A role for the cortex in sleep-wake regulation

Cortical and subcortical circuitry are thought to play distinct roles in the generation of sleep oscillations and global state control, respectively. Here we silenced a subset of neocortical layer 5 pyramidal and archicortical dentate gyrus granule cells in male mice by ablating SNAP25. This markedly increased wakefulness and reduced rebound of EEG slow-wave activity after sleep deprivation, suggesting a role for the cortex in both vigilance state control and sleep homeostasis.

The duration, timing and architecture of sleep are strictly regulated. Early studies based on neurological case reports, transections and electrical stimulation suggested that global state transitions are mediated via a distributed circuitry across the brainstem, the hypothalamus and the basal forebrain. More recent studies using selective targeting of specific neuronal populations based on their gene expression or connectivity patterns, highlighted that the sleep-wake promoting circuitry is highly complex, with distinct subcortical brain regions and neuronal subtypes responsible for specific aspects of wakefulness and sleep 1,2 . Although sleep-wake states are defined by the occurrence of neocortical and hippocampal oscillations, the possibility that neo-and archicortical neurons control vigilance states has been overlooked.
Cortical oscillations and neuronal firing patterns mirror sleep homeostasis 3,4 . Sleep homeostasis refers to the adjustment of the duration and intensity of sleep, to the duration of preceding wakefulness 4 . Electroencephalogram (EEG) slow-wave activity (SWA, EEG spectral power between 0.5-4 Hz) during NREM sleep represents a reliable marker of sleep-wake history 4 and has been proposed to underlie many functions of sleep, such as cellular maintenance and synaptic plasticity. SWA can be regulated in a local, use-dependent manner 5,6 , in line with the view that sleep emerges within cortical networks driven by the local accumulation of metabolic products, such as adenosine 7 . However, slow waves also occur under anaesthesia, in isolated cortical slabs or even ex vivo 8 . Therefore, the capacity to produce slow waves does not automatically imply a causative role for the cortex in physiological sleep or sleep homeostasis, either on a local or global level.
Here we test whether cortical structures have a function in regulating global sleep-wake dynamics. We focused on pyramidal neurons within layer 5 of the neocortex as well as archicortical dentate gyrus granule cells, two cell types involved in the generation of sleep oscillations. Layer 5 pyramidal neurons have been shown essential for the initiation and propagation of neocortical slow waves [9][10][11] , the main type of NREM sleep oscillations. Dentate gyrus granule cells contribute to the generation of hippocampal theta rhythm 12 , the defining oscillatory activity of REM sleep in rodents 13 . Yet neither of these two neo-and archicortical cell populations has so far been implicated in sleep-wake control.
Laminar local field potential (LFP) and multiunit activity (MUA) recordings were performed from the primary motor cortex of male adult wild type (C57BL/6) mice, concomitantly with EEG and electromyography (EMG) monitoring during 24 h undisturbed conditions ( Fig.  1a-e). Consistent with the idea of an active role for layer 5 in generating slow waves 9,10,14 , we found that neurons in layer 5 tended to initiate spiking upon the onset of population ON periods (Fig. 1e, Extended Data Fig. 1). A leading role for layer 5 was indicated by a stronger initial surge of neuronal firing at OFF-ON transitions (Extended Data Fig. 1a), and a shorter latency to the first spike during ON periods, even when the total number of spikes in layer 5 was matched with layer 2/3 (Extended Data Fig. 1b).
To induce a cortex-wide reduction in the output from layer 5 pyramidal neurons, we used a transgenic mouse line, in which a subpopulation (~15-30 %) of pyramidal cells in layer 5 of the neocortex lack the key t-SNARE protein SNAP25 (Rbp4-Cre;Ai14;Snap25 fl/fl ) 15 . Rbp4-Cre is known as a pan layer 5 driver line (Fig. 1f, Supplementary Fig. 1) 16 , but also presents a strong Cre-expression in dentate gyrus granule cells (Extended Data Fig. 2). By contrast, Cre-expression outside of cortex is very sparse. Hypothalamic nuclei with established roles in sleep and circadian regulation show no or very few Cre-expressing cells and we found no overlap with orexin-or melanin-concentrating hormone-expressing cells in lateral hypothalamus (Extended Data Fig. 3-5). As has been shown previously, ablation of SNAP25 virtually abolishes calcium-evoked neurotransmitter release from neurons 15 , rendering the cells functionally silent. Importantly, normal brain development, including cortical layering and axonal path finding, has been shown in SNAP25-ablated mice 15 . We chose this conditional knockout mouse to probe a role for the cortex in sleep-wake regulation because in this model large and widely distributed populations of neo-and archicortical neurons involved in the generation of sleep oscillations and in the communication between cortex and subcortical structures are functionally silenced.
To investigate sleep architecture and electrophysiology in cortical SNAP25-ablated mice we performed chronic EEG, LFP, and MUA recordings as in wild type mice. Laminar MUA revealed a diminished surge of layer 5 firing at the onset of population ON periods (Fig.  1g). In addition, there were significant interaction effects between genotype and cortical layer for both slow wave amplitudes (F (1,8) Fig.   1h,i). The opposite pattern was observed in layer 2/3, where a trend towards increased slow wave amplitudes (t(8)=-2.21, p=0.058, d=-1.40) and elevated slow wave activity (t(8)=3.07, p=0.15, d=-1.94) was observed (Fig. 1h,i). The differential regulation of layer 5 and 2/3 slow waves in cortical SNAP25-ablated mice results in a significantly reduced ratio of slow wave amplitudes and slow wave activity in cortical SNAP25-ablated mice (see insets on Fig.  1h,i). These layer-specific changes to intracortical dynamics are consistent with a reduction of local monosynaptic excitation between layer 5 pyramidal neurons 17 and diminished disynaptic inhibition of layer 2/3 pyramidal neurons 18 . Laminar firing rates and the latency to the first spike in layer 5 were not significantly altered in cortical SNAP25-ablated mice in line with previous reports that abolition of evoked synaptic neurotransmitter release through SNAP25-ablation does not prevent neurons from depolarizing ( Supplementary Fig. 2) 19 . The profound layer-specific changes in LFP slow wave activity during NREM sleep contrasted with a lack of major differences in EEG power spectra during NREM sleep or wakefulness ( Fig. 1j and Supplementary Fig. 3). However, cortical SNAP25-ablated mice presented a leftward shift of the EEG theta-peak frequency during REM sleep (Extended Data Fig. 5). These layer-and sleep-state-specific findings underscore the importance of assessing sleep electrophysiology on the local and global levels 20 .
Beyond the local and global changes in sleep oscillations, we observed profound genotype differences in the daily sleep-wake profile (Fig. 2a). While control animals showed sleep architecture typical for wild type mice (Fig. 2b) 21 , cortical SNAP25-ablated animals presented unusually long wake bouts, that often lasted several hours (Fig. 2b,e). On average, conditional knockout mice spent 13.83±0.39 h awake per day, approximately three hours more than controls (10.57±0.42 h, t(13)=5.55, p<0.001, d=2.96), and the amount of pronounced in the dark period ( Fig. 2d and Extended Data Fig. 6), which is the mouse's circadian active period but also the time of day when the homeostatic sleep drive typically builds up to high levels as a result of prolonged wakefulness 21 . This raises the question whether the increase in wakefulness is due to changes in the homeostatic or circadian process of sleep regulation.
To assess the build-up of the homeostatic sleep drive during spontaneous wakefulness, we compared the levels of EEG SWA during NREM sleep preceding and following individual wake episodes. As expected, a positive correlation was observed in both genotypes, whereby longer spontaneous wake episodes were followed by proportionally higher levels of SWA during NREM sleep. However, the increase of SWA relative to the duration of wake episodes was smaller in conditional knockout mice compared to controls (Fig. 2f, Supplementary Table 1), indicating that the relationship between sleep-wake history and the levels of SWA might be altered in cortical SNAP25-ablated animals.
An established approach to investigate the dynamics of sleep homeostasis is sleep deprivation (SD), which is most commonly performed starting at light onset, when mice in laboratory conditions usually sleep 3,21 . Typically, SD leads to a small increase in sleep amount, especially NREM sleep, and a strong increase in sleep intensity, reflected in SWA during NREM sleep 21,22 . However, we observed a striking difference in this homeostatic rebound between genotypes (Fig. 3a-c). Although many cortical SNAP25-ablated animals had already been spontaneously awake by the time sleep deprivation started ( Supplementary  Fig. 4a), they did not spend more time asleep after SD than controls when sleep deprivation was performed during the first half of the light period (Extended Data Fig. 7). When the time window of the sleep deprivation was shifted to the second half of the light period, the relative amount of sleep over the following 24 hours was reduced in cortical SNAP25-ablated mice compared to controls ( Supplementary Fig. 4b). Moreover, conditional knockout mice presented a marked attenuation of the initial increase of EEG SWA during NREM sleep after sleep deprivation (relative SWA in cortical SNAP25-ablated: 136.77±3.98 %, controls: 180.57±5.13 %, t(11)=6.78, p<0.001, d=-3.87 Fig. 3b,c). Consistent with the notion of a frontal predominance of the homeostatic response to sleep deprivation 23 , the genotype difference in SWA rebound was observed in the frontal EEG and LFPs, but not in the occipital EEG derivation (Extended Data Fig. 8). We also observed that the increase in EEG theta-activity, a measure of 'wake intensity', was attenuated in conditional knockout mice during SD in both the frontal and occipital EEG derivation (Extended Data Fig. 9).
Since the circadian and the homeostatic process interact in the regulation of sleep and wakefulness, we next assessed whether the clock function was altered in conditional knockout mice, using a separate cohort of animals. A standard approach in circadian phenotyping consists in continuously monitoring locomotor activity using a passive infrared recording system under a 12:12 light:dark cycle followed by a release into constant darkness and exposure to a phase-advancing light pulse during the early subjective night (Circadian Time ~13.5) (Fig. 3d) 24 . We replicated the approximately three-hour difference in the amount of wakefulness in this new cohort of animals (cortical SNAP25-ablated: 11.79±0.29 h, controls: 9.10±0.51 h, t(15)=4.957, p<0.001, d=2.35) and observed that the sleep phenotype remained stable in constant darkness (Extended Data Fig. 10). Importantly, Krone  cortical SNAP25-ablated animals remained rhythmic in the absence of light (Fig. 3d). The free-running period was slightly shorter than 24 h and nearly identical in both genotypes (cortical SNAP25-ablated: 23.86±0.04 h, controls: 23.80±0.07 h, t(15)=0.833, p=0.42, d=0.39) and the Chi-Square periodogram analysis did not reveal any difference in circadian amplitude (Fig. 3e). Finally, irrespective of genotype, the light pulse evoked a consistent phase delay, which was comparable between genotypes (Fig. 3f). Taken together, our data demonstrate that cortical SNAP25 ablation leads to a diminished homeostatic sleep drive without affecting circadian regulation of sleep.
The regulation of sleep in mammals is only partially understood because it is unclear where and in what form the need to sleep is encoded, and how it is translated into an adequate compensatory response 25 . Our study reveals a novel role for the cortex in sleep-wake regulation. We show that cortical structures actively contribute to sleep homeostasis and the global control of vigilance states. This supports the hypothesis that brain structures fundamentally involved in sleep regulation extend far beyond the traditionally considered subcortical circuitry 2,7,26,27 . The next step will be to understand how neo-and archicortical neurons interact with established circuits of sleep-wake control. It was recently reported that axons of layer 5 pyramidal neurons in prefrontal cortex produce an axonal terminal field in the lateral hypothalamus 28 . Consistent with this observation, we found dense fine Cre-positive fibres surrounding cell bodies in the lateral hypothalamus (Extended Data Fig.  4). Long ranging projections from prefrontal cortex, which is known to be highly sensitive to homeostatic sleep pressure 23 and a hub for the generation of slow waves during NREM sleep 29 , might represent a direct pathway through which neocortex could modulate the vigilance states control by the lateral hypothalamus. However, layer 5 pyramidal neurons also project to thalamic nuclei, and could influence sleep regulation through corticothalamic loops 30,31 . Given that layer 5 pyramidal neurons have a wide range of efferent connections to target structures involved in sleep-wake control, a systematic and unbiased dissection of the relevant circuits is warranted. Furthermore, a potential involvement of other neocortical cell types and of archicortical dentate gyrus cells of the hippocampus should be considered. Given the critical place of the hippocampus in brain-wide circuitry involved in memory and temporal processing, one could speculate that this structure may have a so far unrecognised role in encoding time spent awake or asleep. Targeted manipulations of the hippocampus and its neocortical projections could shed further light on the relationship between REM and NREM oscillations and their role in global sleep-wake regulation and function 32 .
Our results support the possibility that cortical structures generate sleep drive locally, in an activity-dependent fashion 7 , and raise the question of which mechanisms cortex could use to generate, sense and/or integrate signals of sleep need. Extracellular signals may be found in molecular regulators of inflammation and plasticity 7 , or adenosine levels regulated through neuro-glial interactions. Intracellular processes reflecting wake-dependent increases in sleep need, conserved both in mammalian and non-mammalian species, may represent changes in the synaptic phosphoproteome, endoplasmic reticulum stress, or redox homeostasis. Arguably, such local signals must be integrated to elicit a global homeostatic response 33 , reflected in an occurrence of intense sleep, characterised by elevated cortical SWA and increased sleep propensity. We propose that the wide connectivity of layer 5 pyramidal neurons to other parts of cortex, thalamus, and sleep-wake regulating nuclei in Europe PMC Funders Author Manuscripts hypothalamus 28 and brainstem 16 , places this neuronal population in an ideal position not only to generate SWA, but also to sense and integrate the signals related to sleep need, and ultimately broadcast the information to the subcortical circuitry responsible for sleep-wake switching 1 .

Animals
Rbp4-Cre;Ai14;Snap25 fl/fl is a triple transgenic mouse line, which was designed as a model for functional silencing of cortical layer 5 pyramidal and dentate gyrus granule cells. Snap25 fl/fl is a transgene, with lox-P sites flanking the alternatively spliced exons 5a and 5b of the t-SNARE (target membrane soluble N-ethylmaleimide-sensitive factor attachment protein (SNAP) receptor) gene Snap25. Cre-dependent excision of exon 5a/5b leads to a reduced length gene transcript and non-detectable levels of SNAP25 protein, and cessation of Ca 2+ -dependent evoked synaptic vesicle release 15 . Because this chronic t-SNARE disruption allows cortex-wide silencing of selected cell types while avoiding a mechanical manipulation of cortex, which is known to affect the expression of SWA 34 , we opted for this silencing method. In addition, our choice of this mouse model was guided by previous neurodevelopmental studies on the effects of disrupted evoked neurotransmitter release through ablation of SNARE proteins 19,35,36 as well as by neuroanatomical work conducted in the Rbp4-Cre;Ai14;Snap25 fl/fl mouse line 15 . These studies consistently report that brain development, in particular cortical layering and axonal pathfinding, is unaffected by ablation of SNAP25 15,19 . Moreover, as sleep homeostasis occurs on a timescale of minutes to hours, is most likely a distributed process 33,37 , and it is not yet known whether specific areas of the cortex are more involved than others, we considered the Rbp4-Cre;Ai14;Snap25 fl/fl mouse model most suitable for the aims of this study 16 . Spontaneous behaviour appeared indistinguishable between genotypes, but conditional knockout mice have a lower body weight compared to Cre-negative controls (cortical SNAP25-ablated: 21.3±0.6 g, controls: 24.4±0.6 g, t(18)=2.94, p=0.009), as reported previously 15 .

Chronic electrophysiological recordings
EEG/EMG implants were performed in 7 wild type C57BL/6 mice (WT, age at baseline recording 125±8 days), 12 Rbp4-Cre;Ai14;Snap25 fl/fl mice (cKO, age at baseline recording 90±5 days) and 8 Cre-negative littermates (CTR, age at baseline recording 85±4 days) under isoflurane anaesthesia as described previously 38 . For analysis of sleep architecture based on EEG/EMG recordings, 9 cKO and 6 CTR mice were included. EEG analysis of frontal and occipital spectra was conducted in 8 cKO and 5 CTR mice. Laminar LFP and MUA could be obtained across cortical layers in primary motor cortex (+ 1.1 mm AP (anterior), -1.75 mm ML (left), tilt -15° (left)) of 7 WT, 5 cKO, and 5 CTR mice. All laminar recordings were performed using 16-channel silicon probes (NeuroNexus Technologies Inc., Ann Arbor, MI, USA; model: A1x16-3mm-100-703-Z16) with a spacing of 100 μm between individual channels. All experiments were conducted in young adult male mice, because oestrous cycle, development, and aging affect sleep. The electrophysiological signals revealed typical signatures of wakefulness and sleep states in both genotypes. As expected, the laminar profile of LFPs and MUAs revealed generally activated patterns during waking and rapid eye movement (REM) sleep. Correspondingly, during non-rapid eye movement (NREM) sleep, we observed depth positive LFP slow waves associated with a generalised suppression of spiking activity across cortical layers (Fig.  1c) 20,39 . Electrophysiological acquisition, data processing, and sleep scoring Electrophysiological in vivo recordings-Data was acquired using the 128 Channel Neurophysiology Recording System (Tucker-Davis Technologies Inc., Alachua, FL, USA) and the electrophysiological recording software Synapse (Tucker-Davis Technologies Inc., Alachua, FL, USA), and saved on a local computer. EEG and EMG signals were continuously recorded, filtered between 0.1 -100 Hz, and stored at a sampling rate of 305 Hz. Extracellular neuronal activity was continuously recorded at a sampling rate of 25 kHz and filtered between 300 Hz -5 kHz. Whenever the recorded voltage in an individual laminar channel crossed a manually set threshold indicating putative neuronal firing (at least 2 standard deviations above noise level), 46 samples around the event (0.48 ms before, 1.36 ms after) were stored. Concomitantly with the spike acquisition, LFPs were continuously recorded from the same electrodes and processed with the abovementioned settings for EEG signals.
Offline signal processing-EEG, EMG, and LFP signals were resampled at a sampling rate of 256 Hz using custom-made code in Matlab (The MathWorks Inc, Natick, Massachusetts, USA, version v2017a) and converted into the European Data Format (EDF) as previously described 38 . Spike wave forms were further processed using a custom-made Matlab script and events with artefactual wave forms were excluded from further analysis of neuronal activity.
Scoring of vigilance states-The software Sleep Sign for Animals (version 3.3.6.1602, SleepSign Kissei Comtec Co., Ltd., Nagano, Japan) was used for sleep scoring. EEG, EMG, and LFP recordings were partitioned into epochs of 4 s. Vigilance states were assigned manually to each recording epoch based on visual inspection of the frontal and occipital EEG derivations in conjunction with the EMG. Epochs with recording artefacts due to gross movements, chewing or external electrostatic noise were assigned to the respective vigilance state but not included in the electrophysiological analysis. Overall 18.8±3.5 % of wake, 0.7±0.4 % of NREM, and 0.9±0.4 % of REM epochs contained artefactual EEG signals across all animals included in the EEG spectral analysis, with no significant difference between genotypes. EEG and LFP power spectra were computed using a fast Fourier transform routine (Hanning window) with a 0.25 Hz resolution and exported in the frequency range between 0 and 30 Hz for spectral analysis.

Non-invasive measurement of home cage activity for circadian phenotyping
Circadian characteristics were assessed in a separate cohort of 10 cKO (age at baseline recording 84±1 days) and 7 CTR mice (age at baseline recording 83±1 days). These mice were housed individually and placed under a passive infrared motion detector that continuously recorded activity in time intervals of 100 ms. Individual readings were pooled into one minute bins for all analyses as it has previously been validated that periods of inactivity of >40 s provide a reliable measure of behaviourally defined sleep 40 . Mice were housed in a stable 12:12 h light:dark cycle for at least 7 days before being released into constant darkness. The circadian period and amplitude were assessed over an interval of 7-10 days in the constant dark condition by ChiSquare periodogram analysis using ActogramJ 41 . After at least 10 days in constant darkness, all mice were exposed to a 2h light pulse provided around Circadian Time 13.5 h. The resulting phase delay of the circadian rhythm was quantified in all animals by extrapolating the onsets of activity to the day of the light pulse 24 .

Experimental design and statistical analyses
Sample sizes were initially based on previous studies 6,21 . Power calculations were performed after acquisition of pilot data for the Wellcome Trust grant proposal 203971/Z/16/A indicated an effect size of d = 2.75 for the main outcome parameter NREM sleep time over 24 hours. Applying an intended power of 0.9 and an alpha-error probability of 0.01, the power calculations confirmed the initial sample size estimate of n=8 animals per genotype considering an attrition rate of 25%. Animals were retrieved from the breeder colony upon availability. Based on the breeding scheme we expected that male Rbp4-Cre;Ai14;Snap25 fl/fl mice only represent on average 1/8 of the animals in each litter.
As soon as animals of the desired genotype were available, they were included in the study and the recording capacities were filled with randomly selected Cre-negative controls from the colony, preferably littermates.  42 . To examine potential differences in firing dynamics between superficial versus deep cortical layers in WT animals, repeatedmeasures ANOVAs were conducted with cortical layers (layer 2/3 versus layer 5) and time bins as within-subject factors. To examine potential differences between genotypes, split-plot ANOVAs were conducted; genotype (CTR versus cKO animals) was entered as a between-subject factor, whereas time bins, EEG spectral bins, and vigilance states (wake, NREM, and REM) were entered as within-subject factors. α = 0.05 was adopted for main effects and interactions. Significant two-way interaction terms were followed up with Bonferroni-adjusted post hoc comparisons with αadjusted = 0.05/k, where k represents the total number of pairwise comparisons made 43 . In ANOVAs with multiple time points, the total number of post hoc comparisons was minimised in each case, and hence controlling for the increase in familywise error rate, by pooling across multiple time bins (e.g., 0-10 ms, 11-20 ms, and 21-30 ms in Fig. 1g). For spectral analysis, EEG/LFP power spectra of individual animals were log-transformed before hypothesis testing. Bonferroni correction was not applied for post hoc comparison of spectral data because ANOVAs consisting of 119 EEG spectral bins would be too conservative and reduce statistical power 44 ; thus, α was kept at 0.05 in these cases. In the summary of statistical methods and results (Supplementary Table 1) we report frequency bins with significant differences in post-hoc comparison before (α uncorrected ) and after Bonferroni adjustment of α (α corrected ). In all figure panels illustrating EEG or LFP spectrograms, frequency bins for which the resulting post-hoc comparison was significant after Bonferroni adjustment are highlighted with black asterisks; bins only significant before Bonferroni adjustment are highlighted with grey asterisks to indicate areas of interest bearing in mind that no correction for multiple comparisons was applied. In the circadian analysis, a multivariate ANOVA (MANOVA) was conducted to examine if the linear combination of multiple circadian response variables (period length, periodogram power, and phase shift) differed between genotypes 45

Electrophysiological criteria for analysis of vigilance state episodes and OFF periods
For analyses of mean and maximum duration of sustained wake episodes in the EEG dataset (Fig. 2e), we included wake episodes, which were at least 1 min long allowing brief intrusions of sleep of 1 min or less. For the analysis of mean duration of NREM episodes, we included NREM episodes, which were at least 1 min long allowing brief intrusions of REM sleep or brief awakenings of 1 min or less (Extended Data Fig. 7). To investigate the change in NREM SWA across prolonged wake episodes under undisturbed conditions, we used the EEG dataset from the baseline day and included consolidated periods of waking lasting at least 15 min, whereby short episodes of sleep <1 min were not considered as interruptions. We then performed analyses of NREM sleep SWA in the 15-min time window immediately preceding and following prolonged (>15 min) wake episodes, if both time windows included at least 10 min of artefact-free NREM sleep and no more than 3 min of wakefulness (Fig. 2f). Population OFF periods were defined as periods of total neuronal silence across all electrodes, which lasted at least 50 ms and no more than 4000 ms. Subsequently, the top 20% longest OFF periods were included for final analyses (Fig. 1g, h and Extended Data Fig. 1). The latency to the first spike after the population OFF-ON transition was calculated separately for MUA recorded in layers 2/3 and layer 5. Only ON periods with at least 1 spike in each of the layers occurring within the first 200 ms were included in this analysis (Extended Data Fig. 1).

Histological assessment of laminar probe depth
The tips of laminar implants were stained before surgery with the orange-red fluorescent membrane stain Dil ® (1,1′-Dioctadecyl-3,3,3′,3′-Tetramethylindocarbocyanine Perchlorate; Thermo Fisher Scientific, Waltham, MA, USA) by immersion of the electrode shank into a 20 mg/ml solution (50/50% acetone/methanol) for later histological assessment of the electrode position 46 . After completion of the experiments, microlesions of selected channels on the laminar probe were performed under terminal pentobarbital anaesthesia using the electroplating device NanoZ™ (White Matter LLC, Seattle, WA, USA) applying 10 mA direct current for 25 s to each respective channel. Immediately following microlesioning, mice were perfused with 0.1M phosphate buffered saline (0.9 %) (PBS) followed by 4 % paraformaldehyde in PBS for tissue preservation. A vibrating microtome (Leica VT1000S) was used to section the brains into 50 μm coronal slices. Fluorescent staining was performed with 4′,6-diamidino-2-phenylindole (DAPI). After fluorescence microscopy, implantation sites were mapped using a mouse brain atlas 47  Histological assessment of Cre-expression in cortex, dentate gyrus, and hypothalamus Coronal brain sections from six Rbp4-Cre;Ai14;Snap25 fl/+ mice were prepared using the same procedure as outlined above for the histological assessment of the laminar probe depth. Laser scanning confocal microscope images were acquired (Zeiss LSM710) and image stacks created to assess the cellular morphology across the z-plane of the brain section. The cytoplasmic tdTomato expressed from the Ai14 reporter gene allows reliable identification of Cre+ neurons based on native tdTomato fluorescence 15,49 . The native fluorescence is detectable even in thin neurites and synapses, which can be used to visualise the dendritic and axonal morphology of labelled neurons. The histological assessment of Cre-expression was performed in six Rbp4-Cre;Ai14;Snap25fl/+ mice and the results across all animals were similar to those presented in the representative images.

Immunohistochemistry to determine lateral hypothalamic cell identity
To determine whether Cre+ neurons in the lateral hypothalamus express melanin concentrating hormone (MCH) or orexin/hypocretin (Hcrt), coronal sections from three Rbp4-Cre;Ai14 were prepared as described above. Before counterstaining with DAPI, the sections were stained with rabbit anti-MCH antibody (1:2000; H-070-47; Phoenix Pharmaceuticals) or rabbit anti-Hcrt antibody (1:500, kind gift from Anthony Van den Pol, Yale University). Briefly, sections were incubated in blocking solution (3% or 10% donkey serum [Hcrt or MCH respectively] and 0.3% Triton-X100 in 0.1M PBS) for one hour before incubation with the primary antibody in blocking solution at 4°C overnight (MCH) or for 72h (Hcrt). The primary antibody was revealed by incubating with donkey anti-rabbit-AlexaFluor488 (1:500) antibody in blocking solution for two hours at room temperature. Subsequently, epifluorescence and confocal imaging was performed as described above.  EEG spectral power in the frequency range 4-10 Hz normalised to the mean spectral power over the entire EEG spectrum (0.5 -30 Hz) during REM sleep on the baseline day. Note that peak theta activity is shifted towards lower frequencies in cKOs compared to CTRs in both the frontal and occipital EEG derivations. n=5 CTR and n=8 cKO for EEG spectral analysis. Data in are presented as mean ± SEM (shaded areas   ANOVAs). f) Relationship between wake duration and relative NREM SWA in the frontal EEG derivation during the baseline day. Individual animals are represented with different symbols. n=6 CTR and n=9 cKO for vigilance state analysis (panels c, d, e), n=5 CTR and n=8 cKO for EEG spectral analysis (panel f). Black asterisks indicate post-hoc contrasts with significant differences (*P < 0.05, **P < 0.01, ***P < 0.001), grey asterisk indicates post-hoc comparison with P < 0.05, which does not reach significance after Bonferroni correction for multiple comparisons. Data in panel d are presented as mean ±SEM (shaded areas). Yellow and black bars above panels a, b and d indicate light and dark periods, respectively. Panels c and e represent grouped data including group mean (red line), 95% confidence interval (pink box), and one standard deviation (blue box) with individual data points overlaid. See Supplementary Table 1 for detailed results. a-c) Homeostatic response to 6h sleep deprivation at light onset based on EEG spectral analysis. a) Hypnogram and EEG slow wave activity (0.5-4.0Hz, 4s epochs) of one representative mouse from each genotype (red, yellow and black bars indicate sleep deprivation, light and dark periods, respectively). b) Time course of frontal EEG slow wave activity during NREM sleep after sleep deprivation relative to baseline average (F(11,121) = 7.561, p < 0.001, mixed ANOVA). Note the reduced SWA in cKO compared to CTR mice during the first 2 hours after sleep deprivation. c) NREM sleep spectral power in the frontal derivation during the first 30 minutes after sleep deprivation relative to baseline NREM average (F(118,1298) = 4.068, p < 0.001, mixed ANOVA). Asterisks indicate frequency bins with significant differences in post-hoc comparison before (grey) and after (black) Bonferroni adjustment of α. d-f) Circadian phenotyping based on passive infrared recordings (PIR). d) Representative actograms of one mouse from each genotype. Day 1-3: 12:12h LD cycle. From day 4: constant darkness protocol. Day 13: phase-delaying 2h light pulse. e) Average Chi-square periodogram. Insets: period (h) and periodogram power (Qp). f) Phase shift in activity onset on the day after the light pulse. No significant differences were observed in any of the assessed circadian characteristics. n=5 CTR and n=8 cKO for EEG spectral analysis (panel b,c), n=7 CTR and n=10 cKO for PIR analysis (panels e,f). Black asterisks indicate post-hoc contrasts with significant differences (*p < 0.05, **p < 0.01, ***p < 0.001). Data in panels b, c, e are presented as mean ± SEM (shaded areas). Panels e and f represent grouped data including group mean (red line), 95% confidence interval (pink box), and one standard deviation (blue box) with individual data points overlaid.