Inter-hemispheric oscillations in human sleep

Abstract

Sleep is generally categorized into discrete stages based on characteristic electroencephalogram (EEG) patterns. This traditional approach represents sleep architecture in a static way, but it cannot reflect variations in sleep across time and across the cortex. To investigate these dynamic aspects of sleep, we analyzed sleep recordings in 14 healthy volunteers with a novel, frequency-based EEG analysis. This approach enabled comparison of sleep patterns with low inter-individual variability. We then implemented a new probability dependent, automatic classification of sleep states that agreed closely with conventional manual scoring during consolidated sleep. Furthermore, this analysis revealed a previously unrecognized, interhemispheric oscillation during rapid eye movement (REM) sleep. This quantitative approach provides a new way of examining the dynamic aspects of sleep, shedding new light on the physiology of human sleep.

Figure 1. Conserved topography of the sleep state space.

(a) Summary scatter plot of all sleep states for 14 subjects mapped in a 2-dimensional state space. Each 5s EEG epoch (raw data) is represented by 2 different frequency ratios plotted on log/log axes. Ratio1 = (8.6 to 19.3 Hz)/(1.0 to 10.9 Hz), Ratio2 = (11.5 to 20.3 Hz)/(17.9 to 31.5 Hz). Colour coding of the clusters is based on expert scoring for WAKE (red), NREM stage 1 (yellow), stage 2 (green), stage 3 (blue), and REM sleep (magenta). Projections of the 2d-probability density distributions are plotted for NREM stage 2 and stage 3 (top edge, ratio1) and WAKE and REM sleep (right edge, ratio2). For better visibility, the figure shows 10% of all data after applying a running window average (6 point-Hann Window) on the raw data to filter for short-term fluctuations. (b) Individual sleep trajectories are shown for each subject separately (panel 1–14) and cumulated for all individuals (panel 15, bottom right). Sleep trajectories are smoothed (50 point-Hann Window) for better differentiation of stable (clusters) and transitional sleep states (trajectories). Colour coding is as described in (a).

Figure 2. Trajectory for a single WAKE to NREM stage 2 transition.

(a) Consecutive points in the state space (each representing a 5s EEG-epoch) during a fast state transition (relevant trajectory (red) as shown in embedded figure) Points are colour coded according to velocity (bottom: colour bar legend: arbitrary velocity units). (b) EEG spectral power at different time points during this transition (as indicated by points 1–5 in Figure 2a). Note the uniform and distinctive spectral distribution in slow velocity states. Typical spectral peaks are indicated for point 1 (alpha activity in WAKE: 10.7 Hz) and point 5 (beta activity in NREM2: 13.1 Hz) by a thin red line. During the transitional epochs (points 2–4), the spectra are intermediate.

Figure 3. Subdivision of sleep states by velocity.

States of sleep were subdivided into high velocity states (i.e. transitions and fluctuations) (a), and slow states (i.e. stable clusters) (b), by a heuristic velocity limit. Points are colour coded according to velocity (right side: colour bar legend: arbitrary velocity units). Comparison of the manual classification (c) with automatic probability based classification (d) for a whole night data set for one representative subject is shown. Colour coding: WAKE (red), NREM stage 1 (yellow), stage 2 (green), stage 3 (blue), and REM sleep (magenta).

Figure 4. Laterality Analysis of REM sleep.

(a) Time series data of relative laterality of velocity in corresponding central electrodes (C3 vs C4) for 100 successive REM sleep epochs for one individual. (b) Sample autocorrelation of the same period as shown in (a) demonstrating a stable oscillation with a period of about 40s. Approximate 95% confidence measures for the hypothesis of uncorrelated white noise are indicated by a dotted thin blue horizontal line (p = 0.05, N = 100). (c) Fourier analysis (FFT) of REM sleep autocorrelations as in Figure 4b during one selected REM sleep bout for each subject. (d) Mean spectral analysis of REM sleep autocorrelation of all 14 subjects.