NMDA receptor autoantibodies primarily impair the 1 extrasynaptic compartment

5 Autoantibodies directed against the N-methyl-D-aspartate receptor (NMDAR-Ab) are 6 pathogenic immunoglobulins detected in patients suffering from NMDAR encephalitis. 7 NMDAR-Ab alter the receptor membrane trafficking, synaptic transmission and neuronal 8 network properties, leading to patients’ neurological and psychiatric symptoms. Patients often 9 have very little neuronal damage but rapid and massive (treatment-responsive) brain 10 dysfunctions related to unknown early mechanism of NMDAR-Ab. Our understanding of this 11 early molecular cascade remains surprisingly fragmented. 12 Here, we used a combination of single molecule-based imaging of membrane proteins to 13 unveil the spatio-temporal action of NMDAR-Ab onto live hippocampal neurons. 14 We first demonstrate that different clones of NMDAR-Ab primarily affect extrasynaptic -and 15 not synaptic-NMDAR. In the first minutes, NMDAR-Ab increase extrasynaptic NMDAR 16 membrane dynamics, de-clustering its surface interactome. NMDAR-Ab also rapidly 17 reshuffle all membrane proteins located at the extrasynaptic compartment. Consistent with 18 this alteration of multiple proteins, NMDAR-Ab effects were not mediated through the sole 19 interaction between NMDAR and EphB2 receptor. At the long-term, NMDAR-Ab reduce 20 NMDAR synaptic pool by slowing down receptor membrane dynamics in a cross-linking 21 independent manner. Remarkably, exposing only extrasynaptic NMDAR to NMDAR-Ab was 22 sufficient to produce

not synaptic-NMDAR.In the first minutes, NMDAR-Ab increase extrasynaptic NMDAR membrane dynamics, de-clustering its surface interactome.NMDAR-Ab also rapidly reshuffle all membrane proteins located at the extrasynaptic compartment.Consistent with this alteration of multiple proteins, NMDAR-Ab effects were not mediated through the sole interaction between NMDAR and EphB2 receptor.At the long-term, NMDAR-Ab reduce NMDAR synaptic pool by slowing down receptor membrane dynamics in a cross-linking independent manner.Remarkably, exposing only extrasynaptic NMDAR to NMDAR-Ab was sufficient to produce their full-blown effect on synaptic receptors.
Collectively, we demonstrate that NMDAR-Ab first impair extrasynaptic proteins, and then the synaptic ones.These data shed thus new, and unsuspected, lights on the mode of action of NMDAR-Ab and likely to our understanding of (extra)synaptopathies.

Introduction
Over the past decades, the identification of autoimmune neurological and psychiatric disorders with patients expressing autoantibodies directed against membrane proteins has flourished 1 .The most prominent autoantibody-mediated brain disorder is the N-methyl-D- aspartate receptor (NMDAR) encephalitis, in which patients develop antibodies directed against the extracellular N-terminal domain of the obligatory GluN1-NMDAR subunit of the NMDAR (NMDAR-Ab) [2][3][4][5] .The patients present a spectrum of severe neurological features (e.g.seizures) and psychiatric symptoms (e.g.psychosis) that could result in a persistent coma, illustrating the complexity of the neuropsychiatric features induced by NMDAR-Ab 6 .
Mechanistically, it has been proposed that NMDAR-Ab binding destabilizes the receptor through a weakening of the interaction with EphB2 receptor (EphB2R) 7,8 .Over time, this destabilisation increases the surface diffusion of synaptic NMDAR and promotes their displacement to the extrasynaptic compartment 7,9 .Then, the bivalency of autoantibodies would cross-link receptors and favour their internalization, decreasing NMDAR plasma membrane content and global signalling 5,10 .
Although there is a full consensus that NMDAR-Ab trigger NMDAR hypofunction at hippocampal and cortical neurons 11 , our understanding of the molecular cascade underpinning this phenotype remains quite fragmented, with different scenarios equally supported by existing data 12 .For instance, NMDAR-Ab alter the surface trafficking 7 and nanoorganization 13 of both synaptic and extrasynaptic NMDAR, leaving open the question on whether synaptic and extrasynaptic receptor compartments are concomitantly or sequentially impacted.On one hand, the high density of NMDAR within the postsynaptic density 14 would favour NMDAR-Ab action on the synaptic pool.On the other hand, the high density and crowding of transmembrane proteins located within the synaptic cleft (approx.20 nm wide) would rather restrain NMDAR-Ab (approx.12 nm) penetration within synapses and favour their binding to extrasynaptic NMDAR.Furthermore, NMDAR interact with a dozen of transmembrane receptors and ion channels located within synapses or, for the majority, outside synapses [15][16][17] .We previously showed that NMDAR-Ab weaken the interaction between NMDAR and EphB2R 7 , which regulate the NMDAR synaptic pool [18][19][20] .However, NMDAR and EphB2R are similarly present in both synaptic and extrasynaptic membrane compartments, complicating the interpretation of the corrupted interaction by NMDAR-Ab.In addition, these autoantibodies alter the membrane clustering and surface dynamics of the dopamine receptor, which is a NMDAR protein-protein interactor solely located in the extrasynaptic compartment [21][22][23] .Whether NMDAR-Ab similarly impact other partners of the NMDAR remains an open question.Finally, NMDAR-Ab bivalency would cross-link NMDAR and trigger their internalization, a process that massively reduced the receptor membrane diffusion.Yet, it has been reported that NMDAR-Ab can also increase the surface diffusion of some NMDAR exiting synapses and exploring the extrasynaptic membrane compartment 7 , inconsistent with such a receptor cross-linking process.As other autoantibodies directed against different neurotransmitter receptors can alter their membrane signalling through cross-linking-independent process [24][25][26] , it remains somehow unclear whether NMDAR-Ab action on NMDAR surface organization relies solely on the autoantibody-induced cross-linking.Therefore, our lack of understanding of the spatiotemporal action of NMDAR-Ab on its target and other related membrane proteins strongly hamper our capacity to draw a comprehensive scenario of the molecular cascade triggered by autoantibodies.To directly address this question, we here used a combination of single molecule-based imaging of various membrane proteins to unveil NMDAR-Ab effects onto live hippocampal neurons.We took advantage of the recently developed patients-derived monoclonal NMDAR-Ab to precisely quantify their action over time at the nanoscale.
Cultures were kept at 37 °C in 5 % CO2.Neurons were transfected at DIV 10 using the calcium-phosphate coprecipitation method.Precipitates containing plasmidic DNA were prepared using the following solutions: TE (1 mM Tris-HCl pH 7.3, 1 mM EDTA), CaCl2

Plasmidic DNA
GluN2A-SEP, GluN1-SEP and GluN1-mEos3.2were expressed in a pRcCMV plasmid with SEP and mEeos3.2 in N-terminal.Homer1c-dsRed and Homer1c-GFP were expressed in pcDNA3.1 with dsRed and GFP in N-terminal of the insert.Flag-EphB2R Y504E and Flag-EphB2R Y504 mutants were generated from Flag-EphB2R WT in a pcDNA3 plasmid and previously described 19 .The sh_Control and sh_EphB2R were previously described 28 .GluN1-V5-HRP was generated in a pcDNA3 vector with V5 tag and HRP enzyme in N -terminal.
Clathrin Light Chain-mCherry in N-terminal was previously described 29 .
Expression vectors encoding for heavy and light chains of this clone was introduced into HEK293 cells through transient transfection.The supernatant was utilized to purify the recombinant autoantibodies, following previously established methods 30 .Human control antibody (Control-Ab) is an antibody non-reactive to human tissue derived from mature B cells from the blood of healthy donor 32 .A CSF from 11-year-old female patient with classic acute symptoms of limbic encephalitis (speech and memory disorders and psychiatric disturbances with seizures) and negative global tumor screening was also used in the study.
The patient underwent pre-treatment lumbar puncture at Bordeaux University Hospital (France).The CSF was tested for the presence of NMDAR-Ab using a cell-based assay on human embryonic kidney cells (HEK293) expressing both GluN1 and GluN2B subunits of the NMDAR, as previously described 7 .Cells were fixed (4% PFA, 10 min) and then incubated with patient' CSF (1:50 in saturation buffer, 90 min).The CSF was considered as positive when a clear staining was confirmed by 3 different readers in 3 independent assays.CSF samples was rapidly stored at -80°C after acquisition in the Multi-thematic Biological Resource Centre (Biobank of the Bordeaux University Hospital, France, certification NF S 96-900).The activity of conservation of biological material is declared under n° DC-2020-

Live imaging
Neurons were incubated with Control-Ab, NMDAR-Ab or GFP-Ab (Thermo Fisher Scientific, #A6455) in a wet chamber at 37°C for the indicated time and concentration.
Coverslips containing neurons were then mounted in an open chamber (Ludin chamber, Life Imaging Services), with 500 µl of Tyrode solution (30 mM D-glucose, 120 mM NaCl, 5 mM KCl, 2 mM MgCl2, 2 mM CaCl2 and 25 mM HEPES, pH 7.3-7.4).For time lapse imaging, NMDAR-Ab were directly added in the chamber.Microscope sessions were performed with an inverted confocal spinning-disk microscope Leica DMI8 equipped with 63X oil objective, a sCMOS Prime 95B camera (Photometrics) and an environmental chamber to control temperature and CO2 (37°C, 5 % CO2).A 488 nm laser was used for GluN2A-SEP and GluN1-SEP and a 561 nm laser for Homer-dsRed and CLC-mCherry.The laser power was kept the same between all conditions.The different analysis were done on ImageJ with a homemade macro.Density of GluN1-SEP and Homer clusters were calculated using the mean of clusters density from 2 ROI per neuron.

Coupling of latex beads with antibodies and evaluation of the efficiency of the coupling
4 µg of CML Latex Beads 1.0 µm (Invitrogen, #C37483) were centrifuged at 15000 g for 10 min and washed in 1 mL MES Coupling Buffer (MES 50 mM pH 6; EDTA 1 mM; 0.0005% Tween 20 (Sigma-Aldrich, #P9416)).Beads were re-suspended in 300 µl MES Coupling Buffer and carboxyl group were activated by adding 120 µl of freshly prepared 50 mg/ml EDAC (#E6383, Sigma-Aldrich)-MES Buffer.Beads were mixed on a rotating wheel during 15 min.Activated beads were washed 3 times in PBS-0.0005%Tween20 after each centrifugation at 15000 g for 10 min.Activated beads were re-suspended in 50 µl PBS and monoclonal antibodies were added at 1 mg/ml: Control-Ab, NMDAR-Ab, GluA2 (Merck Millipore,# MAB397) and mixed in a thermomixer (1000 rpm, 4h).Beads coupled with antibody were washed in PBS-0.0005%Tween20 and re-suspended in PBS-BSA 1% and kept at 4°C.To evaluate the efficiency of the coupling, beads were mixed with secondary antibody Alexa 488 (1/500, anti-mouse #A11001, anti-human #A11013) during 15 min on a rotating wheel.Flow cytometry was performed at the TBM Core platform (Bordeaux).Beads coupled to antibody were applied on the neuron with a concentration of 40 µg/ml.

Fab fragment preparation
Fab fragment from monoclonal NMDAR IgG (clone #003-102) were prepared following manufacturer's instructions (Pierce™ Fab Micro Preparation Kit, Thermo Fisher Scientific, #44685).Briefly, IgG were digested for 3-4h with a digestion buffer in a column tube containing immobilized papain with a tabletop rocker at 37°C.Fab fragments were then purified from non-digested IgG and Fc fragments with a Protein A column.Digestion and purification were confirmed with a SDS-PAGE and Fab concentration was determined with absorption at 280nm.

Single quantum dot (QD) tracking
Neurons were first incubated 2h in medium from the cells containing 1 µg/ml anti-EphB2R antibody (R&D system, #AF467) in a wet chamber or directly incubated 10 min with Rabbit anti-GFP antibody (1/25000, Thermo Fisher Scientific, #A6455) followed by 10 min incubation with QD655 coupled to goat anti-rabbit F(ab')2 (1/25000, Thermo Fisher Scientific, #Q11422MP).All incubations were done in tyrode solution supplemented with 1% BSA at 37°C.Coverslips were mounted in tyrode on a heated -chamber for observation.QD were detected by using a mercury lamp and appropriate excitation/emission filters.Images were obtained with an acquisition time of 50 ms (20 Hz) with 500 frames.Signals were detected using an EMCCD camera (Evolve, Photometrics).QD recording sessions were processed with the Metamorph software.The instantaneous diffusion coefficient (D) was calculated for each trajectory from linear fits of the first four points of the mean square displacement (MSD) vs. time function using MSD(t) ≤ r2 > (t) = 4Dt.To determine the distribution of single QD complexes, frame stacks were obtained and after binarisation of the synaptic signal (Homer-dsRed) the QD complexes were automatically located into synaptic or extrasynaptic compartment.

Single particle tracking photoactivation localization microscopy (sptPALM)
Neurons were incubated for the different conditions at 37°C.Coverslip containing neurons were imaged in an open chamber (Ludin chamber, Life Imaging Services) with 1 ml of tyrode solution at 37°C.The chamber was mounted on a Nikon Ti Eclipse microscope (Nikon France S.A.S.) equipped with a Perfect Focus System (PFS), an iLas² TIRF arm (Gataca Systems) an Apo TIRF 100 X oil-immersion objective (NA 1.49) and an ORCA-Fusion BT sCMOS camera (Hamamatsu) with a final pixel size of 65 nm.Transfected cells were detected with Homer-GFP signal (488 nm laser) and GluN1-mEos3.2was photo-activated using 405 nm laser and the resulting photo-converted single molecule fluorescence was excited with a 561 nm laser.Both 405 nm and 561 nm lasers illuminated the sample simultaneously.To keep the number of the stochastically activated molecules constant and well separated during the acquisition, the 405 nm laser power was adjusted.Acquisition was done with Metamorph software, with 2000 frames and exposure time of 50ms with a TIRF illumination to track surface GluN1 subunit-mEos.Detection and reconnection of trajectories (>10 frames) was done with PALM Tracer plugin for Metamorph with a Gaussian fit (xy sigma) to determine the centroid coordinate of a single molecule.Homer-GFP was used as a synaptic marker to discriminate synaptic and extrasynaptic GluN1-NMDAR trajectories.MSD and coefficient diffusion were calculated as described before for single QD tracking.The confinement area was calculated as the averaged MSD at the specified time lag in the plateau (0.5-1 for synaptic trajectories, 1.5-2 for extrasynaptic ones).
Neurons were fixed with PBS-4%PFA/4% sucrose solution for 15 min then washed several times with 50 µM NH4Cl solution and kept at 4°C until imaging.Imaging sessions were also performed on the Nikon Ti Eclipse microscope with the same equipment as described before for sptPALM.Lateral drift was corrected using multicolor fluorescent microspheres (Life Technologies, #T7279 TetraSpeck).For analysis, PALM-Tracer was used to extract exact coordinate of a localisation and SR-Tesseler was used to quantify the clustering of proteins from the localization file as previously described 33 .A minimum of 10 single molecule localizations and density factor of 2 was used to define a cluster (based on comparison with background signals).Synaptic clusters of proteins were determined using Homer-GFP as synaptic marker for GluN1-HRP and Homer/Gephyrin staining for NHS-Ester experiments.

Statistical analysis
Statistical analysis was performed using GraphPad Prism 10 (GraphPad Software, La Jolla, CA, USA).Depending on the data distribution, Mann-Whitney U-test or the two-tailed Student t-test was used to test differences between two groups and One-way ANOVA followed by Tukey's post hoc test or Kruskal-Wallis tests followed by Dunn's post hoc test was used for multiple group comparisons.Two-ways ANOVA followed by Tukey's multiple comparison test was used between groups that have been split on two independent variables.All datas were obtained from at least 3 independents experiments (cultures).

Acute exposure with NMDAR-Ab alters only extrasynaptic NMDAR surface trafficking
Although the long-term effect of NMDAR-Ab onto synaptic receptor pool, transmission, and plasticity has been thoroughly investigated 34 , our understanding of the attack phase, i.e. the first minutes after exposure to NMDAR-Ab, is still limited.To tackle this question, we exposed cultured hippocampal neurons to either control monoclonal antibodies (Control-Ab) purified from the serum of a healthy donor or NMDAR monoclonal autoantibodies (NMDAR-Ab, clone 003-102) derived from NMDAR encephalitis patients (Fig. 1A).As expected, NMDAR-Ab label synaptic and extrasynaptic NMDAR following a 30 min incubation (Fig. 1B and C).To test whether NMDAR-Ab rapidly alter the NMDAR synaptic pool, we performed a time-lapse imaging on hippocampal neurons expressing GluN2A subunit tagged with a super ecliptic pH-sensitive GFP (SEP) at its extracellular N-terminus, which mainly highlight surface GluN2A-NMDAR 35 (Fig. 1D).Over the first 30 min of exposure with NMDAR-Ab, synaptic NMDAR fluorescence intensity was unaffected (Fig. 1E and F), indicating that the NMDAR synaptic pool was not yet affected by NMDAR-Ab in this time window.Although we previously demonstrated that NMDAR-Ab alter NMDAR trafficking in the very first hours 7 , the then used methods (i.e.single nanoparticle tracking) could not specifically pinpoint the locus of prime alteration.We therefore selected another approach that could give us access to a high number of trajectories in each specific compartment (i.e.synaptic and extrasynaptic) simultaneously.For this, we used the superresolution microscopy technique single particle tracking photo-activated localization microscopy (sptPALM) which fulfil these criteria.Hippocampal neurons were transfected with Homer 1c-GFP as a synaptic marker and with GluN1 subunit coupled to the photoconvertible protein mEos3.2which can be effectively targeted by NMDAR-Ab (Supplementary Fig. 1A).Total internal reflection fluorescence (TIRF) illumination was used to mainly activate surface GluN1-mEos and track synaptic and extrasynaptic NMDAR with distinct behaviours (Supplementary Fig. 1B and C).As expected, synaptic trajectories represented only approximately 20% of total trajectories and they were more confined (as indicated by the mean square displacement (MSD) curves) and less diffusive (shift toward blue colour in the diffusion map) than extrasynaptic ones (Supplementary Fig. 1C).In order to evaluate the acute effect of NMDAR-Ab, neurons were acutely (30 min) exposed to Control- Ab various NMDAR-Ab clones, or encephalitis patient CSF (Supplementary Fig. 1D).
Synaptic NMDAR surface trafficking remained unaltered in all conditions (Fig. 1G).In contrast, the extrasynaptic NMDAR surface trafficking was strongly increased by NMDAR-Ab and patient CSF (Fig. 1H).The MSD curve was left-shifted, indicating lower confinement of trajectories with a higher confinement area and the diffusion coefficients were increased when compared to Control-Ab (Fig. 1H).Thus extrasynaptic -and not synaptic-NMDAR were similarly affected by NMDAR-Ab, irrespective to the clones.In addition, the increase surface diffusion strikingly contrasts to the reduced diffusion produced by artificial crosslinker antibodies (Supplementary Fig. 1E).Altogether, these data demonstrate that NMDAR-Ab acutely and specifically increase the surface dynamics of extrasynaptic receptors, possibly through a disruption of the protein-protein interaction between NMDAR and membrane partners 16,17 .

EphB2R is not required for NMDAR-Ab pathogenicity
Previous studies have suggested that the disruption of synaptic NMDAR-EphB2R interaction is instrumental for NMDAR-Ab pathogenicity 8,13,36 .In light of the above data, we thus tested whether this interaction is required for NMDAR-Ab cellular effects.First, surface staining of EphB2R shows that EphB2R is present at synapse as well as in the extrasynaptic compartment (Fig. 2A).Using an antibody directed against the extracellular part of EphB2R (EphB2R-Ab), which disrupt NMDAR-EphB2R interaction 36 , we investigated the impact of EphB2R-Ab on NMDAR surface trafficking using single Quantum Dot (QD) tracking.
Neurons were incubated for 2h with either Control-Ab or EphB2R-Ab and single QD experiments were performed (Fig. 2B).EphB2R-Ab significantly increased synaptic and extrasynaptic NMDAR membrane dynamics (Fig. 2B, C and Supplementary Fig. 2A, B), consistent with the stabilizing role of the NMDAR-EphB2R interaction 20 .We then tested the role of this interaction over a long antibody incubation period (24h) by measuring, in live neurons, the synaptic content of membrane NMDAR (Fig. 2D).Compared to Control-Ab, incubation with NMDAR-Ab reduced by half the NMDAR synaptic density and decreased the cluster area (Fig. 2E).In contrast, synaptic NMDAR cluster density and area were not significantly altered by exposure to EphB2R-Ab, indicating that disrupting NMDAR-EphB2R interaction is not sufficient to deplete NMDAR synaptic pool (Fig. 2E).Consistently, synaptic NMDAR cluster density and area in neurons exposed to both NMDAR-Ab and EphB2R were not different from NMDAR-Ab condition alone (Fig. 2E).In all conditions, the postsynaptic density (Homer 1c cluster) was not affected, indicating that the decrease of synaptic NMDAR is not due to the degradation of the PSD (Fig. 2F) 37 .
To further confirm this conclusion and strongly impair the interaction in a longer period of time, we next used a genetic strategy to modify the interaction.EphB2R-NMDAR interaction is mediated through extracellular phosphorylation of a single tyrosine of EphB2R (p*Y504) 21 .
We generated two EphB2R mutants, previously characterized 19 : EphB2R-Y504E that strongly binds to NMDAR and EphB2R-Y504F that weakly binds to NMDAR compared to EphB2R WT (Fig. 3A).NMDAR surface dynamics was decreased in neurons transfected with EphB2R-Y504E and increased in neurons expressing EphB2R-Y504F compared to EphB2-WT (Fig. 3B).Noteworthy, EphB2R surface trafficking was not altered by the genetic mutations (Supplementary Fig. 3A).Using live imaging, we then tested whether NMDAR cluster density and area were altered by mutants and/or NMDAR-Ab (Fig. 3C).As expected [18][19][20] , EphB2R mutants significantly altered the NMDAR and Homer1c cluster area, without affecting their linear density (Fig. 3D and Supplementary Fig. 3B).Similarly, NMDAR-Ab (24h incubation) decreased NMDAR cluster density and area (Fig. 3D and Supplementary Fig. 3B).Interestingly, NMDAR-Ab-induced decrease in synaptic NMDAR pool was not altered by the EphB2R genotype (Fig. 3D and Supplementary Fig. 3B), suggesting that NMDAR-Ab action is independent on the state of the interaction between NMDAR and EphB2R.
Finally, we knocked down the expression of EphB2R using a previously validated shRNA directed against EphB2R 28 .Transfection with sh_EphB2R decreased by 50% the surface content of EphB2R when compared to sh_Control (Fig. 3E).Transfection with sh_EphB2R significantly decreases NMDAR synaptic cluster area without change in the linear density (Fig. 3F, G and Supplementary Fig. 3C).Incubation with NMDAR-Ab (24h) decreases synaptic NMDAR cluster area and cluster (Fig. 3F, G and Supplementary Fig. 3C).Yet, the same effects were observed in the sh_EphB2R group, with no interaction between the effects of NMDAR-Ab and EphB2R knock-down (Fig. 3G and Supplementary Fig. 3C).Taken together, these results indicate that although EphB2R tunes NMDAR membrane dynamics and synaptic content, NMDAR-Ab effects are mainly independent from the status of the EphB2R-NMDAR interaction.

NMDAR-Ab disorganize NMDAR surface interactome
Whether NMDAR-Ab mediate their effect through altered NMDAR-protein interaction remains thus an open question.Since dozens of membrane proteins have been shown to interact with the NMDAR 38 , we intended to label the surface interactome of the NMDAR at the nanoscale and test whether it is impacted by NMDAR-Ab.To address this challenge, we first generated a construct for a proximity-labelling assay, which consists of a V5-tagged GluN1 subunit coupled to an HRP enzyme (GluN1-HRP).After adding the biotin derivative non-permeant tyramide and H2O2, GluN1-HRP biotinylates only surface protein within a 20 nm radius, revealing the surface protein interactome of NMDAR (SPIN) (Fig. 4A).Neurons expressing GluN1-HRP in presence of tyramide displayed a surface biotinylation revealed by streptavidin, along the dendrite with accumulation in synapses (Fig. 4B).Without tyramide, the staining was virtually absent (Fig. 4B).To characterize the nanoscale organization of SPIN within synaptic and extrasynaptic compartments, neurons were also transfected with Homer-GFP and direct stochastic optical reconstruction microscopy (dSTORM) imaging was performed.Clusters of synaptic SPIN were larger and denser than extrasynaptic ones (Fig. 4C).Yet, extrasynaptic SPIN clusters were numerous, paving the whole dendritic surface.
Then, neurons were incubated during 30 min or 24 h with Control-Ab or NMDAR-Ab before SPIN labelling and imaging.The overall detections were increased in the SPIN after 24h exposure to NMDAR-Ab (Fig. 4D).Acutely (30 min), NMDAR-Ab had no effect on synaptic SPIN, consistent with the above data (Fig. 4E, F and Supplementary Fig. 4A).However, NMDAR-Ab increased extrasynaptic SPIN area both in clusters and nanodomains (Fig. 4G, H and Supplementary Fig. 4B).This increase in area was accompanied by a decrease in local density.These observations are fully consistent with the increased dynamics of extrasynaptic NMDAR following NMDAR-Ab exposure, fuelling the hypothesis that NMDAR-Ab rapidly disrupt extrasynaptic complexes of NMDAR that become laterally dispersed within the membrane.On the long-term (24h), this extrasynaptic alteration fully remains, and propagate to the synaptic compartment where SPIN clusters density greatly increased (Fig. 4E-H).In addition, this chronic exposure to NMDAR-Ab altered the shape of synaptic SPIN clusters, which became fragmented into 2-3 nanodomains with increased local density and decreased area (Fig. 4E and Supplementary Fig. 3A).Because GluN1-Abs can be biotinylated by HRP once bound to the GluN1 subunit, we tested whether the altered SPIN nanoscale organization was due to the presence of antibodies.For this, neurons were acutely (30 min) exposed to an anti-GluN1 subunit antibody (clone 10B11 from rabbit) and SPIN analysis was performed.On the opposite to the effect of patients' NMDAR-Abs, this antibody reduced the area and increased the local density of extrasynaptic SPIN (Supplementary Fig. 4C), indicating that the sole presence of an antibody bound to the GluN1 subunit does not predict patients' NMDAR-Ab effects.Collectively, these data indicate that a short-exposure to NMDAR-Ab impair the nanoscale organization of extrasynaptic NMDAR and related partners, with a propagation of these alterations to the synaptic receptor pool over time.

Neuronal surfaceome is acutely altered by NMDAR-Ab
The fact that membrane NMDAR disorganization propagate from the extrasynaptic to synaptic pool opens the possibility that many other membrane proteins could be affected.We thus label all membrane proteins in live neurons using NHS-Ester 647 (Fig. 5A and Supplementary Fig. 5A).dSTORM experiments were then performed to access the nanoscale organization of all neuronal surface proteins at the surface of hippocampal neurons (Fig. 5B and Supplementary Fig. 5A, B).In the basal condition, we report that the protein surfaceome was comparable between synaptic and extrasynaptic clusters in term of local density and area, indicating, to our surprise, that surface proteins are similarly packed within and outside synapses (Supplementary Fig. 5C and D).NMDAR-Ab (30 min incubation) did not change the overall detection of proteins at the neuronal surface (Fig. 5C).At synapses, NMDAR-Ab modestly impacted protein distribution with a decrease in cluster local density (Fig. 5D and     E).However, NMDAR-Ab severely alter the extrasynaptic protein surfaceome.Protein clusters enlarged and their protein density massively dropped down (Fig. 5F and G).A similar effect, although to a lower extent, was confirmed with another NMDAR-Ab clone (#007-124; Supplementary Fig. 5E and F).Together, these data indicate that a short exposure to NMDAR-Ab is sufficient to strongly disorganize membrane proteins in the extrasynaptic compartment, with an overall de-clustering of proteins consistent with the above increase diffusion of receptors.

NMDAR-Ab do not mimic artificial cross-linking
As mentioned above, NMDAR-Ab have been proposed to act as cross-linker on the receptors, favouring their internalization.Yet, our observations of NMDAR de-clustering and increase lateral dynamics are simply orthogonal to such a scenario.To tackle this discrepancy we exposed GluN1-SEP-expressing hippocampal neurons to either NMDAR-Ab or antibody against GFP (GFP-Ab), which is one of the most potent artificial cross-linker 39 (Fig. 6A).GFP-Ab had no effect the SEP fluorescence per se, or on the postsynaptic density clusters (Supplementary Fig. 6A and B).As expected, GFP-Ab (6h) strongly reduced NMDAR cluster density both in synapse and dendritic shaft, indicating an overall decrease in surface NMDAR (Fig. 6B).However, the NMDAR cluster density onto dendritic shaft remained unaffected by the various NMDAR-Ab clones while the linear density of synaptic NMDAR clusters decreased (Fig. 6B and Supplementary Fig. 6C).This indicates that NMDAR-Ab (6h) disorganize surface NMDAR without altering their membrane content.To further test this possibility, neurons were simultaneously exposed to GFP-Ab and NMDAR-Ab in order to unveil putative competitive or additive effect.NMDAR-Ab and GFP-Ab decrease to a similar amount the linear density of synaptic NMDAR, revealing no additive effect (Fig. 6B).Yet, at the extrasynaptic membrane, the exposure of both NMDAR-Ab and GFP-Ab prevented the reduction in membrane NMDAR content induced by GFP-Ab alone (Fig. 6B).Importantly, these two antibodies were not competing each other for binding on the GluN1 subunit (NMDAR-Ab bind GluN1 subunit, GFP-Ab bind SEP) (Supplementary Fig. 6D and E).Thus, the GFP-Ab-induced cross-linking of NMDAR is distinct from the effects produced by NMDAR-Ab.

A C C E P T E D
To further substantiate these observations, we measured the amount of NMDAR that colocalize with endocytotic pits, labelled by clathrin.For this, neurons were transfected with GluN1-SEP and clathrin light chain mCherry (CLC-mCherry) and exposed to the different antibodies for 6h (Fig. 6C).Then, we measured the percent of clathrin-coated pits containing NMDAR.Neurons exposed to GFP-Ab displayed a significant increase in GluN1-positive CLC-coated pits, indicating an upregulation of NMDAR internalization (Fig. 6D).In contrast, the percent of GluN1-positive CLC-coated pits remained unaltered in neurons exposed to the NMDAR-Ab clones (Fig. 6D and Supplementary Fig. 6F).When simultaneously exposing neurons to GFP-Ab and NMDAR-Ab, the percent of GluN1-positive CLC-coated pits remain unaltered (Fig. 6D), indicating that the NMDAR-Ab prevented the GFP-Ab-induced internalization.These data indicate thus that NMDAR-Ab (6h) do not trigger a cross-linkinginduced NMDAR internalization, but rather a redistribution at the neuronal surface.
The cross-link effect of antibodies onto surface receptor relies on their divalency 37 .To test the effect of NMDAR-Ab divalency on the receptor membrane dynamics, we generated Fab fragment from NMDAR-Ab (NMDAR-Fab) (Fig. 6E).We first report that NMDAR-Fab binds to membrane receptor at high concentration (Supplementary Fig. 6G), consistent with the loss of affinity of Fab compared to full immunoglobulin (Fab2 + Fc) 40 .We then performed sptPALM experiments to investigate the impact of NMDAR-Ab or NMDAR-Fab (6h incubation) on the NMDAR surface trafficking (Fig. 6E).As previously reported, neurons incubated for hours with NMDAR-Ab have reduced NMDAR membrane dynamics 7 .
Interestingly, NMDAR-Fab similarly reduced NMDAR surface trafficking, as displayed by the significant decrease in MSD and coefficient diffusion (Fig. 6E and F).Altogether, these data indicate that NMDAR-Ab rapidly disorganize extrasynaptic NMDAR and reduce over time their membrane dynamics in a cross-linking independent manner.

Targeting only extrasynaptic NMDAR with NMDAR-Ab is sufficient to induce NMDAR synaptic loss
Because extrasynaptic NMDAR are the prime locus of action of NMDAR, we finally tested whether acting specifically on these receptors is sufficient to produce NMDAR-Ab pathogenic effect.To tackle this question, we coupled latex beads (micron wide) to NMDAR-Ab in order to target only extrasynaptic NMDAR (synaptic cleft is 20 nm wide) (Fig. 7A).
Using flow cytometry, we determined that 95% and 85% of the beads were successfully coupled to respectively NMDAR-Ab and Control-Ab, without free NMDAR-Ab in the solution (Fig. 7B and Supplementary Fig. 7A).Ab-beads incubation on the neurons resulted in accumulation of NMDAR-Ab beads but not Control-Ab beads, confirming the specificity and functionality of NMDAR-Ab (Fig. 7C).To determine the effect of NMDAR-Ab beads, neurons expressing GluN1-SEP were exposed to NMDAR-Ab-beads for 24h (Fig. 7D).
Control-Ab beads had no effect on synaptic GluN1 subunit clusters or Homer 1c, indicating that the treatment was not affecting neuronal viability (Supplementary Fig. 7B and C).
As observed above, most of surface NMDAR were extrasynaptic after exposure to NMDAR-Ab-beads (Fig. 7D).This effect was specific to NMDAR-Ab since beads coupled to antibodies against the AMPA receptor subunit, GluA2, were without effect on synaptic GluN1 subunit clusters (Supplementary Fig. 7E).Together, these results indicate that NMDAR-Ab targeting only extrasynaptic NMDAR are sufficient to trigger NMDAR synaptic loss at 24h of exposure, similar to that of NMDAR-Ab targeting both extrasynaptic and synaptic receptors.

Discussion
Understanding the mechanism underpinning the pathogenic effect of NMDAR-Ab from patients with encephalitis is essential for the correlation with clinical symptoms, for the development of innovative therapeutic strategies for autoimmune brain disorders, as well as for gaining further molecular insights into NMDAR-mediated neurological and psychiatric conditions.In this study, we demonstrate that various monoclonal NMDAR-Ab primarily altered the extrasynaptic NMDAR pool, and not the synaptic one.In the initial and acute phase, NMDAR-Ab greatly disorganize extrasynaptic NMDAR, membrane proteins in their close proximity, as well as most surface proteins.NMDAR-Ab increase the dynamics of NMDAR through an overall de-clustering of proteins.Over time, in the chronic phase, NMDAR-Ab increase both synaptic and extrasynaptic NMDAR interactome protein density, reducing the overall membrane diffusion of NMDAR in a cross-linking independent process.Strikingly, NMDAR-Ab full-blown effect was observed when they only target extrasynaptic NMDAR.Collectively, these data fuel a model in which NMDAR-Ab alter NMDAR signalling by first acting at the extrasynaptic compartment (see model, Fig. 7F).Since the NMDAR synaptic pool greatly depends on the lateral diffusion of extrasynaptic receptors, a corrupted trafficking and organization at an extrasynaptic locus will inevitably reduce synaptic NMDAR.Our data support thus the view that NMDAR encephalitis, at its early stage, is an (extra)synaptopathy, providing radically new perspective on the molecular mechanism and potential therapeutic perspective.
We and others have previously shown that NMDAR membrane dynamics and distribution is altered by NMDAR-Ab from patients with encephalitis and autoimmune psychosis 13,36,41 .
Using single QD tracking to determine the lateral diffusion of membrane NMDAR exposed to NMDAR-Ab, we previously showed that NMDAR dynamics was upregulated following exposure to autoantibodies 7 .Yet, this approach allows to track, at a given time, only few NMDAR that exchange between the extrasynaptic and synaptic compartment.Here, we implemented another approach, i.e. sptPALM, since it has several key advantages for our specific question.First, sptPALM provides, at a given time, a large number of trajectories in each compartment, which strongly contrast with single QD tracking.Second, the small size of the mEos fluorophore (3-4 nm) favours the access to the synaptic cleft when compared to a QD-antibody complex (up to 30 nm) 42 .Third, commercial anti-NMDAR antibodies used for single QD-NMDAR tracking may theoretically compete with NMDAR-Ab.Furthermore, sptPALM allows to study the impact of short time incubation with autoantibodies since there is no pre-coupling between nanoparticle and antibodies.Thanks to all these properties, it is now clearly demonstrated that the synaptic NMDAR pool is not altered by NMDAR-Ab within the first tens of minutes, or hour, contrasting with previous claims that did not have the required resolution 9,36 .NMDAR-Ab from NMDAR encephalitis patients are classically seen as "cross-linker", based on two series of observations.First, intact immunoglobulins from NMDAR encephalitis patient decrease synaptic NMDAR content whereas Fab fragment from these immunoglobulins fail to do so 37 .However, as exemplified in this study, the affinity of a Fab fragment is lower than that of a full immunoglobulin 40 , complicating the interpretation of the data.In addition, the concentration of NMDAR-Ab and NMDAR-Fab from patients' immunoglobulins is unknown, further limiting our capacity to draw precise outcomes.This drawback has now been circumvented by the generation of monoclonal antibodies from NMDAR encephalitis patients.Indeed, the concentration of NMDAR-Ab/Fab is now perfectly controlled and adapted to optimal experimental setting.Expectedly, we had to increase to 10-fold the concentration of Fab fragments to obtain a staining similar to that of full immunoglobulins, which is consistent with the loss of affinity and avidity of Fab fragments.The second series of evidence supporting a cross-linking effect of NMDAR-Ab arise from our past study showing that some extrasynaptic NMDAR were slowed -down following hours of exposure to purified immunoglobulins from encephalitis patients 7 .This decrease of surface dynamics resemble, to some extent, to that induced by a commercial anti-NMDAR antibody [43][44][45] .However, the surface dynamics of some extrasynaptic NMDAR was increased by immunoglobulins from encephalitis patients 7 , which clearly contrast to the artificial cross-linker 35,46 .Our current study sheds new and unsuspected light.Indeed, we provide direct evidence that exposure of several hours with NMDAR-Ab decrease NMDAR surface diffusion in a cross-linking-independent manner since NMDAR-Fab produce the same effect.Future studies will be necessary to decipher the mechanism underpinning this NMDAR-Fab-induced slow-down of receptor membrane diffusion.Furthermore, whether receptors become cross-linked at late phase of the disorder (days-to-week), and/or whether a cross-linking process alters receptor cycling between membrane and intracellular stores cannot be excluded.These are important issues beyond the NMDAR autoantibodies because patient autoantibodies directed against different neurotransmitter receptors (e.g.glycine and GABAa receptors) can also alter the receptor-mediated ionotropic transmission as well as internalization through cross-linking-independent process [24][25][26]47 . NMDR-Ab induce a massive reorganization of membrane proteins.Although this observation may be expected for NMDAR and closely-related proteins, it is quite remarkable that the whole protein pool is affected following 30 min exposure.Indeed, NMDAR represent only one protein family among more than 900 families identified at the plasma membrane of cultured hippocampal neurons at this developmental stage 48 .One could have thus predicted that perturbation of the NMDAR surface trafficking and organization could go unnoticed within the whole protein surfaceome.However, we clearly show that NMDAR-Ab are sufficient to "de-clusterize" NMDAR, their interactome (SPIN), and the protein surfaceome in the extrasynaptic compartment.We propose that NMDAR-Ab-mediated effect relies on a broad alteration of numerous membrane proteins as in a dominoes sequence.Disrupting solely the interaction between NMDAR and EphB2R was, for instance, not sufficient to provoke the full-blown effect of NMDAR-Ab.Furthermore, a wide range of alteration of other neurotransmitter receptors and signalling cascades could be expected in such a scenario of corrupted extrasynaptic compartment.Consistently, NMDAR-Ab strongly alter AMPA receptor-and GABAa receptor-mediated transmission and synaptic pools in a process involving extrasynaptic protein-protein interactions 49,50 .Further investigations are surely needed to disentangle the alteration of membrane protein organization in autoimmune brain disorders.
Finally, our findings demonstrate that NMDAR-Ab act independently of cross-linking in the first stage of exposition, but act rather as disorganizer of the extrasynaptic compartment in which other neurotransmitter and neuromodulator systems are strongly impaired 49,50 .Our study points toward new therapeutic strategies in which stabilizing NMDAR at the extrasynaptic compartment may be of great interest, displacing the focus of interest from the synapse to the poorly-understood extrasynaptic compartment.Whether other autoantibodies also alter membrane complex interactome emerge thus an intriguing possibility to fully understand (extra)synaptopathies.
3863 and the activity of transfer of biological material is authorized by the Ministry of Research under AC-2019-3595.The subject and her legal guardian have expressed their Downloaded from https://academic.oup.com/brain/advance-article/doi/10.1093/brain/awae163/7675954 by guest on 10 June 2024 written informed consent, and with the authorization of the local ethics committee of Bordeaux University Hospital.

Figure 4
Figure 4 NMDAR surface interactome nano-organization is modified by NMDAR-Ab incubation.(A) Schematic describing the proximity labelling assay using the GluN1-V5-HRP construct.In presence of H2O2 (2µM) and the biotin derivative non-permeant tyramide

Figure 5 NMDARFigure 6
Figure 5 NMDAR-Ab alter the nano-organization of all surface proteins.(A) Experimental design to label all surface proteins using NHS-Ester Alexa 647.Neurons incubated in live with NHS-Ester display surface staining as shown with the line scan intensity.Scale bar = 2 µm.(B) Endogenous staining of Homer and super-resolved image of NHS-Ester staining using SR-Tessler on a portion of dendrite.The red arrows indicate synapses.Scale bar = 5 µm.(C) Quantification of the total number of detections per surfaceome cluster area of neurons incubated for 30 min with Control-Ab or NMDAR-Ab (1 µg/ml) (mean ± SEM, N = 5 neurons).(D-G) Representative clusters of synaptic and extrasynaptic surface proteins after Control-Ab and NMDAR-Ab incubation (1 µg/ml, 30 min) and quantification of clusters area and local density (median ± min to max, **P<0,01, ***P<0,001, ****P<0,0001 by Mann-Whitney, Control-Ab: N = 5 neurons, synaptic : n = 276 clusters and 244 nanodomains, extrasynaptic: n = 597 and 638; NMDAR-Ab: N = 5, synaptic: n = 123 and 144, extrasynaptic: n = 335 and 836).Scale bars = 50 nm.A C C E P T E D M A N U S C R I P T

Figure 7
Figure 7 Targeting only extrasynaptic NMDAR with NMDAR-Ab bead is sufficient to induce synaptic NMDAR loss.(A) Schematic representation of NMDAR-Ab coupled to 1 µm latex bead that can only bind to extrasynaptic NMDAR.(B)The flow cytometer experiment validates the efficiency of the Ab-bead coupling, in the dot plot graph the red population corresponds to non-aggregated latex beads.From this population is extracted the fluorescence of Anti-human 488 antibody incubated with the different bead conditions: latex bead only, Control-Ab bead and NMDAR-Ab bead.NMDAR-Ab bead and Control-Ab bead curves shifts on the right meaning that the coupling was efficient with respectively 95 % and 85 % of the total beads that were effectively coupled to the Ab.(C) Example images of live neurons incubated for 30 min with Control-Ab bead or NMDAR-Ab bead with quantification of the number of beads per area (mean ± SEM, ****P<0,0001 by student t-test, Control-Ab bead: n = 8 neurons, NMDAR-Ab bead: n = 12) (D) Live imaging of neurons expressing Homer-dsRed and GluN1-SEP treated with Control-Ab bead or NMDAR-Ab bead (40 µg/ml, 24 h), scale bar = 5 µm.(E) Quantification of synaptic GluN1-NMDAR clusters density (mean ± SEM, ***P<0.001by student t-test, Control-Ab bead: n = 25 neurons, NMDAR-Ab bead: n = 29) of neurons treated with the different conditions.(F) Schematic representation of