Characterizing transition state in mouse vigilance with electroencephalogram-electromyogram hypnodensity

Vigilance-state transitions are continuous biological processes, yet conventional rodent sleep scoring relies on discrete epochs that obscure intermediate states. As no standardized framework exists for characterizing these intermediate states in rodents, this study aimed to characterize the temporal dynamics of transitions in mice and validate a machine-learning approach for objective detection.

Chronic electroencephalogram (EEG) and electromyogram (EMG) recordings were obtained from male C57BL/6 N mice. We extracted 56-s windows containing stable transitions between wakefulness (WAKE), non-rapid eye movement sleep (NREMS), and rapid eye movement sleep (REMS). Eight trained experts manually annotated the onset and duration of transitions to establish ground truth and assess inter-rater reliability. Using quantitative EEG/EMG features (e.g. spectral power, complexity, and EMG variance) derived from stable states, support vector machine (SVM) classifiers were trained to predict transition midpoints in independent test animals.

Inter-rater agreement among experts was moderate to low, particularly for WAKE-to-NREMS and NREMS-to-REMS (N-R) transitions, reflecting inherent ambiguity in manual scoring. Temporal analysis revealed distinct dynamics across transition types; N-R transitions were significantly longer than all others, while REMS-to-NREMS transitions were the most abrupt. Despite the variability in human scoring, SVM models trained only on stable-state features successfully predicted expert-defined transition midpoints.

Our approach not only characterizes the recognizable dynamics across transition types in mice, but also provides a reproducible framework for quantifying sleep-wake transitions, which is crucial for studying arousal stability and related impairments in disease.