Abstracts

Rationale: Developmental and Epileptic Encephalopathies (DEEs) are complex, multifaceted disorders; a deeper understanding of their underlying disease mechanisms is likely to be required for the development of disease-modifying therapies. When focusing on genetic haploinsufficiency, we see that relatively modest effects of protein dysfunction at the cellular level may lead to significant alterations at the neuronal network- and behavioral-level. Thinking of these effects as adding up linearly is not sufficient to explain the severe phenotype they produce, often because not all neuronal populations play the same role in network regulation, e.g. SCN1A missense mutation affects only a limited subset of neurons but has profound impact in network activity¹.

Over the years, several readouts of neuronal network activity have been employed to study the effect of DEE-related mutations, from invasive manipulations in preclinical model to more translatable measures such as quantitative-EEG. In this study we look at the final output of neuronal network activity by focusing on behaviors and their fundamental components. Our work leverages advanced machine learning algorithms to analyze long-term video data of mice, aiming to identify behavioral biomarkers beyond seizures and assess the impact of potential gene therapies.

Methods: Videos of controls, acquired epilepsy models and mice carrying DEE-related KO (SLC6A1, STXBP1) were analyzed by combining 2 open-source models: DeepLabCut² for key point estimation and keypoint-MoSeq³ for extracting behavioral syllables in an unsupervised manner. Next, we validated and analyzed the resulting behavioral syllables to extract group-level differences between WT, mice carrying DEE-related mutations. Videos were acquired in accordance with the European Community’s Council Directive (2010/63/EU).


Results: The pipeline successfully identified a range of behavioral syllables (short movement sequences) which revealed distinct behavioral fingerprints between WT and murine models of epilepsy using 15-minute videos. Comparing WT and disease groups, we found that diseased mice exhibited more (hyper)active syllables and that WT mice showed less within-group variability of their behaviors. Leveraging our ability to automate the analysis, we are extending our study to additional DEE models to compare syllable distribution between different etiologies.


Conclusions: Our automated approach to characterizing mouse behaviors can reveal behavioral difference between WT and murine models of genetic epilepsy. We believe this to be the first step in developing a potentially translatable biomarker that goes beyond seizures and takes a more holistic approach to DEEs’ study and therapy. Importantly, this approach may facilitate the development and the assessment of disease-modifying treatments and improve clinical outcomes in DEEs.


Funding: The work was funded by the Walloon Region, Belgium