Cryo-EM structures of the triheteromeric NMDA receptor and its allosteric modulation.

Construct design GluN1, GluN2A and GluN2B constructs used here, deemed GluN1 EM , GluN2A EM and GluN2B EM , were designed based on the previously reported Xenopus laevis NMDA GluN1-Δ1 and GluN2B-Δ2 constructs ( 26 , 27 ). GluN1 EM is generated by removal of the C-terminal of GluN1-Δ2, including the enhanced green fluorescent protein (eGFP), 3C cleavage site (Leu-Glu-Val-Leu-Phe-Gln-Gly-Pro) and an octa-histidine tag. GluN2B EM reverts the K216C mutation in GluN2-Δ2 to a wild-type lysine residue, with a GFP11, SH3 fusion protein and an octa-histidine tag placed at its C-terminus. GluN2A EM was designed based on the GluN2B EM construct, but including the C-terminal 3C cleavage site, eGFP and strepII tag. Due to the coexistence of diheteromeric GluN1/GluN2A and GluN1/GluN2B, we used split GFP to monitor the expression level of triheteromeric NMDAR over diheteromeric NMDAR. Briefly, the GFP can be cleaved into two parts including GFP1-10 and GFP11 and these can be reassembled into a functional intact GFP by fusing them into interacting protein subunits ( 55 ). We fused GFP1-10 and GFP11 into the C-terminal of bicistronic GluN1-GluN2A and GluN2B, respectively. As a result, only triheteromeric GluN1/GluN2A/GluN2B containing both GFP1-10 and GFP11 will fluoresce, and the diheteromeric GluN1/GluN2A or GluN1/GluN2B containing either GFP1-10 or GFP11 will not, which was useful for initial construct screening. We replaced the GFP1-10 in the bicistronic GluN1-GluN2A construct with an intact GFP for large-scale expression because monitoring the GFP fluorescence was important to control the quality of virus production for the bicistronic construct. The final constructs we used for EM experiments are listed in Fig. S1 I . To boost expression levels, GluN1 EM and GluN2A EM constructs were cloned into the same pEG BacMam vector, with the result deemed the GluN1-GluN2A EM construct.

Purification of tri-NMDARs Bacmid and baculovirus of GluN1-GluN2A

EM and GluN2B EM in BacMam vector were generated ( 29 ) and P2 viruses were used to infect suspension HEK293 GnTI - cells at a multiplicity of infection (M.O.I.) of 1:1 (GluN1-GluN2A EM :GluN2B EM ) and then incubated at 37 °C. At 12 h post-transduction, 10 mM sodium butyrate and 2.5 μM MK-801 were added to the culture and the temperature was set to 30 °C. The cells were collected and resuspended in a buffer containing 150 mM NaCl, 20 mM Tris 8.0 in the presence of 1 mM PMSF, 0.8 μM aprotinin, 2 μg/ml leupeptin, and 2 mM pepstatin A (protease inhibitors). The receptor was extracted from whole cell with a buffer containing 150 mM NaCl, 20 mM Tris 8.0, 1% MNG-3, protease inhibitors and 2 mM cholesteryl hemisuccinate (CHS) for 2 h at 4 °C. The solubilized receptors containing GluN1/GluN2A, GluN1/GluN2B and GluN1/GluN2A/GluN2B were incubated with TALON resin to remove strepII-tagged GluN1/GluN2A. The bound GluN1/GluN2B and GluN1/GluN2A/GluN2B receptors were then eluted with 250 mM imidazole at pH 8.0 to streptactin resin. His-tagged GluN1/GluN2B passed through the column and only GluN1/GluN2A/GluN2B was bound to streptactin resin and eluted with buffer containing 5mM desthiobiotin. The receptor was concentrated, mixed with Fab 11D1 at a molar ratio 1:1.2 and was further purified by size-exclusion chromatography in the buffer containing 400 mM NaCl, 20 mM MES pH 6.5, 0.5 mM n-dodecyl β-D-maltoside (DDM), and 0.2 mM CHS. Peak fractions containing the receptor were pooled and concentrated to 4 mg/ml.

Antibody production Monoclonal antibodies against

GluN1 and GluN2B (10B11 and 11D1, respectively) were raised by Dan Cawley (Vector and Gene Therapy Insititute, OHSU) using standard methods. GluN1/GluN2 was purified as described previously ( 26 ) in 1 mM DDM and in the presence of 1 mM glutamate, 1 mM glycine, and 0.2 mM CHS. Purified GluN1/GluN2 was reconstituted into liposomes for immunization as described previously for SERT ( 56 ) ( 57 ), except 400 mM NaCl and 0.8% Na deoxycholate was used for reconstitution of GluN1/GluN2 and excess salt and detergent was removed using PD-10 desalting columns. Mice were immunized with 30 μg of reconstituted GluN1/GluN2B to generate hybridoma cell lines. Antibodies were screened by fluorescence-based size-exclusion chromatography ( 28 ) and western blot to select clones that recognized natively folded GluN1/GluN2B and GluN1/GluN2A protein. The 10B11 and 11D1 monoclonal antibodies were purified from hybridoma supernatants using 4-mercapto-ethyl-pyridine chromatography resin. Fab was generated by papain cleavage of 10 mg mAb at 1 mg/ml final concentration for 2 h at 37 °C in 50 mM NaPi, pH 7.0, 1 mM EDTA, 10 mM cysteine and 1:100 w/w papain. The digest was quenched with 30 mM iodoacetamide at 25 °C for 10 min and Fc was removed from the MAb digest using Protein A. Fab 10B11 was purified by anion exchange using a Hi Trap Q HP column in 20 mM Tris pH 8 and 200 mM NaCl. Fab 11D1 was purified by cation exchange using a Hi Trap SP HP column in 20 mM NaOAc pH 5 and 200 mM NaCl. EM data acquisition and processing Purified GluN1/GluN2A/GluN2B was mixed with 2 mM glycine, 2 mM glutamate, 20 μM MK-801, 0.5 mM EDTA and/or 1 mM Ro (in DMSO) a few hours before grid preparation. Double blotting of a 1.3+2.5 μl sample at a concentration of 4 mg/ml was applied to a glow-discharged Quantifoil holey carbon grid (gold, 1.2/1.3 μm size/hole space, 300 mesh), blotted using a Vitrobot Mark III using 3s blotting time with 100% humidity, and then plunge-frozen in liquid ethane cooled by liquid nitrogen. Images were taken by an FEI Titan Krios electron microscope operating at 300 kV with a nominal magnification of 85k. Images were recorded by a Gatan K2 Summit direct electron detector operated in super-resolution counting mode with a binned pixel size of 1.70 Å or 1.33 Å for Fab-bound data or non-Fab-bound data, respectively. For the non-Ro-bound data, each image was dose-fractionated to 70 frames with a total exposure time of 21 s with 0.3 s per frame. For the Ro-bound data, each image was dose-fractionated to 50 frames with a total exposure time of 20 s with 0.4 s per frame. The images were recorded using the automated acquisition program SerialEM. Nominal defocus values varied from 1.3 to 2.5 μm. For the Ro-bound data, super-resolution counting images were 2x2 binned in Fourier space, motion corrected and summed using MotionCor2 ( 58 ). Defocus values were estimated using Gctf ( 59 ). Approximately 3000 particles were manually picked and subjected to an initial reference-free 2D classification using Relion ( 60 ). Eight representative 2D class averages were selected as templates for automated particle picking for all the Fab-bound data using Gautomatch ( http://www.mrc-lmb.cam.ac.uk/kzhang/Gautomatch/ ). The auto-picked particles were visually checked and false positives were removed. The particles were further cleaned-up by two rounds of 2D classification using Relion. The CTF values of individual particles from selected 2D class averages were estimated using Gctf ( 59 ). For 3D classification in Relion, a Fab-free reference model was generated from the GluN1/GluN2B crystal structure (PDB code: 4TLM) and low-pass filtered to 50 Å using EMAN2 ( 61 ). The 3D classes (5 for Fab-bound data or 6 for non-Fab-bound data) each occupy a similar percentage of particles and share similar overall shape with differences mainly due to the flexibility of the Fab and the intrinsic flexibility between the extracellular domains and transmembrane domain of the receptor. Three-dimensional refinement of individual classes yielded low-resolution reconstructions. By combining particles from all the classes, we obtained a reconstruction at higher resolution. Initial 3D refinement was carried out using Relion. Particles were further refined using Frealign’s local refinement ( 62 ). Subsequently, for all the Fab-bound data, a soft mask around the receptor was calculated and supplied for final refinement using Frealign. The final resolutions reported in Table S1 are based on the gold standard FSC 0.143 criteria ( 60 ). No symmetry was applied during the image processing. A similar procedure was used for the other three datasets with the exception that no additional soft mask was used during 3D refinement of non-Fab-bound data. Model building A homology model of the triheteromeric GluN1/GluN2A/GluN2B receptor was generated with the crystal structure of the diheteromeric GluN1/GluN2B receptor ( 26 ) (PDB code: 4TLM) as a template using the SWISS-MODEL online server ( 63 ). A homology model for the Fab was made using the Fab from an existing crystal structure (PDB code: 4M48) by mutating all the residues to alanine. Both homology models were first rigid-body fitted into the non-Ro EM density map using Chimera ( 64 ), followed by molecular dynamics flexible fitting (MDFF) ( 32 ), which improves the model to map correlation coefficients (CC, without Fab, the same below) from 0.803 to 0.911 (backbone, 0.775 to 0.876). This model was then subjected to Rosetta refinement and a total number of 100 models were generated. The top 10 models with best geometry statistics were inspected for map agreement using Coot ( 65 ). The chosen model has slightly improved CC (0.919 and 0.883 for all atoms and backbone, respectively). In addition, the positioning of loops/linkers is improved and some errors related to secondary structure assignments and geometry are corrected. The model from Rosetta was further manually adjusted in Coot, guided by the crystal structures of intact diheteromeric GluN1/GluN2B (PDB code: 4TLM) ( 26 ) and GluN1/GluN2A LBD domains (PDB code: 2A5T) ( 34 ), as well as by the densities of bulky side chains. The final model has a CC of 0.929 and 0.892 for all atoms and backbone, respectively. For the Ro-bound data, the structure of the non-Ro-bound state was first rigid-body fitted into the EM density map, followed by MDFF. Subsequently, each subdomain of the non-Ro-bound structure, including the R1 and R2 lobes of ATD, D1 and D2 lobes of LBD, and TMD, were aligned to the Ro-bound model and combined to a new Ro-bound model. This model, and the linkers between subdomains in particular, was inspected and manually adjusted in Coot, guided by the non-Ro-bound structure, the crystal structure of intact diheteromeric GluN1/GluN2B (PDB code: 4TLM) and the crystal structure of the LBD of diheteromeric GluN1/GluN2A (PDB code: 2A5T). For validation, FSC curves were calculated between the final models and EM maps. The geometries of the atomic models were evaluated using MolProbity ( 66 ). All figures were prepared using UCSF Chimera and Pymol (Schrödinger) ( 67 ) . Saturation binding experiments MK-801 and Ro binding affinity was determined by scintillation-proximity assay (SPA). SPA experiments were set up in triplicate wells of a 96-well plate. Affinity-purified triheteromeric NMDARs (20 nM) saturated with 2 mM glyine and 2 mM glutamate was incubated with 1 mg/ml copper yttrium silicate (Cu-YSi) beads (Perkin Elmer) and 3 H-labelled MK-801 or Ro (1:9 3 H: 1 H) in SEC buffer (20 mM Tris pH 8, 150 mM NaCl, 1 mM MNG-3 and 0.2 mM CHS) with a final volume of 100 μl. Non-specific binding was determined by the addition of 1 mM ifenprofil (for 3 H-Ro) or 1 mM PCP (for MK-801). Plates were incubated at room temperature until the counting was stable and were read using a MicroBeta TriLux 1450 LSC and luminescece counter. Data were analyzed using GraphPad Prism. Two-electrode voltage clamp electrophysiology (TEVC) The GluN1, GluN2A, and GluN2B constructs for TEVC experiments in the pGEM vector are engineered with the C-terminal tags according to methods developed by the Hansen lab for selective cell-surface expression of recombinant triheteromeric NMDAR ( 22 ). The RNAs were transcribed using the mMessage mMachine T7 Ultra Kit (Ambion). Xenopus laevis oocytes were injected with a total 30–200 ng of mRNA with a ratio of GluN1:GluN2A:GluN2B 2:1:1 and were incubated at 16 °C for 2–3 days in the presence of 50 μM competitive antagonist D-APV and 10 μg/ml gentamicin. Borosilicate pipettes were filled with 3 M KCl. The recordings were performed in a buffer containing 100 mM NaCl, 0.3 mM BaCl 2 , 5 mM HEPES 7.3 and 0.05 mM heavy-metal chelator ethylenediaminetetraacetic acid (EDTA) at −60 mV. All recording experiments were carried out at least 3 times independently.

Supplementary Material Movie S1 Movie S1.

Cryo-EM map and model of non-Ro-bound triheteromeric

NMDAR and its conformational transition upon binding of Ro. The movie first shows the overall density map and the fitted atomic model of the non-Ro-bound triheteromeric NMDAR. The movie then shows the local densities for the ATD-LBD linkers, LBD-TMD linkers and TMD. Subsequently the movie shows the differences between GluN1/GluN2A and GluN1/GluN2B heterodimers. At the end the movie shows a conformational transition induced upon binding of Ro. Supplementary information Figure S1. Construct design, detection and preparation of the triheteromeric GluN1/GluN2A/GluN2B complex. (A-C) Cartoon representations of the triheteromeric GluN1/GluN2A/GluN2B construct. GluN1, GluN2A and GluN2B are shown in orange, red, and blue, respectively. Deletions in the ATD linker of GluN2A and GluN2B are highlighted with a yellow wedge. The point mutations are highlighted in white circles. Mutations of glycosylation sites in GluN1 (N300Q, N350Q, N368D, N440D, N469D, and N769E), in GluN2A (N67Q, N372A, N431A, N529A, and N675A) and GluN2B (N69Q, N343D and N486V) are not shown. Also not shown are mutations in signal peptide for GluN1 (C22A) and GluN2B (M20S, G21R and C22A), and mutation of a potentially reactive cysteine in GluN2B (C581A). Where noted, some constructs contain mutations in the TMD of NR1 (G610R and M816Y). ( D ) Activation of triNMDAR wt current by 0.2 mM/0.2 mM glutamate/glycine at pH 7.3 by TEVC. ( E-H ) Inhibition of triNMDA EM -G610/GluN2A EM /GluN2B EM (E), GluN1 EM -G610-M816/GluN2A EM /GluN2B EM (F) or GluN1 WT /GluN2A EM /GluN2B EM (G) current by 1μM Zn or 3μM Ro, respectively, and the extents of inhibition (H). Current responses were activated by 0.2mM/0.2mM glutamate/glycine at pH 7.3. The extents of Zn or Ro inhibition were determined from measurements on 5 oocytes and calculated using the ratio of current following Zn or Ro (asterisk) application and the glutamate/glycine-induced ‘peak’ current. The error bars represent s.e.m. ( I ) Summary of Xenopus laevis NMDA cryo-EM construct design. (J) SDS gel of purified triheteromeric NMDARs. To differentiate the three subunits, receptors were treated using EndoH and 3C protease prior to loading on the SDS gel. A band at molecular weight 60 kD marked with an asterisk corresponds to EndoH and two bands at lower molecular weights represent 3C protease, cleaved GFP and GFP11-SH3, respectively. Note that the receptors for cryo-EM experiments were not treated by EndoH and 3C protease. (K) Schematic cartoon showing two binding sites for GluN1-specific Fab 10B11 and one binding site for GluN2B-specific Fab 11D1 in the triheteromeric GluN1/GluN2A/GluN2B complex. (L) FSEC traces show a shift of the diheteromeric GluN1/GluN2A complex upon binding to the GluN1-specific Fab 10B11, but no shift with the GluN2B-specific Fab 11D1. (M) The triheteromeric GluN1/GluN2A/GluN2B complex binds to both GluN1-specific Fab 10B11 and GluN2B-specific Fab 11D1; notably, the shift with the GluN1-specific Fab 10B11 is twice that of the GluN2B-specific Fab 11D1. Figure S2. The work-flow of cryo-EM data processing, using the non-Ro-bound data as an example. Particles were auto-picked from 2751 micrographs using Gautomatch and visually checked in Relion. After removing false positives, a total number of 343485 particles were subjected to two rounds of 2D classification in Relion. The CTF values of individual particles (302052) from selected 2D class averages were estimated using Gctf. For 3D classification in Relion, a Fab-free reference model was generated from the GluN1/GluN2B crystal structure (PDB code: 4TLM) and low-pass filtered to 50 Å using EMAN2. The 3D classes each occupy a similar percentage of particles and share similar overall shape with differences mainly due to the flexibility of the Fab and the intrinsic flexibility between the extracellular domains and transmembrane domain of the receptor. Three-dimensional refinement of individual classes yielded low-resolution reconstructions. By combining particles from all the classes, we obtained reconstruction of higher resolution. Initial 3D refinement was carried out using Relion. Particles were further refined using Frealign’s local refinement. Subsequently, for all the Fab-bound data, a soft mask around the receptor was calculated and supplied for the final refinement using Frealign. A similar procedure was used for the other three datasets with the exception that no additional soft mask was used during 3D refinement of non-Fab-bound data. Figure S3.

Cryo-EM analysis of non-Ro-bound triNMDAR

EM in complex with Fab 11D1. (A) Representative electron micrograph. (B) Selected two-dimensional class averages of the electron micrographs. (C) Reference map generated from the GluN1/GluN2B crystal structure (PDB-code: 4TLM) (left), 3D classification with the percentage of particles in each class from 1 to 5, overlay of the five 3D classes (marked with *) with the total number of particles (right). (D) The gold-standard Fourier shell correlation curves for the EM maps with (red) and without (black) masking the Fab. The FSC curve between the atomic model and the final EM map is shown in blue. (E) Angular distribution of particles used for refinement. (F) The three-dimensional map is colored according to local resolution estimation. Figure S4.

Cryo-EM analysis of Ro-bound triNMDAR

EM in complex with Fab 11D1. (A) Representative electron micrograph. (B) Selected two-dimensional class averages of the electron micrographs. (C) Reference map generated from the GluN1/GluN2B crystal structure (PDB-code: 4TLM), 3D classification with percentage of particles in each class, overlay of the five 3D classes (marked with *) with the total number of particles. (D) The FSC curves for the EM maps with (red) and without (black) masking the Fab. The FSC curve between atomic model and the final EM map is shown in blue. (E) Angular distribution of particles used for refinement. (F) The three-dimensional map is colored according to local resolution estimation. Figure S5. Cryo-EM analysis of non-Ro-bound triNMDAR EM –G610 in complex with Fab 11D1. (A) Representative electron micrograph. (B) Selected two-dimensional class averages of the electron micrographs. (C) Reference map generated from the GluN1/GluN2B crystal structure (PDB-code: 4TLM), 3D classification with percentage of particles in each class, overlay of the five 3D classes (marked with *) with the total number of particles. (D) The FSC curves for the EM maps with (red) and without (black) masking the Fab. The FSC curve between the atomic model (non-Ro-bound triNMDAR EM ) and the final EM map is shown in blue. (E) Angular distribution of particles used for refinement. (F) The three-dimensional map is colored according to local resolution estimation. Figure S6. Comparison of non-Ro-bound triNMDAR EM -G610 with non-Ro-bound triNMDAR EM . (A) Density map of non-Ro-bound triNMDAR EM −G610. (B) Overlay of the triNMDAR EM −G610 map with the atomic model of non-Ro-bound triNMDAR EM . The map is shown as a transparent surface and the model is shown in cartoon representation. (C) Overlay of the same map and model as in (B) for each subunit. The non-Ro-bound triNMDAR EM model was rigid-body fitted into the non-Ro-bound triNMDAR EM -G610 density using Chimera. Figure S7.

Cryo-EM analysis of Ro-bound triNMDAR

EM in the absence of Fab (Ro-bound/non-Fab-bound triNMDAR EM ). (A) Representative electron micrograph. (B) Selected two-dimensional class averages of the electron micrographs. (C) Reference map generated from the GluN1/GluN2B crystal structure (PDB-code: 4TLM), 3D classification with percentage of particles in each class, overlay of the five 3D classes (marked with *) with the total number of particles. (D) The FSC curve for the EM map is shown in black and the FSC curve between the atomic model (Ro-bound triNMDAR EM ) and the final EM map is shown in blue. (E) Angular distribution of particles used for refinement. (F) The three-dimensional map is colored according to local resolution estimation. Figure S8. Overlay of the EM map of the Ro-bound/non-Fab-bound triNMDAR EM with the atomic model of Ro-bound triNMDAR EM . (A-D) Superimposition of the map with the model viewed from different orientations. The Ro-bound triNMDAR EM model was rigid-body fitted into the Ro-bound/non-Fab-bound triNMDAR EM density in Chimera. Figure S9. Representative densities of the non-Ro-bound triNMDAR EM . (A–B) Cryo-EM map viewed parallel to the membrane (A) and from the extracellular side (B), respectively. The Fab was included during refinement. (C) The interface between the receptor and the Fab. The atomic model is shown in cartoon representation and EM density is shown as transparent surface. The positions of two NAG (N-linked glycans) sites interacting with the Fab are indicated. (D-E) The densities of the two NAG moieties interacting with the Fab. (F–G) Densities of the helix α6 in GluN1. (H) Densities of the helix α5 in GluN2A. (I) Densities of the helix αK in GluN1. (J) Densities of the M2 in GluN2A. (K) Densities of the M3 in GluN2A. Figure S10. The ATD of the GluN2A subunit. (A) Sequence and secondary structure prediction of GluN2A ATD. (B–C) The densities and atomic model of GluN2A ATD. Figure S11. Comparison of the ATD heterodimers of the non-Ro-bound triNMDR EM with the non-Ro-bound, Ro-bound diheteromeric GluN1/GluN2B and isolated zinc-bound GluN1/GluN2A ATD, respectively. For comparisons in A-D, the structures were superimposed using the R1 lobe of the GluN1 ATD. The distances (Å) between the R2 lobes of the GluN1 and GluN2 subunits and the rotation angles between the two GluN2 R2 lobes are indicated. (A) Comparison of the GluN1/GluN2A ATD heterodimer with non-Ro-bound diheteromeric GluN1/GluN2B. (B) Comparison of the GluN1/GluN2A ATD heterodimer with Ro-bound diheteromeric GluN1/GluN2B. (C) Comparison of the GluN1/GluN2B ATD heterodimer with non-Ro-bound diheteromeric GluN1/GluN2B. (D) Comparison of the GluN1/GluN2B ATD heterodimer with Ro-bound diheteromeric GluN1/GluN2B. (E-F) Comparison of the GluN1/GluN2A ATD heterodimer with the isolated zinc-bound GluN1/GluN2A ATD heterodimer, viewed from the extracellular side of the membrane (E) or in parallel to the membrane (F). The structures were superimposed using the R1 lobe of the GluN2A ATD. Figure S12. Comparison of the LBD heterodimer of non-Ro-bound triNMDAR EM with the soluble LBD heterodimer of diheteromeric GluN1/GluN2A, all bound with glycine and glutamate. The structures are superimposed using the GluN1 LBD. (A) Putative density for bound glutamate in the GluN2A LBD clamshell. (B–D) The positions of the GluN1/GluN2A LBD heterodimer in the triheteromeric NMDAR (A) and its comparison with soluble diheteromeric GluN1/GluN2A (B-C). (E–G) The positions of the GluN1/GluN2B LBD heterodimer in the triheteromeric NMDAR (D) and its comparison with the soluble diheteromeric GluN1/GluN2A (E–F). Figure S13. Comparison of the ATD heterodimer of Ro-bound triNMDAR EM with the non-Ro-bound and Ro-bound diheteromeric GluN1/GluN2B, respectively. The structures were superimposed using the R1 lobe of the GluN1 ATD. The distances (Å) between the R2 lobes of the GluN1 and GluN2 subunits and the rotation angles between the two GluN2 R2 lobes are indicated. (A) Comparison of the GluN1/GluN2A ATD heterodimer with non-Ro-bound diheteromeric GluN1/GluN2B. (B) Comparison of the GluN1/GluN2A ATD heterodimer with Ro-bound diheteromeric GluN1/GluN2B. (C) Comparison of the GluN1/GluN2B ATD heterodimer with non-Ro-bound diheteromeric GluN1/GluN2B. (D) Comparison of the GluN1/GluN2B ATD heterodimer with Ro-bound diheteromeric GluN1/GluN2B. Table S1. Statistics of data collection, 3D reconstruction and models.