To investigate a potential role for PTEN in microglia in shaping animal behaviors, we created mice with inducible Pten deletion in a microglia-specific manner by breeding Cx3cr1 ^CreERT2/+ mice with Pten ^fl/fl mice. To specifically examine PTEN-controlled microglial functions during postnatal neurodevelopment, as detailed in Methods and illustrated in Figure 1A , we injected tamoxifen into the stomach of newborn Cx3cr1 ^CreERT2/+ Pten ^fl/fl and littermate control Cx3cr1 ^CreERT2/+ Pten ^+/+ mice daily from postnatal day 0 (P0) to day 2 (P2). In this system, microglia and other cells in the brain are intact throughout embryonic development. To verify the efficiency of Pten deletion by this procedure, microglia were isolated at different time points to test PTEN protein expression. As compared to cells from littermate control mice, microglia from Cx3cr1 ^CreERT2/+ Pten ^fl/fl mice largely lost PTEN protein as early as one week after the tamoxifen injection and remained so for as long as we tested ( Figures S1A, B ). Taking advantage of the EYFP expression from the knock-in Cx3cr1 ^CreERT2 allele, we used anti-GFP antibody to stain for EYFP on tissue sections and verified that the targeted cells were essentially all Iba-1⁺ microglia but not astrocytes (S100b⁺), oligodendrocytes (Olig2⁺), or neurons (NeuN⁺) ( Figures S1C, D ). For convenience, we subsequently call tamoxifen-treated Cx3cr1 ^CreERT2/+ Pten ^fl/fl and Cx3cr1 ^CreERT2/+ Pten ^+/+ mice mPTEN^KO(0) and mPTEN^WT(0), respectively.

To examine behavioral consequences of postnatal PTEN deletion in microglia, we first compared mPTEN^KO(0) and mPTEN^WT(0) mice (>8 weeks old) in a marble-burying test, which could reveal signs of ASD-like repetitive behaviors and potentially increased anxieties (38). As shown in Figure 1B , the two groups of mice were comparable. On the other hand, when tested in a novel object recognition test (NORT), mPTEN^KO(0) mice failed to exhibit a preference for exploring the novel object, unlike their mPTEN^WT(0) counterparts ( Figure 1C ). In social interaction tests, whereas mPTEN^WT(0) mice showed a clear preference for interacting with a companion mouse than with an inanimate object ( Figure 1D ) and for interacting with a stranger than with a familiar mouse ( Figure 1E ), the mPTEN^KO(0) mice did not show such preferences ( Figures 1D, E ). The failure of mPTEN^KO(0) mice to pass these tests did not result from grossly impaired locomotor abilities or increased anxieties, as they performed as well as the wildtype counterparts in the open-field test ( Figure 1F ). These data demonstrate impaired sociability and exploration of novel objects resulting from postnatal PTEN ablation in microglia.

To characterize changes of postnatal microglial development and function caused by Pten ablation, we first examined their progressive expansion. We chose the cerebral cortex for two reasons. First, microglia from different brain regions are quite heterogeneous (39), and cerebral cortex represents an abundant source of microglia. Second, cortical functions are important for social behaviors (40, 41). As shown in Figures 2A, B , the density of Iba-1⁺ microglia gradually increased to reach a stable plateau by 2-3 months after birth (hereafter “young adult” is used to specifically refer to 2- to 3-month-old mice), with the most rapid expansion taking place during the second week of postnatal development. When Pten was ablated from microglia, these cells accumulated almost twice as fast to reach approximately the plateau density by P14; in young adult mice or older animals, the density of mPTEN^KO(0) microglia in cerebral cortex was only slightly albeit statistically significant higher than that of mPTEN^WT(0) cells ( Figures 2A, B ). We did not observe differences in density or distribution of neurons, astrocytes, or oligodendrocytes in the two types of mice ( Figure S1E ). Individual mutant microglia exhibited markedly different morphology in that their cell bodies were enlarged, with thickened cellular processes extending shorter and fewer branches, indicative of an activated state (42), as seen in both somatosensory ( Figure 2C ) and auditory cortices ( Figure 2D ). Consistent with these morphological patterns, mPTEN^KO(0) microglia from young adult mice expressed higher levels of F4/80 and CD86 ( Figures 2E, F ), which are upregulated in activated microglia (43).

Next, we conducted transcriptomic profiling of mPTEN^WT(0) and mPTEN^KO(0) microglia isolated from postnatal day 7 (P7) mice, postnatal day 14 (P14) mice and 2- to 3-month-old young adult mice. By principal component analyses, we found that mPTEN^WT(0) and mPTEN^KO(0) microglia from the same time of postnatal development were grouped together, whereas microglia of the same genotype, either mPTEN^WT(0) or mPTEN^KO(0), displayed divergent transcriptomes across different time points of postnatal development ( Figure 2G ). Therefore, while microglia-intrinsic Pten deletion led to changes in cell abundance and morphology, it did not completely alter the overall trajectory of postnatal microglial development, which involves progressive changes of the transcriptome at a larger scale (44).

At a finer scale, on the other hand, the number of genes differentially expressed by wildtype and mPTEN^KO(0) microglia increased over time ( Figure 2H , Table S1 ). Prominently, genes encoding C1q components (C1qa, C1qb, and C1qc) were significantly upregulated in mutant microglia at P14 and later ( Figure 2I ). We also observed increased C1q protein by immunostaining on tissue sections ( Figures S2A, B ). Many lysosome-related genes (as defined by the KEGG pathway), including Cd68 and those encoding cathepsins, were expressed at significantly higher levels in PTEN-deficient microglia at P14 and later ( Figure 2J ). Interestingly, many genes upregulated in the DAM signature, which reflects an inflammatory microglia state observed in neurodegenerative diseases including AD and ALS (17, 18), were also expressed at higher levels in mPTEN^KO(0) than in mPTEN^WT(0) microglia at P14 and later ( Figure 2K ). On the other hand, general downregulation of microglia homeostatic genes (e.g. P2ry12, Cx3cr1, Cst3, Tmem119, Olfml3, Hexb, Fcrls, Csf1r, Tgfbr1, Mef2a, Mafb) seen in the DAM signature (44–46) did not occur, as expression of most of these genes were either unchanged or increased in mPTEN^KO(0) microglia ( Figure 2L ). Gene set enrichment analyses further revealed that, at all three time points examined, genes upregulated in mPTEN^KO(0) microglia enriched those in KEGG pathways of oxidative phosphorylation, AD, and the Parkinson’s disease (PD; Figures S3A-C ). A shared feature of these three KEGG pathways is that they all contain many genes encoding mitochondrial components ( Table S2 ), reflecting the fact that mitochondria dysfunctions are implicated in the pathogenesis of AD and PD (47). As shown in Figure 2M , many mitochondria-related genes categorized in the 3 pathways are expressed at higher levels in mPTEN^KO(0) than in mPTEN^WT(0) microglia from P14 or young adult mice.

Combined together, these profiling data suggest that PTEN ablation leads to an activated microglia state that features enhanced oxidative phosphorylation, increased phagocytosis, and possibly altered C1q-dependent synapse pruning (48).

Phagocytosis is important for microglial homeostatic and defense functions (49, 50). PTEN could directly regulate phagosome membrane trafficking, because its enzymatic activities reduce membrane PIP3 concentration (51, 52). KEGG pathway enrichment analysis of upregulated genes in mPTEN^KO(0) microglia reveals lysosome pathway is most significantly enriched at all three time points examined ( Figure S3D ). Consistent with the mRNA-seq results ( Figure 2J ), mPTEN^KO(0) microglia expressed higher levels of intracellular CD68 protein at P14 ( Figures 3A, B ). By flow cytometry, freshly isolated mPTEN^KO(0) microglia were able to phagocytose more zymosan particles in vitro ( Figure 3C ), indicating that the intrinsic phagocytic capability is enhanced in PTEN-deficient microglia.

To directly test whether PTEN deletion from microglia earlier in postnatal life is causative for defective sociability observed later in life, we injected tamoxifen into separate groups of Cx3cr1 ^CreERT2/+ Pten ^fl/fl and littermate control Cx3cr1 ^CreERT2/+ Pten ^+/+ mice on postnatal day 14 to create mPTEN^KO(14) and mPTEN^WT(14) mice. PTEN deletion was verified by Western blotting with sorted microglia, as done in Figure S1B .

Interestingly, microglia remained of overactive features in young adult mPTEN^KO(14) mice. Mutant microglia were more abundant than their wildtype counterpart ( Figure 4A ), and their cellular processes were thickened and shorter with fewer branches ( Figures 4B, C ). The transcriptome of those mPTEN^KO(14) microglia was globally similar to that of microglia from mPTEN^KO(0) mice ( Figure S4A ), which is consistent with the PCA result for mPTEN⁽⁰⁾ mice that microglia from the same age tend to cluster together ( Figure 2G ). Genes encoding C1q components (C1qa, C1qb, and C1qc) were also upregulated in mPTEN^KO(14) microglia ( Figure S4B ).

Despite those similar abnormalities between mPTEN^KO(14) and mPTEN^KO(0) microglia in adult animals, mPTEN^KO(14) mice performed normally in novel object recognition and social interaction tests ( Figures 4D, G ). Further transcriptome analysis showed that the number of differentially expressed genes between mPTEN^KO(14) and mPTEN^WT(14) microglia in young adult mice was greatly reduced, compared to that observed between mPTEN^KO(0) and mPTEN^WT(0) microglia (1411 vs 1831, Table S3 ). Fewer lysosome genes were upregulated in mPTEN^KO(14) microglia as compared to those upregulated in mPTEN^KO(0) microglia ( Figure S4C , Figure 2J ). GSEA analysis indicates that genes differentially expressed between mPTEN^KO(14) and mPTEN^WT(14) microglia were no longer enriched with KEGG pathways implicated in AD and PD, although those implicated in oxidative phosphorylation were still enriched ( Figure S4D ). Consistently, the number of enriched KEGG pathways was dramatically decreased compared to that enriched in mPTEN⁽⁰⁾ mice, although lysosome pathway was still enriched ( Figure S4E ). Most DAM signature genes and mitochondria-related genes that were upregulated in mPTEN^KO(0) microglia were not differentially expressed in mPTEN^KO(14) microglia ( Figures S4F, G ).

Taken together, these data indicate that the first 2 weeks are the most critical period during which PTEN in microglia must be intact in order to safeguard brain circuit development underlying normal sociability.

Because synapses and circuit connections are often altered in ASD (53–55), we further examined synaptic changes as a result of microglial PTEN deletion. We prepared synaptosomes from 2-3-month cerebral cortex and probed different presynaptic and postsynaptic components by Western blotting. A significant reduction in presynaptic VGLUT1 and synaptophysin was observed in mPTEN^KO(0) as compared to control synaptosomes, whereas postsynaptic protein PSD95, the NMDA receptor subunit GluN2B, and Gephyrin were not appreciably altered ( Figures 5A, B ). By immunostaining of tissue sections, we found that the density of VGLUT1 puncta was significantly reduced in deep layers of somatosensory and auditory cortices in P14 mPTEN^KO(0) mice ( Figures 5C, D ). Excitatory synapses, as identified by colocalized VGLUT1 and PSD95 staining, showed a significantly reduced density in these mutant mice ( Figures 5E, F ). To investigate whether the reduction of presynaptic VGLUT1 correlates with more phagocytic microglia at P14, we examined VGLUT1 staining together with Iba-1 and CD68. As shown in Figures 5G-K , Iba-1⁺ microglia in P14 mPTEN^KO(0) L5 contained more VGLUT1 puncta within CD68⁺ vesicles. These data strongly suggest that mutant microglia has already been more actively removing VGLUT1-containing pre-synapses in the early developmental stage.

Given perturbed microglia-synapse interactions observed (mostly L5), we explored early-life functional changes of neurons by electrophysiology. Disrupted intrahemispheric and intracortical functional connectivity are implicated in abnormal social behaviors (56), and L5 pyramidal neurons account for the largest fraction of intracortical output (57). Degraded auditory information processing and transfer would impair speech recognition and communication in human (58, 59). We therefore conducted in vitro recordings from L5 pyramidal neurons in the auditory cortex. Consistent with the decrease of VGLUT1 contents and reduction of excitatory synapses in mPTEN^KO(0) cortex, whole-cell patch clamp recordings from L5 pyramidal neurons in acute slices revealed markedly reduced frequency, amplitude and area of miniature excitatory postsynaptic currents (mEPSC) in L5 pyramidal neurons in P14 auditory cortex ( Figure 5L ). Furthermore, paired-pulse ratios were significantly increased in mPTEN^KO(0) L5 pyramidal cells ( Figure 5M ), suggesting a decrease in release probability of glutamate vesicles from presynaptic terminals. Because levels of postsynaptic AMPA receptor GluA1 and GluA2 subunits showed no difference in synaptosomes of P14 mPTEN^KO(0) and mPTEN^WT(0) mice (data not shown), the decreased mEPSC amplitude potentially results from reduced quantal size (i.e. less glutamate in presynaptic vesicles) due to the reduced level of VGLUT1 expression (60).

These data support the possibility that defects introduced during the critical period of postnatal neurodevelopment underlie behavioral abnormalities later in life.

Consistent with normalized social behaviors of mPTEN^KO(14) mice, synaptosomes from the mPTEN^KO(14) cerebral cortex contained normal amounts of all presynaptic and postsynaptic proteins tested, including VGLUT1 ( Figures S5A, B ). mPTEN^KO(14) cortices in young adult mice had a similar level of VGLUT1 puncta ( Figures S5C, D ), and the density of VGLUT1⁺PSD95⁺ synapses was no longer reduced ( Figures S5E, F ). Therefore, when postponed beyond 2 weeks after birth, PTEN deletion in microglia does not cause neuronal and behavioral abnormalities as seen with earlier deletion.

Pten ^fl/fl (Jax 006440), Cx3cr1 ^CreER (Jax 021160), Thy1-YFP (Jax 003782) were originally purchased from the Jackson Laboratory. All mice were maintained on the C57BL/6 background, and were maintained under specific pathogen-free conditions. All mice analyzed were male mice. All animal experiments were conducted in accordance of government and Tsinghua guidelines for animal welfare and approved by Tsinghua Institution Animal Care and Use Committee.

To induce Pten deletion, newborn littermates derived from a Cx3cr1 ^CreER/CreER Pten ^fl/+×Cx3cr1 ^+/+ Pten ^fl/+ breeding scheme were given 50 μg tamoxifen in sunflower oil (Sigma) daily on P0, P1 and P2 (68). Those of the desired Cx3cr1 ^CreER/+ Pten ^fl/fl and Cx3cr1 ^CreER/+ Pten ^+/+ genotypes were subsequently used as mPTEN^KO(0) and mPTEN^WT(0), respectively. For more delayed postnatal Pten deletion, littermates derived from the same breeding scheme were given two doses of 50 mg/kg body weight by gavage on P12 and P14 to produce mPTEN^WT(14) and mPTEN^KO(14) mice.

Young adult male mice of 2 to 3 months of age were used for all behavioral tests.

Mice were anesthetized and transcardiacally perfused with PBS, the brains were quickly removed and the cortex were isolated in ice-cold HBSS (Gibico). The cortex was homogenized with the Kimble Dounce tissue grinder and then subjected to centrifugation through 30%/70% discontinuous Percoll (GE) gradient. The microglia-enriched fraction at the interface was washed with ice-cold PBS and then resuspended in the MACS buffer (PBS supplemented with 1% FBS and 5mM EDTA) for subsequent procedures. The cells were blocked for 10 min with Fc-blocker (Clone 2.4G2: 20 μg/ml; BioXcell), and then were incubated with the following antibodies for 30 min on ice: PerCPcy5.5 anti-CD45 (104), PE anti-CD86 (GL-1) from Biolegend, PE-Cy7 anti-CD11b (M1/70) from BD Biosciences, eFluor 450 anti-F4/80 (BM8) from eBioscience, rabbit anti-Tmem119 (106-6) from Abcam. For Tmem119, cells were washed and further stained with the secondary antibody AF647 anti-rabbit IgG (catalog: A21245) for 30 min. For intracellular staining of PTEN (138G6, CST) and PE anti-CD68 (eBioscience), the CytoFix/Perm Kit (BD Biosciences) was used according to the manufacture’s protocol. All cytometric data were collected on a LSRII or an Aria III cytometer (BD Biosciences) and were analyzed with FlowJo software (TreeStar). Dead cells were excluded from analysis according to the 7-AAD (Biotium) staining, for intracellular staining zombie yellow (Biolegend) was used to label dead cells.

The library was prepared based on the modified smart-seq2 protocol with modifications to sort 1000 cells instead of single cell. For each group, cerebral cortex isolated from 2-3 male mice were pooled together, microglia were sorted on an Aria III cytometer (BD Biosciences) as 7-AAD^-CD45⁺CD11b⁺ GFP⁺ cells, 1000 microglia were sorted directly into a 0.2-ml PCR tube containing 6 μl lysis buffer. After that, RNA was reverse transcribed into cDNAs, which were subjected to PCR amplification for 14 cycles. The libraries were sequenced by Illumina HiSeq X Ten System. Sequencing data were mapped to Gencode M20 using Salmon (version 0.8.1) (69) and gene expression matrix was processed using the DESeq2 (version 1.22.2) (70) and visualized by showing the first 3 dimensions calculated by plotPCA function. All differential expression analyses were performed by DESeq2 according to the software manual. The standard KEGG pathway genesets are publicly available at https://www.genome.jp/kegg/pathway.html. Gene Set Enrichment Analysis (GSEA) was performed using GSEA_4.2.1 software. The KEGG gene set from MSigDB database was chosen as the reference gene set. Pathways with FDR < 0.05 and p < 0.05 were identified as significant. Bubble plots were generated in R (https://www.r-project.org) using the ‘ggplot’ function in the ggplot2 package. The differentially expressed genes with Log₂FoldChange (mPTEN^KO/mPTEN^WT) and P-Value were used as inputs. The color bar represents Log₂FoldChange (mPTEN^KO/mPTEN^WT) values and the size of each dot represents the P-Value. KEGG pathways of upregulated differentially expressed genes were enriched by using DAVID.

Brains were removed from mice that were sequentially perfused with PBS and 4% paraformaldehyde, post-fixed overnight, dehydrated in 30% sucrose and embedded in OCT (Sakura). Cryosections (40 μm) were blocked in the staining buffer (PBS containing 5% serum and 0.3% Triton-X) for 2 hrs at room temperature, and incubated with primary antibodies diluted in the same buffer overnight at 4°C. Sections were washed and further stained with the secondary antibody in the dark for 2 hrs at room temperature. The sections were counterstained with DAPI (Biotium), washed and mounted with the ProlongGold Antifade reagent (Invitrogen). Images were collected using an Olympus FV1000 upright microscope (20×) or a Nikon A1Rsi spectral confocal system and Ti-E inverted research microscope (100×). The primary antibodies included rabbit anti-Iba1 (catalog: 019-19741, Wako), goat anti-Iba1 (catalog: ab5076, Abcam), rat anti-CD68 (catalog: MCA1957, Bio-Rad), rabbit anti- VGLUT1 (catalog: 13502, Synaptic Systems), mouse anti-NeuN (catalog: MAB377, millipore), mouse anti-PSD95 (catalog: MAB1596, millipore), rabbit anti Olig2 (catalog: ab109186, Abcam), rabbit anti-S100(catalog: ab52642, Abcam), chicken anti-GFP (catalog: GFP-1010, AVES). Secondary antibodies included Alexa Fluor 488-, 555-, 647-conjugated secondary antibodies against rabbit, goat, mouse, chicken or rat IgG (Life Technologies). For 3D reconstruction, images were captured on a Nikon A1Rsi spectral confocal microscope with 100× lens at a 1024×1024 XY resolution and a 0.2-μm Z-step and analyzed with the Imaris software (Bitplane). Microglia including cellular processes were reconstructed automatically in Imaris, verified manually, and subjected to morphometric analyses by the Filament tracer module. For certain experiments, VGLUT1-positive puncta and the VGLUT1/PSD95 colocalization were quantified using the Spot module of Imaris software. Two spots are considered co-localized when the distance between centers of the two spots is less than the diameter of either spot. A spot diameter of 0.2 μm is empirically determined to be suitable for PSD95 and VGLUT1, and this is applied to automatic analyses of all samples to ensure no human bias was introduced. The inclusion of VGLUT1 in CD68+ lysosomes of microglia were analyzed using the Spot and Surface modules of Imaris software.

Mice were perfused with PBS. Isolated cortex pooled from 2 mice per sample were homogenized with a dounce homogenizer in 5ml 0.32M sucrose in 4mM HEPES (pH 7.4) and centrifuged at 1000 g for 10 min at 4°C to remove nuclei and debris. The supernatant was transferred to a new tube and centrifuged at 30000 g for 10 min at 4°C, the crude synaptosome pellet was resuspended in 1 ml of 0.32M sucrose and then layered on top of a discontinuous sucrose gradient (1.0 and 1.2M sucrose) and centrifuged at 120,000 g for 2 hours at 4°C. The synaptosomal fraction was isolated from the gradient interface and diluted into 0.32M sucrose and centrifuged at 200,000 g for 30 min at 4°C. The resulting synaptosomes in the pellet were resuspended in 50 mM HEPES/2mM EDTA solution. The quality of synaptosome preparation was verified by enrichment of synaptic proteins as assessed by Western blotting.

Proteins from isolated synaptosomes were loaded on 10% SDS-PAGE gel and then transferred to PVDF membranes, blocked in 5% nonfat milk, and incubated with primary antibodies, which included rabbit anti VGLUT1 (Synaptic Systems), mouse anti-PSD95 (Millipore), mouse anti-Gephyrin (Synaptic Systems), rabbit anti GluN2B (Synaptic Systems), rabbit anti-Synaptophysin (Abcam) overnight at 4°C. Lysates from microglia were incubated with the primary antibody rabbit anti PTEN (Cell Signaling Technology). Membranes were then incubated with HRP-conjugated secondary antibodies, and immunoblots were developed with enhanced chemiluminescence (Thermo Fisher Scientific). Images were analyzed with ImageJ software.

Microglia freshly isolated from P7 mice were rested in DMEM (Thermo Fisher Scientific, Gibco) supplemented with 2% FBS for 45 min at 37°C and then incubated in the same media containing 0.2 mg/ml pHrodo Zymosan BioParticles (Life Technologies) for 1.5 h at 37°C. After incubation cells were washed and analyzed by FACS.

The sample size was not statistically pre-determined but empirically estimated to be sufficient in accord with standards in the field. No randomization in our animal experiments was necessary, because only age- and sex-matched mice of different genotypes were compared. Experimenters were blinded to the mouse genotype during behavioral tests and electrophysiological recordings. Statistical analyses were performed using Graphpad Prism. Unpaired Student’s t tests were used to compare different groups unless indicated otherwise. Specifically, paired Student’s t tests were performed to analyze data collected in NORT, three-chamber tests ( Figures 1C-E , Figures 4D-F ) and ex vivo phagocytosis assay ( Figure 3C ). Mann–Whitney U-tests were performed to analyze data that show skewed distributions, including the amplitude and area of mEPSC in P14 mice ( Figure 5L ) collected by whole-cell recording in brain slices.