Abstract
Different levels of reduced consciousness characterise human sleep stages at the behavioural level. On electroencephalography (EEG), the identification of sleep stages predominantly relies on localised oscillatory power within distinct frequency bands. Several theoretical frameworks converge on the central significance of long-range information sharing in maintaining consciousness, which experimentally manifests as high functional connectivity (FC) between distant brain regions. Here, we test the hypothesis that EEG-FC reflects sleep stages and hence changes in consciousness. We retrospectively investigated sleep EEG recordings in 14 participants undergoing all stages of non-rapid eye movement (NREM) sleep. We quantified FC with six phase coupling metrics and used the FC coefficients between electrode pairs as features for a gradient boosting classifier trained to distinguish between sleep stages. To characterise FC during each stage of NREM sleep, we compared these metrics regarding their classification accuracy and analysed the ranked feature importance across all electrode pairs. We observed frequency-specific differences in FC between sleep stages for all metrics except the imaginary part of coherence. Alpha coupling decreased from wake to sleep stages N1 and N2, whereas delta coupling increased in deep sleep (N3). FC-based sleep classifiers yielded 51% (phase locking index) to 73% (phase locking value) classification accuracy. Distributed FC patterns in the alpha band ranked highest in terms of feature importance. In a limited sample of 14 subjects, we demonstrated that FC computed from phase information changes significantly across sleep stages. The finding that EEG phase patterns are indicative of sleep stages supports the hypothesis that long-range and spatially distributed phase coupling within frequency bands, especially within the alpha band, is an electrophysiological correlate of consciousness across sleep stages.