Wild-type C57BL/6NTac male mice (3-months-old, Taconic) were bred in a laboratory animal facility at the Institute of Biomedicine and Translational Medicine, University of Tartu. Mice from different litters were housed in 1264C Euro standard type II cages (Tecniplast S.p.A.) measuring 268 × 215 × 141 mm. Cages containing aspen chips and wool for bedding and nesting were replaced once a week. Each cage contained 9–10 animals based on allocation after weaning. Mice were kept under standard conditions with unlimited access to food and water on a 12/12-h light/dark cycle (light on from 7 a.m. to 7 p.m.). All animal procedures in this study were performed in accordance with the European Communities Directive with license No. 171 issued by the Estonian National Board of Animal Experiments.

Mice were randomly assigned into four groups: Control (Ctr) (n = 9), CUS (n = 10), microglial repopulation (repMg) (n = 10), and CUS + repMg (n = 10). After a week of transfer adaptation, the mice were subject to CUS/Ctr and CSF1Ri/Veh treatments (Figure 1A and Supplementary Figure S2B). CUS procedures, as previously described (Cangalaya et al., 2020; Chen et al., 2022; Xuan et al., 2022), were followed. Briefly, mice were exposed to a variable sequence of seven mild and unpredictable stressors once per day (d) for eight consecutive weeks, including food and water deprivation overnight, rat odor and isolation overnight, restraint in a 50-ml tube for 2 h, wet bedding and tilted cage, stroboscopic illumination overnight, flipped light/dark exposure, and swimming at 18°C for 10 min. All stressors were randomly scheduled and changed daily to sustain an unpredictable procedure.

PLX3397 (HY-16749/CS-4256, MedChemExpress) was dissolved at 200 mg/mL in DMSO (D8418, Sigma-Aldrich) stock solution, and an aliquot was freshly diluted with corn oil (#8267, Sigma-Aldrich) by 1: 6.5 before using. Drug-treated mice were individually fed with Veh (100 μL 15% DMSO/85% corn oil+0.5 g Nutella/mouse/day) or PLX3397 (120 mg/kg body weight, e.g., 3 mg/mouse in Veh) (Cangalaya et al., 2020) for voluntary ingestion for 7 consecutive days after 5 weeks of CUS, followed by a 2-week recovery period.

Mice were habituated to ~250 lux room light for 1 h. Each mouse was measured for distance and time traveled in different zones of a digitalized box (44.8 × 44.8 × 45 cm) via software (Technical & Scientific Equipment GmbH, Germany) for 30 min. The floor of the box was cleaned with 70% ethanol and dried thoroughly after each mouse.

The elevated plus maze consisted of open and closed arms (30 × 5 cm each) intersected at a central 5 × 5-cm square platform elevated to a height of 80 cm. Mice were habituated to ~40 lux room light for 1 h. Each mouse was placed on the central platform facing the open arm and recorded for time spent on open or closed arms using EthoVision XT software (Nodules, United States) for 5 min. The arms were cleaned with 70% ethanol and dried thoroughly after each mouse.

Mice were habituated to ~40 lux room light for 1 h. A rectangular three-chamber box made from clear Plexiglas was divided into three identical sections, each side accommodating a lid-covered and wire-structured cup-like container large enough to enclose a single mouse, allowing free exchange of air but not direct physical contact between mice on both sides. A test mouse was first habituated in the central chamber for 5 min and then introduced to a stranger mouse located in a container and left to freely explore the three chambers for the next 10 min. All stranger mice were at the same age as test mice and habituated to the apparatus in advance. The box and wire containers were cleaned with 70% ethanol and dried thoroughly after each test. Social preference was defined as time spent with the stranger mouse, and social contacts were counted when the head and front paws of a test mouse were within 3 cm vicinity of the container wall as recorded by a camera (Nodules, United States).

Mice were habituated to ~40 lux room light for 1 h. An animal was hung on a wooden bar by the tip of the tail using an adhesive tape and recorded with a camera for 5 min. A complete lack of movement or small movements of forefeet and body swinging was counted as immobility, and time (s) of immobility was measured.

Mice were euthanized with CO_2, and the PFC were dissected and immediately stored at −80°C. Human peripheral blood was collected in PAXgene^™ blood RNA tubes. Total RNAs from human blood (Applied Biosystems, United States) and mouse PFC (Molecular Research Center, United States) were extracted, quantified, assessed for purity using a NanoDrop spectrophotometer (Thermo Fisher Scientific, United States) and immediately sent to the Beijing Genomics Institution (BGI) for messenger RNA-seq (after globin mRNA removal and quality control) on the BGIseq-500 platform. The quality of the RNA samples (RIN/RQN ≥ 7.0, 28S/18S ≥ 1.0) was confirmed by BGI. Clean data of at least 20 M clean reads per sample were collected.

RNA-seq data and gene expression analysis were done on the Galaxy and NetworkAnalyst platforms using DESEQ2 (Zhou et al., 2019). Data with variance percentile rank of <15% and counts of <4 were filtered out. Counts per million reads were transformed, normalized, and calculated to Log2 fold changes (Log2FC) for differentially expressed genes (DEGs). DEGs with Benjamini-Hochberg’s false discovery rate (FDR) of <0.05 were subjected to functional clustering analysis in DAVID.¹

C57BL/6Ntac mice were anesthetized with an intraperitoneal injection of end-dose of ketamine or xylazine mixture and transcardially perfused with PBS and 4% paraformaldehyde (PFA). Dissected brains were post-fixed in the 4% PFA at 4°C for 1 day, dehydrated by 30% sucrose, and stored at −80°C. The brains were taken to −20°C for 1 day before cryosectioning.

PFC coronal cryosections in 40 μm-thickness were washed in PBS and incubated in 0.5% triton-x100. After PBS washing, slices were incubated with primary antibodies, including rabbit anti-IBA1 (#SKL6615, Wako, United States, diluted 1:500) and mouse anti-synaptophysin (SYP, sc-17,750, Santa Cruz, United States, diluted 1:250) in blocking buffer (10% goat serum+1%BSA+ 0.3 M Glycine) overnight at 4°C. It was followed by PBS washing and incubation with the secondary antibodies, including goat anti-rabbit IgG H&L-Alexa568 (#ab175471, Abcam, United States), goat anti-mouse IgG H&L-PECy7 (#D2110, Sant Cruz, United States, diluted 1:500), and 0.1 μg/mL DAPI (#ACRO202710100, VWR, United States) for 2 h at room temperature. After PBS washing, slices were mounted onto glass slides with Fluoromount^™ Aqueous Mounting Medium (#F4680-25ML, Sigma-Aldrich, United States).

Images of 512 × 512 pixels were taken by an FV1200MPE laser scanning microscope at 60× magnification (Olympus, United States) with a scanning velocity of 12.5 pixel/μm and Z-stacks (step: 0.5 μm, depth: 20 μm) to analyze the morphology of IBA1⁺ cells and puncta of SYP. The SNT package in ImageJ was used to reconstruct 3D cell morphology, and “Sholl analysis” was used to acquire the ramification index, while the other parameters were acquired through “quick measurements” of SNT, including total branch length and the number of branches. SYP⁺ puncta in the microglial region of interest defined via cell skeleton were obtained by “analyze particles,” and SYP density was calculated as SYP⁺ area (μm²) per image area (3.03 × 103 μm²) or per microglial area. 3D images segregating IBA1⁺ microglia, nuclei, and SYP⁺ puncta were obtained with “surface” and “spots” functions with videos created by “animation” in Bitplane software (Imaris).

Mouse hippocampi were dissected after CO₂ euthanization and gently homogenized through 70 μm cell strainers (#352350, BD Bioscience, United States) in ice-cold PBS with 1% FBS. Homogenates were washed and centrifuged at 500 g for 5 min. Isolated cells were blocked with 10% rat serum in ice-cold PBS for 1 h. Brain cells were stained with 0.5 μL anti-mouse VGLUT2-Alexa488 (#MAB5504A4, Millipore, United States), CD11b-BV421 (#101251, BioLegend, United States), CD45-BV650 (#103151, BioLegend, United States), Glast-APC (#130–123-555, Miltenyi, Germany), and O4-PE (#130–117-357, Miltenyi, Germany) in PBS with 1% FBS on ice for 1 h. Corresponding isotype control antibodies (all BioLegend, United States) included rat IgG2a-Alexa488 (##400,525), IgG2b-BV421 (#400639), IgG2b-BV650 (#400651), IgG2b-APC (#400219), and IgM-PE (#401611). Cells were washed, resuspended in 500 mL PBS, and acquired with a Fortessa flow cytometer (BD Bioscience, United States). Data were analyzed by Kaluza v2.1 software (Beckman Coulter, United States). Astrocytes were defined as Glast⁺ cells, oligodendrocyte precursor cells (OPCs) as O4⁺ cells, and microglia as CD45^lowCD11b^hi cells. Cell populations were calculated as % among total brain cells or microglia as previously described (Piirainen et al., 2021; Chithanathan et al., 2022).

FES patients (n = 74) were recruited from the Beijing Huilongguan Hospital. Patients were diagnosed with schizophrenia according to the Structured Clinical Interview for DSM-IV (SCID) independently by two psychiatrists. The inclusion criteria were as follows: (Green et al., 2020) 18-54-year-old Han Chinese people; (Han et al., 2019) illness duration of ≤3 years (<1 year on average); and (Wang et al., 2020) un-medicated or < 2 weeks of anti-psychotic medication at the time of blood draw. HCs (n = 68) matched for age and sex were recruited from the local community. Candidates who did not meet the recruitment criteria were excluded. All participants provided written informed consent. The study was approved by the Institutional Ethical Committee of Beijing Huilongguan Hospital (No.2017–49).

Past traumatic experiences were evaluated by a childhood trauma questionnaire (CTQ) constituting retrospectively measured 29 items encompassing the following five factors: physical abuse, emotional abuse, sexual abuse, physical neglect, and emotional neglect. Stress levels were evaluated based on a perceived stress scale (PSS) questionnaire constituting 14 items measuring feelings and thoughts during the past month. Sub-scores and total scores (PANSS-P, -N, −G, −T) of the positive and negative syndrome scale were measured independently by two psychiatrists.

Brain structural MRI data were acquired using a Siemens Prisma 3.0 T MRI scanner with a 64-channel head coil. Foam pads were used to minimize head motions. Sagittal 3D magnetization-prepared rapid acquisition gradient echo (MPRAGE) was chosen for imaging, using ENIGMA protocol and FreeSurfer software, with repetition time (TR)/echo time (TE)/inversion time (TI) = 2530/2.98/1100 ms, flip angle (FA) = 7°, field of view (FOV) = 256 × 224 mm², pixel/gap size = 1/0 mm, and matrix size = 256 × 224 bit. After scanning, two radiologists evaluated image quality, and if significant artifacts existed, images were recollected. Representative T1 MRI images at −16 mm from the bregma were obtained from “The Scalable Brain Atlas” (Bakker et al., 2015). Subregions of the hippocampus were drawn based on the Allen Human Brain Atlas.²

Data distributions were examined through the Shapiro–Wilk test. For animal data, two-way ANOVA was used to examine the interaction between CUS and repMg and main effects, with Tukey’s or LSD correction for post-hoc pairwise comparisons. For human data, ANOVA or the Mann–Whitney U test was used for continuous variables and the chi-squared tests for categorical variables. ANCOVA was conducted for MRI data, and correlational analyzes were done using Spearman’s or Pearson’s partial method, with age, sex, and BMI as covariates using SPSS-v27.0 (IBM). Figures were prepared in GraphPad Prism v8.0.1 and online.³ Data were presented as mean ± SEM, and a p-value of <0.05 was considered statistically significant.

To evaluate the effects of CUS and repMg, we performed a battery of mouse behavioral tests (Figure 1A). CUS induced anxiety, as shown by a decreased ratio of central versus total travel distance (p < 0.01, Figure 1B) in OFT and increased time in EPM closed arms (p < 0.05, Figure 1C). In RepMg and CUS + repMg, mice showed similar anxiogenic effects in OFT (both p < 0.05, Figure 1B) and EPM (p < 0.05/0.01, Figure 1C). CUS x repMg interactions were found in OFT (F (1, 35) = 7.545, p < 0.01) and EPM (F (1, 35) = 4.489, p < 0.05). Overall, it suggested that microglial replenishment itself caused anxiety and cannot ameliorate CUS-induced anxiety.

A CUS x repMg interactive effect on sociability was also found in TCT (F (1, 35) = 9.623, p < 0.01). CUS mice displayed less preferential entry into the social chamber and spent time there than in the empty chamber (Figures 1D,E). Mice in the repMg group also entered less into the social chamber (Figure 1D) but spent a longer time once there than in the empty chamber (p < 0.001, Figure 1E). However, mice of the CUS + repMg group showed normal sociability similar to Ctr mice (p < 0.01, Figures 1D,E), indicating that repMg can partially rescue CUS-triggered social deficit.

For depression-like behavior, CUS also interacted with repMg (F (1, 35) = 12.180, p < 0.01). Mice in CUS, but not repMg, showed longer immobility time than Ctr mice (p < 0.05, Figure 1F). Furthermore, CUS + repMg mice showed significantly less immobility time than CUS mice (p < 0.05, Figure 1F), demonstrating the ameliorative effect of repMg on CUS-induced learned helplessness.

Among the four groups, 502 upregulated and 577 downregulated DEGs were identified, involving nervous system development and axon guidance among the top 10 GO-BP pathways (Figures 2A–D and Supplementary Table S1). Notably, expressions of the DEGs in these two pathways were generally recovered in CUS + repMg (Figures 2C,D). Furthermore, compared to Ctr, DEGs particularly affected by repMg were involved in the rhythmic process and regulation of synaptic plasticity, with most synaptic DEGs down-regulated (Supplementary Figure S1 and Supplementary Table S2), whereas those affected by CUS were a small group of angiogenic DEGs (Supplementary Table S3). These results reveal that CUS and repMg both affect brain structural remodeling, but repMg can partially restore brain developmental processes altered by CUS.

To verify the synaptic remodeling function of replenished microglia, we then studied the mouse PFC and hippocampal regions. We first co-stained mouse PFC sections by IHC for SYP, a presynaptic marker, and IBA1 (Figure 3A). CUS x repMg interactions were found on SYP punctal number (F (1, 44) = 5.229, p < 0.05), SYP density (F (1, 44) = 11.480, p < 0.01), and SYP-density per microglia (F (1,76) = 7.469, p < 0.01). CUS-induced loss of SYP puncta (p < 0.05, Figure 3B) and SYP density (p < 0.01, Figure 3C) compared to Ctr, and repMg rescued CUS-triggered SYP decrease (p < 0.05, Figure 3C). Importantly, SYP density inside microglia was higher in CUS than Ctr (p < 0.05, Figure 3D).

CUS and repMg also interacted on microglial ramification index (F (1,76) = 7.963, p < 0.01) and total branch length (F (1,76) = 5.089, p < 0.05). Compared to Ctr, CUS-conditioned microglia had a larger ramification index (p < 0.05, Figure 3E) and longer total branch length (p < 0.01, Figure 3F). In contrast, repMg corrected CUS triggered increases in these parameters (both p < 0.01, Figures 3E,F). No differences were found in the number of branches among the groups (Figure 3G).

We also quantified glial cells by flow cytometry with hippocampal tissues, another important region for stress regulation (Figure 4A and Supplementary Figure S2A). CUS and repMg interacted on microglia% (F (1,24) = 10.680, p < 0.01). However, CUS did not significantly affect microglia% (Figure 4B). Although microglial abundancy did not fully recover in repMg as compared to Ctr (p < 0.01, Figure 4B), the recovery was better in CUS + repMg compared to repMg alone (p < 0.01, Figure 4B). No changes in other glial populations were found (Figures 4C,D), demonstrating the microglia-specific effect of CSF1Ri.

We further quantified the glial engulfment of VGLUT2, a glutamatergic presynaptic marker. CUS x repMg interactions existed on VGLUT2⁺ microglia (F (1, 24) = 10.250, p < 0.01), VGLUT2⁺ astrocytes (F (1, 24) = 7.024, p < 0.05), and VGLUT2⁺ OPCs (F (1, 24) = 9.100, p < 0.01). Compared to Ctr, CUS elevated VGLUT2 in microglia (p < 0.05, Figure 4E), astrocytes (p < 0.01, Figure 4F), and OPCs (p < 0.05, Figure 4G).

Compared to Ctr, replenished microglia were not over-pruning but led to elevated VGLUT2 engulfment by OPCs (p < 0.05, Figure 4G). Notably, repMg normalized VGLUT2⁺ engulfment by stressed microglia (p < 0.001, Figure 4E) and astrocytes (p < 0.01, Figure 4F), thereby supporting the RNA-seq and IHC results and strengthening the idea that repMg can partially rescue CUS-induced synaptic deficits.

These data overall suggest that CUS-conditioned microglia are over-ramified and more phagocytic on presynaptic components, which can be corrected by repMg.

We previously found that some FES patients had higher stress sensitivity (Xuan et al., 2022) and reduced amounts of nonclassical monocytes along with altered signature gene expressions (Chen et al., 2022) compared to HCs, implying their deficits in myeloid reprogramming. To relate the implications of repMg to a clinical setting, we explored a cohort from these FES patients and HCs.

No differences in age, sex, and education years between the two groups were found (Table 1). However, FES patients had lower BMI (p < 0.001) as well as higher CTQ (p = 0.002, Table 1) and PSS (p = 0.001, Table 1) scores than HCs. Interestingly, MRI data showed alterations of hippocampal subregions (Figure 5A and Table 2), but not PFC structures (data not shown), in FES patients, with significantly smaller hippocampal fimbria (left: p = 7.213e-8, right: p = 1.041e-9, Figure 5B) but larger tails (left: p < 0.05, right: p < 0.0001), left presubiculum (p < 0.01), and right fissure (p < 0.05).

Volumes of the right hippocampal fimbria were negatively correlated with PSS scores in FES patients (r = −0.397, p = 0.003, Figure 5E and Supplementary Table S4) but not in HCs. While no correlations with PSS scores were found for the left hippocampal fimbria in both groups (Figure 5D), they were negatively associated with PANSS scores (PANSSN: r = −0.388, p = 0.004; PANSST: r = −0.339, p = 0.013; Figures 5F,G and Supplementary Table S5). No correlations were found between CTQ scores and hippocampal volumes (Supplementary Table S6).

Assuming genes involved in microglial development were similarly altered in CUS-exposed animals and FES patients, we explored blood RNA-seq data on FES patients and HCs. Similar to animal RNA-seq results (Figure 2B), the top GO-BP pathway of human blood DEGs involved nucleotide metabolism and transcription regulation (Supplementary Table S7). Furthermore, we found 181 DEGs involved in immune system development (Figure 6A, Supplementary Table S7), such as hematopoiesis and leukocyte differentiation (Figure 6B). Compared to HCs, these 181 immdev-DEGs overall showed a negative correlation with PSS scores (p < 0.0001, Figure 6C and Supplementary Figure S3B) and a positive correlation with hippocampal fimbria volumes, particularly on the right side (p < 0.0001, Figure 6D and Supplementary Figure S3B) in FES. Among these immdev-DEGs, 18 genes with the most significant correlational coefficients (Zr) kept such trends in FES, which were reversed in HCs in most cases (Figures 6C,D). We further undertook a mediator analysis assuming stress negatively affected hippocampal structures via certain immdev-DEGs. Notably, KCNQ1, encoding a voltage-gated potassium channel, was upregulated in FES patients (FDR < 0.01, Figure 6E) similar to CUS + repMg-exposed animals (Supplementary Figure S3A). Furthermore, KCNQ1 partially mediated the negative regulation of right fimbria volume by PSS (β = −0.442, 95% CI: −1.326 ~ −0.087, Figure 6E).

These data overall mark the right hippocampal fimbria as an important stress-sensitive subregion that immune genes may help regulate in FES, corroborating microglia-mediated brain functions in CUS-devastated animals.