Rationale: Dravet Syndrome (DS) is a neurodevelopmental genetic disorder with severe epilepsy. About eighty percent of DS patients have mutations in the SCN1A gene encoding the voltage-gated sodium channel alpha subunit NaV1.1 which regulates action potential firing in GABAergic inhibitory neurons. The DS phenotype includes hyperactivation of brain activity and is associated with severe febrile and afebrile seizures, neurodevelopmental delay, and sudden unexpected death in epilepsy (SUDEP), which is the major cause of mortality in DS. Predicting SUDEP is crucial for improving the quality of life of DS patients and their families; however, currently, there are no methods for accurately predicting SUDEP in DS.
Methods: Here, we developed a machine learning algorithm to predict SUDEP by collecting and analyzing data obtained using non-invasive behavioral monitoring (motor activity and ultrasonic communication) using Scn1a
+/- mouse model of DS, at a pre-epileptic age. Because previous mouse studies showed that (1) NaV1.1 protein was detected at PND 9 and (2) abnormal brain hyperactivity was observed at PND 16 in DS mice, we hypothesized that behaviors at the pre-epileptic age between PND 9 and 16 would be altered in Scn1a+/- mice, especially in the mice expressing SUDEP.
Results: We found that univariate analysis of parameters from the behavioral analysis at this pre-epileptic age showed no differences between (1) Scn1a
+/- mice and wild-type littermate mice and (2) Scn1a
+/- mice that experienced SUDEP and the ones that survived. However, dimensionality reduction analysis of the behavioral data identified induvial Scn1a
+/- mice that developed SUDEP and those that survived as distinct populations, suggesting there are hidden features that were not apparent in the univariate analysis. Additionally, machine learning-based unsupervised clustering analysis using the pre-epileptic behavioral data enabled us to predict which of the Scn1a
+/- mice developed SUDEP. The accuracy of the prediction was more than 90 % in the Scn1a
+/- mice for model generation and validation.
Conclusions: These results suggest that SUDEP in Scn1a
+/- mice is predictable by observing, measuring, and analyzing certain behaviors. The SUDEP prediction method developed here may be applicable to the other DS mouse models, and importantly, might give a deeper understanding of the molecular mechanisms and biomarkers associated with seizure development and SUDEP in DS.
Funding: This study is supported by the Canadian Institutes of Health and Research.