A Computational Perspective on Brain Dysfunction in Autoimmune Encephalitis Models

Abstract

Autoimmune encephalitides (AEs) are neuroinflammatory disorders associated with seizures, cognitive decline, and sleep disorders ranging from insomnia to degraded sleep architecture. To investigate the symptoms and dynamic changes that autoantibodies cause in AEs, this study investigated and validated a novel rodent passive-transfer model of anti-leucine-rich-glioma-inactivated 1 (anti-LGI1) encephalitis. Juvenile Wistar rats were infused with control, LGI1-antibodies (LGI1-Abs), or anti-contactin-associated protein-like 2 (CASPR2-Abs) antibodies for seven days during a 21-day video-EEG recording period. Intracerebroventricular LGI1-Ab passive-transfer caused spontaneous recurrent seizures following a bimodal distribution. Peaks in seizure frequency occurred acutely at day 1-2, and more chronically at days 5-6. Unilateral tonic forelimb extension, reminiscent of faciobrachial dystonic seizures in patients, was observed in 29% of acute seizures. Seizures were not detected in CASPR2-Ab-treated rats. Support vector machines (SVM) were used to classify sleep-like resting behaviour from EEG data, which yielded an arousal state prediction accuracy of 70-85% when compared against visual classification of combined video-EEG data. LGI1-Ab-treated rats were found to have lower total resting times and shorter continuous periods of rest; CASPR2-Ab-treated rats did not differ from controls. EEG epochs expressing abnormal frequency spectra were also classified using SVM. In LGI1-Ab-treated and CASPR2-Ab-treated rats, periods of rest expressed outlier frequency compositions at a higher rate than controls, though this only held true for LGI1-Ab-treated rats during periods of wakefulness. To examine differences across interictal EEG not captured by standard spectral analyses, the Hurst exponent and Approximate Entropy measures were implemented. Changes in hourly seizure frequency were negatively correlated with the predictability of resting-state EEG among LGI1-Ab-treated rats. Likewise, the entropy of resting-state EEG among LGI1-Ab-treated rats was significantly lower than in controls. Overall, these models were found to have strong face and construct validity and offer a platform for further investigation into the dynamics of AEs.