-subtype glutamate receptors. Nature549, 60–65 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Joshi, I., Shokralla, S., Titis, P. & Wang, L.-Y. The role of AMPA receptor gating in the development of high-fidelity neurotransmission at the calyx of held synapse. J. Neurosci.24, 183–196 (2004). [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Derkach, V. A., Oh, M. C., Guire, E. S. & Soderling, T. R. Regulatory mechanisms of AMPA receptors in synaptic plasticity. Nat. Rev. Neurosci.8, 101–113 (2007). [DOI] [PubMed] [Google Scholar]
7.Henley, J. M. & Wilkinson, K. A. Synaptic AMPA receptor composition in development, plasticity and disease. Nat. Rev. Neurosci.17, 337–350 (2016). [DOI] [PubMed] [Google Scholar]
8.Lu, K. et al. Impairments in endogenous AMPA receptor dynamics correlates with learning deficits in Alzheimer’s disease model mice. Proc. Natl. Acad. Sci. USA.120, e2303878120 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Guo, C. & Ma, Y.-Y. Calcium permeable-AMPA receptors and excitotoxicity in neurological disorders. Front. Neural Circuits15, 711564 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Hanada, T. et al. Perampanel: a novel, orally active, noncompetitive AMPA-receptor antagonist that reduces seizure activity in rodent models of epilepsy. Epilepsia52, 1331–1340 (2011). [DOI] [PubMed] [Google Scholar]
11.Gangwar, S. P. et al. Modulation of GluA2–γ5 synaptic complex desensitization, polyamine block and antiepileptic perampanel inhibition by auxiliary subunit cornichon−2. Nat. Struct. Mol. Biol.30, 1481–1494 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Yelshanskaya, M. V. et al. Structural bases of noncompetitive inhibition of AMPA-subtype ionotropic glutamate receptors by antiepileptic drugs. Neuron91, 1305–1315 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Gan, Q., Salussolia, C. L. & Wollmuth, L. P. Assembly of AMPA receptors: mechanisms and regulation. J. Physiol.593, 39–48 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Sobolevsky, A. I., Rosconi, M. P. & Gouaux, E. X-ray structure, symmetry and mechanism of an AMPA-subtype glutamate receptor. Nature462, 745–756 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Twomey, E. C., Yelshanskaya, M. V., Grassucci, R. A., Frank, J. & Sobolevsky, A. I. Structural bases of desensitization in AMPA receptor-auxiliary subunit complexes. Neuron94, 569–580.e5 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Bissen, D., Foss, F. & Acker-Palmer, A. AMPA receptors and their minions: auxiliary proteins in AMPA receptor trafficking. Cell Mol. Life Sci.76, 2133–2169 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Twomey, E. C., Yelshanskaya, M. V. & Sobolevsky, A. I. Structural and functional insights into transmembrane AMPA receptor regulatory protein complexes. J. Gen. Physiol.151, 1347–1356 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Cull-Candy, S. G. & Farrant, M. Ca2+-permeable AMPA receptors and their auxiliary subunits in synaptic plasticity and disease. J. Physiol.599, 2655–2671 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Priel, A. et al. Stargazin reduces desensitization and slows deactivation of the AMPA-type glutamate receptors. J. Neurosci.25, 2682–2686 (2005). [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Jackson, A. C. & Nicoll, R. A. The expanding social network of ionotropic glutamate receptors: TARPs and other transmembrane auxiliary subunits. Neuron70, 178–199 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Twomey, E. C., Yelshanskaya, M. V., Grassucci, R. A., Frank, J. & Sobolevsky, A. I. Elucidation of AMPA receptor–stargazin complexes by cryo–electron microscopy. Science353, 83–86 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Dürr, K. L. et al. Structure and dynamics of AMPA receptor GluA2 in resting, pre-open, and desensitized states. Cell158, 778–792 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Carrillo, E. et al. Mechanism of modulation of AMPA receptors by TARP-γ8. J. Gen. Physiol.152, e201912451 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Klykov, O., Gangwar, S. P., Yelshanskaya, M. V., Yen, L. & Sobolevsky, A. I. Structure and desensitization of AMPA receptor complexes with type II TARP γ5 and GSG1L. Mol. Cell81, e7 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Soto, D. et al. Selective regulation of long-form calcium-permeable AMPA receptors by an atypical TARP, gamma-5. Nat. Neurosci.12, 277–285 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Coombs, I. D. et al. Cornichons modify channel properties of recombinant and glial AMPA receptors. J. Neurosci.32, 9796–9804 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Gill, M. B. et al. Cornichon−2 modulates AMPA receptor–transmembrane AMPA receptor regulatory protein assembly to dictate gating and pharmacology. J. Neurosci.31, 6928–6938 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Kato, A. S. et al. Hippocampal AMPA receptor gating controlled by both TARP and cornichon proteins. Neuron68, 1082–1096 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Gu, X. et al. GSG1L suppresses AMPA receptor-mediated synaptic transmission and uniquely modulates AMPA receptor kinetics in hippocampal neurons. Nat. Commun.7, 10873 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Kamalova, A., Futai, K., Delpire, E. & Nakagawa, T. AMPA receptor auxiliary subunit GSG1L suppresses short-term facilitation in corticothalamic synapses and determines seizure susceptibility. Cell Rep.32, 107921 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Mao, X., Gu, X. & Lu, W. GSG1L regulates the strength of AMPA receptor-mediated synaptic transmission but not AMPA receptor kinetics in hippocampal dentate granule neurons. J. Neurophysiol.117, 28–35 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
32.McGee, T. P., Bats, C., Farrant, M. & Cull-Candy, S. G. Auxiliary subunit GSG1L acts to suppress calcium-permeable AMPA receptor function. J. Neurosci.35, 16171–16179 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Yelshanskaya, M. V., Li, M. & Sobolevsky, A. I. Structure of an agonist-bound ionotropic glutamate receptor. Science345, 1070–1074 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Baranovic, J. et al. The action of Con-ikot-ikot toxin on single AMPA-type glutamate receptors. J. Gen. Physiol.154, e202112912 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Sun, Y. et al. Mechanism of glutamate receptor desensitization. Nature417, 245–253 (2002). [DOI] [PubMed] [Google Scholar]
36.Armstrong, N., Jasti, J., Beich-Frandsen, M. & Gouaux, E. Measurement of conformational changes accompanying desensitization in an ionotropic glutamate receptor. Cell127, 85–97 (2006). [DOI] [PubMed] [Google Scholar]
37.Chen, S. et al. Activation and desensitization mechanism of AMPA receptor-TARP complex by cryo-EM. Cell170, 1234–1246.e14 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Twomey, E. C. & Sobolevsky, A. I. Structural mechanisms of gating in ionotropic glutamate receptors. Biochemistry57, 267–276 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Miguez-Cabello, F. et al. GluA2-containing AMPA receptors form a continuum of Ca²⁺-permeable channels. Nature641, 537–544 (2025). [DOI] [PubMed] [Google Scholar]
40.Yelshanskaya, M. V., Patel, D. S., Kottke, C. M., Kurnikova, M. G. & Sobolevsky, A. I. Opening of glutamate receptor channel to subconductance levels. Nature605, 172–178 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Twomey, E. C., Yelshanskaya, M. V., Vassilevski, A. A. & Sobolevsky, A. I. Mechanisms of channel block in calcium-permeable AMPA receptors. Neuron99, 956–968.e4 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Gangwar, S. P. et al. Trapping of spermine, Kukoamine A, and polyamine toxin blockers in GluK2 kainate receptor channels. Nat. Commun.15, 10257 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Dawe, G., Musgaard, M., Biggin, P. & Al, E. Distinct structural pathways coordinate the activiation of AMPA receptor-auxiliary subunit complexes. Neuron89, 1264–1276 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Zhao, Y., Chen, S., Yoshioka, C., Baconguis, I. & Gouaux, E. Architecture of fully occupied GluA2 AMPA receptor-TARP complex elucidated by cryo-EM. Nature536, 108–111 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Alekseev, A. et al. Efficient and sustained optogenetic control of sensory and cardiac systems. Nat. Biomed. Eng.10, 277–292 (2026). [DOI] [PMC free article] [PubMed]
46.Shelley, C., Farrant, M. & Cull-Candy, S. G. TARP-associated AMPA receptors display an increased maximum channel conductance and multiple kinetically distinct open states. J. Physiol.590, 5723–5738 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Coombs, I. D. et al. Homomeric GluA2(R) AMPA receptors can conduct when desensitized. Nat. Commun.10, 4312 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Wollmuth, L. P. & Sobolevsky, A. I. Structure and gating of the glutamate receptor ion channel. Trends Neurosci.27, 321–328 (2004). [DOI] [PubMed] [Google Scholar]
49.Salpietro, V. et al. AMPA receptor GluA2 subunit defects are a cause of neurodevelopmental disorders. Nat. Commun.10, 3094 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Gangwar, S. P., Yelshanskaya, M. V., Yen, L. Y., Newton, T. P. & Sobolevsky, A. I. Activation of kainate receptor GluK2–Neto2 complex. Nat. Struct. Mol. Biol.32, 2176–2184 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
51.Daniels, B. A., Andrews, E. D., Aurousseau, M. R. P., Accardi, M. V. & Bowie, D. Crosslinking the ligand-binding domain dimer interface locks kainate receptors out of the main open state. J. Physiol.591, 3873–3885 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Baranovic, J. et al. Dynamics of the ligand binding domain layer during AMPA receptor activation. Biophys. J.110, 896–911 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Esmenjaud, J.-B. et al. An inter-dimer allosteric switch controls NMDA receptor activity. EMBO J.38, e99894 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Yu, J. et al. Hippocampal AMPA receptor assemblies and mechanism of allosteric inhibition. Nature594, 448–453 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Lu, W. et al. Subunit composition of synaptic AMPA receptors revealed by a single-cell genetic approach. Neuron62, 254–268 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Goehring, A. et al. Screening and large-scale expression of membrane proteins in mammalian cells for structural studies. Nat. Protoc.9, 2574–2585 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
57.Soni, R. K. Protocol for deep proteomic profiling of formalin-fixed paraffin-embedded specimens using a spectral library-free approach. STAR Protoc.4, 102381 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Kulak, N. A., Pichler, G., Paron, I., Nagaraj, N. & Mann, M. Minimal, encapsulated proteomic-sample processing applied to copy-number estimation in eukaryotic cells. Nat. Methods11, 319–324 (2014). [DOI] [PubMed] [Google Scholar]
59.Kong, A. T., Leprevost, F. V., Avtonomov, D. M., Mellacheruvu, D. & Nesvizhskii, A. I. MSFragger: ultrafast and comprehensive peptide identification in mass spectrometry–based proteomics. Nat. Methods14, 513–520 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
60.Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr. D. Biol. Crystallogr.66, 486–501 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
61.Afonine, P. V. et al. Towards automated crystallographic structure refinement with phenix.refine. Acta Crystallogr. D. Biol. Crystallogr.68, 352–367 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Pettersen, E. F. et al. UCSF ChimeraX: structure visualization for researchers, educators, and developers. Protein Sci.30, 70–82 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
63.Smart, O. S., Neduvelil, J. G., Wang, X., Wallace, B. A. & Sansom, M. S. HOLE: a program for the analysis of the pore dimensions of ion channel structural models. J. Mol. Graph14, 354–360, 376 (1996). [DOI] [PubMed] [Google Scholar]
64.Roberts, W., Jacobs, R. & Hudspeth, A. Colocalization of ion channels involved in frequency selectivity and synaptic transmission at presynaptic active zones of hair cells. J. Neurosci.10, 3664–3684 (1990). [DOI] [PMC free article] [PubMed] [Google Scholar]
65.Sali, A. & Blundell, T. L. Comparative protein modelling by satisfaction of spatial restraints. J. Mol. Biol.234, 779–815 (1993). [DOI] [PubMed] [Google Scholar]
66.Wu, E. L. et al. CHARMM-GUI membrane builder toward realistic biological membrane simulations. J. Comput. Chem.35, 1997–2004 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
67.Jo, S., Kim, T., Iyer, V. G. & Im, W. CHARMM-GUI: a web-based graphical user interface for CHARMM. J. Comput. Chem.29, 1859–1865 (2008). [DOI] [PubMed] [Google Scholar]
68.Case, D. A. et al. AmberTools. J. Chem. Inf. Model.63, 6183–6191 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
69.Lindorff-Larsen, K. et al. Improved side-chain torsion potentials for the Amber ff99SB protein force field. Proteins78, 1950–1958 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Dickson, C. J. et al. Lipid14: the amber lipid force field. J. Chem. Theory Comput.10, 865–879 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
71.Wang, J., Wolf, R. M., Caldwell, J. W., Kollman, P. A. & Case, D. A. Development and testing of a general amber force field. J. Comput. Chem.25, 1157–1174 (2004). [DOI] [PubMed] [Google Scholar]
72.Case, D. A. et al. Recent developments in amber biomolecular simulations. J. Chem. Inf. Model.65, 7835–7843 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
73.Ryckaert, J.-P., Ciccotti, G. & Berendsen, H. J. C. Numerical integration of the cartesian equations of motion of a system with constraints: molecular dynamics of n-alkanes. J. Comput. Phys.23, 327–341 (1977). [Google Scholar]
74.Feenstra, K. A., Hess, B. & Berendsen, H. J. C. Improving efficiency of large time-scale molecular dynamics simulations of hydrogen-rich systems. J. Comput. Chem.20, 786–798 (1999). [DOI] [PubMed] [Google Scholar]
75.Hopkins, C. W., Le Grand, S., Walker, R. C. & Roitberg, A. E. Long-time-step molecular dynamics through hydrogen mass repartitioning. J. Chem. Theory Comput.11, 1864–1874 (2015). [DOI] [PubMed] [Google Scholar]
76.Darden, T., York, D. & Pedersen, L. Particle mesh Ewald: an N⋅log(N) method for Ewald sums in large systems. J. Chem. Phys.98, 10089–10092 (1993). [Google Scholar]
77.Roe, D. R. & Cheatham, T. E. PTRAJ and CPPTRAJ: software for processing and analysis of molecular dynamics trajectory data. J. Chem. Theory Comput.9, 3084–3095 (2013). [DOI] [PubMed] [Google Scholar]
78.Humphrey, W., Dalke, A. & Schulten, K. VMD: visual molecular dynamics. J. Mol. Graph.14, 33–38 (1996). [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Information^{(25.1MB, pdf)}

Reporting Summary^{(78.8KB, pdf)}

Transparent Peer Review file^{(1.7MB, pdf)}

Source Data file^{(126.8KB, zip)}

Data Availability Statement

[CR1] 1.Hansen, K. B. et al. Structure, function, and pharmacology of glutamate receptor ion channels. Pharm. Rev.73, 298–487 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR2] 2.Yelshanskaya, M. V. & Sobolevsky, A. I. Structural insights into function of ionotropic glutamate receptors. Biochem. Mosc. Suppl. Ser. A16, 190–206 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR3] 3.Reiner, A. & Levitz, J. Glutamatergic signaling in the central nervous system: ionotropic and metabotropic receptors in concert. Neuron98, 1080–1098 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR4] 4.Twomey, E. C., Yelshanskaya, M. V., Grassucci, R. A., Frank, J. & Sobolevsky, A. I. Channel opening and gating mechanism in AMPA-subtype glutamate receptors. Nature549, 60–65 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR5] 5.Joshi, I., Shokralla, S., Titis, P. & Wang, L.-Y. The role of AMPA receptor gating in the development of high-fidelity neurotransmission at the calyx of held synapse. J. Neurosci.24, 183–196 (2004). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR6] 6.Derkach, V. A., Oh, M. C., Guire, E. S. & Soderling, T. R. Regulatory mechanisms of AMPA receptors in synaptic plasticity. Nat. Rev. Neurosci.8, 101–113 (2007). [DOI] [PubMed] [Google Scholar]

[CR7] 7.Henley, J. M. & Wilkinson, K. A. Synaptic AMPA receptor composition in development, plasticity and disease. Nat. Rev. Neurosci.17, 337–350 (2016). [DOI] [PubMed] [Google Scholar]

[CR8] 8.Lu, K. et al. Impairments in endogenous AMPA receptor dynamics correlates with learning deficits in Alzheimer’s disease model mice. Proc. Natl. Acad. Sci. USA.120, e2303878120 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR9] 9.Guo, C. & Ma, Y.-Y. Calcium permeable-AMPA receptors and excitotoxicity in neurological disorders. Front. Neural Circuits15, 711564 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR10] 10.Hanada, T. et al. Perampanel: a novel, orally active, noncompetitive AMPA-receptor antagonist that reduces seizure activity in rodent models of epilepsy. Epilepsia52, 1331–1340 (2011). [DOI] [PubMed] [Google Scholar]

[CR11] 11.Gangwar, S. P. et al. Modulation of GluA2–γ5 synaptic complex desensitization, polyamine block and antiepileptic perampanel inhibition by auxiliary subunit cornichon−2. Nat. Struct. Mol. Biol.30, 1481–1494 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR12] 12.Yelshanskaya, M. V. et al. Structural bases of noncompetitive inhibition of AMPA-subtype ionotropic glutamate receptors by antiepileptic drugs. Neuron91, 1305–1315 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR13] 13.Gan, Q., Salussolia, C. L. & Wollmuth, L. P. Assembly of AMPA receptors: mechanisms and regulation. J. Physiol.593, 39–48 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR14] 14.Sobolevsky, A. I., Rosconi, M. P. & Gouaux, E. X-ray structure, symmetry and mechanism of an AMPA-subtype glutamate receptor. Nature462, 745–756 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR15] 15.Twomey, E. C., Yelshanskaya, M. V., Grassucci, R. A., Frank, J. & Sobolevsky, A. I. Structural bases of desensitization in AMPA receptor-auxiliary subunit complexes. Neuron94, 569–580.e5 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR16] 16.Bissen, D., Foss, F. & Acker-Palmer, A. AMPA receptors and their minions: auxiliary proteins in AMPA receptor trafficking. Cell Mol. Life Sci.76, 2133–2169 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR17] 17.Twomey, E. C., Yelshanskaya, M. V. & Sobolevsky, A. I. Structural and functional insights into transmembrane AMPA receptor regulatory protein complexes. J. Gen. Physiol.151, 1347–1356 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR18] 18.Cull-Candy, S. G. & Farrant, M. Ca2+-permeable AMPA receptors and their auxiliary subunits in synaptic plasticity and disease. J. Physiol.599, 2655–2671 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR19] 19.Priel, A. et al. Stargazin reduces desensitization and slows deactivation of the AMPA-type glutamate receptors. J. Neurosci.25, 2682–2686 (2005). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR20] 20.Jackson, A. C. & Nicoll, R. A. The expanding social network of ionotropic glutamate receptors: TARPs and other transmembrane auxiliary subunits. Neuron70, 178–199 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR21] 21.Twomey, E. C., Yelshanskaya, M. V., Grassucci, R. A., Frank, J. & Sobolevsky, A. I. Elucidation of AMPA receptor–stargazin complexes by cryo–electron microscopy. Science353, 83–86 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR22] 22.Dürr, K. L. et al. Structure and dynamics of AMPA receptor GluA2 in resting, pre-open, and desensitized states. Cell158, 778–792 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR23] 23.Carrillo, E. et al. Mechanism of modulation of AMPA receptors by TARP-γ8. J. Gen. Physiol.152, e201912451 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR24] 24.Klykov, O., Gangwar, S. P., Yelshanskaya, M. V., Yen, L. & Sobolevsky, A. I. Structure and desensitization of AMPA receptor complexes with type II TARP γ5 and GSG1L. Mol. Cell81, e7 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR25] 25.Soto, D. et al. Selective regulation of long-form calcium-permeable AMPA receptors by an atypical TARP, gamma-5. Nat. Neurosci.12, 277–285 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR26] 26.Coombs, I. D. et al. Cornichons modify channel properties of recombinant and glial AMPA receptors. J. Neurosci.32, 9796–9804 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR27] 27.Gill, M. B. et al. Cornichon−2 modulates AMPA receptor–transmembrane AMPA receptor regulatory protein assembly to dictate gating and pharmacology. J. Neurosci.31, 6928–6938 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR28] 28.Kato, A. S. et al. Hippocampal AMPA receptor gating controlled by both TARP and cornichon proteins. Neuron68, 1082–1096 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR29] 29.Gu, X. et al. GSG1L suppresses AMPA receptor-mediated synaptic transmission and uniquely modulates AMPA receptor kinetics in hippocampal neurons. Nat. Commun.7, 10873 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR30] 30.Kamalova, A., Futai, K., Delpire, E. & Nakagawa, T. AMPA receptor auxiliary subunit GSG1L suppresses short-term facilitation in corticothalamic synapses and determines seizure susceptibility. Cell Rep.32, 107921 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR31] 31.Mao, X., Gu, X. & Lu, W. GSG1L regulates the strength of AMPA receptor-mediated synaptic transmission but not AMPA receptor kinetics in hippocampal dentate granule neurons. J. Neurophysiol.117, 28–35 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR32] 32.McGee, T. P., Bats, C., Farrant, M. & Cull-Candy, S. G. Auxiliary subunit GSG1L acts to suppress calcium-permeable AMPA receptor function. J. Neurosci.35, 16171–16179 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR33] 33.Yelshanskaya, M. V., Li, M. & Sobolevsky, A. I. Structure of an agonist-bound ionotropic glutamate receptor. Science345, 1070–1074 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR34] 34.Baranovic, J. et al. The action of Con-ikot-ikot toxin on single AMPA-type glutamate receptors. J. Gen. Physiol.154, e202112912 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR35] 35.Sun, Y. et al. Mechanism of glutamate receptor desensitization. Nature417, 245–253 (2002). [DOI] [PubMed] [Google Scholar]

[CR36] 36.Armstrong, N., Jasti, J., Beich-Frandsen, M. & Gouaux, E. Measurement of conformational changes accompanying desensitization in an ionotropic glutamate receptor. Cell127, 85–97 (2006). [DOI] [PubMed] [Google Scholar]

[CR37] 37.Chen, S. et al. Activation and desensitization mechanism of AMPA receptor-TARP complex by cryo-EM. Cell170, 1234–1246.e14 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR38] 38.Twomey, E. C. & Sobolevsky, A. I. Structural mechanisms of gating in ionotropic glutamate receptors. Biochemistry57, 267–276 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR39] 39.Miguez-Cabello, F. et al. GluA2-containing AMPA receptors form a continuum of Ca²⁺-permeable channels. Nature641, 537–544 (2025). [DOI] [PubMed] [Google Scholar]

[CR40] 40.Yelshanskaya, M. V., Patel, D. S., Kottke, C. M., Kurnikova, M. G. & Sobolevsky, A. I. Opening of glutamate receptor channel to subconductance levels. Nature605, 172–178 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR41] 41.Twomey, E. C., Yelshanskaya, M. V., Vassilevski, A. A. & Sobolevsky, A. I. Mechanisms of channel block in calcium-permeable AMPA receptors. Neuron99, 956–968.e4 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR42] 42.Gangwar, S. P. et al. Trapping of spermine, Kukoamine A, and polyamine toxin blockers in GluK2 kainate receptor channels. Nat. Commun.15, 10257 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR43] 43.Dawe, G., Musgaard, M., Biggin, P. & Al, E. Distinct structural pathways coordinate the activiation of AMPA receptor-auxiliary subunit complexes. Neuron89, 1264–1276 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR44] 44.Zhao, Y., Chen, S., Yoshioka, C., Baconguis, I. & Gouaux, E. Architecture of fully occupied GluA2 AMPA receptor-TARP complex elucidated by cryo-EM. Nature536, 108–111 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR45] 45.Alekseev, A. et al. Efficient and sustained optogenetic control of sensory and cardiac systems. Nat. Biomed. Eng.10, 277–292 (2026). [DOI] [PMC free article] [PubMed]

[CR46] 46.Shelley, C., Farrant, M. & Cull-Candy, S. G. TARP-associated AMPA receptors display an increased maximum channel conductance and multiple kinetically distinct open states. J. Physiol.590, 5723–5738 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR47] 47.Coombs, I. D. et al. Homomeric GluA2(R) AMPA receptors can conduct when desensitized. Nat. Commun.10, 4312 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR48] 48.Wollmuth, L. P. & Sobolevsky, A. I. Structure and gating of the glutamate receptor ion channel. Trends Neurosci.27, 321–328 (2004). [DOI] [PubMed] [Google Scholar]

[CR49] 49.Salpietro, V. et al. AMPA receptor GluA2 subunit defects are a cause of neurodevelopmental disorders. Nat. Commun.10, 3094 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR50] 50.Gangwar, S. P., Yelshanskaya, M. V., Yen, L. Y., Newton, T. P. & Sobolevsky, A. I. Activation of kainate receptor GluK2–Neto2 complex. Nat. Struct. Mol. Biol.32, 2176–2184 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR51] 51.Daniels, B. A., Andrews, E. D., Aurousseau, M. R. P., Accardi, M. V. & Bowie, D. Crosslinking the ligand-binding domain dimer interface locks kainate receptors out of the main open state. J. Physiol.591, 3873–3885 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR52] 52.Baranovic, J. et al. Dynamics of the ligand binding domain layer during AMPA receptor activation. Biophys. J.110, 896–911 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR53] 53.Esmenjaud, J.-B. et al. An inter-dimer allosteric switch controls NMDA receptor activity. EMBO J.38, e99894 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR54] 54.Yu, J. et al. Hippocampal AMPA receptor assemblies and mechanism of allosteric inhibition. Nature594, 448–453 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR55] 55.Lu, W. et al. Subunit composition of synaptic AMPA receptors revealed by a single-cell genetic approach. Neuron62, 254–268 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR56] 56.Goehring, A. et al. Screening and large-scale expression of membrane proteins in mammalian cells for structural studies. Nat. Protoc.9, 2574–2585 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR57] 57.Soni, R. K. Protocol for deep proteomic profiling of formalin-fixed paraffin-embedded specimens using a spectral library-free approach. STAR Protoc.4, 102381 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR58] 58.Kulak, N. A., Pichler, G., Paron, I., Nagaraj, N. & Mann, M. Minimal, encapsulated proteomic-sample processing applied to copy-number estimation in eukaryotic cells. Nat. Methods11, 319–324 (2014). [DOI] [PubMed] [Google Scholar]

[CR59] 59.Kong, A. T., Leprevost, F. V., Avtonomov, D. M., Mellacheruvu, D. & Nesvizhskii, A. I. MSFragger: ultrafast and comprehensive peptide identification in mass spectrometry–based proteomics. Nat. Methods14, 513–520 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR60] 60.Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr. D. Biol. Crystallogr.66, 486–501 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR61] 61.Afonine, P. V. et al. Towards automated crystallographic structure refinement with phenix.refine. Acta Crystallogr. D. Biol. Crystallogr.68, 352–367 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR62] 62.Pettersen, E. F. et al. UCSF ChimeraX: structure visualization for researchers, educators, and developers. Protein Sci.30, 70–82 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR63] 63.Smart, O. S., Neduvelil, J. G., Wang, X., Wallace, B. A. & Sansom, M. S. HOLE: a program for the analysis of the pore dimensions of ion channel structural models. J. Mol. Graph14, 354–360, 376 (1996). [DOI] [PubMed] [Google Scholar]

[CR64] 64.Roberts, W., Jacobs, R. & Hudspeth, A. Colocalization of ion channels involved in frequency selectivity and synaptic transmission at presynaptic active zones of hair cells. J. Neurosci.10, 3664–3684 (1990). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR65] 65.Sali, A. & Blundell, T. L. Comparative protein modelling by satisfaction of spatial restraints. J. Mol. Biol.234, 779–815 (1993). [DOI] [PubMed] [Google Scholar]

[CR66] 66.Wu, E. L. et al. CHARMM-GUI membrane builder toward realistic biological membrane simulations. J. Comput. Chem.35, 1997–2004 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR67] 67.Jo, S., Kim, T., Iyer, V. G. & Im, W. CHARMM-GUI: a web-based graphical user interface for CHARMM. J. Comput. Chem.29, 1859–1865 (2008). [DOI] [PubMed] [Google Scholar]

[CR68] 68.Case, D. A. et al. AmberTools. J. Chem. Inf. Model.63, 6183–6191 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR69] 69.Lindorff-Larsen, K. et al. Improved side-chain torsion potentials for the Amber ff99SB protein force field. Proteins78, 1950–1958 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR70] 70.Dickson, C. J. et al. Lipid14: the amber lipid force field. J. Chem. Theory Comput.10, 865–879 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR71] 71.Wang, J., Wolf, R. M., Caldwell, J. W., Kollman, P. A. & Case, D. A. Development and testing of a general amber force field. J. Comput. Chem.25, 1157–1174 (2004). [DOI] [PubMed] [Google Scholar]

[CR72] 72.Case, D. A. et al. Recent developments in amber biomolecular simulations. J. Chem. Inf. Model.65, 7835–7843 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR73] 73.Ryckaert, J.-P., Ciccotti, G. & Berendsen, H. J. C. Numerical integration of the cartesian equations of motion of a system with constraints: molecular dynamics of n-alkanes. J. Comput. Phys.23, 327–341 (1977). [Google Scholar]

[CR74] 74.Feenstra, K. A., Hess, B. & Berendsen, H. J. C. Improving efficiency of large time-scale molecular dynamics simulations of hydrogen-rich systems. J. Comput. Chem.20, 786–798 (1999). [DOI] [PubMed] [Google Scholar]

[CR75] 75.Hopkins, C. W., Le Grand, S., Walker, R. C. & Roitberg, A. E. Long-time-step molecular dynamics through hydrogen mass repartitioning. J. Chem. Theory Comput.11, 1864–1874 (2015). [DOI] [PubMed] [Google Scholar]

[CR76] 76.Darden, T., York, D. & Pedersen, L. Particle mesh Ewald: an N⋅log(N) method for Ewald sums in large systems. J. Chem. Phys.98, 10089–10092 (1993). [Google Scholar]

[CR77] 77.Roe, D. R. & Cheatham, T. E. PTRAJ and CPPTRAJ: software for processing and analysis of molecular dynamics trajectory data. J. Chem. Theory Comput.9, 3084–3095 (2013). [DOI] [PubMed] [Google Scholar]

[CR78] 78.Humphrey, W., Dalke, A. & Schulten, K. VMD: visual molecular dynamics. J. Mol. Graph.14, 33–38 (1996). [DOI] [PubMed] [Google Scholar]

PERMALINK

Tuning of AMPA receptor activation by inhibitory auxiliary subunits and epilepsy-associated disease mutations

Thomas P Newton

Maria V Yelshanskaya

Muhammed Aktolun

Shanti Pal Gangwar

Laura Y Yen

Alexey A Alekseev

Ivan A Sobolevsky

Maria G Kurnikova

Alexander I Sobolevsky

Abstract

Introduction

Results

Structure of homomeric GluA2 in the open state

Fig. 1. Structure and activation of GluA2 alone.

Open-state structures of GluA2 in complex with inhibitory subunits GSG1L or γ5

Fig. 2. Structures of GluA2-GSG1L and GluA2-γ5-CNIH complexes in the open state.

GSG1L and γ5 architectures and interactions with GluA2

Interactions between GluA2 LBD and auxiliary subunit ECD

Fig. 3. Interactions of LBDs and extracellular domains of auxiliary subunits.

Fig. 4. Functional characterization of GluA2-K697A-GSG1L.

Activation of GluA2-GSG1L and GluA2-γ5-CNIH at the LBD level

Fig. 5. LBDs in AMPAR activation in the presence of different auxiliary subunits.

Opening of the ion channel pore

Fig. 6. Geometry and conductance of the ion channel pore.

Noise analysis

MD simulations of water permeation

Epilepsy-associated disease mutations V626L and T625N in GluA2

Fig. 7. Extent of steady-state activation and deactivation rate in epilepsy-associated mutants.

Auxiliary subunits tune the effects of V626L on receptor desensitization kinetics

Fig. 8. Auxiliary subunits tune the effects of the epilepsy-associated mutation V626L on AMPAR rates of entry and recovery from desensitization.

Discussion

Methods

Plasmids and strains

Protein expression and purification

Cryo-EM sample preparation and data collection

Cryo-EM image processing

Mass spectrometry sample preparation and protein digestion

LC–MS/MS analysis

Mass spectrometry data analysis

Model building, refinement, and validation

Whole-cell patch-clamp recordings

Noise analysis

System preparation for molecular dynamics simulations

Molecular dynamics simulation protocols

Molecular dynamics trajectory analysis

Statistics and reproducibility

Reporting summary

Supplementary information

Source data

Acknowledgements

Author contributions

Peer review

Peer review information

Funding

Data availability

Competing interests

Footnotes

Supplementary information

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases