0194 Sources of variation in the spectral slope of the sleep EEG

Introduction The 1/f spectral slope of the electroencephalogram (EEG) estimated in the gamma frequency range has been shown to reflect neural excitation/inhibition ratio and synchronization level within local neural populations. It was proposed as an arousal marker that differentiates wake, non-rapid eye movement (NREM) sleep, and rapid eye movement (REM) sleep. These stages exhibit progressively steeper 30-45 Hz slopes, interpreted in terms of increasing cortical inhibition. Here we sought to replicate these findings in a larger sample and provide a comprehensive characterization of how slope changes with age, sex, and its test-retest reliability as well as potential confounds that could affect the slope estimation. Methods After stringent exclusions and quality control, our final sample included 10,255 whole-night polysomnograms (PSGs) on 7,312 individuals 2,943 of whom had a second PSG, from the National Sleep Research Resource (NSRR). All preprocessing steps were performed using an open-source Luna package and the spectral slope was estimated by fitting log-log linear regression models on the absolute power from 30 to 45 Hz separately for wake, NREM and REM stages. We described sources of variation in the spectral slope (both within and between individuals) and its relationship to other sleep parameters including power and interhemispheric coherence. Results There was unambiguous statistical support for the hypothesis that, within individuals, the mean spectral slope grows steeper going from wake to NREM to REM sleep. We found that the choice of mastoid referencing scheme modulated the extent to which electromyogenic or electrocardiographic artifacts were likely to bias 30-45 Hz slope estimates, as well as other sources of technical, device-specific bias. Nonetheless, within individuals, slope estimates were relatively stable over time. Both cross-sectionally and longitudinal, slopes tended to become shallower with increasing age, particularly for REM sleep; males tended to show flatter slopes than females across all states. Although conceptually distinct, spectral slope did not predict sleep state substantially better than other summaries of the high-frequency EEG power spectrum (>20 Hz, in this context) including beta band power, however. In contrast to the common conception of the REM EEG as relatively wake-like (i.e. ‘paradoxical’ sleep), REM and wake were the most divergent states for multiple metrics, with NREM exhibiting intermediate profiles. Under a simplified modeling framework, changes in spectral slope could not, by themselves, fully account for the observed differences between states, if assuming a strict power-law model. Conclusion Although the spectral slope is appealing, theoretically inspired parameterization of the sleep EEG, we underscore some practical considerations that should be borne in mind when applying it in diverse datasets. Future work will be needed to fully characterize state-dependent changes in the aperiodic portions of the EEG power spectra, which appear to be consistent with, albeit not fully explained by, changes in the spectral slope. Support (If Any)