Delta-type glutamate receptors are ligand-gated ion channels

Abstract

Delta-type ionotropic glutamate receptors (iGluRs, also known as GluDs) are members of the iGluR ligand-gated ion channel family, yet their function remains unknown¹. Although GluDs are widely expressed in the brain, have key roles in synaptic organization, and harbour disease-linked mutations, whether they retain iGluR-like channel function is debated as currents have not been directly observed^2,3. Here we define GluDs as ligand-gated ion channels that are tightly regulated in cellular contexts by purifying human GluD2 (hGluD2) and directly characterizing its structure and function using cryo-electron microscopy and bilayer recordings. We show that hGluD2 is activated by d-serine and GABA (γ-aminobutyric acid), with augmented activation at physiological temperatures. We reveal that hGluD2 contains an ion channel directly coupled to clamshell-like ligand-binding domains, which are coordinated by the amino-terminal domain above the ion channel. Ligand binding triggers channel opening via an asymmetric mechanism, and a cerebellar ataxia point mutation in the ligand-binding domain rearranges the receptor architecture and induces leak currents. Our findings demonstrate that GluDs possess the intrinsic biophysical properties of ligand-gated ion channels, reconciling prior conflicting observations to establish a framework for understanding their cellular regulation and for developing therapies targeting GluD2.

Structural and in vitro functional studies of the human delta-type ionotropic glutamate receptor GluD2 reveal that it contains an ion channel that is activated by d-serine and GABA (γ-aminobutyric acid).

Main

The delta-type iGluRs GluD1 and GluD2 are critical for synaptic organization throughout various regions of the brain^1,4. This is achieved through a tripartite complex, in which pre-synaptic neurexin presents secreted cerebellin to post-synaptic GluD receptors, forming a trans-synaptic complex^5–10. Although GluD1 and GluD2 share sequence homology and topology with the broader iGluR family, it is unclear whether GluDs are ligand-gated ion channels^{1,3,9,11–14}.

Here we focused on GluD2, which is enriched in cerebellar Purkinje cells and is vital for synaptic organization and maintenance in these cells^1,4,15,16. Mutations in GluD2 are directly correlated with cerebellar atrophy and lead to diseases such as cerebellar ataxia, autism spectrum disorder and schizophrenia^1,17. The protein cerebellin-1 (CBLN1) is secreted from granule cells and binds to pre-synaptic neurexin on parallel fibre terminals, which then forms a trans-synaptic complex with GluD2 at parallel fibre–Purkinje cell synapses^1,4. It has been proposed that the tripartite neurexin–CBLN1–GluD2 complex is necessary for GluD2 channel activity¹³; however, recent work has suggested that neither the tripartite complex nor GluD2 have channel activity^11,12.

Formation of the tripartite complex at parallel fibre–Purkinje cell synapses can lead to signal transduction in the post-synaptic Purkinje cell^{1,4,9,18–24}. We thus expected wild-type GluD2 to maintain iGluR-like channel function, given that the likely ligand of GluD2, d-serine, which is enriched at parallel fibre–Purkinje cell synapses, binds GluD2 and closes the GluD2 ligand-binding domain (LBD) clamshell, an action that would activate the other iGluR subtypes and initiate signal transduction^24–27. In addition, constitutive leak currents caused by mutations in the potential GluD2 ion channel are reduced by d-serine, suggesting desensitization²⁸. d-Serine also gates GluD2 that is mutagenized in its extracellular domain (ECD) or has its amino-terminal domain (ATD) deleted^13,27. Finally, GluD currents stimulated by G-protein-coupled receptors (GPCRs) are reduced by d-serine and blocked by ion channel blockers, suggesting ligand-dependent desensitization of GluD channels^29–34. Nonetheless, despite these data, other studies suggest that GluDs are not functional ion channels owing to the lack of direct observation of ligand-gating events in wild-type receptors^11,12.

One possibility is that GluDs contain ligand-gating machinery, and this machinery is tightly regulated and thus masked in different cellular contexts. To test this idea, we purified full-length hGluD2 and characterized it using cryo-electron microscopy (cryo-EM) and single-channel bilayer recordings. In the resting state of hGluD2, the ATD directly coordinates the LBD above the hGluD2 transmembrane domain (TMD), which houses the ion channel. d-Serine activates hGluD2 and opens the ion channel for conductance. When the LBDs bind d-serine, the ATD maintains coordination of the LBD, resulting in asymmetric opening of the ion channel. We show that the ATD directly affects the hGluD2 ligand response and ultrastructure, and demonstrate how a cerebellar ataxia mutation alters receptor structure and function. Together, these results establish hGluD2 as a bona fide ligand-gated ion channel, setting the foundation for dissecting its cellular regulation, and will enable structure-based drug design targeting hGluD2 in neurological disorders.

Overall architecture

We purified hGluD2 and reconstructed the architecture of hGluD2 in the resting state (in the absence of ligands) with cryo-EM (Fig. 1a, Extended Data Figs. 1–3 and Methods). Overall, hGluD2 resembles an italic ‘Y’ that has a tetrameric, non-domain-swapped arrangement (Fig. 1b). The ATD is at the top of hGluD2. Below the ATD is the LBD, which sits above the TMD that houses the hGluD2 ion channel. The four subunit positions are denoted A to D. The ATD is arranged into local dimers at subunits A–B and subunits C–D (Fig. 1c). Subunits A–B in the ATD sit tightly over the LBD layer, whereas subunits C-D splay away and are conformationally flexible. The A–B ATD local dimer is rigid because the ATD of the A subunit (A-ATD) directly coordinates the hGluD2 LBD (Fig. 1c). This is achieved by the A-ATD physically linking together the A, B and C LBDs through an approximately 1,500 Ų surface. The ATD and LBD are overall asymmetric and the TMD has pseudo-fourfold symmetry around the ion channel (Fig. 1c).

Fig. 1. Cryo-EM and architecture of hGluD2.

a, Composite map of apo hGluD2, with the unsharpened map and Gaussian-filtered map overlaid. b, Surface representation of the hGluD2 structure. c, Slices through each domain presented in b. Flexible regions are outlined with dashed lines. A-ATD, subunit A ATD. d, Overview of features in a single hGluD2 protomer from N terminus to C terminus. e, View of the hGluD2 ion channel composed of helices M2 and M3. Subunit A is omitted for clarity. f, Pore radius profile for apo hGluD2. The dashed line at 1.4 Å represents the radius of a water molecule.

Extended Data Fig. 1. Purification of hGluD2.

a, Construct design schematic. b, Chromatogram from size exclusion chromatography. c, Gel of pooled fractions. d, Example FSEC (Trp fluorescence) of purified hGluD2.

Extended Data Fig. 3. Map and local qualities for apo hGluD2.

a, Gold standard Fourier shell correlation (GSFSC) curves for each whole and local map. Black line is Fourier shell correlation (FSC) = 0.143. Y axis is FSC, X is resolution in Å. b, Heat maps of particle orientation distribution for each map. c, Local resolution maps computed for each voxel, computed FSC = 0.143. d, Pore model fits for each subunit.

The individual hGluD2 protomers maintain iGluR-like topology^1,35 (Fig. 1d): at the start of the amino terminus is the ATD, which is structured into two lobes, R1 and R2. R2 from subunit A coordinates the LBD. Ligands bind to the LBD in the centre of the clamshell, composed of upper (D1) and lower (D2) halves. We refer to the open face between D1 and D2 as the front of the LBD and the opposite side as the back. Because the ATD arranges the LBD layer, the individual LBDs are atypically arranged for an iGluR. Typically, the LBDs are arranged into back-to-back local dimers¹. However, in the case of the tetrameric arrangement of hGluD2, the LBDs are arranged front-to-back in a crescent around the A-ATD (Fig. 1c).

Local hGluD2 ATD dimers resemble crystal structures of isolated GluD1 and GluD2 ATD dimers, with root mean squared deviations (RMSDs) of 1.2 Å and 1.0 Å, respectively (Extended Data Fig. 4a). Similarly, compared to isolated GluD1 and GluD2 LBD crystal structures, hGluD2 has an RMSD of 1.0 Å to each (Extended Data Fig. 4b,c). Notably, isolated GluD2 receptor LBDs are dimeric and arranged in a back-to-back arrangement, and calcium has a critical role in the back-to-back arrangement²⁷ (Extended Data Fig. 4d). However, this arrangement is not present in this conformation of hGluD2. In this arrangement, back-to-back LBD dimers would result in significant clashes throughout the ECD (Extended Data Fig. 4e). Further highlighting this is a cryo-EM structure of the ECD of rat GluD2 in the presence of calcium³⁶, which resembles the hGluD2 structure reported here that was prepared in the absence of calcium (RMSD ≈ 2.0 Å; Extended Data Fig. 4f). Of note, although the ECD has this architecture, we expect it to be primed for binding to potential trans-synaptic binding partners. For example, the crystal structure of the CBLN1–rat GluD2 ATD complex fits well into the hGluD2 structure (ATDs agree to approximately 1.0 Å RMSD) and shows that the CBLN1 binding interface is exposed in the full-length receptor (Extended Data Fig. 4g).

Extended Data Fig. 4. hGluD2 structural comparisons.

a, Comparison of the A-B subunit local ATD dimer in hGluD2 with isolated local ATD dimer crystal structures from GluD1 and GluD2. b, Apo hGluD2 LBD comparison with apo GluD1 and GluD2 isolated LBD crystal structures. c, D-serine bound hGluD2 LBD compared to crystal structures of GluD1 and GluD2 isolated LBDs. d, Crystal structure of a back-to-back GluD2 LBD dimer. e, How the back-to-back LBD dimer would fit into hGluD2. f, Comparison of rat GluD2 ECD to hGluD2. f, Alignment of the Cbln1-GluD2 ATD crystal structure onto hGluD2. h, Comparison of hGluD2_ΔATD to an AMPAR LBD-TMD structure.

The TMD has three transmembrane helices, M1, M3 and M4, and a re-entrant loop helix, M2, between M1 and M3 (Fig. 1d). The pre-M1 helix leads into M1. The LBD is directly tethered to the pre-M1, M3 and M4 helices by LBD–TMD linkers. The carboxy terminus follows M4. The pre-M1 helix cuffs the top of the TMD, with the M1, M2 and M4 helices forming the exterior of the TMD. The central pore is formed by the M3 helices, along with the M2 re-entrant loop helices.

The top of the ion channel is primarily composed of the M3 pore helices, and the bottom is formed by the tips of the M2 re-entrant loop helices and the following selectivity filter (Fig. 1e). The top of the gate is formed by T657 on each hGluD2 M3 helix, and is followed by the SYTANLAAF motif, which is a conserved amino acid sequence across the iGluRs¹. In iGluRs, SYTANLAAF contains hinges that enable opening of the channel gate for ion permeation, and mutations here have deleterious effects. For example, the Lurcher mutation in GluD2, which occurs at the penultimate residue in the SYTANLAAF motif (A654T), causes excitotoxicity, spontaneous activation, and is directly correlated with cerebellar ataxia^28,37,38. The selectivity filter begins with Q618 at the tip of each M2. Mutation of this site (for example, Q618R) in Lurcher GluD2 receptors alters channel selectivity and renders it calcium-impermeable²⁸.

In the apo or resting state of hGluD2, the channel is putatively closed, with a minimum restricted radius (r_min) of approximately 0.9 Å at T657 (Fig. 1f). Below, A653 is restricted to the radius of a water molecule, and the next constriction is approximately 1 Å at T649 (Fig. 1f). Feature-wise, the channel is nearly identical to that of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA)-subtype iGluRs (AMPARs) in the resting state.

Because hGluD2 has a typical AMPAR-like channel³⁵ that is directly connected to LBDs, which enables ligand gating of the other iGluR subtypes, we hypothesized that hGluD2 is a ligand-gated ion channel.

Ligand gating

We reconstituted hGluD2 into 1,2-diphytanoyl-sn-glycero-3-phosphocholine (DPHPC) lipid bilayers and tested whether we could elicit ligand-gated responses at physiological temperature (37 °C) from d-serine, the proposed endogenous ligand for GluD2²⁴, which has been previously shown to induce closure of the LBD clamshell in both GluD1 and GluD2^3,25,26. This action would enable gating in the other iGluR subtypes and has been demonstrated to gate currents from GluD2 receptors with the Lurcher mutation in cells^1,3,9,13,35.

Compared with no ligand treatment, we observe that purified hGluD2 responds to d-serine (Fig. 2a,b), with hGluD2 oscillating between closed (C) and open (O) states. This is both hGluD2- and ligand-dependent (Extended Data Fig. 5a), and is reminiscent of single-channel currents from the other iGluR subtypes¹. The current histogram of hGluD2 was best fit with three components, which reveals the first three sub-conductance levels of the four (O1–O4) that resemble those observed in AMPARs. The observed conductance levels corresponding to mean amplitudes are 16.9 ± 2.7 pS, 33.8 ± 10.3 pS and 64.1 ± 3.5 pS (Fig. 2c). hGluD2 primarily occupying O1 and O2 at 37 °C agrees with AMPARs occupying primarily O1 and O2 at 37 °C in the absence of their auxiliary subunits³⁹ at 37 °C. The mean hGluD2 conductance (±s.d.) is 30.7 ± 16.1 pS (n = 38; Fig. 2d). The open probability (P_O) of hGluD2 in the presence of 10 mM d-serine is 8.85% (Fig. 2e), which is similar to that of AMPARs (without auxiliary subunits)³⁹ at 37 °C.

Fig. 2. hGluD2 bilayer recordings.

a, Recording without ligands at 37 °C (n = 29 individual traces). Total duration shown is 10 s. b, Recordings (10 s) in the presence of 10 mM d-serine at 37 °C (n = 38 individual traces) and 22 °C (n = 42 individual traces). C, closed; O, open. c, Current histogram of d-serine responses fit with three components (n = 38 individual traces, R² = 0.9787) and current histogram with co-application of 100 µM NASPM. Above each component is the corresponding conductance ± s.d. Peak amplitude values generated from 10 s event detections (n = 80 from each group). d, Mean conductance ± s.e.m. 37 °C, 30.67 ± 0.24 pS (n = 4,467 events); 22 °C, 29.76 ± 0.48 pS (n = 1,206 events); 37 °C + 2 mM Ca²⁺, 36.49 ± 0.41 pS (n = 2,242 events); 37 °C versus 22 °C mean conductance P value = 0.2051, 95% CI of difference −0.3463 to 2.173, d.f. = 1,854; 37 °C versus 37 °C + 2 mM Ca²⁺ mean conductance P value < 0.0001, 95% CI of difference −6.948 to −4.700, d.f. = 3,786. e, Mean P_O ± s.e.m. 37 °C, 8.85 ± 1.02% (n = 80 10 s event detections); 22 °C, 1.03 ± 0.14% (n = 142 10 s event detections); 37 °C + 2 mM Ca²⁺, 6.50 ± 1.03% (n = 49 10 s event detections). 37 °C versus 22 °C P_O P value < 0.0001, 95% CI of difference 0.05374 to 0.1027, d.f. = 82.05; 37 °C versus 37 °C + 2 mM Ca²⁺ P_O P value = 0.2374, 95% CI of difference −0.0107 to 0.0578, d.f. = 118.8. Columns represent the mean; error bars (red) represent s.e.m. ****P < 0.0001; NS, not significant.

Extended Data Fig. 5. Additional hGluD2 recordings.

All recordings were performed with a −80 mV holding potential, 37 °C, and 10 s is shown from each example recording unless noted. Also, unless otherwise specified, recordings are from a DPHPC bilayer. a, Example five minute recording of a bilayer with hGluD2 with no ligand treatment (black, n = 29) or a bilayer without hGluD2 after vehicle (GDN buffer) treatment (orange, n = 42). b, Inset i, example recording from hGluD2 in bilayers in the presence of 10 mM GABA. Inset ii, current histogram of hGluD2 GABA responses fit with three components (n = 33 individual traces, R² = 0.9898). Over each gaussian component is the corresponding conductance ± SD. c, Example recording from bilayers with hGluD2 in the presence of 10 mM glycine (n = 10). d, Example recording from bilayers with hGluD2 in the presence of 10 mM glutamate (n = 6). e, Example recording of 100 µM NASPM co-applied with 10 mM D-serine (pink, n = 28) overlayed with 10 mM D-serine treatment in the absence of NASPM (black). f, Recordings from bilayers made of brain lipid total extract. Inset i, five minute recording with hGluD2 with no ligand treatment (black, n = 28) or a bilayer without hGluD2 after vehicle (GDN buffer) treatment (orange, n = 39). Inset ii, example recording in the presence of 10 mM D-serine (n = 26). Inset iii, current histogram of brain lipid bilayer recordings from hGluD2 in the presence of D-serine, displayed as in panel b (R² = 0.9759). g, Recordings from hGluD2_R530k in DPHPC bilayers – inset i, example recording in the presence of 10 mM D-serine (n = 10); inset ii, example recording in the presence of 50 mM D-serine (n = 15; five repeats shown overlayed). h, whole cell recordings of rat GluD2. Inset i, example recording of cells expressing rat GluD2 exposed to 10 mM D-serine at 25 °C, stepped to 37 °C, and stepped back down to 25 °C (n = 5). Inset ii, example recording of lurcher (A654T) rat GluD2 at 25 °C, stepped to 37 °C, and stepped back down to 25 °C (n = 5).

To test whether physiological temperatures affect hGluD2 function, we recorded currents at room temperature or 22 °C (Fig. 2b and Methods). Whereas the mean conductance at 22 °C is 29.8 ± 16.7 pS (n = 42) and not significantly different than at 37 °C (P > 0.05; Fig. 2d), the P_O is reduced more than eightfold to 1.03% from 8.85% (P < 0.0001; Fig. 2e). This change in P_O reflects a temperature coefficient (Q₁₀) of 4.2 (Methods). This is a marked difference from AMPARs and the other iGluRs^39,40, which have a Q₁₀ in the range of 1–2. Therefore, we hypothesize that a possible reason for the difficulty in detecting GluD currents is that they are uniquely temperature sensitive amongst the iGluRs.

Because the LBDs in apo hGluD2 do not maintain the back-to-back dimer arrangement that was observed for calcium binding in isolated LBDs (Extended Data Fig. 4d,e), we investigated the effects of 2 mM calcium on hGluD2 currents. This is of interest because calcium was demonstrated to potentiate constitutive currents through Lurcher GluD2²⁷. In the presence of calcium, we see a modest increase in mean conductance to 36.5 ± 19.6 pS (n = 15, P < 0.0001 versus hGluD2 in the absence of calcium; Fig. 2d). The increase in conductance compared with recordings in the absence of calcium agrees with previous observations of increased potentiation of Lurcher GluD2 currents and may reflect calcium permeation²⁷. Consistent with this idea, the hGluD2 P_O in the presence of d-serine with calcium is not significantly affected (6.5%, P > 0.05 versus hGluD2 in the absence of calcium; Fig. 2e).

We also tested GABA with hGluD2 because it has been identified as a potential ligand for GluD1³. We find that 10 mM GABA also elicits currents from hGluD2 (Extended Data Fig. 5b, n = 33). Whereas the mean conductance in the presence of GABA is increased to 37.3 ± 19.9 pS compared with d-serine (P < 0.001, 95% confidence interval (CI) of difference 5.910 to 7.330, d.f. = 10,570), we find that the P_O is not significantly affected (P_O = 6.7%, P > 0.05, 95% CI of difference −0.0444 to 0.0022, d.f. = 125.1). However, although glycine has been demonstrated to bind to the GluD2 LBD²⁶, we do not observe responses from hGluD2 in the presence of glycine (n = 10; Extended Data Fig. 5c) or glutamate (n = 6; Extended Data Fig. 5d), which is understood to not be a ligand for GluD receptors¹.

To further assess hGluD2, we measured channel block by 1-naphthyl acetyl spermine (NASPM), which has been demonstrated to reduce currents from GluDs^26,27,41. Accordingly, we find that hGluD2 currents are blocked by NASPM, as demonstrated by reduction of currents following co-application of 100 µM NASPM with 10 mM d-serine (Extended Data Fig. 5e). This is reflected in a rightward shift of the current histogram during the co-application (Fig. 2c), reducing the mean conductance to 17.4 ± 3.3 pS (P < 0.0001 versus normal d-serine currents, 95% CI of difference −14.00 to −12.52, d.f. = 3050).

We also tested whether DPHPC bilayers are required for hGluD2 activity by reconstituting hGluD2 in brain lipid bilayers. hGluD2 also exhibits ligand-gating activity in these bilayers, demonstrating that a specific bilayer composition is not required for ion channel activity from purified hGluD2 (Extended Data Fig. 5f). In brain lipid bilayers, the mean conductance is 37.8 ± 21.3 pS (P < 0.0001 versus DPHPC bilayers, n = 26, 95% CI of difference 6.236 to 8.030, d.f. = 5195), and P_O is marginally decreased to 6.4% (P < 0.05, 95% CI of difference −0.0464 to −0.0027, d.f. = 101.9), which may indicate that lipids have a role in regulating GluD2.

Further confirmation of a hGluD2-specific response was tested by mutating the LBD via R530K, which reduces d-serine’s binding affinity²⁷. Indeed, we do not observe responses to 10 mM d-serine from hGluD2(R530K) but do observe responses in the presence of 50 mM d-serine (Extended Data Fig. 5g). These data directly demonstrate that the ligand gating that we observe is from hGluD2. Thus, across all conditions and bilayer compositions, hGluD2 retains ligand-gated activity, pharmacology, P_O and mean conductance that are characteristic of iGluRs, except for the pronounced sensitivity to temperature.

Because GluD2 maintains robust ligand-gating activity in isolation, this supports the idea that GluD currents are tightly regulated in cells, and mutagenesis (for example, the Lurcher mutation) relieves the inhibition by allosteric means^14,28–33. To confirm this, we recorded currents recorded at 37 °C from cells expressing rat GluD2 exposed to 10 mM d-serine (Methods). We do not observe ligand-gated responses in wild-type GluD2 but do observe currents from Lurcher GluD2 that are higher at physiological temperatures (Extended Data Fig. 5h). These data thus agree with our findings of temperature augmentation in bilayers, and with previous findings that GluD ion channel activity is tightly regulated in certain cellular contexts^{11,13,28–33}.

Cryo-EM of activated hGluD2

We prepared hGluD2 for cryo-EM at physiological temperature (37 °C) in the presence of 10 mM d-serine and solved its structure (Extended Data Figs. 6 and 7 and Methods). Overall, activated hGluD2 has a similar overall topology to apo hGluD2, with the exception that the LBD of the D subunit is less flexible and directly contacts the LBD of the C subunit (Fig. 3a) because of d-serine binding (Fig. 3b). All LBDs are bound to d-serine. The features of the cryo-EM map enable definitive characterization of the d-serine binding site (Fig. 3c). The site is identical to the site in the isolated rat GluD2 LBD as determined by X-ray crystallography, and mutation of these residues (for example, R530K) reduces the binding affinity of the LBD to d-serine^{25–27,42,43} (Extended Data Fig. 5g).

Extended Data Fig. 6. Image processing workflow for D-serine bound hGluD2.

Workflow in Cryosparc for D-serine bound hGluD2.

Extended Data Fig. 7. Map and local qualities for activated hGluD2.

a, GSFSC curves for each whole and local map. Black line is Fourier shell correlation FSC = 0.143. Y axis is FSC, X is resolution in Å. b, Heat maps of particle orientation distribution for each map. c, Local resolution maps computed for each voxel, computed FSC = 0.143. d, Pore model fits for each subunit.

Fig. 3. Structure of activated hGluD2.

a, Composite map of hGluD2 bound to d-serine, with the unsharpened map overlaid. b, Slice through the map in the subunit D and C LBD dimer to the d-serine-binding sites. c, Structural model of the d-serine binding site, with the local cryo-EM map overlaid. d, Model of the ion channel pore in the presence of d-serine, the A subunit is omitted for clarity (left) and rotated 90° (right) with all pore helices. e, Pore radius profile. The dashed line at 1.4 Å represents the radius of a water molecule.

As a result of d-serine binding, there are changes in the hGluD2 ion channel (Fig. 3d). Whereas the M3 helices in subunits A and B resemble those in the resting state of the channel, the M3 helices in subunits C and D hinge away from the central pore axis. In subunit C, this hinging occurs in SYTANLAAF at A654 (the underlined residue indicates the Lurcher site), and in subunit D, the hinging occurs at L652 (underlined) in SYTANLAAF. We call these sites the Lurcher hinge (A654) and the L-hinge (L652). As in the resting state, the selectivity filter begins at Q618.

The major consequence of the M3 helix hinging in subunits C and D is opening of the upper gate of the channel (Fig. 3e). Hinging in subunits C and D removes the major constriction via T657, dilating this site (formerly 0.9 Å) to a radius of 2.2 Å. Radii below this site in SYTANLAAF are similarly dilated at A653 and T649 to 1.8 Å and 2.1 Å, respectively. Thus, we expect the channel to be open. This is supported by an identical permeation pathway in CP-AMPAR ion channels that conducts ions even when more restricted⁴⁴.

Activation mechanism

To approximate the activation mechanism, we compared hGluD2 in the apo and d-serine-bound states. Both of these structures maintain a similar, italic Y shape (Fig. 4a). We observe minor changes in the ATD between the states, with the local ATD dimers being maintained as a rigid body but rotating clockwise by 7° (Fig. 4b). Each LBD closes by 16° around d-serine (Fig. 4c), similar to d-serine binding in crystal structures of isolated GluD1 and GluD2 LBDs^3,26 (Extended Data Fig. 4c).

Fig. 4. Activation mechanism.

a, Activated hGluD2 with the apo hGluD2 overlaid in grey. b, Magnified view of the local A–B ATD dimer and the rigid body conformational change between states. c, How d-serine induces clamshell closure in the activated state compared to the apo state. d, Change in the ATD orchestration of the LBD layer upon activation by d-serine. e, The formation of the D–C LBD dimer that orchestrates activation. f, Top-down view of the resting (left) and activated (right) ion channel pore.

The major change during activation by d-serine is the orchestration of the LBD layer. Whereas the subunit A ATD still coordinates the entire LBD layer in each state (Fig. 4d), the centre of mass of the subunit D LBD shifts by 5 Å, which places the subunit D LBD in direct contact with the subunit C LBD. The centre of mass of each other subunit LBD remains relatively stationary. Thus, in the d-serine bound state, an asymmetric, tetrameric LBD is formed around the subunit A ATD, with all LBDs arranged front-to-back relative to one another (Fig. 4d).

The formation of the subunit C and D LBD dimer exhibits the most marked change during activation. The subunit C LBD rotates counterclockwise by 4°, whereas the subunit D LBD rotates clockwise by 19° (Fig. 4e). This places the back of the subunit D LBD in contact with the front of the subunit C LBD. Because of this, the D1–D1 distance (measured by L482–L482) decreases by 11 Å, and the D2–D2 distance decreases by approximately 3 Å (measured by Y734–Y734). This collective change, in addition to LBD clamshell closure around d-serine, results in the D-LBD shifting 14 Å upward (Fig. 4e).

As a result, the ion channel below the LBD is dilated. Although the subunit A and B M3 helices do not exhibit major changes, because of the changes in subunit C and D LBDs, the M3 helices in these subunits unwind to the Lurcher hinge and L-hinge, respectively (Fig. 4f). This hinging of helices in two adjacent subunits for channel opening is distinct from the other iGluR subtypes, which hinge at all four M3 helices or at helices in two opposite subunits for channel opening^39,45,46.

Role of the ATD

The ATD has a substantial contact surface with the subunit A, B and C LBDs (Extended Data Fig. 8a) and orchestrates the overall architecture of hGluD2. To directly test the effects of the ATD, we deleted the ATD from hGluD2 (hGluD2(ΔATD)). We tested hGluD2(ΔATD) response to ligand in bilayers and determined its structure (Extended Data Figs. 9 and 10 and Methods). hGluD2(ΔATD) has similar d-serine responses to hGluD2 (Extended Data Fig. 8b). However, the current histogram is shifted leftward, reflecting a mean conductance of 35.7 ± 19.3 pS (Extended Data Fig. 8c).

Extended Data Fig. 8. Functional and structural role of the hGluD2 ATD.

a, Overview of potential contacts (residues within 5 Å of each other) between the A-ATD and A through C LBDs. b, i, 10 s recording of hGluD2_ΔATD currents in response to 10 mM D-serine. ii, the hGluD2_ΔATD current recording overlayed with the hGluD2 D-serine current recording presented in Fig. 2b. c, Histogram of hGluD2_ΔATD currents in the presence of D-serine fit with three components (n = 21 individual traces, R² = 0.9814). Conductance levels ± S.D. corresponding to each gaussian component are labeled above. d, Composite cryoEM map of hGluD2_ΔATD. e, Slice through the local LBD map of hGluD2_ΔATD. f, Slice through the TMD of hGluD2_ΔATD. g, Pore radius profile of hGluD2_ΔATD. The dashed line represents the 1.4 Å radius of a water molecule.

Extended Data Fig. 9. Image processing workflow for apo hGluD2_ΔATD.

Workflow in Cryosparc for apo hGluD2_ΔATD.

Extended Data Fig. 10. Map and local qualities for apo hGluD2_ΔATD.

a, GSFSC curves for each whole and local map. Black line is Fourier shell correlation FSC = 0.143. Y axis is FSC, X is resolution in Å. b, Heat maps of particle orientation distribution for each map. c, Local resolution maps computed for each voxel, computed FSC = 0.143. d, Pore model fits for each subunit.

The architecture of the hGluD2(ΔATD) LBD is rearranged, but the TMD is maintained (Extended Data Fig. 8d). Local dimers in the LBD are arranged back-to-back between subunits A and B, and between subunits C and D (Extended Data Fig. 8e). The TMD is arranged similarly to hGluD2 (Extended Data Fig. 8f), and the channel is closed (Extended Data Fig. 8g). Because of the rearrangement of the LBD, hGluD2(ΔATD) adopts an ultrastructure that is reminiscent of the LBD and TMD from AMPARs (Extended Data Fig. 4h).

The AMPAR-like architecture of hGluD2(ΔATD) suggests potential effects of trans-synaptic factors on GluD2. For example, binding of CBLN1 to the GluD2 ATD could lift the ATD from the LBD layer, enabling the rearrangement of the receptor that we observe, augmenting ligand response. This indeed seems possible, given that the crystal structure of the CBLN1–GluD2 ATD fusion⁹ fits into our full-length hGluD2 without clashes (Extended Data Fig. 4g).

Ataxia mutation at the LBD–LBD interface

The front-to-back LBD arrangement is fundamental for the architecture of hGluD2 (Fig. 1c). A mutation linked to cerebellar ataxia^38,47 (R710W) in hGluD2 is located at this interface between the A and B subunit LBDs in hGluD2 (Fig. 5a). To test the functional and structural consequences of this mutation, we mutagenized and purified the mutant receptor (hGluD2(R710W)).

Fig. 5. The R710W cerebellar ataxia mutation.

a, Location of the R710W mutation site in wild-type hGluD2 (between A and B subunits only). Inset shows a magnified view of the interface. b, Ten-second recordings of hGluD2(R710W) leak currents in the absence of ligand. c, Histogram of GluD2(R710W) leak currents fit with three components (n = 23 individual traces, R² = 0.9552). Conductance ± s.d. corresponding to each Gaussian component is shown above the bars. d, Left, cryo-EM map of hGluD2(R710W)_apo-leak with the unsharpened map overlaid. Right, views or slices that are viewed perpendicular to the demarcations on the left. Demarcation iv is shown in f, left. The area outlined in orange is the location of the front-to-back local dimer interface containing R710W. e, View of front-to-back local LBD dimer interfaces between subunits D and C (left) and subunits A and B (right). f, Changes in pore shape. Arrangement of the M3 helices in hGluD2(R710W)_apo-leak (perpendicular to demarcation iv in d) (left) and dilation of the pore from hGluD2(R710W)_apo-closed to hGluD2(R710W)_apo-leak (right).

In the absence of ligand, hGluD2(R710W) exhibits spontaneous leak currents (Fig. 5b). Although the current histogram resembles that of wild-type hGluD2 (Fig. 5c), it is shifted leftward, reflecting an increase in the mean conductance to 39.8 ± 22.0 pS. Notably, there is an unfit population left of the third component that may represent the fourth (O4) conductance state observed in AMPARs (Fig. 5c). Thus, R710W appears to be a gain-of-function mutation.

The apo state of hGluD2(R710W) exists in two conformations (Extended Data Fig. 11): closed (hGluD2(R710W)_apo-closed; reconstructed to 3.68 Å overall) and open (hGluD2(R710W)_apo-leak; reconstructed to 3.73 Å overall) (Extended Data Fig. 12a–c and Supplementary Video 1). hGluD2(R710W)_apo-closed is identical to the hGluD2 resting state (RMSD ≈ 1.0 Å; Extended Data Fig. 12d), hGluD2(R710W)_leak is markedly different in its architecture (Fig. 5d). Because hGluD2(R710W)_apo-closed resembles hGluD2_apo (which does not exhibit leak current), we attribute the leak current that we observe to the second state and name it hGluD2(R710W)_apo-leak.

Extended Data Fig. 11. Image processing workflow for apo hGluD2_R710W.

Workflow in Cryosparc for apo hGluD2_R710W.

Extended Data Fig. 12. hGluD2_R710W map quality and patient mutations in hGluD2.

a, GSFSC curves for each whole and local map. Black line is Fourier shell correlation FSC = 0.143. Y axis is FSC, X is resolution in Å. b, Heat maps of particle orientation distribution for each map. c, Local resolution maps computed for each voxel, computed FSC = 0.143. d, Comparison of hGluD2 to hGluD2_{R710W-apo-closed}. e, Mutations compiled and reported in Ref. ³⁸ mapped onto hGluD2. Insets are areas of interest highlighted in panels f-h. f, Mutations that occur at the hGluD2 channel gate. g, Mutations that occur in the ATD region that coordinates the LBD layer. h, Mutations that occur in the front-to-back LBD dimer interface.

The ATD in hGluD2(R710W)_apo-leak is arranged into a dimer of dimers, between the A–B and C–D local dimers between the A and D subunits (Fig. 5d(i)). Both ATD dimers directly coordinate the LBD layer, where the subunit A ATD is in contact with the subunit A, B and D LBDs, and the subunit D ATD coordinates the subunit A, C and D LBDs (Fig. 5d(ii)). The LBD is arranged into two front-to-back LBD local dimers between subunits A and B and between subunits C and D (Fig. 5d(iii)). The LBD interfaces are mediated by R710W in subunits B and C with H802 and N808 in subunits A and D, respectively (Fig. 5e).

The ion channel gate is held in a rhomboid-like shape (Fig. 5f, left). This is achieved by a 14 Å separation between the centre of the subunit A and B and subunit C and D M3 helices, and 12 Å separation of the respective subunit C and A and subunit D and B M3 pairs, effectively dilating the pore from hGluD2(R710W)_apo-closed (Fig. 5f, right). Therefore, we hypothesize that the R710W mutation enables constitutive currents through formation of the C and D front-to-back dimer in hGluD2(R710W)_apo-leak.

Discussion

Our study confers the long-sought ‘license to conduct’ to GluDs, unifying longstanding discrepancies. By demonstrating that purified hGluD2 is intrinsically a ligand-gated ion channel, we provide direct evidence that GluDs possess the molecular machinery necessary for ion conduction and ligand gating, placing them firmly alongside the canonical iGluR subtypes.

The ATD tightly coordinates the hGluD2 LBD, akin to N-methyl-d-aspartate-type iGluRs (NMDARs), in which the ATD has an allosteric role in regulating the ligand response^48–51. This sheds light on how trans-synaptic factors (for example, neurexin–CBLN1) may influence signal transduction by GluDs^1,2,9,19.

In hGluD2, the ATD arranges the LBDs into front-to-back dimers. Similar arrangements occur between dimers in NMDARs and kainate-type iGluRs (KARs), in which disruption of the front-to-back interface leads to desensitization^45,52,53. These insights from NMDARs and KARs could point to a similar mechanism being conserved in GluDs.

A limitation of this study is our focus on hGluD2 in a largely in vitro context. Although this enabled direct investigation of the channel function, recordings from cells at 37 °C show that wild-type GluD2 currents are not readily observed in cellular environments, consistent with prior reports. This reframes a major question for the field: rather than whether GluDs are ligand-gated ion channels, the central challenge is to understand how their activity is regulated within the cellular context. The inability to record currents from wild-type GluDs in heterologous systems probably reflects the action of endogenous regulatory mechanisms. Notably, several studies have implicated G-protein signalling cascades in the activation of GluD currents in situ, with these currents inhibited by NASPM and d-serine^29–34. Together with our findings, this points towards a cytosolic co-factor that is potentially coupled via G-protein pathways, which tightly controls GluD2 ion channel gating, masking its intrinsic activity. Further identification and characterization of such regulatory mechanisms remain a critical avenue for future research and will be essential for resolving the physiological role of GluDs in the brain.

Cytosolic factors have regulatory effects in iGluRs; calmodulin binding to NMDARs negatively modulates the conformation of the NMDAR ECD⁵⁴. In GluDs, a similar phenomenon may be possible, given that mutagenesis or alteration of the GluD2 ECD or TMD (for example, the Lurcher mutation) enables currents to be recorded^13,27,55. Deconvoluting the regulatory mechanisms in situ will require additional studies and is an important future direction for the field. An additional complication in recording currents from GluDs is the likely requirement for physiological temperatures.

The structures of hGluD2 provide a foundation for understanding mutations that are associated with human disease, and in particular cerebellar ataxia³⁸ (Extended Data Fig. 12e). Several mutations (for example, T649A, A654T, A654D and L656V) are located directly in the upper region of M3 and likely have direct effects on gating or selectivity (Extended Data Fig. 12f). Site A654 is the Lurcher hinge, where subunit C hinges during activation (Fig. 3d), and T649 helps to define the permeation pathway (Figs. 1f and 3e). A separate group of mutations (Extended Data Fig. 12g), highlighted by R224Q, occur at the site on R2 of the ATD that is critical for coordination of the LBD (Extended Data Fig. 8a). Given the role of the ATD in hGluD2, these mutations are likely to alter the ATD–LBD coordination. Another site of interest occurs on D2 in the LBD (Extended Data Fig. 12h), highlighted by R710W (Fig. 5). One hypothesis is that the Lurcher mutation enables similar spontaneous activity in GluD2 through a structural rearrangement similar to the one in GluD2(R710W).

It remains to be seen whether GluD1 has intrinsic ligand-gating capabilities like GluD2. Our expectation is that it does, considering the sequence conservation between the subtypes. However, comparison to the cryo-EM structure of rat GluD1 shows that GluD2 and GluD1 have markedly different topologies (Extended Data Fig. 13).

Extended Data Fig. 13. hGluD2 compared to GluD1.

Overall architectural features of GluD1 (left, pdb 6kss) compared to apo hGluD2 (right, this study). Structures are shown as surfaces, with slices through each layer.

Methods

Construct design

Full-length hGluD2 (UniProt ID O43424-1) was fused to a Thr-Gly-Gly linker, a thrombin cleavage site (LVPRGS), an enhanced GFP (eGFP), a Ser-Gly-Leu-Arg-Ser linker, a Strep-Tag II (WSHPQFEK) and a stop codon at its C terminus (detailed in Extended Data Fig. 1a). The construct was introduced into a pEG BacMam vector for baculovirus-driven protein expression in HEK293 cells⁵⁶. hGluD2(ΔATD) was prepared by deleting residues 27–430 from the hGluD2 construct. Both plasmids were synthesized by Twist Bioscience. For hGluD2(R710W) and hGluD2(R530K), site directed mutagenesis was performed by PCR on the hGluD2 construct.

Expression and purification

The hGluD2, hGluD2(ΔATD), hGluD2(R710W) and hGluD2(R530K) bacmids and baculoviruses were made using standard methods⁵⁷. The bacmids were transfected into Expisf9 cells (Gibco, A35243) at 2.5 million cells per ml concentration using ExpiFectamine Sf transfection reagent (Gibco, A38915). The transfected cells were cultured at 27 °C with constant shaking to generate P1 baculovirus. After 5 days, supernatant of the cultured cells containing P1 baculovirus were collected and stored in 4 °C fridge until further use. To increase the virus titre, P2 virus was further gathered. P1 virus (500 μl) was added to 400 ml of fresh Expisf9 cells at a concentration of 10⁶ cells per ml. The infected cells were incubated at 27 °C with constant shaking. The supernatant containing P2 virus was isolated after 5 days and was further extracted via ultracentrifugation (41,300g for 60 min at 4 °C). The pelleted P2 virus was then resuspended with 40 ml of Expi293 expression media (Thermo Fisher, A1435102) and stored at 4 °C until further use.

For protein expression in mammalian cells, 20 ml of concentrated P2 virus was added to Expi293F GnTI- cells (Gibco, A39240) at a density of 3–4 million cells per ml and incubated at 37 °C in 5% CO₂ with constant shaking. At 12–24 h post-infection, the infected Expi293F cells were treated with 10 mM sodium butyrate (Sigma, 303410) and moved to a 30 °C, 5% CO₂ incubator. The cells were then collected at 72 h post-infection by centrifugation (500g for 20 min at 4 °C), washed with 1× PBS and pelleted again (500g for 20 min at 4 °C). The cell pellets were snap-frozen in liquid nitrogen and stored at −80 °C until further use.

For purification, the frozen cell pellets were resuspended at 4 °C in lysis buffer (20 mM Tris-HCl, pH 8.0 and 150 mM NaCl) with addition of protease inhibitors (0.8 µM aprotinin, 2 µg ml⁻¹ leupeptin, 2 µM pepstatin A and 1 mM phenylmethylsulfonyl fluoride). The cells were then lysed on ice with a Misonix sonicator for four cycles (1 s on and 1 s off for 1 min, power level 7). The lysate was clarified by centrifugation (2,500g for 20 min at 4 °C) and the supernatant was further ultracentrifuged to pellet cell membranes (125,000g for 45 min at 4 °C). The membranes were then resuspended in lysis buffer (20 mM Tris-HCl, pH 8.0 and 150 mM NaCl) and mechanically homogenized before solubilized in solubilization buffer (20 mM Tris-HCl, pH 8.0 and 150 mM NaCl, 1% n-dodecyl-β-d-maltopyranoside (Anatrace, D310) and 0.2% brain extract total (Avanti Research, 131101P)) for 2 h at 4 °C with constant stirring. Insoluble material was removed by ultracentrifugation (125,000g for 45 min at 4 °C). The supernatant containing solubilized protein was incubated with 1 ml of Strep-Tactin XT 4Flow resin (IBA, 2-5010) overnight at 4 °C with constant stirring. The following day, the resin was collected via a gravity column and washed with 10 column volumes of glyco-diosgenin (GDN) buffer (150 mM NaCl, 20 mM Tris pH 8.0 and 0.01% GDN (Anatrace, GDN101)). Bound protein was eluted with GDN buffer containing 50 mM biotin (Thermo Scientific, PI29129), collected and concentrated down to 500 µl at 4 °C using a 100 kDa molecular weight cut-off concentrator (Millipore Sigma, UFC9100). The concentrated sample was then digested with thrombin (1:200 mass ratio of thrombin to eluted protein) for 1 hr at room temperature. The cleaved protein was further separated via size-exclusion chromatography on a Superose 6 Increase 10/300 GL column (Cytiva, 29091596) equilibrated with the GDN buffer. The peak fractions corresponding to hGluD2 tetramers were pooled and concentrated up to approximately 4 mg ml⁻¹ (Millipore Sigma, UFC5100) for cryo-EM specimen preparation. The sample was further ultracentrifuged to remove insoluble material after concentration.

Fluorescence-detection size-exclusion chromatography

The pooled fractions of purified protein samples were further examined via fluorescence-detection size-exclusion chromatography⁵⁸ on a high-performance liquid chromatography system equipped with a multi-wavelength fluorescence detector, an autosampler (Shimadzu, 664 SIL40C) and a Superose 6 Increase 10/300 GL SEC column (Cytiva). The tryptophan fluorescence method (excitation at 280 nm and emission at 325 nm) was used to monitor the retention time. Five microlitres of each concentrated sample was diluted into 50 μl with the GDN buffer before loading onto the column. The flow rate was set at 0.5 ml min⁻¹.

Cryo-EM sample preparation and data collection

Home-made grids were used in preparing the samples in this study. In brief, C-flat holey carbon grids (Electron Microscopy Sciences, CF213-50-Au, Electron Microscopy Sciences) were first coated with 50 nm Au by a Leica EM ACE600 sputter coater and then plasma cleaned by Tergeo Plasma Cleaner (Pie Scientific) with Ar/O₂ to remove carbon and make final 1.2/1.3 gold grids with gold mesh, based on previously published methods^59,60. The grids were glow discharged for 120 s with 25 mA current and 10 s hold time in a Pelco Easiglow (Ted Pella, 91000) before sample application.

For the d-serine bound hGluD2, 20 μl of the concentrated hGluD2 sample was first pre-incubated in a thermocycler set at 37 °C for 10 min before use, along with d-serine stock (70 mM). Then, a 3-μl protein sample was spiked with 0.5 μl of the 70 mM d-serine and mixed vigorously (final concentration of d-serine:10 mM) before applied to freshly-discharged grids in a FEI Vitrobot Mark IV (Thermo Fisher Scientific) chamber set at 37 °C and 100% humidity (wait time 30 s; blot force 5; blot time 4 s) and immediately plunge-frozen in liquid ethane. For the apo hGluD2(ΔATD) and apo hGluD(R710W) sample conditions, 3 μl of the pre-incubated (at 37 °C) protein sample was directly applied onto grids in the Vitrobot chamber set at 37 °C and 100% humidity. For the apo hGluD2 sample condition, the protein was kept on ice instead before applied to grids in the Virtobot chamber set at 4 °C and 100% humidity.

All the grids except one sample condition (hGluD2(ΔATD)) were imaged on a 300-kV Titan Krios G3i microscope (Thermo Fisher Scientific) equipped with fringe-free imaging, a Falcon 4i camera with direct electron detector and a Selectris energy filter set to a 10-eV slit width. 10,822 micrographs were collected for apo-state hGluD2 sample dataset with a total dose of 45.00 e⁻ Å⁻², a dose rate of 6.98 e⁻ pixel⁻¹ s⁻¹ and a pixel size of 0.76 Å, and the defocus range was set from −0.9 μm to −2.5 μm. For d-serine-bound hGluD2 sample dataset, 8,371 micrographs were collected with a total dose of 40.00 e⁻ Å⁻², a dose rate of 7.51 e⁻ pixel⁻¹ s⁻¹, and a pixel size of 0.97 Å pixel⁻¹, and the defocus range was set from −0.5 μm to −2.0 μm. For the apo hGluD2(R710W) sample dataset, 8,712 micrographs were collected on Titan Krios with a total dose rate of 40.00 e⁻ Å⁻², a dose rate of 10.05 e⁻ pixel⁻¹ s⁻¹, a pixel size of 0.97 Å and the defocus range was set from −0.6 μm to −2.4 μm. For the apo hGluD2(ΔATD) sample dataset, 4,572 micrographs were collected on a 200-kV Glacios microscope (Thermo Fisher Scientific) equipped a Falcon 4i camera. The micrographs were collected with a total dose of 40.00 e⁻ Å⁻², a dose rate of 7.99 e⁻ pixel⁻¹ s⁻¹ with a pixel size of 1.2 Å, and the defocus range was set from −0.9 μm to −2.5 μm.

Image processing

CryoSPARC⁶¹ v.4.6.0 was used for processing apo-state hGluD2 sample images, and CryoSPARC v.4.6.2 was used to analyse d-serine-bound hGluD2 sample, apo hGluD2(ΔATD) and hGluD2(R710W) datasets. Final particle picking was performed with TOPAZ⁶². Details can be found in Extended Data Figs. 2, 6, 8 and 11.

Extended Data Fig. 2. Image processing workflow for apo hGluD2.

Workflow in Cryosparc for apo hGluD2.

Model building, refinement, and structural analysis

ChimeraX⁶³, Coot⁶⁴, ISOLDE⁶⁵ and PHENIX⁶⁶ compiled by the SBgrid Consortium⁶⁷ were used in combination to perform model building, refinement, and structural analysis. Model building was initiated with a predicted structure of a hGluD2 monomer from the AlphaFold Protein Structure Database (AF-O43424-F1-v4). All visualizations and measurements were performed in ChimeraX. Model quality was assessed with MolProbity⁶⁸. Pore measurements were performed with HOLE⁶⁹. Model quality is reported in Extended Data Table 1.

Extended Data Table 1.

CryoEM data collection, refinement, and validation statistics

Collection, refinement, and validation statistics for all hGluD2 cryoEM data, maps, and corresponding models from this study.

Bilayer recording and analysis

An Orbit mini with temperature control unit (Nanion Technologies) and MECA 4 recording chips (recording chip with 4× 100 µm cavities, Nanion Technologies, 132002) were used for all electrophysiological measurements of purified hGluD2_, hGluD2(ΔATD), hGluD2(R710W) and hGluD2(R530K). Our protocol was adapted from established methods^44,70. First, the Orbit mini was calibrated with a standard test cell chip (Nanion Technologies) after each round of hardware/software start up. A recording chip was then filled with recording solution (HEPES, pH 7.2, 150 mM KCl) in the holding chamber and plugged into Orbit mini. A thin lipid bilayer was painted over each recording cavities located at the bottom of the chamber, using lipid-covered air bubble dispensed by gently pipetting a 10-µl tip previously dipped in the lipid stock solution closely to the cavities. The lipid stock solution was 10 mg ml⁻¹ 1,2-diphytanoyl-sn-glycero-3-phosphocholine (DPHPC, Fisher Scientific, NC0652354) or 100 mg ml⁻¹ brain lipid extract total (Avanti, 131101 C) in decane (Simga-Aldrich, D901-100ML). The capacitances of painted bilayers were kept in the range of 10–25 pF. Purified apo protein was diluted 10,000 times from the stock (~ 4 mg ml⁻¹) into the recording solution, and a total of 1 µl of diluted protein solution was added directly on top of each painted recording cavities via a 10-µl pipette. All the recordings were performed at −80 mV holding potential and a maximum applied current rate at 200 pA. Data were recorded at 1.25 kHz sampling rate using Elements Data Reader 4 (EDR4, Elements) software v.1.7.1. For the hGluD2 and hGluD2(ΔATD) with ligands conditions, 10 mM d-serine or GABA was added before the recording started. For the block condition, 100 µM of NASPM was co-applied with ligand. For recording hGluD2 activity with CaCl₂, the recording solution (HEPES, pH 7.2, 150 mM KCl) was supplemented with additional 2 mM CaCl₂. For the vehicle control, 1 µl of diluted GDN buffer (1:10,000 in the recording buffer) was added to the chip sample chamber and recorded for the indicated length of time.

Individual recorded traces were analysed in Clampfit software version 11.3 (Molecular Devices). These traces were first filtered using an 8-pole Bessel filter with a −3 dB cut-off frequency of 50 Hz and then adjusted for baseline. Only regions of traces containing currents with amplitude within 0 to −10 pA range were subjected to further examination. Single-Channel Search function with 5 detection levels (−2, −4, −6, −8 and −10 pA) in Clampfit was used to identify and perform measurements for each individual current event, as 50% amplitude crossing method was used for idealizing current peaks^71,72. Automatic adjustment for baseline and simultaneous update for all levels (with a 10% level contribution) was applied to minimize effects caused by local baseline drift. Any level change lasting less than 2 ms was further excluded from the final dataset to reduce the noise level. Moreover, a 10-s detection window was used for all measurements to standardize calculations and statistical analysis. Selected data were further compiled and calculated in Microsoft Excel and statistical analysis was done in GraphPad Prism10. To generate initial histograms, frequency distribution was analysed for peak amplitudes of all selected events. Nonlinear regression analysis with multi-component Gaussian distribution was performed to generate fitted curves on histograms, using the following equation:

$$\begin{array}{c}Y=[\mathrm{a}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{u}\mathrm{d}\mathrm{e}1\times \exp (-0.5\times {\left(\right(X-\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}1\left)/\mathrm{s.d}.1\right)}^2\left)\right]\hfill \\ +[\mathrm{a}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{u}\mathrm{d}\mathrm{e}2\times \exp (-0.5\times {\left(\right(X-\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}2\left)/\mathrm{s.d}.2\right)}^2]\dots \hfill \end{array}$$

where s.d. is standard deviation. The default rules implemented in Prism 10 were used to compute and validate all the parameters for the curve fitting, defined as the following: amplitude = value × Y_max, mean = value × X-at-Y_max and s.d. = value × (X_max − X_min), where the default initial value is 1.

The equation conductance (G) = event current peak amplitude (I_peak)/holding potential (V_hold) was used to calculate both overall mean conductance values and individual conductance values of each current event. Again, nonlinear regression analysis with multi-component Gaussian distribution was performed to generate fitted curves on histogram distribution and to calculate mean values ± s.d. of each conductance levels. Unmatched Brown–Frosythe and Welch one-way ANOVA test was used to compare mean conductance or mean open probability of different sample groups, assuming variances among the experimental groups were unequal. The Game–Howell post hoc test was further used to examine significance of the difference found among the tested mean values, as sample size of each group⁷³ was larger than 50. For measuring effects of NASPM treatment, a histogram of peak amplitude was generated based on results of 80 detections (10-s window per event detection) each from both treated and untreated samples. Two-tailed unpaired t-test with Welch’s correction was used to compare mean conductance of untreated and treated wild-type GluD2 with NASPM as well as mean conductance and mean open probability of samples in DPHPC and those in brain total lipids, assuming variances between two groups were unequal. For each 10-s event detection, overall open probability P_O were calculated using Clampfit with a default interval setting (100 ms). Overall mean open probability was then calculated for each sample condition and compared among different samples using either unpaired t-test with Welch’s correction or Welch’s one-way ANOVA test in Prism10. The Q₁₀ value related to P_O was calculated with the following equation:

$$Q_{10}={\left(P_{\mathrm{O}-37^{°}\mathrm{C}}/{\mathit{P}}_{\mathrm{O}-22^{°}\mathrm{C}}\right)}^{\left[10/\right(37^{°}\mathrm{C}-22^{°}\mathrm{C}\left)\right]}$$

All the final graphs were plotted using Graphpad Prism10.

Whole-cell recordings

HEK-293T cells were maintained in Dulbecco’s modified Eagle’s medium (GenDEPOT), supplemented with 10% fetal bovine serum (GenDEPOT) and penicillin-streptomycin (Invitrogen/Life Technologies). Transfections were performed using iMFectin Poly DNA Transfection Reagent (GenDEPOT) with rat GluD2 and rat GluD2 Lurcher (A654T mutant) receptors along with green fluorescent protein (GFP), at a 1:0.2 ratio per 35-mm dish as previously established¹³. Patch-clamp experiments were performed in standard whole-cell and outside out configurations using an Axopatch 200B amplifier (Axon Instruments). Patch pipettes had a resistance of 3 to 8 megohms when filled with an internal solution consisting of 135 mM CsF, 33 mM CsCl, 2 mM MgCl₂, 1 mM CaCl₂, 11 mM EGTA and 10 mM HEPES, adjusted to pH 7.4. The extracellular solution consisted of 150 mM NaCl, 1 mM CaCl₂, and 10 mM HEPES, adjusted to pH 7.4. External solutions were locally applied to lifted cells using a SF-77B perfusion fast-step (Warner Instruments). Currents were sampled at 10 kHz and filtered at 2 kHz with a 16-bit analogue–digital converter (Axon Digidata 1550 A, Axons Instruments) using pCLAMP 10 software (Molecular Devices). Whole-cell patch-clamp recordings were performed 24 h after transfection, with a holding potential of −60 mV. As previously established³⁹, temperature control was achieved with a micro heating VAHEAT stage (Interherence), in addition to a glutamate solution maintained in water baths at each temperature step. To verify temperature, the temperature near the patch was recorded with a BAT-12 Microprobe Thermometer (Physitemp Instruments) within 5 mm of the microelectrode tip.

Reporting summary

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

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-025-09610-x.

Author contributions

E.C.T. conceptualized and supervised the project. H.W., F.A. and E.C.T. designed the experiments. H.W. performed protein expression, purification and specimen preparation for cryo-EM. H.W. collected the cryo-EM data. H.W. and E.C.T. processed the cryo-EM data. H.W. and E.C.T. built the molecular models. H.W., A.K.M. and E.C.T. analysed the molecular models. H.W., F.A. and E.C.T. designed the electrophysiology experiments. H.W. and F.A. performed the electrophysiology experiments. H.W., F.A. and J.K. performed all work on the R710W mutation. H.W., F.A. and E.C.T. performed the electrophysiology data analysis. H.W. and E.C.T. wrote the manuscript, which was then edited by all authors.

Peer review

Peer review information

Nature thanks Stephanie Gantz and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer review reports are available.

Data availability

The cryo-EM reconstructions are deposited into the Electron Microscopy Data Bank (EMDB). The whole maps are the primary cryo-EM maps in each deposition and each local map, as applicable, and half maps are supplied as supplemental files in each deposition. All structural coordinates are deposited in the Protein Data Bank (PDB). The accession codes for each state (EMDB, PDB) are: resting hGluD2 (EMD-49888, 9NWO), activated hGluD2 (EMD-49889, 9NWP), resting hGluD2(ΔATD) (EMD-49890, 9NWQ), hGluD2(R710W)_apo-closed (EMD-70667, 9OOO) and hGluD2(R710W)_apo-leak (EMD-70668, 9OOP). All electrophysiology data are included in the main manuscript and extended data.

Competing interests

Johns Hopkins University has filed a patent (US 63/804,774) for methods and compositions to record currents from GluDs on behalf of H.W., F.A., A.K.M. and E.C.T. The authors declare no other competing interests.

Extended data

is available for this paper at 10.1038/s41586-025-09610-x.

Supplementary information

The online version contains supplementary material available at 10.1038/s41586-025-09610-x.