Mice

Mice were maintained on a 12 h:12 h light:dark cycle with ad libitum access to food and water, at 22 ± 2 °C and 55 ± 10% humidity. All experiments were performed in accordance with the Swiss National Institutional Guidelines on Animal Experimentation and were approved by the cantonal Veterinary Office Committees for Animal Experimentation. Male C57BL6/J mice were used for whole-brain FOS mapping during circadian and sleep-deprivation time points, and both males and females were used for all other experiments. Age-matched mice (2–6 months) of the appropriate genotype were randomly distributed into experimental groups. Knock-in and transgenic lines were obtained from the Jackson Laboratory, including the TRAP2 knock-in line (Fos-2A-iCreERT2, strain 030323)⁴⁰, Ai14 Cre reporter (RCL-tdT, strain 007914), Sert-Cre (Slc6a4^tm1(cre)Xz, strain 014554)⁷⁷, Vgat-IRES-Cre (Slc32a1^tm2(cre)Lowl, strain 028862)⁷⁸ and Vglut2–IRES-Cre (Slc17a6^tm2(cre)Lowl, strain 016963)78.

Sleep deprivation and sleep attempts

Nesting material was removed from the cage at the beginning of sleep deprivation, and replaced at the end of the deprivation period. Mice were housed individually for all deprivation experiments. For grooming-based sleep deprivation, mice were finely misted with Milli-Q water to elicit grooming behaviour every 15 min for up to 6 h. Sleep attempts were interrupted by opening the cage door or applying an additional water mist. For novel-object-based sleep deprivation, a new object was placed in the cage every 15 min to maintain behavioural engagement for up to 6 h. Sleep attempts were interrupted by opening the cage door and displacing the object. Sleep attempts were defined as the adoption of a sleep-like posture and the concurrent emergence of NREM-like slow-wave activity in live EEG recordings, which reverted to wake-like EEG after cage-door opening, water misting or novel object displacement, and could be detected post hoc as sleep bouts by automated state classification (see Extended Data Fig. 7 and Methods sections below).

Whole-brain immunostaining and imaging

For all mapping experiments, mice were housed individually and allowed to habituate to the home cage for several days. For circadian mapping, whole brains were collected every 3 h during the unperturbed light–dark cycle. For deprivation mapping, whole brains were collected 1.5 h, 3 h, 5 h and 6 h into sleep deprivation, as well as 1.5 h and 3 h into the recovery period after 6 h deprivation. Six to seven brains per time point were collected for the circadian experiment and light-phase sleep deprivation, and four brains per time point were collected for dark-phase sleep deprivation. All samples were immunostained and cleared according to the iDisco⁷⁹ workflow (https://idisco.info/) and imaged in horizontal orientation on a MesoSPIM system (Center for Microscopy and Image Analysis, University of Zurich) with corresponding control software (https://github.com/mesoSPIM/mesoSPIM-control) and a Hamamatsu Orca camera, using the following settings: 1.25× zoom, z_stepsize 5.0 μm, pixel size 5.26 μm, 2,048 × 2,048 resolution. Immunostaining was performed using a polyclonal rabbit anti-FOS antibody (Synaptic Systems, 226 003) at 1:2,000 and donkey anti-rabbit Alexa Fluor 647 (Invitrogen, A31573) at 1:800. Autofluorescence from the 561-nm channel was collected for overall brain morphology.

Whole-brain FOS mapping

Whole-brain light-sheet data were registered to an average mouse-brain template using the SHIELD-based workflow²⁴. For each brain sample, striping artefacts were removed from the raw autofluorescence channel before resizing to the dimensions of a 461 × 471 × 323-pixel light-sheet-optimized average brain template²⁵ (https://github.com/Gubra-ApS/LSFM-mouse-brain-atlas) and performing automated non-rigid alignment with sitk-align. Alignment results were inspected and manually corrected by reiteratively adjusting data:atlas correspondences and re-warping images in nuggt (Neuroglancer ground truth; https://github.com/chunglabmit/nuggt, master) using the nuggt-align function²⁴. In parallel, FOS⁺ nuclei were automatically detected in three-dimensional (3D) stacks using a Laplacian of Gaussian filter (skimage.feature.blob_log, scikit-image v.0.19.3, https://scikit-image.org/). Identical alignment parameters were then applied to FOS⁺ nuclei from the same sample. To create heat maps of FOS⁺ cell density for each brain sample, Gaussian blur (sigma = 2.25) was applied to aligned FOS nuclei and the signal was averaged between the left and right brain hemispheres, before performing batch correction based on the ComBat approach⁸⁰. To evaluate distinct patterns of brain activation across experimental conditions, averages for mice within the same experimental time point were normalized across time points within each experiment by dividing by the total activation per voxel. We then segregated each experimental dataset by peak activation time point, resulting in four categories per experimental condition: late wake (zeitgeber time (ZT)15–ZT21 circadian, sleep deprivation (SD) time points SD5h and SD6h for deprivation), sleep (ZT03–ZT09 circadian, recovery (R) time points R1.5h–R3h for deprivation), early wake (ZT12 circadian, SD1.5h–SD3h deprivation) and transition (ZT00 circadian, ZT12 dark deprivation). Finally, hierarchical clustering (clustering depth of 20) was performed for each category, and each cluster was manually categorized as type 1–type 3 on the basis of its mean response pattern during sleep deprivation and recovery. Whole-brain datasets can be browsed at: https://sleep-wake-atlas.scicore.unibas.ch/, and code for all analyses is available at https://gitlab.com/ceda-unibas/sleep-brain-atlas. For Pearson correlation analyses using unperturbed circadian data (Extended Data Fig. 3), voxel intensity correlations with wake duration were calculated for a 2-h time window preceding each experimental time point.

EEG analyses

For EEG implantation, electrodes were stereotactically placed over the right cerebral hemisphere at anteroposterior (AP) −2.25; lateral (L) +1.7 and AP +1.5, L +1.2 relative to bregma. Custom EEG implants were fixed to the skull using superglue and orthodontic resin (Paladur, Kulzer). All mice were allowed to recover from surgery for at least one week in their home cages. Mice were then connected to a flexible recording cable with a commutator and allowed to habituate to custom-made behaviour cages for at least 3 days, before EEG and video tracking were collected continuously for the duration of each experiment. Data were acquired using a 16-channel AC amplifier (A-M Systems, model 3500), filtered (0.3–300 Hz) with a gain of 500, digitized at 200 Hz and recorded with Spike2 (v.9.09a). The EEG signal was downsampled to 100 Hz and partitioned into 2-s epochs. Behavioural states were classified as high-theta wake, low-theta wake, NREM sleep and REM sleep using custom software (offline electroencephalography state space analysis (OESSA); https://github.com/VBits/oessa). EEG recording was used to quantify sleep behaviour in all cases. For the sleep-deprivation FOS-mapping experiments in Fig. 1, only a subset of representative mice was recorded given continuous wake maintenance by manual sleep deprivation. EEG power spectra in the 0–50 Hz range were calculated using a fast Fourier transform at 0.25-Hz resolution. To account for inter-individual variability in EEG signal, spectral data for each mouse were normalized to the total power for each vigilance state and shown as a percentage. Delta power (0.25–4 Hz) time-course data were normalized to mean baseline values from the last four hours of the light phase, when delta power is lowest^2,81. To evaluate the relationship between wake duration and NREM delta power, a three-term filter for wake-enriched episodes⁴² was applied to EEG data from a 5-day undisturbed baseline period: NREM ≥ 10 min preceding wake ≥ 15 min, immediately followed by NREM ≥ 15 min. NREM segments allowed brief wake or REM intrusions of 1 min or less, and wake-enriched segments allowed brief NREM intrusions of 3 min or less.

TRAP induction

4-Hydroxytamoxifen (4-OHT; Sigma H6278) was dissolved in ethanol at 20 mg ml⁻¹ and combined with two volumes of a 4:1 blend of sunflower seed oil:castor oil (Sigma, S5007 and 259853). After mixing at room temperature for 2–3 h, ethanol was evaporated by vacuum centrifugation and the final 10 mg ml⁻¹ solution was injected intraperitoneally (i.p.) at a dose of 50 mg kg⁻¹. Mice were habituated to handling and i.p. injection for 4–5 days before TRAP induction. For deprivation-TRAP experiments, 4-OHT was administered 5 h into a 6-h sleep-deprivation period starting at the beginning of the light phase. For recovery-TRAP, 4-OHT was administered 3 h after the end of the 6-h sleep-deprivation period. For rested control-TRAP, 4-OHT was administered to undisturbed mice 5 h into the light phase. Deprivation-TRAP cells in the MR were labelled during induced grooming-based sleep deprivation, which yielded similar results to novel-object-based deprivation-TRAP (see Extended Data Fig. 5). aMPO deprivation-TRAP cells were labelled using novel-object-based sleep deprivation.

Chemogenetics

Under surgical anaesthesia, AAV5-hSyn-DIO-hM3Dq-mCherry (Addgene 50474), AAV5-hSyn-DIO-hM4Di-mCherry (Addgene 50474), AAV9-hSyn-DIO-hM4Di-mCherry or AAV5-Syn-FLEX-PSAM4-GlyR-IRES-EGFP (Addgene 119741) were stereotactically injected into the MR, aMPO, LPO or LHA with a Nanoject III (Drummond Scientific). Coordinates relative to bregma and injection volumes were as follows: MR: AP −2.28; dorsoventral (DV) −4.90, angled 22° anteriorly relative to the horizontal plane, 100 nl; aMPO: AP +0.80; L ±0.28; DV −5.10, 45 nl bilateral; LPO: AP 0.00; L ±0.70; DV −5.20, 100 nl bilateral; LHA: AP −1.60; L ±1.03; DV −4.95, 100 nl bilateral. For all sleep classification experiments, custom EEG implants were attached immediately following virus injection. For TRAP experiments, 4-OHT was administered 2 weeks after virus infusion, and chemogenetics experiments started 3–4 weeks later. All other chemogenetics experiments commenced 3–4 weeks after virus injection. Mice were habituated to handling and i.p. injection for 3 days before each experiment. The same mice were then injected on consecutive days with either control phosphate-buffered saline (PBS) or 1 mg kg⁻¹ CNO (Sigma, SML2304), a dosage chosen on the basis of the absence of CNO-induced phenotypes in control mice. For experiments mapping MR activation, AAV5-hSyn-DIO-hM3Dq was injected into the MR in TRAP2 mice. Three weeks after deprivation-TRAP induction, CNO was administered at the beginning of the dark phase and mice were euthanized 1.5 h later for iDisco-based FOS analyses. For MR-projection-based manipulations, AAVretro-EF1a-FlpO (Addgene 55637) was injected into the LPO or dorsomedial hypothalamus. After 1–2 weeks, AAV8-EF1a-Con/Fon-hM3Dq-mCherry (Gene Vector and Virus Core, Stanford University) was injected into the MR, and PBS or CNO was administered 3–4 weeks later. Viral targeting was confirmed post-mortem, and mice with off-target expression were excluded from the experiment.

Anterograde tracing

AAV9-hSyn-FLEX-mGFP-2A-Syp-mRuby (Viral Core Facility, Charité–Universitätsmedezin Berlin) was injected into the MR in TRAP2 mice 2 weeks before deprivation-TRAP induction. Brains were collected for sectioning and immunostaining 3–4 weeks after TRAP induction. Synaptophysin-mRuby puncta were registered to Allen Mouse Brain CCFv3 using the QUINT workflow⁸².

Chronic manipulations and cell ablation

For chronic inhibition of deprivation-TRAP cells, AAVdj-CMV-DIO-Kir2.1-2A-ZsGreen or AAVdj-CMV-DIO-EGFP (Gene Vector and Virus Core, Stanford University) was injected into the MR 1–2 weeks before TRAP induction as described above. Sleep behaviour was analysed 3–4 weeks after TRAP induction. For TRAP cell ablation, a mix of AAV9-EF1a-FLEX-DTA (University of Zurich Viral Vector Facility) and AAV5-EF1a-DIO-YFP (Addgene 27056) was injected into the MR 1–2 weeks before TRAP induction. In cell-type-specific ablation experiments, a mix of AAV5-FLEX-taCasp3-TEVp (Addgene 45580) and AAV5-EF1a-DIO-YFP was injected into the MR of Sert-Cre, Vgat-IRES-Cre and Vglut2-IRES-Cre mice. For all ablation experiments, control mice were injected with AAV5-EF1a-DIO-YFP only. Sleep behaviour was analysed 3–4 weeks after TRAP induction or virus infusion, and cell ablation was confirmed post-mortem.

Immunohistochemistry

Mice were perfused with chilled PBS followed by chilled 4% paraformaldehyde (PFA). Brains were post-fixed overnight at 4 °C before cryoprotection in 30% sucrose. Cryosections (40 μm) were washed in PBS, incubated in blocking solution (10% normal donkey serum, 0.2% Triton X-100 in PBS) for 30–60 min at room temperature and incubated for 2–3 days in primary antibody. Sections were then washed at room temperature with 0.2% Triton X-100 in PBS and incubated for 1–3 days in secondary antibody. After washing in 0.2% Triton X-100 in PBS, sections were stained with DAPI and mounted on slides with Fluoromount-G (Invitrogen). Images were acquired with a Zeiss AxioScan.Z1 or a Zeiss LSM800 confocal microscope, using respective AxioScan and ZEN Blue software. Primary antibodies: rabbit anti-FOS (polyclonal, Synaptic Systems 226 003, 1:1,000-1:4,000), rabbit anti-FOS (monoclonal, Synaptic Systems 226 008, 1:2,000), rat anti-FOS (Synaptic Systems 226 017, 1:2,000), chicken anti-GFP (Invitrogen A10262, 1:1,000), rabbit anti-ZsGreen (Takara 632474, 1:1,000), rat anti-RFP (Chromotek 5f8-150, 1:1,000), goat anti-tdTomato (Origene AB8181, 1:1,000), goat anti-5HT (Immunostar 20079, 1:2,000), guinea pig anti-orexin A (Synaptic Systems 389 004, 1:500), rabbit anti-MCH (Phoenix, H-070-47, 1:500) and mouse anti-nNOS (Sigma N2280, 1:500). Primary antibodies were matched with appropriate donkey or goat secondary antibodies conjugated to Alexa 488, Alexa 546, Alexa 555, Cy3, Alexa 647 or Cy5, and used at 1:750-1:1,000 (Thermo Fisher Scientific and Jackson ImmunoResearch).

In situ hybridization

Multiplex HCR (hybridization chain reaction) probe sets for Slc32a1 (Vgat), Slc7a6 (Sert) and Gal and corresponding amplifiers were used according to the manufacturer’s instructions (Molecular Instruments) with modifications. In brief, 40-μm cryosections were collected as described above, incubated in 70% ethanol for 2 h at room temperature and rinsed in 5× SSCT before the pre-hybridization step. After overnight probe hybridization, sections were washed in a 100%/75%/50%/25%/0% probe wash buffer/5×SSCT series at 37 °C, before proceeding to amplification, washes and mounting. For HCR in situ and immunohistochemistry in the same sections, HCR was completed first and sections were fixed in 4% PFA for 10 min at room temperature before proceeding to immunohistochemistry.

Electrophysiology

All recordings and analyses were performed blinded to experimental groups. To label GABAergic or serotonergic cells, AAV9-CAG-DIO-tdTomato or AAV5-hSyn-DIO-tdTomato was injected into the MR in 2–4-month old Vgat-Cre or Sert-Cre mice, respectively. Approximately 2 weeks later, mice were euthanized 6 hours after the beginning of the light phase, immediately after a 6-h sleep deprivation or after being left undisturbed. Coronal slices (200 μm) containing the MR were cut in carbogenated (95% O₂/5% CO₂) ice-cold cutting solution (sucrose 205 mM, glucose 10 mM, NaHCO₃ 25 mM, KCl 2.5 mM, NaH₂PO₄ 1.25 mM, MgCl₂ 7.5 mM and CaCl₂ 0.5 mM) with a Leica VT1200S vibratome (horizontal oscillation amplitude 1.80 mm). Slices were kept at 33.0 ± 1 °C in carbogenated artificial cerebrospinal fluid (aCSF: NaCl 125 mM, KCl 2.5 mM, NaH₂PO₄ 1.2 mM, NaHCO₃ 24 mM, sodium ascorbate 5 mM, glucose 12.5 mM, MgCl₂ 1 mM and CaCl₂ 2 mM, pH 7.4) for 30 min and then kept at room temperature for around 1 h before starting recordings.

During recording sessions, slices were held at 33.0 ± 1 °C in a custom chamber with carbogenated aCSF circulation. Whole-cell patch clamp recordings of tdTomato⁺ MR GABA or serotonin cells were performed in current clamp mode using pClamp11 software with a Multiclamp 700B amplifier (Molecular Devices). Cells were visualized using an upright microscope equipped with gradient contrast infrared visualization (Luigs and Neumann) and a 60× objective. For all experiments, data were digitized by a Digidata 1440a (Molecular Devices) at 10 kHz and filtered at 1 kHz. Patch pipettes (4–8 MΩ) were pulled with a Sutter Instruments P-1000 micropipette puller and filled with the following intracellular solution: 142 mM potassium gluconate, 10 mM HEPES, 1 mM EGTA, 2.5 mM MgCl₂, 4 mM Mg-ATP, 0.3 mM Na-GTP and 10 mM Na-phosphocreatine, with 0.2% biocytin. Passive membrane properties and action potentials were quantified using Clampfit 10 (Molecular Devices).

Sections were fixed immediately after recordings, and recorded cell identity was confirmed post hoc using streptavidin–Alexa488 or streptavidin–Alexa647 (Thermo Fisher Scientific), together with tdTomato immunostaining as described above.

Open field test, elevated plus maze and wood block engagement

For all behavioural assays, AAVdj-CMV-DIO-Kir2.1-2A-ZsGreen or AAVdj-CMV-DIO-EGFP (control) was injected into the MR in Vgat-Cre;Sert-Cre double transgenic mice 3 weeks before experiments. EEG recordings were analysed to confirm sleep loss phenotypes before testing. Both open field and plus maze tests were done at the beginning of the dark phase, when mice naturally spend more time awake. For the open field test, mice were placed in the centre of a 50 × 50-cm square box with 30-cm walls, and video was recorded from above for 10 min at 25 fps. For the plus maze assay, mice were placed at the centre of a maze with 35 cm × 6-cm arms, 74 cm above the ground, and video was recorded for 5 min. Locomotion, time in the open field centre and time in the closed arms were quantified using open source OptiMouse software (https://github.com/yorambenshaul/optimouse)⁸³.

To assess wood block engagement, 1 × 1 × 5-cm wood blocks (Labodia 213-1011) were placed in the home cage for 18 days, and then 3D scanned using a tripod-mounted Shining Einscan Pro HD and turntable. Total volume was quantified from 3D reconstructions.

Contextual fear conditioning

Fear conditioning acquisition and recall sessions were performed at the beginning of the dark phase, when mice naturally spend more time awake. During acquisition sessions, mice were placed in the middle of a 25 × 25-cm square box with a 2% acetic acid odour source and an electrifiable grid floor (Fear Conditioning System, Ugo Basile, with EthoVision XT software v.14-17), and allowed to explore freely. After 3 min, five foot shocks (0.8 mA, 1 s) were delivered at 30-s intervals. Recall was assessed at recent (1 day) and remote (14 days) time points after the acquisition session. For recall sessions, mice were placed in the same training context, and allowed to explore freely for 5 min without any foot shocks. During all sessions, mouse behaviour was recorded with an overhead infrared camera (Basler acA1300-60gm, Basler GenICam). Memory recall was assessed by quantifying the fraction of time spent freezing in the shocked context. Freezing behaviour was automatically classified by the EthoVision XT software, and defined as bouts of complete immobility (apart from breathing) lasting at least 2 s (pixel change < 1–3% adjusted for each mouse), as previously published⁸⁴.

Statistics and reproducibility

Details regarding the number of mice and statistical tests for each experiment are provided in the figure legends. All statistical tests are two-sided. Sample sizes were chosen on the basis of previous experiments and published studies, and all experiments included at least three replicates. Statistical analyses were performed in Python or GraphPad Prism v.10.

Reporting summary

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