Abstract

   Prolonged sleep deprivation (Pr-SD) causes death in many species. While
   various mechanisms related to sleep regulation or this fatal
   consequence of sleep loss have been identified, the core molecular
   basis linking Pr-SD-induced lethality and sleep homeostasis remains
   unknown in mammals. A critical “point of no return (PONE)” status in
   Pr-SD subjects is highlighted in classic research, and characterizing
   PONE status could help uncover this mystery. Using a Pr-SD model and a
   reliable PONE status prediction method, we show that mice in PONE
   exhibit an inability to enter natural sleep, and significant
   disruptions in brain phosphoproteome, independent of deprivation time
   but closely linked to PONE status. Brain kinase or phosphatase
   dysfunction influences PONE status development and leads to
   corresponding sleep aberration concurrently. Daily 80-min recovery
   sleep significantly delays PONE onset and restores brain
   phosphoproteome. The harmful effects of excessive kinase activity on
   PONE development can be eliminated by combining recovery sleep and
   compensatory phosphatase expression. We conclude that sleep is crucial
   for maintaining brain phosphoproteome homeostasis, whose disruption may
   impact both Pr-SD-induced lethality and sleep regulation.

   Subject terms: Proteomics, Bioinformatics, Proteomic analysis,
   Phosphorylation

Introduction

   Sleep is an indispensable behavior conserved across all animal
   species^[56]1–[57]4. Over the past decades, numerous factors that
   regulate the quantity or quality of sleep have been reported, ranging
   from some circadian genes^[58]5 and sleep-related circuits^[59]6 to
   specific kinase signaling pathways^[60]7 and neurotransmitters^[61]8.
   As a homeostatic system, it has been posited that the core molecular
   basis of sleep should govern both its regulation and its primary
   function^[62]9,[63]10. By examining the effects of sleep deprivation on
   varying aspects, many sleep-dependent functions have been proposed,
   including cognition, metabolite clearance, metabolism and
   immunity^[64]10–[65]13. Studies on prolonged sleep deprivation (Pr-SD)
   have also been conducted to dissect the primary function of
   sleep^[66]14, uncovering several mechanisms underlying Pr-SD-related
   lethality, such as the accumulation of reactive oxygen species (ROS) in
   the gut of flies^[67]15, and the cytokine-storm-like syndrome in the
   peripheral system of mice^[68]16.

   Despite the fact that sleep plays a profoundly significant role in the
   central nervous system, up to now, no morphological or functional
   changes in the brain have been identified as mechanisms underlying
   Pr-SD-related lethality^[69]14–[70]17. As a result, the elusive lethal
   mechanism of Pr-SD and the core molecular basis of sleep homeostatic
   regulation are further shrouded in mystery.

   Current methods for studying Pr-SD and its lethality remain limited.
   Gentle handling, a common approach, is ineffective for extended
   deprivation^[71]15,[72]16. The disk-over-water (DOW) method, while
   useful in rats, is hindered by scarce genetic tools for this
   species^[73]14. In mice, DOW is impractical due to restricted mobility
   post-electrode implantation caused by their smaller size. Improved
   models are needed to address these species-specific challenges.

   In DOW of rat studies, a “point of no return (PONE)” status has been
   observed, characterized by the irreversible nature of subjects’ death
   even when the deprivation procedure was terminated^[74]18,[75]19.
   Accurately identifying mice in the PONE status and characterizing the
   profiles of this enigmatic status may contribute to understanding the
   mechanisms underlying Pr-SD-induced lethality. Our PONE status indexing
   analysis, along with proteomic, bioinformatic and genetic studies,
   demonstrated that brain phosphoproteome balance is important for both
   sleep regulation and Pr-SD-induced organismal death.

Results

A Pr-SD sleep deprivation mouse model

   A complete and long-lasting sleep deprivation method is of crucial
   importance for sleep studies. However, the attempt to prolong acute
   sleep deprivation by means of traditional approaches, such as gentle
   handling, is hampered by the phenomenon of rebound sleep and the
   occurrence of local neuron “offline” during homeostatic sleep
   regulation^[76]20,[77]21. Inspired by the seminal DOW studies, which
   demonstrated that rodents wouldn’t sleep in water^[78]22,[79]23, we
   adopted a sustained water aversion method (SWAM) to carry out Pr-SD in
   mice. This method enables efficient performance of more genetic and
   biochemical studies. Mice were individually housed in rectangular cages
   filled with ~0.6 cm of water, and accurate time of death was defined
   when a mouse remained immobile for 5 min (Fig. [80]1a). Indeed, we
   observed a median survival duration of ~108.4 h in SWAM-exposed mice,
   consistent across different mouse substrains (C57BL/6 J, B6J and
   C57BL/6 N, B6N) and ages (8 weeks and 24 weeks) (Fig. [81]1b, c).

Fig. 1. Evaluation of the PONE status during Pr-SD.

   [82]Fig. 1
   [83]Open in a new tab

   a Schematic diagram of SWAM. b, c Survival analysis of SWAM-treated
   mice: C57BL/6 J (B6J, n = 15) vs C57BL/6 N (B6N, n = 16) mice (b), B6J
   cohorts aged 8 weeks (8 W, n = 16) vs 24 weeks (24 W, n = 14) (c). d
   Schematic of SWAM-treated mice undergoing continuous EEG/EMG recording.
   Baseline day (BLD), SWAM onset day (SOD), last full day (LFD). e, f
   Analysis of duration (e) and absolute EEG power spectra (f) of NREMS,
   REMS and wake states of B6J mice (n = 18 per group) on BLD, SOD and
   LFD. g Schematic of PONE status evaluation and PONE index calculation.
   h Representative PONE index development for the mice died on the 6th
   SWAM day during ZT0–5 (yellow, n = 7), ZT6-11 (red, n = 2) or ZT12-23
   (blue, n = 5). i Cumulative mean PONE index value of B6J (n = 15) and
   B6N (n = 15) mice (top), and B6J mice aged 8 W (n = 14) and 24 W
   (n = 17) (bottom). j, k Relative timeline (j) and simulating analysis
   by segmental linear regression (k) for the last four PONE index rating
   in three scenarios. l A plotting of the PONE index values against the
   remaining survival duration from each rating time to the actual death
   point (n = 60). Red line represents mean survival. m Survival analysis
   of SWAM interruption test for mice with PONE index ≤ 2 (n = 26) and
   PONE index ≥ 6 (n = 17). Data are mean ± s.e.m. Log-rank (Mantel-Cox)
   test (b, c, m); two-way ANOVA with Dunnett’s test (e, f right); two-way
   ANOVA with Sidak’s test (f); Exponential growth equation Least squares
   fit (i). *P < 0.05; †P < 0.01; ‡P < 0.001; NS, not significant,
   P > 0.05. n refers to the number of biological replicates.

   Furthermore, we conducted a concurrent electroencephalogram and
   electromyogram (EEG/EMG) analysis during the SWAM test, comparing the
   sleep-wake architecture among baseline day (BLD, the day before the
   SWAM test), SWAM onset day (SOD) and the last full day (LFD, which
   refers to the last complete Zeitgeber time (ZT) 0–23 before the death
   event) (Fig. [84]1d and Supplementary Fig. [85]S1a). Sleep duration
   decreased gradually throughout the SWAM process, from 768.2 min during
   BLD to 361.5 min on SOD, and nearly disappeared on LFD (6.3 min), as
   did the durations of non-rapid eye movement sleep (NREMS) and rapid eye
   movement sleep (REMS) episodes (Fig. [86]1e and Supplementary Fig.
   [87]S1b). Mice experienced about 45 h of complete sleep deprivation
   before death, which was independent of survival duration (Supplementary
   Fig. [88]S1c). Compared to the absolute EEG power of BLD, we observed a
   significant suppression in NREMS delta power on SOD (2–3 Hz), wake
   delta power of SOD (1–3 Hz) and LFD (3–4 Hz) (Fig. [89]1f). These
   observations indicate that SWAM causes a significant reduction in both
   sleep quantity and quality, which underlies inescapable death.

A method to predict PONE status precisely

   Behavioral debilities have been utilized to assess the physiological
   conditions of rats subjected to DOW sleep deprivation^[90]19,[91]23. In
   the SWAM model, we observed similar Pr-SD effects as seen in
   DOW-treated rats^[92]17,[93]19,[94]23. To predict PONE status, we
   employed a combination of 4 behavioral variables to monitor the
   subjects (Fig. [95]1g). More specifically, we deployed 1 Qualitative
   Variable (QV) and 3 Perceptual Variables (PVs), calculating the sum of
   4 rating values to assess mouse status at 3 fixed ZT times (ZT0, 6 and
   11) each day (Fig. [96]1g). To minimize subjective discrepancies as
   much as possible, we defined 2 boundary states which represent “pretty
   healthy” or “extremely feeble” (scored as 0 or 2) for each PV, and
   introduced an intermediate state (scored as 1) including all ambiguous
   states that could not be confidently categorized into either of the 2
   boundary states (Fig. [97]1g). Applying this evaluation method, we
   found a similar PONE index development pattern across mice with varying
   survival times, characterized by a long-lasting initial stationary
   phase followed by a rapid growth phase (Fig. [98]1h). Mice from
   different substrains or ages also exhibited a consistent PONE index
   development curve after mitigating the influence of varying survival
   duration by computing the growth curve of cumulative PONE index (Fig.
   [99]1i and Supplementary Table [100]S1). These findings underscore the
   reproducibility of our scoring system, and consistency of PONE index
   development.

   Next, to further understand the PONE index development, we performed a
   data simulation analysis to quantify the velocity shift of PONE index
   (Fig. [101]1j). We recorded a series of PONE index values for every
   subject. However, we found that the final one was located at different
   ZT time points due to the death events that occurred in different time
   periods (Fig. [102]1h). To analyze the time-dependent development of
   the PONE index accurately, we employed only the last four PONE index
   values for each subject and aligned them within a relative 24-h
   timeline (Fig. [103]1j). The simulation analysis identified an
   inflection point at 1.7, after which PONE index quickly developed to
   the final value of ~5.4 before death (Fig. [104]1k). This suggests a
   transition point likely occurs at a PONE index of 2 in reality. To
   examine whether the PONE index reflects the probability of death, we
   plotted the PONE index against the remaining survival duration from
   each rating time to the actual death time. Results showed that mean
   survival time decreased progressively from ~22.4 h (PONE index = 0) to
   ~12 h (PONE index = 2–4), and further declined to ~6 h once the PONE
   index exceeded 5 (Fig. [105]1l and Supplementary Table [106]S1). To
   test whether we could capture the PONE status successfully, we returned
   mice to their homecages after several days of sleep deprivation and
   PONE index rating (Fig. [107]1a, m). We carried out this SWAM
   interruption test in a paired manner and recorded the mortality of 2
   groups: PONE index ≤ 2 group (whose members had not yet entered the
   rapid growth period) and PONE index ≥ 6 group (whose members
   theoretically experienced at least another 6 h of SWAM after crossing
   the inflection point) (Fig. [108]1k–m and Supplementary Discussions
   [109]1 and [110]2). Interestingly, although both groups of mice
   experienced the same SWAM duration, the mortality rates differed
   significantly — 5.2% vs 92.7% (Fig. [111]1m and Supplementary Fig.
   [112]S1d). This discrepancy suggests that mice with PONE index ≥ 6 have
   entered an irreversible PONE status, jointly representing the two
   distinct poles of status alongside the other group (PONE index ≤ 2)
   (Fig. [113]1a, m). Accordingly, we termed two groups pro-survival (PS)
   and pro-death (PD) to reflect their potential outcomes of survival or
   mortality.

   Together with the simulation data (Fig. [114]1k), these results suggest
   the Pr-SD-induced process may undergo three distinct phases: 1) an
   initial longstanding resistant phase (from start to PONE index = 2)
   that an intrinsic protective mechanism may support the subjects with
   normal behavioral responses; 2) a quick developmental phase (PONE index
   from 2 to 5/6) that the protective mechanism is rapidly out of action
   and the subjects showed abnormal behaviors; 3) the final dying phase
   (PONE index from 5/6 to death) that lethal factors build up and the
   process becomes irreversible (Fig. [115]1a, h, k and m).

Pr-SD causes functional decline of brain electrical activity coupled with
PONE status

   After successfully capturing the point of no return status with the
   SWAM method and the PONE index system, we tried to dissect the causal
   factors related to the emergence of the PONE status. First, we offered
   water-soaked sponges to mice undergoing SWAM as an effort to replicate
   the water stimulus, and to allow for free-will sleep (Supplementary
   Fig. [116]S1e). No mortality or PONE-like status was observed during
   the sponge SWAM, and all mice were released after 20 days
   (Supplementary Fig. [117]S1e). The body weight of mice exhibited a mild
   decrease, and there was no difference between PS and PD mice, which
   could be considered as a side effect of sleep deprivation
   (Supplementary Fig. [118]S1f). We further uncovered that the survival
   duration exhibited no correlation with either the subjects’ body weight
   or basal food intake (Supplementary Fig. [119]S1g, h). Likewise,
   neither the illumination period nor melatonin (which is an antioxidant)
   had an impact on mortality (Supplementary Fig. [120]S1i, j).
   Additionally, the daily food consumption during SWAM exhibited a
   typical circadian pattern, suggesting that the circadian system
   remained intact and that ROS was not a causal factor for mortality
   (Supplementary Fig. [121]S1i–k). Moreover, the blood–brain barrier of
   PD mice did not display abnormal changes as observed in other sleep
   deprivation methods^[122]16 (Supplementary Fig. [123]S1l).
   Collectively, these combined findings indicate that these potential
   confounding factors are not involved in the process of
   mortality^[124]14.

   Next, we focused on the changes in brain electrical activity, and
   examined the absolute EEG power of the last 24 h (L-24h) before the
   time point of actual death and compared it with the BLD and SOD (Fig.
   [125]2a and Supplementary Fig. [126]S2a–g). Consistent with the
   reduction in both sleep quantity and quality (Fig. [127]1e, f), the
   total absolute EEG power, which is mainly dominated by NREMS delta band
   in BLD, showed a significant decrease in SOD and L-24h (Fig. [128]2a,
   b). The ratio of theta (5–8 Hz) vs delta power band was elevated in SOD
   and first half of L-24h (Fig. [129]2b). Notably, the theta/delta ratio
   in L-24h exhibited a plunge at about 12 h before death, and reached its
   lowest point 6 h prior to death, as well as theta, alpha and beta power
   bands (Fig. [130]2b and Supplementary Fig. [131]S2e–g). This pattern
   aligned closely with the three-phase development of the PONE index
   (Figs. [132]1l, [133]2b). These results indicated that the potential
   changes in neuron firing pattern in the brain. Therefore, we measured
   the basic electrophysiological properties of layer 5 pyramidal neurons
   in the primary motor cortex from either ad libitum sleep (AD) or PD
   mice using current clamp recordings (Fig. [134]2c). Approximately 44%
   of PD neurons exhibited significantly higher resting membrane potential
   (RMP > -58 mV) and decreased firing frequencies (Fig. [135]2d, e). To
   demonstrate the differences between mice with different survival
   outcomes, we further examined the EEG data of SWAM interruption test
   (Fig. [136]2f). The none-sleep period between the last NREMS epoch and
   interruption point showed significant difference in PS and PD mice
   (Fig. [137]2g). Notably, only PS mice entered normal sleep eventually
   after returning to the homecage, regardless of varying latency duration
   (Fig. [138]2h). Besides, the absolute EEG power and the theta/delta
   ratio of PD subjects were significantly lower than those in PS mice,
   and could not recover even after SWAM discontinuation (Fig. [139]2i).
   Together, these results strongly indicated a decline in brain
   electrical activity during SWAM process and PONE status.

Fig. 2. Functional changes in the brains of mice with PONE status.

   [140]Fig. 2
   [141]Open in a new tab

   a, b Heat map plot of absolute EEG power (a), mean absolute EEG power
   (b, top), and theta/delta ratio (b bottom) of BLD, SOD and the last
   24 h (L-24h) before the actual death time point (n = 15 per group). c
   Representative trace of action potentials of motor cortex layer 5
   pyramidal neurons from AD mice (top, n = 14), PD mice with normal
   resting membrane potential (RMP) (middle, n = 18), and PD mice with
   RMP > -58 mV (bottom, n = 14). d, e RMP (d left), percentage of neurons
   with RMP > -58 mV (d right) and action potential frequency (e) were
   calculated for AD or PD neurons. f The paradigm of SWAM interruption
   test and relating parameter definitions. g Time within none-sleep
   period of PS and PD mice (n = 13 per group). h Sleep time (left) and
   sleep latency (right) within 24 h after SWAM interruption of PS
   (n = 13) and PD (n = 14) mice. i Mean absolute EEG power (top) and
   theta/delta ratio (bottom) during 24-h periods before (left) and after
   (right) SWAM interruption in PS (n = 13) and PD (n = 14) mice. Data are
   mean ± s.e.m. Two-way ANOVA with Dunnett’s test (b, e); one-way ANOVA
   with Dunnett’s test (d left); two-sided Chi-square (d right); unpaired
   t-test (g, h); two-way ANOVA with Sidak’s test (i). *P < 0.05;
   †P < 0.01; ‡P < 0.001; NS, not significant, P > 0.05. n refers to the
   number of biological replicates.

The phosphoproteomic disruption of brain in PONE status

   Several hours of acute sleep deprivation abolish the daily
   phosphorylation cycle and induce time-dependent cumulative
   phosphorylation of the brain/synaptic proteome^[142]24,[143]25. To
   characterize whether the phosphorylation also accumulates in a Pr-SD
   time-dependent manner or is associated with the PONE status — and to
   identify potential molecular changes corresponding to the decline in
   brain state and the PONE status — we performed quantitative
   phosphoproteomic and proteomic studies using multiplex tandem mass tag
   (TMT) labeling on an Orbitrap platform^[144]24,[145]26,[146]27. Brains
   from AD, PS and PD groups were collected under the SWAM model at
   parallel time points (Fig. [147]3a and Supplementary Tables [148]S2,
   [149]S3). Phosphoproteomic analyses revealed substantial alterations at
   the phosphopeptide level (Q < 0.2 and |log[2](change)| > 0.3) in both
   the PD/PS comparison (5.1% of 29,436) and the PD/AD comparison (19.9%
   of 29,373) (Fig. [150]3b–d). Intriguingly, such changes were absent in
   the PS/AD comparison, even though PS and PD mice underwent the same
   SWAM duration (Fig. [151]3b). This indicates that brain
   phosphoproteomic changes were independent of SWAM duration or
   corresponding stress effects, but rather associated with PONE status.

Fig. 3. Disturbance of brain phosphoproteome during PONE status.

   [152]Fig. 3
   [153]Open in a new tab

   a Experimental design for quantitative proteomic and phosphoproteomic
   studies of Pr-SD model. b–d Volcano plots displaying changed
   phosphopeptides in PS/AD (b), PD/PS (c) and PD/AD (d) groups. In,
   increased [log[2](ratio) > 0.3]; De, decreased [log[2](ratio) < −0.3].
   Multiple unpaired t-test (P value) followed by false discovery rate
   (FDR) (Q value) analysis. e, f Analysis of mean abundance of increased
   (e) or decreased (f) phosphopeptides in PD/AD comparison (d). g, h Venn
   diagram (g), hierarchical cluster and classification analysis (h) of
   317 significantly changed (Q < 0.2) phosphopeptides among three groups
   with total number in parentheses (g). Pr-SD represents all
   significantly changed peptides in PD/PS or PD/AD groups. i–l Global ΔPs
   analysis of phosphoproteins in PD/PS (i) and PD/AD (j) groups, and
   quantitative ΔPs analysis of SNIPPs in PD/PS (k) and PD/AD (l)
   comparisons. Dotted lines represent ΔPs = ±2.4. Numbers of total,
   hyperphosphorylated (Hyper) and hypophosphorylated (Hypo) proteins are
   shown. m Venn diagram showing overlaps of Hyper-phosphoproteins (ΔPs >
   2.4, left) and Hypo phosphoproteins (ΔPs < −2.4, right) among three
   models. n Synaptic percentage analysis for class A–C phosphoproteins
   (left), Hyper-, Hypo- and PD-SNIPPs phosphoproteins (right). The dot
   size represents the number of overlapped kinase motifs and dot color
   shows enrichment factors. One-way ANOVA with Tukey’s test (e, f);
   two-tailed unpaired t-test (k, l); two-sided Chi-square (n). *P < 0.05;
   †P < 0.01; ‡P < 0.001.

   In contrast to the changes observed between 6-h complete deprivation of
   sleep (SD6) and 6-h ad libitum sleep (S6)^[154]24, the faint shifts in
   PS/AD comparison could be attributed to the intermittent sleep bouts
   during Pr-SD (Fig. [155]1e and Supplementary Discussion [156]3), which
   led to a natural restoration of sleep need in the Pr-SD condition and
   accompanying phosphoproteomic alterations^[157]28. However,
   phosphopeptides that showed notable changes in the PD/AD comparison
   also displayed significant group differences in the PS samples (Fig.
   [158]3e, f). This suggests a delicate equilibrium of brain
   phosphoproteins becoming unsettled during the PS state and transforming
   into significant disruptions following the onset of PONE status in the
   PD state. Proteomic analyses showed only a few quantified proteins
   exhibited significant changes in PS/AD (0.95% of 9069), PD/PS (0.31% of
   9075) and PD/AD (2.2% of 9060) comparisons, which were different from
   the phosphoproteomic changes in terms of degree and pattern, and made
   minimal contributions to phosphopeptide changes (Supplementary Fig.
   [159]S3a–i and Table [160]S3). Besides, gene ontology (GO) analysis
   also revealed that the proteomic-enriched components, such as
   fibrinogen complex and lipoprotein particle (Supplementary Fig.
   [161]S3j, k and Table [162]S4), were completely different from
   phosphoproteomic analyses (Supplementary Fig. [163]S4a).

   Our previous study found both SD6 and Sleepy (Slp) mutant mice exhibit
   a global hyperphosphorylation in the comparisons of brain proteomes
   between SD6 and RS3 (SD6 followed by a 3-h recovery sleep) or S6, Slp
   and wild-type (WT) samples^[164]24. We have identified a core set of
   phosphopeptides and hyper-phosphoproteins as the shared substrates in
   the two increased sleep need models (SD6 and Slp) that function as the
   molecular substrates of sleep need^[165]24,[166]29. By applying
   unsupervised cluster analysis to 317 shared peptides in all three
   models (SD6, Slp, and Pr-SD), PD samples were grouped with high sleep
   need group (SD6 and Slp), indicating the response of sleep need-related
   phosphopeptides to Pr-SD (Fig. [167]3g, h). These 317 peptides could be
   further classified into 3 classes (A–C) based on patterns of abundance
   changes (Fig. [168]3h). Subsequently, we conducted phosphorylation
   state change (ΔPs) analysis to elucidate functional alterations of
   phosphoproteins with multiple phosphorylation sites^[169]24,[170]30
   (Supplementary Table [171]S5). 165 Hyper-proteins (ΔPs > 2.4) were
   observed in Pr-SD model, with 43 in the PD/PS comparison and 156 in the
   PD/AD comparison, as well as a profound number of Hypo-proteins
   (ΔPs < 2.4) (Fig. [172]3i, j). In line with the clustering results, the
   sleep-need-index phosphoproteins (SNIPPs), whose phosphorylation states
   closely mirror the shifts in sleep need^[173]24, demonstrated notably
   elevated ΔPs values in both PD/PS and PD/AD comparisons (Fig. [174]3k,
   l and Supplementary Discussion [175]4). We termed the 47 hyper-proteins
   shared in all 3 sleep models as PD-SNIPPs, to highlight the SNIPPs that
   corresponded to PONE status (Fig. [176]3m). Moreover, functional
   analysis revealed most of the class A–C phosphopeptide-corresponding
   proteins, and Hyper-proteins were associated with synaptic functions,
   as indicated by our curated synaptic protein database (Fig. [177]3n and
   Supplementary Table [178]S6). GO analysis showed that these proteins
   were enriched in similar synaptic categories (Supplementary Fig.
   [179]S4a and Supplementary Table [180]S4). All this evidence indicates
   a wide range of synapse- and sleep-related proteins were impacted
   during PONE status, and the KEGG pathway enrichment analysis conducted
   on the 192 core molecular interface proteins (class A–C and PD-SNIPPs)
   also supported this finding (Supplementary Fig. [181]S4b and Table
   [182]S4). Specifically, pathways such as glutamatergic, endocytosis,
   synaptic vesicle release, long-term potentiation, cAMP and calcium
   signaling were particularly relevant. Several core genes — such as
   metabotropic glutamate receptor 5 (Grm5)^[183]31,[184]32, SH3 and
   multiple ankyrin repeat domains protein 3 (Shank3)^[185]33, and
   voltage-dependent R-type calcium channel subunit alpha-1E
   (Cacna1e)^[186]34 — previously identified as sleep regulators and
   associated with synaptic plasticity, were also highlighted.

   In addition to the analysis of phosphoprotein levels, we further
   employed a newly developed method for motif analysis to investigate the
   relevant changes of kinase signaling pathways^[187]35. A tremendous
   kinase disturbance in PD samples was observed, in which 174 of 303
   kinase motifs were significantly enriched (127) or depleted (47)
   (Supplementary Fig. [188]S4c and Table [189]S7). To comprehensively
   explore the kinase signaling patterns in PD samples, we curated 20
   phosphoproteomic datasets with diverse profiles, encompassing scenarios
   such as pharmacological^[190]36–[191]40 or genetic neuron
   manipulation^[192]41, behavior studies^[193]42,[194]43, and cellular
   damage^[195]44 (Supplementary Fig. [196]S4d). All datasets were
   reprocessed using an identical standard and then subjected to kinase
   motif analysis, minimizing potential discrepancies to the greatest
   extent (Supplementary Table [197]S7). Next, we conducted a Functional
   Quadrant Enrichment Analysis (FQEA) to evaluate the kinase signaling
   similarity across all 20 datasets. In brief, kinases showing a
   significant change in both datasets were split into 2 diagonal quadrant
   groups based on their log[2]-transformed frequency factors
   (Supplementary Fig. [198]S4c, e and f). For example, kinases in the
   1^st and 3^rd quadrants were classified into a group with a positive
   correlation (Supplementary Fig. [199]S4f). Subsequently, chi-square
   analysis was performed with the number of kinases in a specific group,
   against the background distribution of kinase points across all
   datasets (Supplementary Fig. [200]S4f).

   FQEA analysis effectively evaluated the similarity of kinase
   disturbance, evident from the strong negative correlation between
   tetrodotoxin (TTX) and bicuculline (BIC) datasets, aligning with their
   respective roles in synaptic up- or down-scaling
   process^[201]36,[202]37 (Supplementary Fig. [203]S4e, g and h). In the
   case of PD datasets (PD/AD and PD/PS), a substantial number of kinase
   motifs displayed positive correlations with sleep-related and BIC- or
   Ca^2+-treated datasets (Supplementary Fig. [204]S4i), suggesting the
   relevant kinase signaling pathways were altered in the brain of PD
   mouse. Conversely, the lack or negative pattern of significant
   correlations between PD and other datasets revealed distinct kinase
   signaling pathway patterns, particularly in comparison with
   stress-related models (chronic unpredictable mild stress, CUMS and
   forced swim stress, FSS) and DNA damage (IR-A549) (Supplementary Fig.
   [205]S4i). Importantly, it is worth noting that the correlation among
   datasets did not depend on the total number of overlapped kinases, but
   rather on the distribution of kinase points within distinct diagonal
   quadrant groups (e.g., in the first line, the number of overlapped
   kinases of TTX-R-24h was more than SRPK2-OE obviously, but no
   significant correlation was observed) (Supplementary Fig. [206]S4i).

   BIC induces homeostatic synaptic downscaling by blocking inhibitory
   synaptic transmission^[207]45. The highly positive correlation in
   kinase signaling pathways between PD and BIC samples demonstrated the
   significant downscaling of synapses in the Pr-SD brain (Supplementary
   Fig. [208]S4f, i). This suggested chronically elevated network activity
   and reduced neuronal firing rates in PD brains, which were in line with
   our electrophysiology results (Fig. [209]2c–e). Besides, we analyzed
   the phosphoproteomic changes of potassium, calcium, sodium, manganese
   ion and proton transporters (Supplementary Table [210]S6). 54% (116 out
   of 213) ion channels showed significant phosphorylation changes
   (Supplementary Fig. [211]S5a–f). Collectively, these findings reveal a
   profound disturbance of phosphoproteome and kinase signaling pathways
   in PONE status, which are closely associated with sleep regulation and
   neuronal activity.

   The contribution of different brain areas to sleep regulation varies,
   raising the question of whether PONE-related phosphoproteomic
   disturbances differ across brain regions. While multiple studies on the
   forebrain have provided valuable insights into sleep regulation,
   knowledge of the cerebellum remains limited. To address this gap, we
   conducted phosphoproteomic analyses of cerebellum samples from AD, PS,
   and PD groups (ADc, PSc, and PDc) (Supplementary Fig. [212]S6a and
   Table [213]S8). Given the smaller scale of the cerebellum experiment
   relative to the whole brain, we applied the corresponding P value
   cutoff of PD/AD comparison to the cerebellar analyses (Fig. [214]3d and
   Supplementary Fig. [215]S6b). In the PDc/PSc and PDc/ADc comparisons,
   6.6% and 10.3% of phosphopeptides, respectively, exhibited significant
   changes (Supplementary Fig. [216]S6c, d). Notably, phosphopeptide
   changes in the cerebellum strongly correlated with those in the whole
   brain in the PD/AD comparison (Supplementary Fig. [217]S6e), indicating
   comparable changes in phosphopeptide levels between the two regions.
   ΔPs analysis further supported this, revealing 72 hyper-phosphorylated
   and 86 hypo-phosphorylated proteins in the PDc/ADc group (Supplementary
   Fig. [218]S6f, g). Additionally, SNIPPs exhibited elevated ΔPs values
   in both the PDc/PSc and PDc/ADc groups, suggesting that
   sleep-need–related phosphorylation dynamics also occur in the
   cerebellum (Supplementary Fig. [219]S6h).

   Next, we compared the extent of phosphoproteomic changes in PD mice to
   other datasets. Overall, the PD samples exhibited the most drastic
   changes in both phosphopeptide and phosphoprotein levels (Supplementary
   Fig. [220]S6i). This indicates that the impact of Pr-SD and PONE status
   on the phosphoproteome was even greater than the effects of BIC
   treatment on cultured neurons. Notably, BIC treatment for 24 h could
   potentially be lethal to the animals, implying that the
   phosphoproteomic alterations observed in PD mice may reflect a more
   profound and widespread disruption than what could be induced by BIC
   treatment in vivo. This suggests that the phosphorylation-based balance
   across all brain regions may be significantly disrupted in PD mice.
   Therefore, we performed kinase enrichment analyses on cerebellum
   samples, revealing significant enrichment or depletion in nearly 50% of
   kinases (Supplementary Fig. [221]S6j, k). FQEA analysis further
   confirmed a similar pattern of kinase pathway changes between
   cerebellum samples and whole brain or BIC-treated samples
   (Supplementary Fig. [222]S6l–o). In the SD6 and Slp models, we
   identified 53 core sleep-regulating kinases that were significantly
   altered across all three comparisons (Supplementary Fig. [223]S6p).
   Many of these kinases were also found in the PD/AD and PDc/ADc
   comparisons, including kinases from the AMP-activated protein kinase
   (AMPK) family, with-no-lysine [K] kinases (WNKs), and cGMP-dependent
   protein kinases (PRKGs), all of which have been implicated in sleep
   regulation (Supplementary Fig. [224]S6p). These findings suggest that
   the cerebellum of PD mice also undergoes significant phosphoproteomic
   disruption, underscoring the widespread impact of Pr-SD on the brain.

SLP and PP2A affect both sleep and PONE status development

   The kinome and phosphatome in PD samples both exhibited profound
   alterations (Supplementary Fig. [225]S7a, b and Table [226]S6), which,
   together with the significant disturbance in kinase signaling under
   PONE status, led us to question whether the surge in brain
   phosphorylation, resulting from the dysfunctions of kinase and
   phosphatase, could affect PONE status and sleep regulation.
   Fortunately, Slp mutant mice (lacking exon 13 of Salt-inducible kinase
   3, Sik3) exhibit relevant phosphoproteomic changes and sleep
   phenotypes^[227]24,[228]29, providing a valuable approach to
   experimentally induce phosphoproteomic disturbances in the mouse brain.
   Hence, we delivered a truncated form of SLP (sSLP, A60-K527, the
   catalytic domain of SLP) into the mouse brain using the AAV-PHP.eB
   virus expression system, driven by the human Synapsin (hSyn) gene
   promoter^[229]46 (Fig. [230]4a and Supplementary Discussion [231]5).
   sSLP expression closely recapitulated the excessive sleep need observed
   in Slp mutant mice, characterized by elevated NREMS delta power and
   duration^[232]29, which were absent in the case of A221, a kinase
   activity-deficient variant of sSLP (Fig. [233]4a–d and Supplementary
   Fig. [234]S7c, d). During the SWAM test, sSLP expression led to notably
   quicker PONE status development and shorter survival duration compared
   to GFP and A221, demonstrating the necessity of kinase activity for
   this phenotype (Fig. [235]4e, f and Supplementary Fig. [236]S7e).
   Furthermore, we expressed active sSLP under the control of Camk2a,
   mDlx, or GfaABC1D promoters to target excitatory neurons, inhibitory
   neurons, and astrocytes, respectively. Among these, only Camk2a
   promoter-driven sSLP expression resulted in an increase in sleep need
   and a reduction in the duration of SWAM (Fig. [237]4g–k and
   Supplementary Fig. [238]S7f–h). Collectively, we have showcased that
   the heightened activity of kinase in excitatory neurons promotes the
   development of PONE status in the SWAM model.

Fig. 4. The effect of SLP on sleep and PONE status development.

   [239]Fig. 4
   [240]Open in a new tab

   a Schematic diagram of the protein structure of SIK3 and truncated
   proteins for AAV expression, and the experimental design for EEG/EMG
   recording and survival tests. b–d Sleep analyses for hSyn-GFP,
   hSyn-A221 and hSyn-sSLP group (n = 13 per group). 24-h absolute delta
   power of NREMS before (pre) and after (post) hSyn-sSLP injection (b).
   Mean absolute NREMS delta power (left) and the change ratio [(‘post’-
   ‘pre’)/ ‘pre’] (right) in mice injected with hSyn-GFP, hSyn-A221 and
   hSyn-sSLP (c). Duration (left) and time difference (right) of NREMS in
   mice injected with hSyn-GFP, hSyn-A221 and hSyn-sSLP (d). e, f Analysis
   of cumulative mean PONE index (e) and survival (f) of SWAM mice
   injected with hSyn-GFP (n = 17), hSyn-A221 (n = 16) and hSyn-sSLP
   (n = 14). g–i Sleep analyses for hSyn-GFP (n = 6), Camk2a-sSLP (n = 9),
   mDlx-sSLP (n = 7) and GfaABC1D-sSLP (n = 7). 24-h absolute delta power
   (g); change ratio of mean absolute NREMS delta power (h); time
   difference of NREMS duration (i). j, k Analysis of cumulative mean PONE
   index (j) and survival (k) of SWAM mice injected with hSyn-GFP
   (n = 13), Camk2a-sSLP (n = 18), mDlx-sSLP (n = 13) and GfaABC1D-sSLP
   (n = 13). Data are mean ± s.e.m. Two-way ANOVA with Sidak’s test (b,
   g); two-tailed paired t-test (c, d, left); one-way ANOVA with Tukey’s
   test (c, d, right); log-rank (Mantel-Cox) test (f, k); one-way ANOVA
   with Dunnett’s test (h, i). *P < 0.05; †P < 0.01; ‡P < 0.001; NS, not
   significant, P > 0.05. n refers to the number of biological replicates.

   In addition to kinase dynamics, phosphatases are another crucial factor
   in maintaining the balance of phosphorylation in the brain. Given the
   substantial number of dephosphorylated peptides in PD samples (Fig.
   [241]3c, d), our hypothesis posited that the disruption of PONE
   status-related phosphoproteomic alterations might result from excessive
   phosphatase function. To systematically identify potential fluctuation
   in phosphatase function, we first conducted a comprehensive
   position-specific scoring matrix (PSSM) search across the entire
   proteome. This involved utilizing three well-established short linear
   motifs (SLiMs) associated with phosphatases (LxxIxE motif for PP2A,
   RVxF motif for PP1, and LxVP and PxIxIT motifs for PP2B) to explore
   their likely substrates^[242]47,[243]48 (Supplementary Fig. [244]S8a
   and Table [245]S9). Then, by computing the relative abundance of
   phosphatase-specific substrates, we observed that only PP2A
   consistently showed signs of elevated activity, as indicated by the
   relatively reduced abundance of its substrates (Fig. [246]5a, b).
   Moreover, we also found increased phosphorylation levels at Ser80/81
   and Ser565 of PPP2R5D, a regulatory subunit of PP2A, which has been
   associated with enhanced phosphatase activity in previous
   studies^[247]49,[248]50 (Fig. [249]5c). Therefore, to confirm the
   effect of phosphatase dysfunction on sleep regulation and the
   development of PONE status, we introduced a single amino acid
   substitution in Ppp2r5d (S565A) using the CRISPR/Cas9 genome editing
   technology (Supplementary Fig. [250]S8b). S565A mutant mice exhibited
   constitutively increased NREMS delta power, while maintaining normal
   sleep duration (Fig. [251]5d and Supplementary Fig. [252]S8c, d),
   indicating a unique role of PPP2R5D in sleep regulation. Echoing the
   EEG findings, S565A mutant mice showed accelerated PONE status
   development and shortened lifespan in SWAM test (Fig. [253]5e, f and
   Supplementary Fig. [254]S8e). Furthermore, the effects of PP2A on sleep
   and PONE status were operative in excitatory neurons, as illustrated by
   inhibiting PP2A activity with Camk2a promoter driven SV40 small T
   antigen AAV system^[255]51 (Fig. [256]5g–i and Supplementary Fig.
   [257]S8f–i). By contrast, mice with point mutations in Ppp1r14a S26A (a
   regulatory subunit of PP1), Ppp3ca S462A or Ppp3cb S472A/S479A (2
   catalytic isoforms of PP2B) exhibited neither sleep abnormalities nor
   PONE status development aberration (Supplementary Fig. [258]S8j-q),
   despite the phosphatase activity of PP1 declined concordantly across
   all sleep models (Fig. [259]5b). S565A mice exhibited subtle changes in
   brain phosphoproteomics, with 85% of peptides showing increased
   phosphorylation levels (Supplementary Fig. [260]S9a, b and Table
   [261]S8). Moreover, 14 proteins, including SNIPPs such as Protein
   piccolo (PCLO) and Microtubule-associated protein 2 (MAP2), were
   identified as hyper-phosphorylated (Supplementary Fig. [262]S9c, d).
   These findings align closely with the elevated NREMS delta power
   observed in S565A mutant mice. Additionally, an increase in
   phosphorylation levels was noted in PP2A substrates (Supplementary Fig.
   [263]S9e), reflecting altered phosphatase activity of PP2A. KEGG
   analysis of proteins with significantly altered peptides revealed that
   pathways such as the cAMP signaling pathway and dopaminergic synapse
   were impacted by the S565A mutation in PPP2R5D (Supplementary Fig.
   [264]S9f). Based on these results, we identified the dual roles of SLP
   and PP2A in sleep regulation and PONE status development, serving as
   prominent instances within the wide spectrum of kinase-phosphatase
   dysregulations observed in PD samples.

Fig. 5. The effect of PP2A on sleep and PONE status development.

   [265]Fig. 5
   [266]Open in a new tab

   a Volcano plots of quantified phosphopeptides of PP2A specific
   substrates in PD/AD comparison. b Relative abundances of the specific
   substrates for three phosphatases in sleep-related comparisons. c
   Volcano plots of quantified phosphopeptides of PP2A family in PD/AD
   comparison. d 24-h (left) and mean (right) relative NREMS delta power
   of Ppp2r5d (S565A) homozygous (HOMO, n = 14) and WT (n = 13)
   littermates. e, f Analysis of cumulative mean PONE index (e) and
   survival (f) for Ppp2r5d (S565A) littermates (n = 15 per group). g 24-h
   absolute delta power (left) and the change ratio (right) of NREMS in
   mice injected with hSyn-GFP (n = 10) and Camk2a-SV40 small T antigen
   (n = 8). h, i Analysis of cumulative mean PONE index (h) and survival
   (i) of SWAM mice injected with hSyn-GFP (n = 12) and Camk2a-SV40
   (n = 11). Data are mean ± s.e.m. Two-way ANOVA with Sidak’s test (d, g,
   left); two-tailed unpaired t-test (d, g, right); log-rank (Mantel-Cox)
   test (f, i). *P < 0.05; †P < 0.01; ‡P < 0.001; NS, not significant,
   P > 0.05. n refers to the number of biological replicates.

Brain phosphorylation balance reconstruction resists PONE status development

   Recovery sleep immediately following SD6 is accompanied by global
   dephosphorylation of brain proteome^[267]24, which may function as an
   essential time window to restore the overall molecular and neuronal
   networks challenged by previous waking experiences and maintain the
   organismal fitness. To further characterize the relationship between
   sleep, phosphoproteomic disturbance, and PONE status, we introduced a
   4-h recovery period (RP4h) to SWAM-exposed mice since the 2^nd SWAM day
   during a fixed time zone (ZT0 to ZT4) (Fig. [268]6a). EEG/EMG analysis
   showed that within the RP4h, there was ~80 min of recovery sleep (NREMS
   69.7 ± 9.4 min; REMS 10.5 ± 2.8 min), which was much shorter compared
   to baseline sleep (Fig. [269]6b–d, Supplementary Fig. [270]S10a, b and
   Discussion [271]4). We found a significantly delayed onset of PONE
   status and mortality under the condition of RP4h, with several mice
   remaining alive even after 15 days (Fig. [272]6e, f). Introducing
   classical sleep deprivation procedure during RP4h completely abolished
   the delay in mortality, which strongly demonstrates the primary role of
   sleep in the curative effect of the recovery period (Supplementary Fig.
   [273]S10c).

Fig. 6. The effect of brain phosphoproteomic reconstruction on PONE status
development.

   [274]Fig. 6
   [275]Open in a new tab

   a Schematic of the 4-h recovery period (RP4h) experiment. b, c Analysis
   of hourly duration of NREMS (b) or REMS (c) for BLD (ZT0-3) and the
   average of all RP4 (n = 16). d Total sleep time on BLD (ZT0–3),
   1st-RP4, L-RP4 (last-RP4) and the average (Ave) of all RP4 (n = 16). e,
   f Cumulative PONE index (e) and population survival (f) of SWAM mice
   without (SWAM) or with recovery (RP4, n = 10 per group). g Schematic
   diagram of the RP4h phosphoproteomic study. h Volcano plot showing all
   phosphorylation changes in the RP0/PS group. In, increased; De,
   decreased. i, j Mean abundance analysis of increased (i) and decreased
   (j) phosphopeptides in the RP0/PS comparison under three conditions. k
   Global ΔPs analysis in RP0/PS comparison. l Comparative ΔPs analysis of
   47 PD-SNIPPs between RP0/PS and RP4/PS comparisons. m Global kinase
   motif enrichment analysis of phosphoproteomic data from RP4/PS groups.
   En, enriched; De, depleted. n–q Cumulative PONE index (n, p) and
   survival analyses (o, q) of SWAM mice with RP4h and injected with AAV.
   n = 11 (n, o, GFP, sSLP and A221); n = 10 (p, q GFP and GFP&sSLP);
   n = 9 (p, q GFP&PPP2CA). r A model illustrating homeostasis of brain
   phosphoproteome linking sleep regulation and functions. Data are mean ±
   s.d. Two-way ANOVA with Sidak’s test (b, c); one-way ANOVA with Tukey’s
   test (i, j); log-rank (Mantel-Cox) test (o, q); two-tailed unpaired
   t-test (l). *P < 0.05; †P < 0.01; ‡P < 0.001; NS, not significant,
   P > 0.05. n refers to the number of biological replicates.

   To examine whether this minimal amount of recovery sleep could also
   restore the phosphorylation state of brain proteome, we harvested mice
   brains in three conditions and all mice were evaluated as being in the
   PS state: 1) the 4th SWAM day without RP at ZT0 (PS); 2) the 10th SWAM
   day at ZT0, the beginning of the 9th RP (RP0); 3) the 10th SWAM day at
   ZT4, the end of the 9th RP (RP4) (Fig. [276]6g). We observed 2,111
   hyper-phosphorylated peptides (Q < 0.2) in RP0 samples, whereas
   alterations in RP4 were the least pronounced (Fig. [277]6h and
   Supplementary Fig. [278]S10d). A significant restoration of
   phosphopeptides abundance was detected in RP4, including both increased
   and decreased ones of RP0/PS comparison (Fig. [279]6i, j). This subtle
   improvement effectively preserved a phosphoproteomic profile resembling
   that of the PS state, which was considered a causal factor in the
   delayed development of PONE status (Fig. [280]6e, f). The
   phosphoproteomic fluctuations between RP0 and RP4 showed patterns
   similar to the phosphorylation changes observed in SD6 models, where
   brain phosphorylation levels varied dynamically in response to sleep
   need^[281]24. This indicates a fragile sleep homeostasis (or brain
   phosphorylation balance) develops in response to the recovery sleep
   (Fig. [282]6i, j). ΔPs analysis further confirmed the restorative
   effect at the phosphoprotein level, particularly highlighted by the
   declined ΔPs value of PD-SNIPPs (Fig. [283]6k, l and Supplementary Fig.
   [284]S10e–i). The hyper-proteins of RP0/PS and hypo-proteins of RP4/PS
   were highly enriched in synaptic-related GO items (Supplementary Fig.
   [285]S10j–l). Besides, we also identified a set of functionally
   depleted kinases in RP4 samples, in contrast to their functional
   enrichment observed in PD samples (Fig. [286]6m and Supplementary Fig.
   [287]S4c). These combined findings on phosphopeptides, phosphoproteins
   and kinase signaling pathways collectively confirm the restorative
   effect of the recovery sleep on the brain phosphoproteome in
   Pr-SD-treated mice.

   Next, we examined the effect of RP4h on sSLP expression mice to figure
   out the curative impact of recovery sleep on kinase dysfunction. The
   results showed that RP4h extended the SWAM duration of sSLP expression
   mice, and delayed PONE status development (Fig. [288]6n, o). However,
   the SWAM defect remained evident in sSLP mice, compared to A221 and GFP
   (Fig. [289]6n, o). This partial rescue was probably due to the
   exorbitant kinase activity of sSLP, which exceeded the modulation
   capacity of RP4h. Therefore, in order to counteract excessive kinase
   activity, we co-expressed a series of PP2A phosphatase activity
   agonists, including PPP2R5D, leucine carboxyl methyltransferase 1
   (LCMT1)^[290]52 and the catalytic subunit of PP2A (PPP2CA) in sSLP
   expression mice. All these PP2A agonists prolonged the survival
   duration to some extent in RP4h treatment, with PPP2CA, in particular,
   fully rescuing mice from accelerated mortality and normalizing the
   development of PONE status (Fig. [291]6p, q and Supplementary Fig.
   [292]S10m, n). Notably, immunoblotting with antibodies against 3
   phosphorylated substrate motifs showed that, complementary expression
   of PPP2CA significantly restored global phosphorylation of substrates
   of AMPK, protein kinase C (PKC) and “ataxia telangiectasia mutated”
   (ATM) compared to sSLP expression (Supplementary Fig. [293]S10o). In
   addition, we selected five antibodies targeting specific
   phosphorylation sites to confirm the changes in the brain following
   sSLP and PPP2CA overexpression, including HDAC4, a substrate of SIK3
   (Supplementary Fig. [294]S10p). The results showed that phosphorylation
   levels of all five sites increased after sSLP expression and
   significantly decreased in the PPP2CA co-expression samples
   (Supplementary Fig. [295]S10p). Taken together, these findings
   demonstrate that sleep facilitates reconstruction of the brain
   phosphoproteome following Pr-SD, highlighting the pivotal role of
   kinase-phosphatase balance in the development of PONE status.

Discussion

   It has been proposed that the core sleep functions should be tightly
   linked to the homeostatic regulation of sleep through an undefined
   molecular interface^[296]9–[297]11. Together with recent
   phosphoproteomic studies on
   sleep^[298]24,[299]25,[300]32,[301]53,[302]54, evidence from varying
   experimental paradigms and multiple temporal dimensions indicates that
   the brain/synaptic phosphoproteome is homeostatically regulated across
   sleep–wake cycles, and that disturbance caused by forced awakening must
   be restored by the following recovery sleep. Functional analyses
   further suggest that the core sleep functions, such as maintaining
   synaptic homeostasis and cognitive fitness during daily circumstances,
   and safeguarding survival in the face of long-lasting sleep loss are
   mechanistically linked to the restorative role of sleep on
   phosphoproteome homeostasis, which also underlies the regulation of
   sleep–wake homeostasis^[303]24,[304]25,[305]32,[306]53,[307]54. Thus,
   we posit that an intrinsic phosphorylation-based feedback regulation
   (PBFR) system^[308]9,[309]53 is hijacked by both homeostatic regulation
   of sleep and the core effects induced by sleep loss, including sleep
   need^[310]24,[311]25,[312]53, synaptic functions^[313]54, and
   organismal survival (Fig. [314]6r). The regulators and effectors of the
   PBFR system may represent a parsimonious molecular interface linking
   sleep regulation and such fundamental cellular/systemic functions.

   The decrease in NREMS delta power during SWAM was a counterintuitive
   result, contrary to expectations based on research into classical sleep
   deprivation, where an increase was anticipated (Supplementary
   Discussion [315]4). We interpreted this as an indication that the
   quality of sleep was poor due to frequent awakenings. Besides, previous
   studies have reported a blunted increase in NREMS delta power when
   sleep deprivation exceeds 9 h^[316]28. Accordingly, we reasonably
   hypothesized that there might be a natural restoration/resilience
   mechanism counteracting the elevation of NREMS delta power, which could
   also be operative in the current case. Moreover, short sleep occurred
   but was severely disrupted or inhibited during SWAM, potentially
   leading to a suppression or disruption of NREMS delta power and sleep
   homeostasis. This, in turn, may have resulted in the loss of sleep
   ability at the PONE status. Notably, the biological role of NREMS delta
   power remains largely unknown, and the question of whether it is merely
   an epiphenomenon or serves some mysterious functions remains unanswered
   despite years of research. Our results may offer new insights and raise
   awareness regarding the significance of NREMS delta power and the sleep
   homeostasis it may represent.

Materials and methods

Animals

   All animal experiments were performed according to procedures approved
   by the Institutional Animal Care and Use Committee of Harbin Institute
   of Technology (HIT/IACUC). Unless otherwise stated, all mice were 8–12
   weeks males in C57BL/6 J background and housed under humidity
   (50 ± 10%), light (12-h light/dark cycle) and temperature (24 ± 2 °C)
   controlled conditions with free access to water and food. Point
   mutation mice (Ppp2r5d S565A, Ppp1r14a S26A, Ppp3ca S462A and Ppp3cb
   S472A/S479A in C57BL/6 L background) were designed and generated at
   Cyagen Bioscience (Guangzhou, China).

Pr-SD

   Pr-SD was conducted based on rodents’ (rats or mice) aversion to water
   environments, as they typically avoid sleeping in
   water^[317]22,[318]23. To achieve continuous sleep deprivation, we
   employed the SWAM. In this method, mice were housed individually in
   rectangular cages filled with ~0.6 cm of sterile water. During the SWAM
   onset day, mice entered a novel environment basically and explored the
   cage at the beginning of several hours. After that, the thin layer of
   water prevented mice from entering into deep sleep and woke mice when
   their noses touched the water, which caused a significant sleep
   duration reduction.

   To ensure animal welfare, several measures were taken. All mice were
   housed in clean and spacious cages prior to the experiments, with a
   two-day acclimation period to allow them to adjust to single housing.
   During the experiments, cages and water were replaced every 24 h
   following the PONE index rating. Additionally, the mice were monitored
   continuously for survival via camera. Survival time was defined as the
   total hours from the start of sleep deprivation until spontaneous
   organismal death (immobility for over 5 min) or the time point of
   euthanasia.

   In line with animal welfare guidelines, mice that reached PONE status
   after index rating were euthanized at three specific time points each
   day to alleviate their suffering, except for experiments directly
   related to establishing the PONE index system.

PONE status evaluation

   Variables used for subjects’ status evaluation were adopted from
   disk-over-water studies^[319]17–[320]19,[321]23. Four behavioral
   variables were rated on a 3-point scale (0, 1 and 2) represent the
   increased severity of debility. Value 0 or 2 was assigned when subjects
   were in a “pretty healthy” or “extremely feeble” state, while value 1
   was assigned when subjects were in an “intermediate” state,
   encompassing all ambiguous conditions that were less confidently
   categorized as any of the two boundary states. The “pretty healthy” and
   “extremely feeble” states of each variables were defined as following.
   (1) frequency and duration for forepaws underwater: less than 2 s in an
   observation of 5 s; more than 10 s in an observation. (2) status of
   mouth stick with water: totally dry; stuck with an obvious droplet. (3)
   response after gentle touch: run away rapidly in 2 s; not run away in
   10 s. (4) food consumption ratio compared with the same time zone of
   previous day: > 60%; < 30%. Due to mice were frequently climbing onto
   the food rack of cage lid on the 1st SWAM day, new chow diet was
   replaced and weighted at ZT 0 of the 2nd SWAM day, thus the first full
   PONE index can be calculated at ZT6 of the 3rd SWAM day. When previous
   day’s food consumption value was not available, the average food
   consumption during the same period of the 2nd SWAM day of WT mice was
   used for calculation. Rating was performed at fixed time point, ZT0,
   ZT6 and ZT11 for each animal, and the PONE index value was calculated
   as the sum of all rating values at indicated ZT.

   When baseline variables were measured, such as body weight and basal
   food intake, mice were individually housed at least 3 days for
   adaptation, and 1-day baseline measurements before SWAM were used as
   basal condition. When analysis the cumulative effect of PONE value, the
   mean PONE value for each experimental group was calculated and used for
   exponential growth regression analysis. When analysis the
   time-dependent development of PONE index, the last four PONE index
   values for each subject were aligned within a relative 24-h timeline
   and analyzed by segmental linear regression to identified the
   inflection point. The remaining survival duration was calculated from
   each rating time to the actual death time, when two sequential ratings
   shown same PONE index value or later one smaller, only the duration
   from the last one was used. For the 4-h recovery period (RP4h)
   experiments, mice were individually housed at least 3 days and resided
   in their homecages during recovery. The recovery was performed since
   the 2nd SWAM day at ZT0 for 4 h per day and released all survived
   subjects after the 15th SWAM day. Representative results and analyses
   of PONE index were listed in Supplementary Table [322]S1.

EEG recording and sleep phenotype analysis

   The sleep phenotype analysis was performed as previously
   described^[323]24,[324]29,[325]55. Under anesthesia by isoflurane (3%
   for induction and 1% for maintenance), 8–10 weeks old mice were
   implanted with EEG/EMG electrodes under stereotaxic control. After
   surgery, mice were individually housed and allowed a minimum 1-week
   recovery. Mice were then acclimatized to the recording chamber around
   1-week before EEG/EMG signal recording. For SWAM and RP4h experiments,
   3 days baseline EEG/EMG recording were conducted and the day before
   SWAM onset was used as basal condition. EEG/EMG signals were monitored
   continuously. Cleaned cage and water was replaced every 24 h. Mice were
   returned to their homecages during recovery. For AAV experiments,
   baseline EEG/EMG recording was conducted on WT mice, then randomly
   divided into different groups and injected with according AAVs as
   indicated. Mice remained in the same recording chambers for 3 weeks,
   then EEG/EMG signals were recorded.

   Raw EEG/EMG recording data were subjected to semi-automated analysis
   followed by manual correction. Fast Fourier transform analysis was used
   to decompose EEG signals to 1 Hz bins from 1 to 30 Hz for following
   analysis. 20 s epoch was used to characterize wake (low amplitude and
   fast EEG, high amplitude and variable EMG), REMS (dominant 5–8 Hz theta
   band EEG and EMG atonia), or NREMS (high amplitude 1–4 Hz delta EEG and
   low EMG tonus) state. Absolute (arbitrary units) EEG power density was
   determined by the delta (1–4 Hz), theta (5–8 Hz), alpha (9–14 Hz), and
   beta (15–30 Hz) band activities during wake, REMS, or NREMS state for
   indicated period. In circadian variation plots, each data point
   represents the mean value of either duration or EEG power density in
   the following 1 hour for the corresponding wake, REMS, and NREMS state.
   Researchers were blinded to the treatments during data analysis and
   only animals with unreadable EEG signals were excluded.

   The LFD was defined as the last complete ZT0-23 before the death event.
   The L-24h was defined as a complete 24 h before the death time point.
   For, example, if a mouse dies at ZT2 of Day 4, “LFD” refers to Day 3,
   representing the last complete time zone from ZT0 to ZT23. “L-24h”
   refers to the 24-h period leading up to death, from ZT2 of Day 3 to ZT2
   of Day 4.

Mass spectrometry sample preparation

   Mass spectrometry samples were prepared as previously described with
   modifications^[326]24. Mouse tissues were quickly dissected, flash
   frozen in liquid nitrogen and stored in –80 °C. For the Pr-SD model,
   whenever PD mouse was identified, whole brain from PS and AD mice were
   collected to ensure PS and PD subjects experienced the same SWAM
   duration. For the RP4h model, all mice used were evaluated as in PS
   state and brains were harvested at three conditions: 1) the 4th SWAM
   day at ZT0 without recovery period (PS); 2) the 10th SWAM day at ZT0,
   the beginning of the 9th recovery period (RP0); 3) the 10th SWAM day
   ZT4, the end of the 9th recovery period (RP4).

   Typically, 5 mL of lysis buffer (50 mM HEPES (pH 7.4), 150 mM NaCl,
   2 mM MgCl[2], 1% SDS) was used for homogenization of one whole brain in
   a glass tissue homogenizer. Protein lysates was mixed with 50 U/mL of
   Universal Nuclease (Thermo Scientific), and then added 2 mL 10% SDS
   buffer. Protein homogenates were incubated for 30 min and centrifuged
   at 15,000 g for 20 min at room temperature, supernatant was transferred
   to a new 15 mL tube. Bicinchoninic acid (BCA) assay was used to
   determine protein concentration (Thermo Scientific). Protein lysates
   were reduced with dithiothreitol and alkylated with iodoacetamide,
   precipitated by chloroform–methanol precipitation method and then
   resuspended in 8 M urea buffer. Protein lysates were digested with
   lysyl-endopeptidase (Lys-c) (Wako Pure Chemical Industries) for 2 h,
   followed by dilution to 2 M urea with 25 mM ammonium carbonate buffer
   (pH 7.8), and trypsin (Promega) digestion overnight at room
   temperature, 1% trifluoroacetic acid (TFA) used to stop the digestion.
   Peptide mixtures were subjected to C18 solid-phase extraction (Sep-Pak,
   Waters) and then vacuum-centrifuged to near-dryness.

   Desalted peptides from 8 mg protein lysates for each sample were
   resuspended in 1 mL phosphopeptide binding buffer (2 M lactic acid/50%
   acetonitrile (ACN)) and centrifuged at 15,000 g for 20 min; supernatant
   was transferred to a new tube for phosphopeptides enrichment. Titanium
   dioxide (TiO[2]) beads (GL Sciences) were washed for three times with
   phosphopeptide binding buffer. Equal amount TiO[2] beads were added to
   each peptide mixtures and incubated with gentle rotation for at least 1
   h. Then TiO[2] beads were washed twice with phosphopeptide binding
   buffer, twice with wash buffer (50% ACN/0.1% TFA). Finally,
   phosphopeptides were eluted twice with 500 µL elution buffer (50 mM
   K[2]HPO[4], pH 10), and acidified with 20% TFA, desalted and
   vacuum-centrifuged to near-dryness.

   Total peptides or enriched phosphopeptides were resuspended in 200 mM
   HEPES (pH 8.5) buffer. Approximately 100 µg total peptides or 50 µg
   phosphopeptides for each sample were used for proteomic or
   phosphoproteomic studies, respectively. Peptide samples were labelled
   with tandem mass tag (TMT) isobaric reagents (Thermo Fisher Scientific)
   for 1 h and quenched with hydroxylamine, then combined into one tube
   and acidified with 20% TFA, desalted and vacuum-centrifuged to
   near-dryness. TMT-labeled sample was resuspended in 400 µL buffer A (1%
   ACN, 10 mM ammonium formate, pH 9.5) for HPLC fractionation. HPLC
   fractions were acidified with 20% TFA and vacuum centrifuged to
   near-dryness, desalted via StageTips and dried by vacuum
   centrifugation.

Mass spectrometry data acquisition and processing

   Mass spectrometry data were collected as previously described with
   modifications^[327]24,[328]26,[329]27,[330]56,[331]57. MultiNotch
   synchronous precursor selection MS3-based TMT method was used on
   Orbitrap-Fusion mass spectrometer platform coupled with EASY-nLC liquid
   chromatography (LC) pump (Thermo Fisher Scientific). Custom made
   pre-column and analytical column were used for sample trapping and
   analytical separation. Peptides were separated at a flow rate of
   300 nL/min using a gradient of 6%–27% ACN (0.1% formic acid) over
   190 min. MS1 spectra (400–1500 m/z) data were acquired at 120,000
   resolutions from Orbitrap with 4e5 AGC target, and 50 ms maximum
   injection time in profile data type. MS1 ions were isolated using the
   quadrupole with a 0.7 m/z isolation window in top speed mode. MS2 scans
   (400–1200 m/z) were acquired from the Ion Trap in Turbo mode with CID
   fragmentation (35% collision energy), 1e4 AGC target, and 50 ms maximum
   injection time in centroid data type. For TMT reporter ions
   quantitation, top ten MS2 fragment ions were selected using synchronous
   precursor selection mode. MS3 scans were acquired from the Orbitrap at
   60,000 resolutions with HCD fragmentation (65% collision energy), 1e5
   AGC target, and 120 ms maximum injection time in profile data type.
   Ions were not accumulated for all parallelisable time.

   The data of cerebellum and S565A mouse brain samples were collected
   with the Ultimate 3000 system and Thermo Orbitrap exploris 480 mass
   spectrometer. The TMT-labeled peptides were separated by using a
   120 min gradient elution at a flow rate 0.3 mL/min with the Ultimate
   3000 system which was directly interfaced with a Thermo Orbitrap
   Exploris 480 mass spectrometer. The analytical column was a
   fused-silica capillary (75 mm ID, 150 mm length; Upchurch, Oak Harbor,
   WA) packed with C-18 resin (300 A˚, 5 mm; Varian, Lexington, MA).
   Mobile phase A consisted of 0.1% formic acid, and mobile phase B
   consisted of 100% acetonitrile and 0.1% formic acid. The Orbitrap
   Exploris 480 mass spectrometer was operated in the data-dependent
   acquisition mode using Xcalibur4.5 software. There was a single
   full-scan mass spectrum in the Orbitrap (300–1800 m/z, 60,000
   resolution) followed by 2 s data-dependent MS/MS scans in an ion
   routing multipole at 36% normalized collision energy.

   Raw mass spectrometry files were searched by Proteome Discoverer
   software (PD, version 2.5, Thermo Fisher Scientific) against a
   composite target/decoy database using SEQUEST^[332]58–[333]60. Total 5
   raw datasets for SD6 and Slp models from Wang et al.^[334]24. were
   reanalyzed using same parameters for cross-reference analysis. Protein
   entries including all isoforms were retrieved from UniProt as the
   target mouse protein database (April 17, 2022; 94,765). MS2 spectra
   were searched with fully tryptic restriction, three maximal missed
   cleavages, ±10 ppm for precursor ion mass tolerance, ±0.8 for fragment
   ion mass tolerance, fixed TMT modifications on the N terminus and
   lysine ( + 229.1629 Da) and carbamidomethylation of cysteine residues
   ( + 57.0215 Da), and dynamic mass shift for oxidation of methionine
   ( + 15.9949 Da), phosphorylation on serine, threonine and tyrosine
   ( + 79.9663 Da). The peptide spectrum matches (PSMs) were filtered to
   achieve 1% protein and peptide FDR (according to Q value) by
   Percolator^[335]61. Phosphorylation sites were localized by ptmRS51,
   and phosphopeptides with the phosphorylation site probability score ≥
   25 were used for following analysis. TMT reporter ion signal-to-noise
   (S/N) values were quantified from MS3 scans with the most confident
   centroid setting for matching peptides using an integration tolerance
   of 20 ppm (Orbitrap). The sum of raw reporter ion values for each TMT
   channel was normalized by assuming equal amount protein input of all
   channels for following data analyses^[336]24.

Mass spectrometry data analysis

   Protein isoform was considered as a different protein for proteomic
   analysis. Phosphopeptide was used for phosphoproteomic analysis
   including unique and composite (contain ≥ 2 phosphorylation sites)
   forms. The normalized abundance data (consuming the protein level over
   each TMT channel was identical) from an experiment were consolidated
   (sum of value) to generate unique protein or phosphopeptide ID and
   consolidated abundance values, which were then scaled (the average
   abundance is one for each subject) to minimize batch effects between
   different proteomic experiments. The scaled data from different
   experiments were integrated together based on the unique subject ID for
   the same comparison (e.g., PD/PS). Then the mean value for each
   experimental group was used to generate the “log[2](fold change)”
   value; and statistical significance (Q < 0.2) was determined by the
   multiple unpaired t-test (P value) analysis following the two-stage
   step-up FDR (Q value) approach^[337]62 unless otherwise stated.
   Notably, no further filter or computation was performed to high-missing
   peptides before statistical analysis. The full description and datasets
   for all phosphoproteomic experiments were listed in Supplementary Table
   [338]S2, and those for all proteomic experiments were listed in
   Supplementary Table [339]S3.

   The phospho-state change (∆Ps) value was calculated as previously
   described^[340]24, which is equal to the “sum of log[2](fold change)
   value” of all statistically significant changed (Q < 0.2)
   phosphopeptides from all protein isoforms encoded by the same gene. If
   none of phosphopeptide’s Q value is less than 0.2, the ∆Ps value will
   be zero. The ∆Ps value for each comparison of brain phosphoproteome was
   normalized with the total quantified phosphopeptides number of SD6/RS3
   comparison for cross-experiment analysis. Due to only few
   phosphopeptides passed the FDR test (Q < 0.2) in RP4/PS comparison, P
   value (P < 0.022, equal to Q < 0.2 in RP0/PS) was used to determine
   statistical significance and ∆Ps analysis for RP4/RP0 comparison. A
   stringent cutoff for ∆Ps value (+/-2.4) was determined by null-tests
   and applied to identify hyper-phosphorylated (Hyper: ∆Ps>2.4) and
   hypo-phosphorylated (Hypo: ∆Ps < −2.4) proteins as described12 to
   represent the concept of cumulative phosphorylation. The full
   description and datasets for ∆Ps analysis were listed in Supplementary
   Table [341]S5.

   GO cellular component enrichment analysis through Gene Ontology
   Consortium and PANTHER classification system^[342]63. KEGG analysis was
   performed with KEGG Mapper ([343]https://www.genome.jp/kegg) after
   converting all candidates Uniprot accession to KEGG genes. All 21,997
   genes of Mus musculus in database were used as reference to determine
   the fold of enrichment. Fisher’s Exact with FDR multiple test
   correction was used to determine statistical significance. A protein
   will be considered as a synaptic protein when the Synaptic Ref Count ≥
   2 based on the synaptic protein database integrated from 11 previous
   proteomic studies^[344]24, gene names were updated according to the
   UniProt mouse protein database (April 17, 2022; 94,765). Cluster 3.056
   was used for hierarchical clustering (Centroid linkage with Euclidean
   distance). The full datasets for gene ontology analysis were listed in
   Supplementary Table [345]S4 and synaptic protein database were listed
   in Supplementary Table [346]S6.

Kinase motif-enrichment and correlation analysis

   Kinase enrichment was performed with previously reported tool
   ([347]https://kinase-library.phosphosite.org/ea)^[348]35. In detail,
   all detected phosphorylation site of each phosphopeptide were
   represented as site sequence with central position format, which means
   the phosphorylation site is at the central position of site sequence
   and the site sequence can include any of the 20 amino acids, ‘X’ for
   masking a position, and ‘_‘ for truncation. For foreground set
   definition, P value threshold was set to 0.05 for collected 20
   experiments, and more stringent Q value (0.2) for sleep related
   phosphoproteome data. To assure foreground set consist of at least 75%
   of significantly changed phosphopeptides, the log fold change threshold
   was different from each data due to log fold change range difference.
   Statistical significance of kinase enrichment analysis was determined
   using one-sided Fisher’s exact tests, and dominant P value was set to
   0.01. The full datasets of kinase motif-enrichment analysis were listed
   in Supplementary Table [349]S7.

   Functional Quadrant Enrichment Analysis (FQEA) was performed to every 2
   of the 25 comparisons. The kinase enrichment result of each comparison
   was filter with 0.01 on dominant P value. All overlapped kinases of 2
   comparisons were plotted on a cartesian coordinate system with dominate
   log[2]-transformed frequency factor value
   (dominant_enrichment_value_log[2]) as x or y coordinate. Then, the
   kinase numbers of 1st or 3rd quadrant (positive correlation direction)
   and 2nd or 4th quadrant (negative correlation direction) were used to
   compute P value of Chi-square test. Noticeably, the background kinase
   number of each quadrant was calculated by plotting 303 kinases of each
   comparison pairs on a coordinate. The FQEA enrichment factor of 2 kinds
   of correlation directions was the quotient of the corresponding kinase
   percentage of significantly changed set and background set. For each
   comparison pairs, the enriched directions were selected based on
   Chi-square P value (0.01) and FQEA enrichment factor.

PSSM search

   A set of reported phosphatase-binding peptides (PBP) for PP1
   (RVxF-binding pocket), PP2A (LxxIxE-binding pocket) and PP2B (LxvP and
   PxIxI-binding pocket) were collected^[350]47,[351]48, which consist of
   107 RVxF, 25 LxxIxE, 34 LxvP and 33 PxIxI motifs. A position-specific
   scoring matrix (PSSM) was constructed from the aligned sequences using
   the PSI BLAST IC scoring scheme with PSSMSearch tool. Then, each PSSM
   was searched against the mouse proteome respectively and the hits were
   filtered based on the PSSM score P value with a cut-off of 0.0001 and
   intrinsic disorder regions which IUPred score is less than 0.3.
   Applying these criteria, we produced 1903 putative phosphatase-binding
   motifs on 1575 proteins and performed following motif analysis on
   phosphoproteomics data with them. The full datasets of PSSM search were
   listed in Supplementary Table [352]S9.

Plasmids construction

   Mouse open reading frame (ORF) of sSLP (Sik3 A60-K527,
   [353]NM_027498.3) was synthesized from Comate Bioscience. SV40 (small T
   antigen, GenBank: [354]LT727623.1) and human LCMT1 (20-338,
   [355]NM_001032391.2) ORF were synthesized from Miaoling Biology. Mouse
   Ppp2ca and Ppp2r5d ORF was purchased from Miaoling Biology ([356]P34019
   and [357]P21643, respectively). Camk2a, mDlx, GfaABC1D promoters were
   from pAAV-CaMKIIa-hM4D(Gi)-mCherry (Addgene, 50477),
   pAAV-mDlx-GFP-Fishell-1 (Addgene, 83900) and pAAV-GfaABC1D-GRAB_Ado1.0
   (Addgene, 153288). A221 (sSLP T221A) mutant was constructed using
   QuickMutation Site-Directed Mutagenesis Kit (Beyotime, D0206). sSLP,
   A221, SV40, LCMT1, Ppp2ca and Ppp2r5d were subcloned into
   pAAV-hSyn-EGFP (Addgene, 50465) by replacing EGFP. Camk2a, mDlx, and
   GfaABC1D promoters were subcloned into pAAV-hSyn-sSLP or pAAV-hSyn-SV40
   by replacing hSyn.

Virus production and purification

   AAVs packaging and purification was performed as previously described
   with modifications^[358]46,[359]64,[360]65. AAVs were generated in the
   AAVpro 293 T cell line (Clontech, 632273) using polyethylenimine “Max”
   (PEI MAX, Polysciences) with pAAV-PHP.eB, pHelper and pAAV/hSyn-target
   gene expressing plasmids. The cell line was not tested for mycoplasma.
   The culture medium containing the transfection reagent was replaced
   with serum-free DMEM at 12–15 h post-transfection. Recombinant AAVs was
   collected at 6 days post-transfection. The cell and medium mixture were
   centrifuged at 17,940 g for 10 min at room temperature. Cell pellets
   were resuspended in 1 mL phosphate buffered saline (PBS), freeze-thawed
   for five times, and centrifuged at 17,940 g for 20 min. The supernatant
   was combined with cleared virus containing medium, and then filtered
   with 0.45 μM sterile filter. Viral media was concentrated using the
   Vivaspin 20 column (Cytiva) and formulated with PBS twice. The
   concentrated viral solution was stored at −80 °C until systemic
   administration. Virus titers were measured using a linearized genome
   plasmid as a standard. Intravenous administration of 100 μl AAV viral
   solution (1.0×10^12 vg/mL) was performed by injecting the virus into
   the retro-orbital sinus under deep anesthesia by isoflurane.

Brain slice preparation and electrophysiology

   For all electrophysiology experiments, brain slices were prepared from
   12 to 16 weeks old mice. The motor cortex slice preparation for
   electrophysiological recordings followed previous neocortical
   studies^[361]66,[362]67. Briefly, brains were rapidly removed after
   decapitation and placed into ice-cold (0–4 °C) artificial cerebrospinal
   fluid (ACSF) solution containing 119 mM NaCl, 2.5 mM KCl, 1 mM
   NaH[2]PO[4], 26 mM NaHCO[3], 1 mM MgCl[2], 25 mM glucose and 2 mM
   CaCl[2], pH 7.4 with NaOH, osmolarity 300–305 mOsm saturated with 95%
   O[2] and 5% CO[2]. The brain was then placed on a custom ramp with a 5°
   tilt angle and slices (400 μM) were cut with a vibratome (VT1200S,
   Leica Biosystems) coronally to retain axonal and dendritic
   arborizations of neurons as intact as possible. The brain slices were
   kept at 37.0 °C in oxygenated ACSF solution for 20 min, then
   transferred to the recording chamber.

   All recordings were obtained from the layer 5 of the motor cortex
   pyramidal neurons. Recording pipettes were pulled from borosilicate to
   resistances of 4–7 MΩ with a Sutter Instruments P-97. Micropipettes
   were filled with an intracellular solution containing the following:
   120 mM potassium gluconate, 10 mM HEPES, 4 mM KCl, 4 mM MgATP, 0.3 mM
   Na[3]GTP, 10 mM sodium phosphocreatine (pH 7.25), osmolarity
   300 mOsm/kg. Motorized micromanipulators (Junior Multpatch, Lugis &
   Neumann) were used to steer pipette electrodes to target cells. Whole
   cell recordings were made with Model 2400 amplifiers (A-M Systems).
   Electrophysiological data acquisition and stimulation command sending
   were performed by synchronized LIH 8 + 8 data acquisition interface
   boards (HEKA Instruments), which were operated by a custom-written Igor
   Pro 6 program.

Immunoblotting

   Western blotting was performed according to standard procedures using
   the corresponding antibodies. Antibodies were used at the optimal
   concentration according to the manufacturer’s instructions. Antibodies
   used in this study included anti-phospho-AMPK Substrate Motif
   (LXRXX(S*/T*)) (#5759, Cell Signaling), anti-phospho-PKC substrate
   motif ((K/R)XS*X(K/R)) (#6967, Cell Signaling), anti-phospho-ATM/ATR
   substrate motif (S*Q) (#9607, Cell Signaling) and anti-HA (C29F4)
   (#3724, Cell Signaling), p-HDAC4 S246 (#ab240643, Abcam), p-KCC2 S940
   (#612-401-E15, Rockland), p-nNOS S1417 (#ab5583, Abcam), p-eEF2 T56/T58
   (#ab82981, Abcam) and p-PRKAR1A S77 (#ab139682, Abcam).

Statistical analysis

   Unless otherwise noted, all experimental subjects were biological
   replicates and at least two independent experiments were performed.
   Mice were randomly assigned to groups. Researchers were blinded to the
   treatments during EEG/EMG data manual correction. No statistical
   methods were used to predetermine sample size. GraphPad Prism, EXCEL or
   Python software was used for statistical test analysis. Following
   one-way or two-way analysis of variance (ANOVA), Tukey’s test compares
   every mean with every other mean; Dunnett’s test compares every mean to
   a control mean; Sidak’s test compares a set of means. Repeated measures
   (RM) or paired tests were performed for matched subject comparisons.
   The complete sample size, statistical test method and results for each
   comparison are reported in each figure legend and fully described in
   Supplementary Table [363]S10.

Supplementary information

   [364]Supplementary information^ (5.8MB, pdf)
   [365]Dataset 1^ (20.2KB, xlsx)
   [366]Dataset 2^ (17.9MB, xlsx)
   [367]Dataset 3^ (4.2MB, xlsx)
   [368]Dataset 4^ (94.1KB, xlsx)
   [369]Dataset 5^ (782.3KB, xlsx)
   [370]Dataset 6^ (281.5KB, xlsx)
   [371]Dataset 7^ (1.1MB, xlsx)
   [372]Dataset 8^ (5MB, xlsx)
   [373]Dataset 9^ (243.3KB, xlsx)
   [374]Dataset 10^ (49.4KB, xlsx)

Acknowledgements