Structural basis of functional transitions in mammalian NMDA receptors

Summary

Excitatory neurotransmission meditated by glutamate receptors including N-methyl-D-aspartate receptors (NMDARs) is pivotal to brain development and function. NMDARs are heterotetramers composed of GluN1 and GluN2 subunits, which bind glycine and glutamate, respectively, to activate their ion channels. Despite importance in brain physiology, the precise mechanisms by which activation and inhibition occur via subunit-specific binding of agonists and antagonists remain largely unknown. Here we show the detailed patterns of conformational changes and inter-subunit/domain reorientation leading to agonist-gating and subunit dependent competitive inhibition by providing multiple structures in distinct ligand states at 4 Å or better. The structures reveal that activation and competitive inhibition by both GluN1 and GluN2 antagonists occur by controlling the tension of the linker between the ligand-binding domain and the transmembrane ion channel of the GluN2 subunit. Our results provide detailed mechanistic insights into NMDAR pharmacology, activation and inhibition, which are fundamental to the brain physiology.

Introduction

Fast excitatory synaptic transmission is the fundamental currency for brain development and high-order brain functions including learning and memory formation. It is largely facilitated by a combination of ionotropic glutamate receptors (iGluRs) including α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors, kainate receptors, and N-methyl-D-aspartate receptors (NMDARs). Synaptic currents and calcium signaling mediated by NMDARs are pivotal to neuroplasticity and are governed by diverse NMDAR subtypes defined by distinct hetero-tetrameric combinations of glycine binding GluN1 (1-4a or b) with either glutamate binding GluN2 (A-D) subunits and/or glycine binding GluN3 (A-B) subunits (Hansen et al., 2018; Paoletti et al., 2013). Dysfunctional NMDARs are implicated in various neurological diseases and disorders including Alzheimer’s disease, depression, stroke, epilepsy, and schizophrenia (Paoletti et al., 2013; Traynelis et al., 2010; XiangWei et al., 2018). Furthermore, recent studies showed that neuroendocrine/ductal pancreatic cancers and breast cancers express NMDARs to mediate signaling for invasive tumor growth and brain metastasis, respectively (Li and Hanahan, 2013; Zeng et al., 2019). Therefore, there are pressing needs to deepen mechanistic understanding of NMDAR functions to facilitate potential development of therapeutic reagents for not only neurological diseases and disorders, but also cancers. The most typical NMDARs contain two copies each of GluN1 and GluN2, which activate upon binding of the co-agonist glycine and the neurotransmitter agonist glutamate (Benveniste and Mayer, 1991; Clements and Westbrook, 1991; Johnson and Ascher, 1987), respectively, with voltage-dependent relief of channel block by magnesium (Mayer et al., 1984; Nowak et al., 1984). NMDARs bind agonists and competitive antagonists at ligand-binding domains (LBDs) while many allosteric modulators bind at either amino terminal domains (ATDs) or LBDs to regulate opening and closing of the ion channel transmembrane domain (TMD) (Figure 1A). There has been progress in understanding the function of NMDARs in the last decade due to availability of structures of both isolated domains (ATDs and LBDs) (Furukawa and Gouaux, 2003; Furukawa et al., 2005; Jespersen et al., 2014; Karakas et al., 2009, 2011; Regan et al., 2019; Romero-Hernandez et al., 2016; Vance et al., 2011; Wang et al., 2020; Yao et al., 2013; Yao et al., 2008) as well as intact NMDAR channels by both X-ray crystallography and electron cryo-microscopy (cryo-EM) (Jalali-Yazdi et al., 2018; Karakas and Furukawa, 2014; Lee et al., 2014; Lu et al., 2017; Regan et al., 2018; Song et al., 2018; Tajima et al., 2016; Zhang et al., 2018; Zhu et al., 2016). All of the structures of agonist-bound intact NMDARs, including GluN1-GluN2B, GluN1-GluN2A, and GluN1-GluN2A-GluN2B NMDARs show extensive ATD-LBD interaction unlike AMPA and kainate receptors where ATD-LBD interactions are minimal, indicating that protein conformational changes in NMDAR ATDs and LBDs are coupled to each other (Hansen et al., 2018; Karakas et al., 2015; Wang and Furukawa, 2019). This structural observation is consistent with the capability of the ATD to control multiple functions including open probability, deactivation speeds, and allosteric inhibition by ATD-targeting compounds (Gielen et al., 2009; Wang and Furukawa, 2019; Yuan et al., 2009).

Figure 1. Domain organization and architecture of NMDARs.

(A) Domain organization of NMDAR subunits. (B) Cryo-EM density of the GluN1b-GluN2B NMDARs bound to glycine and glutamate (the ‘Non-active1’ 3D class) showing the GluN1b-GluN2B ATD heterodimer (orange dashed lines), the GluN1b-GluN2B LBD heterodimer (black dashed lines), and the dimer of the GluN1b-GluN2B LBD heterodimer (cyan dashed lines). (C-E) The structural model of the intact heterotetramer (C), the extracellular domains (D) (ATD top and LBD bottom), and the gate region focusing on the pore forming M3 and M3’ helices (top) and the channel (bottom) (E). A line and a dotted line in (D) show distances between GluN1 α5 and GluN2B α4’ and between GluN1 L2 and GluN2B L1’, which are used to define conformational states of the intact NMDARs. See also Figure S1.

While some patterns of protein conformational alterations between the ATD and the LBD were observed in mid to low range resolution structures (~5-15 Å) (Jalali-Yazdi et al., 2018; Tajima et al., 2016; Zhu et al., 2016), how they actually control the TMD channel is not well defined. With respect to competitive antagonism, previous structures of Xenopus GluN1-GluN2B NMDARs bound to the classical competitive antagonists 5,7-dichlorokynureinic acid (DCKA) and D-2-amino-5-phosphonopentanoate (D-AP5) showed unanticipated robust ‘rupture’ of GluN1-GluN2B LBD dimers strikingly different from the structure of AMPA and kainate receptor competitive antagonist complexes (Zhu et al., 2016). The pattern of LBD coupling to the TMD channel in these structures remains elusive due to low resolution (9-15 Å) where density for the TMDs and the LBD-TMD linker are not resolved. Importantly, there is no clear understanding of how GluN1- and GluN2B-specific competitive antagonists are capable of inhibiting the channel activities while the other subunit is occupied with agonists. Thus, despite advances in NMDAR structural biology, the mechanism leading to channel activation and inhibition remain ambiguous. To overcome this, we improved sample preparation methods to obtain cryo-EM structures of GluN1b-GluN2B NMDARs at overall resolution of 4 Å or better in complex with agonists glycine (Gly) and glutamate (Glu), a GluN1-specific competitive antagonist, L689,560 (L689), and a GluN2-specific competitive antagonist, SDZ-220-040, (SDZ) in the following combinations: Gly/Glu-bound, Gly/SDZ-bound, L689/Glu-bound, and L689/SDZ-bound. The high-resolution structures located antagonists in the LBDs and resolve the majority of side chains in every domain thereby, confidently defining the functional states of the structures and local changes at the gate-region of the TMD channel resulting from agonist- and antagonist-binding and providing a comprehensive mechanistic scheme for functional transitions related to subunit-specific competitive inhibition. Our study provides the blueprint for mechanistic understanding of NMDAR pharmacology, activation, and inhibition.

Results

Improved method for high-resolution structural analysis of intact NMDARs

Structural analysis of multi-domain heteromeric transmembrane proteins from the mammalian sources such as rat GluN1b-GluN2B NMDARs studied here remain technically challenging due to low expression levels, improper multimeric assembly, and instability. To overcome these difficulties, we produced a rat GluN1b-GluN2B NMDAR construct (see Methods) in the Sf9 insect cell/baculovirus expression system where both GluN1b and GluN2B subunits are expressed under the unique combination of the Drosophila Hsp70 promoter and hr1 enhancer element (Rodems and Friesen, 1993) with a p10 3’ untranslated region (UTR) (van Oers et al., 1999) instead of the conventional polyhedrin and p10 late promoters. This modification resulted in production of 0.4 mg of properly assembled GluN1b-GluN2B NMDAR proteins per liter insect cell culture. Sample preparation in lower concentration of lauryl maltose neopentyl glycol (0.002%) than in previous studies (Regan et al., 2018; Tajima et al., 2016) and addition of 0.1% digitonin was crucial for particle integrity in cryo-EM images. For antagonist-bound samples, we observed protein aggregation in electron micrographs when an excessive amount (0.1 mM or above) of L689,560 or SDZ-220-040 was directly added during sample preparation. Thus, instead, we incorporated the antagonists by dialyzing GluN1b-GluN2B NMDAR purified in the absence of ligands against buffers containing 50 μM antagonist. Inclusion of EDTA and adjustment to pH 7.5 were also critical to eliminate ATD-mediated zinc and proton allosteric inhibition to monitor pure competitive inhibition. Finally, we collected large datasets containing 250,000 to 500,000 particle images to observe multiple protein conformations by effective 3D classification in single particle analysis. All of the above collectively contributed to structural analysis at a Fourier Shell Correlation-based resolution as high as 3.6 Å (Table S1), and all of the structures reported here were sufficiently high resolution to resolve cryo-EM densities for antagonists, therefore allowing us to unambiguously define the functional states of individual structures.

Agonist-bound GluN1b-GluN2B NMDARs adopt multiple conformational states

We have previously reported cryo-EM structures of agonist-bound GluN1b-GluN2B, however, due to low resolution (~6.0 Å) those structures did not resolve the linker region between the LBDs and TMDs, preventing assessment of how conformational alteration in the extracellular domains is transduced to the TMD ion channel to regulate activity. Using the improved methods described above, we obtained structures bound to the full agonists, glycine and glutamate, at ~4 Å resolution as estimated by Fourier shell correlation using the 0.143 cutoff (Figures 1 and S1 and Table S1). Here, we observed clear cryo-EM density for all of the domains including the bi-lobed clam shell-like architectures of ATDs (R1/R2) and LBDs (D1/D2), the channel forming TMDs and linker peptides for both GluN1b and GluN2B subunits (Figures 1 and S1). The ATDs and LBDs are assembled as a dimer of the GluN1b-GluN2B heterodimers (Figure 1B; orange and black dashes) whereas the TMDs are assembled as a 1b-2B-1b-2B hetero-tetramer with pseudo-four-fold symmetry with the channel gate formed by the GluN1b-M3 and GluN2B-M3’ helices. Importantly, 3D classification showed three major structures with distinct conformations, which we named ‘Non-active1’, ‘Non-active2’, and ‘Active’, with particle distributions of approximately 41, 40.5, and 18.5%, respectively (Figure S1). The major differences among these 3D classes are the following: 1) subunit arrangement within the GluN1b-GluN2B ATD heterodimers, which can be represented by the distance between GluN1b Lys178 and GluN2B Asn184 in the R2 lobes (α4’-α5 distance) (Figure 1D); 2) relative orientation of the two GluN1b-GluN2B LBD heterodimers, which can be represented by the distance between GluN1b Arg510 in loop1 (L1’) and GluN2B Leu425 in loop2 (L2) in the D1 lobes (L1’-L2 distance) (Figure 1D); and 3) the relative tension of the LBD-M3’ linker of the GluN2B subunits, which directly affects the channel gate and can be measured by the distance between the two GluN2B Gln662 residues from the two GluN2B subunits (Figure 1E). Based on the above criteria, the three 3D classes were extensively compared to one another (Figure 2). Moving from the ‘Non-active1’ to the ‘Non-active2’ conformation, the GluN2B ATD clam shell opens by ~5° but otherwise the structures share similar subunit arrangements in the ATDs (Figure 2A), LBDs (Figure 2B) and TMD, with a closed channel gate (Figures 2C and 2D). In the ‘Active’ conformation, each pair of the GluN1b-GluN2B ATD heterodimers rotate by ~9° compared to ‘Non-active1’ caused by reorientation of GluN2B ATD with respect to GluN1b ATD (Figure 2A). This is subsequently translated to a ‘rolling up’ motion of the GluN1b-GluN2B LBD heterodimers (Figure 2B). This ‘Active’ conformation can be trapped by inter-GluN1b-GluN2B subunit cross-linking at either the ATD or LBD by engineered disulfide bonds (‘Active-SS’) or by methanethiosulfonate crosslinks designed based on our previous low resolution structure and they both result in up-regulation of ion channel activity (Esmenjaud et al., 2018; Tajima et al., 2016).

Figure 2. Conformational variability of agonist-bound NMDARs.

3D classes of the GluN1b-GluN2B NMDAR bound to glycine and glutamate dubbed ‘Non-active1,’Non-active 2,’ and ‘Active’ and the structure of the ‘Active-SS’ construct in magenta (GluN1b) and dark green (GluN2B). Structures of ‘Non-active2,’ “Active,’ and ‘Active-SS’ are aligned to that of ‘Non-active1’ (gray) in panels A-C. (A-B) The conformational states can be defined by the α4’-α5 distance (spheres and lines) between GluN1b and GluN2B ATDs controlled by ‘rolling’ motions of GluN2B ATDs (A), extent of opening of the GluN2B ATD (A), and the L1’-L2 distance (spheres and arrowed dots) between the GluN1b-Glu2B heterodimers controlled by ‘rolling’ motion of the GluN1b-GluN2B LBD heterodimers (B). (C-E) Tension of the GluN2B LBD-M3’ linkers measured by the distances between the two Gln662 Cαs (spheres; panel C) is the major determent for opening or closing of the channel gate (D-E). See also Figures S1–S3, Table S1, and Video S1.

The cryo-EM density for the ‘Active’ class was weaker especially at the LBD-TMD linker region due to small particle numbers compared to the other classes. To overcome this issue, we conducted single particle cryo-EM on the aforementioned mutant construct, ‘Active-SS,’ harboring the GluN1b-Glu719Cys and GluN2B-Leu795Cys mutations (Esmenjaud et al., 2018) at the dimer of the GluN1b-GluN2B heterodimers interface in the LBD-layer (Table S1, Figure 2B, red dashed circle in ‘Active-SS’). This mutant was previously shown to increase equilibrium open probability by 2.5-fold with significant rise in mean open time (Esmenjaud et al., 2018) likely by stabilizing the conformation similar to the ‘Active’ class above. Imaging of the ‘Active-SS’ sample required usage of gold-coated holey grids since they had a higher tendency to adhere to a carbon layer than the WT samples. While the particle orientation for the WT samples was dominated by side views, that for the ‘Active-SS’ sample equally contained both top and side views (Figure S2). Nevertheless, our single particle analysis shows an increased population of a 3D class similar in conformation to the ‘Active’ 3D class but with an increase in resolution from 4.4 to 3.6 Å and improved cryo-EM density (Figure S2). In this structure, the subunit rotation of 11° in the ATD is more pronounced than the ‘Active’ conformation (Figure 2A), which is coupled to the ‘rolling up’ of the GluN1b-GluN2B LBD heterodimers (Figure 2B) and pulling of the LBD-M3’ linkers from the GluN2B subunits to opposite directions (Figures 2C and 2D). Consequently, GluN2B-Ile655 from the two GluN2B subunits at the channel gate are pulled apart to open the gate (Figure 2D, Figure S3, Video S1). The extent of gate opening is similar to that observed in the ‘Pre-open’ state of the GluA2 AMPA receptor complexed to glutamate and cone snail toxins (Chen et al., 2014) but is not as open as agonist bound GluA2 AMPA receptors complexes with the auxiliary subunit Stargazin (Chen et al., 2017; Twomey et al., 2017) (Figure S3C–F). As in GluA2 AMPA receptors, the movement in GluN1b subunits (equivalent to the A/C subunits; Figure S3) is not as robust as for the GluN2B subunits (equivalent to the B/D subunits; Figure S3) and does not appear to be strongly coupled to channel gating.

Antagonist binding and competitive inhibition

One of the unique functional features of NMDAR pharmacology is that binding of either a GluN1-specific glycine antagonist or a GluN2-specific glutamate antagonist is sufficient for full inhibition of the channel activity even when agonists occupy the other subunits (Figures 3A–3B). However, how binding of the GluN1- or GluN2-targeting antagonist results in a similar pattern of channel inhibition remains largely unknown. Here, we attempted to resolve this long standing question using the GluN1-targeting L689,560 and the GluN2-targeting SDZ-220-040 ligands (Figure 3A), which competitively inhibit activities of the wild type channel as well as the CTD truncated channel used in the cryo-EM studies, as assessed by the two-electrode voltage clamp (TEVC) electrophysiology (Figure 3B, Table S2). These antagonist have higher potency (Table S2) than classic antagonists including DCKA and D-AP5 (Figure 3A) where reported K_i values on GluN1-GluN2A NMDARs are 80 nM and 0.71 μM, respectively (Chen et al., 2008; Jespersen et al., 2014). Thus, we sought to reveal the binding modes to explain the high potency of both L689,560 and SDZ-220-040. Furthermore, high occupancy of binding sites with the high-affinity ligands limit plausible conformational variability caused by partial occupancy thereby facilitating single particle cryo-EM. Toward our goal, we first resolved the precise chemistry of antagonist binding and the pattern of conformational changes within the LBDs by obtaining novel x-ray crystallographic structures of the isolated GluN1-GluN2A LBD heterodimers bound to glycine and SDZ-220-040 (2.3 Å) and to L689,560 and glutamate (2.1 Å) (Figures 3C–3H, Table S3). We used the GluN1-GluN2A LBD heterodimers instead of the GluN1-GluN2B LBDs as the tool to obtain high-resolution crystal structures due to technical difficulties in crystallization of the GluN1-GluN2B LBDs proteins. Nevertheless, the GluN2A and GluN2B LBDs have 82%/90% identity/similarity in their primary sequences and are different only by one residue within the agonist/antagonist-binding pockets, thus, structural insights gained in the GluN2A LBD can be applied to study the GluN2B LBD. The subunit arrangement of the isolated GluN1-GluN2A LBD assembly is nearly identical to that of the two GluN1b-GluN2B LBD heterodimers in the intact ion channel where the majority of subunit interactions are mediated at the upper lobe (D1) of the bi-lobe structures (Figures 3C and 3F) with ligands bound to the inter-D1-D2 clefts in each LBD. The binding of L659,560 ‘opens’ the GluN1b LBD clam shell by ~28° compared to the glycine-bound structure (PDB code: 4NF8) (Figure 3D) whereas the binding of SDZ-220-040 ‘opens’ the GluN2A LBD bi-lobe by ~23° compared to the glutamate-bound structure (PDB code: 4NF8) (Figure 3G). The electron density for L659,560, SDZ-220-040, and surrounding residues is of sufficient quality to pinpoint key receptor-ligand interactions (Figures 3E and 3H). Binding of L659,560 to GluN1 involves the following interactions (Figure 3E): 1) polar interactions (2-carboxyl group/GluN1b-Arg544 and carbonyl group/GluN1b-Ser709); 2) hydrophobic interactions (aminophenyl group/GluN1b-Trp752 and–Val705); 3) halogen-pi interactions (7’-chloro group/GluN1b-Phe429 and 5-chloro group/GluN1b-Trp752); and 4) stacking interaction (quinoline backbone/GluN1b-Phe505). Binding of SDZ-220-040 to GluN2A involves the following interactions (Figure 3H): 1) polar interactions (phosphono group/GluN2A-Ser689, -Thr690, and -Tyr730, and amino propanoic group/GluN2A-Arg518, -Thr513, and –Ser511); 2) hydrophobic interactions (bi-phenyl rings/GluN2A-His485, -Tyr730, and -Val713; and 3) halogen-carboxylate interaction (2’-chloro group/GluN2A-Glu413. The information gained from the high-resolution crystal structures above served as excellent models to interpret the cryo-EM density of the antagonists in the context of cryo-EM structures for full length receptors.

Figure 3. Antagonist binding and structures of GluN1-GluN2A LBDs.

(A-B) A GluN1-antagonist L689,560 and a GluN2-antagonist SDZ-220-040 can inhibit NMDARs in the presence of glutamate and glycine, respectively. TEVC recording on the WT and the EM constructs show the similar pattern of inhibition (error bars ±SD) (B). (C-E) Crystal structure of the GluN1-GluN2A LBD complexed to L689,560 and glutamate showing domain opening (D1/D2) of the GluN1 LBD bi-lobe by 28° compared to the glycine-bound GluN1 LBD (D). (F-H) Crystal structure of the GluN1-GluN2A LBD complexed to glycine and SDZ-220-040 showing domain opening (D1/D2) of the GluN2A LBD bi-lobe by 23° compared to the glutamate-bound GluN2A LBD (G). In both structures, electron density of ligands (Fo-Fc omit map contoured at 3σ in green mesh) is sufficiently resolved to pinpoint polar interactions (dotted lines) and hydrophobic interactions (E and H). See also Table S2–S3.

The GluN2 antagonist SDZ-220-040 ‘relaxes’ the LBD-TMD linker in GluN2B

In order to understand how conformational changes observed in the LBDs are represented in the context of the heterotetrameric channel, we obtained cryo-EM structures of intact GluN1-GluN2B NMDARs in the following three conditions: 1) SDZ-220-040 and glycine (Gly/SDZ); 2) L689,560 and glutamate (L689/Glu); and 3) L689,560 and SDZ-220-040 (L689/SDZ).

The Gly/SDZ structures have a major 3D class with C2 symmetry accounting for ~80% of the population and another class with no symmetry but similar to the C2 class. Hence, the comparative analyses are done on the major 3D class (Class 1) (Figure S4, Table S1) against the ‘Active-SS’ structure bound to glycine and glutamate. The Gly/SDZ structure is obtained at 3.6 Å overall resolution as assessed by FSC and shows clear density for SDZ-220-040 (Figures 4A and 4C; cyan surface and mesh, Figure S4). The binding of SDZ-220-040 resulted in opening of the bi-lobe structure of the GluN2B LBD by ~24° compared to the glutamate-bound GluN2B LBD in the ‘Active-SS’ structure (Figure 4B), which is similar to that observed in the crystal structure of the GluN1-GluN2A LBDs (~23°) (Figure 3G). The GluN1b LBD bi-lobes in Gly/SDZ and ‘Active-SS’ are equally closed consistent with binding of glycine in both complexes. The overall conformation of the Gly/SDZ complex is similar to that of the ‘Active-SS’ structure in that the GluN1b-GluN2B ATD subunit orientation is nearly identical (represented by the similar α4’-α5 distance) (Figure 4D), which as a consequence, stabilizes the ‘rolled up’ LBD dimers (as measured by the L1-L2 distance) (Figure 4E). Thus, the only difference between the Gly/SDZ and the ‘Active-SS’ structures is in the extent of bi-lobe opening in the GluN2B LBD. The ~24° opening of the GluN2B LBD in the Gly/SDZ results in ‘loosening’ of the GluN2B LBD-M3’ linker represented by the reduced distances between the GluN2B Gln662 Cαs compared to the ‘Active-SS’ conformation (49.6 Å in Gly/SDZ and 61.0 Å in ‘Active-SS’ (Figure 4F). Therefore, the mechanism of inhibition by the GluN2-specific antagonist SDZ-220-040, is to relieve the ‘tension’ of the GluN2B LBD-M3’ linker, which uncouples the TMD channel from the extracellular domain. Consequently, the hydrophobic gate residues remain tightly associated with each other to ‘lock’ the gate closed (Video S2).

Figure 4. Cryo-EM structure of GluN1b-GluN2B NMDAR complexed to glycine and SDZ-220-040 (Gly/SDZ).

(A) Cryo-EM density of the Gly/SDZ (Class 1) in the same color code as in Fig. 1. (B) Comparison of the GluN2B LBD bi-lobe in the Gly/SDZ (dark green) to that in the ‘Active-SS’ bound to glutamate (gray) showing the antagonist induced domain opening by 24°. (C) Cryo-EM density for SDZ-022-040 (cyan mesh) at the bi-lobe cleft. (D-E) In this structure, the subunit and domain arrangements are similar to those of the ‘Active-SS’ (gray) as represented by the ‘short’ α4’-α5 distance (spheres and a line) in the ATDs (D) and the ‘short’ L1’-L2 distance (spheres and arrowed dots) in the LBDs (E). (F) However, the GluN2B LBD-M3’ linkers are more relaxed compared to those in the ‘Active-SS’ (gray) due to opening of the GluN2B LBD bi-lobes as in panel B. See also Figure S4, Table S1, and Video S2.

Binding of GluN1 antagonist L689,560 favors ‘Non-active1’-like conformation

To understand the mechanism of inhibition mediated by GluN1-antagonists, we explored the structure of the GluN1b-GluN2B NMDAR in complex with L689,560 and glutamate (L689/Glu). The single particle analysis resulted in two major 3D classes, both with C2 symmetry, which are similar to each other (Table S1, Figure S5). The resolution of the L689/Glu structure at 4 Å is not as high as for Gly/SDZ, but is sufficient to confirm the presence of cryo-EM density representing L689,560 (Figures 5A and 5C). In the GluN2B LBD, glutamate was not fully resolved, however, the ‘closed’ clam shell is consistent with the binding of glutamate. By contrast, the GluN1b LBD bi-lobes in all 3D classes are ‘open’ compared to the glycine-bound GluN1b LBDs in the ‘Active-SS’ and other glutamate-bound GluN2B structures reported in the current study by ~12° to 15° (Figure 5B). Notably, the extent of opening is significantly less compared to that observed in the crystal structure of the L689/Glu-bound GluN1-GluN2A LBD at ~28° (Figure 3D). This difference in domain opening likely stems from steric hindrance associated with the packing pattern of the GluN1-GluN2B heterotetramer where the ‘mouth’ of the GluN1 LBD clam shell is in the vicinity of the GluN2B LBD of the other heterodimer. That is, opening of the GluN1b LBDs as much as ~28° would result in steric collision with the neighboring GluN2B LBD, especially around Helices E and G of the GluN1b subunit and Helices K’ and I’ of the GluN2B subunit (Figure 5E). While the extent of bi-lobe opening is different between our cryo-EM structure of the intact GluN1b-GluN2B NMDARs and crystal structure of the isolated LBDs, the binding modes of L689,560 are similar to each other (Figures S6A–B). The greater extent of opening in our crystal structure results from movement of regions that are distant from the L689,560 binding site (Figure S6C), which in the intact heterotetrameric NMDARs would be restricted. This case is different from the previous study on a Drosophila glutamate receptor where two distinct modes of D-AP5 binding resulted in different extent of domain opening in its LBD (Li et al., 2016). Most importantly, unlike the Gly/SDZ complex, the L689/Glu structure has an overall conformation and subunit arrangement distinct from the agonist-bound ‘Active-SS’ state (Figures 5D–F) but similar to the agonist-bound ‘Non-active1’ structure (Figure 5G–I). Compared to ‘Active-SS,’ the GluN2B ATDs and the GluN1b-GluN2B LBD heterodimers in L689/Glu are ‘rolling up’ (Figure 5D) are ‘rolling down’ (Figure 5E), respectively. The L1’-L2 distances in the D1 lobes of the LBDs and the α4’-α5 distance in the R2 lobes of the ATDs between L689/Glu and ‘Active-SS’ are distinct as result (Figures 5D–5E). In contrast, there is little or no difference in the ATD and LBD subunit arrangement between L689/Glu and ‘Non-activel’ as indicated by the similar α4’-α5 and L1’-L2 distances between the L689/Glu and ‘Non-active1’ (Figure 5G–5H). Also similar to ‘Non-active1,’ the GluN2B ATD clam shells in the L689/Glu structure have a ‘closed’ conformation (Figures 5D and 5G; ~9-15° more closed compared to the ‘Active-SS’ structure, and the GluN1b-GluN2B LBD heterodimers are ‘rolling down’ compared to ‘Active-SS’ (Figure 5E); therefore, sufficient tension in the GluN2B LBD-M3’ linkers cannot be generated to open the channel gate (Figure 5F and 5I). Despite changes in the conformation of the GluN1b LBD, the arrangement of the GluN1b LBD-M3 linkers are similar between the ‘Active-SS’, the ‘Non-active1’ and the L689/Glu structures (Figures 5F and 5I). Overall, our structural analysis shows that GluN1-targeting antagonists such as L689,560 inhibit channel activity by opening the GluN1 LBD clam shell and reorienting the GluN1 and GluN2B subunits to stabilize a ‘Non-active1’-like conformation, which results in a closed channel gate (Video S3).

Figure 5. Cryo-EM structure of GluN1b-GluN2B NMDAR complexed to L689,560 and glutamate (L689/Glu).

(A) Cryo-EM density of the L689/Glu (Class 1) in the same color code as in Fig. 1. (B) Comparison of the GluN1b LBD bi-lobe in the L689/Glu (magenta) compared to that in the ‘Active-SS’ bound to glutamate (gray) showing the antagonist induced domain opening by 12-15°. (C) Cryo-EM density for L689,560 (green mesh) at the bi-lobe cleft. (D-F) The subunit and domain arrangements of L689/Glu are distinct from ‘Active-SS’ as shown by rolling motions (arrows), longer α4’-α5 (D) and L1’-L2 (E) distances, and more closed GluN2B ATD bi-lobes (D). (F) Consequently, the GluN2B LBD-M3’ linkers are relaxed. (G-H) The subunit and domain arrangement of L689/Glu are similar to those in the ‘Non-active1’ (light gray) as represented by the similar α4’-α5 (G) and L1’-L2 (H) distances. (I) The GluN1b LBD-M3 linkers are similar between L689/Glu and ‘Active-SS.’ See also Figure S5–S6, Table S1 and Video S3.

Binding of GluN1- and GluN2-antagonists fortifies the hydrophobic gate

Prior cryo-EM studies on Xenopus laevis NMDARs reported structures obtained in the presence of the GluN1- and GluN2-antagonists, DCKA and D-AP5, which display robust dissociation of inter-subunit interactions in the ATD and LBD layers of the receptor (Zhu et al., 2016). In structures of the L689/Glu or the Gly/SDZ complexes reported here, we do not observe such changes; thus, we wondered if the binding of antagonists to both the GluN1 and GluN2 subunits could induce the large conformational change seen in prior studies. We solved the structure of the GluN1b-GluN2B NMDAR in complex with L689,560 and SDZ-220-040 (L689/SDZ) at ~4 Å resolution (Table S1, Figure S7), which clearly showed density for both L689,560 and SDZ-220-040 (Figures 6A–6C and S7). Importantly, the GluN1 and GluN2B subunits remained associated with each other with no disruption of ATD and LBD dimer of dimer assemblies. In this structure, the GluN1 LBD clam shells are open by ~12 to 15° (Figure 6B) similar to the L689/Glu structure but different from the crystal structure of the isolated GluN1-GluN2A LBDs (~28°) (Figure 3D) as in the case of L689/Glu again indicating that domain opening of the GluN1b LBD clam shell in the intact GluN1-GluN2B heterotetramer is restricted by steric hindrance. Domain opening of the GluN2B LBD clam shell is nearly identical to that observed in the both the Gly/SDZ structure and the GluN1-GluN2A LBD crystal structure (Figures 3G and 6B). Further structural inspection revealed that subunit arrangement in the L689/SDZ structure is similar to that of the ‘Active-SS’ structure, represented by similarities in the α4’-α5 distance in the R2 lobes of the ATDs (Figure 6D) and in the L1’-L2 distances in the D1 lobe of the LBDs (Figure 6E) as in the case of Gly/SDZ (Figure 4). However, the ‘open’ GluN2B LBD relaxes the LBD-M3’ linker to disengage conformational signals from the LBD to the channel similar to the Gly/SDZ complex (Figure 6F). A major difference between the L689/SDZ and the Gly/SDZ structures is that the L689/SDZ-bound structure has the LBD-M4’ linkers positioned closer to the ‘middle’ of the heterotetramer above the gate, which results in further stabilization of the hydrophobic gate by formation of extra hydrophobic interaction between GluN2B-Val808 and the hydrophobic gate residues (GluN1b-Val677 and –Leu678 and GluN2B-Ile655) (Figure 6G). Thus, binding of antagonists to both GluN1 and GluN2B stabilizes a ‘super’ closed gate (Video S4).

Figure 6. Cryo-EM structure of GluN1b-GluN2B NMDAR complexed to L689,560 and SDZ-220-040 (L689/SDZ).

(A) Cryo-EM density of the L689/SDZ (Class 1) in the same color code as in Fig. 1. (B) Comparison of the GluN1b LBD bi-lobe complexed to L689,560 (magenta) and the GluN2B LBD bi-lobe complexed to SDZ-220-040 (dark green) against the ‘Active-SS’ complexed to glycine and glutamate (gray) showing antagonist-induced domain opening. (C) Cryo-EM density for L689,560 (green mesh) and SDZ-220-040 (cyan mesh) at the bi-lobe clefts. (D-E) In this structure, the subunit arrangements and domain arrangement is similar to that of the ‘Active-SS’ (gray) as represented by the ‘short’ α4’-α5 distance in the ATDs (D) and the ‘short’ L1’-L2 distance in the LBDs (E). (F) As in the Gly/SDZ, the GluN2B LBD-M3’ linkers are relaxed compared to the ‘Active-SS’ (gray) due to opening of the GluN2B LBD bi-lobe. Little or no change in the GluN1b LBD-M3 linkers but a change in the orientation of the GluN2B LBD-M4’ linkers is observed. (G) Consequently, the hydrophobic interaction at the channel gate is strengthened by GluN2B Val808 (yellow ovals). See also Figure S7, Table S1, and Video S4.

Discussion

Inter-subunit/domain rearrangement is the lynchpin of functional transitions in NMDARs; therefore, precise monitoring of conformational changes in multiple domains of the GluN1-GluN2 hetero-tetramer is essential for mechanistic understanding and reagent development. Here we unraveled detailed functional mechanisms by solving high-resolution cryo-EM structures of GluN1b-GluN2B NMDARs in multiple distinct ligand bound forms. Our study reveals structural features that were not clearly visible in prior studies, including the LBD-TMD linker region, the channel gate, and ligand densities, and thus provides precise and detailed insights in unprecedented detail.

First we showed that following agonist binding NMDARs populate three major conformational states we name ‘Non-active1,’ ‘Non-active2’ and ‘Active,’ which differ from each other in their patterns of inter-GluN1b-GluN2B and inter-ATD-LBD orientations (Figure 2). The relatively scarce population of the ‘Active’ state is consistent with the previous single channel analyses, which estimated the open probability to be around 20% (Banke et al., 2005; Banke and Traynelis, 2003). Disulfide-based conformational trapping of the ‘Active’ conformation (‘Active-SS’) increased the ‘Active’ state population as assessed by 3D classification and resulted in a higher quality structure which facilitated our assessment of channel gating in comparison with the ‘Non-active1’ and the ‘Non-active2 conformations.

In the ‘Active-SS’ conformation, the LBD-M3’ linkers of the two GluN2B subunits are ‘stretched’, creating tension that perturbs the closed channel gate formed by a cluster of hydrophobic residues at the tip of the GluN1b-M3 and GluN2B-M3’ helices (Figures 2C–2D). Furthermore, the GluN1b and GluN2B ATDs orient themselves to close the gap between their R2 lobes (Figure 2A). In both the ‘Non-active1’ and ‘Non-active2’ states the GluN2B LBD-M3’ linkers are too ‘relaxed’ to perturb the channel gate; this is reflected by the fact that the two GluN1b-GluN2B LBD heterodimers are ‘rolled down’ (Figure 2B) and the R2 lobes of the GluN1b and GluN2B ATDs are far apart (Figure 2A). The above patterns of concerted ATD-LBD movements are not observed in the activation process of AMPA receptors (Chen et al., 2017; Twomey et al., 2017), indicating that the gating mechanism is unique in NMDARs. Previous single channel recording and kinetic analyses indicated that binding of glycine and glutamate puts NMDARs in a multiple closed and open channel states as well as desensitized states (Amico-Ruvio and Popescu, 2010; Iacobucci and Popescu, 2018; Maneshi et al., 2017). Whether the ‘Non-active1’ or the ‘Non-active 2’ conformations represent desensitized states or non-desensitized states with a closed channel remains an open question at this point.

The channel gate of the ‘Active-SS’ state is open to a similar extent to the ‘Preopen’ state of the GluA2 AMPA receptor complexed to cone snail toxin (Chen et al., 2014) but less open than observed for GluA2 complex with Stargazin where the end of its M3 helices in the B/D subunits (equivalent to GluN2B subunits) are kinked (Twomey et al., 2017). This indicates that an additional conformational change in the channel gate may occur also for NMDARs. Capturing such a conformation by cryo-EM may require a stabilizer of the ‘open’ channel state equivalent to Stargazin in AMPA receptors. Thus far, several auxiliary subunit proteins for NMDARs have been reported (Ng et al., 2009; Scanlon et al., 2017), however, none of them has been shown to alter ion channel gating. Nevertheless, the activation scheme of NMDARs has similar features to that for GluA2 AMPA receptors in that it is dominated by the movement of the LBD-M3’ linkers of the GluN2B subunits, which are equivalent to the B/D subunits in GluA2 AMPA receptors. Likewise, movement of the LBD-M3 linkers of the GluN1b subunits, which are equivalent to the A/C subunits in GluA2 AMPA receptors is much smaller for both receptor classes. The above observations are also consistent with a previous study showing that perturbation of the GluN2A LBD-M3’ linker by mutation or insertion of Gly residues dramatically reduces the open probability and duration of channel opening (Fedele et al., 2018; Kazi et al., 2014; Ladislav et al., 2018). Our ‘Active-SS’ structure shows that this insertion site is the middle of Helix E’ of GluN2B at the tip of the GluN2B LBD-M3’ linker; thus, it likely deforms the structure of the Helix E’ region and ‘relaxes’ the tension of the GluN2B LBD-M3’ linker in a dramatic fashion.

Next, by obtaining cryo-EM structures of two agonist/antagonist combinations, Gly/SDZ and L689/Glu, we demonstrated that GluN1 and GluN2 antagonists inhibit receptor activation by ‘relaxing’ the GluN2B LBD-M3’ linkers in a similar way, however, via two distinct mechanisms. In both structures, the binding of antagonists stabilizes open-cleft conformations of LBDs consistent with previous structural studies (Furukawa and Gouaux, 2003; Jespersen et al., 2014; Lind et al., 2017; Romero-Hernandez and Furukawa, 2017; Wang et al., 2020) and single molecular FRET studies (Cooper et al., 2015). In the Gly/SDZ complex, the arrangement of GluN1b-GluN2B LBD heterodimers is similar to that of the ‘Active-SS’ state where the GluN1b-GluN2B LBD heterodimers are ‘rolled up’ and the R2 lobes of the GluN1b-GluN2B ATDs are close to each other. However, in the Gly/SDZ complex, unlike the agonist bound ‘Active-SS’ state, binding of SDZ-220-040 opens the GluN2B LBD bi-lobe, which in turn relaxes the GluN2B LBD-M3’ linkers to favor a closed channel gate (Figure 7). In the reverse situation where a GluN1-antagonst, L689,560, and glutamate are bound (L689/Glu), the GluN2B LBD-M3’ linkers become ‘relaxed’ as a result of the receptor entering into the ‘Non-active1’-like conformation where the heterodimers are ‘rolled down’ (Figures 5 and 7). Furthermore, in the L689/Glu complex, the GluN2B ATD bi-lobe closes even more extensively than observed in the ‘Non-active1’ structure, to a similar extent found in the ifenprodil- or zinc-bound GluN2B ATD structures (Karakas et al., 2009, 2011; Regan et al., 2019) indicating tight coupling between ATD and LBD in NMDARs. That is, conformations of ATDs can be controlled from the LBDs and vice versa. Our finding here is likely applicable to the action of other antagonists including DCKA and D-AP5 since they are known to elicit similar opening of the LBD bi-lobe structures (Furukawa and Gouaux, 2003; Jespersen et al., 2014).

Figure 7. Schematic presentation of conformational transitions.

(A-B) The agonist-bound GluN1b-GluN2B NMDAR reside in ‘Non-active’ and ‘Active’ conformations where they differ in the subunit arrangement within the GluN1b-GluN2B ATD heterodimer and the orientation between the GluN1b-GluN2B LBD heterodimers as a result of their rigid-body movement (arrows indicating the movement from ‘Non-active1’ to ‘Active-SS’). These conformations can be defined by the distances between GluN1 α5 (red rectangle) and GluN2B α4’ (yellow rectangle) at the ATDs and/or the distances between GluN1 L2 (orange oval) and GluN2B L1’ (cyan oval) at the LBDs where they are ‘long’ and ‘short’ in ‘Non-active’ and ‘Active,’ respectively. For clarity, only ‘Non-active1’ and ‘Active-SS’ are displayed here. (C) Binding of L689,560 and glutamate to GluN1 and GluN2B, respectively, leads to ‘Non-active1’-like conformation where the inter-subunit distances above are ‘long.’ (D) In contrast, binding of SDZ-220-040 in the presence of glutamate or L689,560 opens the GluN2B LBD cleft and stabilizes ‘Active-SS’-like conformation where the inter-subunit distances above are ‘short.’ In the L689/SDZ, the gate is further closed by extra hydrophobic interactions. Black arrows indicate the rigid-body movement of the ATD and LBD heterodimers from ‘Non-active1’ to ‘Gly/SDZ’ or ‘L689/SDZ.’ Red arrows indicate LBD bi-lobe opening by antagonist binding.

Lastly, we showed that concurrent binding of GluN1- and GluN2-targeting antagonists (L689/SDZ) ‘relaxes’ the GluN2B LBD-M3’ linkers by opening the GluN2B LBD bi-lobes in a similar manner to the ‘Gly/SDZ’ complex. Here, the ATDs and LBDs have the ‘Active-SS’-like conformation as found in the ‘Gly/SDZ’ complex. Another important feature of our study is that the ‘L689/SDZ’ structure clearly shows tightly associated subunits and domains, which is similar to the observation in GluA2 AMPA receptor where the subunits in the homotetrameric assembly remain intact in its antagonist-bound and apo states (Durr et al., 2014; Meyerson et al., 2016; Schauder et al., 2013; Sobolevsky et al., 2009). These observations in iGluRs are in parallel with the recent study on the Cys-loop family where antagonist-bound and apo-state structures resemble each other (Yu et al., 2019). The previous cryo-EM structure of Xenopus laevis GluN1-GluN2B NMDARs obtained in the presence of the GluN1-antagonist, DCKA and the GluN2-antagonist, D-AP5 at 9-15 Å resolution showed robust subunit dissociation for both the ATD and LBD dimer of dimers assemblies (Zhu et al., 2016), which currently stands as an outlier. While the conformation of Xenopus NMDAR proteins may be more susceptible to perturbation by the water-air interface in cryo-EM conditions, there are a number of additional potential factors that could account for the differences including usage of different detergent (e.g. lauryl maltose neopentyl glycol + dodecyl maltoside vs. lauryl maltose neopentyl glycol + digitonin), and different pHs (pH 6.5 vs. pH 7.5). In a more recent study at low pH and in the presence of a high zinc concentration, the GluN1-GluN2A NMDAR subunits also become splayed in the extracellular region (Jalali-Yazdi et al., 2018). The GluN1-GluN2B NMDAR ATDs bind both protons and zinc which allosterically inhibit the channel activity, thus, it is possible that the low pH condition (pH 6.5) applied to the antagonist-bound Xenopus laevis NMDARs may have facilitated subunit dissociation. In our current study, we alleviated the conformational of ATD contributing to allosteric inhibition by removing ambient zinc by EDTA and by using physiological pH at 7.5. Our study demonstrated that the tight coupling between ATDs and LBDs play crucial roles in both activation and competitive inhibition by ligands that bind LBDs. It is interesting to note that auto-immune antibodies against NMDARs including the ones produced in Lupus (Chan et al., 2020; DeGiorgio et al., 2001; Lapteva et al., 2006) and anti-NMDAR encephalitis (Dalmau et al., 2008) bind ATDs to alter the ion channel functions further indicating the tight coupling between ATDs, LBDs, and TMDs. On the other hand, reagents that are designed to control an equilibrium of the conformational transitions observed in this study would be useful in fine-tuning NMDAR functions for therapeutic purposes.

STAR Methods

Lead Contact and Materials Availability

All unique reagents generated in this study are available from the Lead Contact, Hiro Furukawa (furukawa@cshl.edu) with a completed Materials Transfer Agreement.

Experimental Model and Subject Details

Sf9 cells were cultured in HyClone CCM3 cell culture medium (GE Healthcare) at 27°. Xenopus laevis oocytes used in electrophysiology experiments were cultured in 50% HyClone Leibovitz L-15 Medium at 18°C.

Method details

Expression and purification of intact NMDARs

The expression and purification methods of intact NMDAR receptors were adopted from those established previously (Tajima et al., 2016) except that the newly constructed vectors, pFp10_hsp and pUCp10_hsp, harboring sequences for the hr1 and hr5 enhancers, the Drosophila melanogaster HSP70 promoter, and the p10 3’UTR, were used for baculovirus production. Sf9 insect cells at 4 x 10⁶ cells/ml were infected with the recombinant baculovirus harboring both GluN1b and GluN2B (C-terminal truncation) and harvested after 48 hours. The membrane fractions (100 mg/ml) of the infected cells were solubilized in the buffer containing 20mM HEPES-Na pH 7.5, 150 mM NaCl, 1 mM Glycine, 1 mM Na-Glutamate, and 0.5% LMNG for 2 hours at 4°C and centrifuged at 125,000g for 40 minutes. The supernatant was purified using Strep-tactin resin followed by Superose 6 Increase column (GE Healthcare) size exclusion chromatography (SEC) with pre-equilibrated with 20 mM HEPES-Na pH 7.5, 150 mM NaCl, 0.002% LMNG 1 mM glycine, 1 mM Na-glutamate. For antagonist-bound structures, all of the purification steps above were conducted in the absence of glycine and glutamate.

Single particle cryo-EM on GluN1b-GluN2B NMDARs

The protein samples were vitrified on glow discharged C-flat 1.2/1.3 Cu 400 mesh holey carbon grids (Protochips) or UltrAufoil holey gold film grids (Quantifoil) (for ‘Active-SS’) using FEI Vitrobot Mark IV at 15°C and at 85% humidity with a blot time of 4 sec under level 7 blot force. The filter papers were washed with 1 mM EDTA and water and dried before usage. All the micrographs were acquired by Titan Krios (FEI) at Cold Spring Harbor Laboratory operating at 300 kV and the GATAN K2 Summit direct electron detector couple with the GIF quantum energy filter (Gatan Inc.) at 105k magnification (1.37 Å/pixel), with the defocus range of −1.5 to −3.0 μm, and over 50 frames and 15 seconds exposure totaling the dose of 68 e/Ų The movies alignment, CTF estimation, and particle picking were done using the program WARP (Tegunov and Cramer, 2019). The picked particle images were subjected to two to three rounds of 2D classifications to remove ‘junk’ particles. The selected particles are subjected to ab-initio 3D map generation, which are then refined and 3D classified into seven to eight classes without imposing symmetry. 3D classes were inspected using UCSF Chimera (Oshima et al., 2016) and particles from similar 3D classes were merged and further refined and classified. 3D classes that do not contain density for TMDs were excluded from further refinement. If the 3D classes looked symmetrical, C2 symmetry was imposed. At this point, CTF refinement was applied to selected classes followed by another round of 3D refinement. 2D classification, ab-initio 3D map generation, 3D refinement, 3D classification, per particle CTF refinement and B-factor sharpening were done using the program cisTEM (Grant et al., 2018).

Model building was done initially by docking our previous cryo-EM structure (PDB code: 5FXH) into the B-factor sharpened cryo-EM density maps using UCSF Chimera (Oshima et al., 2016) then remodeled using Rosetta 3.7 (Wang et al., 2016). The resulting models were manually inspected and modified to fit into the density using COOT (Emsley et al., 2010). For the areas where the map was poorly resolved, poly-alanine models were introduced to replace the actual primary sequences. The final models were refined against the cryo-EM maps using Phenix real space refinement (Adams et al., 2010) with secondary structure and Ramachandran restraints. The FSCs were calculated by phenix.mtriage. Summary of data collection and refinement statistics are shown in Table S1.

Electrophysiology

cRNA encoding GluN1-4b and GluN2B or GluN1-4bΔCTD and GluN2BΔCTD NMDARs were injected into defolliculated Xenopus laevis oocytes (0.05–0.3 ng total cRNA per oocyte). The oocytes were incubated in recovery medium (50% L-15 medium (Hyclone) buffered by 15mM Na-HEPES at a final pH of 7.4), supplemented with 100 μg mL⁻¹ streptomycin, and 100 U mL⁻¹ penicillin at 18°C. Two electrode voltage clamp (TEVC; Axoclamp-2B) recording was performed between 48 and 96 hours after injection using an extracellular solution containing 5 mM HEPES, 100 mM NaCl, 0.3 mM BaCl₂, 10mM Tricine at final pH 7.4 (adjusted with KOH). The current was measured using agarose-tipped microelectrode (0.4–1.0 MΩ) at the holding potential of − 60 mV. Maximal response currents were evoked by 100 μM of glycine and L-glutamate. Data was acquired by the program Pulse (HEKA) and analyzed by Origin 8 (OriginLab Corp).

X-ray crystallography on GluN1-GluN2A LBDs

The isolated LBD proteins for rat GluN1 and rat GluN2A were obtained using our previously published method (Jespersen et al., 2014). The purified GluN1 and GluN2A LBD proteins were separately concentrated to 6 mg/mL, mixed in at a 1:1 weight ratio, and dialyzed against 10 mM HEPES (pH 7.0), 100 mM NaCl, 1 mM glycine, and 1 mM glutamate. The GluN1-GluN2A LBD proteins were crystallized using hanging-drop vapor diffusion at 18°C for two to three days where reservoir solutions contained 100 mM HEPES (pH 7.0), 60-90 mM NaCl, and 15-20% polyethylene glycol monomethylether 2000 (PEG2000 MME). The crystals were then sequentially soaked against the crystallization buffer containing 1 −100 μM SDZ-220-040 or L689,560. The crystals were flash frozen in liquid nitrogen in the crystallization buffer supplemented with the antagonist of interest and 18% glycerol. X-ray diffraction data was collected at the ID-23-B beamline of the Advanced Photon System at Argonne National Laboratory and processed using HKL2000 (Otwinowski & Minor, 1997). The structures were determined by molecular replacement using the D1 and D2 domains of GluN1 and GluN2A LBDs (PDB code: 4NF8) as search probes. Molecular replacement, structural refinement and model building were performed using PHASER (McCoy et al, 2007), PHENIX (Adams et al, 2010), and Coot (Emsley & Cowtan, 2004). The ligands were built in ChemDraw and CIF files were generated in the Grade server (Global Phasing Ltd). Summary of data collection and refinement statistics are shown in Table S3.

DATA AND SOFTWARE AVAILABILITY

Data Resources

The cryo-EM maps for the Gly/Glu-bound rat GluN1b-GluN2B NMDARs in the non-active1, non-active2, and active conformations have been deposited to the Electron Microscopy Data Bank under the accession codes: 21674, 21673, and 21675, respectively, and the structural coordinates have been deposited to the Protein Data Bank under the accession codes, 6WHS, 6WHR, and 6WHT, respectively.

The cryo-EM maps for the double antagonists, the SDZ/L689-bound rat GluN1b-GluN2B NMDARs in the class 1 and class 2 conformations have been deposited to the Electron Microscopy Data Bank under the accession codes, 21676 and 21677, respectively, and the structural coordinates have been deposited to the Protein Data Bank under the accession codes, 6WHU and 6WHV, respectively.

The two cryo-EM maps for the SDZ/Gly-bound rat GluN1b-GluN2B NMDARs in the class 1 and class 2 conformations have been deposited to the Electron Microscopy Data Bank under the accession codes, 21678 and 21679, respectively, and the structural coordinates have been deposited to the Protein Data Bank under the accession codes, 6WHW and 6WHX, respectively.

The two cryo-EM maps for the Glu/L689-bound rat GluN1b-GluN2B NMDARs in the class 1 and class 2 conformations have been deposited to the Electron Microscopy Data Bank under the accession codes, 21680 and 21681, respectively, and the structural coordinates have been deposited to the Protein Data Bank under the accession codes, 6WHY and 6WI0, respectively.

The cryo-EM map for rat GluN1b-GluN2B NMDARs in the Active-SS conformation has been deposited to the Electron Microscopy Data Bank under the accession code: 21682, and the structural coordinate has been deposited to Protein Data Bank under the accession code, 6WI1.

The electron density map and structural coordinate for the crystal structure of the GluN1/GluN2A ligand-binding domain in complex with L689,560 and glutamate have been deposited to the Protein Data Bank under the accession code, 6USU.

The electron density map and structural coordinate for the crystal structure of the GluN1/GluN2A ligand-binding domain in complex with glycine and SDZ 220-040 have been deposited to the Protein Data Bank under accession code, 6USV.

Supplementary Material

Video S4

Conformational change between ‘Active-SS’ and ‘L689/SDZ,’ related to Figure 6

Video S1

Conformational change between ‘Non-active1’ and ‘Active-SS,’ related to Figure 2

Figure S7

Single particle analysis on the ‘L689/SDZ’ GluN1b-GluN2B NMDARs, Related to Figure 6 (A) A representative image, 2D classes, and the 3D classification workflow. (B) FSC curve of two half maps (top) and map vs model (bottom) for the ‘Class 1’ (cyan), and the ‘Class 2’ (green) 3D classes. (C) The angular distribution plots for each classes. (D) Local resolution for Class 1 calculated by the program ResMap. (E) Zoomed-in views of the LBD-TMD linkers of GluN1b (top) and GluN2B (bottom). (F) TMD segments for GluN1b (top) and GluN2B (bottom).

Figure S1

Single particle analysis on the agonist-bound GluN1b-GluN2B NMDARs, Related to Figure 1 and 2(A) A representative image, 2D classes, and the 3D classification workflow. (B) FSC curve of two half maps (top) and map vs model (bottom) for the ‘Non-active 1’ (cyan), ‘Non-active 2’ (magenta), and the ‘Active’ (green) 3D classes. (C) The angular distribution plots for each classes. (D) Local resolutions calculated by the program ResMap. (E) Zoomed-in views of the LBD-TMD linkers of GluN1b (top) and GluN2B (bottom). (F) TMD segments for GluN1b (top) and GluN2B (bottom).

Figure S3

Ion channel pore analysis on GluN1b-GluN2B NMDARs, Related to Figure 2 (A) Pore radius of the receptors in different conformations of the GluN1b-GluN2B NMDARs calculated using the program HOLE. (B) Representative ion channel conducting pathway. Black arrows point to important residues (in sticks) for gating and ion selectivity annotated in panel A. Note that the pore radius of ‘Active-SS’ could not be measured for Selectivity filter due to insufficient quality of cryo-EM density. (C-D) Structural comparison of the ‘Active-SS’ GluN1b-GluN2B NMDARs (GluN1b-magenta and GluN2B-dark green) with the ‘Pre-open’ GluA2 AMPA receptors (Ala622Thr) bound to a partial agonist, kainate, a positive allosteric modulator, (R,R)-2b, and con-ikot-ikot snail toxin (A/C subunits-yellow and B/D subunits-orange, PDB code: 4U5B). The positioning of gate residues in the M3 helices and the channel lining residues are similar to each other. (E-F) Comparison with the Open’ GluA2 AMPA receptors fused to stargazing and in complex with glutamate and cyclothiazide. Note that the M3 helices in the B/D subunits (orange) kink (arrows in panel D). The tensions of the M3’-LBD linkers in NMDARs and the M3-LBD linkers in AMPA receptors (arrows in panel A and C) created by LBDs are created in different directions due to the distinct orientations between the LBDs in TMDs in those receptors.

Figure S2

Single particle analysis on the ‘Active-SS’ GluN1b-GluN2B NMDARs, Related to Figure 2 (A) A representative image, 2D classes, and the 3D classification workflow. (B) FSC curve of two half maps (top) and map vs model (bottom) for the ‘Active- SS’ (cyan) 3D class. (C) The angular distribution plots for ‘Active-SS.’ (D) Local resolution calculated by the program ResMap. (E) Zoomed-in views of the LBD-TMD linkers of GluN1b (top) and GluN2B (bottom). (F) TMD segments for GluN1b (top) and GluN2B (bottom).

Figure S4

Single particle analysis on the ‘Gly-SDZ’ GluN1b-GluN2B NMDARs, Related to Figure 4 (A) A representative image, 2D classes, and the 3D classification workflow. (B) FSC curve of two half maps (top) and map vs model (bottom) for the ‘Class 1’ (cyan) and ‘Class 2’ (magenta) 3D classes. (C) The angular distribution plots for each classes. (D) Local resolution for Class 1 calculated by the program ResMap. (E) Zoomed- in views of the LBD-TMD linkers of GluN1b (top) and GluN2B (bottom). (F) TMD segments for GluN1b (top) and GluN2B (bottom).

Figure S6

Structural comparison of ‘L689/Glu’ and crystal structure, Related to Figure 5 (A-B) The L689,560 ligand binding sites of the cryo-EM structure (magenta) and the crystal structure (light gray) from different angles showing similar binding modes. (C) Overall GluN1b LBD structures showing a modest structural difference around Helix E and larger changes around the Helix F, D, and H regions (arrows) when D1 (upper lobe) residues are superimposed to one another.

Figure S5

Single particle analysis on the ‘L689-Glu’ GluN1b-GluN2B NMDARs, Related to Figure 5 (A) A representative image, 2D classes, and the 3D classification workflow. (B) FSC curve of two half maps (top) and map vs model (bottom) for the ‘Class 1’ (cyan), and ‘Class 2’ (magenta) 3D classes. (C) The angular distribution plots for each classes. (D) Local resolution for Class 1 calculated by the program ResMap. (E) Zoomed- in views of the LBD-TMD linkers of GluN1b (top) and GluN2B (bottom). (F) TMD segments for GluN1b (top) and GluN2B (bottom).

Video S2

Conformational change between ‘Active-SS’ and ‘Gly/SDZ,’ related to Figure 4

Video S3

Conformational change between ‘Active-SS’ and ‘L689/Glu,’ related to Figure 5

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, Peptides, and Recombinant Proteins
HyClone CCM3 cell culture medium	VWR International Inc. (GE Healthcare)	Cat#16777-272
HyClone Leibovitz L-15 Medium	VWR International Inc. (GE Healthcare)	Cat#82024-284
Digitonin	Calbiochem	Cat#300410
SDZ 220-040	TOCRIS Bioscience	Cat#1251
L689,560	TOCRIS Bioscience	Cat#0742
Experimental Models: Cell Lines
Xenopus laevis (oocyte extraction)	Nasco	N/A
Deposited Data
ratGluN1b/GluN2B NMDAR in non-Active 1 conformation, cryoEM map and molecular model	This study	EMDB: 21674 PDB: 6WHS
ratGluN1b/GluN2B NMDAR in non-Active 2 conformation, cryoEM map and molecular model	This study	EMDB: 21673 PDB: 6WHR
ratGluN1b/GluN2B NMDAR in Active conformation, cryoEM map and molecular model	This study	EMDB: 21675 PDB: 6WHT
ratGluN1b/GluN2B NMDAR, SDZ 220-040, L689,560 bound class 1, cryoEM map and molecular model	This study	EMDB: 21676 PDB: 6WHU
ratGluN1b/GluN2B NMDAR, SDZ 220-040, L689,560 bound class 2, cryoEM map and molecular model	This study	EMDB: 21677 PDB: 6WHV
ratGluN1b/GluN2B NMDAR, SDZ 220-040 bound class 1, cryoEM map and molecular model	This study	EMDB: 21678 PDB: 6WHW
ratGluN1b/GluN2B NMDAR, SDZ 220-040 bound class 2, cryoEM map and molecular model	This study	EMDB: 21679 PDB: 6WHX
ratGluN1b/GluN2B NMDAR, L689,560 bound class 1, cryoEM map and molecular model	This study	EMDB: 21680 PDB: 6WHY
ratGluN1b/GluN2B NMDAR, L689,560 bound class 2, cryoEM map and molecular model	This study	EMDB: 21681 PDB: 6WI0
ratGluN1b/GluN2B NMDAR in Active-SS conformation, cryoEM map and molecular model	This study	EMDB: 21682 PDB: 6WI1
Crystal structure of GluN1/GluN2A ligand-binding domain in complex with L689,560 and glutamate	This study	PDB: 6USU
Crystal structure of GluN1/GluN2A ligand-binding domain in complex with glycine and SDZ 220-040	This study	PDB: 6USV
Recombinant DNA
RatGluN1bEM	This study	N/A
OS-ratGluN2BEM	This study	N/A
ratGluN1bActive-SS	This study	N/A
OS-ratGluN2BActive-SS	This study	N/A
ratGluN1 LBD	This study	N/A
SUMO-ratGluN2A LBD	This study	N/A
GluN1-4b	This study	N/A
GluN2B	This study	N/A
GluN1-4bΔCTD	This study	N/A
GluN2BΔCTD	This study	N/A
Software and Algorithms
SerialEM	Mastronarde, 2005	https://bio3d.colorado.edu/SerialEM/
EPU	ThermoFisher	https://www.fei.com/software/epu-automated-single-particles-software-for-life-sciences/
WARP	Tegunov and Cramer, 2019	http://www.warpem.com/warp/
cisTEM	Grant et al., 2018	https://cistem.org/
UCSF Chimera	Oshima et al., 2016	https://www.cgl.ucsf.edu/chimera/
Rosetta	Wang et al., 2016	https://www.rosettacommons.org/software
COOT	Emsley et al., 2010	https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/
Phenix	Adams et al., 2010	https://www.phenix-online.org/
Pulse	HEKA	https://www.heka.com/about/aboutmain.html#smart-ephys
Origin	OriginLab Corp	https://www.orginlab.com/
HKL2000	Otwinowski & Minor, 1997	https://hkl-xray.com/
PHASER	McCoy et al, 2007	http://www.ccp4w.ac.uk/html/phaser.html
ChemDraw	PerkinElmer	https://www.perkinelmer.com/category/chemdraw
Grade server	Global Phasing Ltd	http://grade.globalphasing.org/
Pymol	Schrodinger	https://pymol.org/2/
Other
Strep-Tacin Sepharose resin	IBA Lifescience	Cat#2-1201-025
Superose 6 Increase column 10/300 GL	GE Healthcare	Cat#29091596
C-flat holey carbon grid	Protochips	Part#CF-1.2/1.3-4Cu-50
UltrAUfoil gold grid	Electron Microscopy Sciences	Cat#Q350AR1.3A

Highlights.

• Cryo-EM structures of GluN1b-GluN2B NMDARs in different liganded states were obtained at 4 Å or better

• Agonist-binding rearranges GluN1-GluN2B dimers to ‘stretch’ the GluN2B LBD-TMD linkers and perturb the hydrophobic channel gate

• A GluN2-antagonist mediates inhibition by ‘relaxing’ the GluN2 LBD-TMD linker

• A GluN1-antagonist mediates inhibition by reorienting GluN1 and GluN2 to result in ‘relaxing’ of the GluN2 LBD-TMD linker