Abstract

   Sleep loss is a key trigger for a manic episode of bipolar disorder
   (BD), but the underlying microglial and molecular mechanisms remain
   unclear. Sleep loss induces microglial and inflammatory responses.
   Microglia, resident macrophages in the central nervous system, regulate
   synaptic pruning by engulfing dendritic spines. Here, we introduce a
   modified paradoxical sleep deprivation (SD) paradigm as a BD mouse
   model. After intermittent 16-h daily SD for 4 days, the mice showed
   mania-like behavior, reduced cytokine/chemokine production,
   mitochondrial damage, microglial loss, decreased synaptic engulfment by
   microglia, and synaptic gain. Single-nucleus RNA sequencing (snRNA-seq)
   revealed cell-type-specific inflammation- and synapse-related gene
   expression profiles in the prefrontal cortex (PFC) and hippocampus of
   SD-treated male mice. Interestingly, much more differentially expressed
   genes were observed in SD-treated female versus male mouse brain,
   especially in the PFC. Pharmacological depletion of microglia by colony
   stimulating factor-1 receptor (CSF1R) inhibitor PLX3397 blocked
   SD-induced inflammation-related and senescence-associated abnormalities
   in a sex-specific manner. Microglial elimination reversed SD-induced
   synapse gain and mania-like behavior in males but not in females.
   However, microglial inhibition by minocycline had no effect on
   SD-induced behaviors in a sex-independent manner. These findings
   demonstrate that microglia-mediated neuroinflammation and synaptic
   pruning contribute to SD-induced mania-like behavior in a mouse model
   of BD in a sex-specific manner.

   graphic file with name 41398_2025_3525_Figa_HTML.jpg

   Subject terms: Molecular neuroscience, Bipolar disorder

Introduction

   People with insomnia or sleep loss often experience mental issues with
   the increasing pressure of study, work, and life [[40]1, [41]2]. Sleep
   loss can trigger a mood switch from depression to mania in bipolar
   disorder (BD) [[42]3]. BD significantly affects patients’ quality of
   work, study, and life and increases individual’s social burdens
   [[43]4]. Patients with BD often experience sleep disturbances. During
   manic episodes, most BD patients (69–99%) experience a decrease in
   sleep demand [[44]3]. Transcriptome sequencing data from peripheral
   blood samples from BD patients indicate that sleep deprivation (SD)
   causes a dysregulated inflammatory response [[45]5] and an increase in
   the levels of inflammatory factors [[46]6]. In addition, abnormal
   expression of inflammation-related genes has been found in
   microglia-like cells differentiated from monocytes derived from
   peripheral blood of BD donors [[47]7].

   A common BD animal model of rapid eye movement (REM) SD is established
   by a modified multi-platform water method, which can induce manic-like
   behavior in rodents [[48]8, [49]9]. Four days of continuous and
   prolonged SD exposure triggers cytokine-storm-like systemic
   inflammation using a “curling prevention by water” paradigm [[50]10].
   Acute 48-h SD leads to an increase in the density of microglia (the
   resident innate immune cells of the central nervous system (CNS)) in
   brain regions in young mice [[51]11]. Microglia regulate synaptic
   density by engulfing dendritic spines [[52]12]. Abnormal synaptic
   transmission and synaptic pruning are observed in sleep-deprived mice
   [[53]13, [54]14]. Sleep fragmentation causes an increase in colony
   stimulating factor-1 (CSF1) levels [[55]15]. CSF1 receptor (CSF1R) in
   the brain is mainly expressed in microglia and regulates the survival
   and distribution of microglia [[56]16]. CSF1R inhibitor reduces
   microglia and complement C1qb, avoiding neuronal damage and synaptic
   pruning abnormalities [[57]17]. However, the detailed mechanisms
   underlying microglia-mediated inflammation and synaptic pruning in BD
   have not been elucidated. Although some studies have suggested
   sex-dependent effects of SD on emotional behaviors [[58]3, [59]18],
   sex-dependent gene expression levels in the brain and treatment
   response in BD remain unresolved.

   Here, we used a modified SD paradigm as a BD model to investigate
   mania-like behavior in mice. As a consequence of intermittent 16-h
   daily SD for 4 days, mice exhibited inflammation- and synapse-related
   gene expression profiles using transcriptome sequencing. Chronic SD
   exposure reduced cytokine/chemokine levels in the plasma with a
   multicytokine assay. Additionally, single-nucleus RNA sequencing
   (snRNA-seq) revealed cell-type-specific changes. Microglial loss and
   synapse gain were found after SD using immunohistochemistry, Golgi
   staining, and transmission electron microscopy. Finally,
   pharmacological depletion of microglia by CSF1R inhibitor inhibited
   SD-induced mania-like behavior in a sex-specific manner and reversed
   SD-induced synapse gain in male mice.

Methods and materials

Animals

   Adult male and female C57BL/6 mice (2 months old) were obtained from
   GemPharmatech (Chengdu, China). CX3CR1^CreERT2 (JAX#020940) and Ai9
   (JAX#007909) mice were obtained from the Jackson Laboratory.
   CX3CR1^CreERT2:: Ai9 mice were used for this experiment.
   CX3CR1^CreERT2:: Ai9 mice were intraperitoneally injected with
   tamoxifen (100 mg/kg) once daily for 4 days at five to six weeks old to
   induce microglia to constantly express tdTomato. The animals were
   housed in animal facilities with food and water available ad libitum
   and were maintained on a normal 12-h light/dark cycle (lights on at
   7:00 AM, lights off at 7:00 PM).

Ethics approval and consent to participate

   All procedures were performed in accordance with the Institutional
   Animal Care and Use Committee of Sichuan University (Approval Number:
   20230504005). All experimental procedures were in compliance with the
   ARRIVE guidelines. The guidelines used in the present study were made
   to minimize animal suffering as well as the number of animals used.

Drug administration

   To deplete microglia in the brain, mice were injected with the CSF1R
   inhibitor PLX3397 (pexidartinib, PLX, MedChemExpress, Cat# HY-16749,
   USA) dissolved in 5% dimethyl sulfoxide (DMSO), 10%
   sulfobutylether-β-cyclodextrin (SBE-β-CD) based on previous studies
   [[60]19]. PLX was administered (i.p. injection) at a dose of 50 mg/kg
   twice a day for 8 days (4-day treatment with PLX, 3-day withdrawal, and
   4-day treatment with PLX). The control mice received vehicle solution
   (5% DMSO/10% SBE-β-CD/85% saline, vol/vol/vol). Minocycline (MCL,
   MedChemExpress, Cat# HY-17412) was administered orally in drinking
   water at 0.5 mg/mL (delivering daily doses of approximately 100 mg/kg)
   for 14 days based on previous studies [[61]20]. Mice were
   intraperitoneally treated with valproic acid (VPA) at a dose of
   200 mg/kg twice a day for 7 days. Administration of PLX, MCL, or VPA
   was stopped 12 h before behavioral testing. To efficiently induce the
   CreER-dependent recombination, tamoxifen (100 mg/kg, MedChemExpress,
   Cat# HY-13757A) dissolved in corn oil was intraperitoneally
   administered for 4 consecutive days as previously described [[62]21].

Sleep deprivation procedure

   All SD-treated mice were first habituated to environmental conditions
   and submitted to rapid eye movement (REM) sleep deprivation stress in a
   Plexiglas box for 16 h daily (starting time: 6:00 p.m., ending time:
   10:00 a.m. the next day) for 4 days. The SD period of 6 p.m. to 10 a.m.
   was adapted based on previous studies [[63]22–[64]24]. Paradoxical SD
   stress was conducted using a modified multiple-platform method,
   according to the protocol described [[65]25]. Briefly, the Plexiglas
   box contained seven visible escape platforms (3 cm in diameter, 1 cm
   above the water surface) surrounded by water. Mice were allowed to
   freely move on the platform with food and water available ad libitum.
   The control mice were kept in their home cages in the same experimental
   room.

Behavioral analysis

   All behavioral measurements were conducted by trained experimenters in
   a blinded manner. All mice were habituated to the test room for at
   least 30 min prior to behavioral testing. Behavioral testing was
   conducted under dim light between 9:00 a.m. and 05:00 p.m. according to
   the protocol described [[66]20, [67]26]. Mice were tested successively
   in the open-field test (OFT) and elevated plus-maze test (EPM). The
   recorded videos were analyzed using EthoVision (12.0) tracking software
   (Noldus, Netherlands). One day after behavioral testing (from
   01:00 p.m. to 05:00 p.m. of the next day), the mice were randomly
   selected and sacrificed.

   In tail suspension test (TST), the mice were suspended by their tails
   using an elastic band based on a previously described procedure
   [[68]27]. A video recording device captured the behavior of each mouse
   throughout the 6-min testing period. Immobility time, defined as the
   time with no body movement, was quantified during the last 4 min of the
   6-min test session.

   In the novel object recognition test (NOR), the test was conducted in
   an open field arena (50 × 50 × 50 cm) based on a previously described
   procedure [[69]28]. During the familiarization phase, mice were allowed
   to explore two identical objects positioned 20 cm apart for 5 min. Five
   hours later, one of the objects was replaced with a novel object during
   the test phase, and mice were allowed to explore both objects for
   5 min. The time spent exploring each object was recorded, and an object
   discrimination index (ODI) was calculated as the time spent exploring
   the novel object divided by the total exploration time.

   In the sociability test (ST), the test involved two phases based on a
   previously described procedure [[70]29, [71]30]. First, mice were
   habituated to the open field (50 × 50 × 50 cm) and empty enclosures for
   5 min. Next, during the sociability phase, a novel male conspecific
   (social object) was placed in one enclosure, while a novel object
   (non-social object) was placed in the other. The experimental mouse was
   then allowed to explore the open field for 5 min, and the time spent in
   each quadrant (social vs. non-social) was recorded. The social
   discrimination index (SDI) was calculated as (time spent with social
   object) / (total exploration time).

Tissue preparation, immunohistochemistry, and imaging

   Twenty-four hours after behavioral testing, mice under anesthesia were
   transcardially perfused with PBS, followed by 4% paraformaldehyde
   (PFA), and the brains were removed, post-fixed in 4% PFA for 24 h.
   Tissues were dehydrated with 30% sucrose and sectioned at 40 μm on a
   Leica microtome (Leica CM3050S, Germany).

   Immunostaining with a rabbit anti-ionized calcium-binding adaptor
   molecule 1 primary antibody (Iba1; #019–19741, Japan, RRID: AB_839504)
   and the specificity of this antibody has been previously validated in
   rodents [[72]31, [73]32]. A series of prefrontal and hippocampal
   sections was collected from experimental mice (at 240 µm intervals).
   These brain sections were processed for immunohistochemical detection
   of Iba1 (microglial marker), as described previously [[74]11]. Briefly,
   the free-floating sections were treated with 0.3% hydrogen peroxide
   (H[2]O[2]) for 30 min to quench endogenous peroxidase activity,
   incubated with 5% normal goat serum (Solarbio, China) for 30 min to
   block nonspecific binding, and immunoreacted with a rabbit anti-Iba1
   antibody (Wako, #019-19741, at 1:1000 dilution, RRID: AB_839504) at
   4 °C overnight. The free-floating sections were performed with
   biotinylated goat anti-rabbit immunoglobulin G (IgG) (at 1:200
   dilution, Cat# BA-1400, RRID: AB_2336187, Vector Laboratories, USA) for
   1 h, avidin-biotin peroxidase complex (at 1:200 dilution, Cat# PK-6100,
   RRID: AB_2336819, Vector Laboratories, USA) for 1 h, and 0.05%
   diaminobenzidine (Sigma-Aldrich) containing 0.03% H[2]O[2] and 0.25%
   nickel ammonium sulfate (Sigma-Aldrich) for 10 min. All sections were
   mounted on gelatin-coated glass, dehydrated in graded ethanol, treated
   with TO-type biological transparentizing agent, coverslipped using
   neutral quick drying glue, and scanned using a digital whole-slide
   scanner (NanoZoomer-XR, Japan). All digital images of brain sections
   were adjusted for cropping, brightness/contrast, and image size using
   Adobe Photoshop 2020 (Adobe Systems, USA). The image files were
   processed in the .tif format. The quantification of microglial density
   and morphology was performed using NIH Image J software as described in
   our previous study [[75]11].

   For immunofluorescence analysis, three sections per sample from the
   hippocampus of CX3CR1^CreERT2:: Ai9 mice were incubated overnight at
   4 °C with the appropriate antibodies (rabbit anti-PSD95 antibody,
   1:500, ab18258, RRID: AB_444362, Abcam; rat anti-CD68 antibody, 1:500,
   MCA1957, RRID: AB_322219, Bio-Rad). Binding sites were visualized using
   Alexa Fluor-conjugated secondary antibodies at 50 μg/mL for 2 h at room
   temperature (Alexa Fluor® 488-AffiniPure donkey anti-rabbit IgG,
   711-545-152, RRID: AB_2313584; Alexa Fluor® 488 AffiniPure goat
   anti-rat IgG, 112-545-003, RRID: AB_2338351; Jackson ImmunoResearch
   Inc). Nuclei were counterstained with Hoechst 33,342 (1:2000; Cat#
   4082, RRID: AB_10626776, Cell Signaling Technology) during incubation
   with the secondary antibody. The slides were coverslipped with
   Citiflour® mounting medium. The fluorescent signals were observed using
   an Olympus VS200 slide scanner and a Spinning-Disc confocal microscopy
   (Nikon/EclipsTi2-E with Andor Dragonfly 200, Nikon) with a
   60 × oil-immersion objective. The imaging parameters (laser power,
   gain, offset, resolution, and steps in the z direction) were consistent
   across all slides.

Three-dimensional (3D) reconstruction

   All images were analyzed using IMARIS 10 software (Bitplane). Three
   images randomly picked from each mouse were reconstructed.
   PSD95 + puncta were reconstructed using the “Spots” function in IMARIS.
   CX3CR1 + microglia (tdTomato) were accurately reconstructed using the
   “Surface” function in IMARIS. The “Split into Surface Objects” function
   in IMARIS was used to assess the number of PSD95 + puncta entirely
   within the microglial surface. Each mouse had at least eight cells
   (n = 5–6 mice per group). The mean result was used for morphological
   analysis.

Transcriptome sequencing and data analysis

   The brains used for RNA sequencing (RNA-seq) were removed from
   euthanized mice within 1–2 h after sleep deprivation, and the
   prefrontal cortex was excised. Samples of the prefrontal cortex from
   eight control and SD-treated mice were subjected to RNA extraction, RNA
   quantification, cDNA library preparation, and library quality
   assessment according to the protocols provided by Novogene (Novogene
   Co., Ltd., Beijing, China). Libraries were prepared for Illumina
   sequencing using the Illumina Nova-seq platform at Novogene. All the
   downstream analyses were based on the high-quality clean Unique
   Molecular Identifiers reads. Furthermore, we performed differential
   expression analysis of two groups using the DESeq R package and
   adjusted P-values using Benjamini and Hochberg’s approach for
   controlling the false discovery rate. We performed Gene Ontology (GO)
   enrichment analysis of differentially expressed genes (DEGs) by the
   GOseq R package with an adjusted P-value < 0.05 and Kyoto Encyclopedia
   of Genes and Genomes (KEGG) pathway enrichment analysis using the DEGs
   to identify functional pathways ([76]http://www.genome.jp/kegg/).

Nuclei isolation and snRNA-seq on the 10x genomics platform

   The brains used for snRNA-seq were removed from euthanized mice within
   1–2 h after SD, and the prefrontal cortex (PFC) and hippocampus (HIP)
   were excised and pooled into a single sample. Samples from the six
   control and SD groups were lysed in lysis buffer and homogenized with a
   glass Dounce homogenizer. Nuclei extraction and library preparation
   were processed according to the protocols provided by Novogene
   (Novogene Co., Ltd., Beijing, China). Single-nucleus cells were
   manually counted on Luna FL using an AO/PI cell viability kit (Logos
   Biosystems) after each centrifugation and resuspension. Single-nucleus
   cell suspensions were loaded into chromium microfluidic chips with a
   10 × Chromium Controller (10X Genomics). RNA from the barcoded cells
   was subsequently reverse-transcribed and sequencing libraries were
   constructed with reagents from a Chromium Single Cell 3’v2 reagent kit
   (10X Genomics) according to the manufacturer’s instructions.
   Single-nucleus RNA sequencing was performed with an Illumina platform
   (Illumina NovaSeq 6000) according to the manufacturer’s instructions
   (Illumina).

snRNA-seq data processing and filtering

   Raw read sequences produced by the Illumina pipeline in fastq format
   were preprocessed through Trimmomatic software as follows: (1)
   low-quality reads were removed; (2) trailing low-quality or N bases
   were removed; (3) adapters were removed; (4) reads below 26 bases long
   were dropped; and (5) reads that could not form pairs were discarded.
   Downstream analyses were performed on the filtered data reanalyzed with
   Cell Ranger (version 7.0)
   ([77]https://github.com/10XGenomics/cellranger) and R package Seurat
   (version 4.3.0)
   ([78]https://satijalab.org/seurat/articles/pbmc3k_tutorial.html).
   Briefly, resulting fastq files were aligned to the mouse genome (mm10
   1.2.0) using Bowtie (version 1.1.2). Unique molecule identifiers were
   counted for individual genes and cells (filtered by CellRanger) to
   construct digital expression matrices. The Seurat package filtered the
   data to screen for genes expressed in more than three cells, as well as
   for cells expressing more than 200 genes.

Dimensionality reduction, clustering, and UMAP

   Integration of multiple sample datasets was performed using the
   canonical correlation analysis (CCA) algorithm within Seurat, followed
   by data scaling via the ScaleData function. Principal component
   analysis (PCA) was conducted on the top 2000 highly variable genes to
   reduce linear dimensionality, with the first 20 principal components
   used in subsequent analysis. Cell clusters were identified using the
   FindClusters function in Seurat with a resolution = 0.5, and
   visualization was achieved through uniform manifold approximation and
   projection (UMAP). Cluster-enriched genes were identified using the
   FindAllMarkers function, with the settings “min.pct = 0.25”.

Functional enrichment analysis, single-cell pseudotime analysis, and
cell‒cell interaction analysis

   Gene functional enrichment analysis was conducted using the
   clusterProfiler package, with the following parameters: ont = BP,
   pAdjustMethod = BH, pvalueCutoff = 0.01, and qvalueCutoff = 0.05.

   Pseudotime analysis was performed using the R package Monocle 2, with
   the starting point determined by CytoTRACE. High dispersion genes
   across cells were utilized to select and arrange cells of interest,
   followed by division into two components using the discriminative
   dimensionality reduction via learning a tree (DDRTree) approach.
   Branch-dependent genes were identified using branched expression
   analysis modeling (BEAM).

   Cell‒cell interactions were assessed using CellChat (version 2.1.0) to
   infer and visualize intercellular communication based on known
   ligand‒receptor pairs. All categories of ligand-receptor interactions
   in the database were utilized, and differences in communication
   probabilities were visualized using the netVisual_bubble function in
   the CellChat R package.

Quantitative real-time polymerase chain reaction (qRT‒PCR)

   At 1–2 h after SD, the prefrontal cortex and hippocampus were quickly
   removed and excised from euthanized mice. The specific sequences of
   primers used for qRT‒PCR are listed in Extended Data Table [79]1 (36
   inflammation-related, 12 senescence-associated, and 11 synapse-related
   candidate genes) and were designed according to reference sequences in
   PrimerBank or the NCBI database with Primer-BLAST. The primer sequences
   were synthesized by Sangon Biotech (Shanghai, China). Total RNA was
   isolated from the PFC and HIP using TRIzol reagent and
   reverse-transcribed into cDNA using a HiScript III All-In-One RT
   SuperMix Perfect for qPCR Kit (Cat# R333, Vazyme, China). qRT‒PCR was
   performed using the QuantStudio™ 3 Applied Biosystem Real-Time PCR
   System and PowerUp SYBR Green Master Mix (Cat. No. A25742; Applied
   Biosystems, USA). The relative mRNA expression levels of selected genes
   were calculated with the 2^−ΔΔCt method [[80]33]. The relative mRNA
   expression levels were normalized to the expression level of GAPDH.
   Each sample was tested in duplicate. The experimental group values were
   normalized to those of the control group.

Multicytokine assay

   At 24 h after SD, the plasma from each treatment and control group was
   collected for a multicytokine assay. Cytokine and chemokine levels were
   evaluated with a Luminex X-200 system (R&D Systems, USA) using a panel
   of 31 mouse cytokines (LX-MultiDTM-31, R&D Systems, USA) in accordance
   with the manufacturer’s instructions. The selected cytokines included
   CXCL13, CCL27, CXCL5, CCL11, CCL24, CX3CL1, GM-CSF, CCL1, IFN-γ, IL-1β,
   IL-2, IL-4, IL-6, IL-10, IL-16, CXCL10, CXCL11, CXCL1, CCL2, CCL7,
   CCL12, CCL22, CCL3, CCL4, CCL20, CCL19, CCL5, CXCL16, CXCL12, CCL17,
   and TNF-α. Data acquisition and analysis were conducted using Luminex
   xPONENT software (Luminex Corporation). The experimental group values
   were normalized to those of the control group.

Electron microscopy

   Synaptic density and mitochondrial damage were determined by electron
   microscopy as described previously [[81]34]. At 1–2 h after SD, the
   mice were deeply anesthetized and transcardially perfused with 4% PFA.
   Hippocampal slices were post-fixed in cold 1% OsO4 for 2 h. Hippocampal
   samples were prepared for electron microscopy using standard
   procedures. Ultrathin sections (70 nm, Ultra 45°, Daitome) were stained
   with uranyl acetate/lead acetate. Synaptic density and mitochondrial
   damage were viewed by transmission electron microscopy (TEM, HT7700,
   accelerating voltage 120 kV). Seven mice were examined in the two
   groups. Synapses were identified by the presence of synaptic vesicles
   and postsynaptic densities. Each mitochondrion was classified as either
   intact (the mitochondrial membranes and internal cristae structures
   were clear and complete) or damaged (loss of the outer membrane or loss
   of the inner cristae structure).

Golgi staining

   At 1–2 h after SD, the mice were deeply anesthetized and transcardially
   perfused with 4% PFA. Golgi staining was performed using the FD Rapid
   GolgiStain™ Kit according to the manufacturer’s recommended protocol
   (PK401, FD NeuroTechnologies). Briefly, hippocampal samples were
   immersed in impregnation solution made by mixing equal volumes of
   solutions A and B for 3 weeks and then transferred to solution C for 5
   days at room temperature in the dark. The hippocampus was sliced using
   a Leica microtome (Leica CM3050S, Germany) at a thickness of 100 μm.
   Each section was rinsed and then placed in staining solution (solution
   D/solution E/double-distilled water = 1:1:2) for 10 min. The sections
   were dehydrated and cover-slipped. Dendritic spines were examined using
   a Nikon ECLIPSE Ni-E microscope (60 × oil objective). The qualitative
   and morphometric properties of the neurons were investigated as
   described previously [[82]35].

Statistical analysis

   Randomization was used to assign animals to experimental groups.
   Furthermore, samples for histopathological and bioinformatics analyses
   were randomly selected from each group. For the histology experiments,
   the damaged brain sections were excluded from the quantitative
   analysis. Drowning mice in the SD model were excluded. All
   histopathological analyses were conducted by trained personnel in a
   blinded manner. Statistical analysis was analyzed using SPSS 25
   software (SPSS, Inc., Chicago, IL, USA). All results are expressed as
   the mean ± standard error of the mean (SEM). For normally distributed
   data, differences between group means were tested using Student’s
   t-test when the experiment contained two groups or one-way analysis of
   variance (ANOVA) followed by Tukey’s post hoc test and two-way ANOVA
   followed by Bonferroni post hoc correction when the experiment
   contained more than two groups. To ensure 80% power to detect a 25%
   difference between groups (Cohen’s d, 5% Type I error rate), a sample
   size of at least eight per group was selected. Significance was set at
   P < 0.05. The data were plotted using GraphPad Prism (GraphPad
   Software, San Diego, CA, USA).

Results

Intermittent SD induces mania-like behavior

   In this study, we provided a new paradigm of intermittent 16-h daily SD
   (from 6 P.M. to 10 A.M. the next day) for 4 days, to mimic sleep loss
   as the result of multiple sleep pressures (Fig. [83]1a, b).
   Intermittent SD in male mice induced significant weight loss, an
   increase in the total distance traveled, and an increase in the
   frequency of entries into the central area in the OFT (Fig. [84]1c–f).
   In the EPM test, SD also significantly increased the total distance
   moved and the frequency of entries into the open arms (Fig. [85]1g–i).
   VPA, a common mood stabilizer for the treatment of bipolar disorder
   [[86]36], reversed these mania-like behavioral alterations in both male
   and female mice after 7 days of VPA pretreatment (Fig. [87]1j–l and
   Extended Data Fig. [88]1). Furthermore, we incorporated the TST and the
   ST to further validate the efficacy of the sleep deprivation-induced BD
   model (Extended Data Fig. [89]2). The TST results showed a significant
   reduction in immobility time in the SD group (Extended Data Fig.
   [90]2a), indicating a marked antidepressant effect (mania-like
   behavior). Furthermore, the NOR test revealed that both the control and
   SD groups exhibited a significant increase in exploration time for the
   novel object (Extended Data Fig. [91]2b). Importantly, there was no
   significant difference in the ODI between the two groups (Extended Data
   Fig. [92]2c), suggesting that SD did not impair novel object
   recognition ability in mice. In the sociability test, the results
   demonstrated that SD mice displayed a significantly increased
   exploration time towards the social mice compared to the novel object
   (Extended Data Fig. [93]2d). Moreover, the SDI was significantly higher
   in the SD group, indicating enhanced social interaction (Extended Data
   Fig. [94]2e).

Fig. 1. Paradoxical sleep deprivation (SD) induces mania-like behavior and
transcriptomic changes, and treatment with valproate (VPA) reverses these
behavioral alterations in male mice.

   [95]Fig. 1
   [96]Open in a new tab

   a, j Schematic of the experimental procedures. b Schematic of the SD
   apparatus. c Body weights of control (Con) and SD mice. d
   Representative traces of movement in the open field test (OFT) of
   control- and SD-treated mice. e, f Quantification of total distance
   moved and frequency of entry into the central area of the open field
   after intermittent 16-h SD for 4 days. g Representative traces of
   movement from the elevated plus-maze test (EPM). h, i Quantification of
   the total distance moved and duration in the open arms. k, l
   Quantification of the total distance moved during the 5-min OFT and
   total cumulative duration of immobility (velocity < 1.75 cm/s). m
   Volcano plot of differentially expressed genes (DEGs) in the prefrontal
   cortex between control and SD mice (n = 4, each group). Red dots
   represent genes with upregulated expression, green dots represent genes
   with downregulated expression, and blue dots represent non-DEGs. The
   labels are given as gene symbols for the top gene by unadjusted P value
   rank. n The top significantly enriched Gene Ontology (GO) categories of
   biological process, cellular component and molecular function between
   the two groups. The X-axis indicates the value of -log10 (P-value). The
   Y-axis indicates GO terms. o The scatter plots show the top 30 enriched
   KEGG pathways between the two groups. The Y-axis indicates KEGG
   pathways. The X-axis indicates the gene ratio. p Validation of DEGs
   expression by quantitative real-time polymerase chain reaction
   (n = 5–6, each group). *P < 0.05, **P < 0.01, ***P < 0.001 (n = 9–12
   each group for behavioral test). Statistical analysis was performed
   using unpaired t test a–i, p or two-way ANOVA followed by Bonferroni
   post hoc correction k, l.

Transcriptome sequencing reveals inflammation- and synapse-related gene
expression profiles in SD-induced BD model

   We first tested for differential gene expression in the PFC between
   control and SD-treated male mice using RNA-seq (Fig. [97]1 and Extended
   Data Fig. [98]3). DEGs (1031 upregulated genes and 782 downregulated
   genes) were identified by RNA-Seq and are shown in a volcano plot (Fig.
   [99]1m). Gene Ontology enrichment analysis revealed that both
   inflammation- and synapse-related DEGs were enriched including the MHC
   class I peptide loading complex, the MHC protein complex, T-cell
   receptor binding, CD8 receptor binding, negative regulation of immune
   system process, synaptic membrane, postsynaptic density membrane,
   dendritic spine, and dendrite development (Fig. [100]1n). In addition,
   KEGG analysis from the RNA-seq data further revealed enrichment of DEGs
   associated with immune responses (e.g., IL-17 signaling pathway, TNF
   signaling pathway, T-cell leukemia virus 1 infection, Epstein‒Barr
   virus infection, and autoimmune thyroid disease) and synaptic
   transmission (e.g., axon guidance and adherens junction) (Fig.
   [101]1o). To validate the RNA-seq data, we confirmed the expression of
   several randomly chosen DEGs by qRT‒PCR (Fig. [102]1p).

SD reduces the cytokine/chemokine levels in the plasma, expression of
microglial marker genes, and microglia density in the brain

   The SD-treated mice showed significantly lower levels of cytokine and
   chemokine levels in the plasma compared with the control group by a
   Luminex-based multicytokine assay (Fig. [103]2a). To examine the
   changes in microglia (CNS-resident immune cells) after chronic SD
   exposure, we performed qRT‒PCR to detect CD11b, CSF1R, and TREM2
   (microglial marker genes) mRNA levels and immunohistochemistry to
   analyze microglial morphology in the brain of male mice (Fig.
   [104]2b–k). We found that the gene expression levels of these
   microglial genes were significantly lower in the PFC and HIP SD-treated
   mice than in the untreated controls (Fig. [105]2b, c). Furthermore,
   intermittent SD stress significantly reduced the density of Iba1-ir
   cells and the microglial soma area in the PFC and HIP (Fig. [106]2da–f,
   i, j) and increased microglial total process length and the number of
   branching points only in the PFC (Fig. [107]2g, h, k).

Fig. 2. Effects of SD on microglial marker gene expression and morphology in
the brain.

   [108]Fig. 2
   [109]Open in a new tab

   a The levels of 31 cytokines and chemokines in the plasma of the two
   groups with a Luminex liquid chip (red, upregulated; blue,
   downregulated; n = 5, each group). b, c The expression of CD11b
   (microglial marker), CSF1R (microglial marker), and Ki67 in the PFC and
   hippocampus (HIP). d Representative high-magnification photomicrographs
   showing Iba1 immunoreactivity in the PFC of control and SD-treated
   mice. e–k Changes in microglial density and morphology in the PFC and
   HIP of the two groups. *P < 0.05, **P < 0.01, ***P < 0.001. Scale
   bars = 50 μm in d. Statistical analysis was performed using unpaired t
   test or Mann-Whitney test.

Single-nucleus RNA sequencing reveals cell-type-specific changes in
SD-induced BD model

   To further identify the cell populations potentially involved in
   SD-induced mania-like behavior, we performed snRNA-seq of approximately
   84,124 nuclei isolated from the prefrontal and hippocampal tissues
   after 16-h daily SD for 4 days (Fig. [110]3a). After raw data quality
   control, alignment and UMI filtering, 75,201 nuclei were retained for
   further analysis (Extended Data Table [111]2 and Extended Data Fig.
   [112]4). As a result, we identified 29 distinct clusters through an
   unsupervised graph-based clustering algorithm (Fig. [113]3b). We next
   confirmed the cell types identified above and annotated the main cell
   types by examining cell marker gene expression reported in previous
   studies [[114]37, [115]38]: microglia (n = 4463; PTPRC, APBB1IP, DOCK8,
   MERTK, CSF1R), excitatory neurons (n = 41,826; marked by CUX2, TLE4,
   RORB, SLC17A7), interneurons (n = 11,604; SST, GAD1, GAD2, SV2C),
   astrocytes (n = 5358; SLC1A2, SLC1A3, GFAP, GLI3, AQP4),
   oligodendrocytes (n = 9027; MBP, PLP1, MOBP), and oligodendrocyte
   progenitor cells (n = 1982; VCAN, LHFPL3, PDGFRA, PCDH15) (Fig.
   [116]3c, d). Moreover, SD stress did not result in an imbalance of
   glial/neuronal ratio or distribution of cell type-specific clusters
   between the two groups.

Fig. 3. Effects of SD on the single-cell transcriptome in the prefrontal and
hippocampal tissues of male mice by single-nucleus RNA sequencing.

   [117]Fig. 3
   [118]Open in a new tab

   a Schematic of the experimental design and analysis. b, c tSNE
   visualization showing clustering of 75,201 nuclei by unsupervised
   cluster (b) and annotated cell types (c). d Bubble plot showing the
   expression levels, distribution of representative cell-type markers and
   numbers annotated by cell types. e Numbers of up- and downregulated
   DEGs in each cell type. f DEGs involved in pathways in each cell type.
   g Representative tSNE depicting clustering of microglial nuclei from
   the brains of control (n = 3) versus SD (n = 3). Ast astrocyte, Exn
   excitatory neuron, Int interneuron, Mic microglia, Oli oligodendrocyte,
   OPC oligodendrocyte precursors.

   Next, we compared cell type-specific gene expression differences
   between control and SD-treated mice, and identified 4181 DEGs across
   six cell types, including 3172 upregulated and 1009 downregulated genes
   (Fig. [119]3e). More upregulated DEGs than downregulated DEGs were
   found in excitatory neurons, interneurons, oligodendrocytes, and
   astrocytes but not in microglia or oligodendrocyte progenitor cells
   (Fig. [120]3e). To gain insight into the biological pathways associated
   with these DEGs, we performed KEGG analysis on DEGs of different cell
   types. The significantly enriched KEGG pathways showed enrichment of
   DEGs in microglia participating in ubiquitin-mediated proteolysis, the
   B-cell receptor signaling pathway, the mTOR signaling pathway,
   autophagy-animal, the T-cell receptor signaling pathway, the chemokine
   signaling pathway, the PI3K-Akt signaling pathway, and the TNF
   signaling pathway (Fig. [121]3f). Additionally, the DEGs in excitatory
   neurons and interneurons were mostly implicated in axon guidance,
   dopaminergic synapse, the calcium signaling pathway, glutamatergic
   synapse, circadian entrainment, morphine addiction, cholinergic
   synapse, and GABAergic synapse (Fig. [122]3f). We performed
   unsupervised cluster to confirm the identity of microglia in control
   versus SD-treated mice (Fig. [123]3g).

Damaged mitochondria, synapse gain, and decreased synaptic engulfment by
microglia during SD

   To evaluate the consequences of SD on synaptic structures, we analyzed
   prefrontal and hippocampal sections from adult male mice using Golgi
   staining and transmission electron microscopy. TEM imaging of the HIP
   revealed that SD caused severe damage of mitochondria (Fig. [124]4a).
   Meanwhile, the results of electron microscopic analysis showed
   SD-induced increase in synaptic density and decrease in the lengths of
   active zones of synapses (Fig. [125]4b–d). Consistent with these
   findings, Golgi staining revealed a significant increase in the total
   spine density in the HIP of SD-treated male mice (Fig. [126]4e–g).
   Additionally, our data showed a relatively low percentage of engulfed
   PSD95 inside microglia after SD exposure (Fig. [127]4h, i).

Fig. 4. Effects of sleep deprivation (SD) on microglia, mitochondria and
synapses in the brains of male mice.

   [128]Fig. 4
   [129]Open in a new tab

   a The percentage of damaged mitochondria in the HIP of control and
   SD-treated male mice by electron microscopy. The green shadow indicates
   mitochondria. b Images of synapses in the HIP by electron microscopy.
   The yellow shadow indicates the presynaptic terminal. The red shadow
   indicates the postsynaptic spine. c, d The number of synapses and the
   lengths of the active zones of synapses (n = 3–4, each group). e, f
   Representative Golgi-Cox staining images of dendritic spines in the
   HIP. g Histograms showing spine densities of dendritic spines. h
   Representative images and 3D reconstruction rendering of microglia
   (red, tdTomato) containing PSD95 + puncta (green) in the HIP from
   CX3CR1^CreERT2:: Ai9 mice treated with SD. The white arrowheads show
   engulfed PSD95 + puncta inside microglia. i Quantification of the
   percentage of engulfed PSD95 puncta inside microglia (n = 5–6 per
   group). The data are shown as the mean ± SEM. The data represent three
   independent experiments. *P < 0.05, **P < 0.01. Scale bars = 200 μm in
   e; 10 μm in f; 5 μm in h. Statistical analysis was performed using
   unpaired t test.

Microglial depletion prevents SD-induced mania-like behavior in a
sex-specific manner

   Interestingly, microglial inhibition by MCL did not prevent SD-induced
   mania-like behavior in either male or female mice (Extended Data Fig.
   [130]5). To determine whether microglial depletion in the brain
   influenced SD-induced behavior, we investigated changes in the
   SD-induced BD model after PLX pretreatment at a dose of 50 mg/kg
   (efficiency of microglia depletion > 80% [[131]20]) (Fig. [132]5a). In
   the OFT and EPM tests, pharmacological depletion of microglia
   completely reversed SD-induced mania-like behavior in male mice (Fig.
   [133]5b, c, e–g). In contrast, microglial elimination had no
   significant effect on SD-induced mania in female mice (Fig. [134]5d, e,
   h, i).

Fig. 5. Effects of the colony-stimulating factor 1 receptor (CSF1R) inhibitor
PLX3397 (pexidartinib, PLX) on SD-induced mania-like behaviors and
inflammation-related gene expression changes in a sex-specific manner in
mice.

   [135]Fig. 5
   [136]Open in a new tab

   a Schematic of the experimental procedures. b, d Representative
   heatmaps showing the cumulative duration spent by each mouse throughout
   the compartment during the 5-min open field test (OFT) in male (left)
   and female (right) mice. c, e Quantification of the frequency of entry
   into the central area of the open field in PLX- and/or SD-treated mice.
   f, h Representative heatmaps showing the cumulative duration spent by
   each mouse throughout the maze in the 5-min elevated plus-maze test
   (EPM). g, i Quantification of the frequency of mouse entry into the
   open arms. j Senescence-associated and inflammation-related gene
   expression changes in the PFC (left) and HIP (right) in PLX- and/or
   SD-treated male and female mice (red, upregulated; blue,
   downregulated). * indicates comparison with Veh + Con. # indicates
   comparison with Veh + SD (n = 6, each group). k Venn diagram
   representing the overlap of DEGs among the four groups in the PFC and
   HIP of male and female mice (blue, Veh + Con vs Veh + SD; yellow,
   Veh + SD vs PLX + SD; red, Veh + Con vs PLX + Con). N.S., no
   significant difference; *P < 0.05, **P < 0.01, ***P < 0.001,
   ****P < 0.0001, ^#P < 0.05, ^##P < 0.01, ^###P < 0.001,
   ^####P < 0.0001. Statistical analysis was performed using two-way ANOVA
   followed by Bonferroni post hoc correction.

Effects of microglial depletion on sex-specific changes in
inflammation-related, senescence-associated, and synapse-related gene
expression in SD-induced BD model

   To further examine the effects of microglial depletion on SD-induced
   inflammation-related and senescence-associated gene expression
   abnormalities, we performed qRT‒PCR to test these genes in the PFC and
   HIP. After 16-h daily SD exposure for 4 days, we identified 16
   inflammation-related and senescence-associated DEGs (4 upregulated and
   12 downregulated genes) in the PFC in males and 25 DEGs (5 upregulated
   and 20 downregulated genes) in females between control and SD-treated
   mice (Fig. [137]5j, k). In the HIP, we identified 18 DEGs (6
   upregulated and 12 downregulated genes) in males and 21 DEGs (3
   upregulated and 19 downregulated genes) in females between control and
   SD-treated mice (Fig. [138]5j, k). Among them, 6 DEGs in the male PFC
   (6/16, 37.5%) and 10 DEGs in the male HIP (10/18, 55.6%) were
   significantly regulated by PLX treatment, while 13 DEGs in the female
   PFC (13/25, 52%) and 6 DEGs in the female HIP (6/21, 28.6%) were
   significantly affected by PLX treatment (Fig. [139]5k). Together, these
   results indicate a sex-specific functional role of microglia in the
   HIP.

   Next, we sought to determine whether there were sex differences in
   synapse-related gene expression in the HIP in SD- and/or PLX-treated
   mice. We identified 2 synapse-related DEGs (1 upregulated and 1
   downregulated gene) in the HIP in males and 2 DEGs (0 upregulated and 2
   downregulated genes) in females between control and SD-treated mice
   (Fig. [140]6a, b). However, synapsin III (Syn3), synaptotagmin-1 (Syt1)
   and synaptotagmin-7 (Syt7) were upregulated after treatment with PLX in
   SD-treated males but not in SD-treated females (Fig. [141]6a).

Fig. 6. Effects of PLX3397 (PLX) on SD-induced microglial and synaptic
changes in the PFC and HIP in male mice.

   [142]Fig. 6
   [143]Open in a new tab

   a Synapse-related gene expression changes in the HIP in PLX- and/or
   SD-treated male and female mice (red, upregulated; blue,
   downregulated). * indicates comparison with Veh + Con. # indicates
   comparison with Veh + SD (n = 6, each group). b Venn diagram
   representing the overlap of DEGs among the four groups in the HIP of
   male and female mice (blue, Veh + Con vs Veh + SD; yellow, Veh + SD vs
   PLX + SD; red, Veh + Con vs PLX + Con). c, f Representative images of
   Iba1 immunoreactivity in the PFC and HIP. d–h Changes in microglial
   density and soma area in the PFC d, e and HIP g, h after treatment of
   male mice with PLX and/or SD. i Representative high-magnification
   Golgi-Cox staining images of dendritic spines in the HIP. j, k
   Histograms showing the total dendritic spine density in males and
   females. *P < 0.05, **P < 0.01. Scale bars = 50 μm in c, f; 10 μm in i.
   Statistical analysis was performed using two-way ANOVA followed by
   Bonferroni post hoc correction.

PLX3397 pretreatment eliminates microglia and reverses SD-induced synaptic
gain in males but not in females

   Subsequently, we conducted immunohistochemistry and Golgi staining to
   observe the effect of CSF1R inhibitor PLX on microglia and neurons. We
   found a significant reduction in the density of microglia and soma area
   in the PFC and HIP (Fig. [144]6c–h). Short-term pretreatment with PLX
   significantly blocked the microglia response to SD stimulation (Fig.
   [145]6d, g). Importantly, the elimination of microglia significantly
   decreased the total spine density in the HIP in SD-treated male mice
   but not in SD-treated female mice (Fig. [146]6i–k).

Discussion

   Here, we introduce the modified SD paradigm as a BD model in mice to
   mimic sleep loss during the whole daytime and partial nighttime in the
   presence of REM sleep pressure. We then used this model to elucidate
   the underlying mechanism by which intermittent SD in male mice triggers
   mania-like behavior and a cytokine-deficiency state. Microglia-mediated
   immune and synaptic dysfunctions were detected not only in the bulk
   transcriptome, but also in the single-cell transcriptome. Furthermore,
   our results confirmed SD-induced mania-like behaviors,
   inflammation-related, senescence-associated, and synapse-related gene
   expression changes in a sex-specific manner. Microglial loss and
   synapse gain were observed in SD-treated mice. Finally, we demonstrated
   that a CSF1R inhibitor can eliminate microglia and prevent SD-induced
   mania-like behavior and synapse gain in male mice.

   The validity of our chronic SD model was demonstrated using a range of
   behavioral paradigms. The OFT and EPM are commonly used to assess
   anxiety-like behavior [[147]39, [148]40], but are also widely employed
   for behavioral assessment of mania-like behavior (hyperactivity and
   decreased anxiety-like behaviors) in BD models [[149]41–[150]43]. This
   hyperactivity may be related to impulsivity in clinical BD patients.
   The TST and the ST further validate the efficacy of the sleep
   deprivation-induced BD model. Results showed a significant reduction in
   immobility time in the SD-treated group in the TST, indicating a marked
   antidepressant effect, which may relate to impulsivity and irritability
   in clinical BD patients. In the sociability test, SD mice displayed a
   significantly increased exploration time towards the social mice
   compared to the novel object. Moreover, the SDI was significantly
   higher in the SD group, indicating enhanced social interaction. This
   may mimic the talkativeness observed in patients with bipolar disorder
   (BD) [[151]44]. The chronic SD model elicits behavioral changes that
   are sustained for several hours (Extended Data Fig. [152]6), while the
   associated microglial reduction is maintained for several days
   (Extended Data Fig. [153]7). Additionally, our findings demonstrate
   that acute and chronic SD exert differential effects on hippocampal
   inflammation-related gene expression (Extended Data Fig. [154]8).

   Previous studies have used acute or severe SD models to investigate the
   effects of sleep loss on brain inflammation [[155]10, [156]45]. Here,
   single-nucleus and bulk RNA-seq analyses emphasized the enrichment of
   DEGs in microglia participating in inflammatory signaling pathways,
   including the TNF signaling pathway, IL-17 signaling pathway, B-cell
   receptor signaling pathway, and autophagy, in the chronic BD model. Liu
   et al. reported that DEGs enriched in the TNF signaling pathway were
   significantly upregulated during acute paradoxical SD [[157]45]. In
   addition, acute SD in mice induces activation of the IL-17 signaling
   pathway [[158]46], impaired B-cell responses [[159]47], and excessive
   autophagy [[160]48]. Four days of prolonged and continuous SD exposure
   leads to a cytokine-storm-like phenotype in mice [[161]10]. In
   contrast, we detected decreased expression levels of inflammatory genes
   in the brain and decreased levels of inflammatory cytokines in the
   plasma after 4 days of intermittent 16-h daily SD exposure. These
   findings expand the understanding of the role of microglia-mediated
   neuroinflammation in BD.

   Importantly, Zandi et al. reported that downregulated neuroimmune and
   synaptic pathways are associated with transcriptional changes in the
   amygdala and anterior cingulate from post-mortem brains of individuals
   with BD [[162]49], which is consistent to our findings in the
   hippocampus of SD-induced mice. Here, we detected the downregulation of
   microglia-specific genes in the PFC and HIP in both SD-treated male and
   female mice. Furthermore, snRNA-seq and bulk RNA-seq analyses showed
   that DEGs related to the synaptic membrane and function were enriched
   in SD-treated male mice. Abnormal synaptic transmission and synaptic
   pruning have been found in sleep-deprived mice [[163]13, [164]14,
   [165]50]. The synaptic homeostasis hypothesis proposes an unsustainable
   increase in synaptic strength during wakefulness and increased synaptic
   pruning during sleep [[166]51, [167]52]. A more recent study showed
   synapse gains during sleep deprivation and synapse loss during sleep in
   zebrafish [[168]53]. Our data confirmed the decreased engulfment of
   synapses in microglia and synapse gains during chronic SD. SD-induced
   behavioral and synaptic changes were prevented by the in vivo depletion
   of microglia with a CSF1R inhibitor (PLX) but not by the inhibition of
   microglia with MCL. Notably, the upregulation of Syn3, Syt1, and Syt7
   induced by PLX may protect against behavioral and synaptic changes in
   SD-treated male mice. Strong evidence has indicated that Syt7
   deficiency in the HIP induces mania-like behavior in mice by
   attenuating N-methyl-D-aspartate receptor activity [[169]54]. These
   findings strengthen the role of neuroinflammation and
   microglia-mediated synaptic pruning in BD. Microglia contribute to the
   inflammatory response and synaptic dysfunction induced by SD
   stimulation. Pretreatment with PLX may block the microglia-mediated
   detrimental processes induced by SD stimulation.

   Previous studies have shown that both Clock knockdown mice and Rev-erbα
   knockout mice exhibit marked mania-like behavior (hyperactivity and
   decreased anxiety-like behaviors) [[170]55, [171]56]. Depleting
   microglia using the CSF1R inhibitor ameliorated neuroinflammatory
   changes and mania-like behavior in Rev-erbα knockout mice [[172]55].
   These literature reports are similar to our findings in SD-induced
   mice. Our findings show decrease in the number of microglia following
   SD contributes to loss of synaptic pruning. Sleep-deprived male mice
   impaired microglia-mediated surveillance, phagocytic microglial
   contraction and synaptic pruning deficits [[173]50, [174]57, [175]58].
   Depletion of microglia with PLX pretreatment, rather than inhibition
   (MCL), blocked these SD-induced abnormalities. These data implicate
   microglial dyshomeostasis as a driver of SD-induced pathogenesis and
   show that pretreatment with the microglial depletion strategy can
   prevent this phenotype.

   Clinically, sex differences are present in BD symptoms and treatment
   response [[176]59–[177]61]. This study further examined sex-specific
   changes in SD-induced behaviors, gene expression levels, and treatment
   responses. We confirmed that VPA reversed SD-induced manic-like
   behavior in a sex-independent manner (Extended Data Fig. [178]1).
   Although long-term pretreatment with MCL has no effect on SD-induced
   behaviors with no sex difference, long-term pretreatment with PLX
   relieves SD-induced mania-like behaviors and synaptic gain in males but
   not females. Interestingly, more DEGs were detected in the PFC and HIP
   in SD-treated females than in SD-treated males, consistent with a
   higher burden of transcriptomic dysfunction observed in human brain
   specimens from females than in those from males with BD [[179]61],
   whereas PLX pretreatment had a much greater impact on DEGs in the HIP
   in SD-treated males than in SD-treated females. Microglial depletion by
   pretreatment with PLX prevents microglia-mediated inflammation-related
   and senescence-associated genes in the hippocampus in male mice.
   Sleep-deprived male mice develop microglia-mediated synaptic pruning
   dysfunction [[180]50], which may be mediated by the C1qb signaling
   pathway [[181]17, [182]62, [183]63] but not by the SIRPα-CD47 signal
   axis [[184]64]. Together, sex differences in BD may result from
   sex-specific differences in microglia-mediated neuroinflammation and
   complement-mediated microglial synaptic elimination. It remains to be
   confirmed whether the C1qb signaling pathway plays a causal role in BD
   in both males and females.

   In summary, our findings provide evidence that microglia-mediated
   inflammation, senescence, and synaptic pruning contribute to
   paradoxical SD-induced mania-like behavior in a mouse model of BD in a
   sex-specific manner. However, it remains to be understood at the
   single-cell level how sex-specific inflammation-related and
   senescence-associated DEGs affect microglia-mediated synaptic pruning
   in the pathophysiology of BD. Further elaborating the causal link
   between neuroimmune, neuroplasticity, and behavioral states will
   provide mechanistic insights into the pathophysiology and
   neurodevelopment of bipolar disorder and pave the way for curative
   therapies for BD patients.

Supplementary information

   [185]Supplementary materials^ (76.9MB, pdf)

Acknowledgements