S62. Computational diagnosis of ion channel dysfunction in NMDA receptor antibody encephalitis

Abstract

Introduction NMDA receptor (NMDAR) antibody encephalitis is an autoimmune disorder causing acute psychosis, confusion and seizures. Electroencephalography (EEG) findings are usually non-specific, with diagnosis confirmed by demonstration of serum/CSF autoantibodies. Improved clinical outcomes depend on early institution of immunotherapies. A non-invasive measure of NMDAR function could facilitate early treatment, and enable tracking of disease activity. Recent biophysical modeling advances allow inference of ion channel dynamics from macroscopic measures of brain activity. Here, we show that clinically relevant diagnoses of NMDAR ion channel dysfunction can be obtained using a microcircuit model fitted to individual human EEG data. Methods We compared resting state EEG cross-spectra from patients with NMDAR encephalitis, encephalopathy unrelated to NMDAR dysfunction, and non-encephalopathic controls. We fit biophysical models (dynamic causal models – DCMs) of connected neuronal ensembles, simulating post-synaptic membrane potentials and EEG spectra through lead field projection. We used canonical variate analysis (CVA) and an empirical Bayesian approach searching candidate models across the groups to reveal specific synaptic deficits. Results Clinical severity was matched between encephalopathic groups. However, there were clear differences in EEG cross-spectra at low (2–4 Hz) and high (28–40 Hz) frequencies, with increased beta-gamma power present in NMDAR antibody encephalitis patients. Subject-specific DCMs mapped model parameters representing dynamics at NMDA, AMPA and GABA receptors to spectral data features. Four inter-connected sources described delta-theta and beta-gamma cross-spectra observed at scalp sensors. Each source comprised a neural mass model with four neuronal subpopulations, with dynamics determined by intrinsic connections mediated via ionotropic receptors. Dynamics along intrinsic and extrinsic connections could reproduce scalp-based EEG recordings. Parameter estimate CVA showed significant differences between patients with and without NMDAR encephalitis in NMDAR parameters. A parametric empirical Bayes (PEB) analysis demonstrated that the group effects of NMDAR encephalitis were specific to NMDA model parameters, in contrast to other encephalopathies where differences were seen across receptor subtypes. Conclusion Neurological and psychiatric practice lack diagnostic probes to assess disease mechanisms at the neuronal level non-invasively in humans. We show that biophysical models of EEG applied to clinical EEG data identify abnormal NMDAR signaling in patients with NMDAR encephalitis but not in patients with other encephalopathies or neurological controls. By comparing model parameter estimates from these patient groups, we selectively identify NMDAR dysfunction in a multi ion-channel model. Given that EEG is ubiquitously available, our findings suggest the feasibility of EEG neuromodeling as a unique assay of brain dysfunction at the molecular-level.