VTA dopaminergic neuronal activity during NREM sleep is modulated by learning and facilitates motor memory consolidation

Authors: Bibi Alika Sulaman, Eric Chen, Aaron Crane, Sangjin Lee, Gideon Rothschild, Ada Eban-Rothschild

Abstract

Memory consolidation enables animals to draw on past experiences to guide future behavior. This process involves system- and synaptic-level changes and predominantly occurs during sleep. While hippocampal-cortical circuits are well studied in this context, the contributions of other systems, including dopaminergic circuits—key players in learning-related processes—remain poorly understood. Ventral tegmental area dopaminergic (VTA^DA^) neurons are active during sleep and suggested to participate in memory consolidation processes; however, causal evidence for this is lacking. Using calcium-dependent fiber photometry, electrophysiology, and chemogenetic and optogenetic manipulations across learning paradigms, we explore the functions of VTA^DA^ neuronal activity during sleep. We show that VTA^DA^ activity during nonrapid eye movement sleep is experience dependent, is enhanced by motor skill and associative learning, facilitates motor skill memory consolidation, and exhibits a motor learning–dependent increase in temporal coordination with cortical spindle oscillations. Our findings uncover a previously unidentified function for VTA^DA^ neurons in consolidating memories during sleep, advancing understanding of this central neuromodulator’s functions in regulating fundamental biological processes.

INTRODUCTION

The ability to learn from experiences relies on memory consolidation, a process that stabilizes and transforms labile neural representations of experiences into lasting memories and knowledge. Memory consolidation involves system-level changes in functional connectivity, neuronal reactivation, synaptic modifications, and transcriptional alterations, with the hippocampus and cortex playing central roles in mammals (1–5). These processes primarily occur during the postlearning (“offline”) period, with sleep long recognized as a critical time for them to occur (1–4).

A central neurophysiological mechanism of memory consolidation is neuronal reactivation, in which firing patterns displayed during wakeful behavior are replayed during subsequent offline periods. Reactivation often occurs during specific phases of synchronous neural oscillations, namely, sharp wave ripples (SWRs; 150- to 250-Hz oscillations lasting 50 to 100 ms), spindles (waxing and waning 10- to 15-Hz oscillations lasting 0.5 to 2 s), and slow oscillations (<1-Hz cortical oscillations), whose coordination is thought to facilitate consolidation during sleep (1, 6). Both the timing of reactivation within these oscillations and the nested structure of SWRs, spindles, and slow oscillations have been causally linked to memory consolidation (7–9).

While initially described and extensively studied in the hippocampus (10–13), emerging evidence suggests that neuronal reactivation events also occur in other brain regions encoding various aspects of waking experiences. These include several cortical regions (14–18) and subcortical structures, such as the amygdala and ventral striatum (19, 20). However, the specific contributions of these circuits to sleep-dependent memory consolidation remain less understood. Recent evidence suggests that neuromodulator signaling during sleep, particularly noradrenaline, regulates memory-related neural oscillations–especially spindles–and can influence consolidation (21–23). However, how distinct neuromodulator systems contribute to memory consolidation and interact with synchronous oscillations during sleep remains unclear.

Dopamine signaling, a central neuromodulator system and crucial component of learning-related processes, plays a key role in encoding internal representations of experiences, including rewarding and aversive stimuli, reward-predictive cues, action initiation, movement, and behavioral choices (24–29). In addition, it plays a central role in associative and motor skill learning (30–35). Traditionally, dopamine and other neuromodulators were thought to be absent during sleep (36, 37). However, recent evidence suggests that, although their levels are reduced, neuromodulator activity persists during sleep and may play diverse roles (21–23, 38). Ventral tegmental area dopaminergic (VTA^DA^) neurons exhibit intermittent activity during nonrapid eye movement (NREM) sleep and prominent activation during rapid eye movement (REM) sleep (39, 40). Various studies suggest that VTA^DA^ neurons contribute to memory consolidation processes during sleep (41–44). For example, activity in VTA^DA^ neurons during wakefulness affects hippocampal reactivation during NREM sleep (41). Putative VTA^DA^ neurons also reactivate taste-related activity during NREM sleep (42). Activity in VTA projections, including both dopaminergic and nondopaminergic fibers, to the nucleus accumbens (NAc) increases during offline cortical reactivation events associated with rewards (44). In addition, pharmacological blockade of dopamine signaling after learning impairs long-term memory formation (43). Despite these findings, the precise roles of dopamine in consolidating different types of learning experiences, the specific time window of its activity, the sleep/wake state involved, and the precise source of dopamine remain unclear.

To address these gaps and probe the functions of VTA^DA^ neuronal activity during sleep, we combine calcium-dependent fiber photometry, polysomnographic and multisite electrophysiological recordings, chemogenetic and optogenetic manipulations, a battery of behavioral and learning tasks, and machine learning–based analysis. We demonstrate that VTA^DA^ neuronal activity during NREM sleep is experience-dependent, increased by motor skill and associative learning, and facilitates motor memory consolidation. In addition, we show that motor skill learning enhances coordination between VTA^DA^ activity and motor cortex spindle oscillations during NREM sleep, with spindles preceding VTA^DA^ neuronal activation.

RESULTS

VTA^DA^ activity during NREM sleep is selectively modulated by experience

We first aimed to determine whether VTA^DA^ neuronal activity during sleep is modulated by waking experiences. To test this, we infused viral vectors carrying the Cre-inducible genetically encoded calcium indicator GCaMP6f into the VTA of dopamine transporter (DAT)–internal ribosomal entry site (IRES)–Cre male and female adult mice and implanted electroencephalographic (EEG)/electromyographic (EMG) electrodes for polysomnographic recordings and fiber optic probes for calcium-dependent recordings (Fig. 1A and fig. S1A). Following recovery and habituation, we subjected the mice to two highly salient experiences at the end of the dark phase [zeitgeber time (ZT) ~22]: an enriched multisensory experience (movie S1) and a predator cue experience (movie S2) and recorded population calcium activity from VTA^DA^ neurons using fiber photometry, along with EEG/EMG signals, during the subsequent sleep phase (Fig. 1, B and C). The enriched multisensory experience involved exposure to an environment containing multiple novel stimuli, including palatable foods, bedding, scents, various enrichment objects, and a running wheel. The predator cue experience involved exposure to an environment with sensory cues from a rat’s home cage, such as soiled bedding and fecal boli. Both experiences provided opportunities for different types of learning. As a control treatment, we gently handled the mice in their home cages to maintain wakefulness and match circadian sleep times between groups (movie S3).

We found a significant increase in VTA^DA^ population activity, reflected in ΔF/F, transient rate, and transient rate × mean transient amplitude, specifically during NREM sleep in the first 1 to 2 hours after exposure to the enriched multisensory experience compared to the control treatment (Fig. 1D). In contrast, there was no significant change in VTA^DA^ neuronal activity during REM sleep or wakefulness in the hours following the enriched multisensory experience (fig. S2, A to C). Sleep architecture was also not significantly affected by the enriched multisensory experience compared to the control treatment (Fig. 1E and fig. S2D), although some animals showed a nonsignificant reduction in total NREM sleep duration—without changes in episode duration—during the first hour after sleep onset (Fig. 1E). However, delta power during NREM sleep was increased following the enriched multisensory experience compared to the control treatment (fig. S2E). Since VTA^DA^ neurons show transient activation at sleep-to-wake transitions (39), we wondered whether the NREM sleep VTA^DA^ transients detected in our analyses might coincide with microarousals. To test this, we examined EMG power during NREM sleep and wake-associated calcium transients (fig. S3, A and B). EMG activity around NREM sleep transients was unchanged (fig. S3A), refuting the possibility that the observed changes in transient rate reflect misclassification of sleep/wake state. To test whether the increase in VTA^DA^ activity during NREM sleep reflected spontaneous day-to-day fluctuations, we compared VTA^DA^ population activity across three baseline days and found no significant differences (fig. S3C), suggesting that the enriched multisensory experience, specifically, modulates VTA^DA^ activity during NREM sleep.

Next, we examined VTA^DA^ population activity following the predator cue experience and found no significant changes compared to the control treatment during NREM sleep, REM sleep, or wakefulness (Fig. 1F and fig. S2, F to H). However, NREM sleep was significantly reduced, and wakefulness increased in the first hour following the predator cue experience (Fig. 1G and fig. S2I). This increase in wake duration, despite unchanged VTA^DA^ activity, further suggests that the observed VTA^DA^ activation during NREM sleep following the enriched multisensory experience is not a by-product of increased wakefulness. Last, the predator cue experience did not alter the EEG power density during postexperience NREM sleep or REM sleep, but it reduced delta power during wakefulness (fig. S2J). Together, these findings suggest that VTA^DA^ population activity during sleep is modulated by enriched, multisensory salient experiences, but not by predator-related salient experiences, and that NREM sleep represents a time window during which experience-dependent processes occur in VTA^DA^ neurons.

VTA^DA^ activity during NREM sleep is modulated by motor skill and place-reward learning

Next, we aimed to determine whether the increase in VTA^DA^ activity during NREM sleep is induced by learning processes (Fig. 2). To explore this, we established two learning tasks that reflect aspects potentially learned during the enriched multisensory experience. Specifically, we focused on motor skill learning and place-reward association. In addition, we decided to further investigate the relationship between VTA^DA^ activity during sleep and negative valence learning using a cued fear conditioning paradigm.

To evaluate motor skill learning, we used a modified balance beam task designed to assess motor coordination and balance (Fig. 2C, fig. S4A, and movie S4) (45, 46). Previous studies have shown that mouse performance on a similar task improves across days (45). Briefly, mice were placed on two parallel steel rods and monitored as they moved toward their old home cages. In a pilot experiment, we observed pronounced motor improvement in mice on the second day of the task, as evidenced by a significant reduction in task completion time and a substantial increase in speed while moving above the rods (fig. S4A and movie S4). To assess place-reward association, we used a four-arm maze with visual and tactile cues distinguishing the different arms (Fig. 2E, fig. S4B, and movie S5). Briefly, we monitored the mice as they explored the maze, identifying their least preferred arm on day 1 and baiting this arm with chocolate rewards on day 2. In a pilot experiment, we found that on the test day (day 3), when no reward was provided, mice showed multiple indicators of place-reward learning, including increased time spent in and more visits to the previously least preferred arm, as well as reduced latency to revisit and fewer visits to other arms before revisiting the previously least preferred arm (fig. S4B). These alterations were not observed in control mice that explored the maze for the same amount of time but received chocolate rewards outside of the maze (fig. S4B). For cued fear conditioning, we exposed mice to either a weak or strong tone-shock pairing in a fear conditioning chamber on day 1. On day 2, we observed a significant increase in their freezing behavior in a novel chamber in response to tone presentations (Fig. 2G).

We then subjected DAT-IRES-Cre female and male adult mice, transduced with GCaMP6f in VTA neurons, and implanted with EEG/EMG electrodes and fiber optic probes, to learning tasks at the end of the dark phase (~ZT 22; Fig. 2, A and B, and fig. S1B). Subsequently, we simultaneously recorded EEG/EMG signals and population calcium activity from VTA^DA^ neurons during the sleep phase (Fig. 2B). Since we identified the first 2 hours following sleep initiation as the time window during which VTA^DA^ neurons exhibit experience-dependent activity changes (Fig. 1D), we focused our analysis on this period.

We found a significant increase in VTA^DA^ population activity—reflected in transient rate × mean transient amplitude, transient rate, and ΔF/F—during NREM sleep but not during wakefulness or REM sleep following motor skill learning compared to the control treatment (Fig. 2D and fig. S5, A and B). Motor learning also induced a significant overall change in the proportion of sleep/wake states during the first hour after sleep initiation compared to gentle handling controls (fig. S6A). However, post hoc comparisons showed no significant differences in NREM sleep time (fig. S6A) or NREM sleep episode duration (fig. S6B). Similarly, we found a statistically significant interaction effect between experience and frequency for NREM sleep power; however, post hoc comparisons showed no significant differences in specific frequency bands when comparing power following the different experiences (fig. S6C). Similar to motor learning, we found a significant increase in VTA^DA^ transient rate × mean transient amplitude during NREM sleep following place-reward association, accompanied by a nonsignificant trend in ΔF/F (P = 0.056; Fig. 2F). During wakefulness after place-reward association, there was a significant decrease in VTA^DA^ transient rate, with no significant change during REM sleep (fig. S5, C and D). Place-reward association did not result in significant alterations in sleep/wake architecture or EEG power density (fig. S6, D to F). VTA^DA^ activity during NREM and REM sleep following both weak and strong cued fear conditioning was not altered compared to the control treatment (Fig. 2H and fig. S5E), consistent with our observations after the predator cue experience (Fig. 1E). In contrast, during wakefulness, we observed a significant increase in ΔF/F and a nonsignificant trend toward increased transient rate × transient amplitude (fig. S5F). Fear conditioning did not significantly affect the total time spent in different sleep/wake states; however, NREM sleep episodes were significantly shorter than in controls, and delta power during wakefulness was reduced (fig. S6, G to I). Together, these findings suggest that VTA^DA^ population activity during NREM sleep is selectively modulated by prior motor skill and place-reward learning but not by cued fear conditioning. Moreover, these activity changes are not linked to consistent alterations in sleep architecture or EEG power density.

Since our learning paradigms included additional salient factors that may not be directly related to learning, such as motor activity and reward consumption, we aimed to further explore their potential contribution to the increased population activity during subsequent sleep. To this end, we recorded VTA^DA^ population activity following simpler one involving an environment with only a running wheel and another focused solely on the chocolate reward, delivered directly to the home cage. In addition, we monitored VTA^DA^ population activity during sleep following the fourth day of training on the balance beam motor task, a time point at which motor skill had plateaued, with no further improvement observed (Fig. 2C). As before, we exposed VTA^DA^-GCaMP6f–expressing mice to these experiences at the end of the dark phase (~ZT 22) and recorded their neuronal activity during the subsequent sleep phase (Fig. 2, I and J, and fig. S1, B and C). We found no significant alterations in VTA^DA^ population activity during NREM sleep, wakefulness, or REM sleep following motor activity on the rods as experts (Fig. 2K and fig. S5, G and H), motor activity on the wheel (Fig. 2L and fig. S5, I and J), or reward consumption (Fig. 2M and fig. S5, K and L). These findings strongly suggest that certain learning experiences increase VTA^DA^ population activity specifically during NREM sleep.

VTA^DA^ activity during the postlearning sleep phase promotes motor memory consolidation

To test the causal role of VTA^DA^ neuronal activity during NREM sleep in motor memory consolidation, we first used a chemogenetic approach (Fig. 3). This allowed us to assess the necessity of VTA^DA^ activity during the postlearning sleep phase, although not specifically restricted to NREM sleep. In a second set of experiments (Fig. 4 and fig. S8; see below), we then examined the function of VTA^DA^ activity specifically during NREM sleep.

We bilaterally infused an adeno-associated virus (AAV) encoding the Cre-inducible inhibitory muscarinic G protein–coupled receptor hM4Di (or one encoding the mCherry reporter protein as a control) into the VTA of DAT-IRES-Cre male and female adult mice and implanted EEG/EMG electrodes (Fig. 3A and fig. S7A). After recovery and habituation, we trained the mice on the balance beam motor task at the end of the dark phase (~ZT 22) and then suppressed VTA^DA^ neuronal activity during the subsequent sleep phase with two clozapine N-oxide (CNO) the first timed to the initiation of presleep behaviors and the second given 3 hours later (Fig. 3B). As an additional control, we administered saline to both mCherry- and hM4Di-expressing mice (Fig. 3B and fig. S7A). Because the motor skill learning task cannot be repeated on the same animals, we used separate cohorts of mice for each treatment.

In all three control groups—saline-treated mCherry-expressing mice, saline-treated hM4Di-expressing mice, and CNO-treated mCherry-expressing mice—mice showed a significant reduction in task completion time and a significant increase in speed above the rods between the first and second days (Fig. 3, C and D), consistent with our previous findings (Fig. 2C and fig. S4A), reflecting motor improvement and suggesting proper consolidation of the motor skills they had learned. However, inhibition of VTA^DA^ neurons during the sleep phase following learning impaired motor memory consolidation, as CNO-treated hM4Di-expressing mice showed no significant improvement in task completion time or speed above the rods between the first and second days (Fig. 3, C and D).

Notably, after two additional days of training without sleep phase–specific inhibition of VTA^DA^ neurons, CNO-treated hM4Di-expressing mice showed motor improvement on the balance beam task, evidenced by a significant reduction in task completion time and a significant increase in speed above the rods between the first and fourth days (fig. S7B). These results suggest that the chemogenetic inhibition of VTA^DA^ neurons did not induce long-term functional alterations that would impair subsequent learning or memory consolidation. Moreover, this finding illustrates that, despite the apparent bias toward faster completion times on the first day (Fig. 3D), hM4Di-expressing mice previously treated with CNO were still able to improve with continued training.

Since VTA^DA^ neuronal activity can influence sleep/wake states (39), we examined whether chemogenetic inhibition might have suppressed sleep, thereby indirectly affecting motor memory consolidation. Rather than suppressing sleep, inhibition of VTA^DA^ neurons during the light phase promoted NREM sleep, consistent with our previous findings during the dark phase (39), while leaving REM sleep unaffected (fig. S7C). These results support the premise that VTA^DA^ neuronal activity during the postlearning sleep phase is critical for motor memory consolidation.

Since VTA^DA^ neurons are strongly implicated in regulating motivation and movement during wakefulness [e.g., (47)], we sought to determine whether the lack of motor skill improvement was due to reduced motivation or movement capacity on the day following inhibition, rather than impaired motor memory consolidation. To test this, we inhibited VTA^DA^ neuronal activity during the sleep phase in expert mice 1 day following the fourth day of training on the balance beam task (Fig. 3E). At this point, VTA^DA^ neuronal activity during NREM sleep following the balance beam experience was no longer elevated (Fig. 2K), and CNO-treated mCherry-expressing control mice no longer showed motor improvement on the task (Fig. 3, F and G). If inhibition of VTA^DA^ neurons during sleep affected next-day motivation, motor capacity, or exploration, then we would expect to see a significant impairment in performance, indicated by an increased time to complete the task and reduced speed in CNO-treated hM4Di-expressing expert mice. However, we found no significant change in task completion time and speed above the rods in expert mice pre– versus post–CNO administration, in both hM4Di- and mCherry-expressing mice (Fig. 3, F and G). This finding strongly suggests that the lack of motor improvement observed following VTA^DA^ inhibition during the sleep phase after learning was not due to impaired motor performance or motivation, underscoring the importance of VTA^DA^ neuronal activity during the sleep phase for motor memory consolidation.

VTA^DA^ activity specifically during postlearning NREM sleep promotes motor memory consolidation

After establishing the importance of the postlearning sleep phase for dopamine’s memory function (Fig. 3), we proceeded to determine whether activity specifically during postlearning NREM sleep—a state during which VTA^DA^ neurons show learning-dependent increases in activity (Figs. 1 and 2)—contributes to motor memory consolidation. To suppress activity in VTA^DA^ neurons with high temporal and spatial resolution, we used an optogenetic inhibition approach, initially using the soma-targeted Guillardia theta anion-conducting channelrhodopsin 2 (stGtACR2) construct (fig. S8) (48) and later the improved step-function bistable inhibitory channelrhodopsin (SwiChR++) construct (Fig. 4) (49).

We first sought to verify the effective inhibition of VTA^DA^ neurons during NREM sleep through optogenetic manipulation by measuring changes in extracellular dopamine levels in the NAc, a major target region of the VTA. To this end, we bilaterally infused an AAV encoding either stGtACR2 or SwiChR++, in a Cre-inducible manner, into the VTA of DAT-IRES-Cre mice and unilaterally infused an AAV encoding the G-protein-coupled receptor activation-based dopamine biosensor GRABDA into the NAc (Fig. 4A and fig. S8A). In addition, we implanted dual fiber optic probes above the VTA for optogenetic inhibition, a fiber optic probe above the NAc for fiber photometry recordings, and EEG/EMG electrodes (Fig. 4A and fig. S8A). To sustain neuronal inhibition throughout NREM sleep episodes, we delivered either a continuous low-intensity (~1 mW) 473-nm blue light pulse for stGtACR2-expressing mice (fig. S8B) or a 1-s 473-nm blue light pulse (~8 mW) every 30 s, followed by a 1-s 635-nm red light pulse (~5 mW) to terminate the inhibition, for SwiChR++-expressing mice (Fig. 4B). Light delivery began within 15 s of NREM sleep onset and ended within 4 s of its termination (Fig. 4B and fig. S8B). Using both viral strategies, we observed a significant reduction in extracellular dopamine levels in the NAc, measured via fiber photometry, during NREM sleep episodes with optogenetic inhibition compared to those without it (Fig. 4, C and D, and fig. S8, C and D). This finding supports the effectiveness of our strategy in suppressing VTA^DA^ neuronal activity during NREM sleep.

To determine whether VTA^DA^ neuronal activity specifically during NREM sleep facilitates memory consolidation, we first bilaterally infused an AAV encoding the Cre-inducible stGtACR2 (or mCherry as control) into the VTA of DAT-IRES-Cre male and female adult mice (figs. S8E and S9A). We also implanted the mice with dual fiber optic probes above the VTA and EEG/EMG electrodes (figs. S8E and S9A). After recovery and habituation, the mice were trained on the balance beam task at the end of the dark phase (~ZT 22) and then subjected to the optogenetic manipulation during NREM sleep episodes for the first 6 hours following sleep onset during the subsequent light phase (fig. S8F). Light delivery began within 15 s of NREM sleep onset and ended either within 4 s of its termination or after 5 min of laser exposure. We successfully targeted ~85% of NREM sleep time in both mCherry-expressing (83.85 ± 3.3%) and stGtACR2-expressing (86.36 ± 0.94%) mice, with minimal off-target manipulations during REM sleep and wakefulness (fig. S8G).

Unlike our previous control mice (Fig. 3D and fig. S4A), the mCherry-expressing control group did not exhibit the expected significant reduction in task completion time between the first and second days on the motor task (fig. S8, H and I). However, they did show a significant increase in speed above the rods across the days (fig. S8J). Optogenetic inhibition of VTA^DA^ neurons during NREM sleep following learning significantly reduced next-day motor performance in stGtACR2-expressing mice compared to mCherry control mice, with only 37.5% (three of eight) of stGtACR2-expressing mice completing the task on the second day, compared to 100% (six of six) of mCherry control mice (fig. S8K). In addition, stGtACR2-expressing mice subjected to NREM sleep–specific optogenetic inhibition of VTA^DA^ neurons after learning showed no significant increase in speed above the rods between the first and the second days on the task (fig. S8L). Notably, both mCherry- and stGtACR2-expressing mice were exposed to comparable amounts of light (fig. S8G) and implanted with the same probes, suggesting that these factors do not account for the lack of motor improvement observed in stGtACR2-expressing mice on the test day. Together, these findings suggest that VTA^DA^ neuronal activity during NREM sleep facilitates motor memory consolidation. However, the reduced performance in the control group and the magnitude of the observed effect in stGtACR2-expressing mice raised concerns that factors related to our experimental methodology—such as prolonged light delivery, potential damage from the dual fiber optic probes, or tethering to multiple cables in the home environment—may have influenced motor performance or learning.

To address the possibility that prolonged light delivery contributed to our observed effects, we repeated the optogenetic inhibition experiment using an alternative viral construct, SwiChR++, which requires only short light pulses for effective inhibition. We bilaterally infused an AAV encoding SwiChR++, or mCherry as a control, into the VTA of DAT-IRES-Cre male and female adult mice and implanted dual fiber optic probes above the VTA along with EEG/EMG electrodes (Fig. 4E and fig. S9B). We trained the mice on the balance beam task at the end of the dark phase (~ZT 22) and then suppressed VTA^DA^ neuronal activity specifically during NREM sleep episodes for the first 6 hours following sleep onset during the subsequent light phase (Fig. 4F). Optogenetic inhibition began within 15 s of NREM sleep onset and ended within 4 s of its termination. We successfully targeted ~90% of NREM sleep time in both mCherry-expressing (90.5 ± 1.1%) and SwiChR++-expressing (88.6 ± 1.4%) mice, with minimal off-target manipulations during REM sleep and wakefulness (Fig. 4G). Notably, EEG power density during inhibited NREM sleep episodes was not significantly altered by optogenetic inhibition of VTA^DA^ neurons (Fig. 4H), consistent with previous findings using both chemogenetic (39) and optogenetic (50) inhibition approaches.

We found that mCherry control mice subjected to NREM sleep–specific optogenetic manipulation showed a significant reduction in task completion time and a significant increase in speed above the rods between the first and second days on the task (Fig. 4, I and J), suggesting proper consolidation of the motor skill they had learned. In contrast, inhibition of VTA^DA^ neurons specifically during NREM sleep after learning impaired motor memory consolidation, as SwiChR++-expressing mice subjected to NREM sleep–specific manipulation showed no significant reduction in task completion time and a diminished increase in speed above the rods between the first and second days compared to the mCherry control mice (Fig. 4, I and J). These findings, along with our previous results from optogenetic manipulations using the stGtACR2 construct (fig. S8, K and L) and chemogenetic manipulations (Fig. 3, C and D), strongly support the premise that VTA^DA^ neuronal activity during NREM sleep following learning facilitates motor memory consolidation.

Notably, after two additional days of training without NREM sleep–specific inhibition of VTA^DA^ neurons, SwiChR++-expressing mice showed motor improvement on the balance beam task (fig. S7D). This was evidenced by a significant reduction in task completion time and a significant increase in speed above the rods between the first and fourth days (fig. S7D). These results suggest that, similar to chemogenetic inhibition (fig. S7B), optogenetic inhibition of VTA^DA^ neurons did not induce long-term functional alterations that would impair subsequent learning or memory consolidation.

Since dopamine signaling has been implicated in regulating REM sleep (51), we investigated whether NREM sleep–specific inhibition of VTA^DA^ neurons affects the tendency to enter or maintain REM sleep (fig. S9, C to J). While stGtACR2-mediated inhibition of VTA^DA^ neurons during NREM sleep significantly reduced both the number of REM sleep episodes and transitions from NREM to REM sleep during the 6 hours of optogenetic manipulation (fig. S9, E and F), SwiChR++-mediated inhibition did not significantly affect the total amount of REM sleep, mean episode duration, the number of episodes, and NREM-to-REM sleep transitions compared to mCherry-expressing controls (fig. S9, G to J). These results suggest that stGtACR2-mediated inhibition is more efficient than SwiChR++-mediated inhibition, as supported by our validation data (Fig. 4, C and D, and fig. S8, C and D). Nonetheless, these findings demonstrate that SwiChR++-mediated inhibition of VTA^DA^ neurons during NREM sleep does not indirectly impact memory by disrupting REM sleep initiation or maintenance.

Postlearning VTA^DA^ neuronal activity is not necessary for place-reward memory consolidation

We next tested the causal relationship between VTA^DA^ neuronal activity during the sleep phase and the consolidation of a place-reward memory. We used the same experimental methodology as for the motor skill learning task, except that mice were trained on a four-arm maze instead of the balance beam task (Fig. 5, A and B, and fig. S7A). We found that on the test day, when no reward was provided, all three control groups—saline-treated mCherry-expressing mice, saline-treated hM4Di-expressing mice, and CNO-treated mCherry-expressing mice—showed multiple indicators of retaining the place-reward memory (Fig. 5, C and D). These included an overall significant increase in time spent in, and number of visits to, the previously least preferred arm, and a significantly reduced latency to revisit, and number of visits to other arms before revisiting, the previously least preferred arm (Fig. 5, C and D), as observed in our previous experiments (Fig. 2E and fig. S4B). Notably, inhibition of VTA^DA^ neurons during the sleep phase after learning did not impair the consolidation of the place-reward association, as CNO-treated hM4Di-expressing mice exhibited behavioral changes similar to those of control mice (Fig. 5, C and D). Together, these findings suggest that VTA^DA^ neuronal activity during the sleep phase is not necessary for the consolidation of a simple place-reward association memory.

We then proceeded to determine whether VTA^DA^ neuronal activity specifically during NREM sleep is necessary for the consolidation of a place-reward association memory, using optogenetic inhibition, which both enables selective targeting of NREM sleep and provides more potent neuronal suppression than chemogenetic inhibition. We used the same optogenetic experimental methodology as for the motor skill learning task, except that mice were trained on a four-arm maze instead of the balance beam task (Fig. 5, E and F, and fig. S8, E and F). We successfully targeted ~91% of NREM sleep time in both mCherry-expressing (90.6 ± 1.2%) and SwiChR++-expressing (92.4 ± 0.8%) mice, with minimal off-target manipulations during REM sleep and wakefulness (Fig. 5G). Using both viral vector strategies, we found that on the test day, when no reward was provided, mCherry control mice showed multiple indicators of retaining the place-reward memory (Fig. 5H and fig. S8M). These included a significant increase in time spent in, and visits to, the previously least preferred arm, as well as a reduced latency to revisit and fewer visits to other arms before revisiting the previously least preferred arm (Fig. 5H and fig. S8M). Notably, VTA^DA^ neuronal activity during postlearning NREM sleep was not necessary for place-reward memory consolidation, as stGtACR2- and SwiChR++-expressing mice subjected to NREM sleep–specific optogenetic inhibition exhibited behavioral changes similar to those of mCherry control mice (Fig. 5H and fig. S8M). Together with our chemogenetic manipulation findings (Fig. 5D), these results suggest that the consolidation of a simple place-reward association is independent of VTA^DA^ activity during NREM sleep. In addition, these findings demonstrate that mice undergoing sustained optogenetic manipulation during NREM sleep are capable of retrieving memories the following day. This suggests that our optogenetic manipulation strategies do not broadly suppress memory-related processes during sleep but rather highlight that VTA^DA^ neuronal activity during NREM sleep is specifically critical for motor memory consolidation.

Motor learning increases the coordination between VTA^DA^ activity and M1 spindles during NREM sleep

Last, we examined the temporal relationship between VTA^DA^ neuronal activity and motor cortex spindle oscillations, known to be modulated by motor skill learning and to contribute to motor memory consolidation (52–55), during NREM sleep in both baseline conditions and after motor skill learning. We simultaneously recorded local field potentials (LFPs) from three motor cortex subregions that control limb movements engaged during the balance beam task—the rostral forelimb area (RFA), caudal forelimb area (CFA), and hindlimb (HL) area—alongside EEG/EMG signals and calcium-dependent fiber photometry from VTA^DA^ neurons in male and female DAT-IRES-Cre adult mice (Fig. 6, A and B, and fig. S10, A to D). Recordings were performed for 2.5 hours during the beginning of the light phase following balance beam motor learning, as well as following two control an undisturbed home-cage experience and a gentle handling experience (Fig. 6, C and D). Gentle handling often induced locomotion during the experience.

To investigate the temporal relationship between VTA^DA^ population activity and spindles, we first aligned sigma power [a commonly used proxy for spindles (21, 56)] to the peak of VTA^DA^ transients during NREM sleep (Fig. 6, E, J, and O). Motor skill learning resulted in stronger coordination between VTA^DA^ transients and sigma power in the RFA and HL area (Fig. 6, E to G and O to Q), but not in the CFA (Fig. 6, J to L). In the RFA and HL area (Fig. 6, F and P), sigma power was significantly elevated during the 0.5- to 1-s window before VTA^DA^ transients during NREM sleep after motor skill learning, compared to both gentle handling and undisturbed controls. In the CFA (Fig. 6K), sigma power was also significantly elevated before VTA^DA^ transients after motor skill learning compared to the undisturbed condition, but this increase was not significantly different from the gentle handling control. These findings suggest that motor skill learning enhances the coordination between spindle oscillations and VTA^DA^ neurons, with spindles preceding VTA^DA^ neuronal activation. We also observed coordination between VTA^DA^ transients and sigma power following the transients (Fig. 6, G, L, and Q), but this appeared to be movement-related rather than learning-related: Following both gentle handling and motor skill learning, sigma power decreased after VTA^DA^ transients across all three motor cortex subregions relative to baseline (Fig. 6, G, L, and Q). Aligning sigma power to an equal number of randomly selected non-VTA^DA^–transient time points revealed no modulation (fig. S10E), supporting the premise that the observed effects are specific to VTA^DA^ transients. Together, these findings suggest that motor skill learning enhances the coordination between VTA^DA^ activity and spindle oscillations in the RFA and HL area, with spindles occurring before VTA^DA^ neuronal activation.

To further investigate the temporal relationship between spindles and VTA^DA^ population activity, we next aligned VTA^DA^ activity to the offset of detected spindle events during NREM sleep (Fig. 6, H, M, and R). With this alignment, we corroborated our previous finding of stronger coordination between VTA^DA^ activity and spindles following motor skill learning in the RFA and HL area (Fig. 6, H, I, R, and S), but not in the CFA (Fig. 6, M and N). In the RFA and HL area, VTA^DA^ activity was significantly increased within 500 ms after spindle offset during NREM sleep following motor skill learning compared to both gentle handling and undisturbed controls, whereas in the CFA, it was not significantly altered relative to either control condition (Fig. 6, I, N, and S). Together, these findings strongly suggest that motor skill learning enhances coordination between VTA^DA^ population activity and spindle oscillations in the RFA and HL area of the motor cortex during NREM sleep, with spindle events occurring before VTA^DA^ neuronal activation.

DISCUSSION

To uncover the functions of VTA^DA^ neurons during sleep, we used calcium-dependent fiber photometry, polysomnographic and electrophysiological recordings, multiple learning paradigms, and chemogenetic and optogenetic manipulations. Our findings demonstrate that VTA^DA^ neuronal activity during sleep is modulated by enriched multisensory experiences, with learning processes—particularly motor skill and place-reward association learning—driving increased VTA^DA^ activity during NREM sleep. We further show that the early hours of postlearning NREM sleep constitute a critical window for experience-dependent processes in these neurons. Using chemogenetic and optogenetic inhibition, we establish a causal role for VTA^DA^ activity during NREM sleep in memory consolidation. Specifically, we establish that VTA^DA^ neuronal activity during postlearning NREM sleep facilitates motor skill memory consolidation. Using simultaneous VTA^DA^ and LFP recordings from three motor cortex subregions, we show that motor learning enhances the coordination between VTA^DA^ activity and cortical spindle oscillations during NREM sleep, with spindles preceding learning-dependent increases in VTA^DA^ activity. To our knowledge, this study provides the first direct evidence that VTA^DA^ neuronal activity supports motor memory consolidation during sleep and that cortical spindle oscillations coordinate with VTA^DA^ transients in a learning-dependent manner. These findings substantially advance our understanding of dopamine’s diverse roles in regulating fundamental biological processes, such as motivation and learning, and its dynamic activity during sleep.

Dopamine is well established as critical for learning processes, including stimulus-reward associations and reward prediction errors (34, 35, 57). While multiple lines of evidence suggest a role for dopamine in memory consolidation, direct testing of this hypothesis remains limited. For instance, neuronal activity patterns associated with certain food stimuli have been found to reactivate in putative dopaminergic neurons of the VTA during NREM sleep following the feeding experience (42). However, given the wide range of stimuli and behavioral variables encoded by VTA^DA^ neurons (24–29, 58), it remains unclear whether this reactivation phenomenon extends to other facets of salient experiences. Furthermore, the identity of the reactivated neurons was uncertain, as electrophysiological measures alone are insufficient for reliably identifying dopaminergic neurons (59). In this study, we demonstrate that certain salient experiences—specifically, exploration of an enriched environment, place-reward learning, and motor skill learning—promote VTA^DA^ neuronal activity during subsequent NREM sleep. Whether this increased activity reflects neuronal reactivation awaits further investigation using ensemble-level recordings with single-cell resolution. Pharmacological manipulation of dopamine signaling during the postlearning phase has also implicated dopamine in long-term memory formation [e.g., (43)]. However, the specific time window during which dopamine operates, the specific sleep/wake state involved, the precise source of dopamine, and the types of learning experiences to which this function applies remained unclear. In this study, we demonstrate that dopaminergic neurons within the VTA play a critical role in the consolidation of motor skill memories, specifically during NREM sleep episodes in the hours following the learning experience.

Motor memory consolidation, which benefits from sleep (60–62), involves the enhancement of various aspects of motor skill performance, including speed and accuracy, and the stabilization of their underlying neural representations (63). This process is thought to rely on a distributed neuronal network that includes the M1, striatum, hippocampus, medial prefrontal cortex, cerebellum, and somatosensory cortex (64–66). Coordination across these regions during sleep appears critical, with spindle oscillations, spindle–slow wave coupling, and reactivation events playing prominent roles [e.g., (60, 67–69)]. Learning of a new motor skill increases spindle activity during NREM sleep (52–54) and triggers motor cortex reactivation events coupled to spindles and slow waves (16). Conversely, disrupting sigma or slow-wave activity during NREM sleep after motor learning impairs motor memory consolidation (55). While coordination between cortical and striatal circuits has been studied in this context, the contribution of subcortical neuromodulator populations—such as dopamine neurons—remains largely unexplored.

VTA^DA^ neurons project to several brain regions implicated in motor memory consolidation (70, 71) and are the primary source of dopamine to M1 (31). Moreover, the VTA^DA^-M1 pathway has been recently implicated in motor skill acquisition (30, 31, 33, 72). Nonetheless, until now, a role for VTA^DA^ neurons in motor skill memory consolidation had not been demonstrated. In this study, we demonstrate that VTA^DA^ signaling during NREM sleep is modulated by motor skill learning and facilitates motor memory consolidation, providing, to our knowledge, the first evidence of VTA^DA^ involvement in this process during sleep. We further show that motor learning increases coordination between motor cortex spindles and VTA^DA^ population activity during NREM sleep, with spindles preceding VTA^DA^ activation. Although direct connections between thalamic spindle–generating circuits and VTA^DA^ neurons have not been identified, these circuits may indirectly influence VTA^DA^ activity. For instance, thalamic projections to the ventral subiculum of the hippocampus have been shown to alter activity in putative VTA^DA^ neurons (56, 73). It will be important for future studies to determine whether the increased spindle–VTA^DA^ coordination during NREM sleep after motor skill learning underlies the causal role of VTA^DA^ signaling in motor memory consolidation, as well as to dissect the mechanisms, ranging from synaptic plasticity to coordinated reactivation, by which dopamine contributes to this process.

Historically, neuromodulation during sleep has been viewed primarily as a regulator of sleep/wake transitions and cortical arousal (36, 37, 74). Our study, together with a growing body of recent work (21–23, 38), challenges this view and highlights emerging roles for neuromodulator signaling during sleep in shaping microstructure, regulating neuronal oscillations, and supporting memory consolidation. Similar to our findings for VTA^DA^ neurons, putative noradrenergic locus coeruleus (LC) neurons exhibit increased activity during the first few hours of postlearning NREM sleep (75), and noradrenergic signaling during this time window has been shown to promote memory consolidation (76). During NREM sleep, LC noradrenergic neuronal activity and noradrenaline extracellular levels in the thalamus and cortex oscillate at an infraslow rhythm, suppressing spindles during peak phases and permitting them during trough phases (21, 22). This clustering of spindles enhances spindle-related functions and promotes memory consolidation (22). Moreover, precise timing of noradrenaline release during NREM sleep is critical, as SWR-triggered high-frequency stimulation of LC neurons disrupts memory consolidation and suppresses SWR and spindle occurrence (77). Together, our findings on VTA^DA^–spindle coordination during NREM sleep after motor learning, along with the aforementioned studies, suggest that neuromodulatory systems interact with oscillatory events during NREM sleep and play key roles in memory consolidation.

Our findings suggest that VTA^DA^ neuronal activity is not modulated by negative valence experiences such as exposure to predator cue or a tone-shock association. This contrasts with the findings of Valdés et al. (42), who reported reactivation in putative VTA^DA^ neurons during NREM sleep following exposure to a negative valence experience involving the consumption of bitter-tasting food. This discrepancy may result from differences in recording methodologies and analyses, the nature of the negative valence experiences (pain/fear-related versus food-related), the species studied (rats versus mice), or other experimental factors. Further research is needed to clarify the extent of negative valence modulation of VTA^DA^ neuronal activity during sleep.

While we observed increased VTA^DA^ neuronal activity during NREM sleep following place-reward association learning, our chemogenetic and optogenetic inhibition experiments suggest that this activity is not necessary for consolidating a place-reward memory. It is possible that the four-arm place-reward association task we used is either too simple and/or not sleep dependent, making it independent of dopaminergic activity during sleep. Notably, putative dopaminergic neurons of the VTA have not been found to reactivate reward-related activity during NREM sleep following training on an appetitive spatial working memory task (78). In addition, optogenetic inhibition of VTA^DA^ neurons during NREM sleep after training on a Barnes maze spatial reference memory task did not impair subsequent performance (50). Together, these findings suggest that VTA^DA^ neuronal activity during NREM sleep is not necessary for the consolidation of certain types of experiences, including reward-related ones. However, whether place-reward association memory is truly independent of VTA^DA^ neuronal activity during NREM sleep or whether the nature of the tasks used thus far minimizes their reliance on dopaminergic mechanisms during sleep remains an open question.

In summary, dopamine signaling plays a pivotal role in regulating various critical functions in animals, including learning, motivation, and arousal. Our findings expand this body of knowledge and reveal a previously unexplored role for VTA^DA^ neuronal activity during NREM sleep in memory consolidation. Precise and rapid movements, along with the refinement of these movements through experience, are essential for animals’ daily activities and survival. Our work provides insights into the functions of VTA^DA^ neurons in consolidating learned motor skills through sleep-dependent processes. Dopamine’s involvement in memory consolidation has also been demonstrated in insects (79–82), suggesting an evolutionarily conserved function, which underscores the fundamental role of dopamine in the transformation of internal representations of experiences during sleep. As alterations in dopamine signaling are associated with neurodegenerative diseases that also involve motor deficits and sleep disturbances (83, 84), understanding these links could pave the way for improved therapeutics and advancements in human health.

MATERIALS AND METHODS

Animals

We bred DAT-IRES-Cre C57BL/6J mice (the Jackson Laboratory, stock no. 006660) in-house. We used both female and male heterozygous mice and wild-type littermates. The mice were at least 8 weeks old at the time of surgeries and between 3 and 5 months old at the time of experiments. We housed mice under a 12-hour light/dark cycle at 22° ± 1°C, with ad libitum access to water and food. Mice were provided with compressed cotton Nestlets nesting material (Ancare, Bellmore, NY, USA), shredded paper Enviro-dri nesting material (Shepherd Specialty Papers, Watertown, TN, USA), and typically a piece of cardboard (one-fourth of a Bio-Tunnel; Bio-Serv, #K3556). All experimental procedures were conducted in accordance with the US National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by the University of Michigan’s Institutional Animal Care and Use Committee (protocol no. PRO00011642).

Surgical procedure

Stereotaxic viral injection and fiber implantation

We anesthetized mice with ketamine and xylazine [100 and 10 mg/kg, respectively; intraperitoneal (ip) injection] and administered lidocaine [4 mg/kg, subcutaneously (sc)] for local anesthesia and carprofen (5 mg/kg, sc) for pain management. We placed the mice in a stereotaxic frame (David Kopf Instruments, Tujunga, CA, USA) and maintained them under isoflurane anesthesia (1 to 2% in O2~). We performed craniotomies and durotomies to expose the brain and then infused viruses into target brain regions at 100 nl/min using a microinjection syringe pump (UMP3T-1, World Precision Instruments Ltd.) with a 33-gauge NanoFil syringe (World Precision Instruments Ltd.). All coordinates were defined relative to the skull surface at bregma unless otherwise specified.

For mice undergoing fiber photometry recordings, 400 nl of either AAV-DJ-EF1α-DIO-GCaMP6f (Gene Vector and Virus Core, Stanford University; 5.33 × 10^11^ genome copies per ml (gc/ml); n = 16 mice) or AAV1-syn-FLEX-jGCaMP7f-WPRE (Addgene, #104492; 2.1 × 10^12^ gc/ml; n = 8 mice) was stereotaxically infused into the VTA [anterior/posterior (AP) = −3.1 mm, medial/lateral (ML) = −0.2 to −0.4 mm, dorsal/ventral (DV) = −4.45 mm]. For mice undergoing chemogenetic manipulations, 400 nl of AAV-DJ-EF1α-DIO-hM4D(Gi)-mCherry (3 × 10^12^ gc/ml) or AAV-DJ-EF1α-DIO-mCherry (Gene Vector and Virus Core, Stanford University; 3.08 × 10^11^ gc/ml) was infused bilaterally into the VTA (AP = −3.2 mm, ML = ± 0.2 to 0.4 mm, DV = −4.45 mm). For mice undergoing optogenetic manipulations, 400 nl of AAVDJ-EF1α-SIO-stGtACR2-FusionRed (University of Michigan Vector Core; 1.04 × 10^12^ gc/ml), AAV8–EF1α–DIO–SwiChR++–enhanced yellow fluorescent protein (eYFP) (Gene Vector and Virus Core, Stanford University; 7.6 × 10^11^ gc/ml), AAV8-EF1a-DIO-mCherry (University of Michigan Vector Core; 3.76 × 10^12^ gc/ml), or AAV-DJ-EF1α-DIO-mCherry (Gene Vector and Virus Core, Stanford University; 3.08 × 10^11^ gc/ml) was bilaterally infused into the VTA (AP = −3.1 mm, ML = ± 0.2 to −0.3 mm, DV = −4.45 mm). For in vivo validation of optogenetic inhibition, in addition to the viral injection of stGtACR2 or SwiChR++ into the VTA, 400 nl of AAV8-hSyn-DA4.4 (BrainVTA, China; 5.13 × 10^12^ gc/ml) was infused into the NAc (AP = 1.5 mm, ML = −0.45 mm, DV = −4.5 mm). We waited at least 10 min before withdrawing the needle from the brain.

Following viral injections, optic fibers were implanted to record fluorescence or to deliver light for optogenetic inhibition. For fiber photometry recordings, a mono fiber optic cannula [400 μm in diameter, 0.36 or 0.48 numerical aperture (NA); Doric Lenses, Quebec, Canada] was implanted above the VTA (AP = −3.35 mm, ML = −0.2 mm, DV = −4.0 mm) or NAc (AP = 1.3 mm, ML = −0.45 mm, DV = −4.1 mm). The cannula was lowered at 0.3 mm/min for the first 3.5 mm and then 0.12 mm/min for the final 0.5 mm. For optogenetic manipulations, we implanted dual fiber optic cannulas (300 μm in diameter, 0.22 NA; Doric Lenses, Quebec, Canada) above the VTA (AP = −3.35 mm, ML = ± 0.2 mm, DV = −3.8 mm) and lowered at the same rate as the mono fibers. We then cemented the optic fiber cannulas to the skull with either C&B Metabond (Parkell) and dental cement or OptiBond (Kerr Corporation, #36519) and flowable light cure composite (Kerr Corporation, #29493) and closed the skin incision with surgical sutures. A second dose of carprofen (5 mg/kg, sc) was administered 1 day after surgery for pain management.

EEG, EMG, and LFP implantation

All mice that were infused with viral vectors were also implanted with EEG and EMG electrodes during the same surgery for polysomnographic recordings. Mice were fitted with two miniature screw electrodes (JI Morris Miniature Fasteners Inc., #F00CE125) positioned at AP = 1.5 mm and ML = 1.5 mm and AP = −2.4 mm and ML = 2.8 mm along with two EMG wire electrodes (Cooner Wire) inserted between the trapezius muscles. All electrodes were previously soldered to a four-pin connector (DigiKey, 833-83-004-10-273101). The implant was secured to the skull with C&B Metabond (Parkell) and dental cement or OptiBond (Kerr Corporation) and flowable light cure composite (Kerr Corporation, #29493).

A subset of the mice implanted with EEG/EMG electrodes was also implanted with LFP electrodes. A bundle of tungsten wires (diameter, 50 μm; California Fine Wire Company, #100211) was lowered into the RFA of M1 at AP = 1.8 mm, ML = −1.5 mm, and DV = −1.0 mm; the CFA of M1 at AP = 0.5 mm, ML = −1.2 mm, and DV = −0.7; and HL area of M1/M2 at AP = −0.7, ML = −0.8, and DV = −0.75 (relative to the surface of the brain). Electrode bundles were mounted on a custom-designed three-dimensional (3D)–printed plate (https://doi.org/10.5281/zenodo.17409953) aligned to the target coordinates and lowered as a unit into the brain. A miniature screw (JI Morris Miniature Fasteners Inc., #F00CE125) was implanted in the occipital bone (AP = −5.6 mm, ML = 0.3) to serve as a reference electrode. All LFP implants were prepared in-house. We measured electrode impedance using an electrode impedance tester (BAK Electronics Inc., #IMP-2A) and gold-plated (31 g/liter; SIFCO Applied Surface Concepts, #80535500) electrodes to achieve a target impedance of less than 20 kilohms before implantation. LFP implantation was performed following viral injection and before EEG/EMG and optic fiber implantation.

Habituation and experimental housing

Following recovery from surgery (7 to 10 days), mice were individually housed in custom PLEXIGLAS recording chambers (39.4 cm by 28.6 cm by 19.3 cm, with open tops) for all experiments, except for the balance beam validation (see below). After the mice had fully built their nests in the recording chambers (~3 days), they were connected to flexible EEG/EMG cables for at least 10 days and to fiber optic cables for at least 6 days, before data collection. Before all behavioral experiments, mice were handled for 3 to 5 days.

Polysomnographic data acquisition and analysis

EEG and EMG signals derived from the EEG/EMG electrodes were amplified (model 3500, A-M Systems) and digitized at 1017.3 Hz using Tucker-Davis Technologies (TDT) RZ5 processor and Synapse software. For a subset of mice in the chemogenetic manipulation experiments, EEG/EMG signals were amplified (model 3500, A-M Systems) and digitized at 512 Hz using the VitalRecorder software (Kissei Comtec America).

The EEG/EMG signals were filtered (EEG, 0.3 to 25 Hz; EMG, 10 to 100 Hz) and downsampled to 512 Hz using a custom MATLAB script (MathWorks, Natick, MA) for further analysis. Sleep/wake states were manually scored using a modified version of the open-access AccuSleep software (85) or semiautomatically scored with a reimplemented Spindle algorithm (86). Data were annotated in 4-s epochs, and at least two consecutive epochs of a distinct state were required for a state change. Wake was defined as epochs with mixed-frequency, low-power EEG oscillations concurrent with EMG activation. NREM sleep was defined as epochs with high-amplitude EEG oscillations, characterized by high delta power (1 to 4.5 Hz) and a lack of EMG activity compared to wakefulness. REM sleep was defined by low delta power, high theta (5 to 9 Hz) power, and low-amplitude EEG oscillations, as well as no EMG activity, except for occasional brief muscle twitches (<4 s). Power spectra were computed using the pwelch function in MATLAB with a 4-s window and 90% overlap. All spectra, except those in Fig. 4, were normalized to total wake power measured between 4 and 6 hours after sleep onset. Spectra in Fig. 4 were normalized to total wake power during the first 6 hours of sleep on the baseline recording day.

Fiber photometry data acquisition and analysis

We sinusoidally modulated blue light from a 465-nm and purple light (isosbestic control) from a 405-nm excitation light-emitting diodes (Doric Lenses) at 220 and 310 Hz, respectively, using the Synapse software and the RZ5 processor. The light was passed via low-fluorescence patch cords (400 μm, 0.48 NA; Doric Lenses) into the Fluorescence Mini Cube (FMC6, FMC4, or ilFMC4; Doric Lenses) and then to the implanted fiber optic probe (connected via a zirconia sleeve). Emitted light from the biosensors was collected through the same cable, routed onto dichroic mirrors in the Mini Cube, and sent to a femtowatt photodetector (Visible Femtowatt Photoreceiver Model 2151, Doric Lenses). This signal was digitized and demodulated at 1 kHz with a sixth-order 6-Hz low-pass filter by Synapse software and RZ5 processor. We measured excitation light intensity (30 μW for 465 nm and 1 to 2 μW for 405 nm) at the tip of a fiber optic probe connected to the patchcord before experiments.

Downsampling and ΔF*/*F

The photometry signal was downsampled using interpolation to match the EEG/EMG sampling rate of 512 Hz (see below). For each recording session, the photometry signal, F, was converted to ΔF/F using the formulaΔFF=(F−F0)/F0where F0 is the median of the fluorescence trace.

Detrending

Our fluorescence recordings included a rapid decay at the start of the recording and a gradual decline in fluorescence over the course of a recording due to photobleaching of the sensor. To address these artifacts, the ΔF/F signal from each session was fitted with a decreasing exponential of the form a×ebx , where a > 0, b < 0, and x is time, or a×ebx+c×edx , where x is time. The fitted exponential was then subtracted from the original ΔF/F signal. This procedure corrected the decay when present and left the signal unchanged when no decay was observed.

Detection of transients

We detected transients similar to those reported in (39, 87, 88). Briefly, two filtered ΔF/F signals were one low-pass filtered at 0 to 0.4 Hz and the other at 0 to 40 Hz. A trace of the derivative of their squared difference was calculated, and candidate transient times were identified by thresholding this signal at mean + 2 SDs of the derivative. Candidate transient times were then filtered to include only events occurring during periods of high ΔF/F, which we defined as events having an amplitude greater than the mean + 2 SDs of the detrended ΔF/F signal. The exact peak time of each transient was then determined by locating the maximal value of the signal within each transient. For the analyses in Figs. 1 and 2 and figs. S2 and S5, the following calcium transient measures were calculated for each session and arousal (i) transient rate, (ii) mean transient amplitude, and (iii) mean transient amplitude × transient rate.

LFP data acquisition and analysis

LFP and EEG/EMG signals were referenced to a skull screw electrode and amplified and digitized at 1017.3 Hz using the Lab Rat and Synapse Lite system (TDT). Before analysis, we detected spindles from electrodes in each motor cortex subregion and selected the channel with the most consistent detections and signal quality across days for analyses.

Alignment of sigma power to VTA^DA^ transients

For each recording, we band-pass filtered the full LFP signal in the sigma range (10 to 15 Hz) and calculated power as the absolute value of the Hilbert transform of the filtered traces. We z-scored power traces using the mean and SD during NREM sleep episodes, downsampled them to 512 Hz to match fiber photometry data, and aligned them to the start of each photometry session using a transistor-transistor logic (TTL) pulse generated by the RZ5 (photometry) system and recorded on the Lab Rat (LFP) system. Last, we smoothed the aligned z-scored power signal with a 1-s moving average filter. We included in our analysis VTA^DA^ transients that occurred during NREM sleep, were flanked by at least 10 s of NREM sleep, and did not fall within NREM-to-REM sleep transition periods (defined as within 20 s before REM onset). We then aligned z-scored sigma power to the peak of each transient. To characterize sigma power modulation around VTA^DA^ transients in undisturbed, control, and postlearning sleep, we calculated for each transient the difference between mean sigma power in the 1-s window after the peak (0 to 1 s from t = 0) and mean power in a baseline window (−3 to −2 s). We also quantified pretransient modulation as the mean sigma power 0.5 to 1 s before the peak. To test whether these effects were specific to transients, we aligned sigma power to an equal number of randomly selected sham time points, placed at least 1 s from real transients and from each other.

Spindle detection

We detected spindles as described in (89) using a complex frequency B-spline function to compute power. Briefly, power was calculated as the absolute value of the squared continuous wavelet transform of LFP data and smoothed with a 200-ms Hanning window. The spindle band was defined as 9 to 16 Hz (central, 10 to 14 Hz). Candidate spindle events were identified when power exceeded mean + 3 SDs, with onset and offset defined as the most recent and next crossings of a lower threshold (mean + 1 SD). Events were retained if they lasted 0.4 to 2 s, contained 5 to 30 cycles in the band-pass–filtered signal, had maximum power of <mean + 20 SD (to exclude outliers), and exhibited greater power in the spindle band than in neighboring frequencies (6 to 8.5 Hz and 16.5 to 20 Hz). Events separated by <10 ms were combined.

To determine how VTA^DA^ activity is modulated around spindles, we aligned fluorescence traces to each spindle event and computed ΔF/F using the median fluorescence from the 1 s preceding spindle onset. We then compared mean ΔF/F during the 0.5 s after spindle offset across undisturbed, control, and postlearning conditions.

Chemogenetic inhibition

CNO dihydrochloride (2.5 mg/kg; Hello Bio Inc.) was prepared in sterile 0.9% saline and stored in a dark compartment at room temperature or −20°C until use. The drug was used within 4 weeks of preparation and thawed for at least 1 hour before use if it was frozen. Mice were habituated to the handling and injection procedures associated with intraperitoneal injections by the experimenters, who held them by the tail and touched their abdomen at the injection site over 3 days, and administered a single intraperitoneal saline injection at least a day before experiments began.

Postlearning chemogenetic inhibition

After completing the learning tasks (see below), the mice were returned to their home cages, placed back in their housing/recording room, and reconnected to the EEG/EMG cables. Mouse behavior was monitored via video recordings, and the first CNO injection was administered once the mice exhibited grooming behavior—a sleep preparatory behavior that typically precedes sleep by ~10 min (90, 91)—in their nest for at least 1 min. A second CNO injection was given 3 hours later to maintain inhibition of VTA^DA^ neurons during approximately the first 6 hours of postlearning sleep.

Optogenetic inhibition

Optogenetic inhibition of VTA^DA^ neuronal activity was achieved via two stGtACR2, an anion-conducting soma-targeted channelrhodopsin (48), and SwiChR++, a blue light– and red light–sensitive inhibitory step function opsin (49). In experiments using stGtACR2, we used a single 473-nm blue light laser (Opto Engine) along with a light intensity splitter (doric mini cube-intensity division, DMC 1x2i, Doric Lenses) to create two beams of equal intensity, which were routed via a 300-μm core dual patchcord (0.22 NA; Doric Lenses) to the implanted bilateral optic fibers. Light intensity measured at the tip of the optic fiber attached to each patchcord was ~1 mW per hemisphere.

For the SwiChR++ experiments, a light intensity splitter (DMC 1x2i, Doric Lenses) was used in reverse, allowing light entry via the two output ports and emission from the single input port. This permitted switching between blue and red lasers (473 and 635 nm; Opto Engine). The output beam from the first splitter was again split into two equal intensity beams via a second splitter and transmitted to the mice. Using TDT Synapse software and the RZ5 processor, we triggered 1 s of blue light (473 nm, 8 mW per hemisphere, measured at the optic fiber’s tip) every 30 s to maintain optogenetic inhibition and 1 s of red light (635 nm, 5 mW) to terminate inhibition.

Postlearning NREM sleep–specific optogenetic inhibition

After completing the learning tasks, mice were returned to their home cages and housing/recording room and reconnected to EEG/EMG and fiber optic cables. The optogenetic manipulations commenced at the first episode of NREM sleep following learning and lasted for 6 hours from sleep onset. We manually detected NREM sleep and triggered the lasers within 15 s of state onset, indicated by the absence of EMG activity, high-amplitude slow oscillations in the EEG, and a sleeping posture observed in video recordings. We terminated the inhibition after observing at least 4 s of either wakefulness or REM sleep, characterized by a reduced EEG amplitude and slow oscillations, EMG activation (wakefulness), or prominent theta oscillations (REM sleep). In the stGtACR2 experiments, lasers were turned off after a maximum of 5 min, even if the mice remained in NREM sleep. Inhibition resumed 20 s following this if the mice remained in NREM sleep or whenever the next NREM sleep episode began. A linear ramp (2 s long) was applied when turning off the blue light to minimize the possibility of rebound excitation in the neurons.

In vivo validation of stGtACR2/SwiChR++ function

We monitored dopamine release in the NAc via GRABDA biosensor recordings while optogenetically inhibiting VTA^DA^ neurons during NREM sleep using the same inhibition protocol as following learning (above). Recordings were conducted during the light phase, lasting 40 min and separated by 40 min. For sham trials, baseline GRABDA fluorescence recordings were obtained without optogenetic manipulations. To quantify the effects of optogenetic inhibition on dopamine release during inhibition trials, we identified all NREM sleep episodes targeted by the lasers and computed ΔF/F using the median fluorescence during the 10 s before laser onset. We then calculated the mean ΔF/F from laser onset to offset. For sham trials, a similar number of NREM sleep episodes were randomly sampled, with laser onsets assigned 15 s after the onset of these episodes and offsets designated as their end points.

Behavioral paradigms

All behavioral experiments were conducted during the last 6 hours of the dark phase. Mice were disconnected from their EEG/EMG and fiber optic cables (if applicable) and transferred to the behavioral room 1 hour before experimentation for acclimation to the new environment, unless otherwise mentioned. Mouse behavior was continuously monitored via video recordings using a camera mounted above the behavioral setups [Logitech HD Pro Webcam C920 at 10 to 15 frames/s (FPS) or Basler acA1440-220um (Basler, #107652) at 100 FPS], using either Synapse, iSpy, or pylon Viewer software. We conducted simultaneous calcium sensor recordings in a subset of experiments.

Enriched multisensory experience

We exposed mice to a novel enclosure (49 cm by 29 cm by 20 cm) containing various stimuli for 90 min (movie S1). The enclosure included palatable food (3 g of Reese’s chocolate and peanut butter cups), novel bedding (either Pure-o’Cel bedding, The Andersons Inc., or Biofresh Comfort Bedding, Lab Supply, #L0110), novel scents [30 μl of 5% limonene and 5% ethyl butyrate (diluted in water, TCI America, #B0759 and L0046, respectively) applied on gauze placed in tea strainers], various enrichment objects (such as tunnel, igloo, and bridge), and a running wheel (either 18 or 10.5 cm in diameter). Mice had access to water throughout.

Predator cue experience

We exposed mice to a used home cage of a male rat for 45 min (movie S2). The cage (45 cm by 23.5 cm by 20.5 cm) was collected within 4 hours before the experiment and contained soiled bedding, fecal boli, and shredded nesting material.

Home cage gentle handling experience

We maintained wakefulness by gently stroking the mice, disturbing their bedding, or tapping the side of the cage with a cotton swab whenever they adopted a quiescent posture and appeared ready to sleep (movie S3). Because mice under both the enriched multisensory and predator cue conditions remained awake for ~30 min after returning to their home cages, we gently handled the control group for a total of 2 hours, ending at ZT 0.5, to match circadian sleep times across groups. For comparison with VTA^DA^ population activity after the place-reward learning task, the gentle handling control was performed under food-restricted conditions (identical to those used for the place-reward task, described below).

Balance beam motor learning task

We used a modified balance beam motor learning task (45, 46). The apparatus consisted of two narrow steel rods (diameter, 4.76 mm; length, 91.44 cm; Home Depot) spaced 3.5 cm apart and suspended ~23 cm above the counter (movie S4). We stabilized the rods with custom 3D-printed plastic holders placed at each end (https://doi.org/10.5281/zenodo.17409953). A vertical barrier was fixed to one end of the apparatus, and the mice’s old home cages were placed at the other. Mice had lived in these cages for at least 2 weeks, ending 1 to 2 weeks before the experiment. After collection, we stored the cages in sealed plastic bags to preserve olfactory cues until the experiments. We recorded mouse behavior from both a side view (Logitech HD Pro Webcam C920, 10 to 15 FPS) and an angled-below view [either Logitech HD Pro Webcam C920, 10 to 15 FPS, or Basler acA1440-220um (Basler, #107652), 100 FPS].

At the beginning of each session, we placed mice in their old home cage for 3 min and then positioned them on the rods at the opposite end, facing the cage. We waited until each mouse placed all four paws on the rods before releasing its tail and monitored behavior throughout the session. If a mouse fell, then we returned it to the rods at the point of the fall. Each mouse performed one session per day. Sessions ended when the mouse traveled more than 60 cm on the rods, 10 min elapsed, or the mouse fell seven times. To complete the task, mice needed to travel at least 60 cm within 10 min. The first day on which this criterion was met was designated as day 1. All mice were run for at least three additional days. If a mouse failed to complete the task within 10 min, then we placed it ~5 cm from its old home cage, allowing it to reenter the cage on its own. Mice that failed to complete the task for 4 consecutive days were excluded from the experiment (one expressing hM4Di-mCherry, one expressing mCherry, and one expressing SwiChR++-eYFP). We excluded seven additional mice from three as outliers (time to complete the task > 1.5 SD), one that failed to learn, one disrupted by a fire alarm during day 1, and two that already performed expertly on day 1.

We assessed motor skill memory by comparing two kinematic variables pre- and postlearning. We used the Detect Any Mouse Model (DAMM) (92) toolbox, which uses instance segmentation to detect and track the location of mice in videos and generates bounding boxes that outline their position and size. We fine-tuned DAMM weights using 310 annotated images from our dataset. We then smoothed the centroid trajectory of the bounding boxes with a locally weighted scatterplot smoothing filter (1.6-s window) and used it for kinematic analyses.

Using custom Python code, we computed two (i) time to traverse 60 cm of the rods in the x dimension and (ii) mean speed while moving above the rods. We excluded epochs when mice were off the rods or “off-task” due to falls or stationary bouts. Fall episodes began when no more than two of the limbs or tail of the mice were touching the rods and ended when the experimenter released the tails of the mice after placing them back on the rods. Stationary bouts were periods when the mice made no movements along the length of the rods, although they may have stretched their upper body, moved between the two rods, or groomed. Both falls and stationary bouts were manually annotated.

Place-reward association learning task

We used a four-arm maze for place-reward association (movie S5). Each arm measured 50 cm (L) by 15 cm (W) by 20 cm (H) with a 15 cm–by–15 cm center region. Arms were distinguished by visual cues on the walls (vertical stripes, horizontal stripes, and circles) and tactile floor cues (rough plastic, grid plastic, and smooth surfaces).

We food restricted mice to 90 to 95% of their initial body weight before the task and maintained them within this range throughout the paradigm. We weighed and fed the mice daily at ZT 23 to minimize sleep disruptions. Several days before training, mice received a small piece of the food reward (Reese’s peanut butter and chocolate) in their home cage.

At the start of each session, we placed the mice in the center of the maze. On day 1, mice explored freely for 10 min, and we identified the least preferred arm offline using EthoVision XT14 (Noldus). On day 2, mice underwent two training sessions (1 hour apart, six trials each). In each session, the least preferred arm was baited with food rewards (0.05 g per trial). Mice explored continuously for 18 min, with a reward placed in the distal half of the baited arm every 2 to 3 min. Each reward was delivered once the mouse left the baited arm; if not consumed within 2 to 3 min, then no new reward was given until the previous one was eaten. After each session, mice returned to their home cages. At the end of the second session, we weighed the mice, provided additional food to maintain weight, and reconnected them to EEG/EMG and fiber optic cables for recordings.

On day 3 (test day), no rewards were delivered. Mice explored the maze for 10 min, and we compared pre- versus postlearning performance (day 1 versus day 3) across four time spent in the rewarded arm, proportion of visits to the rewarded arm, interval between visits to the rewarded arm, and the number of nonrewarded arms visited between rewarded arm entries.

Cued fear conditioning

We used a fear conditioning chamber (31.5 cm by 29 cm by 30 cm) with a metal grid floor, white walls, and a rough plastic wall. The chamber was scented with limonene, present in the stainless-steel wipes used for cleaning (Sanitation Tools, #6497799). A speaker (MF1, TDT), controlled via an RZ5 or RZ2 (TDT) processor, was mounted outside the chamber.

We transferred mice to a behavior room adjacent to the conditioning room, where the tone was inaudible, 1 hour before the experiment. At the start of each trial, we placed mice in the chamber’s center and allowed 2 min of habituation. We then presented a 10-s auditory tone (8000 Hz, 85 dB) paired with a 0.6-mA foot shock delivered in the last 0.5 s of the tone. The shock was generated by an aversive stimulator/scrambler (Med Associates, #ENV-4145) and controlled via the TDT processor. Each mouse received six tone-shock pairings spaced 1 min apart.

We tested memory 24 hours later in a novel context (dark walls, smooth floor, and 5% ethyl butyrate scent; TCI America, #B0759). After 2 min of habituation, mice received six tone presentations (10 s each, spaced 1 min apart).

We used a weaker conditioning protocol in 6 of 10 mice. After 5-min habituation in the same conditioning chamber, mice received six tone-shock pairings, each separated by 10 s. Tones were 1 s long (8000 Hz, 85 dB), and shocks (0.4 mA) lasted 0.5 s. We added 1 min of recovery at the end of each trial with no tones or shocks. Memory was tested in the same novel context as the strong protocol, with 3 min of habituation, followed by six tones (10 s each, 10 s apart).

We quantified freezing with a DeepLabCut (v2.3.8)–based pose estimation approach (93). We tracked 10 key points per snout, head, left/right ears, mid-spine, left/right flanks, and base, midpoint, and tip of the tail. We annotated 515 frames from nine behavioral videos, assigning 80% (412) for training and 20% (103) for testing. We trained a ResNet-50–based neural network (default parameters) for 250,000 iterations, yielding a training error of 3.03 pixels and a test error of 5.48 pixels (frame size, 640 pixels by 480 pixels).

We analyzed the tracked videos in BehaviorDEPOT (94) for velocity-based freezing detection, using the following velocity threshold = 0.52, angle threshold = 12, window width = 10, count threshold = 4, and minimum bout duration = 1 s. We compared detected freezing bouts with manual scoring in a subset of mice (n = 3) and visually inspected additional videos to confirm detection accuracy.

Running wheel experience

We exposed mice to a novel enclosure (49 cm by 29 cm by 20 cm) containing bedding and a running wheel (18 cm in diameter) for 90 min.

Home cage reward consumption experience

We habituated mice to an empty food container (7.8 cm in diameter and 3.8 cm in height) in their home cage for 3 days before the experiment (30 min per day between ZT 23 and 0). On the experiment day, we placed 1 g of Reese’s chocolate reward in the container for 30 min between ZT 23 and 0. As a control for the reward-consumption experience, we exposed mice to the empty container for the same duration at least 1 day before the reward session and kept them awake from ZT 0 to 0.5 to match circadian sleep time following chocolate consumption.

Histology

We deeply anesthetized mice with ketamine and xylazine (100 and 20 mg/kg, ip, respectively) and perfused them with 1× phosphate-buffered saline (PBS) followed by 4% paraformaldehyde (PFA) in PBS. We submerged the heads overnight in 4% PFA at 4°C. After fixation, we removed the implants from the skull, extracted the brains, and cryoprotected them in sucrose solution (30% sucrose in PBS with 0.1% NaN3) at 4°C for at least 48 hours. We sliced brains coronally at 40 μm in thickness on a cryostat (Leica Microsystems) and collected slices containing the VTA and motor cortex. Slices were stored in PBS with 0.1% NaN3 at 4°C until imaging, which we performed using a Zeiss Axio Imager.M2 microscope with ZEN software (Zeiss). To map viral transduction (heatmaps), optic fiber placement, and LFP electrode placement, we overlaid images with the corresponding Allen Mouse Brain Atlas panels in Adobe Illustrator 2020. One mouse was excluded from the SwiChR++ inhibition group due to lack of viral expression beneath the optic fiber.

Quantification and statistical analysis

We analyzed data using Prism 10.4.1 (GraphPad Software) and MATLAB R2022a. We determined sample sizes based on prior publications in the field (22, 38, 39) and, for the behavioral experiments in Fig. 4, through a power analysis (G*Power 3.1.9.7) (95) using data from the chemogenetic inhibition experiments as pilot input. Data analysis was conducted blind to experimental conditions when possible. While we assumed normal data distribution, this was not formally tested in all cases. For comparisons between two groups, we used Wilcoxon signed-rank tests, paired-samples t tests, or unpaired-samples t tests. For comparisons involving more than two groups, we used one-way repeated-measures (RM) analyses of variance (ANOVAs), followed by Dunnett’s multiple comparisons tests. For two-factor analyses, we used RM two-way ANOVAs or mixed-effects model ANOVAs for datasets with missing values (e.g., absence of a specific sleep state at a given time point), followed by Sidak’s multiple comparisons tests. We computed partial η^2^ for ANOVAs and Cohen’s d for paired t tests, or Hedges’ g for t tests with unequal samples, to quantify effect sizes. P < 0.05 was considered statistically significant. Sample sizes are detailed in the figure legends, and the full statistical results are provided in tables S1 to S12. Figures were prepared using Prism 10.4.1 and MATLAB R2022a, with data exported to Adobe Illustrator 27.9 (Adobe Creative Cloud) for final formatting.