Autoimmune seizures and epilepsy

Abstract

The rapid expansion in the number of encephalitis disorders associated with autoantibodies against neuronal proteins has led to an incremental increase in use of the term “autoimmune epilepsy,” yet has occurred with limited attention to the physiopathology of each disease and genuine propensity to develop epilepsy. Indeed, most autoimmune encephalitides present with seizures, but the probability of evolving to epilepsy is relatively small. The risk of epilepsy is higher for disorders in which the antigens are intracellular (often T cell–mediated) compared with disorders in which the antigens are on the cell surface (antibody-mediated). Most autoantibodies against neuronal surface antigens show robust effects on the target proteins, resulting in hyperexcitability and impairment of synaptic function and plasticity. Here, we trace the evolution of the concept of autoimmune epilepsy and examine common inflammatory pathways that might lead to epilepsy. Then, we focus on several antibody-mediated encephalitis disorders that associate with seizures and review the synaptic alterations caused by patients’ antibodies, with emphasis on those that have been modeled in animals (e.g., antibodies against NMDA, AMPA receptors, LGI1 protein) or in cultured neurons (e.g., antibodies against the GABAb receptor).

Seizures, epilepsy, and the concept of autoimmune epilepsy

Many disorders can provoke seizures, which are defined as paroxysmal events due to an excessive, hypersynchronous discharge in central nervous system (CNS) neuronal networks (1). These paroxysmal events can manifest with a broad spectrum of symptoms ranging from convulsions and loss of consciousness to barely perceptible behavioral alterations (2). The term “seizures” should be differentiated from epilepsy, which is a chronic brain disorder characterized by an enduring predisposition to generate epileptic seizures (3, 4), and from epileptogenesis, which consists of the formation of a neuronal network where spontaneous seizures occur (5). Fifty million people worldwide are affected by epilepsy, and about one-third have seizures that do not respond to treatment (6).

The idea that some forms of epilepsy could be autoimmune was suggested 119 years ago (7) and reconsidered in the 1960s and 1970s in experiments showing that the infusion of brain-specific antibodies into the ventricles and brain of cats and monkeys resulted in hyperexcitability and epileptiform activity (8). Over the past 20 years, multiple studies have endorsed the hypothesis that inflammatory brain processes involving components of innate immunity play important roles in the pathophysiology of epilepsy (1, 3). Early observations suggesting the involvement of inflammatory and immune processes in epilepsy include the response of some drug-resistant epilepsies to adrenocorticotropic hormone or steroids (9); the presence of T cells and inflammatory molecules in the brains of patients with Rasmussen’s encephalitis, temporal lobe epilepsy, or cortical dysplasia–related epilepsy (10–12); and the link between febrile seizures and an increase of levels of proinflammatory markers (13). Moreover, patients with autoimmune diseases have a higher risk of epilepsy than the general population (14).

In the 1980s and 1990s the identification of several antigen-specific CNS immune responses in a rare group of cancer-triggered disorders named paraneoplastic syndromes showed that autoimmunity against neuronal proteins caused severe forms of encephalitis that were often associated with seizures (ref. 15 and Table 1). These immune responses are mediated by cytotoxic T cells accompanied by antibodies against intracellular neuronal proteins (16); although the antibodies are not pathogenic, they are useful biomarkers of the disease. Given that most of these diseases have a poor outcome and require monitoring and treatment of the associated cancer, their study has generated little interest as potential models of autoimmune epilepsy.

Table 1. Encephalitis with seizures and autoimmune mechanisms.

The concept of autoimmune epilepsy was reinforced in the mid-1990s by the observation that rabbits immunized with the GluR3 subunit of the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR) developed seizures, GluR3 antibodies, and pathological features resembling those of Rasmussen’s encephalitis (ref. 17 and Table 1). Subsequent inconsistencies in the antibody findings (18) and the refractoriness of most patients’ symptoms to immune modulation suggested that an antibody-mediated pathogenesis was unlikely (19). Currently, Rasmussen’s encephalitis is viewed as an antigen-driven MHC class I T cell–restricted attack against neurons and astrocytes in which the self-protein targets or potential viral antigens are unknown (20–22).

Another important step in the field of autoimmune epilepsy (23) was the identification of antibodies against glutamic acid decarboxylase 65 (GAD65; an enzyme involved in the synthesis of GABA) in patients with stiff-person syndrome (ref. 24 and Table 1). This disorder is associated with muscle rigidity with superimposed spasms, and about 10% of patients develop epilepsy. Subsequent studies showed that GAD65 antibodies also occurred in patients with isolated epilepsy, which is often refractory to treatment (25, 26). GAD antibodies are detected at low titers in 1% of healthy people and at high titers in 80% of patients with type 1 diabetes mellitus and other endocrinopathies. Only very high titers of GAD antibodies are associated with neurological disorders (27).

Experience gained from the above studies and from the field of antibody-mediated neuromuscular diseases (e.g., myasthenic syndromes, neuromyotonia) (28) facilitated, in the mid-2000s, the discovery of a group of CNS diseases in which the antibodies alter the structure and function of receptors, ion channels, or interacting proteins (refs. 29, 30, and Table 1). These neuronal antibody–mediated encephalitides or synaptopathies manifest with syndromes that vary according to the antigen (refs. 27, 31, 32, and Table 2). The immunological triggers can be systemic tumors or viral encephalitis (31, 33) but are unknown in many instances; in some diseases, a genetic link to distinct haplotypes has been shown (34–37).

Table 2. Antibody-mediated encephalitis, general clinical features^A.

In antibody-mediated encephalitis, seizures can be the first, the predominant, or, rarely, the only manifestation of the disease (Table 3), providing a solid foundation for the term “autoimmune epileptic seizures.” In these disorders, the extracellular epitopes are accessible to circulating antibodies, and the antibody pathogenicity has been shown in cultured neurons and in vivo models (31). Despite the severity of the symptoms, 70% to 80% of patients with antibody-mediated encephalitis substantially improve or recover with appropriate treatment (32).

Table 3. Antibody-mediated encephalitis, seizures, and estimated risk of epilepsy^A.

The exact frequency of neuronal antibody–mediated encephalitis is unknown. It has been estimated to constitute 10% to 15% of all cases of encephalitis (which have an annual incidence of 5 to 10 cases per 100,000 persons) (38, 39). A study showed that the incidence of the most frequent neuronal antibody–mediated encephalitis, anti–NMDA receptor (anti-NMDAR) encephalitis, surpassed that of herpes simplex encephalitis (40), and another study showed that the incidence of the second most frequent of these diseases, anti-LGI1 encephalitis, is 0.83 cases per 1 million persons (41).

Enduring predisposition to seizures varies in neuron-specific autoimmunity

In 2014, the International League Against Epilepsy (ILAE) established a new definition of epilepsy requiring “two unprovoked (or reflex) seizures occurring >24 hours apart, or one unprovoked (or reflex) seizure and a probability of further seizures similar to the general recurrence risk (at least 60%) after two unprovoked seizures, occurring over the next 10 years” (4). In 2017, a new classification of epilepsy by the ILAE introduced the concept of “epilepsy of immune etiology” for patients whose epilepsy “results directly from an immune disorder in which seizures are a core symptom of the disorder” (2). According to these definitions, the frequent categorization of the seizures occurring in antibody-mediated encephalitis as autoimmune epilepsy would appear to be incorrect. Indeed, seizures are extraordinarily frequent in the acute, inflammatory-provoked phase of many types of antibody-mediated encephalitis, ranging from 33% to 100% depending on the antigen, but in most patients the seizures are not sustained, and resolve after the encephalitis abates (42–50). Young children and toddlers, who are among the most vulnerable patients for epileptogenesis, rarely develop epilepsy after anti-NMDAR encephalitis (51, 52); possible exceptions are patients who develop complications during the course of the disease (e.g., hypoxia, infection). Therefore, the contribution to the development of epilepsy of anti-NMDAR and other antibody-mediated encephalitis is low (refs. 45, 47–49, 53–55, and Table 3).

Two exceptions regarding the risk of epilepsy within the group of antibody-mediated encephalitis disorders are the encephalitis associated with LGI1 antibodies (which is extremely rare in children) (56) and the encephalitis with GABAa receptor (GABAaR) antibodies, which can affect children and adults. A study showed that after a follow-up of 2 years, 14% of patients with anti-LGI1 encephalitis had seizures and an additional 14% were on antiepileptic drugs (57). Early hippocampal lesions and multiple daily seizures favored the development of hippocampal sclerosis.

Experience with anti-GABAaR encephalitis is more limited because of its recent discovery. However, studies indicate that most patients with this disorder present with refractory seizures or status epilepticus (58, 59). In one study, 2 of 21 patients (10%) died of status epilepticus (59), and in another study, 3 of 9 patients (33%) had seizures after 1 year follow-up (58).

The group of antibody-mediated encephalitis disorders should be separated from other types of autoimmune encephalitis in which the antibodies target intracellular proteins, such as the above-mentioned paraneoplastic encephalitis, or GAD65 antibody–associated encephalitis (60). Clinical and laboratory investigations suggest that when the antigens are intracellular, cytotoxic T cells mediate the predominant pathogenic mechanisms (32). The accompanying antibodies against intracellular antigens do not appear to be pathogenic; for example, neurons exposed to GAD65 antibodies do not show surface labeling or antibody internalization (61). Accordingly, the associated syndromes are often refractory to antiepileptic drugs and immunotherapies and markedly contribute to the number of patients with autoimmune epilepsy seen in outpatient clinics (60, 62). For example, among 13 patients who underwent epilepsy surgery for autoimmune epilepsy refractory to treatment, 11 had antibodies against intracellular antigens (eight GAD65, three onconeuronal) and only 2 against surface antigens (one LGI1, one CASPR2) (63). Pathological studies showed chronic lymphocytic infiltrates in 7, gliosis in 5, neuronal loss in 4, and hippocampal sclerosis in 3 (all 3 with GAD65 antibodies).

In paraneoplastic syndromes and probably in anti-GAD65 encephalitis, the infiltrating T cells cause neuronophagia, granzyme B neurotoxicity, neuronal loss, and gliosis, likely favoring epileptogenesis (64, 65) (Figure 1). This is in contrast with findings observed in patients with anti-NMDAR encephalitis that show milder inflammatory infiltrates, limited or absent neuronophagia, more frequent B cell or plasma cell infiltrates, and deposits of antibodies without activation of the complement system (an ancient immune defense system that triggers antigen phagocytosis and membrane attack) (30, 66, 67). For other antibody-mediated encephalitis disorders, pathological studies are limited to anecdotal case reports; a study that included one patient with anti-LGI1 encephalitis and two with antibodies probably directed against LGI1 suggested that complement fixation played a pathogenic role (65).

Figure 1. Paraneoplastic limbic encephalitis and epilepsy mediated by cytotoxic T cell mechanisms.

(A) Coronal fluid-attenuated inversion recovery (FLAIR) MRI image showing increased signal and volume of the right amygdala and hippocampus, suggestive of limbic encephalitis, in a patient with a history of seminoma and acute-onset seizures associated with Ma2 paraneoplastic antibodies. (B) Coronal FLAIR image 1 year later, showing atrophy of the right hippocampus and medial temporal lobe sclerosis. (C) Subtraction ictal SPECT coregistered to MRI (SISCOM) showing increased ictal perfusion over the right hippocampus and parahippocampal gyrus during a right temporal lobe seizure with epigastric aura, piloerection, and loss of awareness. (D) Coronal FLAIR image showing resection of the temporal pole and right mesial temporal lobe structures. After surgery, the frequency of the seizures decreased, but they did not resolve (Engel’s class III). (E) Inflammatory infiltrates in the surgical specimen; the section of the tissue was immunostained with TIA-1 antibody, a marker of cytotoxic T cells (shown as brown granular staining). Some TIA-1–positive cells are in close apposition with neurons (arrows). Scale bars: 10 μm. Images reprinted with permission from Carreño et al. (63).

In summary, the predisposition to cause enduring seizures in autoimmune encephalitis is dependent on the mechanism that drives the immune response, ranging from a high predisposition in cytotoxic T cell–mediated encephalitis (intracellular antigens) to a moderate or absent predisposition in antibody-mediated encephalitis (surface antigens). Among the latter, the severity of the seizures and likelihood to develop epilepsy vary according to the antigen. Additionally, all these disorders occur with a variable degree of inflammation that could have downstream effects on synaptic function, hyperexcitability, and epileptogenesis.

Downstream synaptic targets of epilepsy-related inflammation

Multiple studies indicate that inflammation, and therefore innate immunity, are involved in epilepsy (refs. 1, 62, 68, and Figure 2). In rodents, induction of seizures or status epilepticus triggers rapid recruitment of inflammatory mediators in the regions of seizure activity and propagation (69, 70). During the process of epileptogenesis, which is ignited in experimental models by acquired brain injuries or by mimicking of infections, proinflammatory cytokines (IL-1β, TNF-α, and IL-6) are first expressed in activated astrocytes and microglia, accompanied by changes in cytokine receptor expression in the same cells and in neurons (71, 72). These events are followed by the induction of COX-2 and prostaglandins (PGE2), with upregulation of components of the complement system in the indicated cells (73). Subsequent changes include the production of chemokines and their receptors in neurons and activated astrocytes (69, 74).

Figure 2. Synaptic dysfunction and hyperexcitability as a result of seizures, inflammation, and antibody-mediated encephalitis.

Diagram showing multiple inflammatory/innate immunity mechanisms triggered by seizures and epileptogenesis, along with inflammation-related transcriptional and nontranscriptional pathways that lead to synaptic dysfunction, changes in plasticity, and hyperexcitability (corresponding with blue and red arrows). In contrast to these mechanisms, the antibody-mediated encephalitides such as those associated with NMDAR, AMPAR, LGI1, or GABAbR autoantibodies (see others in Tables 3 and 4), represent a direct antibody-mediated alteration of the corresponding targets also leading to synaptic dysfunction, impairment of synaptic plasticity, and hyperexcitability (purple arrow). The degree of involvement of inflammatory/innate immunity molecules and pathways of inflammation in antibody-mediated encephalitis is currently unknown.

Seizures activate the perivascular glia and a cascade of cytokine-mediated events that lead to involvement of endothelial cells, with upregulation of IL-1β, IL-1R1, complement system, and multiple adhesion molecules (75, 76) that may direct blood leukocytes into the brain and associate with blood-brain barrier (BBB) leakage (75, 77). An increase of vascular permeability to serum albumin affects astrocyte function via TGF-β1 receptor (TGF-βR), altering potassium buffering and the ability of astrocytes to reuptake glutamate, which in turn results in NMDAR-mediated hyperexcitability (78, 79). TGF-β1 signaling in astrocytes induces upregulation of molecules related to extracellular matrix (ECM) remodeling and a persistent breakdown of perineuronal nets around fast-spiking inhibitory interneurons, predisposing to chronic deficits in inhibitory neurotransmission (80).

Conversely, CNS and systemic inflammation predispose to seizure precipitation (81). Two typical examples include febrile seizures, which involve the release of endogenous cytokines, mainly IL-1β within the brain (82), and the experimental model of increase of release of an endogenous “danger signal” molecule named high-mobility group box-1 (HMGB1), which is produced by stressed neurons (83). The interaction of HMGB1 with Toll-like receptor 4 (TLR4), a receptor of innate immunity, constitutes an important proconvulsant pathway and is a key initiator of neuroinflammation following brain injuries leading to epilepsy (84, 85). Indeed, the expression of HMGB1 and TLR4 is increased in human epileptogenic tissue, and clinical and experimental data suggest that HMGB1 isoforms may serve as biomarkers for epileptogenesis and drug-resistant epilepsy (84, 85).

IL-1β, HMGB1, and the corresponding receptors IL-1R1 and TLR4 have downstream effects that converge with the TNF pathway at the transcription factor NF-κΒ, which regulates the synthesis of cytokines and modulates the expression of genes involved in cell death and survival, neurogenesis, and synaptic plasticity (86, 87). A separate nontranscriptional pathway related to IL-1R1 and TLR activation involves Src and other kinase systems that result in phosphorylation of the NMDAR GluN2B subunit and other receptor-coupled or voltage-dependent ion channels, affecting neuronal excitability (88, 89). Genetic and pharmacological animal models have shown that elevated expression of IL-1β and IL-1R1 also increases neuronal excitability by altering GABAergic and glutamatergic neurotransmission (refs. 72, 89–91, and Figure 2).

Antibody-mediated encephalitis disorders also occur with inflammation and seizures, but compared with other diseases and experimental models, they provide a direct mechanism of synaptic dysfunction and hyperexcitability via specific antibody binding to synaptic receptors and proteins (ref. 92 and Figure 2). Although there are no studies available on the above-described inflammatory pathways in antibody-mediated encephalitis, two reports suggest a role of the accompanying inflammatory mechanisms in anti-NMDAR encephalitis. In one study, the level of the B cell–attracting chemokine CXCL13, which is produced in response to activation of several TLRs, was found to be elevated in the cerebrospinal fluid (CSF) of 70% of patients at early-stage disease (93). The authors postulated that the prodromal viral-like process frequently observed in this disorder could be involved in initiating the production of CXCL13; the same study showed that prolonged or secondary elevation of CXCL13 in CSF was associated with limited response to treatment and relapses (93).

The other study focused on patients who after herpes simplex encephalitis developed anti-NMDAR and other autoimmune encephalitis. This complication occurred in 27% of the patients within 2 months after the viral infection had resolved, and the outcome was substantially worse than that reported in classical (not viral-related) anti-NMDAR encephalitis (33). Indeed, 63% of patients aged 4 years or younger and 13% of those older than 4 years had seizures at 1 year follow-up; moreover, 22% of the younger group developed early infantile spasms (33). Brain MRI showed that 82% of the patients had extensive areas of contrast enhancement, which is rare among cases with classical antibody-mediated encephalitis (49, 53, 57, 59). These findings suggested that entry of complement and other proinflammatory molecules through a disrupted BBB could have contributed to epileptogenesis and worse outcome.

Pathogenic models of antibody-mediated encephalitis

In antibody-mediated encephalitis, the coexistence of antigen-specific antibodies with a variable background of inflammation brings into consideration to what extent the antibodies contribute to patients’ symptoms. Findings that suggest a role of the antibodies include (a) the preferential association with distinct clinical syndromes according to antigen specificity (Table 2), sometimes accompanied by different types of seizures, paraclinical findings (EEG, MRI), speed of recovery, and propensity to epilepsy (Table 3); (b) the pathogenic effects of the antibodies in in vitro and in vivo models (Table 4); (c) the resemblance of the antibody-mediated syndromes or mechanisms to those caused by pharmacological or genetic alteration of the same antigens (Table 4); and (d) the frequent clinical response to treatments focused on removing the antibodies or B cells (Table 3).

Table 4. Comparison of the pathogenic effects of autoantibodies with genetic models of target antigen dysfunction.

Target antigens can be subdivided according to structure and function into ionotropic receptors (e.g., NMDAR, AMPAR, GABAaR), metabotropic receptors (e.g., GABAbR), and synaptic linker proteins (e.g., LGI1).

Autoantibodies against NMDAR and AMPAR change neuronal excitability.

Antibodies against NMDAR, AMPAR, or GABAaR have been shown to cross-link and reduce surface expression of the respective receptor in a dose-dependent manner when applied to cultures of neurons (42, 45, 58). This effect was not observed when antibody Fab fragments were used (94). Antibody-mediated receptor internalization starts after 30–120 minutes (92, 95) with maximal internalization at 12 hours of incubation time in vitro, and is reversible upon removal of antibodies (96).

In anti-NMDAR encephalitis, IgG antibodies are selectively directed against the N-terminal domain of the obligate GluN1 subunit of the receptor (97). In cultured neurons, antibody-mediated internalization leads to a reduction of NMDARs and selectively diminishes NMDAR-mediated currents (ref. 94 and Figure 3A). These effects are specific for GluN1 antibodies, and experiments using human monoclonal GluN1 antibodies revealed similar results (98). Studies with super-resolution stochastic reconstruction microscopy (STORM) revealed that NMDAR antibodies induce clustering of NMDAR in nanodomains in synaptic and extrasynaptic areas preceding their internalization. These changes are subunit dependent, preferentially affecting NMDAR containing GluN1 and GluN2B subunits (92). NMDARs of this subunit composition have longer desensitization kinetics (99) and are believed to be important in synaptic plasticity (100). Concordantly, other studies have shown that human NMDAR IgG antibodies lead to slower diffusion of GluN1/GluN2B heterodimers (101). This has been attributed to antibody-induced disruption of the interaction of NMDARs with EphB2R, which stabilizes NMDAR in the postsynaptic membrane (101, 102). These findings were confirmed in a model based on cerebroventricular infusion of patients’ antibodies to mice via osmotic pumps (103, 104). In this model, and also after stereotactic injection of patients’ CSF antibodies, the levels of NMDAR were reduced in the hippocampus, accompanied by severe impairment of long-term potentiation (LTP) and deficits in learning and memory (103, 105, 106) resembling those observed in mouse models of hippocampal deficiency of NMDARs (Table 4). In the same model, stimulation of EphB2R antagonized the effect of patients’ antibodies, thus providing a potential target-specific treatment strategy (31, 101, 105). In addition to reduction of surface NMDAR expression, direct effects of the antibodies on NMDAR channel function may contribute to pathological NMDAR signaling. In single-channel electrophysiological recordings of GluN1/GluN2B–transfected HEK cells, application of patients’ antibodies prolonged the open probability of NMDAR channels (97). Further studies are needed to determine whether acute changes of NMDAR current flow alter neuronal excitability. Studies with cultured neurons showed that NMDAR antibodies similarly influence the receptor density in excitatory and inhibitory neurons accompanied by a reduction of the overall density of inhibitory synapses (96). It is unknown whether similar changes occur in vivo, and whether the altered excitability would be sufficient to cause epileptic seizures. In a passive-transfer mouse model with a single intraventricular injection of patients’ NMDAR antibodies, mice showed an increased susceptibility to develop seizures upon application of the chemoconvulsant pentylenetetrazol (107).

Figure 3. Synaptic alterations and changes in neuronal excitability induced by autoantibodies against neuronal surface antigens.

(A) Top: Patients’ antibodies (blue) against NMDARs bind to GluN1 subunits, inducing NMDAR clustering and dissociation from Ephrin-B2 receptor (EphB2R), followed by NMDAR internalization. Below: Reduction of synaptic NMDARs affects synaptic plasticity, revealed by decreased long-term potentiation (LTP). In each panel, blue traces depict effects of patients’ antibodies, and gray traces show effects of normal human IgG. (B) Top: Antibodies against AMPAR GluA2 subunit induce internalization of GluA2-containing heterodimers after dissociation from TARPs. AMPAR loss is followed by homeostatic compensation with insertion of Ca²⁺-permeable inward-rectifying AMPARs (e.g., GluA1 monomeric AMPAR), which have higher channel permeability. Below: Nonstationary fluctuation analysis shows an increase in AMPAR channel conductance (steeper hyperbola slope) along with reduced channel number (reduced hyperbola width). Current-voltage relationship of excitatory postsynaptic currents (EPSCs) in neurons preincubated with patients’ GluA2 antibodies reveals incorporation of inward-rectifying AMPARs in the synapse. (C) Top: Anti-LGI1 antibodies react with epitopes in leucine-rich repeat (LRR) and EPTP domains of LGI1, disrupting LGI1’s interaction with presynaptic ADAM23 and postsynaptic ADAM22, and reducing presynaptic voltage-gated Kv1.1 channels and postsynaptic AMPARs. Below: Downregulation of presynaptic Kv1.1 channels increases presynaptic release probability and enhances glutamatergic transmission, resulting in increased evoked EPSCs (eEPSCs) and reduced failure rate of synaptic transmission after minimal stimulation (msEPSCs). (D) Top: Anti-GABAbR antibodies bind to the GABAb1 subunit, which localizes at pre- and postsynaptic membranes and contains the GABA-binding site. Antibody binding does not cause GABAbR internalization but interferes with baclofen-induced GABAbR activation. Below: Baclofen blocks spontaneous network activity of cultured neurons (gray). Anti-GABAbR antibodies interrupt its inhibitory effect (blue).

Similar to the NMDAR, AMPARs are excitatory ionotropic glutamatergic receptors and consist of four subunits. AMPARs mediate the majority of fast excitatory synaptic transmission in the CNS and are mostly composed of two GluA1 and two GluA2 subunits. In contrast to NMDARs, there is no obligatory subunit and there is a larger variability in receptor composition (108). Importantly, the presence of GluA2 determines crucial properties of the receptor: RNA editing of the Q/R site of the GluA2 subunit modifies the pore region of the receptor so that AMPARs containing GluA2 are impermeable to Ca²⁺ and show a linear current-voltage relationship (109). In contrast, AMPARs without GluA2 are Ca²⁺ permeable, have a larger single-channel conductance, and are inwardly rectifying, as intracellular polyamines can block the channel pore at positive membrane potentials (109, 110). Patients with anti-AMPAR encephalitis harbor antibodies against either GluA1 or GluA2 subunits, resulting in a reduction of surface levels of AMPAR (45, 111, 112).

A recent study using patients’ antibodies against GluA2 demonstrated a specific antibody-induced restructuring of AMPAR composition by a synaptic scaling–like mechanism (ref. 95 and Figure 3B). This mechanism has been observed in conditions of neuronal silencing and in cell-specific knockout models of AMPAR subunits (Table 4). Patients’ GluA2 antibodies led to internalization of GluA1/GluA2 heterodimeric AMPARs followed by synaptic insertion of inwardly rectifying AMPARs with increased channel conductance. In cultured neurons, confocal and STORM microscopy showed a reduction of GluA2 but not GluA1 subunits. These observations were confirmed in mice after intraventricular and hippocampal transfer of patients’ antibodies. Patch-clamp electrophysiological analyses of ionic current in hippocampal neurons revealed a decrease of the levels of AMPARs, whereas the remaining receptors showed increased single-channel conductance (Figure 3B). Importantly, application of patients’ GluA2 antibodies in GluA1-knockout mice also led to reduced levels of AMPAR, but the replacement with GluA1-AMPARs of higher conductance was no longer present, suggesting that GluA1 homomeric receptors are responsible for the synaptic scaling–like mechanism observed in wild-type mice (95). Interestingly, recent studies showed that in rat models of chronic temporal lobe epilepsy there was a relative increase in inwardly rectifying non-GluA2 AMPARs, which was linked to neuronal excitotoxicity and seizure development (113, 114). Determining whether the rearrangement of AMPAR subunits observed in the model of anti-AMPAR encephalitis results in increased neuronal excitability and enhanced seizure susceptibility is a goal of future studies.

Antibodies against LGI1 induce presynaptic and postsynaptic pathology.

Limbic encephalitis with antibodies against LGI1 is the second most common form of autoimmune encephalitis, resulting in memory deficits and several types of epileptic seizures, which are often preceded by faciobrachial dystonic seizures (31). Binding of autoantibodies against LGI1 cannot induce internalization of the antibody-antigen complex, because LGI1 is a neuronally secreted protein without direct membrane anchoring (115). LGI1 contains three leucine-rich repeats (LRRs) that are flanked by two cysteine-rich regions at the N-terminal, and seven-bladed propeller structures or epitempin (EPTP) repeats at the C-terminus (116). LGI1 forms a trans-synaptic complex that includes the presynaptic disintegrin and metalloproteinase domain–containing protein 23 (ADAM23) (which interacts with Kv1.1 potassium channels) and the postsynaptic ADAM22 (which interacts with AMPARs) (117). Mutations of LGI1 are associated with an inherited form of epilepsy called autosomal dominant lateral temporal lobe epilepsy (ADTLE) that usually presents with acoustic or visual hallucinations and partial seizures (118, 119).

Antibodies of patients with anti-LGI1 encephalitis bind to the LRR and EPTP domains of LGI1 (refs. 120, 121, and Figure 3C). In cultured neurons, these antibodies reversibly decrease postsynaptic clusters of ADAM22 by interfering with the interaction of LGI1 and ADAM22 (120). Using an animal model based on cerebroventricular transfer of patients’ IgG antibodies, a more complex pathophysiology involving pre- and postsynaptic LGI1-dependent signaling has been revealed (121). In the hippocampus of infused mice, total and postsynaptic levels of AMPARs were reduced, confirming previous in vitro findings (120). In addition, the levels of presynaptic Kv1.1 were also decreased, indicating antibody-induced disruption of presynaptic LGI1/ADAM23/Kv1.1 signaling (121). This involvement of presynaptic Kv1.1 channels resulted in increased neuronal excitability with higher presynaptic release probability and reduced synaptic failure rate, leading to increased glutamatergic transmission, which likely enhances the susceptibility to develop seizures (121). An increase of neuronal excitability was also reported in a previous in vitro study using the IgG fraction of a patient with antibodies presumably against LGI1 (122). Moreover, mice infused with LGI1 antibodies developed severe memory dysfunction with concomitant impairment of synaptic LTP. Interestingly, these changes were independent of Kv1.1 signaling, suggesting they were caused by altered postsynaptic AMPAR recruitment induced by patients’ LGI1 antibodies (121).

These functional and molecular findings resemble those obtained in genetic mouse models of LGI1 deficiency or mutations (Table 4), but the clinical features in anti-LGI1 encephalitis are different from those in ADTLE. A lower degree of LGI1 disruption in the autoimmune model along with coexisting inflammatory changes in patients with anti-LGI1 encephalitis could explain some of these differences. It has also been shown that most of the mutated forms of LGI1 related to ADTLE are no longer secreted by neurons (115, 123, 124), indicating fundamental differences in the pathophysiology of the autoimmune and genetic LGI1 models.

Antibodies against the GABAbR are selective GABAb1R antagonists.

Different from NMDAR and AMPAR, the ionotropic receptors for the excitatory neurotransmitter glutamate, the GABAbR is a G protein–coupled receptor for the inhibitory neurotransmitter GABA. The encephalitis with antibodies against GABAbR is associated with early and prominent epileptic seizures (47, 48). GABAbRs are heterodimeric receptors composed of a GABAb1a or GABAb1b subunit together with a GABAb2 subunit. The GABAb1 subunit contains the GABA binding site and determines receptor localization, and the GABAb2 subunit activates the G protein (125). GABAbRs are located mainly at the perisynaptic membrane and can serve as auto- and heteroreceptors, influencing synaptic function in the range of seconds to minutes (126). Genetic models of GABAbR deficiency show several abnormalities in neuronal and synaptic function (Table 4). Antibodies from patients with anti-GABAbR encephalitis bind to several epitopes in the N-terminal region of the GABAb1a and GABAb1b subunits (47, 127). Unlike antibodies targeting ionotropic receptors, patients’ GABAbR antibodies do not induce receptor internalization in cultured neurons; instead, the antibodies interfere directly with inhibitory GABAbR function, as they antagonize the effects of the GABAb1 agonist baclofen (ref. 127 and Figure 3D). Since baclofen usually reduces the frequency of miniature excitatory postsynaptic currents in cultured neurons, this antibody-induced effect is most likely mediated by presynaptic mechanisms (31, 127). These findings suggest a pathogenic mechanism of patients’ antibodies leading to severe refractory seizures. Future studies should assess the pre- and postsynaptic effects of patients’ antibodies in an animal model and whether they alter the regulatory function of GABAbRs in network activity.

Current challenges and future investigations in autoimmune epilepsy

There is a pressing need to clarify the definition of autoimmune epilepsy. It is frequently implied that any disorder with seizures and autoantibodies is autoimmune epilepsy (128–130). Consequently, most autoimmune encephalitides are routinely categorized as autoimmune epilepsy irrespective of the disease provoking the seizures, type of antibody, or definition of epilepsy (128, 131). This extensively used assumption is inaccurate and has led to the development of score systems for antibody prevalence in epilepsy (APE) that are based on the same clinical insights and diagnostic criteria used for antibody-associated encephalitis, resulting in an important selection bias (132). Indeed, patients with multiple symptoms of autoimmune encephalitis have the highest APE scores, whereas those with pure or predominant seizures have the lowest (e.g., faciobrachial dystonic seizures, or drug-resistant temporal lobe epilepsy with GAD65 autoimmunity).

The acute phase of most antibody-mediated encephalitis with seizures can last several months, yet the risk of epilepsy is small (53, 54). Patients with these diseases should have a reasonable follow-up (we propose 1 year) before the diagnosis of epilepsy is considered in those who continue having seizures or need sustained antiepileptic medication. The length of this follow-up has not been previously established and is open to reassessment; however, during this observation period, patients should be considered to have an autoimmune seizure disorder, but not epilepsy. This is important for two reasons: first, a premature diagnosis of epilepsy can lead to unnecessary and prolonged use of antiepileptic medication; and second, according to the ILAE, epilepsy might “resolve” but not be “cured,” thus, it becomes a preexisting condition that confers important socioeconomic implications (4).

A separate problem is the patients with new-onset seizures of unclear etiology who are antibody negative. These include most patients with Rasmussen’s encephalitis, subsets of patients with new-onset refractory status epilepticus (NORSE) (133, 134), and patients with idiopathic seizures and inflammatory CSF findings (Table 1). Without biomarkers of adaptive immunity, a definite diagnosis of autoimmune seizures cannot be established. The presence of CSF pleocytosis or oligoclonal bands and the clinical response to steroids or immune modulation (e.g., plasma exchange) are not reliable indicators of autoimmunity because they can occur in nonautoimmune inflammatory diseases in which the indicated pathways of innate immunity are involved (e.g., interferonopathies) (135), or in disorders of unclear etiology (e.g., seronegative limbic encephalitis or central nervous system vasculitis) (136–138).

Antibodies against astrocytes (glial fibrillary acidic protein [GFAP]) and oligodendrocytes (myelin-oligodendrocyte glycoprotein [MOG]) are associated with meningoencephalomyelitis and neuromyelitis optica spectrum disorders, but in some cases they occur with seizures (refs. 139, 140, and Table 1). Particularly, MOG antibodies are detected in approximately 50% of children with acute disseminated encephalomyelitis, and less frequently in a form of cortical encephalitis with seizures (141, 142). Whereas these antibodies are rarely included in serological screening panels for suspected autoimmune seizures or epilepsy (62), others that are included should be discontinued. For example, antibodies against voltage-gated potassium channels (VGKCs) are not useful biomarkers of brain-specific autoimmunity unless antigen-specific assays demonstrate that the targets are LGI1 or CASPR2 (two proteins complexed to VGKC) (143, 144). This and the fact that many patients with antibody-mediated encephalitis do not develop epilepsy explain why, in epilepsy clinics, the number of cases with genuine autoimmune epilepsy is limited to those with GAD65 antibodies, and even a smaller number of cases with LGI1 or other antibodies (145–147). Although Hashimoto encephalopathy is an ill-defined disorder (148) and the autoimmune mechanisms are unclear, this disorder is often considered in the differential diagnosis of autoimmune epilepsy (refs. 149, 150, and Tables 1 and 5).

Table 5. Differential diagnosis of seizures and epilepsy of suspected autoimmune etiology in children and adults.

A common feature of all types of autoimmune epileptic seizures is the refractoriness to antiepileptic drugs unless immunotherapy is concurrently used (130). It is currently unclear whether some antiepileptics are better than others in patients with these disorders.

A task for the future is to determine whether genetic factors, or variable involvement of inflammatory pathways, may enhance the likelihood of seizures in patients with autoimmune encephalitis. It is also unclear why the spectrum of autoimmune encephalitis is different in children as compared with that in adults (135); in practice, these differences are important because they change the approach to differential diagnosis (Table 5).

The antibody-mediated encephalitides represent a new biomedical frontier, helping to better understand the role of ion channels, receptors, and other synaptic proteins in neurological function and seizures. The associated antibodies can be used to determine how blocking, reducing the levels, or altering the surface dynamics of specific synaptic proteins changes neuronal excitability or synaptic plasticity or can potentially induce seizures. Although several models of antibody pathogenicity have been developed, no animal model of antibody-mediated clinical seizures is yet available. Given that the autoantibodies are frequently synthesized within the CNS (66, 151), treatments designed to remove systemic antibodies are often suboptimal, resulting in protracted clinical courses (53). A better understanding of the physiopathology of these diseases should lead to novel treatment strategies. This is supported by experiments showing that an agonist of EphB2, a tyrosine kinase that regulates excitatory synapse formation (152), was able to antagonize the effect of patients’ NMDAR antibodies (101, 105), or that a positive allosteric modulator of NMDAR accelerated the recovery of NMDAR function in neurons exposed to patients’ antibodies (153). For anti-GABAbR encephalitis, preliminary studies show that activation of the B2 subunit, bypassing the B1-blocking effect of the antibodies, lessens the increased neuronal excitability (127). A current challenge is to extend these types of studies to other antibody-mediated encephalitis, with the goal of developing drugs to be tested in clinical trials.

Version 1. 02/04/2019

Electronic publication

Version 2. 03/01/2019

Print issue publication