Rocking-induced sleep enhancement promotes motor learning through transcriptional and synaptic remodelling

TL;DR

Rocking during sleep improves motor learning in mice by enhancing sleep and changing brain genes and synapses.

Contribution

This study shows that rocking-induced sleep enhancement improves motor learning through transcriptional and synaptic changes in the motor cortex.

Findings

Eleven days of rocking increased sleep duration and consolidation in mice.

Improved sleep correlated with better motor learning performance.

Rocking led to transcriptional changes in glutamatergic signaling and increased excitatory synapse density in the motor cortex.

Abstract

Growing evidence shows that sleep can be enhanced in a non-invasive, drug-free manner through sensory stimulation. While modalities such as auditory and vestibular stimulation effectively increase sleep, the cognitive and cellular consequences of such enhancement remain unclear. Here, we investigated the effects of vestibular stimulation via rocking on sleep architecture, motor learning, cortical gene expression, and synaptic organization in the motor cortex. Eleven consecutive days of rocking enhanced sleep in mice, increasing both sleep duration and consolidation. These improvements were accompanied by greater motor learning performance, and the degree of learning enhancement positively correlated with total sleep amount. At the molecular level, improved learning was associated with transcriptional changes in genes involved in glutamatergic signalling and synaptic plasticity, alongside an increased density of excitatory synapses in the motor cortex. Together, these findings demonstrate that sleep enhancement via rocking facilitates learning by promoting neuroplastic mechanisms in the motor cortex.

Enhancing sleep with vestibular stimulation improves motor skill learning in mice and is associated with plasticity-related gene expression and synaptic changes in the motor cortex.

Figures

Rocking increases NREM sleep time during stimulation time.A Experimental design of the polysomnographic experiment. Wake (W), NREM, and REM sleep quantification through polysomnographic staging for baseline (*n* = 5) and rocking time (*n* = 5) for whole day (B), only light period (C), and only dark period (D). P < 0.01, *P < 0.001. Within the box plots lines and errors bars represent mean ± std. Here and thereafter: ns = not significant.

Rocking increases sleep time and reduces wake-up episodes.A Experimental design. Here and thereafter: yellow bars indicate 12 h of lights ON, grey bars 12 h of lights OFF. B Representative motion activity profiles from S (upper panel) and SE (lower lower) mice. C Sleep amount relative to baseline in the S and SE groups across the 11-day experiment. Mean ± std. Averaged sleep amount (D) and sleep-to-wake transitions (E) in the S and SE groups. Each dot is a single animal, bar are mean ± std. From C to E, data are shown for SE (*n* = 15) and S (*n* = 12) mice. P* < 0.05.

Rocking improves motor learning on the complex wheel task.A Photograph of the complex wheel and schematic representation of its irregular rung pattern. Maximum (B) and average (C) running speed of S and SE mice across the 11-day experiment. Mean ± std. Motor learning quantified using maximum (D) or average (E) speed. Correlation between learning (maximum (F) and average (G) speed) and sleep amount (normalized to baseline) in S and SE mice. H Correlation between learning (calculated using maximum speed) and sleep fragmentation. I Average bout duration. J Average number of running bouts. K Produ

Rocking during wake had no impact on motor learning.A Experimental design. B Motion activity profiles of a control (non-rocked, upper panel) and rocked (lower panel) mouse. C Time course over the 11 days of the percentage of sleep during the 4-h experiment for non-rocked and rocked mice. The small inset indicates the percentage of sleep during the 4-h experiment averaged across the 11 days of observation. D Average sleep amount during the light phase averaged across the 11 days of observation for non-rocked and rocked mice. E Maximum speed profile of non-rocked and rocked mice over the course

Rocking with motor learning is linked to transcriptional changes in synaptic and motor-related genes.A Volcano plot displaying upregulated (red) and downregulated (blue) transcripts, with selected transcripts labeled (S, *n* = 4; SE, *n* = 4). B Left. Gene ontology (GO) enrichment analysis for biological processes, highlighting highly enriched categories (blue) and moderately enriched categories (green). Right. Protein-protein interaction (PPI) network of putative proteins from the most enriched GO category (synaptic cluster), indicating significant functional associations. C GO enrichment ana

Rocking with motor learning is associated with an increased density of glutamatergic synapses.A Representative images of S and SE mice showing cortical VGLUT1 immunostaining. Scale bar: 10 µm. B Quantitative estimation of VGLUT1 positive puncta density in S and SE mice. P* < 0.05. C Representative images of S and SE mice showing cortical VGAT immunostaining. Scale bar: 7 µm. D Quantitative estimation of VGAT positive puncta density in S and SE mice. Example of perisomatic VGAT analysis (E, scale bar: 7 µm) and its estimated density in S and SE mice (F). Correlation between learning (maximum (G

Rocking enhances sleep but does not change the density of synapses in the motor cortex and CA1.A Experimental design. B Motion activity profiles of a S (upper panel) and SE (lower panel) mouse. C Sleep amount relative to baseline in the S and SE groups across the 11-day experiment. Averaged sleep amount (D) and sleep-to-wake transitions (E) in the S and SE groups. P* < 0.05, *P* < 0.01. F Representative images of S and SE mice showing cortical VGLUT1 immunostaining. Scale bar: 10 µm. Quantitative estimation of VGLUT1(G), VGAT (neuropilar, H), and VGAT (perisomatic, I) positive puncta density i

Full text

Introduction

Sleep is a fundamental and evolutionarily conserved process critical for maintaining cognitive function, synaptic plasticity, and overall brain health^1^. Over the past decades, various strategies have been explored to enhance sleep quality and amplify its restorative functions. While pharmacological interventions have traditionally been at the forefront, increasing attention has been directed toward non-pharmacological methods, particularly those leveraging stimulation during specific sleep stages^2–4^.

Among these, both electrical and magnetic stimulations have been shown to enhance slow-wave activity (SWA), a hallmark of deep non-REM (NREM) sleep^5,6^. However, due to their invasive or technically demanding nature, their clinical and practical utility has been limited. As a result, research has shifted toward sensory-based approaches, which employ naturalistic, non-invasive stimuli to modulate sleep. Acoustic stimulation, delivered in a phase-locked manner during NREM sleep, has emerged as a particularly promising method^7^. Multiple studies have demonstrated that rhythmic auditory stimuli can reliably enhance SWA and improve memory consolidation in young, healthy individuals with stable sleep architecture^8,9^.

However, the efficacy of acoustic stimulation significantly diminishes in populations with fragmented sleep, such as older adults, individuals with insomnia, or those prone to frequent nocturnal awakenings. Even in healthy subjects with increased arousability, the response to acoustic-based sleep enhancement is often limited^10^. Neurophysiological studies suggest that these limitations stem from the concurrent activation of arousal systems in response to auditory stimuli during sleep, which can disrupt sleep continuity or diminish the effectiveness of slow-wave enhancement^11,12^. As a result, individuals with a low arousal threshold may be overly sensitive to acoustic stimulation during sleep, leading to frequent awakenings or sleep disruptions.

An alternative and potentially more tolerable sensory approach involves vestibular stimulation, typically in the form of gentle rocking. Rocking has been shown to promote sleep onset and increase the amount of time spent in NREM sleep in both humans and animal models^13–15^. Unlike acoustic stimuli, vestibular input appears to exert its effects without strongly activating arousal systems, thereby minimizing sleep disruption^13^. However, while rocking can facilitate the initiation and maintenance of NREM sleep, its ability to enhance slow-wave amplitude is more limited compared to acoustic stimulation^14,15^.

Most studies on acoustic stimulation during sleep have focused on behavioural outcomes, particularly memory and learning, with mixed and sometimes inconsistent results^16–18^. Evidence that vestibular stimulation may also support memory consolidation is limited to a few studies in humans and flies^14,19^. A critical knowledge gap remains regarding the cellular and systems-level changes induced by sensory-based sleep enhancement. Clarifying these mechanisms is essential to understanding the true impact of such interventions on brain function and to determining whether the observed behavioural benefits are supported by underlying physiological changes.

In this study, we combined vestibular stimulation with a motor learning task to investigate brain changes induced by this sleep-enhancing approach. We found that sleep enhancement promoted motor learning, with sleep-enhanced mice displaying a steeper learning curve. This improvement was accompanied by the upregulation of plasticity-related genes and an increase in glutamatergic synapses within the motor cortex.

Results

Rocking increases sleep duration and consolidation

Previous studies from our lab and others have shown that 1 Hz rocking promotes sleep in mice by activating the vestibular system, leading to increased NREM sleep through reduced wakefulness and faster sleep onset, without affecting REM sleep^13,15^. To verify the effectiveness of rocking in our current experimental context, we conducted a polysomnographic experiment recording one day of baseline followed by a day of rocking, with rocking active only during the light phase (8:00 AM–8:00 PM). Consistent with earlier findings, rocking during the light period increased NREM sleep duration while reducing wake time, with no significant impact on REM sleep (Fig. 1).Fig. 1. Rocking increases NREM sleep time during stimulation time.A Experimental design of the polysomnographic experiment. Wake (W), NREM, and REM sleep quantification through polysomnographic staging for baseline (n = 5) and rocking time (n = 5) for whole day (B), only light period (C), and only dark period (D). **P < 0.01, ***P < 0.001. Within the box plots lines and errors bars represent mean ± std. Here and thereafter: ns = not significant.

Building on these findings, we employed rocking to enhance sleep in our study. We designed an experiment in which mice were habituated to the experimental environment for at least 5 days (with the last three used as baseline). A group of mice was then rocked for 12 h during the light phase (Sleep enhancement [SE] group), while the control group (Sleep [S]) did not receive any stimulation. Both groups were then exposed to a motor learning task using a complex wheel during the dark phase. This alternating pattern of rocking and motor learning was repeated for 11 consecutive days (Fig. 2A). We utilized a video motion tracking system to monitor sleep and wake activity (Fig. 2B), avoiding the need to keep the mice implanted with EEG/EMG electrodes for extended periods and to minimize implant-induced tissue damage and inflammation. As shown in other works, this method reliably estimates sleep and wake with over 90% concordance with EEG, but cannot differentiate between NREM and REM sleep^20,21^.Fig. 2. Rocking increases sleep time and reduces wake-up episodes.A Experimental design. Here and thereafter: yellow bars indicate 12 h of lights ON, grey bars 12 h of lights OFF. B Representative motion activity profiles from S (upper panel) and SE (lower lower) mice. C Sleep amount relative to baseline in the S and SE groups across the 11-day experiment. Mean ± std. Averaged sleep amount (D) and sleep-to-wake transitions (E) in the S and SE groups. Each dot is a single animal, bar are mean ± std. From C to E, data are shown for SE (n = 15) and S (n = 12) mice. *P < 0.05.

We observed a gradual and sustained increase in sleep duration throughout the 11-day rocking period, with no decline over time (two-way repeated measure ANOVA: effect of time, F(5.601, 140.0) = 3.796, P = 0.002; effect of condition, F(1,25) = 5.836, P = 0.023; time × condition interaction, F(10, 250) = 1.1, P = 0.36). In contrast, sleep duration in the S group remained stable, fluctuating around baseline levels (Fig. 2C). This effect was further supported by a higher average sleep amount in the SE group compared to the non-rocking S group (unpaired t-test: P = 0.023; Fig. 2D). Moreover, the SE group exhibited fewer sleep-to-wake transitions, indicative of less fragmented and more consolidated sleep, compared to the S group (unpaired t-test: P = 0.017; Fig. 2E).

Thus, rocking over 11 days improved both sleep quantity and stability.

Rocking during sleep promotes motor learning

To promote motor learning, we utilized a custom-made, 3D-printed, Arduino-based complex wheel. Its key feature was an inhomogeneous rung pattern, requiring mice to adapt and learn how to run on the wheel effectively (Fig. 3A).Fig. 3. Rocking improves motor learning on the complex wheel task.A Photograph of the complex wheel and schematic representation of its irregular rung pattern. Maximum (B) and average (C) running speed of S and SE mice across the 11-day experiment. Mean ± std. Motor learning quantified using maximum (D) or average (E) speed. Correlation between learning (maximum (F) and average (G) speed) and sleep amount (normalized to baseline) in S and SE mice. H Correlation between learning (calculated using maximum speed) and sleep fragmentation. I Average bout duration. J Average number of running bouts. K Product of average bout duration × number of bouts. L Total distance travelled. Correlation between average bout duration × number (M) or total distance travelled (N) and learning. For all bar graphs, data represent mean ± std. S (n = 12), SE (n = 15). *P < 0.05, **P < 0.01.

Motor performance was analyzed by measuring the maximum and average running speed across the 11-day experiment period in both the SE and S groups. Both groups showed a gradual increase in speed, indicating progressive motor adaptation to the complex wheel task. However, the SE group consistently outperformed the S group, demonstrating a faster rate of improvement and reaching peak performance earlier (Fig. 3B, C). To quantify learning, we calculated the % improvement, defined as the ratio between maximum (or average) speed on the last and first day of wheel use. SE mice showed significantly greater improvement from baseline compared to S mice (unpaired t-test: maximum speed, P = 0.016; average speed, P = 0.04; Fig. 3D, E).

We next examined the relationship between sleep and motor learning by correlating total sleep duration with % improvement. Positive correlations were observed in both groups for average and maximum speed (max: SE: r = 0.52, P = 0.048; S: r = 0.48, P = 0.12; avg: SE: r = 0.62, P = 0.03; S: r = 0.45, P = 0.097; Fig. 3F, G). In contrast, the fragmentation index, used as a measure of sleep consolidation, showed a negative but non-significant correlation in both groups (SE: r = −0.2, P = 0.47; S: r = −0.24, P = 0.46; Fig. 3H). To formally assess whether these relationships differed between groups, we fitted a linear model including a Group × Sleep interaction. The interaction was not significant (max: β = 30.68, SE = 152.15, t = 0.20, P = 0.84; avg: β = −30.5, SE = 131.58, t = −0.23, P = 0.82), indicating similar slopes between SE and S groups for both maximum and average speed. Together, these results suggest that longer sleep duration was similarly associated with greater motor learning in both conditions.

Finally, to assess whether the enhanced learning in SE mice was driven by increased wheel use, we analyzed multiple measures of running activity, including the average bout duration, the average number of bouts, their product (bout duration × number), and the total distance traveled. No significant difference was observed between groups in average bout duration (unpaired t-test: P = 0.36, Fig. 3I), but SE mice exhibited a higher average number of bouts than S mice (unpaired t-test: P = 0.004, Fig. 3J). However, neither the bout duration × number nor the total distance differed significantly between groups (bout duration × number: unpaired t-test: P = 0.13; distance: unpaired t-test: P = 0.18, Fig. 3K, L). Importantly, neither of these measures correlated with learning performance (bout duration × number: SE: r = −0.13, P = 0.64; S: r = −0.04, P = 0.91; distance: SE: r = -0.08, P = 0.78; S: r = 0.15, P = 0.64; Fig. 3M, N), whereas sleep amount did.

These findings indicated that rocking during sleep facilitated motor learning on the complex wheel task, and this effect was primarily related to increased sleep, rather than greater time or activity on the wheel.

Rocking during wake does not affect motor learning

To directly test whether the facilitation of motor learning depended on sleep rather than on a direct effect of vestibular stimulation, we applied rocking during the dark (wake) phase and then assessed learning. Specifically, both groups of mice underwent a habituation period, followed by natural, undisturbed sleep during the light phase. At the start of the dark phase, one group received 1 Hz rocking for 4 h, while the other group remained unstimulated. To ensure that the mice stayed mostly awake during this period, an operator directly observed them and gently tapped on their cage whenever they appeared drowsy. However, since mice are naturally very active at dark onset, human intervention was minimal. After the 4-h period, both groups had unrestricted access to the complex wheel for the remainder of the dark phase (8 h). This protocol was repeated daily for 11 consecutive days (Fig. 4A). Representative 24 h motion activity graphs for S and SE mice are represented in Fig. 4B. The amount of sleep during the 4-h experiment was minimal throughout the entire duration and did not differ between the rocked and non-rocked groups (unpaired t-test: P = 0.97, Fig. 4C). Moreover, sleep during the light phase was comparable between the two groups (unpaired t-test: P = 0.56, Fig. 4D).Fig. 4. Rocking during wake had no impact on motor learning.A Experimental design. B Motion activity profiles of a control (non-rocked, upper panel) and rocked (lower panel) mouse. C Time course over the 11 days of the percentage of sleep during the 4-h experiment for non-rocked and rocked mice. The small inset indicates the percentage of sleep during the 4-h experiment averaged across the 11 days of observation. D Average sleep amount during the light phase averaged across the 11 days of observation for non-rocked and rocked mice. E Maximum speed profile of non-rocked and rocked mice over the course of the experiment. F Motor learning in non-rocked and rocked mice. For all the graphs data are shown for non-rocked (n = 6) and rocked (n = 7) mice. Bars represent mean ± std.

At the complex wheel task, both groups exhibited a progressive increase in maximum speed over the course of the experiment (mixed-effects model analysis: effect of time: F(2.351, 21.39) = 18.14, P < 0.0001, Fig. 4E), with no significant difference between groups (effect of condition: F(1, 11) = 0.64, P = 0.44, Fig. 4E). This lack of an effect from rocking was further confirmed by the normalized improvement analysis, which showed comparable motor learning between the two groups (unpaired t-test: P = 0.79, Fig. 4F).

These findings indicated that rocking occurring during wakefulness did not significantly impact motor learning.

Rocking with motor learning is associated with increased transcription of synaptic genes

To assess the cellular changes associated with sleep-enhanced learning, we performed standard RNA-sequencing (RNA-seq) on the motor cortex extracted from the S and SE groups. Principal component analysis (PCA) of the RNA-Seq data effectively captured the differences between the samples from the two experimental groups, with samples from the same group clustering closely together, indicating high consistency in the gene expression profiles within each group.

RNA-seq analysis identified 139 differentially expressed (DE) transcripts between the two groups, with 73 upregulated and 66 downregulated (Fig. 5A). To determine the biological significance of these DE transcripts, we performed a gene ontology (GO) enrichment analysis, which revealed that the predominant functional categories were related to the regulation of synaptic transmission and behaviour (Fig. 5B). Next, we focused on the synaptic category to explore potential relationships between the DE transcripts in this group. A protein-protein interaction (PPI) analysis using STRING^22^ revealed significant putative interactions among many transcripts, suggesting a strong functional network (Fig. 5C). Within this cluster, GO enrichment analysis identified key transcripts involved in glutamatergic synaptic transmission, such as Grm2, as well as transcripts associated with dopaminergic neuromodulation (Drd1, and Drd2) and adenosine neurotransmission (Adora2a), which is implicated in sleep regulation. Finally, to assess the potential phenotypic implications of these DE transcripts, we performed a mammalian phenotype ontology (MPO) enrichment analysis. This approach allowed us to systematically map the identified genes to known phenotypic traits observed in genetically modified mouse models, providing insight into potential functional consequences. The analysis revealed a significant enrichment of phenotypes related to learning, motor behaviour and coordination, indicating that the differentially expressed transcripts are functionally linked to neural processes underlying these behaviours (Fig. 5D).Fig. 5. Rocking with motor learning is linked to transcriptional changes in synaptic and motor-related genes.A Volcano plot displaying upregulated (red) and downregulated (blue) transcripts, with selected transcripts labeled (S, n = 4; SE, n = 4). B Left. Gene ontology (GO) enrichment analysis for biological processes, highlighting highly enriched categories (blue) and moderately enriched categories (green). Right. Protein-protein interaction (PPI) network of putative proteins from the most enriched GO category (synaptic cluster), indicating significant functional associations. C GO enrichment analysis of the synaptic cluster, revealing key biological processes. D Mammalian Phenotype Ontology (MPO) analysis, showing enriched phenotypes related to learning, memory, and motor behavior. For (B–D), the x-axis represents the enrichment score (signal), which indicates how strongly each GO term is overrepresented in our gene set relative to the background set.

These findings indicated that improved motor learning promoted by sleep enhancement was linked to transcriptional changes in genes associated with synaptic transmission, learning, and motor behaviour.

Rocking with motor learning promotes synaptic changes in the motor cortex

To validate our transcriptomic analysis at the synaptic level, we performed immunofluorescence staining on layer 5 of the motor cortex in both the S and SE experimental groups, each of which underwent motor learning at the complex wheel task. This approach allowed us to directly assess the morphological changes of the synaptic landscape and investigate how these changes correlate with sleep enhancement and motor learning. To assess glutamatergic and GABAergic synapse density in the motor cortex, we used VGLUT1 (Vesicular Glutamate Transporter 1) as a marker for excitatory synapses and VGAT (Vesicular GABA Transporter) for inhibitory synapses. Confocal analysis of cortical neuropil revealed a significantly higher density of VGLUT1-positive puncta in SE mice compared to S mice (Linear mixed model, P = 0.01, Fig. 6A, B). In contrast, densitometric analysis of VGAT-positive puncta, whether dispersed in the neuropil or localized perisomatically, showed no significant differences between the groups (Linear mixed model, neuropilar: P = 0.59; perisomatic: P = 0.85, Fig. 6C–F).Fig. 6. Rocking with motor learning is associated with an increased density of glutamatergic synapses.A Representative images of S and SE mice showing cortical VGLUT1 immunostaining. Scale bar: 10 µm. B Quantitative estimation of VGLUT1 positive puncta density in S and SE mice. *P < 0.05. C Representative images of S and SE mice showing cortical VGAT immunostaining. Scale bar: 7 µm. D Quantitative estimation of VGAT positive puncta density in S and SE mice. Example of perisomatic VGAT analysis (E, scale bar: 7 µm) and its estimated density in S and SE mice (F). Correlation between learning (maximum (G) and average (H) speed) and VGLUT-1 puncta density in S and SE mice. For all graphs, data were extracted from S (n = 5) and SE (n = 5) mice. Bars represent mean ± std.

Finally, to investigate whether the increased density of glutamatergic synapses in SE mice was directly associated with motor learning, we performed a correlation analysis between VGLUT1 puncta density in the motor cortex and learning performance, quantified as both maximum and average running speed on the complex wheel. While both metrics of motor learning showed positive trends with VGLUT1 density (maximum speed: r = 0.44, P = 0.2; average speed: r = 0.46, P = 0.19), these correlations did not reach statistical significance (Fig. 6G, H).

Thus, rocking with motor learning was associated with an increased density of glutamatergic synapses.

Rocking without motor learning did not impact the density of glutamatergic synapses

Our findings suggested that rocking during sleep enhances motor learning, an effect accompanied by transcriptional changes linked to synaptic plasticity and increased glutamatergic synaptic density in the motor cortex. To test the hypothesis that sleep enhancement alone, independent of motor learning, could drive synaptic changes, we carried out an experiment in which mice experienced 11 days of rocking without engaging in the motor learning task (Fig. 7A, B). Sleep enhancement was achieved consistently throughout the 11-day experiment (mixed-effects analysis: condition effect: F(1,11) = 8.253, P = 0.015, Fig. 7C), resulting in a significant increase in overall sleep duration (unpaired t-test: P = 0.02, Fig. 7D) and a marked reduction in sleep fragmentation (unpaired t-test: P = 0.004, Fig. 7E) in SE mice compared to controls, with no effects on wake time during the dark phase (wake bouts: P = 0.7; normalize wake duration: P = 0.9, Supplementary Fig. 1). However, despite the clear impact of rocking on sleep, densitometric analysis of VGLUT1 did not show any significant differences in synaptic density between the groups (P = 0.63, Fig. 7F, G). Similarly, analysis of VGAT puncta, whether dispersed in the neuropil or perisomatic, revealed no changes between SE and S mice (Linear mixed model: neuropilar: P = 0.86; perisomatic: P = 0.48, Fig. 7H, I). To determine whether SE induced changes in synapse number outside the primary motor cortex, we measured VGLUT1 and VGAT puncta density in the stratum radiatum of the hippocampal CA1 region, which is critically involved in learning. This region was also selected due to prior evidence of sleep-dependent changes in synaptic density^23^. In our analysis, SE did not significantly alter the number of VGLUT1- or VGAT-positive puncta (Linear mixed model: VGLUT1: P = 0.14; VGAT (neuropilar): P = 0.46; VGAT (perisomatic): P = 0.84) in this hippocampal region (Fig. 7J–M).Fig. 7. Rocking enhances sleep but does not change the density of synapses in the motor cortex and CA1.A Experimental design. B Motion activity profiles of a S (upper panel) and SE (lower panel) mouse. C Sleep amount relative to baseline in the S and SE groups across the 11-day experiment. Averaged sleep amount (D) and sleep-to-wake transitions (E) in the S and SE groups. *P < 0.05, **P < 0.01. F Representative images of S and SE mice showing cortical VGLUT1 immunostaining. Scale bar: 10 µm. Quantitative estimation of VGLUT1(G), VGAT (neuropilar, H), and VGAT (perisomatic, I) positive puncta density in S and SE mice. From C to I, data are derived from S (n = 5) and SE (n = 8 for sleep analysis, n = 6 for immunohistochemistry) mice. J Representative images of S and SE mice showing CA1 stratum radiatum VGLUT1 immunostaining. Scale bar: 7 µm. Quantitative estimation of VGLUT1(K), VGAT (neuropilar, L), and VGAT (perisomatic, M) positive puncta density in S (n = 5) and SE (n = 6) mice. Bars represent mean ± std.

Thus, sleep enhancement without motor learning did not lead to detectable changes in synaptic density in either the motor cortex or the CA1 region of the hippocampus.

Discussion

Here, we demonstrated that vestibular stimulation through rocking over an eleven-day period reliably enhanced sleep in mice. Improvements in sleep quantity and quality were associated with accelerated motor learning in a task that typically requires several days of training to reach a performance plateau. Rocking with motor learning was linked to upregulation of genes involved in synaptic plasticity, as well as increased density of glutamatergic synapses in the motor cortex.

The idea that rocking promotes sleep is longstanding, rooted in intuitive practices such as rocking infants or resting in a hammock^11^. However, only recently has this form of sensory stimulation been subjected to rigorous scientific analysis. Recent studies have confirmed that rocking facilitates sleep onset and increases the time spent in NREM sleep, typically at the expense of wakefulness, with minimal effects on REM sleep or sleep intensity^13–15^. In rodent models, sleep enhancement via rocking is commonly achieved by applying low-frequency horizontal motion during their natural rest phase. To date, experiments have used continuous stimulation, and it remains an open question whether closed-loop systems, where rocking is triggered only during sleep, might lead to different and even more effective results.

We found that enhanced sleep improved motor learning and that the rate of improvement was related to the amount of sleep. In a previous study, we found that performance on the complex wheel task is highly sensitive to prior sleep, with sleep-deprived mice showing a markedly slower learning curve compared to well-rested controls^24^. While the detrimental effects of sleep deprivation on motor learning have been well documented, especially in humans^25–28^, studies investigating the behavioral benefits of sleep enhancement are still limited. To date, only one previous study has directly examined this relationship by enhancing sleep through acoustic stimulation and assessing motor performance using the single-pellet reaching task^29^. Our behavioral findings are consistent with that study and extend its conclusions by employing a different sensory modality, vestibular stimulation, and a more complex, dynamic motor task.

An important consideration is whether the enhanced performance of SE mice could reflect differences in wheel engagement rather than the effects of sleep itself. Although SE mice initiated running bouts more frequently, the total time spent running and distance covered did not differ significantly between groups. Moreover, these measures did not correlate with learning outcomes, in contrast to the clear relationship observed between sleep amount and performance improvement. Thus, it is unlikely that the improved learning observed in SE mice was simply due to increased physical activity or greater task exposure; rather, the data support the hypothesis that sleep enhancement directly facilitated the acquisition of this complex motor skill.

We also considered the possibility that differences in wake duration during the dark phase might have influenced motor learning. However, both EEG recordings and motion-based sleep/wake estimates indicated that the amount of wakefulness during the dark period was similar in SE and S mice. Nevertheless, we cannot exclude the possibility that the “quality” of wakefulness differed between the two groups, which could, in principle, have affected learning.

Finally, to further exclude nonspecific effects of vestibular stimulation, we performed a control experiment in which rocking was applied exclusively during wakefulness. This manipulation did not improve motor performance, supporting the interpretation that enhanced motor learning in SE mice was mediated by improved sleep rather than by the direct effects of rocking or increased wake activity.

Consistent with this interpretation, sleep-enhanced mice that showed improved motor performance also exhibited upregulation of several genes implicated in learning, memory consolidation, and synaptic function. Of note, some of the differentially expressed genes are recognized as key modulators of synaptic strength and plasticity, such as Grm2, Mef2c, Sytl2 and Drd1. For instance, Grm2 encodes for the metabotropic glutamate receptor (mGluR2), which plays a significant role in synaptic plasticity by modulating glutamate release and neuronal excitability^30,31^. Interestingly, deleting or inhibiting the Grm2/3 metabotropic glutamate receptor (mGluR2/3) in mice leads to alterations in sleep patterns, including reduced sleep duration and increased sleep fragmentation^32,33^. Sleep enhancement also promoted the expression of Mef2c, a member of the MEF2 family of transcription factors, which plays a crucial role in synaptic plasticity by regulating the number and function of synapses, and is involved in the development and refinement of neuronal circuits^34,35^. Another synaptic marker was Sytl2, a gene encoding for synaptotagmin-like protein, which is involved in RAB27A-dependent vesicle trafficking and secretion in neurons^36^. Drd1 coding for dopamine receptor 1 is implicated in motor learning, influencing motor sequence learning. Specifically, DRD1 activation enhances BDNF sensitivity and TrkB translocation and enhances glutamatergic transmission through the NMDA receptors, thus facilitating long-term plasticity. Moreover, acute sleep deprivation can induce changes in dopamine release and synaptic plasticity, including an increase in dendritic spine density in the medial prefrontal cortex (mPFC)^37–39^. Conditional knock-out of DRD1 in mPFC pyramidal neurons can abolish these dendritic spine density changes, highlighting the role of DA in these sleep deprivation-induced changes^39^. Another gene of interest was Adora2a, which encodes the adenosine A2A receptor. This receptor not only modulates glutamate release but also plays a central role in the homeostatic regulation of sleep and wake states^40,41^.

From a functional standpoint, gene enrichment analysis of the differentially expressed transcripts revealed significant clustering around biological processes related to glutamatergic synapses and synaptic plasticity. These transcriptomic findings were corroborated by immunohistochemical analyses, which demonstrated a selective increase in glutamatergic synapse density in the motor cortex of sleep-enhanced animals. To quantify excitatory synapses, we employed VGLUT1 immunostaining to label presynaptic glutamatergic terminals. Although this method does not directly assess synaptic strength, it is a widely accepted proxy for estimating synaptic density^42–44^. In contrast, no significant changes were observed in GABAergic synapse density, suggesting a selective enhancement of excitatory transmission.

The observed increase in glutamatergic synaptic density in the SE group is likely the result of improved motor learning rather than the direct effect of sleep enhancement. Indeed, sleep enhancement in the absence of motor learning did not lead to changes in synapse number within the motor cortex. Although we cannot exclude the possibility of more subtle changes in synaptic strength, such effects may have gone undetected due to the limitations of confocal microscopy. Specifically, quantifying changes in the size of punctate staining through immunofluorescence lacks sufficient resolution and reliability for capturing nuanced modifications in synapse structure, which usually requires analysis at the electron microscopy^45–47^. Further investigation, targeting the molecular and ultrastructural effects of rocking alone, will help determine whether this method of sleep enhancement also induces synaptic strength remodelling, as observed during baseline sleep^1,48–50^.

It is also possible that other brain regions are more sensitive to the effects of sleep enhancement when synapse number is considered. For example, a recent study using three-dimensional electron microscopy showed that, unlike in the motor cortex, sleep reduces the number of synapses in the CA1 stratum radiatum relative to short sleep deprivation^23^. In contrast, our data from the same hippocampal region did not reveal any difference in synapse density between S and SE mice, suggesting that rocking-induced sleep enhancement does not influence hippocampal synapse number under our experimental conditions. This apparent discrepancy with previous findings may arise from several factors. The earlier study reported effects of sleep only for non-perforated synapses, whereas our confocal approach did not distinguish between synapse subtypes. Moreover, our comparison involved two sleeping conditions, where the potential effect size of sleep-dependent downscaling is likely smaller and possibly below the detection threshold of our method. Finally, it is conceivable that differences in synapse number emerge primarily when sleep is compared with sleep deprivation, rather than between two conditions of normal or enhanced sleep.

Taken together, these findings suggest that the increased synaptic density observed in our primary experiment is not directly attributable to sleep enhancement itself, but rather to the motor learning experience facilitated by improved sleep. Within this framework, enhanced sleep appears to create a neurophysiological environment that is more favourable for learning, while it is motor skill acquisition that drives increases in synaptic density.

The biological mechanism underlying the promoting effect of sleep on motor learning has not been investigated in this study, but previous evidence has shown that this effect may involve the reactivation and synchronization of neural ensembles in motor-related brain regions during NREM sleep, particularly during sleep spindles and slow-wave activity, and in association with dopaminergic neuromodulation, as well as microstructural remodelling in the striatum and prefrontal cortex^51–57^.

Some aspects of the study should be considered when interpreting the results. First, baseline sleep measurements were always collected prior to the experimental phase, and their order was not randomized, which could have introduced minor order effects. Second, although our control experiments substantially limit nonspecific arousal, activity-related, and wake-time confounds, the absence of a dedicated sham-rocking condition modestly limits the strict attribution of vestibular specificity. While previous work showing that mice with defective otolith function fail to exhibit rocking-induced sleep enhancement supports a primary role for vestibular signaling^13^, additional sensory inputs (e.g., noise or vibration) cannot be fully excluded in the present setup. Third, although we controlled for differences in physical activity, other unmeasured factors, such as stress levels or wake quality, may have influenced the results. Fourth, synaptic changes were assessed using immunofluorescence and confocal microscopy, providing reliable estimates of synapse density but not capturing ultrastructural or subtle changes in synaptic strength; higher-resolution approaches such as electron microscopy would allow a more detailed characterization of synaptic modifications. Lastly, only male mice were studied, limiting the generalizability of our findings across sex and age.

In conclusion, our findings demonstrate that vestibular stimulation through rocking is an effective, non-invasive method to enhance sleep in mice. This sleep enhancement not only promoted longer and more consolidated sleep but also facilitated motor learning in a task requiring sustained training. Enhanced learning was accompanied by upregulation of genes involved in synaptic plasticity and a selective increase in excitatory synapse density within the motor cortex. Importantly, our control experiments showed that these neuroplastic changes were not directly induced by sleep enhancement alone, but rather resulted from the motor learning experience enabled by improved sleep quality.

Methods

Animals

Male C57BL/6 mice were used for this study. Mice were maintained on a 12 h light/12 h dark cycle (lights on at 8:00 am, lights off at 8:00 PM) with food and water ad libitum. At 40 days of age (post-natal day (P)40), mice were selected for the experiment and housed individually in a cage composed of four separate chambers. Each chamber was a 25 × 25 cm square and hosted a single mouse. Chamber walls were transparent and equipped with small holes to facilitate the diffusion of smells and limit the effects of social isolation. In each chamber, mice were constantly video-monitored. Mice were allowed to habituate to this new environment for a week before starting the sleep manipulation. All procedures were in accordance with the guidelines laid down by the European Communities Council Directives (2010/63/EU) for the care and use of laboratory animals under an approved protocol (609/2023-PR) by the Veterinary Health Dept of the Italian Ministry of Health.

Experimental design

Mice were randomly assigned to one of two groups: a control group allowed to sleep undisturbed (S) and a sleep enhancement group (SE). Groups were balanced for body weight. The study included a pilot polysomnographic experiment and three main experimental conditions, each preceded by a 5-day habituation period followed by 11 days of experimentation.

In the polysomnographic experiment, electroencephalographic (EEG)/electromyographic (EMG)-implanted mice underwent one day of baseline recording followed by one day of rocking, restricted to the light phase (8:00 AM–8:00 PM).

In the first experimental condition, designed to assess the impact of rocking on sleep and motor learning, SE mice were exposed to 12 h of rocking during the light phase (8:00 AM–8:00 PM), while S mice remained undisturbed. During the dark phase, both groups had access to a complex running wheel.

In the second condition, aimed at isolating the effect of rocking per se on motor learning, one group (Rocking ON) was exposed to 4 h of rocking at the beginning of the dark phase (8:00 PM–12:00 AM). During this period, an operator ensured wakefulness by gently tapping the cages only when animals showed clear signs of drowsiness. Control mice (Rocking OFF) did not receive rocking stimulation but were monitored under the same procedure, including cage tapping when necessary, to ensure comparable handling and wakefulness across groups. Importantly, the need for such intervention was rare, as mice are typically active during this phase of the dark cycle. All animals then had access to the complex wheel for the remaining 8 h of the dark phase.

In the third condition, designed to dissociate the effect of rocking on sleep from potential wheel-running effects, SE mice underwent rocking during the light phase (8:00 AM–8:00 PM), while S mice remained undisturbed. Unlike the first condition, no wheel task was introduced during the dark phase.

Sleep enhancement

Sleep enhancement was carried out by placing the mouse cage on a flat reciprocal shaker that allows horizontal movements (HS 260 Control, IKA, Switzerland). Rocking occurred at 1 Hz with a fixed peak displacement of ±20 mm. These parameters were chosen based on prior evidence indicating that they maximize the sleep-enhancing effect while preserving normal sleep architecture^13^. The control group was housed under identical conditions and placed on the same shaker platform; however, the apparatus remained stationary throughout the experimental period.

Polysomnographic recordings

A subset of mice (n = 5) was implanted with electrodes to assess the efficacy of the rocking procedure in this model. Briefly, mice were anesthetized with isoflurane (2.5% for induction, 1.5% for maintenance) and implanted with electrodes over the frontal and parietal cortices to monitor EEG activity. An additional electrode was placed above the cerebellum to serve as a reference. To record muscle activity, silver wire electrodes were inserted into the neck muscles bilaterally for EMG acquisition. Following surgery, mice were housed individually in transparent Plexiglas cages for the duration of the experiment.

After a recovery period, recordings began at the onset of the light phase. Continuous data acquisition was performed using the Open Ephys system for 48 h. EEG signals were band-pass filtered between 0.1–40 Hz, while EMG signals were filtered between 10–50 Hz. The experimental protocol included one baseline day followed by one day of sleep enhancement through 1 Hz rocking, applied during the light phase.

Sleep scoring was conducted offline by visual inspection of 4-second epochs using SleepSign software^58^. Behavioral state classification and further analysis were performed using custom MATLAB scripts.

Video tracking of sleep and wake motion activity

Mice motion activity was continuously detected with infrared cameras for the entire experiment duration and stored in real-time. In designing the shaking apparatus, care was taken to ensure that cage motion did not compromise behavioral tracking. Cameras were mounted above each cage and moved synchronously with it during rocking, enabling the detection algorithm to precisely monitor mouse locomotion without artifacts from cage displacement relative to the surrounding environment.

Motion activity was quantified by custom-made video-based motion detection algorithms (MATLAB MathWorks) with a time resolution of 1 s. As previously shown, this method was not able to distinguish NREM sleep from REM sleep, but it consistently determined the total sleep time in mice with more than 90% accuracy with respect to EEG chronic recordings^20,21^.

Sleep fragmentation was quantified using a sleep fragmentation index derived from motion activity, defined as the number of sleep-to-wake transitions per hour. For each mouse, total sleep and wake time and the sleep fragmentation index were normalized to baseline values derived from the final 3 days of the habituation period. This normalization accounted for potential variability in camera sensitivity across setups, facilitated pooled data analysis, and improved the detection of small effect sizes, thereby reducing the number of animals required for the study.

Complex wheel task

In experiments assessing motor performance and learning, all mice had access to a complex wheel during the dark period (8 PM to 8 AM). This wheel has a total number of rungs are 38 with a wheel radius of 6.35 cm. In the task, we randomly removed 16 rungs to make 2 identical complex sequences of rungs in one rotation. This arrangement required mice to adapt their paw placement to maintain proper coordination and balance while running.

Under normal conditions, mice typically learn to navigate the complex wheel and reach peak running speed within approximately one week. In this study, the training period was extended to 11 days to ensure that all animals, including slower learners, had sufficient time to achieve stable performance and fully consolidate the acquired motor skill. The wheel used in this study was specifically designed and 3D-printed for the project.

The wheel was equipped with an Arduino Mega 2560 microcontroller, which recorded the absolute position of the wheel with a frequency of 100 Hz. This data was analyzed offline to extract both running speed and usage patterns. For each animal, all running bouts were identified and characterized. A running bout was defined as a continuous period of activity lasting at least 5 s, with significant breaks (>5 s) used to separate consecutive bouts. For each bout, the average speed was calculated as the total distance run divided by the bout duration, providing a measure of the animal’s running “pace.” The daily average speed was defined as the mean of all bout average speeds, while the maximum speed was defined as the highest average speed achieved in any running bout within a given day, ensuring that the speed was maintained for at least 5 s. The total daily distance was computed as the sum of the distances run across all bouts. Motor learning was quantified by calculating the ratio between the maximum speed achieved on the last day of training and that recorded on the first day the mouse accessed the wheel.

Tissue collection for RNA-sequencing

S and SE mice were sacrificed between 8:00 and 9:00 AM on the 12th day, following completion of the final complex wheel session. Each mouse brain was quickly extracted, and the motor cortex was quickly dissected on a cold platform and immediately stored at −80 °C. Frozen tissue was then sent to an external facility (AZENTA Life Sciences) for standard bulk RNA sequencing.

RNA-Sequencing analysis

All RNA-sequencing analyses were conducted using the Galaxy Europe platform (https://usegalaxy.eu). During the data processing and quality control phase, the quality of the raw sequencing reads was initially assessed using FastQC, which provided an overview of sequence quality scores, GC content distribution, and potential adaptor contamination. To ensure high-quality data for downstream analysis, we used Trimmomatic to identify and eliminate low-quality bases, adaptor sequences, and other artifacts. Only high-quality reads passing the filtering criteria were retained for further analysis. The next step was read alignment and gene expression quantification. The cleaned reads were aligned to the reference genome (mm39 for the mouse genome), ensuring accurate mapping of reads to gene loci. The alignment quality was verified using SAM tools to remove improperly mapped reads. Gene expression was then quantified using Feature Counts, which assigned reads to annotated genes based on the corresponding GTF annotation file. The resulting count matrix was used for differential expression analysis. To identify genes that were differentially expressed between the S and SE groups, DESeq2 was used. Gene expression levels were normalized, and statistical tests were performed to detect significant differences. Genes using the threshold of 0.05 on the adjusted p-value (Benjamini-Hochberg correction) and a log2FC > 0.38 or <−0.38 were considered significantly differentially expressed. Moreover, to determine the biological significance of differentially expressed genes, functional enrichment analysis was conducted using STRING^22^. This analysis helped identify enriched Gene Ontology (GO) terms and protein-protein interaction networks related to the differentially expressed genes. Enrichment analysis provided insights into the functional roles and potential regulatory mechanisms associated with gene expression changes.

Immunohistochemistry

At the end of the experiment (8AM-9AM of the 12th day), mice were sacrificed under general anesthesia and perfused with a solution of phosphate buffer and paraformaldehyde at 4%. Mice brain tissue was allowed to fix for a day at 4 °C and then was cut on a vibratome in 40 μm coronal sections. Sections were then rinsed in a blocking solution [10% normal goat serum (NGS) and 0.02% Triton X-100] for 1 h and incubated at room temperature for 2 h then overnight (4 °C) in the PBS solution containing anti-VGLUT1 (1:1000, AB5905, Millipore) and anti-VGAT (1:500, 131 011, SYSY). The following day, sections were washed with PBS three times then block with 10% NGS for 20 min, then probed with secondary antibodies: anti-guinea pig Alexa Fluor 488 (1:1000), anti-mouse Alexa Fluor 594 (1:1000). Sections were mounted on slides and cover slipped with a medium containing DAPI.

Image acquisition and analysis

Z-stack images were acquired using a Nikon Eclipse confocal microscope (Tokyo, Japan). For each section, five microscopic fields were randomly selected and captured as 1024 × 1024-pixel images (Z-step: 300 nm) from layer 5 of the mouse motor cortex using a UPlan FL N 60X oil objective. To enhance the signal-to-noise ratio, each image was averaged over four consecutive scans during acquisition. The experimenter was blinded to group assignments during both image acquisition and subsequent quantitative analysis. Fluorescence imaging employed excitation lasers at 405 nm for DAPI (nuclear marker), 488 nm (FITC) for VGLUT1, and 561 nm (Texas Red) for VGAT. Image analysis was performed semi-automatically using custom-written macros in ImageJ/Fiji. From each Z-stack, the image with the best resolution was selected for analysis, and the non-specific background signal was excluded using standardized thresholding procedures.

Quantification of VGLUT1-positive puncta was performed within neuropil areas, excluding cell bodies, to assess excitatory synapse density. VGAT-positive puncta were quantified using two approaches: (1) across the neuropil and (2) in the perisomatic region, defined as a 1.50 µm band surrounding the DAPI-stained soma outline.

Statistics and reproducibility

Motion and complex wheel data were analysed with parametric statistics carried out by using GraphPad Prism9 (GraphPad Software, USA) and R. Alpha was set at 0.05 and appropriately corrected for multiple comparisons. Statistical analysis of synaptic data was conducted using a linear mixed-effects model, with the mice as a random variable and condition as a fixed variable. Model parameters were estimated via maximum likelihood using the lme4 package in R, and statistical significance was assessed via likelihood ratio tests. Statistical details are summarized in the Supplementary Data file 1.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Transparent Peer Review File Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Reporting Summary