Structural basis of GluK2 kainate receptor activation by a partial agonist

Abstract

Kainate receptors (KARs) belong to the family of ionotropic glutamate receptors (iGluRs) that regulate neurotransmitter release and excitatory synaptic transmission in the central nervous system. Despite their critical roles in synaptic signaling and disease, the detailed gating mechanisms of KARs are not completely understood. Here, we present cryo electron microscopy structures of homomeric rat GluK2 KAR in an unliganded apo state and in complexes with a partial agonist domoate. Partial agonist-bound GluK2 populates multiple conformations, including intermediate and desensitized states. Moreover, we demonstrate that the N-glycans at the amino-terminal domain (ATD)-ligand binding domain (LBD) extracellular domain interface modulate receptor gating properties by interfering with cation binding at the LBD dimer interface. Together, these studies provide insights into the unique gating mechanisms of KARs.

Introduction

Kainate receptors (KARs), along with α-amino-3-hydroxy-5-methylisoxazole-4-propionate receptors (AMPARs), N-methyl-D-aspartate receptors (NMDARs), and delta receptors, are subtypes of ionotropic glutamate receptors (iGluRs). These ligand-gated ion channels are widely expressed in the central nervous system where they mediate excitatory synaptic transmission and plasticity^1,2. KARs play crucial roles in high-level cognitive functions, such as learning and memory, and are implicated in various brain diseases³, including epilepsy^4,5 and schizophrenia^6–9. They are expressed pre-, post-, and extra-synaptically¹⁰, and also regulate neurotransmitter release, and network excitability^3,10–17. KARs are cation-permeable receptors assembled from five subunits, GluK1 to GluK5^18,19. Each subunit comprises four domains: amino-terminal domain (ATD), ligand binding domain (LBD), transmembrane domain (TMD), and intracellular C-terminal domain (CTD)¹⁰. The function of KARs is regulated by various ligands, including agonists, competitive antagonists, and allosteric modulators^20–22. Additionally, interacting proteins and auxiliary subunits such as neuropilin and tolloid-like (NETO) proteins play crucial roles in modulating KAR trafficking and gating^23–27.

Extensive studies over the past decades have highlighted the similarities and differences in the gating mechanisms and conformations of KARs and AMPARs^1,28,29. Notably, some KAR partial agonists exhibit distinct characteristics compared to AMPAR partial agonists, suggesting a unique mechanism of partial agonism in KARs. Generally, partial agonists bind to receptors but activate them less effectively than full agonists, even at saturating concentrations^20,30–32. iGluRs possess a bi-lobate LBD, where agonists bind and induce closure of the LBD cleft. In AMPARs, a strong correlation exists between the degree of agonist-induced LBD cleft closure and agonist efficacy. Studies have shown that AMPAR partial agonists stabilize non-conducting conformations of the LBDs, which still permit channel gating^33,34. In contrast, the degree of bi-lobe closure does not always directly correlate with agonist efficacy in KARs. For instance, in heteromeric GluK2/GluK5 receptors, the LBD bi-lobe conformation of GluK5 bound to the partial agonist AMPA is similar to that of the glutamate-bound GluK2 LBD³⁵. Another unique feature of KAR LBD conformations is their high degree of conformational flexibility, which distinguishes them from other iGluRs³⁶. This flexibility may contribute to the unique gating properties of KARs.

Particularly, some KAR partial agonists, such as (2S,3S,4S)-3-carboxymethyl-4-[(1Z,3E,5R)-5-carboxy-1-methyl-hexa-1,3-dienyl]-pyrrolidine-2-carboxylic acid (domoate), exhibit distinct kinetic properties. Previous studies have demonstrated that domoate generates characteristic large steady-state currents, along with slow deactivation and recovery from desensitization^37,38. It also produces a prolonged inhibition of GluK5-containing KARs³⁹. Crystal structures of the isolated GluK2 LBD complexed with domoate have shown that domoate stabilizes the LBD bi-lobe in a more open conformation compared to the glutamate-bound LBD⁴⁰. While various structures stabilized in various functional states has been determined^28,41–49, it remains unclear whether domoate and glutamate induce distinct conformational changes in the context of the full-length KAR, as well as how partial agonism is conserved or differs between KARs and AMPARs^33,50. Functionally, previous electrophysiological studies indicate that receptors activated by partial agonists show distinct intermediate states toward channel opening and desensitization compared to glutamate activated receptors^33,35,51,52.

Additionally, earlier studies have shown that external monovalent ions also act as allosteric modulators of KARs, a feature not conserved in other iGluRs^53–59. Ions bind to the LBD dimer interface, stabilizing dimer formation^60–64. Specifically, cations regulate the responsiveness to a range of full and partial agonists⁶⁵. Furthermore, N-linked glycosylation on KARs^66–70 can significantly alter the activity of KARs as well^43,68,71. Altered N-glycosylation in KARs and other iGluRs contributes to abnormal neuronal signaling and connectivity, as observed in schizophrenia^72–74. These findings highlight ions and N-glycosylation as critical modulators for KARs.

Here, we present structures of the intact GluK2 KAR in the apo state and in complex with domoate, capturing receptors in desensitizing intermediate and fully desensitized states. These structures reveal the conformational changes of GluK2 along the desensitization/recovery pathway, highlighting the differences between full and partial agonist activation. Additionally, we uncover functional-state-dependent interactions of N-glycans at the ATD-LBD dimer interface. Our data demonstrate that the glycan mildly modulates KAR gating by interfering with cation binding at the LBD dimer interface. Together, our studies provide insights into the gating mechanisms of GluK2 KAR activated by a partial agonist and the detailed modulation of GluK2 by N-glycans.

Results

Overall structure of unliganded apo GluK2 KAR

To better understand the conformational changes, we first determined the apo/resting state structure of the full-length rat GluK2 using cryo-electron microscopy (cryo-EM) (Extended Data Fig. 1, Extended Date Table 1, PDB ID: 8F0O). The structure is comparable to the glutamate-bound, four-fold symmetrical GluK2 structure (PDB ID: 5KUF)⁴² with a root mean square deviation (RMSD) of 1.53 Å (Extended Data Fig. 2). In this conformation, the LBD bi-lobes in the subunit AC pair are tightly closed, resembling the glutamate bound GluK2 LBD structures^75–77, while the LBDs in the BD subunit pair 11.2–11.7° are more open, similar to the domoate bound GluK2 LBD conformation⁴⁰. Previous studies have proposed that agonist binding on one or two subunits within the tetrameric receptor is sufficient to desensitize KAR and AMPARs^78–82. Moreover, recovery from desensitization becomes significantly slower once receptors are fully desensitized^78,83,84. Due to the relatively low resolution of the LBD layer, ligand densities are not well observed. Based on the LBD bi-lobe closure and overall conformation, this structure likely represents either a desensitized state triggered by residual agonist binding or an apo state recovering from desensitization. We therefore named this class as “no added ligands, desensitized” to distinguish it from the later determined apo state.

To eliminate the potential contamination by ambient agonists, we further optimized our protein preparation protocols. We next expressed rat GluK2 wild-type (GluK2_WT) in HEK293S GnTl⁻ cells, solubilized proteins in the presence of DNQX, a competitive KAR/ANPAR antagonist, to reduce desensitization^46,85. Proteins on the affinity column chromatography were washed extensively with buffers containing 1 mM DNQX and high salt concentration (500 mM), shifting the receptors to a low agonist affinity state⁵³. The detergent was also exchanged from DDM to digitonin during the strep-tactin affinity purification. The 3D classification yielded four distinct classes, including the major class displaying the ATD, LBD, and TMD layers (Extended Data Fig. 3). The structure of the major class was solved to 4.0 Å, as estimated by the 0.143 Fourier shell correlation (FSC) gold standard (Extended Data Fig. 4). We further conducted refinement focused on the ATD and the LBD-TMD layers separately, which improved the resolution to 3.3 Å and 3.8 Å, respectively. In this structure, the LBD layer exhibits an asymmetric organization, with two LBDs forming a dimeric configuration on one side and a disrupted dimer (with two separated monomeric LBDs) on the other side. This contrasts with the traditional dimer-of-dimers conformation observed in the potentiator-bound^46,47 or antagonist-bound⁴² GluK2 (Fig. 1 A, B). This asymmetric conformation is rather reminiscent of previously observed structures of glutamate-bound GluK2⁴⁷, a competitive antagonist bound GluK2 complexed to NETO2⁸⁵, and antagonist bound GluK3^43,44.

Figure 1. Cryo-EM structure of unliganded apo GluK2 and comparison of the LBDs between unliganded apo and agonist or antagonist bound GluK2 LBDs.

(A) Cryo-EM map of the tetrameric unliganded apo GluK2 viewed perpendicular to the membrane. The A–D chains are colored in cyan, yellow, green, and purple, respectively. (B) Top view of the unliganded apo GluK2 LBD (top) and the antagonist bound GluK2 LBD (bottom). (C) Comparison of the LBD D1-D1 interface in the apo GluK2, with subunits colored as in panel A, superimposed onto the BPAM344-bound GluK2 structure (PDB ID: 8FWQ)⁴⁶ shown in light gray (left). A close-up view of the D1-D1 interface highlights salt bridges, depicted as black dashed lines (right). (D) Cryo-EM models of individual LBDs in the unliganded apo GluK2 are superimposed onto the crystal structure of isolated GluK2 LBD bound to glutamate (PBD ID: 2XXR)⁷⁷ in light gray. The D1 lobes are used for the superimposition of LBDs. The LBDs adopt more open conformations in the unliganded apo state compared to the glutamate-bound LBD, as highlighted by the arrows. (E) Cryo-EM models of individual LBDs in the unliganded apo GluK2 are superimposed on the cryo-EM reconstruction of DNQX-bound GluK2 (dark gray, PBD ID: 7F5B)⁸⁵. The LBDs adopt more closed conformations in the unliganded apo state compared to the competitive antagonist-bound LBDs, as indicated by the arrows. (F) Ligand binding site comparisons. A zoom-in view of the EM density map of the ligand binding site in the unliganded apo state (left) compared to the same site in the competitive antagonist DNQX bound GluK2 (right)⁸⁵, highlights the agonist binding pocket formed by residues E378, E440, Y488, Y764.

To date, no apo KAR LBD structure has been solved, likely due to its conformational flexibility⁴⁹. Previously determined structures of intact KARs in apo states have been resolved only at relatively low resolutions, down to 7.5 Å or lower^44,45. GluK2 structure in the absence of orthosteric ligands, but with a potentiator, BPAM344, has also been reported⁴⁶. Our 3.8 Å structure of apo GluK2 overcomes the limitation. BPAM344-bound GluK2 structures demonstrated that the potentiator stabilizes the LBD dimer association, preventing the D1-D1 LBD rupture typically associated with receptor desensitization^46,86. In our apo GluK2 structure, the LBD dimer forms interactions between D1 lobes in the BC subunits in the absence of the potentiator, similar to those observed in the crystal structure of agonist-bound isolated GluK2 LBD forming a dimer⁷⁷. The D1-D1 lobes in the apo-state structure is superimposable with those in the BPAM344-bound GluK2 structure⁴⁶ (Fig. 1C) rather than those observed in desensitized AMPARs, where the D1-D1 lobes of the LBD dimers are dissociated^50,87–90.

Importantly, in the apo GluK2 structure, the LBD clamshells are significantly (19.3–24.0°) more open than the glutamate-bound LBDs but 3.9–8.4° less open than the antagonist bound LBDs (Fig. 1D, E), in contrast to the closed LBD conformations in the “no added ligands, desensitized” GluK2 structure described above (Extended Data Fig. 2 C). While strong ligand densities are present in the DNQX-bound GluK2-NETO2 structures solved to 3.9 Å⁸⁵, no clear density of ligands was observed in the apo GluK2 structure resolved to 3.8 Å. This indicates that we successfully washed out DNQX during the purification, resulting in no or negligible ligands bound to the receptors (Fig. 1F, Extended Data Fig. 5 a–d). The ion channel pore resembles that of the glutamate-bound GluK2, with a narrow point at Thr652, and the channel remains closed (Fig. 2A, C). We concluded that this structure likely represents an unliganded apo state.

Figure 2. Ion channel structure of apo and domoate-bound GluK2 KARs.

(A, B) Pore profiles of the receptors in the apo and glutamate-bound desensitized GluK2 structure (PDB ID: 5KUF)⁴² (A), and the domoate-bound GluK2 (B) structures are shown, with two M3 helices and their pore-lining residues highlighted. The pore radii were calculated using HOLE (http://www.holeprogram.org)¹¹⁸. Regions with a pore radius less than 1.15 Å are colored in red, water-accessible sections with a pore radius between 1.15 Å and 2.5 Å are colored in green, and sections with a radius greater than 2.5 Å are colored in purple. (C) Pore diameter comparison between the apo and glutamate-bound desensitized GluK2 (PDB ID: 5KUF)⁴². (D) Pore diameter of domoate-bound GluK2 KARs.

(Source Data Figure 2)

Clearly, this apo state GluK2 conformation is distinct from the classic apo state structures of other iGluRs as well as antagonist bound KAR structures. Previous molecular dynamics (MD) simulation and mutagenesis analysis of isolated GluK2 LBD indicate that the GluK2 LBDs are conformationally flexible, capable of adopting both open and closed bi-lobe conformations in the absence of agonists³⁶, distinguishing KARs from other iGluR subfamilies. The LBD bi-lobe closure is likely associated with the LBD dimer dissociation and in-plane rotation of the LBDs. Given this information, our apo GluK2 structure suggests that the LBD D1-D1 rupture and complete dissociation of the LBD dimer at the physiological NaCl concentration (where cation binding sites of receptors are not saturated)⁶⁰, coupled with LBD bi-lobe closure, could occur in the absence of ligands. However, additional information, including the ratio of receptors adopting the classical two-fold and non-two-fold conformations in the physiological membrane, as well as kinetic data, is required to confirm the mechanism.

Activation of GluK2 KAR by partial agonist.

KARs desensitize rapidly within 10 ms in response to a saturating concentration of glutamate. Therefore, glutamate-bound KAR structures are considered to represent desensitized conformations. To gain deeper insights into the gating mechanisms, we utilized domoate, a highly potent KAR partial agonist^38,39,91,92. We performed whole-cell patch-clamp electrophysiological recordings on GluK2_WT expressed in HEK293T cell, and compared the peak current amplitudes and desensitization kinetics of receptors activated by either domoate or glutamate (Fig. 3A–C). Consistent with earlier observations³⁷, domoate (10 μM) evoked GluK2 current amplitudes were approximately three times smaller than those evoked by glutamate (10 mM) at a holding potential of −70 mV. The mean GluK2 peak current density was 960 ± 60 pA/pF (n=13) for glutamate and 330 ± 40 pA/pF (n=9) for domoate. The desensitization of domoate-activated GluK2 (τ_{desensitization}: 40 ± 2 ms, n = 9) was significantly slower compared to that of glutamate-activated GluK2 (τ_{desensitization}: 8 ± 0.3 ms, n = 13). Furthermore, the steady-state to peak current ratio was 45 ± 5 % (n=9) for domoate-evoked currents, whereas no steady-state current was observed for glutamate-evoked GluK2_WT (n=13). These observations suggest that domoate stabilizes receptors in intermediate states that are likely distinct from those of glutamate-activated receptors.

Figure 3. Function and structure of GluK2 KAR activated by partial agonist domoate.

(A) The chemical structures of L-glutamate and domoate. (B, C) Representative whole-cell currents recorded at a holding potential of −70 mV from HEK 293T cells expressing wild-type rat GluK2 in response to a 1-second application of L-glutamate or a 10-second application of domoate. (D) Cryo-EM map of the tetrameric GluK2 in complex with domoate, viewed perpendicular to the membrane, with subunits colored as in Figure 1. This panel shows the EM map for ATD, LBD, and TMD layers. (E) Top view of LBD models in the same reconstructions as in panel D.

(Source Data Figure 3)

Partial agonist bound KARs adopt multiple conformations.

To understand the conformational alternations induced by domoate, we next determined the domoate-bound GluK2 KAR structures. To increase protein stability and resolution, we used the previously described rat GluK2 with mutations that stabilize tetramer formation (GluK2_{I567V/C576V/C595S})^42,45,93. We observed that GluK2_{I567V/C576V/C595S} retains basic functional properties similar to GluK2_WT (Extended Data Fig. 6). We expressed GluK2_{I567V/C576V/C595S} proteins in Spodoptera frugiperda (Sf9) insect cells using the EarlyBac method⁹⁴. We successfully purified the tetrameric receptors using Strep-tactin Sepharose in n-dodecyl-β-D-maltoside (DDM), which was subjected to single-particle cryo-EM. The reported EC₅₀ of domoate for GluK2 is 0.29 ± 0.03 μM⁹⁵. Thus, we incubated the purified proteins with a high concentration (5 mM) of domoate prior to grid freezing. The 3D classification revealed four classes of domoate-bound GluK2 KAR complexes (Extended Data Fig. 7, Extended Date Table 1) with overall resolutions of 3.9–4.3 Å, without applying symmetry (Extended Data Fig. 4 M–X). The 3D reconstructions of all classes showed a typical KAR tetramer assembly with ATD, LBD, and TMD layers (Fig. 3D).

Two major classes exhibited nearly four-fold rotational symmetry of the LBD layers in the presence of domoate, generated by large LBD rigid-body movements. Because these conformations are similar to the glutamate-bound GluK2 structure⁴² with the ion channel pores tightly closed, we consider they represent domoate-bound fully/deeply desensitized conformations, which we have named “desensitized 1” and “desensitized 2”. In contrast, two minor classes, including an asymmetrical conformation and a four-fold symmetrical conformation, represent intermediate states between the active and fully desensitized states, or during recovery from desensitization. We have named these two minor classes “intermediate 1” and “intermediate 2”. The particle distributions for the desensitized 1, desensitized 2, intermediate 1, and intermediate 2 classes were 30.6%, 30.9%, 12.6%, and 22.3%, respectively (Extended Data Fig. 7).

The local resolution of the ATD layers in all four classes was higher than those of the LBD and TMD layers likely due to their conformational stability (Extended Data Fig. 4 M–P). The conventional clamshell structure of ATDs has upper and lower lobes (R1 and R2). As previously shown, AB and CD subunits formed ATD dimers in all four configurations (Fig. 3D, Fig 4A, B). Comparing ATD layers revealed that the ATDs of all classes are superimposable onto the glutamate-bound fully desensitized state (PBD ID: 5KUF)⁴² (Extended Data Fig. 8A). Superimposing TMDs and analyzing ATD position in the full-length structures reveals a 12 Å swing compared to the glutamate-bound GluK2 (Fig. 8B). The ATD layers were at least 7.2 Å away from the LBD layers in the absence of allosteric modulators and auxiliary subunits, similar to previously reported homomeric KAR structures^41–43. Unlike AMPARs^96,97 and NMDARs^98–100, ATD swings do not affect KARs gating, particularly desensitization or desensitization recovery because there are no functional state-dependent ATD–LBD interactions. In contrast, the LBD layers in each class show distinct conformations (Fig. 3E, Fig 4E). TMD layers in all classes create pseudo-four-fold symmetry with minor differences (Fig. 4F). Therefore, we will focus primarily on the LBD-TMD layers in the following sections.

Figure 4. Cryo-EM structure of domoate-bound GluK2 KAR in the intermediate 1 state.

(A) Side view of the domoate-bound GluK2 in the intermediate 1 conformation, colored as in Figure 1. Domoate molecules are shown in magenta. (B) Top view of the ATD subunit arrangement in the intermediate 1 class, as viewed from the extracellular side. (C) Comparison of the apo (gray) and domoate-bound (cyan, subunit A) LBDs in the intact GuK2 structures. The domoate molecule is colored in light pink. The domoate-bound GluK2 LBD adopts a more closed cleft conformation compared to the apo GluK2 LBD. (D) Zoom-in view of the domoate binding site. (E) Top view of LBDs in the intermediate 1, active (PDB ID: 9B35)⁴⁷, and desensitized 1 states. In this panel, the ligands have been removed from the structures to highlight the conformation of the LBDs. (F) Top view of the TMD in the intermediate 1 conformation.

Figure 8. The N275 glycan modulates channel properties Na⁺ dependently.

(A) Representative whole-cell recordings from HEK293T cells expressing GluK2_WT and GluK2_T277A, exposed to extracellular Na⁺ concentrations ranging from 0 to 160 mM. Receptors were activated by 10 mM glutamate for 1 second (black bar).(B) Mean desensitization rate constants for GluK2_WT and GluK2_T277A at 40 mM (4.6 ± 0.2 ms, n=5 for WT, 4.2 ± 0.2 ms, n=5 for T277A), 80 mM (6.7 ± 0.3 ms, n=5 for WT, 5.9 ± 0.2 ms n=6 for T277A), and 160 mM Na⁺ (7.6 ± 0.3 ms, n=8 for WT, 6.5 ± 0.2 ms, n=5 for T277A). (C) Mean deactivation rate constants for GluK2_WT and GluK2_T277A at 80 mM (1.7 ± 0.1 ms, n=5 for WT, 2.9 ± 0.1 ms, n=11 for T277A) and 160 mM Na⁺ (3.9 ± 0.3 ms, n=6 for WT, 3.6 ± 0.1 ms, n=7 for T277A). (D) Two-pulse glutamate (10 mM) recovery experiments for GluK2_WT and GluK2_T277A at 80 mM (n=5 for WT, n=5 for T277A) and 160 mM Na⁺ (n=5 for WT, n=7 for T277A). The amplitude of the second glutamate response is normalized to the first application. The continuous line represents a fit with the Hodgkin-Huxley equation. (E) Comparison of recovery time constants (τ) from desensitization (derived from Hodgkin-Huxley equation fits) for GluK2_WT and GluK2_T277A at 80 mM (2.1 ± 0.1 s, n=5 for WT, 1.9 ± 0.2 s, n=6 for T277A) and 160 mM Na⁺ (1.3 ± 0.1 s, n=5 for WT, 1.1 ± 0.1 s, n=7 for T277A). Data are presented as mean ± SEM (error bars). The number of biological replicates (n) is indicated for each condition and is represented by a circle in bar graphs. Statistical significance was determined using a two-sided, two-sample t-test; p-values are shown in each panel, with significance defined as p<0.05. P values not shown in the figure panel B WT comparing 80 mM and 160 mM Na⁺ p=0.07003 and WT vs T277A at 40 mM p=0. 181. Panel C, GluK2_WT vs GluK2_T277A at 160 mM Na⁺, p=0.302.

(Source Data Figure 8)

Domoate binding induces LBD bi-lobe closure and LBD layer rearrangement in full-length GluK2.

We observed clear densities for domoate within the agonist binding pocket (Fig 4C, D) of all four classes, except for subunit D in the intermediate 1 conformation (Extended Data Fig. 5e–l). We fitted the domoate models using the crystal structure of domoate-bound GluK2 LBD (PDB ID: 1YAE)⁴⁰ as a guide, although the precise ligand binding modes were not clearly resolved due to resolution limitations. Previous structural studies have shown that domoate binding partially closes the LBD cleft, with a D1–D2 closure degree of 24–25°, compared to the glutamate bound LBD with a D1–D2 closure degree of 33–36°^40,49,101. In our structures, domoate closes LBD similar to the crystal structure (PDB ID: 1YAE)⁴⁰, with an RMSD of 1.6–2.2 Å. In intact GluK2 structures, we observed minimal D1–D2 twisting motions, with D1–D2 lobes twists ranging from 0.4 to 1.5 Å, in contrast to the previously suggested movements associated with partial agonism¹⁰².

Comparing domoate-bound GluK2 structures showed numerous domain rearrangements. The intermediate 1 conformation displayed an asymmetric organization similar to the previously reported asymmetrical architectures of KARs, including the agonist-bound GluK2¹⁰³ and the competitive antagonist-bound GluK2-NETO2 and GluK3 structures^43,44,85. Subunit D in this conformation does not exhibit discernible ligand densities in its agonist binding pocket at this resolution, whereas the other three subunits show clear densities (Extended Data Fig. 5 e–h). The BC pair’s LBDs dissociate when subunit B’s LBD is rotated 110° anticlockwise in the horizontal plane compared to the open-state GluK2 structure (PDB ID: 9B35)⁴⁷. These LBD are stabilized in a deep, fully desensitized conformation, similar to the glutamate-bound GluK2 structure (PDB ID: 5KUF)⁴² (Fig. 5A). In contrast, the AD pair in the intermediate 1 conformation forms a dimer with separated D1–D1 lobes, resembling the desensitized conformation observed in AMPARs^50,104 rather than the open-state configuration with intact D1–D1 lobes (Fig. 4E, 5A)^{33,89,90,105–107}. Viewed from the side, we observe the entire LBDs of subunit A and B are rolled up by 8° and 7°, respectively, while the LBD of subunit C is rolled down by 8° (Fig. 5 B–E), further contributing to the conformational asymmetry. Overall, this class likely represents an intermediate state that is either undergoing desensitization or recovery from desensitization. The intermediate 2 class displayed quasi-four-fold symmetrical LBDs with domoate molecules bound to all four subunits (Fig. 5A). Perpendicular to the membrane, subunit D’s LBD rotated upward by 11°, whereas the D1 domain of subunit C showed minimal conformational changes, with a 4° downward rotation compared to the glutamate-bound GluK2 structure (Fig. 5B–E), resulting in minor conformational asymmetry.

Figure 5. Conformational variability of domoate-bound GluK2.

Four classes of the domoate-bound GluK2 termed intermediate 1, intermediate 2, desensitized 1, and desensitized 2 conformations are shown. Subunit A, B, C, and D of domoate-bound GluK2 are colored in cyan, yellow, green, and purple, respectively. Structures of domoate-bound GluK2 are aligned to the glutamate-bound GluK2 structure (PDB ID: 5KUF, gray)⁴² in panel A, B and D. (A) Extracellular view of the LBD layers for domoate-bound GluK2 structures. The subunit arrangements of the LBDs are represented in cartoon form. (B, D) Side view of the A–C or B–D subunit pairs, depicting are the conformational changes in the LBDs compared to the glutamate-bound GluK2. (C, E) Schematic representation of the conformational changes in domoate-bound GluK2 structures.

The desensitized 1 class has a four-fold symmetrical structure of tetrameric LBDs in the top view. This class adopts conformation nearly identical to the glutamate-bound GluK2 structure (PDB ID: 5KUF)⁴² with minor differences, including the small-angle (2.5–6°) further in-plane rotations of the LBDs (Fig. 5A) and a slight (5.2 Å) swing of the ATDs (Extended Data Fig. 8). Observing TMD-LBD layers perpendicular to the membrane revealed slight LBD domain rotations (~2°) compared to the glutamate-bound GluK2 structure⁴². This structure suggests that both full and partial agonists can lead receptors to adopt similar fully desensitized states at equilibrium, albeit via distinct kinetics or desensitization pathways. Lastly, the desensitized 2 conformation also displayed approximately four-fold symmetry. When viewed from the extracellular side, all four LBDs underwent an anticlockwise rotation by 2–15° more than the glutamate-bound GluK2 structure (Fig. 5A). Domoate binding caused the LBD D2 lobes in the AC pair to rotate downward, while the D1 and D2 lobes in the BD pair rotated upward by 8°, resulting in the D2 lobes of the BD pair moving away from the membrane (Fig. 5B–E). Consequently, the AC and BD pairs created unique LBD arrangements. Only domoate-bound GluK2 showed these conformations, not glutamate.

Desensitization ring formation.

Previous studies have demonstrated that KARs possess a unique ring-like motif comprised of G and E helices (Fig. 6A), which defines the receptor desensitization and recovery from desensitization^42,45. Each domoate-bound GluK2 structure showed distinct arrangements of the G/E helices and M3-LBD linkers, suggesting that they may represent different functional states (Fig. 6B–F). As described, the intermediate 1 structure exhibits an asymmetric LBD arrangement, consisting of one dimer and two monomers. The upward motion of the LBDs of subunits A and B, combined with the downward rotation of subunit C, rearranged the EG helices (Fig. 5), Consequently, the EG helices are not positioned close enough to form inter-subunit interactions (Fig. 6C, D). The intermediate 2 class adopts a four-fold symmetrical conformation. However, a closer investigation revealed that the minor downward rotation of LBD D2 in subunit C and the 10° upward rotation of LBD in subunit D generated a unique asymmetrical gating ring, while two desensitized conformations exhibited symmetrical ring formation (Fig. 6C, D). The G helices of the AC subunits moved 2 Å closer to each other, while those of the BD subunits moved apart by 4 Å. Consequently, there is no interaction between subunit A–B and B–C due to the movement of subunit B in this conformation.

Figure 6. Desensitization ring in GluK2.

(A) Top and side views of the LBD layer in the domoate-bound desensitized state conformation, highlighting the desensitization ring formed by the E/G helices of subunits A–D, colored in cyan, yellow, green, and purple, respectively. (B) Inter-protomer electrostatic interactions between the G and E helices in the domoate-bound desensitized 1 and 2 conformations. (C) Desensitization ring in the open/active GluK2 structure (PDB ID: 9B35)⁴⁷ (disrupted), and domoate-bound GluK2 structures, depicting are conformational changes in the LBDs (and E/G helices). (D) Schematic representation of the desensitization rings in GluK2 structures. The gray dash line represents the position of the G helices of the subunit A. (E) Plot of the A–C versus the B–D distances measured between Cα of Ser670, comparing the domoate-bound intermediate and desensitized states. (F) Side views of the LBD-M3 linkers comparing the intermediate 1 versus open GluK2 (left), and desensitized 2 versus glutamate-bound desensitized GluK2 (PDB ID: 5KUF)⁴² structures. Tension of the LBD-M3 linkers measured by the distance between the two Ala671 Cα (spheres) is a major determinant for the opening or closing of the ion channel gate.

In contrast, the domoate-bound GluK2 desensitized 1 class, similar to the glutamate-bound desensitized GluK2 structure (PDB ID: 5KUF)⁴², contains a two-fold symmetrical desensitization ring when viewed from the side. This ring features alternating staggered helices stabilized by numerous inter-protomer interactions (Fig. 6C, D). Lastly, the desensitization ring in the desensitized 2 conformation also displayed two-fold symmetry but formed a distinct staggered ring. Due to the downward rotation of the AC pair and the approximately 10° upward rotation of the BD pair, the position of the AC and BD pairs were swapped (Fig. 5B–E). Regardless, the ring formation was stabilized by several inter-subunit contacts, including several electrostatic interactions in this conformation (Fig. 6B), which appear to stabilize the overall desensitized conformations. Together with the earlier mutagenesis analysis showing that the LBD D2 lobe controls desensitization and recovery from desensitization¹⁰⁸, and that disrupting inter-subunit interactions in the desensitization ring accelerates desensitization recovery⁴², our results support the significant role of the desensitization ring.

Ion channel in domoate bound GluK2 structures.

The TMD consists of three transmembrane helices, M1, M3, and M4, consistent with previously reported structures of the iGluR superfamily (Fig. 4A, F)¹. The M3 helices line the ion permeation pathway at the extracellular side and function as an activation gate, while the M1 and M4 helices surround the M3 helices and face the membrane lipids. We observed densities for the side chains of the conserved SYTANLAAF motif consistent with previous reports (Extended Data Fig. 5 B)^42,46,47,85. In our structures, the M2 helices and the M1–M2 loops, which observed in some structures^46,47,103, were not well resolved due to the conformational flexibility or resolution limitation. The central pore of the intermediate 1 structure was the widest in all domoate-bound GluK2 structures (Fig. 2B, D). Because of the asymmetrical LBD-M3 linker positions due to the remaining LBD dimer of subunit AD, the ion channel pore is asymmetrical with a mildly expanded ion channel pathway. Compared to the channels in the apo and glutamate-bound desensitized GluK2 (PDB ID: 5KUF)⁴² structures (Fig. 2A), the pore is noticeably wider throughout the ion channel gate, particularly around Thr660 (Fig. 2B). The pore radius at Thr660 was approximately 1.4 Å, which corresponds to the radius of a water molecule. However, the pore radius remains significantly narrower than that of fully open iGluR structures with pore dilation, where the radii are approximately 2.5 Å in KAR⁴⁷ and greater than 3 Å in AMPAR¹⁰⁴ and NMDAR¹⁰⁹. This suggests that the ion channel in the intermediate 1 state is at the threshold between open and closed (Fig. 2B, D). Similarly, the central pore in the intermediate 2 conformation has a narrow point at Thr660 that is slightly narrower than 1.4 Å, suggesting that the ion channel is on the verge of opening. In contrast, in the domoate-bound desensitized state conformations, the ion channels are more tightly sealed at the bundle crossing of the M3 helices at Thr652, Ala656, and Thr660, although the pore radius at Thr652 is not as narrow as in the apo or glutamate-bound desensitized conformation (PDB ID: 5KUF)⁴²(Fig. 2B).

Asn275 glycan is positioned near the anion and cation binding sites at the LBD dimer interface.

Rat GluK2 contains nine putative N-linked glycosylation sites (Uniprot P42260). Previous structures demonstrate densities for N-linked glycosylation carbohydrates^{42,44,45,47,110,111}. In the EM maps, we observed residual densities of four N-linked glycans at Asn275, Asn378, and Asn412 on the ATD, and Asn430 on the LBD of the AC subunits, while the BD subunits showed only two evident glycan densities at Asn378 and Asn412. As reported earlier⁴³, we observed two N-glycans at Asn275 and Asn430 positioned at the interface of the ATD–LBD, where they mediate the inter-domain interactions (Fig. 7A). We identified that proteins expressed in Sf9 and HEK293S GnTl⁻ cells have four to six different glycan forms of N-glycans, with the most abundant subtypes: 8 hexoses and 2 N-acetylhexosamines [HexNAc(2)Hex(8)]; and 9 hexoses and 2 N-acetylhexosamines [HexNAc(2)Hex(9)], as determined by capillary column liquid chromatography with tandem mass spectroscopy (LC-MS/MS) (Fig. 6B). Note that the HEK293S GnTl⁻ cell line cannot synthesize complex N-glycans due to its lack of N-acetylglucosaminyltransferase I (GnTI) activity. Compared to recombinant proteins expressed in Sf9 and HEK cells, their natural equivalent creates more complex and heterogeneous glycans. This MS analysis indicates that bulky, complex glycans are present at the ATD–LBD domain interface, while the glycan densities are largely invisible by cryo-EM due to their flexibility. Our EM maps clearly assign two N-acetylhexosamines (HexNAc) and three hexoses (Hex) of the N275 glycan at the ATD–LBD interface in the intermediate 1 and desensitized conformations (Fig. 7C), as in GluK2 and GluK3 structures^42,43. However, the N275 glycan forms distinct intra- or inter-protomer interactions in these two conformations.

Figure 7. N-Glycans on ATD interacts with the anion and cation binding sites at the LBD dimer interface.

(A) EM density maps of N-glycans at Asn275 in the ATD of the intermediate 1 and desensitized conformations. N275 glycans are colored in red. The domoate-bound intermediate 1 structure contains the LBD dimer, with the N275 glycan forming inter-domain, inter-protomer interactions with residues on the LBD D1 lobes of subunit A and D (left). The desensitized GluK2 LBDs form an approximate four-fold symmetrical conformation with the N275 glycan forming an intra-protomer, inter-domain interactions (right). (B) Quantification of N-glycans on purified GluK2 KARs expressed in Sf9 insect cells and HEK293S GnTl⁻ mammalian cells using LC-MS/MS. (C) Cryo-EM density for resolved Asn275 and Asn430 glycans in the intermediate 1 structure. (D) Amino acid sequence alignment of GluK1–5 and GluA1–2. The conserved N275 residue is marked with (●), and conserved residues contributing to the anion, and cation binding in GluK1-GluK5 KARs are marked with (▲). (E) Structural alignment of the intermediate 1 and desensitized 1 structures, focusing on the N275 glycan at the ATD-LBD interface. (F) Top and side views of the LBD dimer, with the N275 glycan positioned at the LBD dimer interface in the intermediate 1 conformation. (G) Residues interacting with the glycan, as well as residues involved in cation and anion binding, are shown. The cation and anion binding sites are highlighted in orange and gray, respectively.

Sequence alignment shows that all KARs and AMPARs contain the N275 glycan (Fig. 7D). The N275 glycan is surrounded by numerous polar and bulky residues, which appear to stabilize the glycan between the ATD lower domain, the ATD–LBD linker, and the LBD upper domain. While the N275 glycan in the intermediate 1 conformation faces downward, the same N275 glycan in the desensitized conformation faces upward, demonstrating function-dependent placement (Fig. 7E).

The LBD dimer of the AD subunit in the intermediate 1 conformation has separated D1–D1 lobes, unlike the active state GluK2 conformation (PDB ID: 9B35)⁴⁷. For example, the distances between the Arg775 Cα in the intermediate 1 and active conformations are 20.7 Å and 13.2 Å, respectively. Interestingly, the N275 glycan at the middle of the LBD dimer interacts with both protomers approximately two-fold symmetrically, mediating inter-subunit and inter-domain interactions in the intermediate 1 conformation (Fig. 7F). Moreover, the spatial restriction facilitates the positioning of the N275 glycan near the top of the anion and cation binding sites located at the D1–D1 interface (Fig. 7G) and the N275 glycan electrostatically interacts with the allosteric ion binding sites Met770, Arg775, and Asp776. Thus, we anticipated that the N275 may affect sodium and chloride binding to the LBDs. Since the N275 glycan forms only intra-protomer interactions in the four-fold desensitized conformation, unlike the active GluK2 structure (PDB ID: 9B35)⁴⁷, where it is excluded from the D1–D1 interface, we propose that the N275 glycan–LBD interactions are function-dependent.

N-glycosylation at Asn275 modulates receptor channel properties.

As cations and anions stabilize active LBD dimer formation and modulate channel gating, we hypothesized that the N275 glycan affects channel properties in a Na⁺ concentration-dependent manner. To test this hypothesis, we recorded whole-cell patch-clamps of wild-type and glycosylation knockout (KO) mutant receptors, and evaluated how the N275 glycan affects KAR gating. We then examined channel properties at various extracellular Na⁺ concentrations to assess the Na⁺-dependent modulation. To generate an N275 glycan KO mutant, threonine 277 was mutated to alanine, disrupting the consensus N-glycosylation sequence (N-X-S/T where X≠ P)^43,52. The receptor currents were evoked by 10 mM glutamate. N275 glycan KO mutant (GluK2_T277A) showed a 16 % higher peak amplitude than GluK2_WT (Extended Data Fig. 9 A, B). GluK2_T277A also shifted the glutamate dose-response curve. The EC₅₀ values for GluK2_WT and GluK2_T277A were 180 ± 20 μM and 120 ± 20 μM, respectively, indicating that the N275 glycan knockout mutation reduces the EC₅₀ by approximately 40% (Extended Data Fig. 9 C–E). Additionally, GluK2_T277A exhibited approximately 30% and 20% faster desensitization and deactivation rates, respectively, as well as a 36% faster recovery from desensitization compared to GluK2_WT (Extended Data Fig. 9F–L). This is consistent with prior reports^43,71.

Earlier studies have demonstrated that extracellular Na⁺ greatly affects KAR desensitization, with receptor desensitization occurring after Na⁺ unbinding^42,62. We further investigated whether the N275 glycan modulates channel properties in a Na⁺ concentration-dependent manner. Previous research indicates that the large impermeant cation, N-methyl-D-glucamine (NMDG⁺), does not bind to the cation binding sites on KARs due to steric hindrance¹¹². Therefore, we replaced extracellular Na⁺ with NMDG⁺ and analyzed the channel properties of GluK2_WT and GluK2_T277A. Receptors were activated by 10 mM glutamate under three different Na⁺:NMDG⁺ ratios ([Na⁺:NMDG⁺] = [160:0], [80:80], and [40:120] mM), with all solutions maintaining a total of 160 mM Cl⁻.

Both GluK2_WT and GluK2_T277A exhibited Na⁺ concentration-dependent responses (Fig. 8A). Across all Na⁺ concentrations, the GluK2_T277A mutant showed increased mean current amplitude compared to GluK2_WT, with increases of 44%, 107%, and 150% at 160 mM, 80 mM, and 40 mM Na⁺, respectively (Fig 8A). GluK2_T277A exhibited 14% and 9% faster desensitization rates at 160 mM and 80 mM Na⁺, respectively, compared to GluK2_WT, but no significant difference at 40 mM Na⁺ (Fig. 8B). In contrast, the deactivation and desensitization recovery rates of GluK2_T277A were similar to those of GluK2_WT, with the exception of a minor difference in τ_deactivation at 80 mM Na⁺ (Fig. 8C–E). Overall, our observations show that the N275 glycan decreases GluK2 sensitivity to glutamate and slows desensitization in a Na⁺-concentration-dependent manner, likely due to slower Na⁺ unbinding. In contrast, modulation of deactivation and desensitization recovery by the N275 glycan occurs independently of Na⁺ concentration.

Discussion

Here, we unveiled the structure of GluK2 both in the absence of ligands and in complex with the partial agonist domoate. First, we determined a ligand-free apo GluK2 KAR structure, which exhibits an asymmetric arrangement with the LBD bi-lobe ~24.0° more open and ~8.4° more closed compared to the glutamate- and DNQX-bound LBDs, respectively. Unlike the energetically favorable LBD bi-lobe open conformations observed in apo AMPARs and some apo NMDARs^98,113–116, earlier computational studies showed that isolated apo GluK2 LBD exhibits extreme flexibility and preferentially adopts closed-cleft conformations³⁶. Since the LBDs dimerize back-to-back through their D1 lobes, closure of the LBDs may leads to separation of the D1 lobes, presumably resulting in the large in-plane rotation of the LBDs like we observed in the subunit AD pair of our apo-state structure. The apo GluK2/GluK5 KAR structure determined by cryo-EM at 7.5 Å⁴⁵ and the apo GluK3 structure analyzed by negative stain data (resolution not described)⁴⁴ showed the two-fold symmetric conformation of LBD layers. However, relatively low-resolution data suggests a high level of conformational heterogeneity and flexibility of KARs, indicating the possible co-existence of multiple conformations in the apo-state. Our apo GluK2 structure, along with previous reports, suggests that at least a subset of GluK2 may adopt asymmetric arrangements rather than the traditional dimer-of-dimer conformation. This observation is consistent with functional data indicating that, at physiological ion concentrations, the allosteric ion-binding sites on GluK2 LBDs are not fully saturated and that approximately one-third of the receptor population remains inactive⁶⁰. Another possibility is that this structure represents an inactive state induced by our purification method. Although we attempted to remove potential endogenous ligands bound to the receptors during purification—and the structure appears to represent a ligand-free state with the LBD bi-lobed open—it could also correspond to an apo state that represents an intermediate, recovering from the desensitized to the resting states. Based on our structural analysis, we were unable to distinguish between these two potential functional states. Therefore, further experiments analyzing the conformation of resting-state KARs within the physiological membrane will fulfill the shortfall.

Structure–function studies of KARs currently lack the extensive body of information available for AMPAs and NMDRs. Therefore, we aimed to investigate how the partial agonist domoate induces conformational changes and modulates KAR gating differently from classical full agonists. We observed that domoate stabilizes receptors in a range of conformations, likely representing intermediate states between the active and fully desensitized states or during the recovery from desensitization. Previous studies have suggested that the desensitization ring formation is crucial for regulating KAR desensitization, and its disruption serves as a key mechanism for receptors to return to their resting state from desensitization^42,108,117. Indeed, in domoate-bound GluK2 intermediate states, we observe a disrupted desensitization ring, with the ion channel pore not as tightly closed as in the fully desensitized or apo states, but not as widely open as in the fully open/active conformation. This observation strongly supports a correlation between the formation of the desensitization ring and the transitions involved in the ion channel’s closing and opening. This conformational transition is not observed in the presence of glutamate, suggesting a distinct mechanism between full and partial agonists.

While multiple studies have demonstrated that N-linked glycosylation regulates the localization and functional properties of KARs^43,71, the detailed molecular mechanisms remain unknown. Given that we observed the N275 glycan at the ATD-LBD domain interface and the LBD dimer interface interacting with residues forming the cation and anion binding sites in one of the domoate-bound GluK2 transient state structures, we investigated how this glycan modulates GluK2 gating. Our patch-clamp recordings revealed that the N275 glycan decreases glutamate affinity and slows desensitization, recovery from desensitization, and deactivation kinetics. Specifically, modulation of agonist affinity and desensitization occurs in a Na⁺ ion concentration-dependent manner. This implicates the physiological and pathophysiological roles of the N275 glycan.

In summary, our structural and functional studies reveal the unique gating mechanism of KARs, which is not conserved in other iGluRs. The partial agonist, which exhibits a steady-state current, allowed us to obtain structures representing intermediate states, revealing distinct mechanistic regulation by partial and full agonists. We analyzed the role of the desensitization ring formation, which is crucial for stabilizing the deep desensitized conformation with the ion channel tightly closed. Finally, our electrophysiological experiments support an extracellular Na⁺-dependent regulation mechanism of KARs by the glycan at the LBD dimer interface.

Methods

Plasmid construction

The full-length wild-type Rattus norvegicus GluK2 KAR (GenBank 54257, Uniprot code P42260, GRIK2_RAT), synthesized by Genscript, was cloned into the pFW-CMV vector and fused with a human rhinovirus 3C protease recognition site and a C-terminal Twin strep affinity tag for the large-scale protein expression in mammalian cells. For the protein expression in insect cells, the rat GluK2 gene was RNA-edited at position 567 (I to V) and had two mutations of C576V and C595S, which promote protein expression^42,93. For the large-scale expression of GluK2_{I567V/C576V/C595S} proteins, the gene was cloned into the pFp10_hsp vector harboring sequences for the hr1 enhancer, the Drosophila melanogaster HSP70 promoter, and the p10 3’ UTR, which has been established previously^94,119, and combined with a human rhinovirus 3C protease recognition site and a Twin strep tag at the C-terminus.

Cell culture

Mammalian cell culture for electrophysiological recordings:

Human embryonic kidney 293T cells (ATCC, Cat No. CRL-3216) were cultured in Dulbecco’s modified Eagle medium (CORNING) supplemented with 10% FBS at 37°C in a 95% O₂–5% CO₂ atmosphere. Wild-type cDNAs from rat GluK2 and mutant DNAs were cloned into a pCAG-IRES-EGFP (Addgene plasmid #119739) vector. HEK293T cells were transfected with 1 μg/μl cDNA using the TransIT2020 transfection reagent (Mirus) per 24–48 hrs according to the manufacturer’s instructions and then dissociated using Accutase (Innovative Cell Technologies, Inc). After resuspension, the cells were placed on 35 mm poly-D-lysine-coated dishes (Neuvitro), and electrophysiological records were obtained 4 h later⁶².

Mammalian cell culture for structural studies:

Suspension-adapted human embryonic kidney 293S GnTI⁻ (ATCC, Cat No. CRL-3022) cells were cultured in Gibco FreeStyle^™ 293 Expression Medium supplemented with 2% vol/vol fetal bovine serum (FBS), 100 units/ml penicillin,100 μg/ml streptomycin and 250 ng/ml Amphotericin.

Insect cell culture for structural studies:

Sf9 insect cells (Novagen, Cat No. 71104) were cultured in Sf-900 II SFM (Gibco) or HyClone CCM3 cell culture medium (Cytiva) supplemented with 100 units/ml penicillin,100 μg/ml streptomycin at 27°C. The cells were split twice a week and used until the 30th passage.

Electrophysiological recordings

All experiments were conducted in whole-cell configuration from transfected cells. The external solution contained (in mM) 145 NaCl, 2.5 KCl, 1.8 CaCl₂, 1 MgCl₂, 5 glucose, and 5 HEPES units. The internal pipette solution contained (in mM) 105 NaCl, 20 NaF, 5 Na₄BAPTA, 0.5 CaCl₂, 10 Na₂ATP, and 5 HEPES units. The pH and osmotic pressure of the external and internal solutions were adjusted to 7.4 and 300–290 mOsmol/Kg, respectively. 1 mM L-glutamate (TCI Chemicals, Cat. No. G0188) or 10 μM domoate (Sigma Aldrich, D6152; Sifga, CRM-03-DA) was applied using a theta glass tubing mounted on a piezoelectric stack (MXPZT-300 series, Siskiyou) driven by a HEKA EPC10 amplifier for 1 s to L-glutamate and 10 s to domoate. All recordings were performed with HEKA EPC10 amplifiers (HEKA Elektronik, Lambrecht, Germany) using a thin-wall borosilicate glass pipette (2–5 MΩ) coated with dental wax to reduce electrical noise. Currents were recorded with a holding potential of −70 mV, the sampling frequency was 10 kHz, and the signal was filtered at 2.6 kHz. Data acquisition was performed using PULSE software (HEKA Elektronic, Lambrecht, Germany, https://scicrunch.org/resolver/SCR_018399, https://www.heka.com/). All experiments were conducted at room temperature (22–24°C).

The microscopic rate of desensitization (tau desensitization) was measured by the exponential fit to the decay of current from ~80% of its peak amplitude (I_peak) to baseline of recordings where glutamate was applied by 1 s and domoate for 10 s taking just the first exponential component of the recording which shown a desensitization kinetics, omitting the steady state current for the domoate activated receptors. The desensitization kinetics were fitted by using the single exponential, one-term fitting (Levenberg-Marquardt). The percentage of desensitization was quantified only to domoate recordings, using the formula:

$$\mathrm{\%}desensitization=\left[100-\left(\frac{I_{steady-state}}{I_{\max peak}}\mathrm{*}100\right)\right]$$

where the I steady state was the maximum current taken at the end of the domoate application. The macroscopic deactivation rate (tau deactivation) was measured from 1 ms glutamate application recordings by exponential fit to the decaying current from ~80% of its peak amplitude to baseline.

Two protocols were employed to evaluate recovery from desensitization. The first protocol involved a two-pulse application, each application of 10 mM Glutamate for 100 ms. The interval between those pulses was changing with a Δ100 ms ranging from 50 ms to 2 s with. The second protocol utilized Δ2 s, extending from 50 ms to 20 s. In all cases, the first peak was considered the control (representing 100% activation), while the second peak indicated recovery at a specific time. Data from both protocols was combined, and a graph depicting the recovery percentage versus the time between sweeps was generated. The experimental data was then fitted to with the Hodgkin-Huxley equation: I_t=(I_max ^1/m – I₀ ^1/m)exp(−t/τ))^m, where the I_t is the peak current at a given pulse intervals, I₀ is current at zero time, τ is the recovery time constant and m is an exponent the value of which corresponds to the number of kinetically equivalent rate-limiting intermediate that contributes to the recovery time course⁸⁰.

Statistical analyses were performed using OriginPro 2025 (Origin Lab, RRID:SCR_014212, https://www.originlab.com/Newst). Statistical significance was calculated using a two-sided two-sample t-test, with the significance assumed if P < 0.05.

Expression and purification of GluK2 KAR

The GluK2 bacmids and baculovirus were generated using standard methods^94,120. The P1 and P2 baculovirus were produced in Sf9 cells. Three liters of Sf9 insect cells at 4 × 10⁶ cells/ml or five liters of HEK293S GnTl⁻ cells at 3.0 × 10⁶ cells/ml were infected with the recombinant baculovirus harboring GluK2_{I567V/C576V/C595S} or GluK2_WT at 27°C or 37 °C and 5% CO_2. To the HEK cell culture, 10mM sodium butyrate were added and temperature was decreased to 30 °C at 12 hrs post-infection. Sf9 and HEK cells are harvested at 48 hours and 72 hours post-infection, respectively.

For preparing the proteins of the GluK2_{I567V/C576V/C595S} construct, the cells were lysed with a Qsonica sonicator (4 min, 15 s on/off intervals, power level 25–30%) in a buffer containing 50 mM Tris-Cl, pH 8.0, 150 mM NaCl, and 1 mM PMSF. The cell debris was removed with 20 minutes centrifugation at 6,500 g followed by ultracentrifugation at 180,000 g. The cell membrane was solubilized using the buffer containing 50 mM Tris-Cl pH 8.0, 150 mM NaCl, and 15 mM dodecyl-β-D-maltoside (DDM, Anatrace) for 2 hours at 4°C and followed by the centrifugation at 180,000 g for 45 minutes. The Streptactin XT sepharose beads (1.0–1.2 ml per liter culture, IBA) was packed into a glass column, and the solubilized proteins were loaded by gravity flow at approximately 0.5 ml/min. The unbound proteins were washed with 15 CV of the buffer containing 50 mM Tris-Cl, pH 8.0, 150 mM NaCl, and 2 mM DDM, and eluted with 50 mM biotin. Subsequently, the eluted GluK2 sample was concentrated and loaded onto a Superose 6 increase 10/300 (GE Healthcare) column pre-equilibrated with 50 mM Tris-Cl pH 8.0, 150 mM NaCl, and 1 mM DDM. The peak fractions were pooled and concentrated to 4 mg/ml for the cryo-EM studies. All steps were performed at 4°C.

For preparing the proteins of the GluK2_WT, cell pellets were resuspended in the lysis buffer containing 20 mM Tris-Cl pH 8.0, 300 mM NaCl, 100 μM DNQX, and 2mM PMSF. Cells were lysed by sonication. Cell debris and unbroken cells were removed by centrifugation at 7,000 g for 20 minutes. The supernatant was subjected to ultracentrifugation at 200,000 g for 20 minutes to isolate cell membranes. Cell membranes were homogenized and solubilized for 2 hours in the buffer containing 20 mM Tris-HCl pH 8.0, 300 mM NaCl, 100 μM DNQX and 1% DDM. The insoluble material was removed by ultracentrifugation at 200,000 g for 20 minutes. The supernatant was mixed with 5 ml of streptavidin-linked beads. The mixture was rotated for 1 hour at 4 °C. The protein-bound beads were extensively washed with the buffer consisting of 20 mM Tris-Cl pH 8.0, 500 mM NaCl, 1 mM DNQX and 0.02 % DDM at room temperature, followed by the further wash with the buffer consisting of 20 mM Tris-Cl pH 8.0, 300 mM NaCl, and 0.05% digitonin at room temperature. The flow rate was adjusted at 0.2–0.3 ml/min. The protein was eluted with the buffer containing 20 mM Tris-Cl pH 8.0, 300 mM NaCl, 0.05% digitonin, and 50 mM biotin. The eluted protein was concentrated and loaded onto a Superose 6 10/300 GL size-exclusion column equilibrated with the buffer containing 20 mM Tris-Cl pH 8.0,150 mM NaCl, and 0.05 % digitonin. The peak fractions corresponding to tetrameric Gluk2 were pooled and concentrated to 3.4 mg/ml for grid freezing.

Mass spectroscopy

For protein digestion, the bands were cut from the gel with a punch machine and washed/destained in 50% ethanol and 5% acetic acid. The gel pieces were then dehydrated in acetonitrile, dried in a Speedvac chamber, digested with trypsin by adding 5 μl of 10 ng/ μl proteases in 50 mM ammonium bicarbonate, and incubated overnight digestion at room temperature. The peptides that were formed were extracted from the polyacrylamide in two aliquots of 30 μl each, with 50% acetonitrile with 5% formic acid. These extracts were combined and evaporated to <10 μl in a Speedvac chamber and then resuspended in 1% acetic acid to form a final volume of approximately 30 μl for the LC-MS analysis. The LC-MS system used was a Finnigan LTQ-Obitrap Fusion Lumos hybrid mass spectrometer. The HPLC column was a Dionex 15 cm x 75 μm id Acclaim Pepmap C18, 2 μm, 100 Å reversed-phase capillary chromatography column. Five microliters of the extract were injected, and the peptides eluted from the column by using an acetonitrile/0.1% formic acid gradient at a flow rate of 0.25 μl/min were introduced into the source of the mass spectrometer online. The digest was analyzed using the data-dependent multitask capability of the instrument by acquiring full-scan mass spectra to determine peptide molecular weights and product ion spectra to determine the amino acid sequence in successive instrument scans. These data were searched spectra specifically against the rat GluK2 sequencing using all CID or EThcD spectra collected in the experiment to search specifically against the proteins of interest with the search programs Sequest (CID) or MSFragger (EThcD) program considering N-linked glycosylation as a variable modification.

TEM sample preparation and data collection

The protein samples were vitrified on glow-discharged UltrAufoil holey-gold film grids (Quantifoil) using FEI Vitrobot Mark IV at 4–22°C, at 100% humidity, with a blot time of 2–8 s under level 10 blot force (domoate bound GluK2) or at 4°C, at 100% humidity, with a blot time of 3 s under level −5 blot force (apo GluK2). PELCO easiGlow^™ was used to render all grids hydrophilic by glow discharging for 60 s at 30 mA. In addition, 2–4 μl of the receptor sample (3–4 mg/ml) was applied onto the grid. All micrographs were acquired using Titan Krios (FEI) at the Stanford-SLAC Cryo-Electron Microscopy Facility at 300 kV and the GATAN K3 Summit direct electron detector coupled with the GIF quantum energy filter (Gatan Inc.) at 81,000 magnification in super-resolution mode (0.53/1.06 Å/pixel), with the defocus range of −1.0 to −2.0 μm (domoate-bound GluK2) or −0.8 to −1.6 μm (apo GluK2), and over 50–60 frames and 2.5–5 s exposure, yielding a total dose of 60 e⁻/Ų (domoate-bound GluK2) or 50 e⁻/Ų (apo GluK2). Automated data collection was conducted using EPU software (Thermofisher Scientific).

Data processing

Domoate bound GluK2 and no added ligands desensitized GluK2 Data:

The movie alignment and subsequent data processing were performed using Relion 3.1,4.1, and Beta¹²¹ (RRID:SCR_016274, https://www3.mrc-lmb.cam.ac.uk/relion/index.php/Main_Page) or Cryosparc v3.2 and 3.31 (Structura Biotechnology Inc.)¹²² with various wrappers. CTF estimation was performed using wrapper GCTF 1.06 (RRID:SCR_016500, https://www2.mrc-lmb.cam.ac.uk/download/gctf_v1-06-and-examples/)¹²³ and CTFFIND 4.1 (RRID:SCR_016732, http://grigoriefflab.janelia.org/ctffind4)¹²⁴, and the initial particles were selected automatically. The final particle selection was performed using the Topaz¹²⁵ particle picker. The selected particle images were subjected to several rounds of 2D classifications to remove misaligned and damaged particles. The selected particles were subjected to iterative rounds of homogeneous and nonuniform refinements using an imported reference map (EMD-8289) low-passed to 30 Å. Subsequent heterogeneous refinement was performed to classify the particles in distinct 3D classes with C1 symmetry. The 3D classes were inspected using UCSF Chimera (RRID:SCR_004097, http://plato.cgl.ucsf.edu/chimera/)¹²⁶, and the particles from similar 3D classes were merged, further refined, and classified. The 3D classes that did not contain density for the TMDs were excluded from further refinement. If the 3D classes looked symmetrical, C2 symmetry was imposed, and the particles were further refined using per-particle CTF refinement and B-factor sharpening. The final map was refined using a standalone program deepEMhancer¹²⁷. Model building was initially performed by docking the structural coordinates of GluK2 experimental cryo-EM density maps using UCSF ChimeraX (RRID:SCR_015872, https://www.cgl.ucsf.edu/chimerax/)^128,129. The resulting models were manually inspected using COOT (RRID:SCR_014222), where bond lengths and torsion angles were corrected to accommodate the missing sidechains and loops and the rotamer orientations were corrected to avoid steric clashes. The prepared structure was further modified to fit into the experimental density map using COOT¹³⁰. The final models were refined against the cryo-EM maps using Phenix real-space refinement (RRID:SCR_014224)¹³¹ with a secondary structure and Ramachandran restraints.

Apo GluK2 Data:

9053 movies were collected and processed using cryoSPARC v4.5.0 (RRID:SCR_016501, https://cryosparc.com/). Patch Motion Correction algorithm was used for movie frames alignment. Following Motion Correction, movies were subjected to patch CTF estimation for Contrast transfer function (CTF). After CTF estimation, micrographs having CTF resolution better than 10 Å were considered for further processing. Approximately 500 micrographs were used for blob picking and particle extraction, followed by multiple rounds of 2D classification to create templates for template picking. The total number of 1,198,495 particles were picked from the micrographs using results from Template picker and extracted with the 280 box-size and then binned to the bin-4-pixel box size. Junk particles were removed after several rounds of reference-free 2D classification. Initial reference was created using ab initio reconstruction. Bad classes were removed by multiple rounds of heterogeneous refinement. 176, 501 particles from the good class were re-exacted with bin-1-pixel box size and further subjected to non-uniform refinement resulting in a 3.88 Å reconstruction. Subsequently, focused masks around the ATD and LBD-TMD were separately created and used for the local refinements of the corresponding regions. Finally, 3.1 Å reconstruction of ATD region and 3.63 Å reconstruction of LBD-TMD region were generated by local refinement with focused masks.

Model modeling, refinement and validation

Refinement and analysis were performed with a combination of Chimera, Coot and PHENIX. ATD model (PDB ID: 8FWS)⁸⁵ was used as staring model and rigid-body fit into the ATD reconstruction using Chimera. For LBD-TMD reconstruction, LBD and TMD domains of DNQX-bound structure (PDB ID: 7F56)⁴⁶ were separately isolated and underwent rigid-body fitting into the corresponding position of each protomer. Then, each domain was joined into a single model in Chimera. The composite map was generated in Chimera using ATD focused map and LBD-TMD focused map. ATD and LBD-TMD model were individually rigid-body fit into composite map in Chimera. All initial models were then subjected to real-space refinement using Phenix. The refined models were further manually examined and adjusted in Coot. Finally, all models were refined in real space in Phenix. Phenix comprehensive validation was used to validate all refined models. Figures were prepared using Chimera X.

Extended Data

Extended Data Fig. 1. Single-particle analysis of GluK2 (no added ligand, desensitized).

A representative EM image, 2D classes, and the 3D classification/refinement workflow.

Extended Data Fig. 2. Structural analysis of no added ligands, desensitized GluK2 compared to the glutamate bound desensitized GluK2 KAR structure.

(A, B) Side view of three-dimensional reconstruction and cartoon representation of no added ligands desensitized GluK2 KAR. (C) The GluK2 LBD structures of subunits A in the “no added ligands, desensitized GluK2” in cyan compared to the crystal structure of isolated LBD of glutamate bound GluK2 LBD (PDB ID: 3G3F)126. The GluK2 LBD ofsubunits A in this structure are stabilized in bi-lobe closed conformations similar to the glutamate bound crystal structures of isolated GluK2 LBD. (D) The top view of the LBD layer, and the side view of LBD-TMD layers for the “no added ligands, desensitized GluK2” superimposed onto the glutamate-bound GluK2. (E) Top view of the G and E helices, and the side view of the desensitization ring consisting of the G and E helices with the TMD M3 helices. (F) Side view of the LBD–TMD layers. Top and middle: subunit AC and BD LBD pairs shown as viewed from the side with the distances between the center of masses (COMs) of D1 or D2 lobes of the subunit AC pair and the subunit BD pair for the “no added ligands, desensitized GluK2” and glutamate bound GluK2. Bottom: Shown are the distances between the COMs of the E helices and the top of the TMD M3 helices for the apo and glutamate bound GluK2 viewed from the side.

Extended Data Fig. 3. Single-particle analysis of apo GluK2.

A representative EM image, 2D classes, and the 3D classification/refinement workflow.

Extended Data Fig. 4. Local and global resolution estimates and post-processing analysis of unliganded apo and domoate-bound GluK2.

(A–D, M–P) Local resolution estimation calculated by ResMap. (E–H, Q–T) Fourier shell correlation (Glu) curves with unmasked curve in black, masked curve in red, and map-to-modelcurve in blue. The threshold was set at 0.143 for unmasked and masked curves, and 0.5 for the map-to-model curve. (I–L, U–X) Angular distribution for apo and domoate-bound GluK2.

Extended Data Fig. 5. Ligand and transmembrane densities in cryo-EM structures.

(A) a–d: Cryo-EM densities of LBDs in the unliganded apo state. The magenta circle indicates the agonist-binding site. e-l: Cryo-EM densities of domoate within the ligand binding pocket in domoate-bound GluK2 structures. (B) Cryo-EM densities of the M1, M3, and M4 helices in the apo and domoate bound GluK2 in the desensitized 2.

Extended Data Fig. 6. Functional characterization of GluK2I567V/C576V/C595S forstructural studies.

(A) Representative whole-cell currents at −70 mV holding potential in HEK293T cells expressing rat GluK2WT (wild-type) or GluK2I567V/C576V/C595S in response to 1 s applications of 10 mM glutamate (black bar). (B) Desensitization kinetics. Bar graphs show the mean desensitization time constants (τdesensitization) for GluK2WT (n = 14) and GluK2I567V/C576V/C595S (n = 11) obtained by fitting an one term exponential function to the decay kinetics. (C) Representative whole-cell currents evoked by 1 ms application of 10 mM glutamate (black bar) in rat GluK2WT and GluK2I567V/C576V/C595S. (D) Deactivation kinetics. Bar graphs showing the mean deactivation time constants (τdeactivation) for GluK2WT (n = 11) and GluK2I567V/C576V/C595S (n = 8), obtained by fitting an exponential function to the decay kinetics. (E) Glutamate dose-response curves for GluK2WT and GluK2I567V/C576V/C595S recorded in the whole-cell configuration. Continuous lines represent the fit to the Hill equation. (F) EC50 quantification from (E). GluK2WT EC50: 182 ± 18 μM (n = 6) and GluK2I567V/C576V/C595S EC50: 239 ± 63 μM (n = 5). Data are presented as mean ± SEM (error bars). Each circle represents an independent experiment. The number of biological replicates (n) is indicated for each condition. Statistical significance was assessed using a two-sided, two-sample t-test, with p-values shown in each panel; p < 0.05 was considered significantly different. (Source Data Extended Data Fig. 6).

Source data

Extended Data Fig. 7. Single-particle analysis of the domoate bound GluK2I567V/C576V/C595S.

A representative EM image, 2D classes, and the 3D classification/refinement workflow.

Extended Data Fig. 8. Comparison of the amino-terminal domain in domoate-bound GluK2 structures.

(A) Structural superposition of ATDs in domoate-bound GluK2 KARs and glutamate-bound GluK2 (PDB ID: 5KUF)42, showing the top and side views. (B) Differences in relative positioning of ATD layers in the full-length domoate-bound GluK2 compared to the glutamate-bound GluK2 KAR (PDB ID: 5KUF)42.

Extended Data Fig. 9. Effect of N-linked glycosylation on KAR gating properties.

(A) Representative whole-cell currents recorded at a holding potential of −70 mV from HEK293T expressing GluK2WT and the N275 glycan knock-out mutant of GluK2 (GluK2T277A) in response to 1 s of 10 mM glutamate application (black bar). (B) Comparison of the peak current amplitudes between GluK2WT (n = 12) and GluK2T277A (n = 8). (C) Representative whole cells recordings from HEK293T cells expressing GluK2WT and GluK2T277A evoked by a 1 s application of glutamate at concentrations ranging from 0.001–50 mM glutamate (black bar indicates glutamate application). (D) Dose-response curves for GluK2WT (n = 6) and GluK2T277A (n = 7). Continuous lines represent the fit to the Hill equation (E) EC50 quantification from (D). GluK2WT (181 ± 18 μM, n = 6) and GluK2T277A. (118 ± 20 μM, n = 7) (F, H) Representative normalized whole cell currents recorded at −70 mV from HEK293T cells expressing GluK2WT or GluK2T277A in response to 1 s (F) or 1 ms (H) glutamate applications. (G, I) Desensitization and deactivation kinetics. Time constants were obtained by fitting a single-exponential function to 80 % of the baseline-to-peak decay kinetics. Desensitization: GluK2WT (7.5 ± 0.3 ms, n = 14) vs GluK2T277A (5.44 ± 0.41 ms, n = 11). Deactivation: GluK2WT (3.1 ± 0.08 ms, n = 11) vs GluK2T277A (2.49 ± 0.17 ms, n = 10). (J) Two-pulse glutamate recovery experiments. Representative current traces from GluK2WT and GluK2T277A showing recovery from desensitization with inter-pulse intervals ranging from 50 ms to 20 s. (K) Recovery quantification. The amplitude of the second glutamate application in a two-pulse experiment is reported as a normalized percentage of the first glutamate application and was fitted to the Hodgkin-Huxley equation. (L) Comparison of the τ (tau) recovery. The tau recovery value was obtained from fits to the Hodgkin-Huxley equation: GluK2WT 716.23 ± 24 ms, n = 11, and GluK2T277A 637.28 ± 19 ms, n = 10. Data are presented as mean ± SEM (error bars). Each black circle represents an independent experiment. The number of biological replicates (n) is indicated for each condition. Statistical significance was determined using a two-sided, two-sample t-test; p-values are shown in each panel, with significance defined as p < 0.05. (Source Data Extended Data Figure 9).

Source data

Extended Data Table 1.

Cryo-EM data collection, refinement, and validation statistics.

	Domoate-bound desensitizing intermediate 1	Domoate-bound desensitizing intermediate 2	Domoate-bound desensitized 1	Domoate-bound desensitized 2	Apo, ATD	Apo, LBD-TMD	Apo, full-length	Apo, composite full length	No added ligands Desensitized
	(EMDB-29926) (PDB 8GC2)	(EMDB-29928) (PDB 8GC4)	(EMDB-29927) (PDB 8GC3)	(EMDB-29929) (PDB 8GC5)	(EMDB-45237) (PDB 9C5Y)	(EMDB-45238) (PDB 9C5Z)	(EMDB-45410) (PDB 9CAZ)	(EMDB-45239) (PBD 9C60)	(EMDB-28775) (PDB 8F0O)
Data collection and processing
Microscope	Titan Krios	Titan Krios	Titan Krios	Titan Krios	Titan Krios	Titan Krios	Titan Krios	Titan Krios	Titan Krios
Camera	K3	K3	K3	K3	K3	K3	K3	K3	K3
Magnification	81,000	81,000	81,000	81,000	81,000	81,000	81,000	81,000	81,000
Energy filter	Gatan	Gatan	Gatan	Gatan	Gatan	Gatan	Gatan	Gatan	Gatan
Energy filter slit width (eV)	20	20	20	20	20	20	20	20	20
Collection software	EPU	EPU	EPU	EPU	EPU	EPU	EPU	EPU	EPU
Voltage (kV)	300	300	300	300	300	300	300	300	300
Cumulative exposure (e–/Å2)	60	60	60	60	50	50	50	50	60
Exposure rate (e–/Å2/frame)	1.2	1.2	1.2	1.2	1.0	1.0	1.0	1.0	1.2
Defocus range (μm)	−2.0 – −1.0	−2.0 – −1.0	−2.0 – −1.0	−2.0 – −1.0	−0.8 – −0.6	−0.8 – −0.6	−0.8 – −0.6	−0.8 – −0.6	−2.0 – −1.0
Pixel size (Å)	0.53	0.53	0.53	0.53	1.06	1.06	1.06	1.06	0.53
Symmetry imposed	C1	C1	C1	C1	C1	C1	C1	C1	C2
Number of micrographs	12,486	12,486	12,486	12,486	9,315	9,315	9,315	9,315	4,021
Initial particle images (no.)	327,912	327,912	327,912	327,912	1,198,495	1,198,495	1,198,495	1,198,495	532,380
Final particle images (no.)	39,636	73,190	100,389	101,484	171,688	171,688	171,688	171,688	179,490
0.143 FSC half map masked (Å)	4.3	4.0	3.9	3.9	3.26	3.79	3.99		3.26
0.143 FSC half map unmasked (Å)	7.3	6.2	4.5	4.5	3.39	3.89	4.17		3.34
Refinement
Refinement package	Phenix	Phenix	Phenix	Phenix	Phenix	Phenix	Phenix	Phenix	Phenix
Initial model used (PDB code)	5KUF	5KUF	5KUF	5KUF	8FWS	7F56	8FWS, 7F56	8FWS, 7F56	5KUF
0.5 FSC model resolution masked (Å)	4.3	4.1	4.0	3.9	3.4	4.0	4.2	3.6	4.2
0.5 FSC model resolution unmasked (Å)	4.5	4.3	4.3	4.2	3.5	4.0	4.3	3.8	4.5
Model resolution range (Å)	50 – 4.3	50 – 4.0	50 – 4.3	50 – 3.9	2.5 – 4.5	3.5 – 7.0	3.5 – 7.0	2.5–7.0	50–3.0
Map sharpening B factor (Å2)	−10	−10	−10	−10	−91	−114.6	−131.2		−10
Model composition
Non-hydrogen atoms	24,346	24,139	24,150	24,150	12,302	11,451	23,751	23,765	24,030
Protein residues	2,993	2,968	2,968	2,968	1,532	1,442	2,978	2,978	2,968
Ligands ((N-acetylgucosamine))	45	43	44	44	7	6	11	12	20
CC map vs. model (%)	0.83	0.84	0.84	0.84	0.79	0.73	0.72	0.74	0.83
R.m.s. deviations
Bond lengths (Å)	0.006	0.006	0.006	0.007	0.003	0.003	0.003	0.003	0.005
Bond angles (°)	0.809	0.865	0.792	0.786	0.530	0.583	0.585	0.583	1.078
Validation
MolProbity score	1.52	1.59	1.53	1.49	1.87	1.93	2.11	1.93	1.53
Clashscore	4.29	4.72	4.31	3.75	9.84	12.36	16.27	12.36	3.91
Poor rotamers (%)	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
Ramachandran plot
Favored (%)	95.56	94.97	95.41	95.31	94.95	95.3	94.08	95.3	94.90
Allowed (%)	4.44	5.03	4.59	4.69	5.05	4.7	5.92	4.7	5.10
Disallowed (%)	0.10	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
C-beta deviations	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
EMRinger Score	1.41	1.98	2.01	1.39	2.37	0.94	0.82	1.84	0.27
CaBLAM outliers (%)	2.78	3.29	2.50	2.09	1.78	3.19	2.66	2.56	2.84

Data availability

Cryo-EM density maps and atomic coordinates for GluK2-apo and GluK2-domoate were deposited in the electron microscopy data bank under the accession codes listed below. Source data are provided with this paper: Domoate-bound GluK2 in intermediate 1 (EMDB: 29926, PDB: 8GC2), Domoate-bound GluK2 in intermediate 2 (EMDB: 29928, PDB: 8GC4), Domoate-bound GluK2 in desensitized 1 (EMDB: 29927, PDB: 8GC3), Domoate-bound Gluk2 in desensitized 2 (EMDB: 29929, PDB: 8GC5), GluK2 ATD, apo (EMDB:45237, PDB:9C5Y), GluK2 LBD-TMD, apo (EMDB:45238, PDB:9C5Z), GluK2 full-length, apo (EMDB:45410, PDB:9CAZ), GluK2 composite full-length, apo (EMDB:45239, PDB:9C60), and GluK2 with no added ligands, desensitized (EMDB: 28775, PDB 8F0O). All other data supporting the findings of this study are available from the corresponding author on reasonable request.