Abstract
Antibodies against N-methyl-d-aspartate receptors (NMDARs) are most frequently detected in persons with autoimmune encephalitis (AE) and used as diagnostic biomarkers. Elucidating the structural basis of monoclonal antibody (mAb) binding to NMDARs would facilitate the development of targeted therapy for AE. Here, we reconstructed nanodiscs containing green fluorescent protein-fused NMDARs to label and sort individual immune B cells from persons with AE and further cloned and identified mAbs against NMDARs. This allowed cryo-electron microscopy analysis of NMDAR–Fab complexes, revealing that autoantibodies bind to the R1 lobe of the N-terminal domain of the GluN1 subunit. Small-angle X-ray scattering studies demonstrated NMDAR–mAb stoichiometry of 2:1 or 1:2, structurally suitable for mAb-induced clustering and endocytosis of NMDARs. Importantly, these mAbs reduced the surface NMDARs and NMDAR-mediated currents, without tonically affecting NMDAR channel gating. These structural and functional findings imply that the design of neutralizing antibody binding to the R1 lobe of NMDARs represents a potential therapy for AE treatment.
Similar content being viewed by others
Main
The N-methyl-d-aspartate receptors (NMDARs) are excitatory glutamate-gated ion channels highly expressed in the hippocampus with an essential function in learning and memory1,2,3. In recent years, autoantibody-mediated crosstalk between the peripheral immune systems and the brain has attracted much attention in both basic and clinical neuroscience. Anti-NMDAR autoantibodies were frequently detected in the serum or cerebrospinal fluid (CSF) of persons with autoimmune encephalitis (AE)4, as shown by their strong reactivity to the rodent hippocampal neuropil5. Persons with AE carrying anti-NMDAR antibodies typically show severe symptoms including psychosis, memory deficit, seizure, dyskinesias, reduced consciousness and autonomic dysfunction6. Clinical evidence indicates that viral infection7 or peripheral tumorigenesis5 could initiate the immune process and production of autoantibodies, which cross the blood–brain barrier and directly bind to NMDARs on the synaptic membrane4,8. To date, clinical treatments of persons with AE are restricted mainly to nonspecific immunotherapies, including steroid administration, plasma exchange and intravenous injection of immunoglobulins4.
Over the past decade, the pathogenesis of NMDAR-related AE has been studied from behavioral to synaptic levels. Mouse models of AE displaying memory deficit and neurological abnormalities have been generated by passive cerebroventricular transfer of patients’ total IgGs9,10 or active immunization with NMDAR-incorporated liposomes11. At the synapse level, anti-NMDAR autoantibodies disrupted the association between NMDARs and cell adhesion proteins Ephrin and EphB2 (refs. 12,13,14) and subsequently caused the internalization of NMDARs and downregulation of synaptic functions15,16,17,18,19,20. Single-molecule imaging also indicated that patients’ IgGs increased the surface NMDAR clustering and disrupted synaptic organization20,21. Most of these studies were carried out using total IgGs from patients displaying variable phenotypes, leading to some inconsistent results16,17,22,23. Cloning and identification of patient-derived monoclonal antibodies (mAbs) may greatly help to elucidate the molecular action of autoantibodies on the NMDARs.
In this study, a sorting and cloning protocol was developed to increase the success rate of obtaining patient-specific mAbs. We first reconstituted green fluorescent protein (GFP)-tagged NMDARs (NMDARGFP) in nanodiscs and used fluorescence-activated cell sorting (FACS) to isolate single immune cells of patients that expressed NMDAR-specific auto-mAbs on their surfaces. We then cloned the sequences of the heavy-chain and light-chain pairs of auto-mAbs against NMDARs from isolated immune cells. Finally, we performed structural determination of NMDAR–fragment antigen binding (Fab) and NMDAR–mAb complexes by cryo-electron microscopy (cryo-EM) and small-angle X-ray scattering (SAXS), respectively, and validated the pathogenic actions of these auto-mAbs. These studies led to the identification of two distinct auto-mAb-binding epitopes on human NMDARs and the stoichiometry of NMDAR–mAb complexes, as well as the demonstration of auto-mAb-mediated downregulation of surface NMDARs and function.
Cloning of patient-derived auto-mAbs against NMDARs
A previous study cloned a panel of mAbs from total B cells from the CSF of persons with NMDAR-related AE, with ~10% of isolated single B cells producing anti-NMDAR antibodies24. To increase the cloning efficiency for auto-mAbs from human samples, we developed an antigen-based FACS for isolating individual B cells from persons with AE. To obtain those cells with NMDAR-specific mAbs on their surfaces, we established a cell marking strategy using conformationally stabilized tetrameric NMDARGFP. Human GluN1–GluN2A NMDARGFP was initially purified in a detergent-solubilized buffer as previously reported25,26. To preserve the integrity of immune cells from detergent exposure, we reconstituted NMDARGFP into nanodiscs consisting of soybean polar lipids and membrane scaffold protein MSPΔH5 (ref. 27). We found that a mixture of NMDARGFP, lipids and MSPΔH5 at a molar ratio of 1:240:4 could successfully stabilize tetrameric-formed NMDARGFP in nanodiscs, which displayed a monodispersed fluorescence size-exclusion chromatography (FSEC) profile in the detergent-free buffer (Fig. 1a–c).
a, Workflow of cloning human auto-mAbs from persons diagnosed with NMDAR encephalitis. The method includes the reconstruction of purified NMDARGFP into nanodiscs, memory B cell (MBC) and antibody-secreting cell (ASC) enrichment from human PBMCs, NMDARGFP-labeled FACS to isolate individual B cells, reverse transcription followed by nested PCR and recombinant expression of mAb and Fab protein. b, SEC of GluN1–GluN2A NMDARGFP constructed in nanodiscs, with tetrameric NMDARGFP protein stabilized in detergent-free PBS buffer. c, FSEC profiles and Coomassie blue staining gel of the reconstructed NMDARGFP in nanodiscs, NMDARGFP in detergent and scaffold protein MSP∆H5 alone. The experiments were repeated for three times with similar results. d, Representative FACS plots of B cells with (left) or without (right) NMDARGFP labeled. B cells highlighted in the red box were individually collected for antibody cloning. e,f, Representative FSEC profiles (e) and Coomassie blue staining gel (f) of recombinant expressed mAb5F6, Fab5F6, mAb2G7 and Fab2G7. The experiments were repeated for three times with similar results. g, SPR binding curves for mAb5F6 (left) and mAb2G7 (right) interacting with GluN1–GluN2A NMDAR. The estimated KD of mAb5F6 and mAb2G7 was 4.31 μM and 11.3 μM, respectively. RU, resonance units.
The fluorescence-marked NMDARGFP protein was then used for FACS to label the targeted immune B cells from persons with AE. We collected fresh peripheral blood samples from five persons with NMDAR AE with psychiatric symptoms, cognitive deficits, seizures and movement abnormalities (Supplementary Table 1). Peripheral blood mononuclear cells (PBMCs) were harvested and then non-B cells were depleted by biotin-conjugated antibodies (Miltenyi Biotec) to their markers, including CD2, CD14, CD16, CD36, CD43 and CD235a. Further FACS of B cells (memory B cells and antibody-secreting cells) labeled with IgD and CD27 markers, in the presence or absence of NMDARGFP, identified ~0.08% B cells that were positively labeled with NMDARGFP (Fig. 1d). These GFP-marked cells were individually collected to produce complementary DNA (cDNA) by reverse transcription. The sequences encoding variable regions of light and heavy chains were amplificated by nested PCR24,28 and further subcloned into the mAb expression vector containing the constant region29, allowing the expression of recombinant mAbs. In total, we obtained 11 patient-derived mAbs (Supplementary Table 2), four of which were confirmed for NMDAR-specific binding on the basis of immunostaining of live HEK293T cells expressing surface NMDARs. Among them, mAb5F6 and mAb2G7 displayed monodispersed FSEC profiles (Fig. 1e) and were used for all subsequent experiments. To assess the binding affinity of auto-mAbs on NMDAR, surface plasmon resonance (SPR) analysis revealed that mAb5F6 and mAb2G7 had a KD of 4.31 μM and 11.3 μM for binding to human GluN1–GluN2A NMDARs, respectively (Fig. 1g). Immunostaining of mouse brain slices further demonstrated that both mAbs displayed intensive immunoreactivity to the hippocampal neuropil at 60 μg ml−1 (Extended Data Fig. 1a,b). Thus, our NMDARGFP-labeled FACS strategy achieved the identification of NMDAR-targeted auto-mAbs from persons with AE. The advantage of this sorting strategy is the improved cloning rate of targeted B cells, while the disadvantage is the additional step of preparing nanodiscs with GFP-fused NMDARs.
Structural determination of NMDAR–Fab complex
We next sought to determine the precise epitope on NMDAR for the binding of AE-relevant mAbs. Functional NMDARs are tetramers comprising two obligatory GluN1 subunits and two alternative GluN2 (2A–2D) or GluN3 (3A and 3B) subunits1,30. To identify the subunit recognition of patient auto-mAbs, we cotransfected mammalian HEK293T cells with two plasmids encoding GFP-fused GluN1 and alternative GluN2 or GluN3 subunits. Through cell-based immunostaining, we found that mAb5F6 and mAb2G7 costained with all subtypes of GFP-fused NMDARs (Fig. 2a,b), indicating that both antibodies targeted to the obligatory GluN1 subunits.
a,b, Immunocytochemistry staining on live HEK293 cells expressing recombinant NMDARs associated with GluN1 and GluN2 (2A–2D) or GluN3 (3A and 3B) subunits (in green) stained with mAb5F6 (a), mAb2G7 (b) (in red) and DAPI (in blue). Scale bars, 15 μm. c,d, The 2D class-average images of NMDAR–Fab5F6 (c) and NMDAR–Fab2G7 (d) complexes, with triangles indicating the Fabs. e,f, Side and top-down views of cryo-EM density of the NMDAR–Fab5F6 (e) and NMDAR–Fab2G7 (f) complexes. The NTDs, LBDs and TMD are indicated. The length and width of the cryo-EM density of NMDAR–Fab complex are marked. g,h, Cartoon illustration of NTD–Fab5F6 (g) and NTD–Fab2G7 (h) structures; R1 and R2 lobes are circled with dashed lines.
Next, we carried out single-particle cryo-EM analysis on NMDARs in complex with Fabs, which were purified from mAbs treated with papain digestion (Fig. 1a,e,f). Fab5F6 or Fab2G7 was incubated with detergent-solubilized human GluN1–GluN2A NMDARs at a molar ratio of 3:1 or 5:1, in the presence of agonists Gly and Glu at the saturating concentration of 1 mM each. The mixed proteins were biochemically characterized by FSEC and Coomassie blue staining (Extended Data Fig. 1c,d), then loaded on the grids and plunge-frozen for cryo-EM data collection.
Two-dimensional (2D) class-average images of both NMDAR–Fab5F6 or NMDAR–Fab2G7 datasets showed that Fabs bound to the N-terminal domain (NTD) of the receptor, distant from the transmembrane domain (TMD) surrounded by detergent (Fig. 2c,d). Through the process of three-dimensional (3D) classification, we noticed that there were subclasses of NMDAR–Fab5F6 and NMDAR–Fab2G7 complexes with either one Fab or two Fabs bound (Extended Data Fig. 2). These different classes were then independently refined without symmetry applied, yielding four EM maps of NMDAR–Fab complexes at a resolution ranging from 3.8 Å to 4.5 Å (Table 1), estimated by Fourier shell correlation (FSC) curves (Extended Data Fig. 2).
In the NMDAR–Fab5F6 complex, Fab5F6 was found to interact with the side of the NTD, extending the width of NMDAR alone (126 Å) to 218 Å (Fig. 2e). In the NMDAR–Fab2G7 complex, Fab2G7 bound to the top of the NTD with an orientation parallel to the center z axis and increased the height of NMDAR alone (161 Å) to 202 Å (Fig. 2f). Notably, both epitopes were located at the exposed R1 lobe of GluN1-NTD (Fig. 2g,h), which is consistent with the finding that most anti-NMDAR antibodies bind to the NTD of GluN1 subunits4,23.
The NTD of GluN1 forms a bilobate clamshell-liked structure with R1 and R2 lobes tethered by hinge linkers31 (Fig. 2g,h) and displays conformation mobility responsible for the channel function32. We, thus, investigated whether Fab binding could induce a conformational change in individual NTDs or ligand-binding domains (LBDs) by comparing the conformation of current structures to the agonist-bound GluN1–GluN2A receptor structure resolved previously33. We found that the center-of-mass (COM) distances of R1–R2 lobes of the NTD or D1–D2 lobes of the LBD remained unchanged between the Fab-bound and Fab-free GluN1–GluN2A heterodimer (Extended Data Fig. 3). These data implied that Fab binding had a minimal effect on the conformational change of NTDs and LBDs.
Characterization of mAb binding epitopes on GluN1-NTD
In the NMDAR–Fab5F6 complex, the R1 lobe of GluN1-NTD intensively interacted with Fab5F6 through its complementarity-determining regions (CDRs) in variable light chain (Vl) and heavy chain (Vh). In detail, GluN1Arg36 on the α1 helix formed a hydrogen bond with Ser28 and a salt bridge with Asp92 in the Vl of Fab5F6. GluN1Thr35 and GluN1Lys37 formed hydrogen-bonding interactions with Asp92 and Phe91 on CDR3, respectively. GluN1Asp100 on the β3–α3 loop formed a hydrogen bond with Arg31 on CDR1. Moreover, the finger-like CDR3 loop in VH inserted into the cavity underneath the α1 helix of the R1 lobe, with Tyr105, Asp106 and Tyr108 forming hydrogen bonds and hydrophobic contacts with GluN1Thr122 and GluN1His38 (Fig. 3a–c). To confirm these binding interactions, we performed immunostaining on live HEK293T cells expressing GluN1–GluN2A NMDARs with site-directed mutagenesis on the GluN1 subunit. The result showed that charge-reversed GluN1R36E and polarity-reversed GluN1K37A and GluN1H38A substitutions blocked Fab5F6 binding to NMDARs. Additionally, receptors with GluN1T35A, GluN1D100K and GluN1T122V substitutions significantly decreased the staining signal by Fab5F6 (Fig. 3d,e).
a, Cartoon representation of the interface between GluN1-NTD and Fab5F6. Dashed boxes highlight zoomed-in views with the EM map sharpened with a B factor of −90.0 Å2 and shown with a contour level of 0.022. Key residues on GluN1-NTD and the Vh and Vl of Fab5F6 are highlighted in blue, purple and gray, respectively. Dashed lines indicate hydrogen bonds and salt-bridge interactions. b, Schematic diagram of the NMDAR–Fab5F6 binding interfaces plotted by LigPlot+. Green dashed lines represent polar interactions. Red dashed lines indicate salt bridges. Circular sectors represent hydrophobic interactions. c, CDR sequence of heavy chain and light chain of Fab5F6. Residues involved in the interaction are colored. d, Representative images from immunostaining on HEK293T cells expressing WT and GluN1 mutant NMDARs (in green) stained by mAb5F6 (in red) and DAPI (in blue). Scale bar, 15 μm. e, Bar graphs plotted with the fluorescence ratio by stained antibody of mAb5F6 versus NMDARGFP intensity. The fluorescence intensity ratio of WT NMDARGFP was normalized to 1. n = 10–21 independent images. Significance was calculated by one-way ANOVA. Data are presented as the mean ± s.e.m. of three different replicates.
In the NMDAR–Fab2G7 complex, the α1–β2 region was mainly involved in the contact with the CDR3 in Vl and CDR2 in Vh of Fab2G7. In particular, GluN1Lys51 formed a hydrogen-bonding interaction with Tyr94 on Vl. GluN1Arg52 formed salt-bridge contacts with Asp59, while GluN1Gln48 formed a hydrogen bond with nearby Thr55 on Vh (Fig. 4a–c). Immunostaining analyses further confirmed that GluN1Q48A and GluN1K51A substitutions abolished the staining, while the charge-removed GluN1R52A substitution significantly decreased the staining by Fab2G7 (Fig. 4d,e). Altogether, these functional data confirmed that Fab5F6 and Fab2G7 bound to distinct epitopes dominantly located at two edges of the α1 helix, which is less conserved between GluN1 and four GluN2 subunits (Extended Data Fig. 4).
a, Cartoon representation of the interface between GluN1-NTD and Fab2G7. Dashed boxes highlight zoomed-in views with the EM map sharpened with a B factor of −80.0 Å2 and shown with a contour level of 0.022. Key residues on GluN1-NTD and the Vh and Vl of Fab2G7 are highlighted in blue, red and green, respectively. Dashed lines indicate hydrogen bonds and salt-bridge interactions. b, Schematic diagram of the NMDAR–Fab2G7 binding interfaces plotted by LigPlot+. Green dashed lines represent polar interactions. Red dashed lines indicate salt bridges. Circular sectors represent hydrophobic interactions. c, CDR sequence of heavy chain and light chain of Fab2G7. Residues involved in the interaction are colored. d, Representative images from immunostaining on HEK293T cells expressing WT and GluN1 mutant NMDARs (in green) stained by mAb2G7 (in red) and DAPI (in blue). Scale bar, 15 μm. e, Bar graphs plotted with the fluorescence ratio by stained antibody of mAb2G7 versus NMDARGFP intensity. The fluorescence intensity ratio of WT NMDARGFP was normalized to 1. n = 9–35 independent images. Significance was calculated by one-way ANOVA. Data are presented as the mean ± s.e.m. of three different replicates.
Auto-mAb downregulates NMDAR density and function
We next sought to determine whether patient-derived auto-mAbs affect neuronal surface NMDAR morphology and NMDAR-mediated currents. We first evaluated the density of surface NMDAR clusters on cultured hippocampal neurons treated by patient-derived mAb incubation at 0.5-h and 24-h timescales. We found that a 24-h incubation with mAb5F6 and mAb2G7 (at 60 μg ml−1 concentration) resulted in approximately 35% and 42% reduced density of immunostained surface NMDARs, respectively, as compared to that found after 0.5-h incubation (Fig. 5a,b). We also investigated the postsynaptic density protein 95 (PSD95) level and found no significant change between the 0.5-h and 24-h incubations of patient-derived auto-mAbs (Fig. 5a,c).
a, Representative images of hippocampal neurons showing the density of NMDAR clusters (in green), PSD95 (in red) and DAPI (in blue) after 0.5-h (top) and 24-h (bottom) incubation of human mAb5F6, mAb2G7 and mAb4F11 (at 60 μg ml−1 concentration). White boxes highlight the representative dendrites for fluorescence intensity in an area of 25 × 10 μm2. Scale bar, 15 μm. b,c, Quantification of the density of NMDAR cluster (b) and PSD95 (c) per area (25 × 10 μm2) of dendrite after 0.5-h and 24-h incubation of human mAb5F6, mAb2G7 and mouse mAb4F11. Data are shown as violin plots (n = 25–30 cells and 3–5 dendrites per condition). Significance was calculated by a two-tailed Student’s t-test. d,e, Representative traces (d) and current amplitude quantification (e) of whole-cell responses to 300 μM NMDA and 100 μM Gly puffed onto cultured hippocampal neurons. Significance was calculated using a two-tailed paired t-test. Data are presented as the mean ± s.e.m. of three different replicates. f, Representative patch-clamp recording traces on HEK293 cells expressing GluN1–GluN2A NMDARs, puffed with 300 μM NMDA and 100 μM Gly for 5 ms. The recording traces before (black) and 2 min (blue) or 5 min (pink) after incubation of PBS, human mAb5F6, mAb2G7 and mouse mAb4F11 are shown. g,h, Quantification of current amplitude (g) and decay τweighted (h) on HEK293 cells, before and 2 or 5 min after incubation with PBS (n = 10), human mAb5F6 (n = 8), mAb2G7 (n = 8) and mouse mAb4F11 (n = 8). Significance was calculated using a two-tailed paired t-test between the before and 5 min after groups.
For a parallel comparison, we generated a mouse-derived mAb also targeting the GluN1-NTD region. Briefly, BALB/c mice were actively immunized with GluN1–GluN2A NMDAR proteins and mAbs were obtained using the hybridoma technique (Extended Data Fig. 5a,b). Among them, mAb4F11 displayed a strong immunocytochemistry reactivity to NMDARs with a binding affinity of 16 nM verified by SPR (Extended Data Fig. 5c,d). Cryo-EM structural analysis and molecular dynamics (MD) simulation of the NMDAR–Fab4F11 complex revealed that the mouse-derived mAb4F11 bound to both R1 and R2 lobes of GluN1-NTD (Extended Data Fig. 5g–i). Unliked the patient-derived mAbs, 0.5-h and 24-h incubation of mouse-derived mAb4F11 on cultured hippocampal neurons exhibited no significant change in surface NMDAR density (Fig. 5a,b). These results are in line with previous morphological observations, suggesting that mAbs derived from persons with AE downregulated the surface NMDAR distribution, with a molecular mechanism presumably differing from the mouse-derived mAb.
We next preformed whole-cell patch-clamp recording of NMDA–Gly-evoked NMDAR currents on cultured hippocampal neurons after 24-h incubation of patient-derived mAb5F6 and mAb2G7, mouse-derived mAb4F11 and PBS. The groups treated with human mAb5F6 (1,131.3 ± 90.5 pA) and mAb2G7 (1,115.2 ± 158.4 pA) exhibited a significant reduction in NMDAR-mediated currents compared to the groups incubated with mouse-derived mAb4F11 (1,803.4 ± 179.2 pA) and control PBS (1,688.5 ± 208.3 pA) (Fig. 5d,e). These results suggested that mAbs derived from persons with AE downregulated NMDAR-mediated synaptic transmission. Altogether, we demonstrated that our individual mAbs could directly induce chronic alteration of NMDAR clustering and function, a phenotype similar to that induced by previously reported polyclonal or monoclonal autoantibodies from patient serum and CSF16,17,18,19,20,24,34.
To investigate whether auto-mAb binding induces acute activity and kinetic changes on NMDARs, we performed whole-cell patch-clamp recordings on HEK293 cells expressing full-length recombinant GluN1–GluN2A receptors and compared the current amplitude and deactivation time constant before and after incubation with 60 μg ml−1 mAbs for 2 or 5 min. Our data revealed that neither human-derived nor mouse-derived mAbs significantly altered the amplitude and kinetics of NMDA–Gly-evoked tonic NMDAR currents (Fig. 5f–h). Altogether, these results implied that long-term incubation of human auto-mAbs induced the downregulation of synaptic NMDARs, without causing a tonic effect on channel activity, indicating that receptor internalization represents a major pathogenic mechanism of AE.
Stoichiometry of NMDAR–antibody interaction
To further elucidate the conformation of the NMDAR–mAb complex in a physiological state, synchrotron SAXS was conducted to analyze the binding pattern between NMDARs and human mAbs. SAXS averages signals from diverse conformations in solution, yielding a low-resolution (10–20 Å) overview of macromolecule shapes with multiple consistent ensembles. We prepared the complex of GluN1–GluN2A NMDARs mixed with mAb5F6 or mAb2G7 (NMDARs versus mAb at a molar ratio of 1:5), using free NMDARs, mAb5F6 and mAb2G7 as controls. Ab initio model building from SAXS data yielded two low-resolution envelopes with heart-shaped and bouquet conformations for NMDAR–mAb5F6 and NMDAR–mAb2G7 complexes, respectively (Fig. 6a). The maximum particle dimension (Dmax) was further calculated by indirect Fourier transformation. The Dmax of free NMDARs, mAb5F6 and mAb2G7 was 204 Å, 158 Å and 160 Å, while that of NMDAR–mAb5F6 and NMDAR–mAb2G7 complexes was 317 Å and 254 Å, respectively. This information indicated that NMDAR and mAbs formed a larger complex with distinct conformations in the physiologically relevant liquid condition.
a, SAXS profiles of NMDAR–mAb5F6, free mAb5F6, NMDAR–mAb2G7, free mAb2G7 and free NMDAR (from left to right). Dot plots in gray are experimental scattering data, while colored lines are the theoretical scattering intensities calculated from atomic structures. The inserted figures illustrate the corresponding ab initio SAXS models. b,c, Pie chart of protein population in the mixed liquid condition for NMDAR–mAb5F6 (b) and NMDAR–mAb2G7 (c). Structures of NMDAR–mAb with different stoichiometry (1:2 or 2:1), free NMDAR and free mAb are indicated according to the related populations. d, Schematic model for NMDAR-related AE pathogenesis. Left, the physiological function of NMDARs for the calcium influx at the synapse. With the accumulation of pathological auto-mAbs in the synaptic cleft, NMDARs on the plasma membrane are crosslinked and subsequently internalized into the endocytic vesicles, resulting in the downregulation of synaptic function.
To analyze the stoichiometry of the NMDAR–mAb complex, we used the OLIGOMER algorithm implemented in ATSAS 3.2.1 with the input of atomic structures comprising free NMDAR, free mAb and the NMDAR–mAb complex at different ratios (1:1, 1:2 and 2:1). The model of NMDAR–mAb was built according to the superimposition of Fab modules in the structure of the NMDAR–Fab complex. SAXS analysis of the NMDAR–mAb5F6 complex (χ2 = 84.04) showed that 82.0% of the particle population comprised one NMDAR binding with two mAb5F6, while 18.0% of the population consisted of two NMDARs clustered with one mAb5F6 (Fig. 6b). In the NMDAR–mAb2G7 complex, the result (χ2 = 4.06) showed that the particle population was composed of four major components with 18.8% free mAb2G7, 24.3% free NMDAR, 11.6% one mAb2G7 gluing together two NMDARs and 45.3% two mAb2G7 sticking to the two epitopes on one NMDAR (Fig. 6c). These data suggested that patient-derived mAbs clustered NMDARs with different interaction patterns in solution and presumably induced receptor clustering on the neuronal surface (Fig. 6d).
Discussion
To date, more than 20 neuronal membrane proteins have been identified as targets for the pathological autoantibody in persons with AE35. Different autoantibodies mediate multifaceted pathogenic mechanisms including receptor internalization17,36, disruption of protein–protein interaction37, alteration of channel activity38,39, complement activation40 and cellular cytotoxicity41,42. The binding of these autoantibodies to various ion channels including excitatory glutamate receptors43,44 and inhibitory GABAA45 and glycine39 receptors altered the synaptic neurotransmission and excitatory–inhibitory balance. Recently, the structural elucidation of the GABAA receptor–Fab complex indicated that mAbs derived from persons with AE bound to the orthostatic and allosteric sites on the GABAA receptor and directly inhibited channel activity46. Our results showed that mAbs interacted with the peripheral GluN1-NTDs of NMDAR, which varies from the LBD or ion channel gate, without acutely changing the channel activity as seen with anti-GABAA receptor antibodies.
Structural identification of 3D epitopes helps us better understand the diverse pathological mechanisms in the subclasses of persons with AE. It should be noted that both Fab5F6 and Fab2G7 contacted the surface-exposed area on the R1 lobe of GluN1-NTD, which is calculated to have more exposed surface (34,995.7 Å2) than GluN2A-NTD (33,883.6 Å2) in the GluN1–GluN2A receptor structure33. Both epitopes discovered in this study are distinct from the previously predicted binding pocket buried in the back hinge of GluN1-NTD23, a region that we speculate is presumably less accessible for the direct binding of patient auto-mAbs.
NMDARs form a heterotetrameric assembly with a large extracellular region comprising eight clamshell modules of NTDs and LBDs, which provide abundant epitopes for autoantibody binding. Although only two patient-derived mAbs were investigated here, our epitope discoveries are consistent with previous studies demonstrating that GluN1-NTD serves as the primary target for most anti-NMDAR AE auto-mAbs22,23,24. Recent studies also showed that auto-mAbs targeting GluN2A or GluN2B subunits47,48 have been found in persons with NMDAR encephalitis, schizophrenia49 and systemic lupus erythematosus (SLE)50,51. Auto-mAbs discovered in persons with SLE were identified to bind the Asp-Trp-Glu-Tyr-Ser motif on GluN2A-NTD and act as positive allosteric modulators51, revealing a different pathogenesis mechanism compared to anti-GluN1 auto-mAbs from persons with AE. Cloning more patient-derived mAbs from autoimmune-related disorders would increase the knowledge of autoantibody diversity and clinical heterogeneity.
Methods
Plasmid construction
For protein expression, C-terminal domain (CTD)-truncated human GluN1 (residues Met1–Gln847; GenBank: NP_015566) and GluN2A (residues Met1–Phe841, GenBank: NP_000824) were cloned into the pEG-BacMam vector. A 3C protease cleavage site (LEVLFQGP), enhanced GFP, His affinity tag (HHHHHH) and Strep II affinity tag (WSHPQFEK) were placed at the C terminus of GluN1 and GluN2A subunits, respectively. A tail fragment of the GluA2Tyr837–Lys847 AMPA receptor was inserted into GluN2A to improve the expression level and thermostability25,52.
For immunocytochemistry staining, CTD-truncated human GluN1 (GFP-fused) and GluN2 (2A–2D) or GluN3 (3A and 3B) subtype were cloned into the pEG-BacMam vector. Full-length human GluN1 and GluN2A (GFP-fused) were cloned into PCI-neo-based vectors53. Site-directed mutagenesis was introduced on PCI-neo-based vectors using Takara KOD-FX DNA polymerase.
Immunocytochemistry staining
HEK293T cells were transiently transfected with PCI-neo-based GuN1–GluN2A plasmid. Then, 48 h later, the cells on coverslips were gently washed with PBS and incubated with mAbs (60 μg ml−1) for 30 min. The cells were washed with PBS and fixed with 4% paraformaldehyde for 5 min and subsequently incubated with secondary antibodies (Invitrogen, A-21089) at a dilution of 1:1,000 for 20 min and then washed. The coverslips were mounted with ProLong Gold Antifade reagent with DAPI (Thermo Fisher) and imaged with an Olympus FV3000 confocal microscope. Images were quantified using Image J 1.54f software.
For brain slice immunocytochemistry staining, C57BL/6 mouse brain was extracted without perfusion and fixation. The brain was chopped along the sagittal plane, fixed in 4% paraformaldehyde for 1 h at room temperature (RT), cryoprotected with 40% sucrose for 48 h at 4°C, embed in optimal cutting temperature (OCT) compound, snap-frozen in isopentane chilled with liquid nitrogen and cut into 15-μm slices. The slices were rehydrated with PBS for 5 min, blocked with PBS containing 10% normal goat serum for 30 min at RT and subsequently incubated with either PBS (as vehicle) or mAbs at a series of concentrations of 6.7, 20, 60 and 180 μg ml−1 overnight at 4 °C. Thereafter, Alexa Fluor 488-conjugated secondary antibodies (1:1,000; Molecular probes) were used to label mAb for 1 h at RT and cell nuclei were visualized with DAPI. Images were acquired by a fluorescence microscope (Olympus BX51). The mean fluorescence of three randomly selected hippocampal CA1 regions per concentration was quantified. Relative fluorescence intensity was calculated by dividing the fluorescence of the mAb at each concentration by the background fluorescence of the PBS group.
NMDAR expression and purification
For GluN1–GluN2A NMDAR expression and purification, recombinant baculovirus was produced according to the standard protocol of the Bac-to-Bac TOPO expression system (Invitrogen, A11339) using sf9 insect cells. Suspended HEK293S GnTI– cells were infected using P2 virus at a density of 3.5 × 106 cells per ml. Then, 12 h after infection, 10 mM sodium butyrate was added to the culture medium to improve the expression level of NMDARs and 10 µM MK-801 was added to prevent the cell cytotoxicity caused by NMDAR activation. Cells were collected at 48 h and sonicated in TBS buffer (150 mM NaCl and 20 mM Tris, pH 8.0) supplemented with protease inhibitor cocktail (1 mM PMSF, 0.8 µM aprotinin, 2 mM pepstatin A and 2 µg ml−1 leupeptin). After centrifugation at 8,000g for 20 min, the supernatant was collected by ultracentrifugation at 165,000g for 1 h. To remove MK-801, the membrane sample was dialyzed in TBS buffer supplied with 1 mM Gly and Glu for 3–4 days.
The membrane was homogenized and solubilized in TBS containing 1% l-MNG, 2 mM CHS and protease inhibitor cocktail for 1.5 h. After ultracentrifugation at 200,000g for 1.5 h, the supernatant was collected and incubated with Strep-Tactin beads for 1.5 h. The protein was eluted with TBS buffer supplemented with 1% l-MNG, 2 mM CHS and 5 mM d-desthiobiotin and digested with 3C protease (1:20 molar ratio) overnight to remove the GFP tag. The receptors were further purified by SEC using a Superose Increase 6 10/300 GL column in buffer containing 150 mM NaCl, 0.1% digitonin, 5 µM CHS, 0.1 mM CHAPSO, 50 µM EDTA, 1 mM Gly–Glu and 20 mM HEPES (pH 8.0). The peak fraction was concentrated to 4.5 mg ml−1 for cryo-EM experiments.
Nanodisc reconstruction
MSPΔH5 was purified according to a previously published protocol27. Purified GluN1–GluN2A NMDAR tagged with GFP was diluted to ~1 mg ml−1 and mixed with purified MSPΔH5 and soybean lipid extract (Avanti Polar Lipids) at a 1:240:40 molar ratio. The mixture was incubated for 40 min at 4 °C. Bio-Beads (60 mg ml−1; Bio-Rad) were washed with methanol, double-distilled H2O and HBS (150 mM NaCl, 20 mM HEPES and 50 μM EDTA, pH 8.0) before being added to the protein–MSP–lipid mixture. The Bio-Beads were replaced with a new portion after incubating with protein–MSP–lipid mixture for 2 h and then the mixture was incubated at 4 °C overnight. The nanodisc reconstructed protein mixture was further purified by SEC on a Superose 6 Increase 10/300 GL column with HBS plus 1 mM Gly–Glu.
NMDARGFP-labeled FACS
PBMCs were isolated from peripheral venous blood by Ficoll-Paque (GE Healthcare) density gradient centrifugation according to the manufacturer’s instructions. B cells were roughly enriched by non-B cell depletion using a B cell isolation kit (Miltenyi, 130-091-151). Purified B cells were stained for 45 min with Zombie yellow (1:100; BioLegend) and fluorescently labeled antibodies human anti-CD27–PE and human anti-IgD–APC (1:25; BioLegend) with or without NMDARGFP in nanodiscs (60 μg ml−1) in the experimental group and control group at 4 °C. FlowJo 10.8.1 was used for single-cell sorting data analysis. Single cells were sorted using MoFlo Astrios EQ (Beckman) and collected into 96-well PCR plates containing 2.3 μl of TE buffer per well supplemented with protein K (20 mg ml−1) at a ratio of 15:1 (v/v).
Auto-mAb sequence cloning
cDNA was synthesized in a total volume of 10 μl per well in the original 96-well sorting plate. Total RNA from single cells was first incubated at 56 °C for 1 h in an Eppendorf cycler with the lid set at 66 °C. Then, 1 μl of oligdT primer (10 µM) and 1 μl of dNTPs (10 mM) were added and incubated for 10 min at 95 °C. The reverse transcription reaction was initiated with 100 U of SuperScript II reverse transcriptase (Invitrogen), 10 U of RNAse inhibitor (Promega), 5 mM DTT, 8% PEG8000, 6 mM MgCl2 and 1 μM TSO, with the reaction proceeding in a cycle at 42 °C for 90 min, 50 °C for 2 min, 42 °C for 2 min and 70 °C for 15 min. cDNA was then obtained.
To amplify the variable region gene of immunoglobulin heavy chain (IgH) and putative immunoglobulin light chains (IgK or Igλ), PCR was performed with 40 μl per well containing 0.2 µM primer mix, 0.3 mM each dNTP and 2 U of HotStar Taq DNA polymerase (Qiagen). The PCR products were sequenced with reverse primers for IgH, IgK and Igλ24. Sequences were compared to GenBank to identify the variable region with the highest sequence homology. The variable region was cloned into the pF1 vector, which contained the corresponding constant region29.
Auto-mAb and Fab production
For recombinant expression of mAb5F6 and mAb2G7, transfection was carried out on HEK293 cells at a density of 2 × 106 cells per ml using PEI. The culture medium was collected 5 days after transfection. Recombinant antibodies were purified by rProtein G beads (SA020025, Smart-Lifesciences).
For Fab preparation, the purified antibodies were diluted to 1 mg ml−1 with PBS before adding 1 mM EDTA and 10 mM l-Cys. Papain was added at a ratio of 20:1 (w/w) and digested at 37 °C for 3 h. The mixture was incubated with rProtein A beads (SA012025, Smart-Lifesciences) for 1 h at 4 °C with rotation and then the flowthrough was collected. Fab was purified by removing papain using a 10-kDa molecular weight cutoff filter (Millipore). mAbs and Fabs were qualified by SDS–PAGE and FSEC detection.
Mouse-derived mAb generated using hybridoma technology
BALB/c mice were immunized with purified GluN1–GluN2A NMDAR proteins. The purified NMDAR proteins were mixed 1:1 with Freund’s complete adjuvant. Mice were subcutaneously injected with 100 μg of emulsified proteins per mouse and immunized every 2 weeks. After the third injection, spleen cells were collected and fused with Sp2/0 cells to generate hybridomas using PEG1450. The positive clones were screened by immunocytochemistry staining. Ascites was prepared by intraperitoneal injection of hybridomas into mice and mAbs were purified from ascites by rProtein G beads. The mAbs were digested by papain; next, the Fc fragment was removed by rProtein A beads and applied to an anion-exchange column with running buffer (20 mM Tris-HCl, pH 10.8). The purified Fab fragment was obtained with elution buffer (1 M NaCl and 20 mM Tris-HCl, pH 10.8).
SPR
The purified GluN1–GluN2A NMDAR was diluted with 10 mM sodium acetate (pH 5.5) to a concentration of 2 µg ml−1. Then, NMDAR was injected at a constant flow rate of 10 µl min−1 for 900 s and immobilized to CM5 sensor chips (GE Healthcare). The SPR measurements were performed using a Biacore 8K instrument (GE Healthcare) in running buffer containing 10 mM HEPES pH 8.0, 150 mM NaCl, 1 mM Gly–Glu and 50 μM EDTA at a flow rate of 30 µl min−1. To determine the binding affinities, antibodies were analyzed using concentration–response experiments. Antibodies (2 µM) were serially diluted twofold in running buffer. The resulting increasing concentrations of the mAbs were injected over NMDAR in the running buffer for 240 s. The surface was washed with running buffer for 270 s between each binding cycle. The experimental data were analyzed by fitting both a 1:1 steady-state affinity model and a kinetic binding model to determine the KD values of various mAbs in Biacore 8K evaluation software.
Cryo-EM sample preparation and data acquisition
For the NMDAR–Fab5F6 complex, Fab5F6 was added at a molar ratio of 1:3 to NMDAR. For the NMDAR–Fab2G7 complex, Fab2G7 was added at a molar ratio of 1:5 to NMDAR. For the NMDAR–Fab4F11 complex, Fab4F11 was added at a molar ratio of 1:5 to NMDAR. All complexes were incubated for 1 h before grid preparation. Samples (2.5–3 µl) were applied to glow-discharged Quantifoil 1.2/1.3 300 mesh grids and then blotted and plunge-frozen in liquid ethane using an FEI Vitrobot with 100% humidity at 8 °C.
The grids were initially screened by an FEI Talos 120C microscope. For the NMDAR–Fab4F11 complex, images were manually collected using a CETA electron detector. For the NMDAR–Fab5F6 and NMDAR–Fab2G7 complexes, the data were collected on 300-kV FEI Titan Krios cryo-EM. Images were collected using a Gatan K3 direct electron detector in super-resolution mode with an energy filter (Gatan) at ×64,000 magnification (unbinned pixel size of 0.54 Å per pixel). The image was collected using EPU 3.2 with a defocus range of −1–−2 μm and a total dose of ~60e− per Å2 for NMDAR–Fab5F6 and ~70e− per Å2 for NMDAR–Fab2G7.
Cryo-EM data processing
The raw dataset was collected using super-resolution mode and the 2 × 2 Fourier space binned raw data stack was motion-corrected and dose-weighted by MotionCor2 1.4.0 (ref. 54). The contrast transfer function (CTF) was accurately estimated by Gctf_v1.06 (ref. 55). Approximately 80 micrographs were initially picked out for reference-free particle picking using Laplacian-of-Gaussian-based autopicking in RELION 3.1.1 (ref. 56).
Approximately 7,000 particles were chosen for reference-free 2D classification. The selected 2D class averages were used as templates for autopicking in RELION. The particles were then taken for multiple rounds of 2D classification in RELION 3.1. The initial model was generated de novo in RELION. The 3D classification was followed with C1 symmetry to prevent reconstruction bias. After multiple rounds of 3D classification, 3D classes were inspected using UCSF Chimera57 and particles from 3D classes with one-Fab or two-Fab conformations were further selected for 3D refinement with C1 symmetry. CTF refinement and Bayesian polishing were applied to the reconstructed 3D classes and followed by another round of 3D refinement. All maps were postprocessed using a solvent mask with B factors that were automatically estimated. Resolutions were estimated with the gold-standard FSC criterion of 0.143. Local resolution estimation was conducted in RELION 3.1.
Model building and structural analysis
The structure of human GluN1–GluN2A NMDAR (Protein Data Bank (PDB) 7EOQ) was divided into individual lobes and docked into cryo-EM density maps using UCSF Chimera57. The models of human Fab5F6, Fab2G7 and mouse Fab4F11 were generated de novo by uni-Fold (https://github.com/dptech-corp/Uni-Fold) through the Hermite platform (https://hermite.dp.tech). The missing loops were manually built using Coot 0.9.7 (ref. 58). For areas with poorly resolved density, the residues were replaced by alanines. All structural refinements were carried out by real-space refinement in Phenix 1.20-4459 (ref. 59) with secondary-structure and geometry restraints.
For the structural analysis, the COMs were generated by COM scripts in PyMOL. The COM of each lobe was calculated by connecting the residues of Leu34–Phe147 and Tyr281–His358 segments for R1, Gly148–Ser280 and Pro359–Asp403 segments for R2, Asn404–Val539 and Thr758–Leu796 segments for D1 and Glu530–Ser538 and Asp660–Ala757 segments for D2 in the GluN2A subunit and the residues of Lys25–Pro144 and Gly274–Phe348 segments for R1, Tyr144–Asn273 and Ala349–Gln393 segments for R2, Met394–Phe533 and Ser756–Gln796 segments for D1 and Lys534–Glu545 and Arg663–Arg755 segments for D2 in the GluN1 subunit. The Cα atoms of all residues were used for the structural comparisons. The angles between space vectors connecting the COMs of R1 and R2 or D1 and D2 were calculated using a Python script. All figures were generated by Pymol 2.5.2 (Schrödinger), Chimera 1.16 (ref. 57) and Chimera X 1.6.1 (ref. 60).
Immunofluorescence staining and confocal imaging
Hippocampal cultured neurons were prepared from Sprague–Dawley (SD) rats on embryonic day 17. The dissected hippocampus was sectioned into fragments and digested with 0.25% trypsin. After resuspension, the cells were plated at a density of 4 × 105 cells per ml on poly(d-Lys)-coated glass slides and cultured in neurobasal medium with 2% B27. Neurons were cultured for 12–14 days in vitro and used for immunocytochemistry staining.
To stain surface NMDAR clusters, hippocampal neurons were incubated with 60 μg ml−1 human mAbs for 30 min or 24 h. Neurons were washed and fixed in 4% paraformaldehyde with 4% sucrose in PBS (pH 7.4) for 5 min and blocked in PBS with 10% FBS for 1 h at RT. The neurons were then incubated with antihuman or antimouse secondary antibodies for 1.5 h at RT to stain the surface NMDAR (Jackson ImmunoResearch, Alexa Fluor 488 antihuman and antimouse IgG; 1:300). Next, the neurons were washed by PBS and blocked in PBS with 10% FBS and 0.3% Triton X-100 for 1 h at RT. Next, the neurons were incubated with anti-PSD95 antibody (Abcam; 1:500) overnight at 4 °C. The neurons were then washed and incubated with antirabbit secondary antibody (Abcam, Alexa Fluor 594 goat antirabbit IgG; 1:500) for 1 h at RT. The coverslips were mounted with ProLong Gold Antifade reagent with DAPI (Thermo Fisher) and imaged with an Olympus FV3000 confocal microscope.
For quantification of human mAb binding on the mutant NMDAR construct, the fluorescence intensity of the wild-type (WT) NMDARGFP was normalized to 1. Significance was calculated by one-way analysis of variance (ANOVA). The density of the surface NMDAR cluster per area (25 × 10 μm2) of dendrite was quantified using ImageJ software. Data were displayed as a violin plot. Significance was calculated by a Student’s t-test.
Patch-clamp recordings
To record NMDAR-mediated currents, we conducted whole-cell patch-clamp recordings on cultured hippocampal neurons at 13–14 days in vitro after 24-h incubation with patient-derived mAbs, mouse-derived mAbs and PBS. Throughout the recording sessions, the cells were bathed in an external solution containing 138 mM NaCl, 4 mM KCl, 2 mM CaCl2, 10 mM glucose and 10 mM HEPES (pH adjusted to 7.2 with KOH; 290–310 mOsm), together with 1 μM tetrodotoxin and 100 μM picrotoxin. The recording pipettes were filled with an intracellular solution containing 120 mM CsMeSO4, 20 mM CsCl, 10 mM HEPES, 0.2 mM EGTA, 10 mM sodium phosphocreatine, 4 mM Na2-ATP and 0.4 mM Mg-ATP (pH adjusted to 7.2 with CsOH; 290–310 mOsm).
The recordings were conducted at RT with voltage-clamp mode, maintaining a holding potential of −70 mV using an Axopatch 700B amplifier (Axon Instruments). The signals were digitized through a Digidata-1550 interface (Axon Instruments) for data acquisition and analyzed using pClamp 11.2 (Axon Instruments). To evoke NMDAR-mediated currents, a glass pipette with a similar resistance (~4 MΩ) was filled with a solution containing 300 μM NMDA and 100 μM Gly and positioned near the neurons under recording. The agonists were puffed to activate NMDARs for a duration of 2 s. The recording electrode resistance in the bath solution was maintained between 3 and 5 MΩ. Recordings with series resistance greater than 25 MΩ or changes exceeding 20% were discarded. All statistical analyses were performed using Prism (GraphPad Prism 9.0 Software). Data were presented as the mean ± s.e.m. Multiple groups were compared by one-way ANOVA. Statistical significance was determined.
Patch-clamp recordings were performed on HEK293 cells transfected with GluN1, GluN2A and GFP plasmids at a ratio of 20:20:1. The recording electrode resistance in the bath solution was maintained between 3 and 5 MΩ. Coagonists 300 μM NMDA and 100 μM Gly were perfused to lifted cells by the rapid and automated solution change system (RSC-200, Bio-Logic) for 5 ms to activate NMDAR-mediated currents. Deactivation time constants were calculated using dual-exponential fits and two time constants (τfast and τslow) were obtained. Weighted time constants (τweighted) were calculated according to the equation τweighted = (τfast Afast + τslow Aslow)/(Afast + Aslow), where Afast and Aslow are the fitted amplitudes of the fast and slow components. Clampfit 10.6 was used for electrophysiological recordings and data analysis. The values for amplitude and τweighted were compared using a paired t-test.
SAXS
SAXS experiments were performed at beamline BL19U2 of the National Facility for Protein Science Shanghai (NFPSS) at Shanghai Synchrotron Radiation Facility (SSRF). The wavelength, λ, of X-ray radiation was set as 1.033 Å. Scattered X-ray intensities were collected using a Pilatus 1M detector (DECTRIS). The sample-to-detector distance was set such that the detecting range of momentum transfer (q = 4π sinθ/λ, where 2θ is the scattering angle) of SAXS experiments was 0.013–0.25 Å−1. To reduce the radiation damage, a flow cell made of a cylindrical quartz capillary with a diameter of 1.5 mm and a wall of 10 µm was used. SAXS data were collected as 20× 1-s exposures and scattering profiles for the 20 passes were compared at 10 °C using 60 μl of sample in SEC buffer (50 mM NaCl, 0.1% digitonin, 5 µM CHS, 0.1 mM CHAPSO, 50 µM EDTA,1 mM Gly–Glu and 20 mM HEPES, pH 8.0). For individual NMDAR, mAb5F6 and mAb2G7, the concentration was set to 0.6 mg ml−1 for SAXS experiments. For the NMDAR–mAb5F6 and NMDAR–mAb2G7 complexes, the concentration of NMDAR was 0.6 mg ml−1. mAbs were added at a molar ratio of 1:5. The 2D scattering images were viewed using ALBULA 3.2.0 and were further converted to 1D SAXS curves through azimuthally averaging after solid angle correction. The images were then normalized with the intensity of the transmitted X-ray beam using the software package BioXTAS RAW 2.2.1 (ref. 61). Scattering data were further analyzed using ATSAS 3.2.1. In detail, background scattering was subtracted using PRIMUS62. The ab initio shapes were determined using DAMMIF63 with 15 DAMMIF runs for each experimental group and DAMAVER64 was used to analyze the normalized spatial discrepancy (NSD) among the 15 models. The lowest-NSD model was used as the representative. The theoretical scattering intensities of atomic structure were calculated using CRYSOL65.
MD simulation
The model of mouse Fab4F11 was generated de novo by Uni-Fold (https://github.com/dptech-corp/Uni-Fold) through the Hermite platform (https://hermite.dp.tech). The NTD–Fab4F11 model was obtained by rigid-body fitting into the low-resolution cryo-EM density and was solvated in 0.15 M NaCl solution. The structurally sound model with the lowest MODELLER objective function was chosen for subsequent simulations. GROMACS66 version 2021.3 with the CHARMM36 force field67 and TIP3P water model68 was used to perform the simulations. The long-range electrostatics (< nm) were modeled using the particle mesh Ewald method69. The system was equilibrated under an NVT ensemble for 500 ps with an integration time step of 1 fs, followed by a 50-ns NPT simulation with an integration time step of 2 fs. The Nose–Hoover thermostat70 and Parrinello–Rahman barostat71 were used for temperature and pressure coupling, respectively. Covalent bonds were constrained to their equilibrium length by the LINCS algorithm for all simulations. We referred to a multisampling strategy72 and conducted three rounds of separate MD simulations, with each lasting 200 ns. Residue-wise solvent-accessible surface area (SASA) was calculated using Biopython73. The NTD total SASA is the sum of the SASA of all residues in the corresponding domain.
Patient and animal study ethical approval
The study was approved by the Institutional Review Boards of Huashan Hospital affiliated with Fudan University, of Ruijin Hospital affiliated with Shanghai JiaoTong University and of the Center for Excellence in Brain Science and Intelligence Technology (CEBSIT), Chinese Academy of Sciences (CAS). Written informed consent was received from participants or their representatives.
All experiments with mice and rats were approved by the Animal Care and Use Committee of CEBSIT, CAS. Mice and rats were bred with standard chow and water and raised in rooms with controlled temperature of ~22 °C and humidity of 50% under a 12-h light–dark cycle.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
Cryo-EM density maps and corresponding coordinates were deposited to the EM Data Bank (EMDB) and PDB under the following accession codes: EMD-36335 and PDB 8JIZ for the NMDAR–Fab5F6 two-Fab-bound state, EMD-36336 and PDB 8JJ0 for the NMDAR–Fab5F6 one-Fab-bound state, EMD-36337 and PDB 8JJ1 for the NMDAR–Fab2G7 two-Fab-bound state and EMD-36338 and PDB 8JJ2 for the NMDAR–Fab2G7 one-Fab-bound state. SAXS data were deposited to the Small-Angle Scattering Biological Data Bank (SASBDB) under the following accession codes: SASDR42 for the NMDAR–mAb5F6 complex, SASDR32 for the NMDAR–mAb2G7 complex, SASDR52 for GluN1–GluN2A NMDAR, SASDR62 for mAb2G7 and SASDR72 for mAb5F6. No code was developed in this study. Source data are provided with this paper.
References
Hansen, K. B. et al. Structure, function, and pharmacology of glutamate receptor ion channels. Pharm. Rev. 73, 298–487 (2021).
Paoletti, P. Molecular basis of NMDA receptor functional diversity. Eur. J. Neurosci. 33, 1351–1365 (2011).
Talukder, I., Borker, P. & Wollmuth, L. P. Specific sites within the ligand-binding domain and ion channel linkers modulate NMDA receptor gating. J. Neurosci. 30, 11792–11804 (2010).
Dalmau, J. et al. An update on anti-NMDA receptor encephalitis for neurologists and psychiatrists: mechanisms and models. Lancet Neurol. 18, 1045–1057 (2019).
Dalmau, J. et al. Paraneoplastic anti-N-methyl-d-aspartate receptor encephalitis associated with ovarian teratoma. Ann. Neurol. 61, 25–36 (2007).
Graus, F. et al. A clinical approach to diagnosis of autoimmune encephalitis. Lancet Neurol. 15, 391–404 (2016).
Pruss, H. et al. N-methyl-d-aspartate receptor antibodies in herpes simplex encephalitis. Ann. Neurol. 72, 902–911 (2012).
Diamond, B., Huerta, P. T., Mina-Osorio, P., Kowal, C. & Volpe, B. T. Losing your nerves? Maybe it’s the antibodies. Nat. Rev. Immunol. 9, 449–456 (2009).
Planaguma, J. et al. Human N-methyl d-aspartate receptor antibodies alter memory and behaviour in mice. Brain 138, 94–109 (2015).
Wright, S. et al. Epileptogenic effects of NMDAR antibodies in a passive transfer mouse model. Brain 138, 3159–3167 (2015).
Jones, B. E. et al. Autoimmune receptor encephalitis in mice induced by active immunization with conformationally stabilized holoreceptors. Sci. Transl. Med. 11, eaaw0044 (2019).
Dalva, M. B. et al. EphB receptors interact with NMDA receptors and regulate excitatory synapse formation. Cell 103, 945–956 (2000).
Henderson, J. T. et al. The receptor tyrosine kinase EphB2 regulates NMDA-dependent synaptic function. Neuron 32, 1041–1056 (2001).
Nolt, M. J. et al. EphB controls NMDA receptor function and synaptic targeting in a subunit-specific manner. J. Neurosci. 31, 5353–5364 (2011).
Planaguma, J. et al. Ephrin-B2 prevents N-methyl-d-aspartate receptor antibody effects on memory and neuroplasticity. Ann. Neurol. 80, 388–400 (2016).
Mikasova, L. et al. Disrupted surface cross-talk between NMDA and Ephrin-B2 receptors in anti-NMDA encephalitis. Brain 135, 1606–1621 (2012).
Moscato, E. H. et al. Acute mechanisms underlying antibody effects in anti-N-methyl-d-aspartate receptor encephalitis. Ann. Neurol. 76, 108–119 (2014).
Rosch, R. E. et al. NMDA-receptor antibodies alter cortical microcircuit dynamics. Proc. Natl Acad. Sci. USA 115, E9916–E9925 (2018).
Hughes, E. G. et al. Cellular and synaptic mechanisms of anti-NMDA receptor encephalitis. J. Neurosci. 30, 5866–5875 (2010).
Ladepeche, L. et al. NMDA receptor autoantibodies in autoimmune encephalitis cause a subunit-specific nanoscale redistribution of NMDA receptors. Cell Rep. 23, 3759–3768 (2018).
Jezequel, J. et al. Dynamic disorganization of synaptic NMDA receptors triggered by autoantibodies from psychotic patients. Nat. Commun. 8, 1791 (2017).
Castillo-Gomez, E. et al. All naturally occurring autoantibodies against the NMDA receptor subunit NR1 have pathogenic potential irrespective of epitope and immunoglobulin class. Mol. Psychiatry 22, 1776–1784 (2017).
Gleichman, A. J., Spruce, L. A., Dalmau, J., Seeholzer, S. H. & Lynch, D. R. Anti-NMDA receptor encephalitis antibody binding is dependent on amino acid identity of a small region within the GluN1 amino terminal domain. J. Neurosci. 32, 11082–11094 (2012).
Kreye, J. et al. Human cerebrospinal fluid monoclonal N-methyl-d-aspartate receptor autoantibodies are sufficient for encephalitis pathogenesis. Brain 139, 2641–2652 (2016).
Wang, H. et al. Gating mechanism and a modulatory niche of human GluN1–GluN2A NMDA receptors. Neuron 109, 2443–2456 (2021).
Zhang, Y. et al. Structural basis of ketamine action on human NMDA receptors. Nature 596, 301–305 (2021).
Bibow, S. et al. Solution structure of discoidal high-density lipoprotein particles with a shortened apolipoprotein A-I. Nat. Struct. Mol. Biol. 24, 187–193 (2017).
Kreye, J. et al. A therapeutic non-self-reactive SARS-CoV-2 antibody protects from lung pathology in a COVID-19 hamster model. Cell 183, 1058–1069 (2020).
Liu, X. et al. Human immunoglobulin G hinge regulates agonistic anti-CD40 immunostimulatory and antitumour activities through biophysical flexibility. Nat. Commun. 10, 4206 (2019).
Paoletti, P., Bellone, C. & Zhou, Q. NMDA receptor subunit diversity: impact on receptor properties, synaptic plasticity and disease. Nat. Rev. Neurosci. 14, 383–400 (2013).
Farina, A. N. et al. Separation of domain contacts is required for heterotetrameric assembly of functional NMDA receptors. J. Neurosci. 31, 3565–3579 (2011).
Zhu, S., Stroebel, D., Yao, C. A., Taly, A. & Paoletti, P. Allosteric signaling and dynamics of the clamshell-like NMDA receptor GluN1 N-terminal domain. Nat. Struct. Mol. Biol. 20, 477–485 (2013).
Jalali-Yazdi, F., Chowdhury, S., Yoshioka, C. & Gouaux, E. Mechanisms for zinc and proton inhibition of the GluN1/GluN2A NMDA receptor. Cell 175, 1520–1532 (2018).
Sharma, R. et al. Monoclonal antibodies from a patient with anti-NMDA receptor encephalitis. Ann. Clin. Transl. Neurol. 5, 935–951 (2018).
Pruss, H. Autoantibodies in neurological disease. Nat. Rev. Immunol. 21, 798–813 (2021).
Peng, X. et al. Cellular plasticity induced by anti-α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor encephalitis antibodies. Ann. Neurol. 77, 381–398 (2015).
Petit-Pedrol, M. et al. Encephalitis with refractory seizures, status epilepticus, and antibodies to the GABAA receptor: a case series, characterisation of the antigen, and analysis of the effects of antibodies. Lancet Neurol. 13, 276–286 (2014).
Lancaster, E. et al. Antibodies to the GABAB receptor in limbic encephalitis with seizures: case series and characterisation of the antigen. Lancet Neurol. 9, 67–76 (2010).
Crisp, S. J. et al. Glycine receptor autoantibodies disrupt inhibitory neurotransmission. Brain 142, 3398–3410 (2019).
Bien, C. G. et al. Immunopathology of autoantibody-associated encephalitides: clues for pathogenesis. Brain 135, 1622–1638 (2012).
Brilot, F. et al. Antibodies to native myelin oligodendrocyte glycoprotein in children with inflammatory demyelinating central nervous system disease. Ann. Neurol. 66, 833–842 (2009).
Asavapanumas, N. & Verkman, A. S. Neuromyelitis optica pathology in rats following intraperitoneal injection of NMO-IgG and intracerebral needle injury. Acta Neuropathol. Commun. 2, 48 (2014).
Lai, M. et al. AMPA receptor antibodies in limbic encephalitis alter synaptic receptor location. Ann. Neurol. 65, 424–434 (2009).
Landa, J. et al. Encephalitis with autoantibodies against the glutamate kainate receptors GluK2. Ann. Neurol. 90, 101–117 (2021).
Pruss, H. & Kirmse, K. Pathogenic role of autoantibodies against inhibitory synapses. Brain Res. 1701, 146–152 (2018).
Noviello, C. M., Kreye, J., Teng, J., Pruss, H. & Hibbs, R. E. Structural mechanisms of GABAA receptor autoimmune encephalitis. Cell 185, 2469–2477 (2022).
Wollmuth, L. P., Chan, K. & Groc, L. The diverse and complex modes of action of anti-NMDA receptor autoantibodies. Neuropharmacology 194, 108624 (2021).
Dalmau, J., Geis, C. & Graus, F. Autoantibodies to synaptic receptors and neuronal cell surface proteins in autoimmune diseases of the central nervous system. Physiol. Rev. 97, 839–887 (2017).
Steiner, J. et al. Increased prevalence of diverse N-methyl-d-aspartate glutamate receptor antibodies in patients with an initial diagnosis of schizophrenia: specific relevance of IgG NR1a antibodies for distinction from N-methyl-d-aspartate glutamate receptor encephalitis. JAMA Psychiatry 70, 271–278 (2013).
DeGiorgio, L. A. et al. A subset of lupus anti-DNA antibodies cross-reacts with the NR2 glutamate receptor in systemic lupus erythematosus. Nat. Med. 7, 1189–1193 (2001).
Chan, K. et al. Lupus autoantibodies act as positive allosteric modulators at GluN2A-containing NMDA receptors and impair spatial memory. Nat. Commun. 11, 1403 (2020).
Lee, C. H. et al. NMDA receptor structures reveal subunit arrangement and pore architecture. Nature 511, 191–197 (2014).
Li, D. et al. GRIN2D recurrent de novo dominant mutation causes a severe epileptic encephalopathy treatable with NMDA receptor channel blockers. Am. J. Hum. Genet 99, 802–816 (2016).
Zheng, S. Q. et al. MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nat. Methods 14, 331–332 (2017).
Zhang, K. Gctf: real-time CTF determination and correction. J. Struct. Biol. 193, 1–12 (2016).
Scheres, S. H. RELION: implementation of a Bayesian approach to cryo-EM structure determination. J. Struct. Biol. 180, 519–530 (2012).
Pettersen, E. F. et al. UCSF Chimera—a visualization system for exploratory research and analysis. J. Comput. Chem. 25, 1605–1612 (2004).
Emsley, P. & Cowtan, K. Coot: model-building tools for molecular graphics. Acta Crystallogr. D Biol. Crystallogr. 60, 2126–2132 (2004).
Liebschner, D. et al. Macromolecular structure determination using X-rays, neutrons and electrons: recent developments in Phenix. Acta Crystallogr. D Struct. Biol. 75, 861–877 (2019).
Pettersen, E. F. et al. UCSF ChimeraX: structure visualization for researchers, educators, and developers. Protein Sci. 30, 70–82 (2021).
Nielsen, S. S. et al. BioXTAS RAW, a software program for high-throughput automated small-angle X-ray scattering data reduction and preliminary analysis. J. Appl. Crystallogr. 42, 959–964 (2009).
Petoukhov, M. V. et al. New developments in the ATSAS program package for small-angle scattering data analysis. J. Appl. Crystallogr. 45, 342–350 (2012).
Franke, D. & Svergun, D. I. DAMMIF, a program for rapid ab-initio shape determination in small-angle scattering. J. Appl. Crystallogr. 42, 342–346 (2009).
Volkov, V. V. & Svergun, D. I. Uniqueness of ab initio shape determination in small-angle scattering. J. Appl. Crystallogr. 36, 860–864 (2004).
Franke, D. et al. ATSAS 2.8: a comprehensive data analysis suite for small-angle scattering from macromolecular solutions. J. Appl. Crystallogr. 50, 1212–1225 (2017).
Abraham, M. J. et al. GROMACS: high performance molecular simulations through multi-level parallelism from laptops to supercomputers. SoftwareX 1–2, 19–25 (2015).
Best, R. B. et al. Optimization of the additive CHARMM all-atom protein force field targeting improved sampling of the backbone φ, ψ and side-chain χ1 and χ2 dihedral angles. J. Chem. Theory Comput. 8, 3257–3273 (2012).
Jorgensen, W. L., Chandresekhar, J., Madura, J. D., Impey, R. W. & Klein, M. L. Comparison of simple potential functions for simulating liquid water. J. Chem. Phys. 79, 926–935 (1983).
Essmann, U. et al. A smooth particle mesh Ewald method. J. Chem. Phys. 103, 8577–8593 (1995).
Parrinello, M. & Rahman, A. Polymorphic transitions in single-crystals––a new molecular-dynamics method. J. Appl. Phys. 52, 7182–7190 (1981).
Hess, B., Bekker, H., Berendsen, H. J. C. & Fraaije, J. G. E. M.LINCS: a linear constraint solver for molecular simulations. J. Comput. Chem. 18, 1463–1472 (1997).
Sinitskiy, A. V. & Pande, V. S. Computer simulations predict high structural heterogeneity of functional state of NMDA receptors. Biophys. J. 115, 841–852 (2018).
Cock, P. J. et al. Biopython: freely available Python tools for computational molecular biology and bioinformatics. Bioinformatics 25, 1422–1423 (2009).
Acknowledgements
We thank Y. Kong and L. Pan for assistance with the 120-kV microscope, H. Wu and L. Quan for assistance with FACS, Q. Hu, D. Xiang, Y. Wang and Y. Zhang for help with confocal imaging (CEBSIT, CAS) and S. Lan for assistance with SPR (Center for Excellence in Molecular Cell Science, CAS). We thank the staff at Shuimu BioSciences, Ltd. for their assistance with 300-kV cryo-EM data collection and the staff of the BL19U2 beamline for assistance with SAXS data collection (NFPSS and SSRF). We thank H. Prüss (Universitätsmedizin Berlin) for generously providing antibodies for preliminary tests and L. Fubin (Shanghai Jiaotong University) for providing the pF1 expression vector. We also thank M. Poo and M. Zhang for helpful discussion. Financial support is gratefully acknowledged from the STI2030 Major Project (2022ZD0212700), National Natural Science Foundation of China (32221003) and Talent Plan of Shanghai Branch, CAS to S.J.Z. and the CAS Youth Interdisciplinary Team to S.J.Z., H.W. and T.-F.Y.
Author information
Authors and Affiliations
Contributions
H.W. purified the protein, constructed the NMDAR nanodiscs and carried out cryo-EM and two-electrode voltage-clamp recordings. C.X. performed the NMDARGFP-labeled FACS and cloned the mAb sequence with assistance from J.H., J.B.Z. and M.M.J. H.W. and C.X. performed the SPR experiment and confocal imaging with assistance from Z.W.K. B.D. performed immunostaining on mouse brain slices. J.J.D. and T.-F.Y. performed patch-clamp recordings. H.W. and N.L. performed the SAXS experiments. B.D., Q.M.Z., S.C. and X.J.C. collected the data and blood samples of persons with AE. Q.R.W. and H.W. carried out the MD simulation. S.J.Z. conceptualized the project and supervised all experiments. H.W. and S.J.Z. wrote the manuscript with input from all authors.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Structural & Molecular Biology thanks Laurent Groc, Wei Lu and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Primary Handling Editor: Sara Osman, in collaboration with the Nature Structural & Molecular Biology team.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
Extended Data Fig. 1 Biological characterization of patient-derived monoclonal auto-antibodies.
a. Mouse hippocampal immunofluorescence staining with different concentrations of mAb5F6 and mAb2G7. The fluorescence intensity of mAbs and DAPI are shown in green and blue, respectively. The zoom-in views of boxed areas in CA1 region are presented in the upper layer. b. Plotting relative fluorescence intensities in CA1 region versus different concentrations of mAb5F6 (in purple) and mAb2G7 (in blue). Data are presented with mean ± SEM, n = 3 independent replicates. c. FSEC profiles of GluN1-GluN2A NMDARs alone (in black), and in complex with Fab5F6 (in violet) or Fab2G7 (in blue). d. Coomassie blue staining of purified GluN1-GluN2A NMDARs alone, and in complex with Fab5F6 or Fab2G7. The experiments were repeated for three times with similar results.
Extended Data Fig. 2 Cryo-EM data processing of NMDAR-Fab complex structures.
a, d. Pipelines of single-particle analysis and reconstruction of the NMDAR-Fab2G7 (a) and NMDAR-Fab5F6 (d) complex. Particles with clear characteristic orientations from the 2D class average were selected. 3D classifications are carried out with multiple rounds of C1 symmetry of multiple rounds, and the classes with similar conformation (shown in dashed line square) are merged for 3D refinement with C1 symmetry. Surface clippings are shown from top to bottom. Euler angle distributions of particles in cryo-EM density maps for each state is shown next to the cryo-EM density. b, e. Fourier shell correlation curves (FSCs) in cryo-EM density maps of two-Fab2G7 (b) and two-Fab5F6 (e) bound states with (in blue) or without (in green) masking. A gold-standard FSC value of 0.143 is implied for the final resolution. Cryo-EM density maps are coloured based on the local resolution estimation by Relion 3.1. c, f. Fourier shell correlation curves (FSCs) in cryo-EM density maps of one-Fab2G7 (c) and one-Fab5F6 (f) bound states with (in blue) or without (in green) masking. A gold-standard FSC value of 0.143 is implied for the final resolution. Cryo-EM density maps are coloured based on the local resolution estimation by Relion 3.1.
Extended Data Fig. 3 Structural comparison of NTD and LBD in Fab-bound and Fab-free states.
a, c. Schematic representation of GluN2A (in orange) and GluN1 (in blue) subunits in tetrameric assembly, with lines indicate COM distances of R1-R2 lobes of the NTD and D1-D2 lobes of the LBD. b, d. COM distance plots of R1-R2 (left panel) and D1-D2 (right panel) lobes in GluN2A (b) or GluN1 (d) subunits from the structures of NMDAR-Fab5F6 and NMDAR-Fab2G7 in two-Fab and one-Fab bound states, and NMDARWT (PDBID: 6MMP). Filled and empty squares, circles and triangles indicated the Fab-bound state and Fab-free state, respectively.
Extended Data Fig. 4 NTD sequence alignment of in GluN1, GluN2A and GluN2B subunits.
Sequence alignment of GluN1-, GluN2A- and GluN2B-NTD. Residues involved in mAb5F6 and mAb2G7 binding is highlighted in purple and red, respectively. Conserved residues are highlighted in gray.
Extended Data Fig. 5 Production of a mouse-derived anti-GluN1 NMDAR antibody.
a. Monoclonal antibodies were produced by active immune mice with purified intact GluN1-GluN2A NMDARs in digitonin. The spleen cells were isolated and fused with Sp2/0 cells to produce hybridoma. Monoclonal antibody secreting hybridoma cell lines were further screened by immunocytochemistry staining on live HEK293T cells cotransfected with the obligatory GluN1 subunit and GluN2 or GluN3 subunit NMDARs. b. FSEC profiles of the mAb4F11 and Coomassie blue gel staining of the purified mAb4F11 and isolated Fab4F11. The experiments were repeated for three times with similar results. c. Immunocytochemistry staining of mAb4F11 binding to NMDAR transfected with GluN1-GluN2A NMDARs (green). mAb4F11 was labelled with Alexa Fluor 594 anti-mouse IgG secondary antibody (red). The experiments were repeated for three times with similar results. d. SPR measurements of the mAb4F11 (concentration: 0.347 to 6.7E-03 μM) interacting with GluN1-GluN2A NMDAR. The data were fitted with a kinetic affinity fit model with a 1:1 binding affinity (KD) indicated. The estimated KD of mAb4F11 was 16 nM. Ru, resonance units. e. Representative micrograph (left panel) and 2D images (right panel) of NMDAR-Fab4F11 on Talos 120 kV cryo-electron microscopy captured by CETA. The black circle points out the featured EM density of Fab4F11. The representative micrograph was randomly chosen, and the 3D model was reconstructed based on the particles from 37 micrographs. f. 3D reconstruction and fitted atomic model of the NMDAR-Fab4F11 complex. Fab4F11 binding to the GluN1-NTD could be visualized. g. r.m.s.d. trajectories for Fab4F11-GluN1-NTD on Cα atoms based on the initial structure within the total simulation time of 600 ns. Each iteration of 200 ns is repeated for three times. h. Cartoon illustration of Fab4F11-GluN1-NTD complex obtained from MD simulation. First round of MD result was shown. Dash line highlighted the interactions formed by VH CDRs (in yellow), GluN1-R1 (in red) and GluN1-R2 (in purple) lobes. i. COM distance between VH CDRs and GluN1-R1 (in red) shows that the major contributor of the fluctuations in the RMSD across different repeats of MD simulations is the relative movement between VH and GluN1-R1. Moreover, COM distance between CDRs and GluN1-R2 (in purple) shows a consistent interaction throughout all trajectories.
Supplementary information
Source data
Source Data Fig. 1
Unprocessed gels.
Source Data Fig. 3
Statistical source data.
Source Data Fig. 4
Statistical source data.
Source Data Fig. 5
Statistical source data.
Source Data Extended Data Fig. 1
Unprocessed gels.
Source Data Extended Data Fig. 3
Structural measurement.
Source Data Extended Data Fig. 5
Unprocessed gels.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Wang, H., Xie, C., Deng, B. et al. Structural basis for antibody-mediated NMDA receptor clustering and endocytosis in autoimmune encephalitis. Nat Struct Mol Biol (2024). https://doi.org/10.1038/s41594-024-01387-3
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41594-024-01387-3
