Monoclonal humanized monovalent antibody blocking therapy for anti-NMDA receptor encephalitis

Abstract

Anti-NMDA receptor (NMDAR) encephalitis is a devastating disease with severe psychiatric and neurological symptoms believed to be caused by pathogenic autoantibodies that bind to the N-terminal domain (NTD) of the NMDAR GluN1 subunit (GluN1-NTD) crosslinking adjacent NMDARs and driving their internalization. Here we describe ART5803, a humanized monovalent antibody, as a potential therapy for anti-NMDAR encephalitis. ART5803 binds with a high affinity (K_D = 0.69 nM) to GluN1-NTD without affecting NMDAR activity or inducing internalization. ART5803 blocks NMDAR internalization induced by patients' pathogenic autoantibodies, and restores NMDAR function. A marmoset animal model was developed using sustained intracerebroventricular (ICV) administration of a human pathogenic autoantibody to evoke behavioral and motor abnormalities. ART5803 ICV infusion or peripheral injections rapidly reversed these abnormalities. These data, together with the pharmacokinetic profile in cynomolgus monkeys, indicate a therapeutic potential for intravenous (IV)-administered ART5803 as a fast-acting and efficacious option for anti-NMDAR encephalitis.

Fig. 1. Generation of ART5803, a humanized monovalent antibody, and its binding characterization to the GluN1-NTD.

a The proposed mechanism of action for the pathogenic autoantibody (bivalent, two-armed, red) and therapeutic antibody (monovalent, one-armed, blue). Bivalent pathogenic anti-NMDAR autoantibodies bind to GluN1-NTD and crosslink adjacent NMDARs, which triggers receptor internalization. Monovalent therapeutic antibodies bind to GluN1-NTD, block binding of pathogenic autoantibodies, and prevent crosslinking and internalization of NMDARs. b ART5803 schematic design. Variable regions (antigen binding sites) are shown in blue. “Hole” mutations (T366S / L368A / Y407V) are introduced in the CH3 domain of the parental antibody heavy chain. A “Knob” mutation (T366W) was introduced in the CH3 domain of the second Fc fragment. LALA mutations (L234A / L235A) are introduced in the CH2 domains to avoid antibody-dependent cell-mediated cytotoxicity (ADCC), antibody-dependent cell-mediated phagocytosis (ADCP) or complement-dependent cytotoxicity (CDC). c Competitive ELISA between biotinylated #003-102 Ab and ART5803, #003-102 Ab and control Ab on GluN1-NTD. Data are mean ± SD of n = 2 internal replicates. Source data are provided as a Source Data file. d Epitope mapping by X-ray co-crystallography. Ribbon model of #003-102 Ab (Fab’) (red) on NMDAR (green). NTD terminal domain, LBD Ligand Binding Domain. e The zoomed-in view of the Fab binding site of #003-102 Ab (red) at the R2 lobe of GluN1-NTD (green) as determined by X-ray crystallography. Interacting residues for #003-102 Ab are from the Vh domain except where indicated with VK. #003-102 Ab residues are labeled with the Kabat numbering system. f Space filling model of GluN1-NTD of NMDAR (green) with #003-102 Ab (red) epitope by X-ray crystallography and #003-102 Ab (pink) and ART5803 (blue) epitopes as determined by HDX-MS.

Fig. 2. ART5803 blocks NMDAR internalization and NMDAR hypofunction driven by pathogenic autoantibody (#003-102 Ab).

a HEK293 cells expressing human NMDAR GluN1 and GluN2B subunits, were used in an NMDAR internalization assay to assess antibody effects on GluN1 surface expression by flow cytometry. Representative histograms of data used to generate Fig. 2b, show changes in geometric mean fluorescence intensity (GeoMFI) of allophycocyanin fluorescence (APC-A) normalized for visual comparison. Dotted lines indicate 100% GluN1 surface expression. GluN1/GluN2B expressing cells were treated for 2 h with either #003-102 Ab, ART5803 or control antibody (0-10 µg/mL). b NMDAR internalization by #003-102 Ab, ART5803 and isotype control at 2 h post-treatment. Data are mean of n = 4 independent replicates ± SEM. c NMDAR internalization by ART5803, bivalent (two-armed) ART5803 and Fab ART5803 at 2 h post-treatment. Data are mean of n = 4 independent replicates ± SEM. d ART5803 block (#003-102 Ab + ART5803 co-incubation for 2 h) and rescue (#003-102 Ab pre-treatment for 2 h, ART5803 post treatment for 2 h) of #003-102 Ab induced NMDAR internalization. Data are mean of n = 3 independent replicates ± SEM. e Assessment of ART5803’s agonistic or antagonistic effects on NMDAR was performed using a Fluorometric Imaging Plate Reader (FLIPR) assay to measure intracellular Ca²⁺ levels with a calcium-sensitive fluorescent dye in HEK293 cells expressing human NMDAR GluN1 and GluN2B subunits. In agonistic activity assessments, ART5803, human IgG isotype control antibody, MK-801, or assay buffer only were added, and fluorescence data was acquired over 5 min. Sequentially for antagonistic activity assessments, NMDA or assay buffer only was added, and fluorescence data was acquired over 5 min. Data are mean ± SEM of n = 3 independent replicates and were analyzed by unpaired, two-tailed Student’s t-test. ** p < 0.01, ns = not significant. RFU relative fluorescence units. f Blocking activity of ART5803 on NMDAR hypofunction (Ca²⁺ influx reduction) induced by #003-102 Ab. Cells preincubated with control antibody and subsequently treated with control antibody (conditions on the left side of graph) were either stimulated with assay buffer only or NMDA to demonstrate baseline activity and maximum activation, respectively. Data are mean ± SEM of n = 3 independent replicates. Statistical analysis were performed using unpaired, two-tailed, Student’s t-tests. *p < 0.05, ** p < 0.01. RFU relative fluorescence units. Source data is provided as a Source Data file.

Fig. 3. ART5803 rescued NMDAR internalization and spine shrinkage caused by pathogenic autoantibody (#003-102 Ab) in mouse hippocampal neurons.

a Schematic of experimental paradigm. Super-ecliptic pHluorin (SEP)-tagged NMDAR is fluorescent at the surface (green) but is quenched when internalized (gray). Created in BioRender. b–d Images of dendrites from CA1 pyramidal neurons in hippocampal slice cultures transfected with SEP-GluN2A (green), GluN1 (unlabeled), and TdTomato (red) directly prior to (baseline) and after (3 h and 14 h) incubation with (b) pathogenic #003-102 Ab, or (c) ART5803, or (d) pathogenic #003-102 Ab alone followed by ART5803 at 3 h. The scale bar is 2 µm. e, f Reduction of (e) surface NMDAR expression and (f) spine size driven by pathogenic #003-102 Ab (left, red; 12 cells) was rescued by the addition of ART5803 at 3 h (right, red and purple; 9 cells). ART5803 alone (middle, blue; 7 cells) had no impact on surface NMDAR expression or spine size. Each point is from an individual spine, and bars represent mean ± SEM. Paired ordinary one-way ANOVA with Tukey’s multiple comparisons test. *p < 0.05; **p < 0.01, ***p < 0.001, #p = 0.1, ns = not significant. Source data are provided as a Source Data file.

Fig. 4. ICV infusion of ART5803 reverses abnormal behaviors induced by pathogenic autoantibody (#003-102 Ab) in marmoset disease model.

a Experimental design of ART5803 intracerebroventricular (ICV) infusion efficacy study. Continuous ICV infusion of pathogenic #003-102 Ab (10 µg/h) into the third ventricle in marmosets for 4 weeks. Two weeks after ICV infusion of #003-102 Ab, continuous ICV infusion of ART5803 (n = 9) or control antibody (n = 3) at 10 µg/h was administered alongside the ICV infusion of #003-102 Ab for an additional 2 weeks. Over the course of the study, abnormal behaviors were evaluated using the abnormal rating scale. b Time course of abnormal behaviors (Abnormal Rating Scale, maximum score 22) in marmosets with ART5803 ICV infusion at Day 0, Day 14, and Day 28. Each point is from an individual marmoset (n = 9) and bars represent mean ± SEM. Statistical analysis was performed between timepoints using two-tailed, Wilcoxon matched-pairs signed rank test. **, p < 0.01. c Comparison of abnormal behaviors between ART5803 and control antibody ICV-infused groups at Day 0, Day 14 and Day 28 of the study. Each point is from an individual marmoset and data are mean ± SEM of n = 3 (#003-102 Ab + Control antibody) or n = 9 (#003-102 Ab + ART5803). Statistical analysis was performed using unpaired, two-tailed, Student’s t-tests. *, p < 0.05; ns = not significant. Source data are provided in Supplementary Table S5.

Fig. 5. IP injections of ART5803 reverse abnormal behaviors induced by pathogenic autoantibody (#003-102 Ab) in marmoset disease model.

a Experimental design of ART5803 intraperitoneal (IP) injection efficacy study. Continuous ICV infusion of pathogenic #003-102 Ab (10 µg/h) into the third ventricle in marmosets for 3 weeks. Six days after ICV infusion of #003-102 Ab, marmosets showing robust abnormalities were divided into two treatment groups (Grouping on Day 6), ART5803 (n = 8) and control vehicle (n = 7) and dosed with either 400 mg/kg ART5803 or vehicle via IP twice a week for 2 weeks starting on Day 7. Over the course of the study, abnormal behaviors were evaluated using the abnormal rating scale. b Time course (days) of abnormal behaviors in marmosets with vehicle or ART5803 IP injections at Day 0, Day 6, Day 14 and Day 21. Each point is from an individual marmoset in the vehicle treated group (n = 7) and ART5803 treated group (n = 8), and bars represent mean ± SEM. Statistical analysis was performed using two-tailed, Wilcoxon matched-pairs signed rank test. *, p < 0.05; **, p < 0.01; ns = not significant. c Comparison of abnormal behaviors between ART5803 and vehicle IP injected groups at Day 0, Day 6, Day 14 and Day 21 of the study. Each point is from an individual marmoset and data are mean ± SEM of n = 7 (Pathogenic antibody + Vehicle) or n = 8 (Pathogenic antibody + ART5803). Statistical analysis was performed using unpaired, two-tailed, Student’s t-tests. *, p < 0.05; ns = not significant. Source data are provided in Supplementary Table S7.

Fig. 6. IP injections of ART5803 reverse GluN1 reduction induced by pathogenic autoantibody (#003-102 Ab) in marmoset disease model.

a Representative immunostaining patterns of human IgG, with anti-human IgG, 3,3’-diaminobenzidine (DAB) staining as brown, and nuclei hematoxylin staining as blue, in sagittal brain sections of marmosets without treatment (treatment naïve control) or treated with either #003-102 Ab ICV infusion at 10 µg/h for 4 weeks (#003-102 Ab ICV) or ART5803 IP injections at 400 mg/kg twice a week for 2 weeks (ART5803 IP). The treatment naïve control marmoset brain serves as negative background control. Slides were scanned at 40x magnification. The scale bar is 5 mm. FC = frontal cortex. b Comparison of GluN1 (NMDAR) and GluR2 (AMPA receptor) levels by immunostaining in the frontal cortex of marmosets without treatment (treatment naïve control, n = 3) or treated with #003-102 Ab ICV plus vehicle IP injections (n = 7) or ART5803 IP injections (n = 8), quantified using DAB positive superpixel segmentation in QuPath. Data are represented as mean ± SEM. Statistical analysis was performed using unpaired, two-tailed, Student’s t-tests. *p < 0.05; ns = not significant. Representative heat maps demonstrating DAB intensity, with black indicating low signal, and yellow, high signal. Scale bars are 5 mm. Source data are provided as a Source Data file. c Comparison of GluN1 (NMDAR), GluR2 (AMPA receptor), PSD95 (synapse marker) and GAPDH levels by western blot analysis in the cerebral cortex (whole) of marmosets without treatment (treatment naïve control, n = 3) or treated with #003-102 Ab ICV plus vehicle IP injections (n = 7) or ART5803 IP injections (n = 8). Quantitative comparisons between samples on different blots were normalized to the treatment naive control for total protein. All samples were from the same experiment and blots were processed in parallel. Data are represented as mean ± SEM. Statistical analysis was performed using two-tailed, unpaired Student’s t-tests. *p < 0.05; ns = not significant. Uncropped blots are reported in Supplementary Fig. S10.

Fig. 7. ART5803 blocks NMDAR internalization induced by patient-derived polyclonal autoantibodies and patients’ sera and CSF in NMDAR-expressing HEK293 cells.

a Using HEK293 cells expressing human NMDAR GluN1 and GluN2B subunits, the NMDAR internalization assay evaluated the effects of antibodies on GluN1 surface expression, measured by flow cytometry. Recovery of surface GluN1 expression after incubation with a patient monoclonal antibody mixture for 18 h, followed by ART5803 co-incubation for an additional 12 h. Data are mean ± SD of n = 2 internal replicates. b Effect of ART5803 (blocking, co-incubation) on Patient 2-6 CSF-like monoclonal antibody mixtures induced internalization for 24 h. Data are mean ± SD of n = 2 internal replicates. c NMDAR binding activity of patients’ sera and CSF samples to HEK293 cells expressing human NMDAR GluN1 and GluN2B subunits, measured by flow cytometry (502 Ab equivalent pseudo-concentration). Each ID correlates to a sera/CSF matched patient. R² and p values were calculated by simple linear regression (n = 6; data presented as mean). Simple linear regression best fit line is solid, and the 95% confidence bands are dotted. Patient #2 was excluded from this graph due to the lack of serum data. CSF = cerebrospinal fluid. d The effect of ART5803 (blocking, co-incubation) on NMDAR internalization induced by patients’ sera and CSF for 20 h. Each graph is titled with a unique patient identifier and is matched between serum and CSF. Serum dilutions of 3% and 10% are indicated by ^ and ^^ respectively. CSF dilutions of 10% and 40% are indicated by * and ** respectively. Red bars indicate no treatment, and blue bars indicate treatment with ART5803. The mean ± SD of n = 2 internal replicates is represented for all samples except for serum #1, which the mean ± SD from n = 2 independent experiments. Source data are provided as a Source Data file.

Fig. 8. Cynomolgus monkey PK and population PK (PopPK) modeling with human scaling and simulation.

a Mean ( ± SD) ART5803 overlay of serum and CSF concentrations in cynomolgus monkeys following a single IV dose at 100 and 1000 mg/kg (n = 3 each dose). SD = standard deviation, CSF = cerebrospinal fluid, BLQ = below limit of quantification. b Simulated ART5803 concentrations in human CSF via IV weekly dosing based on cynomolgus monkey PopPK modeling. The model-predicted CSF concentration–time profiles after IV administration are depicted (0.1 – 100 mg/kg every 7 days). Approximate doses greater than 40, 60 and 100 mg/kg were required to achieve target concentration 0.6 – 2.5 µg/mL in CSF (the top light green band). Source data are provided as a Source Data file.