Predictability of sleep slow oscillation emergence and spatial extent from pre-onset neural dynamics.

Slow oscillations (SOs; ~ 0.5-1.5 Hz) are a hallmark of non-rapid eye movement sleep and are known to support memory consolidation. Although their dynamics are well characterized, processes that precede SO initiation and predict their spatial extent remain largely unknown, despite their potential utility for closed-loop intervention. Using high-density EEG from 29 healthy adults, we examined neural activity in the 2-s interval preceding the SO trough. SOs were classified into Global and Frontal subtypes, which differ in propagation. We quantified instantaneous spectral power, time-frequency dynamics, and cross-frequency coupling including phase-amplitude coupling (PAC). Theta-band power (4-8 Hz) emerged as the strongest individual predictor of SO initiation, although delta-band activity provided complementary predictive information. Spectral parameterization indicated that this pre-onset theta-band elevation primarily reflects broadband shifts in the aperiodic background rather than a modulation of a discrete oscillatory component. Theta-band power exhibited a sustained rise beginning ~ 1.6 s before onset and differentiated SO subtypes with moderate-to-strong effect sizes (Cohen's d = 0.45-0.77). Mechanistically, Global SOs were preceded by enhanced delta-theta PAC and broad low-frequency synchronization, whereas Frontal SOs showed elevated theta/alpha-to-beta/low-gamma coupling, reflecting locally enhanced coupling that restricts propagation. A logistic regression classifier using pre-onset theta power achieved > 95% accuracy distinguishing SOs from surrogate events and ~ 83% accuracy differentiating SO subtypes. These findings demonstrate that pre-onset low-frequency spectral dynamics, particularly theta-band power, can predict SO occurrence and might be leveraged in neuromodulation.