0074 Sleep Slow Oscillations Differ in Depth Profiles Based on Their Coupling with Spindles: A Classification Study

Abstract Introduction Slow oscillations (SOs) are large amplitude events in the electroencephalogram and contribute to sleep functions including homeostasis and consolidation of episodic memory. Complexes of one SO and one sleep spindle, where a spindle follows an SO trough within a short delay, are causally connected to sleep-dependent episodic memory improvement. In this study, we compare the cortical-subcortical brain currents that underlie SOs paired to spindles (SO+) and unpaired SOs (SO-), with the goal of identifying whether networks of activations are different in the two cases. This knowledge is essential to understanding the mechanisms of memory consolidation across brain regions. Methods Full-night polysomnography data (64 channel EEG) was acquired in the Mednick Lab for 22 typical young adult participants. Individual SOs and spindles were detected with in-house algorithms at every EEG channel. SO current sources at multiple time instances were found using source estimation software (Brainstorm) and encoded in a matrix representation of region-by-time, where each matrix was labeled as SO+ or SO-. Region-by-time features with strongest differentiation between SO+ and SO- were identified with multivariate feature ranking; classification with K-nearest neighbors and random forest evaluated with Matthew’s Correlation Coefficient (MCC) tested differentiability in a pooled dataset. We analyzed light (S2) and deep NREM (SWS) sleep separately. Results All classification algorithms achieved high MCC scores, suggesting structural differentiation between SO+/SO-. Within-individual models did not achieve high MCC scores when used to classify another individual’s SOs, refuting generalizability across subjects. Subcortical/cortical activity in SO+ vs. SO- was also found to be distinguishable up to 1 second before the SO trough. Additional analysis showed overall higher region-by-time feature interdependence underlying SO+, with the most significant features being the bilateral cortex, hippocampus, putamen, and pallidum. Conclusion Successful classification of SO+/SO- indicates the presence of structural differences in cortical-subcortical activation during SO that are spindle-coupled compared to uncoupled. Combined with feature analysis identifying distinct network activations in coupled vs uncoupled SOs, our study suggests a potential functional difference. Ongoing work includes investigating the various feature dependencies that exist within our dataset as well as conducting confirmatory analyses in a separate data subset. Support (if any)