Crystal structure of a heterotetrameric NMDA receptor ion channel

Production and structural study of heterotetrameric NMDA receptors

NMDA receptors are obligatory heterotetramers composed of two copies each of GluN1 and GluN2. Structural studies of heteromultimeric eukaryotic membrane proteins from a recombinant source have been hindered by difficulties in properly assembling multiple membrane proteins in recombinant expression host cells. After extensive exploration of expression methods, we succeeded in obtaining homogeneously assembled heterotetrameric NMDA receptors by expressing modified GluN1-4a and GluN2B subunits (GluN1a/GluN2B_cryst; fig. S1–4, see Supplementary Methods) in Sf9 insect cells using a recombinant baculovirus containing both of those subunits under the Hsp70 promoter from Drosophila melanogaster. Expression under conventional late promoters such as P10 and Polyhedrin promoters hampers proper heteromeric assembly of the NMDA receptor subunits (fig. S2). The GluN1a/GluN2B_cryst construct forms an ion channel that is opened upon glycine and L-glutamate application and allosterically regulated by ifenprodil and polyamines similarly to the wild-type receptor (fig. S5–6). GluN1a/GluN2B_cryst was crystallized in the presence of a GluN1 agonist, glycine, a GluN2 agonist, L-glutamate, and an ATD-binding allosteric inhibitor, ifenprodil. The structure was initially solved at 5.7 Å by molecular replacement using the extracellular domain structures as search probes (Supplementary Methods). Overall the extracellular domains of the receptor were well resolved and electron density for the TMD was of sufficient quality to conclude that TMD helices of GluN1a/GluN2B_cryst receptor are arranged similarly to those of GluA2 (fig. S7) (10). To improve the x-ray diffraction quality, we stabilized the heterotetramer by forming disulfide cross-links between subunits (fig. S1). Based on the 5.7 Å structure, we engineered a GluN2B Ser214Cys mutation to form a disulfide bond between the ATDs of two GluN2B subunits and pairs of mutations, the GluN1a Thr561Cys/Phe810Cys and GluN2B Asp557Cys/Ile815Cys, to tether the M1 helices to the M4 helices of the neighboring subunit at the TMD. The disulfide cross-linked mutant receptor (GluN1a/GluN2B_crystx) is trapped in an inhibited state, which could be unlocked by application of reducing agents (fig. S6). The cross-linking improved the diffraction limit to better than 4 Å, which resulted in electron density sufficient to build most of the GluN1a/GluN2B NMDA receptor including the entire extracellular domains, TMD, linkers between ATD and LBD and between LBD and TMD except some residues in the cytoplasmic loops (GluN1a 583–604, 617–622 and 834–847, and GluN2B 570–601, 616–629 and 841–852), in the loop connecting the LBD to the TMD (GluN2B 541–548 and 803–806) and in the extracellular loops (GluN1a 95–104 and 442–444, and GluN2B 440–450) (Fig. 1, S3 and S4). The model for the TMD was built using GluA2 TMD as a guide and residue assignment was verified using selenomethionine labeling (fig. S8, table S1), and electron density for aromatic residues and cross-link sites. Even though structural refinement was conducted using the most advanced refinement methods for treating low resolution data (16, 17), we suggest cautious interpretation of our structural model at the TMD, since there is some level of positional uncertainty intrinsic to a 4 Å model. No significant difference in the architecture between GluN1a/GluN2B_cryst and GluN1a/GluN2B_crystx was observed demonstrating that disulfide cross-linking of the subunits did not alter the overall structure.

Overall Structure

The GluN1a/GluN2B NMDA receptor bound to glycine, L-glutamate, and ifenprodil is shaped like a hot-air balloon where the balloon and basket correspond to the extracellular domains and the TMD, respectively (Fig. 1). There is a clear boundary between the layers of LBD and TMD, while the ATD and LBD appear as a single unit. The GluN1a and GluN2B subunits assemble as the staggered GluN1-GluN2-GluN1-GluN2 (1-2-1-2) heterotetramer in every domain as previously predicted (10, 18, 19) (Fig. 1–2). The GluN1a/GluN2B ATD and LBD heterodimers are similar to isolated GluN1b/GluN2B ATD complexed to ifenprodil (rmsd 0.9 Å) (8) and isolated GluN1/GluN2A LBD complexed to glycine and L-glutamate (rmsd 1.1 Å) (12). Observed electron density for glycine, L-glutamate, and ifenprodil supports the view that the structure represents the allosterically inhibited state (fig. S9). The assembly of the NMDA receptor tetramer as a dimer-of-dimers at the extracellular region is similar to the organization of the GluA2 AMPA receptor (10, 20) and likely of other iGluR members. Despite the similarity in the pattern of tetrameric arrangement, the overall shape of GluN1a/GluN2B NMDA receptor is distinct from that of the “Y” shaped GluA2 AMPA receptor (Fig. 1) (10). This is attributed to tight packing of the ATD and LBD in GluN1a/GluN2B NMDA receptors. In contrast ATD and LBD interact minimally in GluA2 AMPA receptors.

Organization of GluN1 and GluN2B subunits

There are two key features in the pattern of subunit arrangement. First, pseudo-symmetry mismatch is present between the extracellular region and the TMD. The GluN1a and GluN2B subunits are arranged in a 1-2-1-2 orientation with two-fold symmetry between the two GluN1a/GluN2B heterodimers in the ATD and LBD, but with pseudo-four-fold symmetry in the TMD (Fig. 2). A similar subunit arrangement is observed in the GluA2 homotetrameric structure that has two-fold symmetries within and between homodimers of the ATD and LBD and fourfold symmetry in the TMD (10) indicating that symmetry mismatch may be common to iGluR structures. The second important feature is swapping of dimer pairs between the ATD and LBD layers (Fig. 2A and B). In the ATD layer, heterodimer pairs assemble as GluN1a (α)/GluN2B (α) and GluN1 (β)/GluN2B (β) whereas in the LBD layer, they assemble as GluN1a (α)/GluN2B (β) and GluN1 (β)/GluN2B (α) (Fig. 2A and B). A similar pattern of domain swapping is also observed in the homotetrameric GluA2 AMPA receptor where subunits assemble as a dimer of A/B and C/D homodimers at the ATD and as a dimer of A/D and B/C dimers at the LBD (Fig. 2A and B) (10). Overall the conformations of GluN1a and GluN2B subunits approximately correspond to those of A/C and B/D subunits in the GluA2 AMPA receptor homotetramer. This assignment is based on the observation that the ATD-LBD linker and the M3-LBD linker are respectively more “distal” and “proximal” from the overall two-fold symmetry axis for GluN1a than GluN2B similar to the orientation of the A/C than B/D subunits in the GluA2 AMPA receptor (Fig. 2 and fig. S10).

Though they share a dimer-of-dimers arrangement, the modes of subunit association in each domain differ significantly between NMDA and AMPA receptors. In the ATD, the two GluN1a/GluN2B heterodimers interact with each other at two interfaces involving upper lobes of the two GluN1a subunits (α and β) and lower lobes of the two GluN2B subunits (α and β). In contrast the GluA2 receptor has only one interface between Subunit B and D (Fig. 2A and fig. S11). The ATDs of NMDA receptor and AMPA receptor subunits have low sequence identity and this is reflected in the large differences in structures (7, 8). Consequently, the ATD layer is more compactly packed in GluN1a/GluN2B than in GluA2. At the LBD, the dimer-of-dimers arrangement is comparable between the GluN1a/GluN2B and GluA2 receptors. The GluN1a/GluN2B heterodimers interact between the lower lobe of GluN1a and the upper lobe of GluN2B at the two equivalent sites (Fig. 2B and S11). In GluA2, the equivalent regions are much more loosely packed, but instead, there is another closely packed region involving Subunit A and C (10). Whether these distinct modes of interactions at the LBD represent an architectural difference between NMDA and AMPA receptors or different functional states remains an open question. In the previous study, the GluA2 AMPA receptor was crystallized in the presence of an antagonist (ZK200775) whereas the GluN1a/GluN2B NMDA receptor was crystallized in the presence of the allosteric inhibitor, ifenprodil, and agonists, glycine and L-glutamate. Finally, at the TMD, the subunits form pseudo four-fold symmetrical interactions between the M3 helices at the center of the ion channel and between M4 helices of one subunit and M1 and M3 helices of the adjacent subunit, similar to the GluA2 AMPA receptor (Fig. 2C and fig. S11). Within the pore, the M2 helices are close to the M1 and M3 helices of the same subunit and the M3 helices of the adjacent subunit.

Inter-subunit interfaces and function

The crystal structure of the heterotetrameric GluN1a/GluN2B NMDA receptor shows inter-subunit interfaces that are distinct from those in the GluA2 AMPA receptor. Thus, we have tested whether the distinct interfaces observed in our crystal structure are present in intact NMDA receptors in the membrane environment. We engineered cysteine residues and tested for spontaneous disulfide bond formation at the following subunit interfaces: (I). GluN2B (ATD)-GluN2B (ATD); (II). GluN1a (ATD)-GluN2B (LBD); (III). GluN1a (LBD)-GluN2B (ATD); (IV). GluN1a (LBD)-GluN2B (LBD); and (V). GluN1a (TMD)-GluN2B (TMD) (Fig. 3). The Western blot experiments on the mutant GluN1a/GluN2B NMDA receptor proteins have shown that all of the single cysteine mutant pairs in the extracellular region form disulfide cross-links resulting in GluN2B–GluN2B homodimer formation (I) and GluN1a-GluN2B heterodimer formation (II, III and IV), whereas the double cysteine mutant pair in the TMD forms GluN1a-GluN2B-GluN1a-GluN2B disulfide cross-links resulting in heterotetramer formation (V) in the absence of a reducing agent (Fig. 3B). In the presence of reducing agents, or when a subunit with the cysteine mutation is expressed with a subunit with no mutation, bands representing monomers of GluN1a and GluN2B appear (fig. S12). This indicates that the disulfide cross-links are formed by engineered cysteine pairs, and validates the physiological relevance of the subunit arrangement observed in our crystal structure.

What are the molecular determinants that favor the 1-2-1-2 arrangement over 1-1-2-2 arrangement? Previous studies have shown that the ATD is important for allosteric modulation and for controlling open probability and deactivation speeds (4, 5), but not required for formation of functional ion channels (21). Furthermore, interactions of helices in the TMD are similar to those in the homotetrameric GluA2 AMPA receptor structure, thus, the structural determinant for the 1-2-1-2 subunit arrangement may reside in the LBDs. In silico construction of GluN1a-GluN2B heterotetramers in the 1-1-2-2 format by superposing the GluN1 LBD structure onto the GluN2B LBD structure and vice versa revealed steric clashes in both the GluN1a-GluN1a and the GluN2B-GluN2B interfaces (fig. S13). The GluN1a-GluN1a interaction is prevented by a collision between Loop1 of one GluN1a and Helix G of the other, whereas the GluN2B-GluN2B interaction is disfavored by a steric hindrance between Helix K′ of one subunit and Helix E′ and F′ of the other (fig. S13). Thus, while the TMD is essential for tetramerization, the structural features in the LBD appear to favor the 1-2-1-2 arrangement.

The GluN1a/GluN2B heterotetrameric structure shows that some residues and motifs previously shown to play important roles in function are located at the interface between GluN1-GluN2B heterodimers (fig. S14). One example is Loop 1′ of GluN2B located within the LBD. The equivalent motif in GluN2A has been suggested to play a major role in negative cooperativity between the glycine and L-glutamate binding sites (22). Another example is a point mutation on Helix E of the GluN1 LBD lobe (Asp669Asn), which has previously been shown to affect gating properties by altering sensitivity to pH, spermine and ifenprodil (23). Our cross-linking results also showed “trapping” GluN1a-GluN2B ATD attenuates the ion channel activity. Taken together these results suggest that rearrangement between the two GluN1a-GluN2B dimers may be important for activity. Consistent with this, a recent study on GluA2 AMPA receptor showed that rearrangement of two GluA2 homodimers at the LBD regulates ion channel gating activity (24).

Inter-domain interaction between ATD and LBD and function

One of the major functional differences between NMDA receptors and non-NMDA receptors is the ATD-mediated regulation of ion channel activity present in the former and absent in the latter. In NMDA receptors, binding of allosteric modulators at the ATD alters agonist potency at the LBD indicating tight functional coupling between the ATD and the LBD (25). This coupling is structurally well represented by the more extensive ATD-LBD interaction in NMDA receptors (3,107 Ų) than in AMPA receptors (1,470 Ų). The crystal structure shows the two major sites, Site-II and –III, mediating the tight ATD-LBD association mainly through hydrophilic interactions even though the exact mode of residue-by-residue contacts cannot be pinpointed due to limitation in resolution (Fig. 3D and E). In Site-II, GluN1a ATD and the GluN1a/GluN2B LBD heterodimer are packed together through interaction between the loop extending from Helix α5 to Strand β7 at GluN1a ATD and GluN2B Helix J′ at the GluN1a-GluN2B interface at LBD (Fig. 3D). The region around GluN1a Helix α5 is where the twenty one residue long loop that contains numbers of basic amino acids would be present in the GluN1b splice variant. This loop, encoded by exon 5, has been shown to accelerate the deactivation time course (26), and to influence allosteric modulation by decreasing potency of protons, polyamines, and Zn²⁺ (27). In Site-III, the GluN1a LBD and the GluN1a/GluN2B ATD heterodimer are packed together by GluN1a Loop 2 “wedging” into the interface between GluN1a ATD and GluN2B ATD. Furthermore, GluN1a Helix F and G, GluN2B Loop 1′, and the loop extending from GluN2B Helix α4′ stack onto each other to further stabilize the ATD-LBD interaction (Fig. 3E). Helix F in GluN1 has been implicated in gating control, thus, this may be a key locus where ATD may have an impact on gating properties (28). Overall, the ATD and the LBD are in a tight arrangement that is suited to transmit structural changes between domains (fig. S15).

Transmembrane domain and the extracellular vestibule

The TMD of the GluN1a/GluN2B NMDA receptor forms a heterotetrameric ion channel with pseudo four-fold symmetry similar in overall shape to the homotetrameric ion channel of the GluA2 AMPA receptors (rmsd 2.2 Å) except the M4 helix in GluN2B (Fig. 4A). The tetrameric crossing of the M3 helices occludes the ion penetration pathway to a similar degree in the presumed allosterically inhibited GluN1a/GluN2B NMDA receptor to the closed GluA2 AMPA receptor (Fig. 4A). This crossing of the M3 helices occurs around the highly conserved SYTANLAAF motifs in iGluRs, mutations of which are known to modify gating properties (29). The ion channel pore (M1 to M3) of the GluN1a/GluN2B NMDA receptor shows high structural similarity to that of KcsA potassium channels in a closed conformation (30) (rmsd 2.4 Å) despite the low sequence identity (19%) (fig. S16). In contrast, shaker (31) and MthK (32) potassium channels in open conformation do not superpose well, mainly due to bending of their TM2 helices compared to the M3 helices in GluN1a/GluN2B NMDA receptor (Fig. 4B and fig. S16). On the basis of structural similarities with the potassium channel, we speculate that gating of the GluN1a/GluN2B NMDA receptor may involve rearrangement of M3 helices.

One of the hallmarks of NMDA receptor function is the high permeation of calcium ions, which plays a major role in neuronal plasticity as well as excitotoxicity. The crystal structure complexed with holmium or gadolinium, lanthanides known to occupy calcium binding sites in many biological macromolecules (33, 34), show binding between the LBD-TMD linkers from the two GluN1 subunits around the center of the ion channel (Fig. 5). A set of acidic residues in GluN1 (DRPEER motif) located in this region is critical for the high calcium flux characteristic of NMDA receptors (35). Thus, the lanthanide binding site along with the previous electrophysiological study further confirms the presence of the calcium pool located right outside of the ion channel. A similar charge-based ion pooling mechanism has been suggested for cation conductance in P2X4 receptors and nicotinic acetylcholine receptors (36, 37). Despite extensive efforts, the regions of the TMD that determine voltage-dependent Mg²⁺ block and Ca²⁺ permeation (38) are not resolved in this crystal structure. Structure-based understanding of cation selectivity and voltage-dependent Mg²⁺ block is thus a question that remains to be addressed.

Conclusion

The crystal structure of GluN1a/GluN2B NMDA receptors in the current study reveals the patterns of inter-subunit and –domain interactions, which are different from those observed in AMPA receptors. These differences widely reflect functional differences. The structure will serve as a template for designing experiments that addresses functional questions specific to NMDA receptors. Finally, the defined subunit interfaces should serve as an important blueprint for design of therapeutic compounds.

Supplementary Material