PMC

The decline of timescales during sleep deprivation occurs over broad cortex areas with a particular dominance in frontal and parieto-occipital regions. All plots correspond to results from the alpha band (8–12 Hz). (a) Exemplary time courses from two individual EEG channels indicating a strong decline in autocorrelation values during sleep deprivation. (b) Topographic distribution of changes in lag-1 autocorrelation during sleep deprivation (difference between average values from 0–6 hours and average values from 33–39 hours). Each marker corresponds to one EEG channel. While a decrease in autocorrelation is evident in all channels, changes a particularly prominent in frontal and parieto-occipital regions. (c) Topographic distribution of sleep deprivation changes in DFA scaling exponents. (d) Topographic distribution of average power in the alpha band.

Changes in EEG power during sleep deprivation. (a) EEG power density at the beginning of sleep deprivation (0–6 hours, blue) and end of sleep deprivation (33–39 hours, red; grand average across 8 subjects and all channels, error bars indicate s.e.m.). Local slopes of the PSD were obtained in the ranges indicated by the solid lines. Right: Local slopes indicate a more shallow PSD at the end of sleep deprivation compared to beginning of sleep deprivation and after recovery sleep. (b) Power changes in frequency bands. Straight black lines indicate linear regression results; bars indicate the average power values at the beginning (0–6 hours) and at the end (33–39 hours) of sleep deprivation. Of the three frequency bands, only the alpha band (8–12 Hz) showed no significant change during sleep deprivation as judged by linear regression and difference between bars (ΔPower).

Influence of signal power on estimating long-range temporal correlations. DFA scaling exponents exhibited a significant correlation with signal power in all frequency bands investigated (left column). Middle column: DFA scaling exponents (black) follow a very similar trajectory to signal power (green) particularly when power is changing significantly, as is the case in the theta (top) and beta (bottom) bands. Right column: DFA increases between early (0–6 hours, blue) and late (33–39 hours, red) sleep deprivation vanish in the theta and beta bands when only channels with no significant power change (power controlled channels) are considered for DFA estimation. Conversely, the decline in DFA scaling exponents in the alpha band remains. This suggests that it is the strongly increasing signal power in theta and beta bands that leads to an apparent increase of DFA scaling exponents in these frequency bands.

Decline in signal autocorrelation and long-range temporal correlations during sleep deprivation. (a) Autocorrelation function of the signal envelope in the alpha band (0–6 hours, blue; 33–39 hours, red; inset, autocorrelation functions until 1 s). (b) Faster autocorrelation decay during sleep deprivation (0–39 hours of sleep deprivation and after consecutive recovery sleep, rec). The solid black line corresponds to the mean across all channels from all 8 subjects, error bars indicate s.e.m. (c) Detrended fluctuation analysis of the of the signal envelope in the alpha band. Surrogate white-noise data are depicted in grey. DFA scaling exponents were obtained from linear fits in double-logarithmic coordinates aver the depicted fitting range. (d) Decline in long-range temporal correlations estimated by the DFA scaling exponent. The inset shows results for different polynom detrending orders. Results settle for detrending order of three and higher as judged by linear regression analysis and the difference between early (0–6 hours) and late (33–39 hours) sleep deprivation (ΔDFA).