Abstracts

Rationale: A central challenge in epilepsy research is to understand epileptiform propensity across time. While seizures are observed to cluster in certain behavioral states, quantified state definitions and underlying circuit mechanisms are not known. As a result, tailoring treatments to brain states is not yet possible. It has been shown that continuous fluctuations between multiple substates in wakefulness and sleep can be captured using pupil size¹. The fact that epileptiform activity is theorized to “hijack” normative circuit dynamics² that control brain state begs the question: how do epileptiform discharges covary with ongoing fluctuations in state? Understanding the relationship between epileptic networks and internal brain state dynamics will lead to better quality of life for patients through improved seizure forecasting and closed-loop neuromodulation. In this work, we characterize epileptiform propensity in heterozygous Scn8a mice head-fixed on a cylindrical treadmill, using pupil size as a continuous index of brain state.

Methods: Head-fixed Scn8a mice were monitored for 1–2-hour sessions with surface EEG, continuous infrared pupillometry, and encoder for treadmill movement. Pupil size was measured with a deep learning pipeline and normalized to the 99th percentile of max pupil size in each session³. Spike wave discharges (SWDs) were detected with a custom algorithm and then manually reviewed. For each session, SWD probability given a pupil state (p(SWD|State)) was derived from Bayes’ Theorem and equals the SWD probability (p(SWD)) multiplied by the pupil state probability given SWD (p(State|SWD)) and divided by the state probability (p(State)). The p(SWD) = the number of SWDs / the session time. The p(State|SWD) = the mean pupil size pre-SWD / the number of SWDs, and the p(State) = the mean pupil size across the session / the cumulative number of states. Mean pupil sizes were divided into 10% bin widths from 0 to 120%. All probabilities were converted to a percentage. We assumed that pupil changes during SWDs as state changes and were thus included in pupil size calculations. This assumption remains to be further assessed.

Results: SWD median duration was 4.2s (IQR25-75: 2.69,6.17s, N=10 mice, n=4299 SWDs, n = 1-4 sessions/mouse). There was strong coupling between pupil diameter and increased walking velocity. SWDs occurred predominantly during still periods (figure 1) with a maximal probability at mean pupil diameters of 20-30% (p < 0.001, one-way ANOVA) (figure 2).