Sleep is more than rest for plasticity in the human cortex

Abstract

Sleep promotes adaptation of behavior and underlying neural plasticity in comparison to active wakefulness. However, the contribution of its two main characteristics, sleep-specific brain activity and reduced stimulus interference, remains unclear. We tested healthy humans on a texture discrimination task, a proxy for neural plasticity in primary visual cortex, in the morning and retested them in the afternoon after a period of daytime sleep, passive waking with maximally reduced interference, or active waking. Sleep restored performance in direct comparison to both passive and active waking, in which deterioration of performance across repeated within-day testing has been linked to synaptic saturation in the primary visual cortex. No difference between passive and active waking was observed. Control experiments indicated that deterioration across wakefulness was retinotopically specific to the trained visual field and not due to unspecific performance differences. The restorative effect of sleep correlated with time spent in NREM sleep and with electroencephalographic slow wave energy, which is thought to reflect renormalization of synaptic strength. The results indicate that sleep is more than a state of reduced stimulus interference, but that sleep-specific brain activity restores performance by actively refining cortical plasticity.

NREM, sleep, interference, synaptic plasticity, texture discrimination, slow-wave activity

Statement of significance

We still have a limited understanding of why we sleep, and disruptions of sleep represent a prominent health problem. One fundamental question remains controversial—namely, whether sleep-specific brain activity promotes adaptation of behavior and underlying neural plasticity (sleep hypothesis), or alternatively, whether sleep only provides a window of reduced stimulus interference passively protecting novel neural representations (rest hypothesis). Here, we used visual texture discrimination as a model of primary cortex plasticity and provide evidence that sleep is more than a state of reduced interference, which actively restores task performance by refining underlying neural plasticity.

Introduction

Sleep is ubiquitous in animals and humans and vital for healthy functioning. Thus, sleep after training improves performance on various tasks in comparison to equal periods of active wakefulness [1, 2], including visual texture discrimination [3–11]. However, the mechanisms remain controversial.

One influential line of research posits that sleep-specific brain activity patterns observed as electroencephalographic (EEG) slow waves during non-rapid eye movement (NREM) sleep improve task performance through a downselection of non-information bearing synapses that have been potentiated towards saturation during prior wakefulness, refining the information-to-noise ratio in task-relevant neural representations under conditions of limited energy and space (synaptic homeostasis hypothesis of sleep-wake regulation) [2, 12–16].

In contrast, another line of research regards sleep as an evolutionary adaptive yet inactive state, which improves task performance only through passive protection of task-relevant neural representations from destructive interference from novel input inherent to active wakefulness (rest hypothesis) [17–20]. For instance, in an auditory learning task, periods of daytime sleep, and passive waking with reduced sensory interference yielded similar performance benefits [21].

To further test these contrasting concepts, we used a well-characterized visual texture discrimination task (TDT) [4, 6, 22–25]. In this task, repeated within-day testing results in deterioration of perceptual performance [4, 5, 7, 8, 23, 26]. This deterioration has been shown to be (1) retinotopic (locally limited to the trained quadrant of the visual field) and (2) binocular (deterioration after monocular training transfers from the trained eye to the other at the same location) [23, 26]. These features indicate that this type of deterioration must emerge from experience-dependent synaptic modifications in the cortex, presumably at the level of orientation-gradient sensitive cells in the primary visual cortex. More specifically, initial training is assumed to strengthen task-related synaptic connections, with further within-day testing leading to runaway potentiation and saturation of surrounding, non-causally engaged connections and deterioration in performance [27]. Prior work has shown that sleep promotes the restoration of this performance deterioration and performance gains compared to active wakefulness (with performance gains even transferring from one eye to the other [28]) [3–9, 29]. One study demonstrated a restoration of performance deterioration also after brief periods of daytime sleep compared to active waking [8]. This study additionally implemented a reduced stimulus interference condition (blindfolded wake condition without visual input). This condition did not differ significantly from active waking. Yet critically, no direct statistical comparison between this passive wake and the sleep condition was reported [8], leaving the fundamental question open whether sleep is more than a state of reduced interference.

To further disentangle the effects of sleep-specific brain activity and reduced interference, we used a 1-hour daytime sleep laboratory paradigm comparing brief periods of NREM sleep to passive and active waking. A daytime sleep paradigm was chosen to minimize adverse effects from sleep deprivation and circadian bias that are inherent to whole-night studies. We tested the primary hypothesis that periods of sleep in comparison to equal periods of passive waking with maximally reduced interference would restore decrements in texture discrimination performance across repeated within-day testing. Moreover, we hypothesized a positive correlation between the restorative effect of sleep and EEG slow-wave energy, thought to reflect synaptic renormalization.

Methods

Participants

A total of 66 healthy individuals (18 to 30 years) participated in the study. All participants gave written informed consent. The study had been approved by the local ethics committee and was conducted in accordance with principles of the Declaration of Helsinki. All participants had normal or corrected-to-normal vision and no history of neurological, mental, or physical illness. Based on predefined criteria [30, 31], eight participants with a sleep efficiency <50% in the sleep group and five participants with >10 wake signals (presented at first signs of sleep onset during online polysomnographic monitoring) in the passive wake group were excluded to ensure clear-cut group differences for primary comparisons. This led to a sample of 53 participants for direct group comparisons (sleep group, n = 13, nine female, four male; passive wake group, n = 13, eight female, five male; active wake group, n = 15, nine female, six male; location switch group [active wake], n = 12, nine female, three male).

Study design

Using a parallel group sleep laboratory study design, the effects of periods of daytime sleep, passive waking, and active waking on performance in a texture discrimination task (TDT) were compared across repeated within-day testing (Figure 1, a and b). Participants were assigned to either a sleep group with polysomnographic monitoring from 1:30 pm to 2:30 pm, a passive wake group (maximally reduced interference) with polysomnographic monitoring from 1:30 pm to 2:30 pm (online monitoring with discrete wake signals at first signs of sleep onset), an active wake group (video, physical activity), or a location switch group (active waking, with switch between the lower left quadrant at training to the upper right quadrant at retest), which served as a control condition for retinotopic specificity of the observed effects. Texture discrimination and general cognitive performance (alertness and working memory) were assessed in the morning before and in the afternoon after the sleep/wake intervention. Participants of the sleep group were instructed to sleep. Participants of the passive wake group were instructed to lie awake in bed in the dark sleep laboratory with maximally reduced sensory input and motor activity (online visual polysomnographic monitoring with discrete auditory wake signals at first signs of sleep onset). Of note, we did not control for internal interferences (e.g. thoughts and imaginary processes). Participants in the active wake group watched a video and played table tennis (elevated/controlled level of interference) under supervision of staff members.

Figure 1.

Texture discrimination task, study design and results. (a) Trial sequence in the texture discrimination task. During each trial, participants fixated on a circle in the center of the screen and activated the target frame that consisted of a central fixation letter and the peripheral target texture in the lower left quadrant. (b) Parallel group sleep laboratory study design. (c) Effects of sleep (n = 13), passive waking (n = 13), active waking (n = 15), and active waking with location switch (n = 12) on texture discrimination performance, measured as change of SOA75 (difference; Training minus Test). Sleep was observed to not only restore performance in comparison to active waking, as has been shown previously, but also in direct comparison to passive waking. In the active wake group with location switch, performance did not deteriorate and did not differ from the sleep group. (d) No significant correlation between sleep polysomnographic or spectral analysis parameters and texture discrimination performance changes was observed for the sleep group participants (n = 13 with ≥ 50% sleep efficiency) that were included in the main analysis (all p-values > .6). However, after additional inclusion of participants (n = 8) with < 50% sleep efficiency, that were initially excluded from the main analysis of group comparison to ensure clear-cut differences between the sleep and wake conditions, a significant positive correlation between slow wave energy and restoration of texture discrimination performance was observed (n = 21; r = .6, p = .007). Of note, exclusion of three outliers (marked with a hashtag symbol) that may have driven the observed correlation did not change the result (r = .5, p = .036). SOA75, threshold of stimulus-to-mask onset asynchrony (ms) with 75% correct texture detection; SEI, sleep efficiency index. Mean ± SEM; *p < .05.

Texture discrimination task

The effects of periods of daytime sleep and waking on performance changes (Training – Retest) in the TDT were compared during repeated within-day testing (Figure 1, a and b). The texture discrimination procedures closely followed those described by Karni and Sagi [22], who provided the original software. Procedures were customized for adaptive testing according to Censor, Karni, and Sagi [4] and used as previously reported [32]. A trial sequence in the texture discrimination task is depicted in Figure 1a. During each trial, participants fixated on a circle in the center of the screen and activated the target frame that consisted of a central fixation letter (a randomly rotated L or T), and the peripheral target texture in the lower left quadrant (a vertical or horizontal array of three diagonal bars). The central task of identifying the letter ensures good fixation and shows only a modest improvement with practice, whereas the detection of the peripheral target represents the plasticity-dependent component of the task. An immediate auditory feedback was provided for letter identification to ensure central fixation. No feedback was given for texture discrimination. In the morning, participants were trained in blocks of 15 trials at 1000 ms and 400 ms stimulus-to-mask onset asynchrony (SOA), until they reached a criterion of >90% correct responses. Then, successive blocks of 40 trials each were presented with decreasing SOAs to determine the SOA at which 75% of the responses were correct (SOA75). The SOA was reduced in steps of 40 ms as long as >90% of responses were correct and then in steps of 20 ms, until the criterion of two blocks with <70% correct responses was reached. In the afternoon, successive blocks of 40 trials each were presented with decreasing SOAs, as described for the morning. For each participant and session, the learning curve (correct, %, versus SOA) was fitted with a logistic function between the chance level of 50% and 100% with the parameters SOA75 and scale (describing the inverse slope at SOA75) [32]. This adaptive testing, as previously described [4, 32], was applied to determine SOA75 as quickly and accurately as possible. As high performers show substantially lower SOA75 than low performers, they would have to endure an excessively large number of unnecessary trials (i.e. the number of samples from the nearly flat region of the curve above 90%, which do not contribute to the accuracy of the SOA75 estimation) to approach their limit, if the step was chosen to be constant and small [22]. Between-session changes in performance were expressed in a change (difference; Training session – Test session) of SOA75 (SOA at which 75% of the target textures were correctly detected).

Electroencephalographic recordings

Sleep and passive waking were monitored from 1:30 pm to 2:30 pm in the sleep laboratory using polysomnography and scored according to standard criteria [33, 34]. All recordings included an EEG, horizontal and vertical electrooculography, submental electromyography, and electrocardiography. The following parameters of sleep continuity and architecture were assessed: time in bed (period between lights off and lights on), total sleep time (time spent in NREM sleep stages N1, N2, SWS, or REM sleep, with sleep onset defined by the first epoch of N2, SWS, or REM sleep), as well as the time spent in waking and in sleep stages N1, N2, N3, and REM sleep. Additionally, sleep EEG spectral analysis was performed to assess power spectra in the standard frequency bands [35]. EEG spectral analysis was performed on the C3-M2 electrode derivation in the same 30-second epochs for which also sleep stages had been determined. In brief, data were recorded with a sampling rate of 256 Hz and a resolution of 16 bits. Signals were recorded with a low-pass at 75 Hz and a high-pass of 0.2 Hz (corresponding to a time constant of 0.8 second). Spectral estimates for each epoch were obtained as the average of 28 FFT windows overlapping by half (512 data points, 2.56 seconds) covering a 30-second epoch to obtain the spectral power within that epoch, resulting in a spectral resolution of 0.5 Hz. A Welch taper was applied to each FFT window after demeaning and detrending the data in that window. The spectral power values were then log-transformed (base e) to achieve normality distribution and continuously stored on disk. All subsequent steps, including statistical analyses, were performed on these logarithmic values. Artifact rejection used an automatic method discarding epochs due to abnormal total or gamma-band power values relative to a 10-min moving window, as described in detail previously [36]. Spectral energy was calculated by multiplying power in a specific spectral band with total sleep time (log[µV²/Hz]*min).

Attention and working memory

Attention [37] and working memory performances [38] were assessed prior to the training and retest session to control for potential confounding effects. For attentional performance, a standardized reaction time test was conducted, in which a cross appears on the monitor at randomly varying intervals and to which the participant responds as quickly as possible by pressing a key [37]. Working memory performance was assessed using a standard digit span test, in which participants have to repeat a sequence of numbers back to the examiner [38]. The change of performance (retest - training) was calculated for statistical analyses.

Data analysis

Data are reported as means ± SDs, if not indicated otherwise. For the primary analysis, a one-way analysis of variance (ANOVA) with the between-subject factor Group (sleep vs. passive wake) was conducted with change of texture discrimination performance (SOA75_Training− SOA75_Test) as outcome parameter. The same analysis was applied for comparing performance changes between the sleep and active wake group, as well as the passive and active wake group (secondary analyses). Partial eta square (⁠${\mathrm{\eta }}_{\mathrm{p}}^2$⁠) values were calculated as effect sizes for ANOVA effects (low < 0.06; medium ≥ 0.06 and < 0.14; large ≥ 0.14). In case of violation of the normality and variance homogeneity assumption for residuals in the ANOVA, a nonparametric Kruskal-Wallis one-way analysis of variance (Kruskal–Wallis test) was conducted instead of a one-way ANOVA. Spearman’s rank correlation coefficients were calculated to characterize the relationship between changes in texture discrimination performance and sleep parameters. IBM SPSS 24 was used for statistical analysis. Power calculation was done using G*Power 3.1.9.6 (F test, ANOVA, one-way). The level of statistical significance for the primary analysis was set at p < .05 (two-tailed). For secondary analyses no adjustment of p-values was performed.

Results

The effects of periods of daytime sleep, passive waking, and active waking on performance in the texture discrimination task (TDT) were compared across repeated within-day testing (Figure 1, a and b).

Sleep-wake data

The chosen sleep laboratory period of 1 hour during daytime ensured that participants exhibited virtually no REM sleep (Table 1). Only three participants in the sleep group exhibited very brief periods of REM sleep, allowing for the comparison of well-defined periods of NREM sleep with periods of passive or active waking.

Table 1.

Sleep-wake data

	Sleep (n = 13)	Passive wake (n = 13)	Active wake (n = 15)	Location switch (Active wake) (n = 12)
Time in bed, minutes	60.5 ± 0.6	60.0 (0.5)	0.0 ± 0.0	0.0 ± 0.0
Total sleep time, minutes	47.4 ± 7.9	0.0 ± 0.0	0.0 ± 0.0	0.0 ± 0.0
Stage N1, minutes	7.7 ± 3.9	0.5 (2.0)	0.0 ± 0.0	0.0 ± 0.0
Stage N2, minutes	29.7 ± 8.5	0.0 ± 0.0	0.0 ± 0.0	0.0 ± 0.0
Slow wave sleep, minutes	12.6 ± 11.6	0.0 ± 0.0	0.0 ± 0.0	0.0 ± 0.0
REM sleep, minutes	0.0 (0.3)	0.0 ± 0.0	0.0 ± 0.0	0.0 ± 0.0

Data represent mean ± SD of normally distributed data or median (IQR), if normal distribution was not given. Of note in the sleep group, only 4.7 ± 3.3 minutes of stage N1 sleep were contributing to the total sleep time, as sleep onset of nocturnal sleep was defined by the first epoch of N2, N3, or REM sleep.

Sleep is more than rest

As the main finding and in line with our primary hypothesis (Figure 1c), we observed a significant restorative effect of sleep-specific brain activity on texture discrimination performance in direct comparison to passive waking with maximally reduced interference (sleep group, n = 13, vs. passive wake group, n = 13; F_1,24 = 7.1, p = .013, ${\mathrm{\eta }}_{\mathrm{p}}^2=.229$⁠). Importantly, there was no baseline difference in texture discrimination performance between the sleep (223.2 ± 96.8 ms) and passive wake group in the morning (174.5 ± 95.9 ms; χ ²(1) = 2.1, p = .144; Kruskal-Wallis test).

Secondary analyses

As a control and replication of prior work [8], we observed a significant restorative effect of sleep in comparison to active waking (Figure 1c; passive wake group, n = 13, vs. active wake group, n = 15; F_1,26 = 4.3, p = .047, ${\mathrm{\eta }}_{\mathrm{p}}^2=.143$⁠). No significant difference in change of performance was observed between the passive and active wake group (χ ²(1) = 0.7, p = .394). Direct comparison of texture discrimination performance at baseline in the morning yielded no significant difference between the active wake group (182.4 ± 78.3 ms) and the sleep group, as well as between the active wake and the passive wake group (all p-values > .2).

To exclude that the group differences were due to unspecific variables, such as fatigue, we compared indices of general cognitive performance (attention and working memory) in the morning and afternoon. No significant differences were observed between groups at baseline and for change in attentional and working memory performance between the morning and afternoon sessions (p > .2 for all comparisons).

Importantly, to demonstrate retinotopic specificity of performance changes across periods of waking, we trained additional participants (location switch group [active wake], n = 12) in the lower left quadrant, but retested them in the upper right quadrant of the visual field after an equivalent period of active waking. In contrast to the active wake group without location switch and in agreement with previous studies [23, 26], performance change did not differ from the sleep group (χ ²(1) = 0.2, p = .624), suggesting a retinotopic effect (Figure 1c).

Control analyses

Of note, for the comparisons between the sleep and the passive wake group, as well as between the sleep and the active wake group, the normality assumption of residuals for conducting an ANOVA was met. However, the change of performance (dependent variable) itself was not normally distributed in neither the sleep, nor the passive wake or the active wake group. To further explore the data, we conducted additional nonparametric tests for ANOVA results of the primary and secondary analysis. While the primary comparison between the sleep and the passive wake group was robust with regard to nonparametric testing (χ ²(1) = 6.7, p = .010), the comparison between the sleep and active wake group was not robust (χ ²(1) = 1.7, p = .189). This difference was due to three outliers in the active wake condition and one outlier in the sleep condition, which were, however, not excluded from the main analysis, as reaction times in the morning and afternoon were within the normal range and the change of performance levels was therefore considered to be valid. For full descriptive characterization of changes in TDT performance, medians and interquartile ranges (IQR) for each group, please refer to Table 2.

Table 2.

Change of performance in the texture discrimination task

	Sleep (n = 13)	Passive wake (n = 13)	Active wake (n = 15)	Location switch (Active wake) (n = 12)
Change of performance, SOA75Training-Test	12.1 (36.9)	-12.5 (51.7)	-1.1 (70.9)	8.0 (43.6)

Data represent median (IQR). SOA75, threshold of stimulus-to-mask onset asynchrony (ms) with 75% correct texture detection.

Association between sleep parameters and performance restoration

No significant correlation between sleep polysomnographic or spectral analysis parameters and texture discrimination performance changes was observed for the n = 13 participants of the sleep group (all p-values > .6). In a next step, additional participants (n = 8) with a sleep efficiency < 50%, who had been excluded from the main group comparisons based on predefined criteria [30, 31] to ensure clear-cut differences between the sleep and wake conditions, were included in this correlation analysis to increase the variance of sleep parameters and performance changes in texture discrimination, resulting in a sample of n = 21 participants. In this sample, we observed a significant positive correlation between spectral energy in the slow-wave frequency band and performance restoration in texture discrimination (r = .6, p = .007; Figure 1d). Of note, exclusion of three outliers (Figure 1d; marked with a hashtag symbol) that may have driven the observed correlation did not change the result (n = 18; r = .5, p = .036). Furthermore, we observed a significant positive correlation between performance change in texture discrimination and total sleep time (r = .6, p = .005), as well as total time spent in stage 2 sleep, which constituted the dominant sleep stage in our 1-hour window of daytime sleep (r = .5, p = .013), with both correlations indicating a bolstered restoration of performance with increasing time spent in NREM sleep. Furthermore, significant correlations between change in texture discrimination and spectral energy in the theta, alpha, beta 2, and gamma frequency band were observed (all p-values < .05), indicating that effects of spectral energy might have been driven by the observed correlation with total sleep time.

Despite the limited number, we further explored the group of participants with a sleep efficiency of <50% (n = 8, TST: 15.6 ± 9.8 min). This group showed an absence of a restorative effect on performance levels (SOA75_{Training−Test}: −48.8 ± 59.2 ms). This effect differed significantly from the group of participants with a sleep efficiency of >50% (20.1 ± 45.4 ms; F_1,19 = 9.1, p = .007, ${\mathrm{\eta }}_{\mathrm{p}}^2=.323$⁠). There were no significant differences between the group of participants with a sleep efficiency of < 50% and the passive wake (-28.5 ± 47.4 ms; χ ²(1) = 1.3, p = .247) or active wake group (-27.3 ± 70.2 ms; χ ²(1) = 2.2, p = .138). Please refer also to Figure 1c. This suggests that only sleep containing enough consolidated NREM sleep promotes performance restoration.

Discussion

The results of this study provide, in line with our primary hypothesis, evidence that sleep-specific brain activity is superior to an equal period of rest (passive wakefulness with maximally reduced sensory input and motor activity) with respect to the restoration of performance in visual texture discrimination, a model of synaptic plasticity in the human cortex. The daytime paradigm allowed for differential testing of a well-defined period of NREM sleep without the adverse effects of sleep deprivation in the wake comparisons or a circadian bias with respect to time points of testing [39]. The superiority of a period of NREM sleep to an equal period of active wakefulness (secondary hypothesis), as shown previously [8], was replicated in the current study.

Our results provide direct evidence against the rest hypothesis, which suggests that sleep provides only a window of inactivity with reduced stimulus interference [17, 18, 40, 41]. Rather, our results indicate that sleep-specific brain activity is required to restore the wake-associated deterioration of visual discrimination. Our findings extend prior work on the effect of sleep-specific brain activity on plasticity changes in the human cortex, suggesting that sleep might ensure a global reinstatement of neural network stability and consolidation of local learning-induced plasticity changes in the human cortex [15, 16].

Of note, we controlled for alternative explanations, such as unspecific fatigue, by demonstrating a restriction of deterioration across wakefulness to the trained location in the visual field (retinotopic effect). Specifically, this deterioration is thought to result from over-potentiation and saturation of non-causally engaged synapses during repeated testing across wakefulness [23, 26]. The observed effects may also include higher visual [42–44] or other cortical areas [45, 46]. Downselection of non-causally engaged synapses during NREM slow-wave sleep provides a plausible mechanism through which sleep might improve the signal-to-noise ratio and restore performance [2, 14]. The current data are consistent with this concept by demonstrating an association between EEG slow-wave energy and the restorative effect of sleep on the behavioral readout of the network that is required for visual discrimination. However, this effect might have been driven by the observed correlation with total sleep time or other sleep parameters, and future studies are needed to further determine the neural mechanisms. Longer sleep periods containing REM sleep have been reported to be required for actual task improvement [5–7, 10], beyond the restoration level observed in the current study. Notably, we observed effects of brief periods of NREM sleep. However, this effect was restricted to sleep periods with sufficient time spent in NREM sleep. For participants, who were below a predefined cutoff of < 50% sleep efficiency in our 1-hour sleep opportunity window [30, 31], no restoration of performance was observed. This suggests that not sleep per se, but only sleep containing enough consolidated NREM sleep promotes performance restoration. Of note, we focused on the maximal reduction of external interference and did not control for internal interference, such as thoughts and imaginary processes. This constitutes a limitation of our study. Future studies are needed, potentially using meditation- or mindfulness-based approaches, to control for this aspect.

The findings have critical implications for real-life conditions of intense training and related neural refinements and indicate that sleep cannot be replaced by periods of rest.

Together, our results provide evidence that sleep-specific brain activity is more than rest and reduced interference for restoring primary cortex plasticity in humans.

Acknowledgments

We thank Y. Bonneh and D. Sagi for providing the software for the Texture Discrimination Task (TDT). The authors also wish to thank the technical staff at the University of Freiburg Medical Center for their help in conducting the study.

Conflict of interest statement. None declared.

Author contributions

C.N. designed the experiment with advice of H.P., J.H., D.R., and B.F. H.P., J.H., and L.F. collected the data. B.F. supervised data collection. C.N. supervised the entire project. H.P., J.G.M., and B.F. analyzed the data. All of the authors interpreted the results. H.P., J.H., J.G.M., A.S., and C.N. wrote the manuscript, with contributions from all other authors.

Data availability

The data that support the findings of this study are available from the corresponding author upon request.