Fig. 1: Study design and behavioural results.

a Participants learned three different cognitive tasks: (i) an attentional task (n = 20): the psychomotor vigilance test (PVT), (ii) a visual task (n = 18): the texture discrimination test (VDT), and (iii) a procedural task (n = 20): the mirror tracing test (MTT). Then, participants were allowed to sleep for ~7 h, and acoustic clicks locked to eye movements (EM) were applied during REM periods in the stimulation procedure (STM), or the sound was muted in the control procedure (CNT). After sleep, participants were tested on the same cognitive tasks. The order of the nights (CNT, STM) and the order of the tasks were randomized between participants, but the tasks were applied in the same order within participants for pre- and post-sleep tests. b Automatic detection of EM. One EOG channel was constantly monitored, and the EM detection was activated when in stable REM sleep. An amplitude threshold (50–100 µV) was manually set, and when the EOG amplitude crossed this level, an auditory click was applied, followed by a 2.5-s pause in the stimulation. Clicks were muted in the CNT condition. c No differences were evident in the overnight performance gain for the attention task (PVT). Changes in the overnight performance gain and the effect sizes for STM and CNT conditions in the visual (VDT) (d) and procedural (MTT) (e) tasks. Multivariable linear models suggest that the time spent in tonic REM predicts overnight changes in visual task performance (% gain = 13.5 + 20.5 C–0.66tR) (f), whereas the time spent in phasic REM predicts the performance gains on the procedural task (% gain = 31.9.5–12.0 C–0.28pR) (g). Error bars indicate mean ±95% CI and Δ visualizes effect sizes by the difference of means. The distribution curve for the effect size indicates the resampled distribution of ∆ given the observed data. For the multivariate linear regressions in f and g: C = condition (CNT/STM), tR = tonic REM and pR = phasic REM, shading corresponds to 95% CI for the responses.