Research ArticleENCEPHALITIS

Autoimmune receptor encephalitis in mice induced by active immunization with conformationally stabilized holoreceptors

See allHide authors and affiliations

Science Translational Medicine  10 Jul 2019:
Vol. 11, Issue 500, eaaw0044
DOI: 10.1126/scitranslmed.aaw0044

An active model of autoimmune encephalitis

Autoantibodies targeting the neurotransmitter N-methyl-d-aspartate receptor (NMDAR) have been found in patients with encephalitis. Although treatments reducing antibody titers have been developed, their effect is often incomplete. The lack of models recapitulating the pathophysiology of the human disease hinders the development of effective treatments. Now, Jones et al. developed a mouse model of NMDAR encephalitis using active immunization with intact NMDARs embedded in liposomes. Characterization of the model showed alterations resembling human anti-NMDAR encephalitis, including the presence of hippocampal inflammation, immune cell infiltration, and cognitive impairments. Moreover, the authors showed that the presence of T cell and B cell infiltrates is necessary for the induction of the disease.


Autoimmunity to membrane proteins in the central nervous system has been increasingly recognized as a cause of neuropsychiatric disease. A key recent development was the discovery of autoantibodies to N-methyl-d-aspartate (NMDA) receptors in some cases of encephalitis, characterized by cognitive changes, memory loss, and seizures that could lead to long-term morbidity or mortality. Treatment approaches and experimental studies have largely focused on the pathogenic role of these autoantibodies. Passive antibody transfer to mice has provided useful insights but does not produce the full spectrum of the human disease. Here, we describe a de novo autoimmune mouse model of anti-NMDA receptor encephalitis. Active immunization of immunocompetent mice with conformationally stabilized, native-like NMDA receptors induced a fulminant encephalitis, consistent with the behavioral and pathologic characteristics of human cases. Our results provide evidence for neuroinflammation and immune cell infiltration as components of the autoimmune response in mice. Use of transgenic mice indicated that mature T cells and antibody-producing cells were required for disease induction. This active immunization model may provide insights into disease induction and a platform for testing therapeutic approaches.


Behavioral changes, psychosis, memory impairment, and seizures have been recognized as a pattern in several forms of encephalitis (1). In some cases, this discrete clinical syndrome has been attributed to herpes simplex virus, but the underlying cause in other cases has remained unknown. In 2007, autoantibodies targeting N-methyl-d-aspartate (NMDA) receptors were found in a subset of these patients (2, 3). NMDA receptors are ligand-gated Ca2+ permeable ion channels expressed postsynaptically at excitatory synapses in neurons throughout the central nervous system (CNS). These receptors play critical roles in synaptic plasticity and development, while NMDA receptor antagonists disrupt memory formation and cause schizophrenia-like symptoms (47). Thus, loss of function of these receptors can lead to both structural and functional changes in the brain (810). NMDA receptors are heterotetrameric composed of two obligate GluN1 subunits and two variable GluN2 subunits (GluN2A-D) arranged around a central pore. GluN1/GluN2A/GluN2B triheteromers are the most common subunit combination in forebrain excitatory neurons (4, 11). With recognition of anti-NMDA receptor encephalitis as a clinical syndrome, diagnostic tests have revealed that the disease is relatively common across a broad range of ages (1216). Initially considered as one of the paraneoplastic disorders, it is now clear that many affected patients do not have detectable tumors (12, 17), underscoring our lack of understanding of the disease etiology. Patient-derived serum or cerebrospinal fluid (CSF) indicates a role for antibodies directed at NMDA receptor subunits in the pathogenesis, leading to treatments to reduce antibody titers with plasmapheresis or immunosuppression (15). However, recovery after standard treatments can be prolonged and incomplete (12, 1820).

Passive transfer of antibodies from affected patients to mice has indicated that NMDA receptor antibodies can cause hypofunction in NMDA receptor–mediated synaptic transmission (17, 2123). Yet, the initiating immunological factors and the potential role of immune cell infiltrates and neuroinflammation in the disease process are difficult to determine using existing models (21, 2426). A robust animal model of the disease has the potential to address such issues.

To develop a mouse model of autoimmune encephalitis, we postulated that immunization with fully assembled receptors could be important in triggering the disease. Thus, we used active immunization with intact native-like NMDA receptors composed of GluN1-GluN2B tetramers embedded in liposomes. Subcutaneous injection of these NMDA receptor–containing proteoliposomes induced fulminant encephalitis within 4 weeks in young adult mice. The mice demonstrated behavioral changes, seizures, and histological features of neuroinflammation and immune cell infiltration that were most prominent in the hippocampus. The presence of NMDA receptor antibodies was confirmed by immunohistochemistry (IHC) and Western blot. CD4+ T cell infiltration was an early feature, and both mature T cells and B cells were required for disease induction.


Encephalitis induced by active immunization with NMDA receptor holoprotein

We used purified GluN1/GluN2B NMDA fully assembled tetrameric receptors (holoreceptors) embedded in liposomes (NMDA receptor proteoliposomes) to immunize C57BL6 adult mice (Fig. 1A). The production of highly purified NMDA holoreceptors and the generation of proteoliposomes are described in detail in Methods. Proteoliposomes, such as those used in the current study, are an established tool for investigating the structural and biophysical properties of membrane-bound protein complexes and for facilitating an antigen-specific antibody response (2735). The native-like conformation of NMDA receptors in our proteoliposome preparation was validated in a recent cryogenic electron microscopy study in which the same method was used to resolve the structure of tetrameric NMDA receptors with atomic precision (34).

Fig. 1 Clinical phenotype in proteoliposome-treated mice.

(A) Timeline of treatment and behavioral testing. Adult wild-type mice received subcutaneous injections at days 0 and 15 with proteoliposome (purple) or control [liposome (green) or saline (blue)]. (B) Clinical signs were plotted from the first injection. *P < 0.0001. (C) Clinical signs in proteoliposome-treated mice. (D) Kaplan-Meier survival plot for proteoliposome- and control-treated mice. *P < 0.0005.

Using a standard immunization protocol, mice received a subcutaneous injection of NMDA receptors in proteoliposomes (see Methods for protein/liposome ratio and injection volume), followed by a booster 2 weeks later (36). Littermate control cohorts were injected only with liposomes or with saline. Overt neurological signs began to appear by 4 weeks, and by 6 weeks after immunization, nearly all of the proteoliposome-treated mice (86%) exhibited abnormal home cage behavior (Fig. 1B).The most prominent behavioral changes were hyperactivity (86%), followed by tight circling (50%), overt seizures (21%), and hunched back/lethargy (11%) (Fig. 1C and movies S1 to S5). Cumulative clinical scores (described in Methods) were significantly higher in proteoliposome-treated mice [P < 0.0001 (proteoliposome versus liposome), P < 0.0001 (proteoliposome versus saline), and P > 0.9999 (liposome versus saline); Kruskal-Wallis test with Dunn’s multiple comparisons post hoc; n = 28 per treatment group; Fig. 1B]. Proteoliposome-treated mice also had increased mortality by 6 weeks after immunization (n = 8 of 56 proteoliposomes and n = 0 of 56 controls; P < 0.0005; log-rank test; Fig. 1D). The results were not limited to the specific holoprotein, as we observed a similar distribution of clinical signs in mice immunized with a rat GluN1-GluN2A proteoliposome preparation (50% hyperactivity, 20% circling, 20% seizure, and 10% hunching/lethargy; n = 10; fig. S1A).

To assess behavioral phenotype, we used a battery of standardized tests. Open-field testing (Fig. 2A) confirmed a hyperactive locomotor phenotype with nearly double the distance traveled in proteoliposome-treated mice [proteoliposome = 5472 ± 525.4 cm (n = 26), liposome = 3207 ± 111 cm (n = 28), and saline = 3093 ± 84.9 cm (n = 28); P = 0.0002 (proteoliposome versus liposome), P < 0.0001 (proteoliposome versus saline), and P > 0.9999 (liposome versus saline); Kruskal-Wallis test with Dunn’s multiple comparisons post hoc]. Proteoliposome-treated mice also showed a high degree of variability in the open field ranging from near immobility to extreme hyperactivity. Nest building, indicative of complex stereotyped behavior, was severely compromised in the proteoliposome-treated mice (Fig. 2B). At 6 weeks after immunization, control mice created precise nests, whereas proteoliposome-treated mice barely disturbed the nestlets as indicated by lower nesting scores at 24 and 48 hours, respectively: proteoliposome = 1 ± 0.29 and 1.48 ± 0.39 (n = 27), liposome = 4.60 ± 0.14 and 4.78 ± 0.11 (n = 28), and saline = 4.21 ± 0.14 and 4.75 ± 0.12 (n = 28) [24 hours: P < 0.0001 (proteoliposome versus liposome), P < 0.0001 (proteoliposome versus saline), and P = 0.3603 (liposome versus saline); 48 hours: P < 0.0001 (proteoliposome versus liposome), P < 0.0001 (proteoliposome versus saline), and P > 0.9999 (liposome versus saline); Kruskal-Wallis with Dunn’s multiple comparisons post hoc]. In the zero maze, often used as a measure of anxiety-like behavior, proteoliposome-treated mice spent more time in the normally aversive open area [percentage of time in the open area: proteoliposome = 27.90 ± 3.70 (n = 18), liposome = 16.08 ± 0.71 (n = 28), and saline = 16.32 ± 0.96 (n = 28); P = 0.0065 (proteoliposome versus liposome), P = 0.0081 (proteoliposome versus saline), and P > 0.9999 (liposome. versus saline); Kruskal-Wallis with Dunn’s multiple comparisons post hoc; Fig. 2C]. There was no statistical difference in the total distance moved in the open or closed areas, indicating that hyperactivity could not explain the observed phenotype [open area: proteoliposome = 175.4 ± 26.18 cm, liposome = 118.7 ± 9.41 cm, and saline = 130.4 ± 15.67 cm (P = 0.0786, proteoliposome versus liposome; P = 0.0629, proteoliposome versus saline; and P > 0.9999, liposome versus saline); closed area: proteoliposome = 572.6.4 ± 27.29 cm, liposome = 494.8 ± 17.83 cm, and saline = 524.1 ± 22.17 cm (P = 0.1140, proteoliposome versus liposome; P = 0.4312, proteoliposome versus saline; and P > 0.9999, liposome versus saline); Kruskal-Wallis with Dunn’s multiple comparisons post hoc; Fig. 2C]. These results indicated that immunization with NMDA receptor holoprotein induces a behavioral phenotype compatible with encephalitis based on the criteria used to diagnose anti-NMDA receptor encephalitis in humans (13).

Fig. 2 Behavioral assessment.

(A) Representative movement traces in the open field from a control mouse (field outlined in green) and two proteoliposome-treated mice (field outlined in purple) to examine hyperactivity. Total distance moved in the treatment groups are plotted on the right: proteoliposome (purple), liposome (green), and saline (blue) purple. (B) Representative images of a nest created by a control mouse compared to nests in two proteoliposome-treated mice. Nests were scored at 24 and 48 hours and quantified as shown on the right. (C) Mice were assessed in the zero maze for time spent in the open area (left) and for the total distance moved in the open and closed areas. **P < 0.0001.

Neuroinflammation and peripheral leukocyte infiltration

The histopathology of reported cases of human anti-NMDA receptor encephalitis is heterogeneous and can include immune cell infiltrates, neuroinflammation, and occasionally neuronal loss (2426, 37). In proteoliposome-treated mice at 6 weeks after immunization, hematoxylin and eosin (H&E)–labeled perivascular cuffing was observed in multiple CNS regions including the hippocampus and neocortex of five of the six mice examined [perivascular cuffing measure: proteoliposome = 0.83 ± 0.17 and control = 0 ± 0; P = 0.0152 (proteoliposome versus control), Mann-Whitney test; n = 6 per group]. Representative images of perivascular cuffing are shown in Fig. 3A (right). Patchy areas of cell death were observed in brain sections in one of the six mice included in the histological analysis [cell death measure: proteoliposome = 0.17 ± 0.17 and control = 0 ± 0; P > 0.9999 (proteoliposome versus control), Mann-Whitney test; n = 6 per group]. Figure 3B (right) shows areas of cell death in a proteoliposome-treated mouse. No evidence of inflammation or cell death was present in assessed H&E controls (Fig. 3, A and B, left). Individual data points from the H&E assay and all subsequent experiments with an n < 20 are included in table S1.

Fig. 3 Histological assessment.

(A) Representative images of perivascular cuffing (right, black arrowhead) in hippocampal and neocortical tissue from a proteoliposome-treated mouse. Matched samples from liposome-treated controls are shown on the left. (B) Representative tissue sections were used to examine for areas of cell loss including karyolysis (single white arrowhead) and pyknosis (double white arrowhead) in a proteoliposome-treated mouse (right) compared to control (left). Scale bars, 100 μm.

Immunolabeling with glial fibrillary acidic protein (GFAP) and Iba1 revealed an inflammatory response in assessed proteoliposome-treated mice at 6 weeks after immunization (Fig. 4, A and B, right) as measured by total GFAP immunoreactivity in the hippocampus [proteoliposome = 42.5 × 108 ± 8.80 × 108 and control = 6.33 × 108 ± 1.64 × 108; P = 0.0022 (proteoliposome versus control), Mann-Whitney test; n = 6 per group]. Iba1 labeling revealed foci of microgliosis compared to the tiling of diffuse labeling in controls as expected for healthy mice (Fig. 4B). Total Iba1 immunoreactivity in the hippocampus of these mice was 60.9 × 108 ± 14.3 × 108 in proteoliposome-treated mice compared to 28.5 × 108 ± 1.64 × 108 in controls (P = 0.0022, Mann-Whitney test; n = 6 per group).

Fig. 4 Glial cell labeling in proteoliposome-treated mice.

(A) Immunofluorescence for GFAP (white) in proteoliposome-treated mice (right) compared to controls (left). The higher magnification insets from the hippocampus show individual astrocytes. The histogram shows quantification of GFAP immunofluorescence (see Methods). (B) Labeling with the microglial marker Iba1 (white) in proteoliposome-treated mice (right) compared to controls (left). Higher magnification inset shows microglia in the hippocampus of proteoliposome-treated mouse compared to control. Immunofluorescence was quantified for Iba1 as for GFAP. Scale bar, 1000 μm (inset, 100 μm). **P = 0.0022.

By 6 weeks after immunization, there was also pronounced CNS infiltration by peripheral immune cells in the proteoliposome-treated mice assessed by IHC. Labeling with the pan-leukocyte marker CD45R was robust in the hippocampus, striatum, thalamus, amygdala, and neocortex of these mice (Fig. 5, A and C). Control mice included in this assay showed sparse CD45R labeling [P = 0.0022 (proteoliposome versus liposome), Mann-Whitney test; n = 6 per group, for all anatomical regions analyzed; Fig. 5, A and C]. Peripheral immune cells including activated macrophages/microglia (Galectin3+), plasma cells (CD138+), helper T cells (CD4+), and B cells (CD20+) were increased in the brains of assessed proteoliposome-treated mice as well, whereas cytotoxic T cells (CD8+) were sparse or absent (Fig. 5B, right). In contrast liposome- or saline-treated mice assessed by IHC lacked immunoreactivity to the same immune cell markers [P = 0.0022 (proteoliposome versus liposome), Mann-Whitney test; n = 6 per group, for all immune cells analyzed; Fig. 5, B and C]. Mean cell densities for control- and proteoliposome-treated mice included in the preceding analyses are summarized in Table 1, with individual values provided in table S1.

Fig. 5 Immunohistochemical labeling of CNS immune cell infiltrates.

(A) Labeling with the pan-leukocyte marker, CD45R, in coronal sections from a proteoliposome-treated mouse (right) and control mouse (left). Insets show labeling of individual CD45R+ cells (white) from the indicated region of the hippocampus. CD45R+ cells in controls (left). Scale bar, 1000 μm (inset, 100 μm). (B) Immunohistochemical labeling for a battery of immune cell markers in brains of proteoliposome-treated mice and control mice (CD8+, CD4+, CD20+, CD138+, and Gal3+). Proteoliposome-treated mice showed immune cell infiltrates as indicated with the insets from the hippocampal region. Scale bar, 500 μm (inset, 200 μm). (C) Left: CD45R+ density (cells/μm3) in striatum (Str), cortex (Ctx), amygdala (Amyg), hippocampus (Hipp), and thalamus (Thal) in proteoliposome-treated (black) and control mice (gray). Right: CD8+, CD4+, CD20+, CD138+, and Galectin3+ cell densities (cells/μm3) in the hippocampus of proteoliposome treated (black) and control mice (gray). Only Gal3+ cells were of sufficient density to be apparent above zero on the histogram in control mice. **P = 0.0022.

Table 1 Immune cell densities and quantification of immune cell infiltration.

Mean CD45R+ cell densities across sampled anatomical regions in control- and proteoliposome-treated mice at 6 and 3 weeks after immunization (±SEM). Mean immune cell subtype densities in the hippocampus of control- and proteoliposome-treated mice at 6 and 3 weeks after immunization (±SEM).

View this table:

Although mice examined at 3 weeks after immunization did not show prominent clinical features, neuroinflammation, and immune cell infiltration could be detected in some animals at this early time point (Table 1 and fig. S2, A to E). Figure S2 (B to E) shows representative images and quantification of glial markers and immune cell infiltrates. Quantification of GFAP and Iba1 immune reactivity at 3 weeks after immunization showed no significant differences between treatment groups despite evidence of glial nodules in proteoliposome-treated mice [GFAP fluorescence intensity: proteoliposome = 34.74 × 108 ± 8.85 × 108 and liposome = 19.12 × 108 ± 1.63 × 108 (P = 0.2857, proteoliposome vesus liposome; Mann-Whitney test; n = 5 per group); Iba1 fluorescence intensity: proteoliposome = 40.56 × 108 ± 6.44 × 108 and liposome = 22.66 × 108 ± 1.08 × 108 (P = 0.0952, proteoliposome versus liposome; Mann-Whitney test; n = 5 per group); fig. S2B, right]. Immune cell quantification at 3 weeks after immunization revealed significantly increased CD45R+ and CD20+ cells in the hippocampi of assessed proteoliposome-treated mice [P = 0.0286 and P = 0.0476, proteoliposome versus liposome, for hippocampus (Hipp) CD45R+ (n = 4 per group) and Hipp CD20+ (n = 5 per group), respectively, by Mann-Whitney test; fig. S2, C to E]. Table 1 and table S1 contain means and individual values, respectively, for all immune cell quantifications at the 3-week time point.

Receptor autoantibodies in serum from proteoliposome-treated mice

To assess the specificity of the immunoglobulin G (IgG) isolates from the serum of proteoliposome-treated mice, we used human embryonic kidney (HEK) 293 cells to express and stain for NMDA receptor subunits. HEK cells were transfected with single subunit constructs or combinations of GluN1/GluN2A and GluN1/GluN2B. Using this HEK cell assay, we tested all serum/IgG isolates from proteoliposome-treated mice included in the behavioral and IHC analyses. We observed colabeling of NMDA receptor–expressing HEK293 cells using a commercially available GluN1 antibody and IgG derived from all proteoliposome-treated mice at 6 weeks after immunization, with no colabeling seen in serum/IgG isolates from a randomly selected group of controls [proteoliposome = 1 ± 0 and control = 0 ± 0; P < 0.0001 (proteoliposome versus control), Mann-Whitney test; n = 26 per group]. Figure 6A (bottom) shows representative images of HEK cells coimmunolabeled with a GluN1 antibody and IgG derived from a proteoliposome-treated mouse at 6 weeks after immunization. We used the same HEK cell assay to assess the serum from a cohort of mice at 3 weeks after immunization (fig S3). At the 3-week time point, all but one proteoliposome-treated mouse showed colabeling with a GluN1 antibody, indicating the presence of antibodies before we observed overt clinical signs [proteoliposome = 0.80 ± 0.20 and control = 0 ± 0; P = 0.0476 (proteoliposome versus control), Mann-Whitney test; n = 5 per group]. Figure S3A shows representative HEK cell colabeling for a GluN1 antibody and IgG from a proteoliposome- and control-treated mouse at 3 weeks after immunization. To ensure that the colabeling observed in our HEK cell assay was specific for the GluN1 subunit, we repeated the staining in HEK cells expressing only the GluN1 subunit with a subset of serum/IgG samples from each treatment condition. Again, IgG derived from each proteoliposome-treated mouse included in the assay colabeled with a GluN1 specific antibody at 6 weeks after immunization [proteoliposome = 1 ± 0 and control = 0 ± 0; P = 0.0286 (proteoliposome versus control), Mann-Whitney test; n = 4 per group; fig. S4A]. Serum samples derived from all mice treated with rat GluN1-GluN2A proteoliposomes also showed GluN1-specific colabeling, with no colabeling observed in controls at 6 weeks after immunization [proteoliposome = 1 ± 0 and liposome = 0 ± 0; P < 0.0001 (proteoliposome versus liposome), Mann-Whitney test; n = 10 proteoliposomes and 8 liposomes; figs. S1C and S4B (HEK GluN1-GluN2A and HEK GluN1 only, respectively)]. Individual values from all HEK293 cell assays are shown in table S1.

Fig. 6 Detection of NMDA receptor subtype–specific serum antibodies and localization of IgG binding to NMDA receptors.

(A) HEK293FT cells transfected with rat GluN1/2A subunits were labeled with anti-GluN1 antibody (left, top and bottom; green) or with IgG from proteoliposome-treated mice (bottom middle, red). Merge panel shows colocalization (bottom right, yellow). IgG derived from control mice (top middle, red) showed no labeling. Scale bar, 15 μm. (B) Dendrites of cultured hippocampal neurons with the expected punctate pattern of synapses using an anti-GluN1 antibody (left, green) compared to IgG from proteoliposome-treated mice (bottom middle, red). The merge image is shown at the bottom row (right, yellow). IgG from liposome-treated mice (top middle) compared to puncta observed with anti-GluN1 (top left and right, green). Dendritic shafts were labeled with anti-MAP2 antibody (gray). Scale bar, 5 μm. (C) Pattern of immunoreactivity in hippocampal tissue sections for a commercial NMDA receptor antibody (GluN2A) in an untreated wild-type mouse (top) compared to labeling using purified IgG derived from a proteoliposome-treated mouse (bottom) and purified IgG derived from a liposome-treated mouse (middle). Scale bar, 500 μm. (D) A battery of recombinant green fluorescent protein (GFP)–tagged NMDA receptor (NMDAR) subunits from Xenopus (XI) and rat (R) was blotted and detected with anti-GFP antibodies (top, red). Incubation with serum (1:100) derived from a liposome-treated mouse (middle, green). Serum (1:100) derived from a proteoliposome-treated mouse showed bands corresponding to GluN1 subunit isoforms in Xenopus and rat (Xl.GluN1-3a, R.GluN1-1a, and R.GluN1-1b), Xenopus GluN2B, and Xenopus GluN1 lacking the ATD domain (XI.NR1-3a ΔATD). Blots are from representative liposome-treated control mice and proteoliposome-treated mice with clinical signs of disease.

In cultured mouse hippocampal neurons [days in vitro (DIV) 14 to DIV 21], only the IgG isolate from proteoliposome-treated mice colocalized with a GluN1 subunit–specific antibody in IgG samples derived from proteoliposome-treated mice at both the 6- and 3-week time points [neuron ICC (immunocytochemistry): P < 0.0001 (proteoliposome versus control) and P = 0.0476 (proteoliposome versus liposome), at 6-week (n = 26 per group) and 3-week (n = 5 per group) time points, respectively, by Mann-Whitney test]. The punctate labeling that we observed along dendrites and at dendritic spines follows the expected distribution of NMDA receptors at synapses (4, 8, 38, 39). Figure 6B shows colabeling of proteoliposome-derived IgG (6 weeks after immunization) and a GluN1 antibody along a dendritic shaft. Figure S3B shows the same colabeling for proteoliposome IgG from the 3-week time point.

Hippocampal NMDA receptors are triheteromeric, composed of GluN1, GluN2A, and GluN2B subunits, and are widely distributed in the neuropil (40, 41). As a demonstration of the tissue distribution of proteoliposome-derived IgG labeling, we examined IgG staining pattern in naïve mouse brain sections as compared to the pattern of a GluN2A subunit–specific antibody, which was effective in these floating tissue sections (6 weeks after immunization; Fig. 6C). In the samples that we tested in this assay, purified IgG (red) from liposome controls showed no labeling, whereas proteoliposome-derived IgG (red) showed the same staining pattern as the NMDA receptor antibody (green) [proteoliposome = 1 ± 0 and control = 0 ± 0; P = 0.0286 (proteoliposome versus control), Mann-Whitney test; n = 4 per group]. Staining for mouse IgG as a proxy for the presence of autoantibodies outlined the hippocampi of mice included in this assay (fig. S5), consistent with the expected high levels of NMDA receptor expression in hippocampus and the IgG deposits observed in anti–NMDA receptor encephalitis (proteoliposome = 1 ± 0 and control = 0 ± 0; n = 3 per group) (2, 21).

To confirm the NR1 labeling in the HEK293 cell assays from each mouse, we used serum from two proteoliposome-treated and two liposome-treated mice at 6 weeks after immunization to examine bands on Western blots. Bands corresponding to purified recombinant rat and Xenopus GluN1 subunit protein, as well as Xenopus GluN2B, were observed. Although a putative pathogenic epitope on the GluN1 amino-terminal domain (ATD) has been identified in some human cases, immunoreactivity to GluN2A and GluN2B subunits has been also reported in a subset of cases (2, 17, 42, 43). For the mouse shown in Fig. 6D, serum also labeled a Xenopus GluN1 subunit that lacked the ATD domain, suggesting the presence of polyclonal antibodies in at least some of the mice. Serum from control-treated mice included in Western blot did not recognize NMDA receptor subunits (proteoliposome = 1 ± 0 and liposome = 0 ± 0; n = 2 per group; Fig. 6A, middle).

Serum from proteoliposome-treated mice did not acutely block NMDA receptor function, as assessed by whole-cell currents in cultured hippocampal neurons (Fig. 7, A and B). NMDA (50 μM) was coapplied by local flow pipes either with serum from liposome-treated mice or with serum from proteoliposome-treated mice (1:100 dilution). The NMDA-evoked current in the presence of serum from proteoliposome-treated mice was 95.9 ± 6.8% of that evoked by the combination of NMDA and serum from liposome-treated mice in the same neuron [P = 0.23, paired t test (n = 8); P = 0.2231 (proteoliposome serum) and P = 0.1413 ( liposome serum), Shapiro-Wilk normality test]. In contrast, a 24-hour incubation with serum from proteoliposome-treated mice reduced synaptically activated NMDA receptors, which underlie the slow components of excitatory postsynaptic currents (EPSCs) and drive overall network activity. As shown in Fig. 7C (top left), the slow components of EPSC barrages from neurons incubated in serum from liposome-treated mice were reduced by the NMDA receptor antagonist, D-AP5, as indicated by the rapid decay of the spontaneous EPSCs (sEPSCs; Fig 7C, top right). However, after 24-hour incubation in serum from proteoliposome-treated mice, sEPSCs had reduced NMDA receptor–mediated currents and thus were less sensitive to block by D-AP5 (Fig. 7C, bottom right). For neurons incubated in serum from liposome-treated mice, total charge from sEPSCs was reduced to 44.1 ± 7.9% of control charge by D-AP5 (Fig. 7D). In contrast, D-AP5 reduced total charge to only 85.6 ± 6.0% of control charge in neurons incubated with serum from proteoliposome-treated mice [P < 0.005, paired t test (n = 8 per group); P = 0.3893 (proteoliposome serum) and P = 0.5722 (liposome serum), Shapiro-Wilk normality test; Fig. 7D]. These results demonstrate a marked reduction in NMDA receptor function after 24-hour incubation with serum from proteoliposome-treated mice. We also stained the same 24 hour–treated cultures for postsynaptic density-95 protein (PSD-95), a key structural component and marker of dendritic spines at excitatory synapses, and GluN1 to assess their colocalization (Fig. 7E) (44). The 24-hour incubation did not affect total synaptic puncta but did result in a decrease of >50% in GluN1 immunoreactivity (Fig. 7E), consistent with the reduction in functional measures of NMDA receptor activity [PSD-95 puncta per micrometer: proteoliposome = 0.54 ± 0.019 and liposome = 0.62 ± 0.059 (P = 0.4244, Mann-Whitney test; n = 8 per group); NR1 and PSD-95 puncta per micrometer: proteoliposome = 0.18 ± 0.034 and liposome = 0.56 ± 0.064 (P = 0.0003, Welch’s t test); P = 0.8788 (proteoliposome) and P = 0.9482 (liposome), Shapiro-Wilk normality test; n = 8 per group]. Thus, antibodies generated in proteoliposome-treated mice can lead to NMDA receptor hypofunction after a 24-hour exposure.

Fig. 7 Electrophysiological effects in neurons after acute and chronic exposure to serum from proteoliposome-treated mice.

(A) Representative whole-cell currents in a neuron evoked by acute flow-pipe application of NMDA and serum from liposome-treated or NMDA and serum from proteoliposome-treated mice, respectively. Serum dilution was 1:100. (B) Quantification of evoked whole-cell NMDA currents in neurons after acute serum application. n.s., not significant. (C and D) sEPSCs were recorded after 24-hour incubation (“chronic”) in serum from liposome or proteoliposome-treated mice. (C) Representative sEPSC traces in cells treated with serum from liposome-treated before and after D-AP5 application (top, left and right traces, respectively) and in cells treated with proteoliposome-derived serum (bottom, left and right traces, respectively). (D) Quantification of D-AP5–induced reduction in sEPSCs shown in histogram. ***P < 0.005. (E) Immunocytochemical labeling of dendrites, PSD-95, and GluN1 puncta in cultured hippocampal neurons after 24-hour incubation in serum from liposome- and proteoliposome-treated mice. PSD-95 immunoreactivity (left, top and bottom; red). GluN1-labeled puncta from control- and proteoliposome-treated mice (middle, top and bottom, respectively; green). Overlap of GluN1 and PSD-95 immunolabeling (right, top and bottom; yellow). Dendrites are labeled with anti-MAP antibody (gray). Scale bar, 10 μm. Quantification of PSD-95– and GluN1-positive synaptic puncta per micrometer of dendrite shown on the left and right histograms (liposome, gray; proteoliposome, black). Quantification of PSD-95– and GluN1-positive synaptic puncta per micrometer of dendrite shown on the left and right histograms (liposome, gray; proteoliposome, black). ***P = 0.0003.

T cells and B cells are necessary for proteoliposome-induced encephalitis

Studies of anti–NMDA receptor encephalitis in human cases have focused on the role of antibodies (2, 21, 22). Our results in proteoliposome-treated mice are consistent with the presence of B cell infiltrates and NMDA receptor autoantibodies. In human cases, perivascular T cell infiltrates, primarily CD4+ helper T cells, have been reported; however, parenchymal infiltrates appear to be infrequent (21, 24). To distinguish the roles of B cells and T cells in proteoliposome-treated mice, we used two well-characterized mutant mouse lines that lack either mature B cells or mature T cells (45, 46). Consistent with a role for B cells in the pathophysiology, proteoliposome treatment of MuMt mice, which lack the capacity to generate an antigen-specific antibody response, showed no behavioral or histological abnormalities at 6 and 12 weeks after immunization [GFAP fluorescence intensity: proteoliposome = 51.00 × 108 ± 2.80 × 108 and liposome = 60.08 ± 3.15 × 108 (P = 0.0572, proteoliposome versus liposome; Welch’s t test; n = 6 per group); Iba1 fluorescence intensity: proteoliposome = 24.81 × 108 ± 5.81 × 108 and liposome = 25.81 ± 5.81 × 108 (P = 0.2520, proteoliposome versus liposome; Welch’s t test; n = 6 per group); CD45R+: P > 0.9999 (Mann-Whitney test; n = 6 per group); cell-based colabeling assays: P > 0.9999, proteoliposome versus control (Mann-Whitney test; n = 6 per group); fig. S6]. To evaluate the role of mature T cells, we used Tcrα (“CD4, CD8 cell KO”] mice, which lack mature T helper cells and cytotoxic T cells. All proteoliposome-treated Tcrα mice survived to 12 weeks without clinical signs of disease (Fig. 8A). Likewise, Tcrα mice did not show histopathological evidence of gliosis or cell infiltrates, and serum lacked detected anti–NMDA receptor antibodies [GFAP fluorescence intensity: proteoliposome = 60.47 × 108 ± 4.08 × 108 and liposome = 59.35 ± 4.39 × 108 (P = 0.8573, proteoliposome versus liposome; Welch’s t test; n = 4 per group); Iba1 fluorescence intensity: proteoliposome = 21.97 × 108 ± 4.03 × 108 and liposome = 20.07 ± 3.77 × 108 (P = 0.8573 and P = 0.5282, proteoliposome versus liposome; Welch’s t test; n = 4 per group); CD45R+: P > 0.9999, proteoliposome versus liposome (Mann-Whitney test; n = 4 per group); cell-based colabeling assays: P > 0.9999, proteoliposome versus control (Mann-Whitney test; n = 4 per group); Fig. 8, B to D]. Individual data and group statistics for all data shown in the B cell and T cell mutant analyses can be found in table S1. A cohort of immunocompetent wild-type mice immunized in parallel with the MuMt and Tcrα showed the expected phenotype, indicating the potency of the immunogen. These results not only confirm the expected role of B cells but also indicate a requirement for mature T cells in disease pathogenesis.

Fig. 8 Response of T cell mutant mice to proteoliposome treatment.

(A) Tcrα mutant mice were immunized in parallel with a cohort of wild-type mice. Graphs show clinical observations and mortality rates by 12 weeks after immunization (liposome, green; proteoliposome, purple). (B) CD45R (top), GFAP (middle), or Iba1 (bottom) immunolabeling in proteoliposome- and liposome-treated Tcrα mice. Scale bar, 500 μm. (C) Immunocytochemical colabeling of GluN1 and purified IgG from control- and proteoliposome-treated Tcrα mice in hippocampal neuronal cultures (top) and HEK cells expressing GluN1-GluN2A subunits (bottom). The upper two rows show GluN1-positive puncta (left, green) along a dendrite (MAP2, gray) or lack of staining for IgG from liposome-treated mice (top middle, red) or IgG from proteoliposome-treated mice (bottom middle, red). Scale bar, 5 μm. The lower two rows show anti-GluN1 antibody labeling of HEK293FT cells expressing rat GluN1/2A subunits (left, green) and absent labeling for IgG from proteoliposome- or liposome-treated mice (bottom two, middle). Scale bar, 15 μm.


Our results demonstrate that active immunization with NMDA receptor holoproteins induces a disease state in mice that recapitulates the core features of human anti–NMDA receptor encephalitis including the presence of pathogenic anti-GluN1 autoantibodies (13). In general, active immunization to develop animal models has played an important role in the study of neurological disorders including autoimmune diseases such as myasthenia gravis and multiple sclerosis (47, 48). For example, peptide fragments from myelin have long been used to generate experimental autoimmune encephalomyelitis with its incumbent clinical signs to test the molecular basis and therapeutic approaches in multiple sclerosis (47, 49, 50). Likewise, active immunization with neuromuscular junction proteins can cause myasthenic-like features in mice (51). Although synaptic membrane proteins have been implicated in autoimmune encephalitis, studies of these diseases have largely been limited to passive transfer approaches (1). Our de novo autoimmune anti–NMDA receptor encephalitis model represents an additional approach to examining the pathophysiology and developing treatments for the human disease.

The etiology in some cases of encephalitis was an enigma for decades. The discovery of NMDA receptor antibodies in a subset of these patients not only was a surprise but also provided an opportunity for a better understanding of the causes and treatment strategies as well (17). The diagnostic criteria used to identify anti–NMDA receptor encephalitis (13) include a number of features that were also present in our mouse model including behavioral changes, movement abnormalities, seizures, and the presence of antibodies against the GluN1 subunit of NMDA receptors. The clinical signs and histopathology ranged from mice with marked behavioral impairments to occasional mice that lacked clinical signs but had histological features and anti–NMDA receptor antibodies. As shown in the HEK293 cell assays, all proteoliposome-treated mice had antibodies to GluN1 at 6 weeks after immunization, but in those mice tested by Western blot, we also saw immunoreactivity to GluN2B or to a construct that lacked the ATD of GluN1. This pattern suggests the possibility of a polyclonal response by the time fulminant symptoms were present (6 weeks after immunization). This issue and the possibility of epitope spreading will be important to examine in the future. However, consistent with the human disease, a GluN1 epitope was predominant.

Antibodies recognizing NMDA receptor subunits have been observed in some other contexts. For example, systemic lupus erythematosus (SLE) is associated with a range of antibodies to nuclear antigens and antibodies recognizing the NMDA receptor GluN2A subunit (52, 53). However, unlike the NMDA receptor antibodies found in our mice and in patients with human anti–NMDA receptor encephalitis, anti-GluN2A/B antibodies derived from patients with SLE do not appear to alter synaptic responses but rather induce cell death (52, 53).

Although histopathological reports are available for only a small fraction of human cases of anti–NMDA receptor encephalitis, immune cell infiltrates and neuroinflammation have been observed in some patients (21, 24) but were less prominent in other cases (26). The presence or absence of inflammatory cell infiltrates thus likely depends on the time point at which the tissue is examined. Neuroinflammation in our mice was most prominent during the fulminant stages of the disease (6 weeks after immunization compared to 3 weeks). In the human disease and in our mouse model, cytotoxic T cell involvement can occur but was not a prominent feature in mice included in this assessment (2426, 37). In addition, consistent with human cases, IgG infiltration of the hippocampus could be seen in proteoliposome-treated mice in which tissue slices were incubated with anti-IgG antibodies (2, 21).

Previous experimental studies of anti–NMDA receptor encephalitis used passive transfer approaches with patient-derived CSF/IgG (2123, 54). These studies provide compelling evidence for the involvement of NMDA receptor antibodies in pathogenesis (22, 23, 43, 54). Intraventricular infusion of human NMDA receptor antibodies in mice decreased NMDA receptor density in the hippocampus (23, 55), but these mice did not exhibit movement disturbances or spontaneous seizures (23, 5557). These observations suggest that passive infusion of NMDA antibodies is not sufficient to fully mimic the clinical syndrome. Although no animal model is expected to perfectly mimic a human disease, our results indicate that active immunization of immunocompetent mice with NMDA receptor holoproteins replicates several signs of human encephalitis. The active immunization model also allows an examination of the time course of the disease process from the point of induction.

The current hypothesis for the clinical phenotype is that antibody-mediated internalization of NMDA receptors leads to hypofunction at a network level. The available evidence suggests that antibodies from human cases cause internalization of NMDA receptors in vitro and reduce NMDA responses (21, 22, 58). Thus, therapies in use or proposed have been directed at removing antibodies, inhibiting immunosuppression, or preventing receptor internalization (1). The hypofunction hypothesis gains further support from the behavioral side effects observed with the use of NMDA receptor antagonists, but general NMDA receptor hypofunction does not easily account for the presence of unprovoked, spontaneous seizures given the presence of NMDA receptors on both excitatory and inhibitory neurons and the complexity of the circuits involved (6, 22, 56, 59, 60). More experiments will be necessary to the relative contribution of receptor internalization to neurological signs and whether other factors contribute to the spectrum of clinical signs and histopathology.

Although previous studies of anti–NMDA receptor encephalitis have largely focused on the role of B cells and antibodies, the role of CD4+ T cells in autoimmune encephalitis is an area of growing interest (61). T cells could promote neuroinflammation and potentiate B cell– and plasma cell–mediated antibody responses. In human cases, CSF cytokine/chemokine profiles support a role for CD4+ T cell involvement (6265). For example, interleukin-17 (IL-17), a proinflammatory cytokine produced by T helper 17 CD4+ T cells, is prominent in the CSF of human cases and may perform the dual role of blood-brain barrier disruption and up-regulation of IL-6, a pro–B cell and plasma cell cytokine (62, 66, 67). Although cytotoxic T cells were not a prominent histological feature in our mice, the absence of disease in the Tcrα mice lacking mature CD4+ or CD8+ cells supports an important role for some population of T cells in disease pathogenesis. Thus, therapies designed to reduce T cell–mediated inflammation, such as blocking IL-17 signaling, as well as those aimed at reducing B cell activation, are worthy of further investigation.

The trigger for an autoimmune reaction to NMDA receptors remains unclear. Immunization using peptide fragments to produce NMDA receptor antibodies has not resulted in reports of clinical disease. In a recent study, mice immunized with NMDA receptor peptides did not show clinical signs, but the authors suggested that blood-brain barrier integrity may prevent circulating NMDA receptor antibodies from entering the CNS (68). The immunogens in our case were the tetrameric Xenopus laevis GluN1/GluN2B or rat GluN1/GluN2A receptor in native-like heteromeric assembly (69). The X. laevis subunits had been altered to maximize protein stability, by removal of the intracellular C termini (69), and were capable of binding glutamate or glycine, which may also have improved protein stability. The rat subunits that we used had no mutations except for removal of the intracellular C termini and thus unaltered topology in the ATD, one of the putative sites of pathogenic antibody interactions in human cases (42, 43). The use of holoprotein immunogens with intact extracellular domains likely played a role in the high incidence of disease in our mice and may be relevant considerations for other membrane proteins implicated as causes of encephalitis. Our results suggest that disease induction depends on conformationally restricted epitopes. This idea is consistent with previous studies in HEK293 cells, which showed that an assembled NMDA receptor was necessary for reactivity of antibodies from human cases (2, 17). In human cases, the source of intact NMDA receptors to trigger the autoimmune response is unknown, but it could be ectopic expression from a tumor or membrane debris after an insult causing neuronal cell loss. For example, anti–NMDA receptor encephalitis has been reported after viral infection (12, 70). An association with viral infection was also suggested in the case report of Knut, a polar bear at the Berlin zoo (71).

Despite the prevalence of anti–NMDA receptor encephalitis, many experimental questions remain unanswered given the lack of a de novo autoimmune animal model that recapitulates the signs and symptoms. The mouse model described here provides such a platform and has already provided several new insights. For example, the use of conformationally stabilized holoproteins appears to be a critical component of immunogenicity, and our results already indicate a complex pathogenesis. Thus, the initial steps in disease induction, the roles of specific immune components, and potential new therapies can now be tested.

In the future, it will be interesting to examine immunized mice at earlier time points to look for specific memory deficits, more subtle cognitive impairments, and the evolution and atomic localization of autoantibody epitopes. We did not examine some other aspects that have been reported in the human disease such as autonomic dysfunction or the cause of death in some of our mice. Furthermore, we did not investigate aberrant electrophysiological activity at the network and synaptic level in vivo or in brain slices derived from affected mice. However, these questions are all addressable using this mouse model, which offers significant advantages for further exploration of such issues.


Study design

We examined the effect of active immunization with NMDA receptor native-like holoproteins on normal adult mice. The aim of the study was to investigate autoimmunity to NMDA receptors in the context of anti–NMDA receptor encephalitis. Littermate controls (liposome or saline) of both sexes were used for all interventions. Results from liposome and saline controls did not differ and were thus combined for statistical analysis of “control” in some experiments as indicated. Criteria were established in advance on the basis of the pilot studies for issues including data inclusion, outliers, selection of end points, and sample size (see the “Statistics” section). All analyses were blinded. Observations within each animal were averaged, and the value for n replicates reflects the number of animals. Detailed methods for each experimental technique and analysis are included in the Supplementary Materials including animal use, NMDA receptor expression and purification, NMDA receptor proteoliposome preparation and immunization, behavioral assessments, histology and IHC, serum collection/IgG purification and Western blots, in vitro assays and IHC, and electrophysiology and quantification of synaptic puncta.


Sample size was determined on the basis of previous experiments of this type, with an effect size of 20% and a power of 0.8. Tests of normality were used to determine the appropriate test. Multiple comparisons used one- or two-way analysis of variance (ANOVA) or nonparametric ANOVA as indicated for each experiment. Exact P values are provided, and both the number of animals and the number of observations are indicated as appropriate. Data unless otherwise indicated are plotted as means ± SEM. All statistical analyses were completed using Prism 7 software (GraphPad).



Fig S1. Rat GluN1-GluN2A holoprotein and disease induction.

Fig S2. Neuropathological assessment at 3 weeks after immunization.

Fig S3. IgG immunolabeling at 3 weeks after immunization.

Fig S4. GluN1 subunit–specific immunocytochemistry in HEK cells.

Fig S5. IgG deposits in CNS at 6 weeks after immunization.

Fig S6. Response of B cell mutant mice to proteoliposome treatment.

Table S1. Individual values for all analyses with n < 20.

Movie S1. Representative liposome-treated control mouse in home cage.

Movie S2. Proteoliposome-treated mouse with locomotor hyperactivity.

Movie S3. Proteoliposome-treated mouse with aberrant circling.

Movie S4. Proteoliposome-treated mouse during a clinical seizure.

Movie S5. Proteoliposome-treated mouse with prominent hunched back and lethargy.


Acknowledgments: We thank the OHSU histology core for processing of samples, R. Woltjer for advice concerning interpretation of the histology, J. Raber and S.W. Boutros for assistance with the behavioral assessments, G. Marracci and P. Chaudhary for advice on immune cell markers, W. Sun for wild-type NMDA receptor subunit constructs, and F.J.-Y. for NMDA receptor amino acid sequence alignments. Funding: This work was supported by NS080979 and the Ellison Medical Foundation to G.L.W. and the NINDS imaging core facility (P30NS061800) and NS038631 to E.G. E.G. is an Investigator of the Howard Hughes Medical Institute. Author contributions: B.E.J. and G.L.W. conceptualized the study and performed the experimental design. A.G. (Gouaux Lab) prepared the Xenopus NMDA receptor proteoliposome and performed the Western blot analysis. F.J.-Y. (Gouaux Lab) prepared the rat NMDA receptor proteoliposome. K.R.T. conducted the electrophysiological assessments. N.J.O. assisted with MuMt mouse IHC analysis. B.E.J. completed all other experiments. E.G. provided material support for the preparation of NMDA receptor proteoliposomes. Competing interests: All authors declare that they have no competing interests. Data and materials availability: All summary data and individual measurements are included in the paper and associated supplementary figures and tables. All materials and methods are included in the Supplementary Materials.

Stay Connected to Science Translational Medicine

Navigate This Article