Induction of cortical on/off periods in awake mice fulfills sleep functions Authors: Kort Driessen, Fabio Squarcio, Giulio Tononi, Chiara Cirelli Published: 2026 ABSTRACT In mammals, slow-wave sleep is characterized by synchronized neuronal activity that alternates between on and off periods. Slow-wave activity (SWA) and synchrony reflect sleep need, are correlated with synaptic strength in cortical circuits and promote synaptic downselection and memory consolidation. Here we assessed whether these core benefits of sleep can be obtained during waking. We locally induced alternating on/off periods during wakefulness using optogenetics in mice. This led to a local ipsilateral reduction in SWA and synchrony during subsequent sleep, and to reduced markers of synaptic strength. Moreover, bilateral induction of off periods over sensorimotor cortex during sleep deprivation restored memory consolidation. Thus, inducing on/off activity during wakefulness is sufficient to reduce local sleep need and fulfill core functions of sleep. Main Sleep is a universal behavioral state typically characterized by quiescence and decreased responsiveness to stimuli. It is also homeostatically regulated, becoming deeper and often longer after extended periods of wakefulness1. Sleep is thought to perform essential functions that the brain cannot carry out when connected to the environment2,3. Chief among these is synaptic homeostasis, which prevents synaptic saturation, restores the ability to learn4–6 and promotes memory consolidation4–7. Nonrapid eye movement (NREM) sleep, which accounts for ~80% sleep in adults, has been closely linked with synaptic homeostasis8,9 and associated memory functions10,11. A defining feature of NREM sleep is the synchronized, low-frequency alternation between periods of high population spiking activity of cortical neurons (on periods) and periods of population silence (off periods)12–14. This widespread, synchronized, bistable activity pattern underlies the occurrence of slow waves in recordings of cortical field potentials such as the electroencephalogram (EEG). Slow-wave activity (SWA), the EEG power in the delta range during NREM sleep (0.5–4 Hz), is a well-established marker of sleep need and homeostasis, increasing after sleep deprivation (SD) and learning and declining during sleep in both rodents and humans1. There is substantial evidence, in both animals and humans, that the synchronous on/off activity patterns that underlie slow waves in NREM sleep may be key mechanistic mediators of sleep’s restorative functions8,9,11,15–18. This raises the question whether, if one could induce synchronous on/off activity patterns locally in individuals who are otherwise awake, some of the homeostatic functions of sleep and associated benefits may be carried out without disconnecting from the environment. Indeed, an evolutionary proof of concept is offered by some cetaceans19, fur seals20 and bird species21 that can enter NREM sleep with one hemisphere while the other hemisphere remains vigilant22. To address this question experimentally, we developed a method to locally induce NREM sleep-like on/off patterns in one hemisphere of awake mice using complementary optogenetic approaches, while recording unit activity from homotopic networks in both hemispheres. As shown below, this resulted in reduced SWA and synchrony in subsequent NREM sleep on the stimulated side. We further show that a broad induction of NREM sleep-like cortical on/off periods in awake animals renormalizes molecular markers of excitatory synaptic strength and improves memory consolidation, thus fulfilling key homeostatic functions of sleep in awake, behaving animals. Results Recording and manipulation of homotopic cortical networks We implanted adult mice with bilateral linear silicon probes, which allowed us to record laminar multi-unit activity (MUA) and local field potential (LFP) in local cortical networks of homotopic brain regions. An optic ferrule was attached to one of the probes (hereafter termed the ‘optrode’) so that the cortical network recorded by that probe could be optogenetically manipulated independently of its homotopic partner (hereafter termed the ‘contralateral control’ probe; Fig. 1a).Fig. 1Chronic recording and optogenetic manipulation of local cortical networks.a, Top left, implant schematic. Two silicon probes are inserted into the same coordinates exactly contralateral (that is, homotopic) to each other. One probe is connected to an optic ferrule for the delivery of light in optogenetic experiments (‘optrode’) and the other has no attached ferrule and can only record (‘contralateral control’). Top right, representative coronal section showing the fluorescence tracks where the probes were implanted. Area in blue box is enlarged at bottom. Scale bar = 1 mm (top right). b, Top, example hypnogram and corresponding muscular activity (EMG) from a continuous 48-h recording. Bottom, SD through new objects for 5 h, beginning at lights on. Recovery sleep is defined as the first 1 h of cumulative NREM sleep after SD. SWA (spectral power between 0.5 and 4 Hz) for optrode and contralateral control probe is expressed relative to the NREM average during the 12-h baseline. Please note the tight SWA correspondence between the two probes. Dotted line indicates the 12-h NREM baseline SWA mean (indicated at left by an asterisk). Red arrow indicates the homeostatic decline in SWA as the mouse spends time asleep. c, Example LFP and MUA data in wake, NREM sleep and REM sleep from a mouse with probes implanted in secondary motor area (M2). Layer boundaries across the 16 channels are shown for both probes at the far left. d, SOM+ mice express ChR2 in somatostatin-producing interneurons. The optogenetic activation of these neurons (dark blue cells indicated by arrows) leads to inhibition of all neuron types in the cortical network. e, Example SOM+ mouse average field potential response to 200 80-ms light pulses. Data represent mean ± s.e.m. Data are from one representative mouse with implant in M2. f, Example data from the same mouse as in e. Bottom, raster plot with each row showing MUA collapsed across all channels for one 80-ms light pulse. Middle, spike count in 10-ms bins normalized to maximum value, data represent mean ± s.e.m. Top, spike counts during each full 80-ms epoch (prestim, stim, poststim) normalized to maximum value. g, ACR mice express soma-targeted Guillardia theta anion-conducting ChR1 (stGtACR1) in CaMKIIα-producing neurons (mainly pyramidal neurons, light blue cells indicated by arrows). When stGtACR1 is optogenetically activated, all CaMKIIα neurons in the cortical network are inhibited. h, Same as in e, but for one representative ACR mouse with an implant located in M2. i, Same as in f, but for the ACR mouse shown in h. Schematics in a, d and g created in BioRender; Driessen, K. https://biorender.com/ajppjxu (2026). Chronic recording and optogenetic manipulation of local cortical networks. a, Top left, implant schematic. Two silicon probes are inserted into the same coordinates exactly contralateral (that is, homotopic) to each other. One probe is connected to an optic ferrule for the delivery of light in optogenetic experiments (‘optrode’) and the other has no attached ferrule and can only record (‘contralateral control’). Top right, representative coronal section showing the fluorescence tracks where the probes were implanted. Area in blue box is enlarged at bottom. Scale bar = 1 mm (top right). b, Top, example hypnogram and corresponding muscular activity (EMG) from a continuous 48-h recording. Bottom, SD through new objects for 5 h, beginning at lights on. Recovery sleep is defined as the first 1 h of cumulative NREM sleep after SD. SWA (spectral power between 0.5 and 4 Hz) for optrode and contralateral control probe is expressed relative to the NREM average during the 12-h baseline. Please note the tight SWA correspondence between the two probes. Dotted line indicates the 12-h NREM baseline SWA mean (indicated at left by an asterisk). Red arrow indicates the homeostatic decline in SWA as the mouse spends time asleep. c, Example LFP and MUA data in wake, NREM sleep and REM sleep from a mouse with probes implanted in secondary motor area (M2). Layer boundaries across the 16 channels are shown for both probes at the far left. d, SOM+ mice express ChR2 in somatostatin-producing interneurons. The optogenetic activation of these neurons (dark blue cells indicated by arrows) leads to inhibition of all neuron types in the cortical network. e, Example SOM+ mouse average field potential response to 200 80-ms light pulses. Data represent mean ± s.e.m. Data are from one representative mouse with implant in M2. f, Example data from the same mouse as in e. Bottom, raster plot with each row showing MUA collapsed across all channels for one 80-ms light pulse. Middle, spike count in 10-ms bins normalized to maximum value, data represent mean ± s.e.m. Top, spike counts during each full 80-ms epoch (prestim, stim, poststim) normalized to maximum value. g, ACR mice express soma-targeted Guillardia theta anion-conducting ChR1 (stGtACR1) in CaMKIIα-producing neurons (mainly pyramidal neurons, light blue cells indicated by arrows). When stGtACR1 is optogenetically activated, all CaMKIIα neurons in the cortical network are inhibited. h, Same as in e, but for one representative ACR mouse with an implant located in M2. i, Same as in f, but for the ACR mouse shown in h. Schematics in a, d and g created in BioRender; Driessen, K. https://biorender.com/ajppjxu (2026). We recorded LFP and MUA data from the bilateral silicon probe implant, as well as electromyogram (EMG), continuously for 48 h, and determined whether the mouse was awake, in NREM sleep, or in REM sleep for the duration of the recording (Fig. 1b,c). Beginning at light onset of day 2, mice were sleep deprived for 5 h by introducing new objects into the home cage. This SD protocol reliably kept mice awake for the full 5-h period (Supplementary Fig. 1) and reliably resulted in a large ‘rebound’ in SWA in subsequent NREM sleep, which then declined as the mouse slept—a clear demonstration of SWA reflecting the buildup and subsequent dissipation of the pressure to sleep (Fig. 1b). Measures of sleep pressure, such as SWA, were tightly coupled in both probes (Fig. 1b,c). We used two optogenetic mouse models to induce NREM sleep-like patterns of on/off activity in one local cortical network independently of neurons in the homotopic cortical network (Supplementary Video 1). One strategy involved recruiting somatostatin-expressing neurons, a class of cortical interneurons known to have a key role in the generation of sleep slow waves23. A complementary approach involved directly inhibiting excitatory pyramidal neurons. In the first model, we crossed mice carrying a transgene encoding channelrhodopsin-2 (ChR2) with mice expressing the enzyme Cre under the control of the somatostatin promoter (Fig. 1d). In the generated mouse line (hereafter termed ‘SOM+’ mice), somatostatin-expressing neurons could be optogenetically activated, resulting in positive-going, NREM sleep-like slow waves (Fig. 1e) in the LFP, and the associated inhibition of spiking activity in nearly all other neurons in the local cortical network recorded by the optrode (Fig. 1f). In the second model, we crossed mice expressing soma-targeted Guillardia theta anion-conducting ChR1 (stGtACR1 (ref. 24)) with mice expressing the enzyme Cre under the control of the CaMKIIα promoter (Fig. 1g). In the generated mouse line (hereafter termed ‘ACR’ mice), optogenetic stimulation led to potent inhibition of nearly all cortical neurons, resulting in negative-going slow waves (Fig. 1h), and the associated inhibition of nearly all spiking activity in the local cortical network recorded by the optrode (Fig. 1i). Induction of off periods during wakefulness reduces SWA in subsequent NREM sleep We first wondered whether the sustained induction of NREM-like off periods in a local cortical network of an awake mouse could reduce SWA only in that network in subsequent NREM sleep. To test this hypothesis, we performed a 5-h SD through new objects as described in Fig. 1b, but starting 30 min before the end of the SD, we induced off periods similar to those seen in NREM sleep (Methods), while the SD continued and the mouse remained awake (Supplementary Video 1). General aspects of behavior, such as the likelihood of pausing or preference for turning left or right, were unaffected by off-period induction (Supplementary Fig. 2). This manipulation was equally effective in supragranular and infragranular channels (Extended Data Fig. 1c,f) and did not cause an immediate or otherwise notable rebound in firing after, between or during off-period induction pulses (Extended Data Fig. 1a–e). On the other hand, in both SOM+ mice and ACR mice, off-period induction resulted in a large increase in SWA during the stimulation period on the optrode side, similar to that seen in NREM sleep, and not on the homotopic contralateral control side (Fig. 2a–c and Extended Data Fig. 2a,b). Induced off periods were associated with suppression of high frequencies and were followed by oscillatory responses in the theta and sigma ranges. These changes, especially in SOM+ mice, resembled those seen during NREM sleep (Extended Data Fig. 3). In both SOM+ and ACR mice, however, induced off periods were followed by fewer spindles than spontaneous off periods (Extended Data Fig. 3c,d).Fig. 2Off-period induction during wake reduces SWA and synchrony in subsequent sleep.a, Experiment overview; in the last 30 min at the end of 5 h of SD, off periods are optogenetically induced. SWA for optrode and contralateral control probe, EMG and hypnogram are shown across the duration of the SD experiment and recovery sleep. Dashed line indicates the SWA mean over the 12-h NREM baseline (indicated at left by an asterisk). Black triangle indicates the first 45-s break in between bouts of off induction (Methods). b, LFP traces and MUA during off-period induction for an example optrode (top, blue) and contralateral control probe (bottom, gray) from one representative mouse. Please note the tight correspondence between laser pulses (blue arrows at top) and algorithmically detected off periods (red arrows, bottom). c, Off period frequency (left) and SWA (right) during the induction period, relative to the 12-h baseline NREM mean, for all SOM+ mice (top, dark blue) and ACR mice (bottom, light blue). SWA is normalized to the contralateral control probe. Dotted gold line represents mean in all box plots. d, Left, average SWA in 1 h of cumulative NREM sleep, averaged across all channels of all SOM+ animals. Values are calculated relative to the 12-h baseline NREM mean and then normalized to the contralateral control probe for visualization. Right, average PSD during 1 h of cumulative NREM sleep, across all channels of all SOM+ animals. Values are relative to the 12-h baseline NREM mean, shading represents ±s.e.m. e, Example NREM traces from a representative SOM+ animal in baseline and recovery sleep. Asterisks mark slow waves. Please note the tight correspondence in baseline, and dissociation in recovery. f, Same as in d, but for all ACR animals. g, Same as in e, but for a representative ACR animal. h, Left, example traces from one representative SOM+ animal showing field potential slopes at off-period initiation; data show an average of ±s.e.m. of traces from the first 1,500 off periods in baseline and recovery sleep. Vertical dotted line marks off-period start. Right, average field potential slope at off-period initiation across all SOM+ and ACR mice. Data in h–j are for the first 30 min of the NREM recovery, when values were most reliably above baseline. i, Same as in h, but for the peak off-period amplitude. Data are aligned to field potential peak during the off period, shown by vertical dotted line. j, Same as in h, but for the slope at off-period termination. Vertical dotted line marks off-period end. k, STTC, a rate-independent measure of neural synchrony, is calculated for a pair of spike trains. The measure is high when a large proportion of spikes from train-1 have at least one other spike from train-2 within ±5 ms (light green boxes). In the ‘high synchrony’ example, four of five spikes from train-1 have at least one partner spike in train-2 within ±5 ms, whereas in the ‘low synchrony’ example, only one of five spikes has a partner. l, STTC is calculated across all unique channel pairs for each probe of each mouse. Shown is the full STTC matrix for one representative SOM+ mouse’s contralateral control probe (top) and optrode (bottom) for the 1-h NREM recovery period (right) and the circadian-matched NREM period from the baseline (left). Values are expressed as a difference between the condition mean and the 12-h baseline NREM mean. m, The difference matrix (optrode–contralateral control) averaged across all SOM+ mice for the 1-h NREM recovery period (middle) and the circadian-matched period from the baseline (left). Right, all-channel STTC average across all SOM+ mice for the 1-h NREM recovery period relative to the full 12-h baseline NREM mean. n, Same as in m, but for ACR mice. In a–m, n = 10 SOM+ mice, and n = 8 ACR mice. In SWA and PSD panels, n = 9 ACR mice. In c, d, f, h–j, m and n, *P < 0.05; **P < 0.005; ***P < 0.0005. For all box plots in this figure, the box itself spans from the 25th percentile of the data (Q1) to the 75th percentile of the data (Q3), the solid horizontal line represents the median, the dashed horizontal line represents the mean, the bottom whisker extends to the furthest data point within 1.5× IQR (inner quartile range, Q3-Q1) below Q1 and the top whisker extends to the furthest data point within 1.5× IQR above Q3. For detailed information on all statistical tests in this figure, see Supplementary Table1. PSD, power spectral density. Off-period induction during wake reduces SWA and synchrony in subsequent sleep. a, Experiment overview; in the last 30 min at the end of 5 h of SD, off periods are optogenetically induced. SWA for optrode and contralateral control probe, EMG and hypnogram are shown across the duration of the SD experiment and recovery sleep. Dashed line indicates the SWA mean over the 12-h NREM baseline (indicated at left by an asterisk). Black triangle indicates the first 45-s break in between bouts of off induction (Methods). b, LFP traces and MUA during off-period induction for an example optrode (top, blue) and contralateral control probe (bottom, gray) from one representative mouse. Please note the tight correspondence between laser pulses (blue arrows at top) and algorithmically detected off periods (red arrows, bottom). c, Off period frequency (left) and SWA (right) during the induction period, relative to the 12-h baseline NREM mean, for all SOM+ mice (top, dark blue) and ACR mice (bottom, light blue). SWA is normalized to the contralateral control probe. Dotted gold line represents mean in all box plots. d, Left, average SWA in 1 h of cumulative NREM sleep, averaged across all channels of all SOM+ animals. Values are calculated relative to the 12-h baseline NREM mean and then normalized to the contralateral control probe for visualization. Right, average PSD during 1 h of cumulative NREM sleep, across all channels of all SOM+ animals. Values are relative to the 12-h baseline NREM mean, shading represents ±s.e.m. e, Example NREM traces from a representative SOM+ animal in baseline and recovery sleep. Asterisks mark slow waves. Please note the tight correspondence in baseline, and dissociation in recovery. f, Same as in d, but for all ACR animals. g, Same as in e, but for a representative ACR animal. h, Left, example traces from one representative SOM+ animal showing field potential slopes at off-period initiation; data show an average of ±s.e.m. of traces from the first 1,500 off periods in baseline and recovery sleep. Vertical dotted line marks off-period start. Right, average field potential slope at off-period initiation across all SOM+ and ACR mice. Data in h–j are for the first 30 min of the NREM recovery, when values were most reliably above baseline. i, Same as in h, but for the peak off-period amplitude. Data are aligned to field potential peak during the off period, shown by vertical dotted line. j, Same as in h, but for the slope at off-period termination. Vertical dotted line marks off-period end. k, STTC, a rate-independent measure of neural synchrony, is calculated for a pair of spike trains. The measure is high when a large proportion of spikes from train-1 have at least one other spike from train-2 within ±5 ms (light green boxes). In the ‘high synchrony’ example, four of five spikes from train-1 have at least one partner spike in train-2 within ±5 ms, whereas in the ‘low synchrony’ example, only one of five spikes has a partner. l, STTC is calculated across all unique channel pairs for each probe of each mouse. Shown is the full STTC matrix for one representative SOM+ mouse’s contralateral control probe (top) and optrode (bottom) for the 1-h NREM recovery period (right) and the circadian-matched NREM period from the baseline (left). Values are expressed as a difference between the condition mean and the 12-h baseline NREM mean. m, The difference matrix (optrode–contralateral control) averaged across all SOM+ mice for the 1-h NREM recovery period (middle) and the circadian-matched period from the baseline (left). Right, all-channel STTC average across all SOM+ mice for the 1-h NREM recovery period relative to the full 12-h baseline NREM mean. n, Same as in m, but for ACR mice. In a–m, n = 10 SOM+ mice, and n = 8 ACR mice. In SWA and PSD panels, n = 9 ACR mice. In c, d, f, h–j, m and n, *P < 0.05; **P < 0.005; ***P < 0.0005. For all box plots in this figure, the box itself spans from the 25th percentile of the data (Q1) to the 75th percentile of the data (Q3), the solid horizontal line represents the median, the dashed horizontal line represents the mean, the bottom whisker extends to the furthest data point within 1.5× IQR (inner quartile range, Q3-Q1) below Q1 and the top whisker extends to the furthest data point within 1.5× IQR above Q3. For detailed information on all statistical tests in this figure, see Supplementary Table1. PSD, power spectral density. In the first hour of NREM sleep following SD, SWA recorded by the optrode was reduced relative to SWA on the contralateral control probe, in both SOM+ (Fig. 2d) and ACR (Fig. 2f) mice. This difference is also evident in the raw traces from matched channels of the same animal, where an extremely tight correspondence in slow-wave magnitude and the general structure of the LFP is present during baseline NREM sleep but not during recovery NREM sleep (Fig. 2e,g). Furthermore, the reduction in SWA on the optrode recovered to equivalence with the contralateral control probe within a few hours of NREM sleep (Extended Data Fig. 2c) and was specific for slow frequencies (that is, was not a broadband reduction; Fig. 2d,f, right), not extending to other frequency bands tested (except the theta band for ACR mice; Extended Data Fig. 2d). After SD, overall sleep architecture (time spent in NREM and REM sleep, sleep bout duration, number of microarousals) was not affected in SOM+ and ACR mice (Extended Data Fig. 4). Spindle incidence and duration were slightly higher on the optrode compared to the contralateral control probe in ACR but not in SOM+ mice (Supplementary Fig. 3). When we performed the same off-period induction protocol in wild-type control mice, we did not see any change in overall firing rate, nor an increased frequency of off periods on the optrode relative to the contralateral control (Extended Data Fig. 5a). During the 1-h NREM recovery period, we also observed no change in SWA, firing rate or a general change in power spectral density (Extended Data Fig. 5b). Thus, the local induction of off periods during wakefulness leads to a local reduction of sleep pressure, as indexed by local SWA. Induction of off periods during wakefulness reduces network synchrony in subsequent NREM sleep Next, we investigated the mechanisms underlying the decline of SWA in recovery sleep after off-period induction during wake. We first determined that the frequency and duration of off periods during the first 1 h of recovery sleep did not differ between the optrode side and the contralateral control (Methods; Extended Data Fig. 6a,b). We then asked whether a decrease in the synchrony of neurons switching between on and off periods may explain the overall reduction in SWA, as predicted by modeling studies25,26. We examined features of the LFP that reflect neuronal synchrony, including the LFP slope at off-period initiation, the peak LFP amplitude within an off period and the LFP slope at off-period termination, all of which were reliably enhanced in the first 30 min of NREM recovery sleep compared to the 12-h baseline. In both SOM+ and ACR mice, we found that all three measures were reduced on the optrode compared to the contralateral control probe (Fig. 2h–j), indicating reduced neural synchrony after off-period induction. To examine synchrony at the level of neuronal firing (MUA), we computed the spike-time tiling coefficient (STTC), a firing-rate-independent measure of neural synchrony27. On every channel-wise pair of MUA spike trains from each probe, the STTC measures the proportion of spikes from train-1 that have at least one spike from train-2 within a certain window (±5 ms), normalized by the chance level implied by each train’s own firing rate, thus providing a more rate-robust measure of spike train synchrony than traditional binned correlations (Fig. 2k). An example STTC matrix for the 1-h NREM recovery period and the 1-h NREM circadian-matched period (the 1 h of NREM sleep starting exactly 24 h before the recovery period) for both the optrode and contralateral control probe is shown for a representative SOM+ animal in Fig. 2l. Again, we found a tight correspondence between the homotopic probes in the baseline period, followed by a dissociation during the recovery period (Fig. 2m,n, left). Across all participants, the STTC was significantly reduced on the optrode relative to the contralateral control when averaging across all-channel pairs for SOM+ (Fig. 2m) and ACR (Fig. 2n) mice. Similar results were obtained when STTC analyses were performed on all possible pairs of spike-sorted single units. Specifically, in both SOM+ and ACR mice, the majority (~2/3) of all pairs showed lower synchrony on the optrode relative to the contralateral control probe, although we note that these single-unit changes occurred over a smaller overall range of raw STTC values. Notably, these unit pairs were also those that were homeostatically regulated during baseline NREM sleep, that is, their STTC value was larger in early baseline sleep (high-sleep-pressure condition) compared to late baseline sleep (low-sleep-pressure condition). By contrast, STTC values in the remaining approximately one of three unit pairs neither changed between early and late baseline sleep, nor after off induction in wake (Extended Data Fig. 7). Thus, the decrease in SWA produced by the local induction of off periods is accompanied by a concomitant decrease of LFP, MUA and single-unit-activity markers of neuronal synchrony. Equivalent tonic reduction of firing rate cannot account for the effects of on-/off-period induction Is the reduction of SWA and cortical synchrony obtained by inducing off periods during wakefulness due to an overall reduction of firing rates, or does the sleep-like alternation of on/off periods matter? If sleep pressure were to increase because of tonic activity during wakefulness, reduced firing during NREM sleep might be sufficient to relieve that pressure28,29, regardless of how reduced firing is achieved. To address this issue, we used a third optogenetic approach paired with bilateral silicon probe recordings of homotopic cortical areas. We crossed CaMKIIα-Cre mice with mice expressing the inward chloride pump halorhodopsin (eNpHR3.0; Fig. 3a)30.Fig. 3Tonic inhibition using halorhodopsin does not reduce SWA or synchrony in subsequent sleep.a, Halorhodopsin mice express Natronomonas pharaonis-derived halorhodopsin (eNpHR3.0) in CaMKIIα-producing neurons (mainly pyramidal neurons, green cells indicated by arrows). When eNpHR3.0 is optogenetically activated, all CaMKIIα neurons in the cortical network are inhibited through a light-driven chloride pump. b, Experiment overview. In all halorhodopsin mice, unilateral tonic inhibition of overall MUA firing rate was performed for the final ~30 min of SD. All other aspects of the experimental design were identical to the off-period induction experiments previously described (Methods). Dotted line indicates the 12-h NREM baseline SWA mean (indicated at left by an asterisk). c, Raw data of 6 s from a representative mouse showing tonic inhibition of MUA using continuous activation of eNpHR3.0 on the optrode, and not on the contralateral control probe. Scale bars at right of each plot = 2 mV. d, Firing rate, off-period frequency and SWA on the optrode relative to the contralateral control during tonic inhibition using eNpHR3.0. Firing rate is expressed relative to the first hour of SD, off-period frequency and SWA are calculated relative to the 12-h NREM baseline mean and SWA is expressed normalized to the contralateral control for visualization. e, SWA and PSD during the 1-h NREM recovery period. PSD data represent mean ± s.e.m. f, Example NREM LFP traces in baseline and recovery sleep; please note that the tight correspondence between optrode and contralateral control LFP continues from baseline into recovery sleep. Shown is one matched layer 5 channel on each probe of a representative mouse (same mouse from c), with implant in M2. g, Total MUA firing rate, off-period duration and off-period frequency during the 1-h NREM recovery period. h, Field potential slopes at off-period initiation and termination, and slow-wave peak values, during the 1-h NREM recovery period. In e, g and h, all data are expressed relative to the 12-h NREM baseline mean. i, STTC difference (1-h NREM recovery period mean − 12-h NREM baseline mean). Right, the data in channel-wise form, expressed as the difference between the optrode and the contralateral control. In all panels, n = 7 Halo mice. In d, e, g, h and i, **P < 0.005; NS, P > 0.05. For all box plots in this figure, the box itself spans from the 25th percentile of the data (Q1) to the 75th percentile of the data (Q3), the solid horizontal line represents the median, the dashed horizontal line represents the mean, the bottom whisker extends to the furthest data point within 1.5× IQR (inner quartile range, Q3-Q1) below Q1 and the top whisker extends to the furthest data point within 1.5× IQR above Q3. For detailed information on all statistical tests in this figure, see Supplementary Table 1. Schematic in a created in BioRender; Driessen, K. https://biorender.com/ajppjxu (2026). Tonic inhibition using halorhodopsin does not reduce SWA or synchrony in subsequent sleep. a, Halorhodopsin mice express Natronomonas pharaonis-derived halorhodopsin (eNpHR3.0) in CaMKIIα-producing neurons (mainly pyramidal neurons, green cells indicated by arrows). When eNpHR3.0 is optogenetically activated, all CaMKIIα neurons in the cortical network are inhibited through a light-driven chloride pump. b, Experiment overview. In all halorhodopsin mice, unilateral tonic inhibition of overall MUA firing rate was performed for the final ~30 min of SD. All other aspects of the experimental design were identical to the off-period induction experiments previously described (Methods). Dotted line indicates the 12-h NREM baseline SWA mean (indicated at left by an asterisk). c, Raw data of 6 s from a representative mouse showing tonic inhibition of MUA using continuous activation of eNpHR3.0 on the optrode, and not on the contralateral control probe. Scale bars at right of each plot = 2 mV. d, Firing rate, off-period frequency and SWA on the optrode relative to the contralateral control during tonic inhibition using eNpHR3.0. Firing rate is expressed relative to the first hour of SD, off-period frequency and SWA are calculated relative to the 12-h NREM baseline mean and SWA is expressed normalized to the contralateral control for visualization. e, SWA and PSD during the 1-h NREM recovery period. PSD data represent mean ± s.e.m. f, Example NREM LFP traces in baseline and recovery sleep; please note that the tight correspondence between optrode and contralateral control LFP continues from baseline into recovery sleep. Shown is one matched layer 5 channel on each probe of a representative mouse (same mouse from c), with implant in M2. g, Total MUA firing rate, off-period duration and off-period frequency during the 1-h NREM recovery period. h, Field potential slopes at off-period initiation and termination, and slow-wave peak values, during the 1-h NREM recovery period. In e, g and h, all data are expressed relative to the 12-h NREM baseline mean. i, STTC difference (1-h NREM recovery period mean − 12-h NREM baseline mean). Right, the data in channel-wise form, expressed as the difference between the optrode and the contralateral control. In all panels, n = 7 Halo mice. In d, e, g, h and i, **P < 0.005; NS, P > 0.05. For all box plots in this figure, the box itself spans from the 25th percentile of the data (Q1) to the 75th percentile of the data (Q3), the solid horizontal line represents the median, the dashed horizontal line represents the mean, the bottom whisker extends to the furthest data point within 1.5× IQR (inner quartile range, Q3-Q1) below Q1 and the top whisker extends to the furthest data point within 1.5× IQR above Q3. For detailed information on all statistical tests in this figure, see Supplementary Table 1. Schematic in a created in BioRender; Driessen, K. https://biorender.com/ajppjxu (2026). Using these halorhodopsin mice, we were able to tonically inhibit nearly all neurons in the unilateral cortical network during the final ~30 min of SD (Fig. 3b–d), affecting to a similar degree supragranular and infragranular channels (Extended Data Fig. 1g–i). Crucially, this large reduction of overall firing rate did not differ from the one seen during off-period induction in SOM+ and ACR mice (Extended Data Fig. 8a,b). However, in halorhodopsin mice NREM sleep-like on/off periods were not triggered and SWA in recovery sleep was not significantly affected on the optrode side relative to the contralateral control probe (Fig. 3d,e and Extended Data Fig. 8c,d). This was again evident in the raw NREM LFP traces, as the tight correlation across matched channels seen in the NREM baseline period continued into the NREM recovery period following SD (Fig. 3f). Across all other measures examined for the off-period induction experiments, no significant effects were found on the optrode after the tonic inhibition of firing rate during wake. Thus, during the 1-h NREM recovery period after tonic inhibition, off-period duration and frequency, overall firing rate, LFP slope at off-initiation and off-termination, slow-wave peak amplitude and spike-time synchrony showed no significant differences between optrode and contralateral control (Fig. 3g–i and Extended Data Fig. 7i,j). We further investigated the effects of tonic inhibition in a subset of SOM+ mice. We performed a separate experiment in which we tonically reduced firing rate via continuous sinusoidal stimulation of SOM-ChR2 neurons (Extended Data Fig. 9a,b). Again, this resulted in an overall reduction of firing rate comparable to the off-period induction protocol (Extended Data Fig. 9c), but few synchronized off periods (Extended Data Fig. 9d), and no significant reduction of SWA in subsequent NREM recovery sleep (Extended Data Fig. 9e). Critically, when directly comparing within the same mice the ability to reduce SWA in NREM recovery sleep, the induction of on/off periods during wake was more effective than a tonic reduction of firing rate, in every mouse tested (Extended Data Fig. 9e, right). These results provide converging evidence, across two in vivo optogenetic approaches, that the on/off bistable activity patterns that characterize NREM sleep, but not mere reduction of firing, can release sleep pressure and reduce neural synchrony in a local cortical network even while the animal is awake and behaving. As such, all further experiments and analyses focus solely on the off-period induction protocol, and on the ability of on/off activity patterns to fulfill key homeostatic sleep functions when induced during wake. Induction of off periods during wakefulness leads to a reduction of molecular markers of synaptic strength Modeling studies also predict that the homeostatic decline of SWA and synchrony during sleep (together with the decline of slow-wave amplitude and slope, and overall firing rate; Extended Data Fig. 6c) can be accounted for by a net reduction in excitatory synaptic strength in cortical networks25,26. Such a reduction has been confirmed experimentally using electrophysiology31,32, ultrastructural measurements of synaptic size and density33, as well as molecular markers8,9,31. We therefore took advantage of the same molecular markers of synaptic strength to determine whether the induction of off periods during wakefulness can have similar effects as sleep on cortical synaptoneurosomes. We implanted SOM+, ACR and wild-type control mice with EEGs over parietal cortex. On one side of the brain, four optic ferrules were implanted within a 3-mm diameter circle over parietal cortex, with each ferrule 1 mm from the ‘opto-EEG’ (Fig. 4a), thus enabling the induction of NREM sleep-like on/off bistable activity throughout a large area of unilateral parietal cortex. These mice underwent the same SD procedure as those implanted with silicon probes, except after the 30 min of unilateral off-period induction at the end of SD, mice were immediately killed and were not allowed to sleep. A mirrored 3-mm diameter circle of cortex was collected from each side of the brain, centered over each of the bilateral EEGs (Fig. 4a, left). Using this tissue, synaptoneurosomes were extracted and two established markers of excitatory synaptic strength34 were measured—the amount of cortical synaptic expression of AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid) receptors containing the GluA1 subunit and the amount of phosphorylation of GluA1 at serine 845 (pGluA1(845)).Fig. 4Off-period induction during wake reduces molecular markers of cortical excitatory synaptic strength.a, Experiment overview. SOM+, ACR and control mice are implanted with bilateral EEGs over parietal cortex and four optic ferrules on one side. Off-period induction in one half of the brain occurs at the end of SD (final 30 min), followed by brain collection. A 3-mm-diameter circle of cortical tissue from each brain hemisphere (pink circular region) was used to extract synaptoneurosome. b, SWA during SD in the opto-EEG and the contralateral control EEG for SOM+, ACR and control mice. Data represent mean ± s.e.m. c, Average EEG SWA during off-period induction, normalized to contralateral control EEG. d, Synaptic levels of GluA1 for both hemispheres in SOM+, ACR and control mice. Representative immunoblots are shown for each experimental group. Data were normalized by the intensity of the GAPDH band and expressed as means ± s.e.m. e, As in d, but for synaptic levels of phosphoGluA1. Brightness and contrast were adjusted for all blot snippets in this figure, solely for space and clarity considerations. Please note that, for each group, the example GluA1 and pGluA1 immunoblots shown in d and e are from the same participant and the same gel, and as a result the GAPDH bands are identical. In all panels, n = 8 mice in all groups. The red arrows indicate the bands used for quantification. In c–e, **P < 0.005; ***P < 0.0005; NS, P > 0.05. For all box plots in this figure, the box itself spans from the 25th percentile of the data (Q1) to the 75th percentile of the data (Q3), the solid horizontal line represents the median, the dashed horizontal line represents the mean, the bottom whisker extends to the furthest data point within 1.5× IQR (inner quartile range, Q3-Q1) below Q1 and the top whisker extends to the furthest data point within 1.5× IQR above Q3. For detailed information on all statistical tests in this figure, see Supplementary Table 1. Schematic in a created in BioRender; Driessen, K. https://biorender.com/ajppjxu (2026). Off-period induction during wake reduces molecular markers of cortical excitatory synaptic strength. a, Experiment overview. SOM+, ACR and control mice are implanted with bilateral EEGs over parietal cortex and four optic ferrules on one side. Off-period induction in one half of the brain occurs at the end of SD (final 30 min), followed by brain collection. A 3-mm-diameter circle of cortical tissue from each brain hemisphere (pink circular region) was used to extract synaptoneurosome. b, SWA during SD in the opto-EEG and the contralateral control EEG for SOM+, ACR and control mice. Data represent mean ± s.e.m. c, Average EEG SWA during off-period induction, normalized to contralateral control EEG. d, Synaptic levels of GluA1 for both hemispheres in SOM+, ACR and control mice. Representative immunoblots are shown for each experimental group. Data were normalized by the intensity of the GAPDH band and expressed as means ± s.e.m. e, As in d, but for synaptic levels of phosphoGluA1. Brightness and contrast were adjusted for all blot snippets in this figure, solely for space and clarity considerations. Please note that, for each group, the example GluA1 and pGluA1 immunoblots shown in d and e are from the same participant and the same gel, and as a result the GAPDH bands are identical. In all panels, n = 8 mice in all groups. The red arrows indicate the bands used for quantification. In c–e, **P < 0.005; ***P < 0.0005; NS, P > 0.05. For all box plots in this figure, the box itself spans from the 25th percentile of the data (Q1) to the 75th percentile of the data (Q3), the solid horizontal line represents the median, the dashed horizontal line represents the mean, the bottom whisker extends to the furthest data point within 1.5× IQR (inner quartile range, Q3-Q1) below Q1 and the top whisker extends to the furthest data point within 1.5× IQR above Q3. For detailed information on all statistical tests in this figure, see Supplementary Table 1. Schematic in a created in BioRender; Driessen, K. https://biorender.com/ajppjxu (2026). In both SOM+ and ACR mice, EEG recordings confirmed that the off-period induction was accompanied by a large increase in SWA in the corresponding hemisphere (Fig. 4b,c). In both SOM+ and ACR mice (but not in controls), the cortical hemisphere that underwent off-period induction also had significantly lower synaptic levels of GluA1-containing AMPA receptors (Fig. 4d) and their phosphorylation at serine 845 (Fig. 4e and Supplementary Fig. 4). In both direction and magnitude, these changes mirror those seen after 6–7 h of natural sleep9,31. Because in these experiments mice were not allowed any sleep after SD, the changes in AMPA receptors must have been triggered by the forced induction of NREM sleep-like on/off activity patterns during wakefulness. Induction of off periods during SD recovers memory performance in a new floor texture recognition (FTR) task There is substantial evidence that sleep has a critical role in learning and memory consolidation7. Moreover, at least some of these sleep benefits may be mediated by sleep-dependent synaptic downselection. For example, after learning a complex wheel task, adult mice that are allowed to sleep immediately after learning perform better in the next training session than mice that are sleep deprived, and the weakening of nontask-engaged excitatory synapses is correlated with that improvement (revealed through two-photon microscopic imaging of super ecliptic pHluorin-GluA1 expression35). This raises the question of whether inducing sleep-like on/off periods during wakefulness can provide a memory or learning benefit similar to that seen after sleep. To test this possibility, we used an FTR task that is known to benefit from sleep and involves neuronal activity localized to secondary motor cortex (M2) and primary sensory cortex (S1)15. We focused on SOM+ mice because of the involvement of somatostatin-positive interneurons in the generation of NREM sleep slow waves in natural conditions23 and implanted them with bilateral optic ferrules over M2 and S1 (Fig. 5a). Mice were exposed to an arena with identical floor textures on either side of the chamber for 15 min (‘acquisition’ period). After acquisition, mice were returned to their home cage and either allowed to sleep freely (sleep group), sleep deprived for 1 h (SD group) or sleep deprived for 1 h with concurrent induction of NREM sleep-like off periods (SD + off induction group). Mice were recorded in their home cages, and we verified the presence of sleep in the group of mice immediately allowed to return to their cage and sleep (Supplementary Fig. 5). We also verified that neither overall time spent asleep nor locomotor activity differed between SD and SD + off induction mice in the 1-h and 6-h periods after SD (Extended Data Fig. 10f–j). Next, 24 h after acquisition mice were placed back into the arena, where now half of the arena contained a floor with a new texture (‘recall’ period; Fig. 5b). We then used a neural network trained to recognize the mouse’s position within the arena to quantify the amount of time the mouse spent on each side of the arena within the first 5 min of the recall period36. Replicating previous findings15, we found that mice allowed to sleep immediately after acquisition performed better than mice sleep deprived for 1 h after acquisition (Fig. 5c,d). Moreover, in SOM+ mice that underwent bilateral induction of NREM sleep-like off periods during 1 h of SD after acquisition, performance was rescued back to the level of the mice allowed to sleep (Fig. 5c,d). Within each of the three groups, the time spent asleep before acquisition (either 1 h or 6 h before) did not correlate with performance during recall (Extended Data Fig. 10a–e). Thus, inducing sleep-like on/off periods during wakefulness can benefit memory consolidation similar to sleep.Fig. 5Bilateral off induction during SD recovers memory performance in a novel FTR task.a, Implant schematic. Optic ferrules are implanted bilaterally over M2 and S1 cortices. b, After acquisition of the FTR task, mice return to the home cage where they are either allowed to sleep, sleep deprived for 1 h or sleep deprived while off periods are induced for 1 h. The next day, mice are tested on FTR. c, Time-resolved plot showing the nose position within the arena over the 5-min recall period for a representative mouse from the sleep, SD and SD + off-induction groups. d, Performance (ratio of time spent on new side relative to familiar side) for each of the three groups (n = 9 sleep mice, n = 13 SD mice, n = 8 SD + off-induction mice). In d, *P < 0.05. For all box plots in this figure, the box itself spans from the 25th percentile of the data (Q1) to the 75th percentile of the data (Q3), the solid horizontal line represents the median, the dashed horizontal line represents the mean, the bottom whisker extends to the furthest data point within 1.5× IQR (inner quartile range, Q3-Q1) below Q1 and the top whisker extends to the furthest data point within 1.5× IQR above Q3. For detailed information on all statistical tests in this figure, see Supplementary Table 1. Schematics in a and b created in BioRender; Driessen, K. https://biorender.com/ajppjxu (2026). Bilateral off induction during SD recovers memory performance in a novel FTR task. a, Implant schematic. Optic ferrules are implanted bilaterally over M2 and S1 cortices. b, After acquisition of the FTR task, mice return to the home cage where they are either allowed to sleep, sleep deprived for 1 h or sleep deprived while off periods are induced for 1 h. The next day, mice are tested on FTR. c, Time-resolved plot showing the nose position within the arena over the 5-min recall period for a representative mouse from the sleep, SD and SD + off-induction groups. d, Performance (ratio of time spent on new side relative to familiar side) for each of the three groups (n = 9 sleep mice, n = 13 SD mice, n = 8 SD + off-induction mice). In d, *P < 0.05. For all box plots in this figure, the box itself spans from the 25th percentile of the data (Q1) to the 75th percentile of the data (Q3), the solid horizontal line represents the median, the dashed horizontal line represents the mean, the bottom whisker extends to the furthest data point within 1.5× IQR (inner quartile range, Q3-Q1) below Q1 and the top whisker extends to the furthest data point within 1.5× IQR above Q3. For detailed information on all statistical tests in this figure, see Supplementary Table 1. Schematics in a and b created in BioRender; Driessen, K. https://biorender.com/ajppjxu (2026). Discussion We show here that sleep-like neuronal off periods can be reliably induced in local cortical networks while mice are awake and behaving. Off periods induced during sustained wakefulness locally reduce sleep pressure, as assessed by a decline of SWA and neuronal synchrony, compared to the contralateral homotopic cortex. Sleep pressure is relieved if off periods alternate with on periods every few hundred milliseconds, as it happens in NREM sleep, while an equivalent overall reduction of neuronal firing rate obtained through tonic inhibition is not effective. The induction of sleep-like off periods during wakefulness also leads to a decrease in excitatory synaptic strength, as assessed by a reduced expression and phosphorylation of GluA1-containing AMPA receptors, similar to what happens after NREM sleep. Finally, inducing off periods bilaterally over motor and sensory cortices during waking after learning restores performance in a recognition task, similar to what sleep does. Thus, local off periods occurring during wakefulness can offer key benefits normally provided by sleep. During NREM sleep, the induction and maintenance of the characteristic bistability across synchronous on/off periods is enabled by the low level of activity of all major arousal systems, which leads to neuronal disfacilitation. Blocking noradrenergic but not cholinergic transmission during wakefulness is sufficient to trigger local off states in the rat barrel cortex37, while in fur seals unihemispheric slow-wave sleep is associated with lower levels of acetylcholine, but not of noradrenaline, in the sleeping hemisphere compared to the waking hemisphere20, consistent with in vitro evidence that both noradrenaline and acetylcholine strongly oppose cortical on/off bistability38. Active inhibition by cortical SOM+ interneurons also has a role in promoting bistability23,39. SOM+ interneurons are activated by strong synchronous firing of pyramidal neurons40,41 through progressively facilitating synapses42–44. When enough SOM+ interneurons are recruited, they act as ‘master regulators’ of cortical excitability by inhibiting all other cell types but themselves45–47. As shown here, inhibition of pyramidal neurons through SOM+ interneurons is sufficient to induce local off periods during waking for as long as we tried (30–60 min). Direct inhibition through stGtACR1 can also produce off periods, although the mechanisms are likely less physiological, as indicated by the opposite polarity of the field potential compared to spontaneously occurring and SOM+-induced off periods (Fig. 1h). The occurrence of off periods during NREM sleep underlies SWA13,14. In fact, increased SWA is associated with more synchronous and longer off periods14. SWA is a reliable marker of sleep pressure in both rodents and humans48,49. There is also abundant evidence that the level of SWA, as well as the amplitude and slope of individual slow waves, reflects changes in net cortical synaptic strength, increasing with the duration of wakefulness and decreasing homeostatically during sleep4,31,50,51, in line with modeling studies25,26. An intriguing possibility is that SOM+ interneurons may be involved in mediating the increased occurrence, duration and synchrony of off periods associated with increased sleep pressure. Synaptic activation of NMDA receptors on the dendrites of cortical pyramidal neurons leads to long-term potentiation of the inhibitory synapses established by SOM+ interneurons, suggesting a mechanism through which wake-induced potentiation of excitatory cortical synapses could lead to stronger SOM+-mediated dendritic inhibition and increased propensity to enter an off state52. It has been suggested that SWA and associated off periods do not simply reflect sleep pressure as a consequence of net synaptic potentiation, but may be causally involved in its homeostatic regulation by enacting synaptic renormalization4,53. In vitro experiments in rodent54,55 and human56 cortex show that bistable, on/off conditions are typically associated with synaptic depression, although the underlying mechanisms remain unclear. In vivo experiments under urethane anesthesia are consistent with this conclusion57, but the evidence in naturally sleeping animals is limited and indirect58. The present results demonstrate in vivo that sleep-like on/off alternations lead to the weakening of excitatory glutamatergic synapses. This is directly confirmed by the reduced levels of GluA1-containing AMPA receptors and their phosphorylation, and indirectly by the reduced amplitude and synchrony of on/off periods and associated slow waves during subsequent sleep. The effect on synchrony was not obtained by an equivalent reduction in firing rate induced by tonic inhibition, suggesting that repeated sequences of on/off transitions may be critical for synaptic depression. Moreover, single-unit analysis suggests that homeostatically driven changes in neural synchrony may occur in the majority, but not all, of neurons. Remarkably, off-period induction was effective in relieving local sleep pressure when performed in awake animals. Assuming that the overall levels of noradrenaline and acetylcholine were those typical of wakefulness (high), rather than of NREM sleep (low), our results would suggest that low levels of neuromodulators may be permissive for enabling bistable activity, but not necessary for producing synaptic renormalization, as long as on/off periods can be triggered through other means. In vitro experiments have indicated that low levels of noradrenaline coupled with low levels of brain-derived neurotrophic factor may be conducive to synaptic depression and the removal of AMPA receptors from the synaptic membrane59–61. It will be important to establish whether the local induction of off periods may indirectly affect the local level of neuromodulators. Previous experiments in mice found that the removal of AMPA receptors depends on the activation of constitutive, ligand-independent mGluR1/5 signaling after the accumulation of Homer1a in the dendritic spines, which requires the drop in noradrenaline levels that is typically observed at sleep onset, or the buildup of adenosine in the course of SD59. In our experiments, adenosine levels likely increased on both sides of the cortex after 4.5 h of SD, but synaptic strength decreased on the side with off induction. The weakening of cortical synapses may thus require activity bursts, perhaps linked to activation of mGluR1/5 signaling, as observed in cerebellar and hippocampal synapses62. In previous work, we showed that extended periods of sustained wakefulness, especially if coupled with intense learning, lead to the spontaneous occurrence of on/off patterns of activity, which we termed ‘local sleep’63. Local sleep has been demonstrated in both rodents63 and humans64. Its spontaneous occurrence is sporadic and can disrupt performance if it occurs in the wrong brain region at the wrong time63. The present findings raise the question whether sustained periods of local sleep during extended wakefulness (provided they do not substantially interfere with ongoing activities63) may offer some of the benefits of sleep by regulating local synaptic strength and by preventing negative consequences on cellular processes. A systematic induction of on/off transitions over specific networks in an otherwise awake brain may be much more effective in restoring homeostasis and performance than sporadic off periods. Species such as bottlenose dolphins19,65, beluga whales66, fur seals67, mallard ducks68 and frigate birds69 exhibit unihemispheric NREM sleep, presumably enabling sleep’s restorative functions to take place in one hemisphere while the other maintains vigilance70. Conceivably, technologies capable of inducing on/off periods on demand in select brain regions may provide some of the benefits of sleep without paying the full price of sensory and motor disconnection. On the other hand, even if sleep can be induced and regulated locally, the overall disconnection enforced by sleep is likely necessary to ensure that synaptic renormalization and memory consolidation can take place not just for local circuits, but at the systems level, enabling the spontaneous activation of brain-wide circuits6. Methods Animal husbandry All surgical and experimental protocols described in this study were approved by the University of Wisconsin Institutional Animal Care and Use Committee. Mice were housed on a 12-h light/12-h dark cycle, room temperature was maintained at 23 °C ± 1 °C and 50% humidity. All experiments used adult mice (>8 weeks at time of surgery); male and female mice were used in roughly equal proportions in all experiments, based on the availability at the time of surgery. The ACR mouse line was generated by crossing R26-LNL-GtACR1-Fred-Kv2.1 male mice (Jackson Labs, 033089)24,71 with B6.Cg-Tg(Camk2a-cre)T29-1Stl/J female mice (Jackson Labs, 005359)72. Heterozygous offspring were used as wild-type controls. The SOM+ mouse line was generated by crossing Ai32(RCL-ChR2(H134R)/EYFP)73 male mice (Jackson Labs, 024109) with Sst-IRES-Cre female mice (Jackson Labs, 013044)74. The halorhodopsin mouse line was generated by crossing Ai39(RCL-eNpHR3.0/EYFP)73 male mice (Jackson Labs, 014539) with B6.Cg-Tg(Camk2a-cre)T29-1Stl/J female mice (Jackson Labs, 005359)72. Immunohistochemical labeling was performed on each of the three mouse lines to confirm expected expression pattern. Selected images for each of the three mouse lines are shown in Supplementary Fig. 6. Surgery Preparation of mice for silicon probe electrophysiology experiments All silicon probes used in this study consisted of 16 channels in a linear array, with 50-µm channel spacing (NeuroNexus Technologies, A1x16-3mm-50-177-CM16LP). The ‘optrodes’ referred to in these experiments were manually constructed by physically attaching an optic ferrule (Doric Lenses, MFC_200/240-0.22_8mm_MF2.5_DFL) to a silicon probe. The desired target location in the brain for the electrodes on the silicon probes was planned, and the optic ferrule was cemented to the probe such that the tip of the ferrule would end up just at the surface of the brain when the probe was inserted to its final depth (that is, the ferrule did not penetrate into the brain). The ferrule was positioned such that its tip was ~150 to 200 µm from the shank of the probe, with a 12°–15° angle relative to the shank. Mice were implanted with bilateral silicon probes in homotopic cortical areas (that is, at the same anterior–posterior (AP) and dorsal–ventral (DV) coordinates, and at mirrored medial–lateral (ML) coordinates) of either the frontal cortex or parietal cortex. The stereotaxic coordinates relative to bregma (the DV axis is relative to the brain’s surface) for the implant locations were as follows: (1) for frontal animals with an implant in secondary motor area (M2)—AP + 2.4 mm, ML ± 1.6 mm, DV = −1.15 mm, θ = 3°; (2) for parietal animals with an implant in primary somatosensory cortex (S1)—AP = −1.4 mm, ML ± 2.0 mm, DV = −0.97 mm, θ = 0°; and (3) for parietal animals with an implant in posterior parietal cortex—AP = −2.0 mm, ML ± 2.15 mm, DV = −0.92 mm, θ = 4°. Before the surgery, mice were given one dose of ceftriaxone (100 mg kg−1) and one dose of dexamethasone (2.5 mg kg−1). They were allowed ad libitum oral carprofen (Bio-Serv) for one day before and 3–6 days after surgery. Mice were anesthetized using 1–2% isoflurane and body temperature was maintained using an electronic heating pad. Mice were fixed onto a stereotaxic rig (Kopf Instruments, model 942), and the skull was exposed. A circular craniotomy of 1-mm diameter was drilled over the target location for each probe. One stereotaxic arm was then used to position each probe at the target location over the craniotomy, and the probes were slowly lowered to the required depth in the brain. The craniotomies were sealed with a biologically safe silicone polymer (‘Kwik-Sil’; World Precision Instruments), and the probes were cemented to the skull with dental cement (‘C&B Metabond’; Parkell). Reference and ground screws were placed over the cerebellum. Stainless steel EMGs were implanted into the nuchal muscles at the end of the surgery. All mice were given at least 1 week of full undisturbed recovery after surgery. Once mice had recovered, they were briefly anesthetized with 1–2% isoflurane and connected to the electrophysiology acquisition system and the implanted ferrule(s) were connected to a patch cord (Doric Lenses) for delivery of light in optogenetic experiments. Preparation of mice for quantitative immunoblotting experiments For mice used for quantitative immunoblotting experiments, EEGs and optic ferrules were implanted. First, two 0.7 mm burr holes were drilled and two stainless steel screw EEGs were implanted over parietal cortex (AP = −2.0 mm, ML ± 2.5 mm). Then, four 0.5 mm burr holes were drilled, each 1 mm away from one of the EEGs on only one side of the brain (Fig. 4). Optic ferrules were slowly lowered into the hole until they reached the brain surface, and the hole was sealed with a clear silicone polymer (Kwik-Sil) before all EEGs and ferrules were further cemented to the skull. Reference and ground screws were implanted over cerebellum. Surgical procedures were otherwise identical to those described above for mice receiving bilateral silicon probe implants. Preparation of mice for novel FTR experiments For mice used in tests of novel FTR, four optic ferrules were implanted. Two ferrules were implanted bilaterally over M2 (AP + 2.20 mm, ML ± 0.65 mm), and two were implanted bilaterally over S1 (AP = –0.73 mm, ML ± 1.95 mm). All other surgical procedures were otherwise identical to those described in the above two sections describing surgical preparation. Electrophysiology acquisition and analysis Electrophysiology data acquisition For acquiring EEG and EMG signals, copper hookup wire was used to connect the animal to the electrophysiology acquisition system. Raw EEG and EMG data were acquired and stored at 1,017 Hz. Silicon probes were connected to the electrophysiology acquisition system through low-profile headstages (Tucker-Davis Technologies, LP16CH-Z). Raw data from the silicon probes were acquired and stored at 24.414 kHz. All electrophysiology data were acquired using a battery-powered amplifier (Tucker-Davis Technologies, PZ5) in conjunction with a signal processor (Tucker-Davis Technologies, RZ2) and standard acquisition software (Synapse Neurophysiology Suite; Tucker-Davis Technologies). Electrophysiology data processing Generally, all electrophysiological analyses were performed using custom Python code. Code required to reproduce statistical tests and plots in the study (as well as in all statistical source data) is available at the GitHub repository (https://github.com/CSC-UW/off_PERIOD_INDUCTION_MATERIALS). For MUA-based analyses, raw 24.4 kHz data acquired from the silicon probes were bandpass filtered between 300 Hz and 12 kHz, and a global common median reference was applied. Noise levels on the filtered data were estimated using median absolute deviations, and a spike threshold was set at four times the estimated noise level. For every channel, peaks were detected when the threshold was crossed, and spike times were recorded. Where firing rates are reported, firing rate was computed in 2 s windows across the duration of the recording. All MUA preprocessing and detection were done using standard routines from the SpikeInterface Python package75. For LFP-based analyses, raw 24.4 kHz data were low-pass filtered using a finite impulse response filter and decimated to 400 Hz. Raw EEG and EMG data were acquired at 1017 Hz and not decimated before analysis. After all data processing, bad channels were manually marked, interpolated where possible (through a simple spatial interpolation using the voltages on neighboring channels75), and excluded from a given analysis if interpolation could not resolve data quality or stability problems, although this was rare. Throughout the paper, when silicon probe data or analyses derived from them are presented at the level of comparisons between animals and probes (for example, time-frequency analyses, firing rate analyses, synchrony analyses, etc.), data always reflect the average of all viable channels across the probe, unless otherwise noted. Time-frequency analyses Time-frequency analyses were performed on LFP or EEG data following the preprocessing steps outlined above. Power spectral density was estimated with a short-time Fourier transform using 4 s Slepian (DPSS) windows with time–half-bandwidth product NW = 4 and 50% overlap (2 s hop). SWA was computed by integrating the spectral power in the delta band (0.5–4 Hz). SWA was expressed relative to the NREM mean of the 12-h light-period baseline value. Additionally, for time-frequency visualizations aligned with on/off periods or laser pulse onsets/offsets, the LFP was decomposed using a Morlet wavelet convolution implemented via fast Fourier transform-based convolution. Wavelet power was computed at 80 linearly spaced center frequencies from 1 to 80 Hz (1 Hz steps) with a frequency-adaptive number of cycles that scaled linearly from three cycles at 1 Hz to ten cycles at 80 Hz. To generate event-locked spectrograms, epochs were extracted from the continuous wavelet power time series in windows time locked to event onsets (for example, off-period starts or optogenetic pulse onsets). Epochs were averaged across events and normalized to the mean power at each frequency and channel computed over NREM sleep during the circadian-matched baseline period or late SD period. Off-period detection and synchrony analyses Off periods were first detected at the single-channel level using a given channel’s MUA spike train. At the single-channel level, an on period was defined as any period of continuous spiking activity with no interspike intervals greater than or equal to 50 ms. Off periods were defined as any period of total silence lasting at least 50 ms, with an on period of at least 30 ms in duration on either side (before or after) of the off period (that is, an off period cannot be surrounded by two single-spike ‘on periods’). Global off periods were then detected as periods when most channels (12 or more of 16) were synchronously off for between 50 and 400 ms. Global on periods were defined as periods in which <12 channels were synchronously off for between 50 and 2,000 ms. Unless otherwise noted, all off-period analyses presented throughout the article refer to the global off-period definition. In rare cases with very low MUA on a number of channels, the definition of off period used above, which requires off periods to be preceded and followed by more than one spike, resulted in an extremely low off-period detection rate. In these cases, off periods were defined as a period of zero MUA spiking across all channels lasting at least 50 ms, as in ref. 14, and the same definition and detection method were always applied to both probes for a given participant (optrode and contralateral control). To detect off-period initiation slopes, trace snippets for each channel were extracted and aligned relative to all off-period start times within a given condition (for example, 1-h NREM recovery sleep). Then, an average waveform was calculated for all channels across all off periods in the condition. The first-order derivative of these traces was computed after applying a Savitsky–Golay filter, and the maximal value of the first-order derivative within ±10 samples of the off-period start sample was taken as the maximal initiation slope for that channel. The same procedure was followed for the off-period termination slope, except traces were aligned to the off-period end, rather than the start. To detect the off-period-associated slow-wave peak values, trace snippets for each channel were extracted and aligned to the midpoint of the off period. Average waveforms were calculated for all channels across all off periods, and the peak value of the field potential within ±20 samples of the off-period midpoint sample was recorded. STTCs were computed as described in ref. 27. Briefly, for a given pair of spike trains A and B, we computed the proportion of spikes in each train that had at least one coincident partner spike in the other train within a ±5 ms window (TA, TB), and the fraction of the total recording time that already lay within ±5 ms of any spike in the same train (PA, PB, that is, the union of (t – Δt, t + Δt) around spikes of A or B, respectively), which were then combined to yield the symmetric and rate-corrected metric below:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{STTC}=\frac{1}{2}\left(\frac{{T}_{{\rm{A}}}-{P}_{{\rm{A}}}}{1-\,{T}_{{\rm{A}}}{P}_{{\rm{A}}}}+\frac{{T}_{{\rm{B}}}-{P}_{{\rm{B}}}}{1-\,{T}_{{\rm{B}}}{P}_{{\rm{B}}}}\right)\,$$\end{document}STTC=12TA−PA1−TAPA+TB−PB1−TBPB STTC was computed for every unique channel pair on a per-epoch basis (that is, for each continuous bout within a condition) and all epochs were averaged across the entire condition. STTC values were Fisher z transformed (z = atanh(r)) after clipping r between (−1 + ε, 1 − ε), where ε = 1 × 10−6 before computing condition means or differences in the coefficient (for example, between baseline and recovery sleep, or between optrode and contralateral control) or running statistical tests. For analyses of single-unit-based synchrony, NREM sleep data were concatenated across the early baseline, circadian-matched baseline and NREM recovery sleep periods, and spike sorted using MountainSort5 implemented through the SpikeInterface package, using default parameters75,76. Data were preprocessed as for MUA data and whitened before spike sorting. After spike sorting was completed, to ensure units could be stably tracked over all periods of interest, only units with a presence ratio of 1 were retained (that is, for each unit, spikes had to be present across the entire concatenated recording, which resulted in the filtering out of ~52.1% of all detected units). After filtering, the mean number of units per probe is 12.6 with an s.d. of 5.71 units. Spindles Spindle detection was implemented on the LFP signals described above through the ‘YASA’ (‘yet another spindle algorithm’) package77, with only minor tweaks to default parameters applied. In short, three thresholds needed to be crossed simultaneously for a spindle to be detected, all of which were computed continuously from the LFP signal. First, at least 30% signal’s total power in the broadband frequency (1–30 Hz) must have been contained in the sigma band (9–16 Hz). Second, the moving Pearson correlation coefficient between the raw LFP and the sigma-filtered LFP needed to be at least 0.65. Third, the moving root mean square (RMS) of the sigma-filtered LFP needed to be greater than 2 s.d. above the mean (Supplementary Fig. 3a). For all NREM spindle analyses, only NREM data were used (that is, wake data were excluded from the detection pipeline). Finally, any two spindles occurring within 500 ms of each other were merged into a single spindle, and spindles were only included if their duration was between 0.5 and 2 s77. Sleep scoring Arousal states (wake, NREM sleep, REM sleep or intermediate state in the transition to REM sleep) were manually assigned using custom software written in Python (https://github.com/CSC-UW/sleepscore), based on a fork of visbrain sleep78. When EEGs were available, scorers used all EEG channels to make state decisions. When only LFPs from silicon probes were available, experiments used a fraction of channels (~1/4) spanning the probes to make state decisions. Ambiguous or unclear states were marked as ‘unsure’, although this constituted a very small minority of all labeled data (<0.5%). Artifact was manually marked where obvious and feasible, but, because of the variance across channels and probes, it was largely not possible to comprehensively label artifactual periods manually. As such, bad channels were interpolated individually as needed, and artifactual channels or epochs were further excluded per experiment. Scorers’ labeling data were blinded to animal and experiment identity where possible. Unless otherwise noted, the ‘baseline’ condition refers to all periods scored as NREM sleep during the 12-h light period on the day before the experiment. The ‘recovery sleep’ condition refers to the first 1 h of cumulative NREM sleep after SD. The ‘circadian-matched baseline’ condition refers to the first 1 h of cumulative NREM sleep beginning exactly 24 h before the start of the recovery sleep condition. SD and optogenetics The day before all electrophysiology experiments described in this paper, a 24-h baseline recording was acquired, starting at light onset on the day before the experiment (that is, 24 h before the experiment begins). The experiment consisted of a 5-h SD by exposure to new objects, beginning at light onset on the day of the experiment. After placing a new object in the home cage, the mouse explored the object for a period of time; when the mouse became disinterested in the object, stopped exploring or looked as if it was preparing for sleep (nesting, grooming, etc.), the old object(s) was removed, and a new object(s) was placed into the cage. This process was repeated for the entirety of the 5-h SD. During the final 30 min of the SD, off periods were selectively and optogenetically induced in the cortical network recorded by the optrode. Notably, the SD procedure continued throughout the 30 min off-period induction. The pattern of off-period induction was planned to start with longer induced off periods, and progressively move to shorter durations, to mimic the natural decline in off-period duration seen after SD in NREM sleep14. Indeed, the temporal protocol we used typically produced a rise and then subsequent decline in SWA, as is naturally seen in NREM sleep (Extended Data Fig. 2a). There were 30–45 s breaks in between bouts of off induction, to limit the duration of continuous light exposure and to mimic the wake-like brain activity seen during REM sleep or during brief arousals. The off-period induction pattern consisted of the following stages: (1) stage 1—pulses delivered at 1 Hz, with a laser-on time of 180 ms, for 10 min total; laser-off break—45 s; (2) stage 2—pulses delivered at 2 Hz, with a laser-on time of 140 ms, for 10 min total; laser-off break—45 s; (3) stage 3—pulses delivered at 3 Hz, with a laser-on time of 100 ms, for 5 min total; laser-off break—45 s; (4) SD ends—at this point the formal SD stopped, all objects were removed from the cage and the mouse was free to sleep at any time; and (5) stage 4—pulses delivered at 3 Hz, with a laser-on time of 80 ms, for 5 min total. After each 5-min bout of stage 4, the laser was turned off for a 30-s break, after which another bout began. The above-mentioned stage 4 continued until the mouse actually fell asleep. In short, the SD formally ended (that is, efforts to keep the mouse awake ended) 5 h after light onset, but off-period induction continued until the mouse actually began to sleep, at which point it was immediately and manually terminated. Thus, the actual duration of off-period induction varied slightly across mice, typically 30–45 min. The recording continued until light onset of the following day. For off-period induction in the synaptoneurosome experiments, the induction protocol was largely identical to that used in silicon probe electrophysiology experiments, and off periods were again only induced during the final 30 min of the 5 h SD. Specifically, it consisted of the following stages: (1) stage 1—pulses delivered at 1 Hz, with a laser-on time of 180 ms, for 10 min total; laser-off break—45 s; (2) stage 2—pulses delivered at 2 Hz, with a laser-on time of 140 ms, for 10 min total; laser-off break—45 s; (3) stage 3—pulses delivered at 3 Hz, with a laser-on time of 100 ms, for 5 min total; laser-off break—45 s; and (4) stage 4—pulses delivered at 3 Hz, with a laser-on time of 80 ms, for 5 min total; laser-off break—30 s. Critically, in these experiments, the SD did not end, and mice were not allowed to sleep. Immediately after stage 4, the mouse was manually disconnected from optic patch cables and EEG/EMG tethers (pulses continued at 3 Hz/80 ms as in stage 4 for this brief ~1 min disconnection period), briefly anesthetized and killed for synaptoneurosome collection. A 470 nm laser (OEM Laser Systems DPSSL Driver) was used for all off-induction experiments and any experiments involving either ACR or SOM+ mice. Power at the ferrule tip was measured using a photodiode power sensor (Thorlabs, S120C) before optrode construction and surgical implantation. Power at the ferrule tip varied across mice and was adjusted within experiments at the discretion of the experimenter to achieve effective off-period induction and/or tonic inhibition. These adjustments were limited to a small range, and the power in any optogenetic experiment (excluding the halorhodopsin experiments) never exceeded ~1.5 mW. For off-period induction in ACR mice, power at the ferrule tip was generally between 250 and 550 µW and was never greater than ~700 µW. For off-period induction in SOM+ mice, power at the ferrule tip was generally between 350 and 700 µW and was never greater than ~1 mW. For off-period induction in the synaptoneurosome experiments, slightly higher powers were used to maximize the diffusion of light and the area of affected tissue. Power at the ferrule tip was generally between 0.6 and 1.2 mW, and was never greater than 1.5 mW. For tonic inhibition experiments, the temporal pattern of light delivery was designed to match relatively closely the total MUA firing rate reduction observed in the off-induction experiments, but without inducing many synchronized on/off periods as in the off-period induction experiments. Therefore, beginning 30 min before the end of SD, rather than inducing off periods as above, we delivered light continuously for 20 s, and then turned the light off for 60 s. This pattern was repeated for ~30 min while the SD continued, as in the off-induction experiments. After 30 min, the SD formally ended and the mouse was free to sleep, and this tonic inhibition protocol continued until the moment the mouse first fell asleep, at which point the inhibition was immediately and manually terminated and the mouse was left undisturbed, as in the off induction experiments. For the experiments involving tonic inhibition using SOM+ mice, an analog voltage signal with a sinusoidal profile was used to drive the laser, at a frequency of either 0.5 Hz (one mouse) or 40 Hz (four mice). In all experiments, laser power at the ferrule tip was between ~200 and 700 µW, and the distance between any given consecutive peak and trough was always between ~100 and 200 µW. For the experiments involving tonic inhibition using halorhodopsin mice, a 561 nm laser was used (OEM Laser Systems DPSSL Driver); the power at the ferrule tip was never more than 12 mW and was typically in the range of 3–6 mW. The same experimental design and temporal pattern of light delivery were used in the SOM+ tonic inhibition experiments, with two differences. First, the laser was driven digitally according to a TTL pulse, so when the laser was on, it was on at a consistent power rather than varying with the sinusoidal profile used in the SOM+ tonic inhibition. Second, in half of these mice, we used a 3-h SD with inhibition during the final 30 min, rather than a 5-h SD. This was done because of the known effects of sleep/wake history on intracellular chloride79 and our own experience suggesting that, in some mice, it was easier to achieve effective inhibition at lower irradiance levels when sleep pressure was not as high as after 5 h of continuous SD. These parameters and variations are largely taken from and in line with other published literature using these tools71,79. For analyses of mouse behavior during off-period induction (Supplementary Fig. 2), both manual video scoring and analysis of EMG signals were performed. Briefly, for the analysis of mouse turning behavior, videos from the off-period induction epoch of a subset of mice were manually reviewed, and all right and left turns were marked. A turn was marked when the mouse made a continuous turn of >90 degrees in either direction. For EMG analysis of pausing behavior during SD and off-period induction, we used an amplitude-based approach with adaptive thresholding. Raw EMG signals were bandpass filtered (20–500 Hz, fourth-order Butterworth, zero phase) to isolate muscle activity and remove movement artifacts. The RMS envelope was computed using a 50 ms sliding window and normalized using robust z scoring. Detected pauses were initiated when the z-scored RMS dropped below −1.0 and terminated when it rose above 0. Candidate pauses were constrained to 0.5–3 s in duration. To ensure that detected pauses reflected brief stillness during otherwise active behavior, rather than prolonged quiescence, pauses were retained only if the mean z-scored RMS in the 5 s preceding or following the pause exceeded zero. For all optogenetic experiments, data collection and analysis were not performed blindly to the experimental conditions, because the identity of the optogenetic construct and the presence of laser stimulation or SD were necessary for the experimenter to correctly run the protocol. Quantification of synaptic GluA1 levels Synaptoneurosome preparation After the 30 min of off-period induction at the end of SD, mice were immediately deeply anesthetized with 1–3% isoflurane and decapitated. The left and right cortices were isolated, and a circular sample 3mm in diameter was extracted from both hemispheres, centered over the EEG electrode, to collect the majority of the affected tissue. Crude synaptosomes were prepared from fresh cortical tissues using Syn-PER synaptic protein extraction reagent (Thermo Fisher Scientific, 87793) supplemented with protease and phosphatase inhibitor cocktail (Halt Protease and Phosphatase Inhibitor Cocktail, EDTA-Free (100X), 78441) following the manufacturer’s instructions. Briefly, the samples were homogenized using a 2 ml Dounce tissue grinder with ten up-and-down even strokes. The homogenate was first centrifuged at 1,200g for 10 min to eliminate cell debris, followed by a second centrifugation at 15,000g for 20 min. The pellets, containing synaptosomes, were gently resuspended in the respective amount of Syn-PER synaptic protein extraction reagent. After extraction, the samples were stored at −80 °C until used for western blot analyses. Quantitative immunoblotting For quantification of GluA1 and pGluA1 levels in the synaptoneurosome, we followed a protocol previously described in ref. 9. Protein concentration was measured using a Qubit 2.0 Fluorometer (Invitrogen, Life Technologies, Q32866). Aliquots were thawed on ice and denatured in a buffer containing Coomassie G250 and phenol red (Invitrogen NuPAGE) at 70 °C for 10 min. Then, protein samples (10 μg) were loaded and separated electrophoretically using a 1.5-mm thick 4–12% Bis–Tris gel together with NuPAGE MOPS SDS Running Buffer (both by Invitrogen NuPAGE). The gels were then electrotransferred onto PVDF membranes (Invitrolon, Novex, LC2005) through wet transfer. Membranes were blocked using 5% (wt/vol) not-fat dry milk in 0.1% (vol/vol) Tween 20 in PBS (PBS-T) for at least 40 min at room temperature, and then incubated overnight at 4 °C with the different primary antibodies (anti-GluA1 ser845—PhosphoSolutions, p1160-845; anti-GluA1—Santa Cruz, sc-55509 and anti-GAPDH—Santa Cruz, sc-32233). Bound antibodies were detected using horseradish peroxidase-conjugated antirabbit and antimouse secondary antibodies (mouse antirabbit IgG-HRP—Santa Cruz, sc-2357 and m-IgGκ BP-HRP—Santa Cruz, sc-516102), and the uniformity of sample loading was confirmed through Ponceau S staining and immunodetection of GAPDH, used as a loading control. Typhoon 9410 (Amersham Biosciences) was used to obtain digital images through chemiluminescence reaction (ECL reagents). A semiquantitative measurement of the band intensity was conducted using ImageJ software (https://imagej.net/ij/) and expressed as a ratio of band intensity with respect to the loading control. Histology and probe localization After the final experimental session, all mice with bilateral silicon probe implants were anesthetized with isoflurane (2–3%) and then intracardially perfused with PBS followed by 4% paraformaldehyde in PBS. Brains were extracted, postfixed overnight, cryoprotected by exposure to increasing concentration of sucrose in PBS at 4 °C and then sliced in coronal sections (thickness = 40–50 μm) using a cryostat (Thermo Fisher Scientific, CryoStarTM NX50). After mounting onto slides, sections were left to dry overnight, mounted and coverslipped with standard mounting medium for fluorescence microscopy (SouthernBiotech, Fluoromount-G), and imaged with an upright epifluorescence microscope (Leica, DM2500). Using the acquired images, probes were identified, labeled and registered to the Allen Mouse Brain Common Coordinate Framework80 using the openly available histological software ‘Histological E-data Registration in rodent Brain Spaces’81. This allowed us to obtain a precise position for every channel in the common coordinate framework space. Slight adjustments or alignments between probes were further made using electrophysiological features such as current source densities computed on averaged potentials time locked to on-period initiations82,83. Immunohistochemistry Free-floating coronal brain sections (40–50 µm) from perfused mice were processed for immunofluorescence using standard procedures. Briefly, sections were washed thrice for 10 min in 1× PBS to remove residual fixative, then incubated in blocking buffer (1× PBS, 0.3% Triton X-100, 5% normal donkey serum) for 1 h at room temperature. Primary antibodies were diluted in the same blocking buffer and incubated with sections overnight at 4 °C with gentle rocking. The following primary antibodies were used: for SOM+ mice, chicken anti-GFP (Aves Labs, GFP-1020; 1:4,000) and rat antisomatostatin (MilliporeSigma, MAB354; 1:100); for ACR mice, mouse anti-CaMKIIα (Abcam, ab22609; 1:1,000) and chicken anti-mKate2 (Aves Labs, mKate2-0020; 1:1,000); and for halorhodopsin mice, mouse anti-CaMKIIα (Abcam, ab22609; 1:1,000). Sections were then washed thrice for 10 min in PBS-T (0.3% Triton) before incubation with the appropriate species-matched donkey secondary antibodies (Alexa Fluor 488 or 594; Thermo Fisher Scientific), diluted 1:500–1:1,000 in PBS-T with 2% donkey serum for 2 h at room temperature in the dark. Sections were washed twice for 10 min in PBS after secondary incubation, and nuclei were stained by incubating sections in 0.5 µg ml−1 Hoechst 33342 (Thermo Fisher Scientific, R37605) in PBS for 20 min, followed by two final PBS washes for 10 min. Sections were mounted on slides using ProLong Glass antifade mountant (Thermo Fisher Scientific, P36982) and coverslipped for imaging using 1.5 coverslips. Images were acquired using a confocal microscope (Nikon AXR) at the University of Wisconsin Optical Imaging Core, with appropriate excitation/emission settings for Alexa Fluor 488, Alexa Fluor 594 and Hoechst, and processed using ImageJ/FIJI. FTR task An FTR task previously characterized in ref. 15 was used to assess memory consolidation. Performance in this task mainly depends on sensorimotor cortex without a major involvement of the hippocampus15. The arena consisted of an acrylic chamber measuring 53 cm in length by 32 cm in width, with black walls 35 cm high. A USB camera mounted above the arena was used to record the mouse’s position. During both testing and acquisition, an identical and simple cylindrical object (a small glass bottle filled with colored stones) was placed on both sides of the arena. All floors used were constituted of clear polycarbonate plastic, which sat on top of a white acrylic background. The floor was split into two halves along the long axis, and as such each half of the floor (size 32 cm × 26.5 cm) consisted of a clear polycarbonate sheet (~6 mm thick) with either a smooth finish or a pebbled finish (both materials from ePlastics). Experiments were counterbalanced so that half of participants were exposed to two smooth floors during the acquisition and then the pebbled floor served as the new texture in the test, and half of the participants were exposed to two pebbled floors during the acquisition and the smooth floor served as the new texture during the test. All SOM+ mice used in the FTR task were habituated to handling for 7 days before the acquisition phase of the test. Habituation consisted of placing a gloved hand in the mouse’s home cage for 10–15 min per day, during which time the mouse was allowed to freely explore, as has been shown to be critical in other novelty-recognition tasks of memory consolidation84. Mice were monitored by video in their home cage to ensure that sleep patterns were regular, and to ensure that mice in the sleep condition met a minimum criterion of 30% time spent asleep (mean = 45.5%, s.d. = 14.1%) during the 1 h when other mice were being sleep deprived. To accurately estimate total sleep time based on video alone, we created a pose-tracking model using DeepLabCut (DLC36), using a large number of frames across a wide variety of different cameras and mice, all from an overhead view of the mouse’s home cage. The model was trained to predict the position of a number of ‘nodes’ on the mouse visible from overhead, including the nose, head, ears, center, tail, shoulders and hips. After training, the model accurately labeled the positions of these nodes in an overhead shot of the mouse. We used this model to predict each mouse’s position in every frame (after downsampling to 1 frame per second) of a video that started the day before acquisition and continued until testing was completed. We then used these predictions to compute a form of actigraphy (‘DLC actigraphy’; Supplementary Fig. 5) that could accurately estimate a mouse’s total sleep time. After getting the X–Y prediction across all frames for all nodes, we computed for each node the movement vector across all frame changes (that is, where ΔX and ΔY represent the change in the X and Y position of a node from one frame to the next, the movement vector is equal to (ΔX2 + ΔY2)). The movement was then averaged across all nodes to get the average movement of the mouse, and that averaged movement was robustly z scored. A threshold was set at (−mean/s.d.) of the z-scored movement, and any times falling below that value were labeled as sleep, while times above that value were labeled as wake. We validated the ability of the DLC-actigraphy method to predict total sleep time using overhead videos of mice that also had synchronized EEG or LFP and EMG recordings, and that had manually labeled vigilance states, which served as ground truth. Across 21 recordings from 11 individual participants with good overhead recordings, which constituted ~395 h of total recording time, this method predicted total sleep time for most recordings within 5% manually labeled hypnogram, with an average error (compared to the manual hypnogram) of 0.55% (Supplementary Fig. 5d,e). We used a similar but separate DLC model to track the position of the mice in the arena, which produced highly accurate and reliable position predictions, likely due at least in part to the constrained environment and low variance of frames to be labeled. All labeling of each mouse’s position and quantification of arena position and novelty ratios associated with the FTR task were based on the unmodified predictions of this model. On the day of acquisition, at the lights-on time, mice were placed into the arena with two identical floor textures and left to freely explore for 15 min. To reduce behavioral variability during the acquisition phase, only mice that spent at least 1/3 of the total acquisition time (that is, 5 min) on both sides of the arena were used for the full experiment. After the acquisition phase, these mice were removed from the arena, placed back into their home cage and either allowed to sleep, sleep deprived or sleep deprived concurrently with off-period induction for 1 h. For the induction, to mimic real NREM sleep as closely as possible, the temporal sequence of off periods used to drive the laser matched exactly (that is, with identical timings and durations) the one previously recorded during the first hour of recovery NREM sleep after SD on a control probe of a mouse with a very high-quality implant in M2. Laser power at the ferrule tip in these experiments was similar to that used in the synaptoneurosome experiments. Next, 24 h after the start of the acquisition phase, mice were placed back into the arena again for 15 min. Using the model predictions of the mouse’s nose position, the novelty ratio during the 5-min testing phase (first 5 min of exploration during the test) was quantified and expressed as the ratio of time on the side of the arena with the new floor texture compared to the total time in the arena during the testing phase (5 min). Statistical analysis The results of every statistical test, linked to the corresponding figure and panel, are reported in Supplementary Table 1. The table reports the type of test run, the test statistic, the P value, the effect size method and value, and n for all statistical tests performed. All statistics were implemented in Python, largely using the SciPy and Pingouin statistical libraries85,86. Statistical source data and code to reproduce statistical tests and plots are available in the GitHub repository (https://github.com/CSC-UW/off_PERIOD_INDUCTION_MATERIALS). No participants or data were selectively excluded from any analyses, except for a single ACR mouse that was included in bandpower analyses due to the presence of high-quality LFPs but was excluded from other electrophysiological analyses that relied on off-period detection or spike-time synchrony. Excluding this mouse entirely did not affect the results of any relevant analysis. Animals were allocated to groups based on genotype and availability/age at the time of surgery. Where possible (for example, assignment to sleep/SD in behavioral experiments), animals were randomly assigned to a condition. No statistics in this paper are derived from technical replicates alone or from fewer than three biological replicates. All significance testing used in this paper was two tailed with α = 0.05. For all significance tests where multiple comparisons were not necessary, normality was first assessed using a Shapiro–Wilk test, and the appropriate parametric or nonparametric tests were performed. Specifically, when data did not violate parametric assumptions per the Shapiro–Wilk test (P > 0.05), a paired or unpaired t-test was performed as appropriate; otherwise, the corresponding nonparametric test was used (Wilcoxon signed-rank test or Mann–Whitney U test, respectively). If the same null hypothesis was evaluated across, for example, channels or frequency bins, P values were adjusted with the Benjamini–Hochberg false-discovery-rate procedure. Sample sizes were not predetermined using statistical methods, but are in line with previously published similar work14,15,23,79 and sample sizes used across all experiments are reported in Supplementary Table 1 and in the figure captions. In the case of multiple group comparisons, normality was assessed for all components using Shapiro–Wilk, and equal variance was assessed using Levene’s test of homoscedasticity. If normality and homoscedasticity assumptions were met, one-way or two-way analysis of variance (ANOVA) tests were performed as appropriate. Significant omnibus effects were followed with pairwise post hoc contrasts using Tukey’s honestly significant difference. For statistical testing of spike-time synchrony at the level of individual spike-sorted unit pairs, we fit a linear mixed-effects model with a dyadic multimembership random-effects structure. The outcome variable was the Fisher z-transformed difference in STTC between rebound sleep and late baseline sleep for each unit pair. The model included a fixed effect for probe identity (optrode versus contralateral control, coded as a binary indicator), random intercepts and random slopes for probe identity by mouse, and a multimembership random intercept for unit identity (to account for the nonindependence of unit pairs sharing a common unit). Models were fit using restricted maximum likelihood. Degrees of freedom and P values for the fixed effect of probe identity were computed using the Kenward–Roger approximation (pbkrtest package), which provides improved small-sample correction for both the denominator degrees of freedom and the fixed-effect covariance matrix. All tests were two tailed with α = 0.05. Analyses were performed in R (version 4.5.2) using lme4 (version 1.1-38), lmerTest (version 3.1-3), pbkrtest and lmerMultiMember (version 0.11.8). Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41593-026-02318-9. Supplementary information Supplementary InformationSupplementary Figs. 1–6. Reporting Summary Supplementary Video 1The video shows 10 s of raw data (LFP and MUA) as animal is awake during SD (left), in NREM sleep (middle), and is awake with concurrent off-period induction (right). Please note the dissociation between the two probes during off-period induction, and that during off-period induction, the animal remains awake and interacts with the environment. Supplementary Table 1Complete statistical information for all statistical tests run in this study, with corresponding figure and panel labels. Reported for each test is the number of mice, the test type, the test statistic, the P value and the effect size method and value. All further materials can be found in the publicly available source data. Supplementary Figs. 1–6. Reporting Summary The video shows 10 s of raw data (LFP and MUA) as animal is awake during SD (left), in NREM sleep (middle), and is awake with concurrent off-period induction (right). Please note the dissociation between the two probes during off-period induction, and that during off-period induction, the animal remains awake and interacts with the environment. Complete statistical information for all statistical tests run in this study, with corresponding figure and panel labels. Reported for each test is the number of mice, the test type, the test statistic, the P value and the effect size method and value. All further materials can be found in the publicly available source data.