Emergence of high-connectivity states before epileptic seizures: Multi-patient validation, physiological correlates, and network modeling

Epilepsy affects approximately 50 million people worldwide, making it one of the most prevalent neurological disorders. Predicting seizures through reliable pre-seizure biomarkers could greatly enhance neuromodulation therapies for drug-resistant patients. Recent research using stereo-electroencephalography (sEEG) has revealed specific changes in network dynamics during pre-seizure periods. In particular, our previous work demonstrated that day-time dependent alterations preceding seizures are largely driven by an increased occurrence of recurrent, short-lasting (0.6s) high-connectivity network configurations, termed High-Connectivity States (HCS). In this study, we replicated these findings using long-term sEEG recordings from a publicly available database and another patient undergoing presurgical evaluation with depth electrodes. For the latter patient, we leveraged all available monitoring information to validate the robustness of HCS alterations across different recording configurations. Additionally, we compared HCS dynamics with other sEEG measurements (e.g., power activity) and electrocardiogram-derived variables (e.g., heart rate) and developed a low-dimensional statistical network model to capture key network features of the pre-seizure period. Our results confirmed the specific emergence of HCS hours before seizure onset, consistent across monopolar, bipolar, and common average referential recordings. Notably, preictal HCS probability alterations were distributed across the frequency power spectrum and exhibited stronger correlations with heart rate fluctuations than with grand-average functional connectivity. The statistical model indicated that preictal HCS probability changes arise from shifts in network topology rather than changes in mean connectivity, highlighting the role of complex interactions between epileptogenic and non-epileptogenic brain regions. These findings suggest that integrating HCS probability with both invasive and non-invasive monitoring variables could establish it as a promising biomarker for an early detection of upcoming epileptic seizures.