Elevated protein synthesis in microglia causes autism-like synaptic and behavioral aberrations

Mutations that inactivate negative translation regulators cause autism spectrum disorders (ASD), which predominantly affect males and exhibit social interaction and communication deficits and repetitive behaviors. However, the cells that cause ASD through elevated protein synthesis resulting from these mutations remain unknown. Here we employ conditional overexpression of translation initiation factor eIF4E to increase protein synthesis in specific brain cells. We show that exaggerated translation in microglia, but not neurons or astrocytes, leads to autism-like behaviors in male mice. Although microglial eIF4E overexpression elevates translation in both sexes, it only increases microglial density and size in males, accompanied by microglial shift from homeostatic to a functional state with enhanced phagocytic capacity but reduced motility and synapse engulfment. Consequently, cortical neurons in the mice have higher synapse density, neuroligins, and excitation-to-inhibition ratio compared to control mice. We propose that functional perturbation of male microglia is an important cause for sex-biased ASD. The main cell types involved in autism spectrum disorders through elevated protein synthesis are not well identified. Here, the authors show that overexpression of translation initiation factor eIF4E in microglia results in autism-like behaviour in male, but not female, mice.

A utism spectrum disorders (ASD) are a group of neurodevelopmental disorders with deficits in two core domains: social interaction and communication, and repetitive or restrictive behaviors 1,2 . ASD are frequently associated with comorbidities, such as intellectual disability and anxiety 3,4 . Prevalence of ASD is 1 in 59 children under 8 years of age, and males are four times more likely than females to be identified with ASD 5 . The biological basis for this sex bias is not clear. There is a strong genetic basis for ASD [6][7][8] . One recognized common biochemical pathway underlying ASD is dysregulation of protein synthesis 3 .
Eukaryotic mRNAs contain a modified guanosine (m 7 Gppp), termed a cap, at their 5′ end. Translation of mRNAs requires binding of a translation initiation factor eIF4E to the cap. This factor interacts with the scaffolding protein eIF4G, which is essential for formation of the translation initiation complex by bridging an mRNA to the ribosome 9 . Hypophosphorylated forms of eIF4E-binding proteins (4E-BPs) repress translation initiation by sequestering eIF4E and consequently disrupting formation of the translation initiation complex 9,10 . Although most mRNAs need basal amounts of eIF4E to be translated, eIF4E preferentially increases translation of a selective set of mRNAs with extensive secondary structures at their 5′ untranslated regions 11 . The mammalian target of rapamycin complex 1 (mTORC1) phosphorylates 4E-BPs and releases eIF4E from the 4E-BP complex, thus stimulating translation 9,10 .
There are four main negative regulators upstream of mTORC1: PTEN (phosphatase and tensin homolog), TSC1 (tuberous sclerosis complex 1), TSC2 and neurofibromatosis type 1 3 . Inactivating mutations in the genes for these proteins cause ASD in a subset of patients 12,13 . Furthermore, fragile X syndrome is caused by silencing of the FMR1 gene, which encodes fragile X mental retardation protein (FMRP) 14 , and the mTORC1-eIF4E pathway is over-activated in fragile X syndrome patients diagnosed with ASD 15 . These single-gene disorders account for over 3% of all ASD cases 3 . These discoveries suggest that elevated translation might cause ASD in a subset of individuals. Importantly, a causal relationship between elevated translation and ASD-like behaviors has been recently established in mice. Deletion of the 4E-BP2coding Eif4ebp2 gene or overexpression of eIF4E under the promoter of beta tubulin (βT-Eif4e) increases protein synthesis in the mouse brain and leads to ASD-like behaviors 16,17 .
Microglia are derived from myeloid progenitors generated in the yolk sac and migrate into the brain when neurons and other glial cells only begin to be produced during early embryogenesis 18,19 . After the closure of the blood brain barrier, microglia self-renew without the contribution of circulating monocytes in the healthy brain 20,21 . Until recently, the primary function of microglia was thought to migrate to inflammation sites and engulf debris from dead or dying cells 22 . Recent work indicates that homeostatic microglia also play important roles in synaptic development and function [23][24][25][26] .
As synthesis of synaptic proteins is necessary for enduring synaptic plasticity and synaptic dysfunction can lead to ASD, it has been proposed that inactivating mutations in negative translation regulators cause ASD by enhancing translation of mRNAs in neurons 27,28 . Because mRNA translation is elevated in all cells of the body in Eif4ebp2 knockout and transgenic βT-Eif4e mice 16,17 , it remains, however, to be determined whether translational dysregulation in neurons is sufficient to cause ASD and whether translational dysregulation in glial cells contributes to autism manifestations. In this study, we elevated mRNA translation by overexpressing eIF4E and showed that exaggerated translation in microglia is sufficient to cause ASD-like phenotypes in mice via its detrimental impact on microglia-neuron interactions.

Results
Intact sociability in mice with elevated neuronal translation. We generated a conditional eIF4E overexpression allele at the Rosa26 locus (R26 Eif4e ) that expresses Myc-tagged eIF4E in a Credependent manner ( Supplementary Fig. 1a, b). eIF4E-Myc interacted with eIF4G as efficiently as untagged eIF4E (Supplementary Fig. 1c), indicating that the Myc tag does not interfere with the function of eIF4E. As it is generally believed that elevated protein synthesis in neurons causes ASD-like behaviors 27 , we first overexpressed eIF4E in neurons by crossing R26 Eif4e mice with Syn1-Cre mice, which selectively express Cre in neurons as early as embryonic day 12.5 29 , to generate Syn1-Cre;R26 Eif4e/Eif4e mice (termed NN 4E mice thereafter) (Fig. 1a). Levels of total eIF4E (eIF4E + eIF4E-Myc) in the NN 4E hippocampus were more than twice as high as those in control mice (Fig. 1b). Interestingly, neuronal transgenic eIF4E expression significantly reduced endogenous eIF4E levels ( Fig. 1b and Supplementary Fig. 2a). In fact, we found that one copy of the R26 Eif4e allele was not sufficient to significantly increase levels of total eIF4E in the brain due to this negative feedback regulation (1.01 ± 0.10 for Syn1-Cre; R26 Eif4e/+ vs. 1.00 ± 0.09 for R26 Eif4e/+ , p = 0.92, n = 5 mice per genotype). Using surface sensing of translation (SUnSET) which measures incorporation of puromycin into nascent polypeptides 30 , we found that protein synthesis rate was doubled in cultured NN 4E hippocampal neurons over control neurons (Fig. 1c). These results indicate that neuronal protein synthesis is increased in NN 4E mice.
We assessed if elevated protein synthesis in neurons leads to abnormal behaviors in mice. As revealed in rotarod and open-field tests ( Supplementary Fig. 2b, c), NN 4E mice had intact motor function. We examined anxiety-like behaviors by performing open field, elevated plus maze, and light-dark box tests. Anxious mice tend to avoid open, exposed, and brightly illuminated areas. Although NN 4E and control mice spent comparable time in the light chamber in light-dark box tests ( Supplementary Fig. 2e), NN 4E mice of both sexes spent significantly less time in the center of the open field than control mice in the first 5 min of a 30-min open field test (Fig. 1d). In addition, male NN 4E mice stayed longer in the closed arm of an elevated plus maze than control mice ( Supplementary Fig. 2d). These results indicate that elevated neuronal protein synthesis increases anxiety. In agreement with the requirement of protein synthesis for long-term memory 31 , NN 4E mice showed more freezing in the contextual chamber 24 h after fear condition training than control mice ( Supplementary  Fig. 2g). Conversely, working memory was unaffected in NN 4E mice as measured by T-maze alternation ( Supplementary Fig. 2f). We then ran marble burying and 3-chamber sociability tests to determine if elevated neuronal protein synthesis leads to repetitive behaviors and deficits in social interaction. Although female NN 4E mice showed increased repetitive behaviors in marble burying tests relative to control mice (Fig. 1e), male NN 4E mice were unaffected ( Supplementary Fig. 2h). Unexpectedly, NN 4E mice of both sexes stayed longer in the chamber with a holder holding a stranger mouse than in the chamber with an inanimate object (empty holder) and spent more time in investigating the stranger mouse than the empty holder, as did control mice in 3-chamber sociability tests (Fig. 1f, g). These behavioral results indicate that elevated neuronal protein synthesis does not lead to deficits in social interaction associated with ASD. This suggests that social interaction impairments observed in Eif4ebp2 knockout and transgenic βT-Eif4e mice 16,17 are attributable to elevated protein synthesis in glial cells.
Elevated microglial translation leads to ASD-like behavior. We next overexpressed eIF4E in astrocytes ( Supplementary  Fig. 3a-c), which is the largest population of glia in the brain, using a Cre transgene driven by the astrocyte-specific promoter for glial fibrillary acidic protein (GFAP-Cre) 32 which becomes active during late embryogenesis and the first postnatal week 33 . We found that astrocytic eIF4E overexpression did not alter repetitive and social behaviors in mice of either sex (Supplementary Fig. 3d-f). Thus, we shifted our efforts to microglia, which make up~10% of brain cells 34 .
We employed Cx3cr1 CreER mice 25 to overexpress eIF4E in microglia. To determine the efficiency for tamoxifen to activate Cre recombinase in microglia, we crossed Cx3cr1 CreER/+ mice to Rosa26 Ai9/+ mice 35 to generate Cx3cr1 CreER/+ ;Rosa26 Ai9/+ mice. A single tamoxifen injection (180 mg/kg, subcutaneously) was administered to Cx3cr1 CreER/+ ;Rosa26 Ai9/+ mice at P0. Tamoxifen activated CreER to excise the loxP-flanked transcription blocker of the Rosa26 Ai9 allele in microglia, leading to tdTomato expression ( Supplementary Fig. 4a). There is nearly 100% colocalization (n = 3 mice) between tdTomato and the microglial marker, ionized calcium binding adapter molecule 1 (Iba1), in the hippocampus and cortex ( Supplementary Fig. 4b), indicating extremely high specificity in the expression of the Cx3cr1 CreER gene. Furthermore, the recombination activity of the CreER protein is tightly controlled by tamoxifen, as we did not detect tdTomato expression in Cx3cr1 CreER/+ ;Rosa26 Ai9/+ mice in the absence of tamoxifen injection ( Supplementary Fig. 4c). This observation indicates that a single tamoxifen injection in newborn pups is sufficient to activate Cre recombinase selectively in microglia of Cx3cr1 CreER mice.
Newborn pups from R26 Eif4e/Eif4e × Cx3cr1 CreER/+ ;R26 Eif4e/Eif4e crosses were treated with tamoxifen (180 mg/kg, subcutaneously) to generate control (R26 Eif4e/Eif4e ) and MG 4E (Cx3cr1 CreER/+ ; R26 Eif4e/Eif4e ) mice that should express eIF4E-Myc in microglia (Fig. 2a). Immunoblotting analysis of brain lysates revealed eIF4E-Myc expression in MG 4E mice ( Supplementary Fig. 4d). Immunohistochemistry revealed that eIF4E levels were lower in microglia than in neurons in control mice, but the transgenic overexpression increased microglial eIF4E levels more than two folds ( Supplementary Fig. 5a). To further assess microglial eIF4E   6). Analysis of purified microglia revealed that two copies of Cre-excised R26 Eif4e alleles increased levels of total eIF4E by 21-28% and protein synthesis rate by 28-35% in both male and female MG 4E microglia compared to control microglia (Fig. 2b). Interestingly, eIF4E-Myc expression did not reduce endogenous eIF4E levels in microglia (Fig. 2b), as it did in neurons ( Supplementary Fig. 2a) and astrocytes ( Supplementary  Fig. 3c). These results indicate that transgenic eIF4E-Myc    expression elevates microglial protein synthesis to a similar extent in both sexes of mice. There is a difference in the extent of microglial eIF4E overexpression between the two aforementioned assays. To understand the discrepancy, we established cultures enriched for neurons, astrocytes, or microglia. To our surprise, eIF4E levels in cultured microglia were higher than those in cultured neurons ( Supplementary Fig. 5b), although eIF4E levels are higher in neurons than in microglia in vivo on the basis of immunohistochemistry ( Supplementary Fig. 5a). This result indicates that cell culturing somehow upregulates microglial eIF4E expression, which should cause underestimation of transgenic eIF4E expression. Therefore, it is likely that microglia in MG 4E mice have more than twice as much eIF4E and protein synthesis as those in control mice, as do neurons in NN 4E mice and astrocytes in AC 4E mice.
Microglia are the most abundant CNS macrophages, which also include perivascular macrophages, meningeal macrophages and choroid-plexus macrophages 37 . In addition to CNS macrophages, CX3CR1 is expressed in peripheral monocytes and inflammatory macrophages 25,37 . In contrast to fast turnover of peripheral CX3CR1-expressing cells, CNS macrophages are longliving 37 . We ran behavioral tests when MG 4E mice were at 2-3 months of age when the peripheral monocytes present at the time of tamoxifen injection should have been turned over, so that peripheral cells express little or no elF4E-Myc in tested mice ( Supplementary Fig. 5c).
MG 4E mice had normal locomotion in open field tests, indicating normal motor function ( Supplementary Fig. 7a). They did not show anxiety-like behaviors in open field tests (Fig. 2c), elevated plus maze tests, or light-dark box tests ( Supplementary  Fig. 7b, c). Interestingly, MG 4E mice displayed sexual dimorphism in 3-chamber sociability tests. Male MG 4E mice displayed social interaction deficits, as they spent similar amounts of time investigating a stranger mouse and an inanimate object (Fig. 2d), whereas female MG 4E mice had normal social interaction behaviors (Fig. 2e). To further characterize social behaviors of male MG 4E mice, we performed social recognition tests in the home cage-like environment, an assay which relies on the mouse's innate tendency to investigate a novel social partner and decrease the investigation of a known social stimulus (social habituation) 38 . Consistent with the social interaction deficit observed in 3-chamber sociability tests, male MG 4E mice spent less time investigating a novel stimulus mouse in trial 1 (Fig. 2f). Furthermore, social habituation (trial 4 vs. trial 1) and dishabituation (novel stimulation mouse vs. trial 4) were impaired in male MG 4E mice (Fig. 2f, g), indicating that their ability to adapt to novel social stimuli was attenuated, a phenomenon which has been linked to the pathognomonic social impairment and behavioral rigidity in ASD patients 39 . Again, female MG 4E mice performed normally in social recognition tests ( Supplementary  Fig. 7f, g). These results indicate that elevated protein synthesis in microglia leads to social interaction deficits only in male mice.
We evaluated learning and memory in MG 4E mice by conducting contextual fear conditioning and novel object recognition tests. Both sexes of control and MG 4E mice showed similar long-term hippocampus-dependent contextual fear memory ( Supplementary Fig. 7d). However, male, but not female, MG 4E mice spent less time investigating a novel object than control littermates (Fig. 2h, i and Supplementary Fig. 7h), indicating recognition memory deficits in male MG 4E mice. Thus, male MG 4E mice exhibit selective deficits in cognitive functions.
To determine whether MG 4E mice display repetitive behaviors, one of the core domains of ASD, we performed marble-burying and self-grooming tests. Although both sexes of MG 4E mice performed normally in marble-burying tests ( Supplementary   Fig. 7e), male, but not female, MG 4E mice spent significantly more time grooming than control mice ( Fig. 2j and Supplementary Fig. 7i).
One copy of the Cx3cr1 gene is disrupted in Cx3cr1 CreER/+ mice 25 . To rule out the possibility that the observed ASD-like phenotype in MG 4E mice results from Cx3cr1 haploinsufficiency, we examined the behaviors of WT and Cx3cr1 CreER/+ mice in three-chamber sociability tests ( Supplementary Fig. 8a-d) and novel object recognition tests ( Supplementary Fig. 8e-j). Both male and female Cx3cr1 CreER/+ mice were normal in these behavioral tests.
Collectively, these results indicate that elevated protein synthesis in microglia leads to ASD-like phenotypes in mice, including male bias, deficits in social interaction, increased repetitive behaviors, and impaired cognitive functions.
Elevated protein synthesis alters microglial morphology. We next sought to investigate why microglial eIF4E overexpression increased protein synthesis in both sexes, but only caused ASDlike behaviors in males. We performed Iba1 immunohistochemistry to examine the impact of elevated protein synthesis on microglia in the medial prefrontal cortex (mPFC), hippocampus and striatum, the three brain regions that have been implicated in ASD pathophysiology [40][41][42] . We found that microglia in the mPFC were more numerous and larger in 2-week-old male MG 4E mice than male control littermates ( Fig. 3a-c). The same phenotype was observed in the hippocampus and striatum (Supplementary Fig. 9a-f). In contrast, elevated microglial protein synthesis did not affect the density and size of microglia in the mPFC, hippocampus and striatum of 2-week-old female MG 4E mice ( Fig. 3d-f and Supplementary Fig. 9g-l). Semi-automatic quantitative morphometric 3D measurements of microglia confirmed the increase in size and complexity of microglia in 2-week-old male MG 4E mice (Fig. 3g). By 6 weeks of age, microglia in the mPFC, hippocampus and striatum in male MG 4E mice remained larger than those of male control mice; however, their density became normal or slightly lower compared to control mice ( Inactivating mutations in the PTEN and FMR1 genes account for a large percentage of human syndromic ASD 3 . To determine if microglial changes observed in MG 4E mice also occur in mice that model these two syndromes, we measured microglial density and size in 2-week-old male Pten +/− and Fmr1 knockout (KO) mice. Microglia in the mPFC, hippocampus and striatum were significantly larger in Pten +/− and Fmr1 KO mice than WT littermates (Fig. 4a, b and Supplementary Figs. 11a, 12a). Moreover, microglial density was increased in the Pten +/− mPFC and Fmr1 KO striatum ( Fig. 4a and Supplementary Fig. 12a). The lack of elevated microglial density in all brain regions of 2-week-old male Pten +/− and Fmr1 KO mice could be due to a smaller degree of protein synthesis increase relative to MG 4E mice. We notice that microglial density in tamoxifen-treated male R26 Eif4e/Eif4e (control) mice is lower than that in untreated male WT mice at 2 weeks of age (compare Figs. 3c, 4a, b), which is possibly due to growth retardation of mice caused by tamoxifen injection at P0.
The microglial morphological phenotype observed in 2-weekold male Pten +/− and Fmr1 KO mice also displays sexual dimorphism. While microglia in 2-week-old female Pten +/− mice show some morphological alterations, the phenotype is not as prominent as observed in male Pten +/− littermates. Compared to control mice, microglia in 2-week-old female Pten +/− mice were larger in the mPFC and the striatum but not in the hippocampus, and their density was not significantly altered in all the three brain regions ( Fig. 4a and Supplementary Fig. 11b). Furthermore, the density and size of microglia was normal in 2-week-old female Fmr1 KO mice ( Fig. 4b and Supplementary Fig. 12b). Similar to what was observed in MG 4E mice, the morphological phenotype in male Pten +/− and Fmr1 KO mice is attenuated with age, so that the phenotype in these mice disappeared by 6 weeks of age (Fig. 4c, d and Supplementary Figs. 13,14).
Synaptic alterations in male MG 4E mice. As microglia play important roles in neuronal development, the observed changes in microglial number and morphology could alter synaptic development and function in male MG 4E mice, which then leads to ASD-like behaviors. To investigate this possibility, we examined synaptic structures in layer 2 of the mPFC prelimbic area (PrL) in 6-week-old male control and MG 4E mice using serial block-face scanning electron microscopy (Fig. 5a). Threedimension reconstruction of dendrites revealed that male MG 4E mice had higher spine density, with contributions from both non-synaptic and synaptic spines compared to control mice (Fig. 5b). Male MG 4E mice had smaller spines and synapses on average (Fig. 5c). To examine synaptic maturation, we counted multiplesynapse spines (MSS) and multiple-synapse boutons (MSB), which are believed to be the structural basis of synaptic multiplicity 26 . Synaptic multiplicity increases in the hippocampus during postnatal development and is considered an indicator of synapse maturation 43 . We found that male control and MG 4E mice had comparable MSS and MSB density in layer 2 of the PrL (Supplementary Fig. 15a, b), suggesting that reduced synapse size in male MG 4E mice is not indicative of immaturity.
Because dendritic spines are the postsynaptic sites for the vast majority of excitatory synapses 44 , higher spine density indicates more excitatory synapses. Indeed, transmission electron microscopy analysis revealed more asymmetric (excitatory) synapses in the PrL of 6-week-old male, but not female, MG 4E mice compared to control mice (Fig. 5d). This analysis also found a normal density of symmetric (inhibitory) synapses in the PrL of both male and female MG 4E mice (Fig. 5d). Thus, exaggerated protein synthesis in microglia increases the number of excitatory synapses in the mPFC in male mice.  Comparable microglial density and size in the mPFC between female control and MG 4E mice. n = 6 control mice and 5 MG 4E mice. Five to ten microglia from each mouse were randomly selected for measurement of cell size (cross section area). n.s. not significant by two-sided t-test. Scale bar, 100 μm. g Three-dimension reconstruction of microglia in the mPFC of 2week-old male mice. n = 7 mice per group. Three to five microglia from each mouse were reconstructed for determination of cell size (volume) and number of processes. **p = 0.0037 and ***p = 0.0005 by two-sided t-test. Scale bar, 20 μm. h-m Microglia density and size in the mPFC of 6-week-old male (h-j) and female (k-m) MG 4E mice. Ten microglia in each brain region of each mouse were randomly selected for measurement of cell size (cross section area). Male, n = 6 mice per genotype; female, n = 5 control mice and 6 MG 4E mice. Two-sided t-test: ***p < 0.001 and n.s. not significant. Scale bar, 100 µm. n Three-dimension reconstruction of microglia in the mPFC of 6-week-old male mice. Four microglia from each mouse were reconstructed. Six control mice and five MG 4E mice. *p = 0.0385 by two-sided t-test. Scale bar, 20 µm. All data are shown as mean ± s.e.m. Source data are provided as a Source Data file.
Increased spine density has been observed in ASD patients 45,46 . To determine if spine density is also increased in other brain areas of male MG 4E mice and at other ages, we employed the Thy1-GFP transgene 47 to label isolated neurons in 2-week and 6-weekold mice. We found that male, but not female, MG 4E mice had higher spine density in pyramidal neurons of mPFC layer 5 and hippocampal CA1 area at both ages ( Fig. 5e and Supplementary  Fig. 16a). These results indicate that elevated microglial protein synthesis increases the density of excitatory synapses in multiple brain areas.
We next examined whether levels of synaptic proteins are altered in the mPFC and hippocampus of MG 4E mice. We found that adult male and female control and MG 4E mice had comparable levels of presynaptic synaptophysin and postsynaptic neuroligin 3, neuroligin 4, PSD95 and GluA1 in the hippocampus and mPFC (Supplementary Figs. 15c, 15d, 16b, 16c). Neuroligins are cell-adhesion molecules important for synaptogenesis and have been implicated in ASD 48 . Levels of neuroligins 1-4 are increased in the hippocampus of Eif4ebp2 knockout and transgenic βT-Eif4e mice, which was interpreted as a result of increased mRNA translation in neurons 16 . Interestingly, levels of neuroligins 1 and 2 in the hippocampus and mPFC were significantly higher in male, but not female, MG 4E mice, compared with sex-matched control mice ( Fig. 5f and Supplementary Fig. 16b, c). This finding suggests that the observed increase in hippocampal levels of neuroligins in Eif4ebp2 knockout and transgenic βT-Eif4e mice could be a result of microglial alterations rather than increased eIF4E activity in   Female, n = 6 mice per genotype. Five to ten microglia from each condition were randomly selected for measurement of cell size (cross section area). Two-sided t-test: male, cell density *p = 0.0317, cell size **p = 0.001; female, cell density p = 0.0725, cell size *p = 0.0232, n.s. not significant. Scale bar, 100 µm. b Microglial density and size in the mPFC of 2-week-old male and female Fmr1 KO mice. Male, n = 7 mice per genotype. Female, n = 4 WT mice and 5 KO mice. 5-10 microglia from each condition were randomly selected for measurement of cell size (cross section area). Two-sided t-test: **p = 0.0046, and n.s. not significant. Scale bar, 100 µm. c, d Microglial density and size in the mPFC of 6-week-old male Pten +/− (c) and Fmr1 KO (d) mice. c n = 5 mice per condition; d n = 6 WT mice and 5 Fmr1 KO mice. Five to ten microglia from each condition were randomly selected for measurement of cell size (cross section area). n.s., not significant by two-sided t-test. Scale bar, 100 μm. All data are shown as mean ± s.e.m. Source data are provided as a Source Data file.  Fig. 17). The increase in levels of neuroligins 1 and 2 is likely due to enhanced transcription in neurons, as levels of mRNAs for the two proteins show a trend of increase in male MG 4E mice ( Supplementary Fig. 15e).
To investigate if alterations in synaptic structure and proteins lead to abnormal synaptic function in male MG 4E mice, we recorded miniature excitatory postsynaptic currents (mEPSCs) and miniature inhibitory postsynaptic currents (mIPSCs) in layer 5 pyramidal neurons of the mPFC. The amplitude, but not the frequency, of mEPSCs was increased in male MG 4E mice relative to control mice (Fig. 6a). In contrast, the amplitude of mIPSCs were slightly reduced in male MG 4E mice, as indicated by a significant but small left shift of its cumulative curve (Fig. 6b). These results suggest an alteration in the excitation/inhibition (E/ I) balance, which is believed to contribute to ASD 3 , in the mPFC. Indeed, we found that the normalized total charge transfer in male MG 4E mice relative to control mice was significantly larger for mEPSCs than for mIPSCs (Fig. 6c). As expected on the basis of synapse density and levels of synaptic proteins, female control and MG 4E mice had comparable mEPSCs, mIPSCs and charge transfer in the mPFC ( Supplementary Fig. 16d, e).
Collectively, our microscopic, biochemical and electrophysiological analyses reveal that elevated protein synthesis in microglia leads to synaptic alterations that are associated with ASD in male mice only, including increased spine density, increased levels of neuroligins and E/I imbalance. These synaptic changes likely contribute to deficits in social interaction, repetitive behaviors, and cognitive impairments in male MG 4E mice.
Microglia are out of homeostatic state in male MG 4E mice. To understand how microglial eIF4E overexpression alters synaptic structure and function, we analyzed hippocampal gene expression in male control and MG 4E mice at 2 and 6 weeks of age using RNA-Seq. Weighted gene correlation network analysis (WGCNA) 49 revealed 11 modules of highly co-expressed genes in control and MG 4E mice ( Supplementary Fig. 18a). Of note, the magenta module displayed a set of transcripts that were significantly upregulated in 2-week-old MG 4E mice (Fig. 7a). Pathway enrichment analysis of genes within the magenta module revealed top KEGG pathways related to antimicrobial activities,  including herpes simplex infection, antigen processing and presentation, phagosome, influenza A and autoimmune thyroid disease (Fig. 7b, c). Some of the disease-associated microglial genes found in neurodegenerative conditions 50 were also upregulated in 2-week-old MG 4E mice (28 out of 138 genes in the magenta module; Fig. 7c). By 6 weeks of age, the magenta module eigengene was no longer significantly upregulated in MG 4E mice, suggesting that this set of transcripts were only upregulated during postnatal development. Interestingly, the herpes simplex virus infection and autoimmune thyroid disease pathways in the magenta module have been reported to be risk factors for ASD [51][52][53] . In addition to the upregulated genes ( Supplementary Fig. 18b, c), we identified a group of downregulated genes by comparing 2-week-old male MG 4E mice to control littermates ( Fig. 7d and Supplementary  Fig. 18b, d). Many of the downregulated genes (11/40, or 27.5%) are markers for homeostatic (M0) microglia 54 , including P2ry12, P2ry13, Cx3cr1, Tmem119, and Slco2b1 (Fig. 7e).
In addition to M0 state, microglia can achieve two other polarized states (M1 and M2) that are characterized by specific molecular signatures 54,55 . It is proposed that M1 microglia produce pro-inflammatory cytokines and M2 microglia are antiinflammatory 55 . We found that very few M1 and M2 markers were upregulated in 2-week-old male MG 4E mice (Fig. 7f, g). This indicates that elevated protein translation does not move microglia to M1 or M2 polarized states.
We validated the RNA-Seq results by measuring mRNA levels of a few select genes in 2-week and 6-week-old control and MG 4E mice using quantitative RT-PCR (Fig. 7h-j). The RT-PCR analysis also found that the observed gene expression changes were absent in 2-week-old female MG 4E mice ( Supplementary  Fig. 18e). Taken together, the RNA-Seq and RT-PCR analyses show that elevated protein synthesis shifts microglia from the homeostatic state to a functional state with properties that have been observed in neurodegenerative conditions.
Male MG 4E microglia with altered phagocytosis and motility. Microglia are the primary phagocytes in the brain that engulf invading pathogens and cellular debris 56 . Upregulation of antimicrobial pathways including the phagosome pathway (Fig. 7c) prompted us to hypothesize that the phagocytic capacity of microglia is enhanced in 2-week-old male MG 4E mice. Indeed, purified microglia from male MG 4E mice phagocytized more than twice as many Aβ  aggregates as those from male control mice, but microglia from female control and MG 4E mice had comparable phagocytosis (Fig. 8a).
Given that microglia have been shown to prune synapses through phagocytosis during postnatal development 23,24 , we were puzzled by how elevated spine density (Fig. 5b, e) and enhanced microglial phagocytosis (Fig. 8a) co-exist in male MG 4E mice. We noticed that the downregulated genes are associated with leukocyte cell-cell adhesion, leukocyte migration, and cell-cell adhesion ( Supplementary Fig. 18d). Thus, we examined migration of microglia to an injection site of Aβ  aggregates in 2week-old male control and MG 4E mice. We found that the number of microglia migrated to the Aβ  injection site was significantly reduced in male, but not female, MG 4E mice relative to control mice (Fig. 8b, c). These results indicate that elevated protein synthesis impairs the motility of microglia in male mice.
Microglial processes display two modes of motility: constant extension and retraction to survey the brain (baseline motility) and extension to the source of attractants such as ATP released from a damaged site (induced motility) 57 . Because synapse pruning involves contacts of microglial processes with synapses 58 , it should be affected by the motility of microglial processes. As elevated protein synthesis impaired the motility of whole microglia, it could also affect the motility of microglial processes. We monitored the baseline motility and ATP-induced motility of microglial processes in organotypic hippocampal cultures established from P14-P18 control and MG 4E mice harboring the CD68-EGFP transgene 59 which is selectively expressed in microglia. Irrespective of sex, microglial processes in control and MG 4E mice had comparable baseline processes motility ( Supplementary Fig. 18f). However, ATP-induced motility was impaired in male, but not female, MG 4E mice (Fig. 8d).
Synapses to be eliminated may release a factor to initiate their interactions with microglial processes, and thereby deficits in the induced motility in male MG 4E mice would impair engulfment of synapses. We assessed synapse engulfment by measuring the immunoreactivity of postsynaptic protein Homer1 inside microglia. Three-D reconstruction of Iba1 and Homer1 immunoreactivities revealed that the volume of Homer1 immunoreactivity inside microglia was significantly reduced in male, but not female, MG 4E mice relative to control mice (Fig. 8e).
Collectively, these results indicate that induced motility of microglia is impaired in male MG 4E mice, which might lead to reduced microglial engulfment of synapses and subsequently increased density of excitatory synapses, despite of enhanced microglial phagocytosis capacity in these mice.

Discussion
Our results indicate that elevated neuronal protein synthesis does not lead to deficits in social interaction, but it produces some ASD-related behaviors. NN 4E mice of both sexes displayed elevated anxiety, a common ASD comorbidities 4 . In addition, as revealed by marble burying tests, elevated neuronal protein synthesis increased repetitive behavior in female but not male mice. These behavioral phenotypes are interesting, given the observation that elevated microglial protein synthesis impaired social behaviors without elevating anxiety in male mice. These results indicate that different ASD-like and related behaviors have distinct cellular basis. They also suggest that ASD manifestations could be different in females and males.
How does elevated protein synthesis in microglia lead to ASDlike synaptic and behavioral aberrations only in male mice? Our results indicate that the key mechanism is a sex-dependent response of microglia to elevated protein synthesis. Male microglia alter their transcriptome, which leads to changes in their function and morphology, but female microglia do not. Our transcriptomic analysis reveals that eIF4E overexpression shifts microglia from the homeostatic state to an undefined functional state only in male mice. The alterations in male microglia should result from elevated translation of some mRNAs in-these cells. However, our data do not exclude a remote possibility that elevated translation in peripheral immune cells impacts microglia differently in males and females through circulating factors,      Fig. 8 Microglial surveillance and synapse engulfment in MG 4E mice. a Phagocytosis of FAM-Aβ  in cultured control and MG 4E microglia. Male, n = 18 for control and n = 28 for MG 4E ; female, n = 14 for control and n = 13 for MG 4E . **p = 0.0017 and n.s. not significant (p = 0.4919) by two-sided t-test. b, c Migration of microglia into FAM-Aβ  injection sites in 2-week-old male (b) and female (c) control and MG 4E mice. Microglia clustering index is defined as (density of Iba1 + cells in the FAM-Aβ-covered area)/(density of Iba1 + cells in a contralateral site). Male, n = 6 control mice and 7 MG 4E mice; female, n = 4 per genotype. Two-sided t-test: male mice, contralateral microglia density, **p = 0.0013; clustering index, **p = 0.0026, n.s. not significant. Scale bars, 100 μm. d Microglial surveillance in response to ATP treatment in male and female MG 4E microglia at P14-P18. Microglial baseline motility was recorded for 5 min, then ATP was bath-applied to brain slices and recorded for another 10 min. Data was normalized to the mean process motility of first 5 min. Male, n = 15 microglia from 3 control mice and 20 microglia from 3 MG 4E mice; female, n = 18 microglia from 3 control mice and 26 microglia from 3 MG 4E mice. Male, two-way ANOVA for genotype during ATP treatment, F (1, 330) = 9.566, p = 0.002; female, two-way ANOVA for genotype during ATP treatment, F (1, because monocytes and peripheral macrophages do overexpress eIF4E in MG 4E mice in the first few postnatal weeks. Homeostatic microglia are constantly surveying their microenvironment with extremely motile processes and their processes rapidly converge at the site of injury in the brain 60,61 . The motility and convergence of microglial processes in response to ATP released from damaged cells are dependent on the purinergic receptors such as P2Y12 61,62 . It is likely that homeostatic microglia employ a similar mechanism to reach synapses tagged for elimination. We found that eIF4E overexpression downregulates many genes specific to homeostatic microglia including genes for purinergic receptors (P2ry12 and P2ry13) 54 in young male mice. As this gene expression change would predict, eIF4E overexpression dramatically diminishes the induced motility of microglia. The loss of the induced motility would impair the ability of microglia to find synapses tagged for elimination so that microglia could not efficiently prune synapses. Indeed, we found that microglia in male MG 4E pups contain a lower amount of engulfed Homer1 than those in control pups. This could be the main reason for observed increases in spine density and synapse density in male MG 4E mice. This conclusion does not exclude the possibility that elevated microglial protein synthesis increases the rate of spine formation in male mice. We found that microglia in male MG 4E mice have more processes relative to control mice, which would increase points of microglial contacts with neurons. The contact has been found to induce synapse formation in the cortex 63 . It would be important to identify the proteins that lead to alterations in transcription, morphology, and function in MG 4E microglia when their synthesis is elevated and to understand why elevated synthesis of the proteins only affects male microglia in future studies.
Our RNA-Seq data indicate that microglia in male MG 4E mice upregulate expression of several cytokines, including C-X-C motif chemokine 10 (CXCL10), CXCL16, C-C motif chemokine 12 (CCL12) and interleukin 1 alpha (Il1α). We speculate that these cytokines could upregulate the expression of neuroligins and modify the function of glutamate receptors via posttranslational modifications, leading to the observed increase in the amplitude of mEPSCs. The increase in the function of excitatory synapses, along with increased density of excitatory synapses, would lead to E/I imbalance. Deficits in synapse pruning would lead to the presence of excess, imprecise and inefficient neuronal connections. Both E/I imbalance and synapse pruning impairment have been linked to ASD 3 and could be responsible for deficits in social interaction and repetitive behaviors in male MG 4E mice.
It is generally believed that PTEN haploinsufficiency elevates protein synthesis by increasing mTORC1 activity 3 . However, a recent study shows that genetic ablation of mTOR complex 2 (mTORC2), but not mTORC1, activity reverses the pathophysiology of Pten +/− mice 64 . It is likely that the complete ablation of mTORC1 activity is not the best way to reverse mTORC1 hyperactivity in Pten +/− mice. It is known that mTORC2 can fully activate Akt via Ser473 phosphorylation 65 , thereby increasing the activity of mTORC1, an Akt downstream target. Therefore, it could be a good strategy to treat ASD associated with elevated protein synthesis by targeting mTORC2 to reduce mTORC1 activity. There are conflicting reports on the effect of elevated protein synthesis on basal synaptic transmission. Enhanced mTORC1 activity via expression of a constitutively active form of Rheb was reported to reduce the frequency of spontaneous EPSCs in pyramidal neurons of the anterior cingulate cortex 66 . However, increases in both the frequency and the amplitude of mEPSCs were found in the hippocampal CA1 region of Eif4ebp2 knockout mice, where mRNA translation is elevated 16,67 . The discrepancy could result from differences in the extent of translation enhancement, brain region, and recorded synaptic property in these studies. We found that elevated protein synthesis in microglia via eIF4E overexpression increases the amplitude of mEPSCs in the mPFC without affecting the frequency of mEPSCs. It would be interesting to investigate if elevated protein synthesis in multiple types of brain cells is required to affect both frequency and amplitude of mEPSCs.
Transcriptomic analyses of post-mortem brains find that genes associated with activated microglia are upregulated in the cortex of ASD patients 68,69 . Furthermore, knockdown of the microgliaspecific Cx3cr1 gene leads to a transient reduction in microglia, deficits in synapse pruning and ASD-like behaviors 26 . These findings provide some evidence for involvement of microglia in ASD; however, ASD-associated causal genetic variants in microglial genes have not been identified yet. Protein synthesis is elevated as a result of ASD-causing mutations in genes encoding negative translation regulators (such as in PTEN, TSC1/2, and FMR1) 12,13,15 . Our results show that elevating protein synthesis in microglia leads to ASD-like synaptic and behavioral aberrations in mice. Therefore, our study suggests that ASD-associated mutations in ubiquitously expressed genes could lead to ASD through their primary effects on microglia. In addition, our findings may provide some insights into the pathophysiology of ASD associated with other risk factors. Environmental insults, such as microbial infection during pregnancy, lead to the emergence of core ASD symptoms in the adult offspring, especially in males [70][71][72][73] . Because microglia are at the interface between the brain and environment, our findings suggest that environmental insults could increase the risk of ASD by altering functional states of microglia in the offspring. We propose that augmented responses of microglia to biochemical perturbations induced by gene mutations or microbial infection are an important underlying pathogenesis for ASD. Elucidation of the sexual difference in microglial response to the perturbations in future studies could provide innovative strategies for prevention and treatment of ASD. Generation of the R26 Eif4e mouse strain. To generate a transgenic mouse strain overexpressing eIF4E from the Rosa26 locus, we generated a targeting construct by replacing the tdTomato coding sequence with the Eif4e coding sequence in the Ai9 construct 35 (a gift from Hongkui Zeng, Addgene plasmid # 22799). To distinguish between endogenous and overexpressed eIF4E, we inserted a DNA sequence for the Myc tag (gccGAACAAAAACTCATCTCAGAAGAGGATCTGaatagctag) immediately before the stop code of the eIF4E coding sequence. The targeting vector was linearized and transfected into C57BL/6 embryonic stem cells. G418-resistant ES clones were screened with Southern blots. Two positive ES clones were injected into blastocysts to obtain chimeric mice. The mouse strain was maintained on the C57BL/6 background.
Tamoxifen injections. Tamoxifen (Sigma, T5648) was dissolved in corn oil at a concentration of 40 mg/ml. P0 mice were subcutaneously injected with tamoxifen solution around the lower neck region one time at 180 mg/kg (~7 μl per mouse).
Microglial engulfment of Homer1. Coronal brain sections from 2-week-old male and female mice were incubated with rabbit anti-Iba1 (Wako, 019-19741, 1:500) and chicken anti-Homer1 (Synaptic Systems, 160006, 1:500) overnight, followed by Alexa Fluor 594 (703-585-155, Jackson ImmunoResearch, 1:500) and 649 (711-605-152, Jackson ImmunoResearch, 1:500) conjugated secondary antibodies. Images were acquired on a Nikon C2+ confocal laser scanning microscope using a 40× oil objective, laser power and gain were kept consistent throughout the experiment. Z-stack images were acquired with a 0.8-μm interval for 21 steps. Images were analyzed using Amira software (Thermo Fisher Scientific) to create a 3D surface rendering of microglia with a threshold to ensure that microglial processes were accurately reconstructed. This rendering was used to mask the Homer1 channel, and the overlapped volume was considered as Homer1 engulfed by microglia.
Cleared cell lysates were incubated with either anti-eIF4E antibody (A301-154A, Bethyl Laboratories, 1:100) or IgG on ice for 2 h. Protein G-Agarose (11719416001, Roche) was then added to each sample, and the mixture was incubated at 4°C under rotary agitation for 4 h. Immunoprecipitates were washed with the buffer five times and eluted with the SDS-loading buffer for immunoblotting analysis.
Microglial culture. Purification of primary microglia was performed using an immunopanning protocol 36 . Briefly, a mouse brain at P10 was digested with 2 ml of buffer containing 2.4 mg papain (Worthington, LS003126) and 1 mg deoxyribonuclease I (Sigma, DN25) at 37°C for 30 min. The brain was then mechanically dissociated by gentle trituration using a 1 ml pipette. Cell suspension was transferred to a 10 cm culture dish pre-coated with rat anti-mouse CD45 antibody (Invitrogen, Cat. 17045182) and incubated for 45 min at room temperature. The culture dish was then washed with Dulbecco's PBS 8-10 times. Culture medium (DMEM with 10% FBS, 1% GlutaMAX, 1% penicillin/streptomycin) was added to the dish, and the dish was returned to an incubator until harvest.
Astrocyte culture. Astrocytes were isolated from brains of control and AC 4E mice at P3. Brains were isolated and mechanically dissociated in HBSS and papain (Worthington, LS003126) for 20 min at 37°C. Afterwards, papain was inhibited with Dulbecco's modified Eagle's medium (DMEM) containing 10% FBS. Cell suspensions were subsequently plated into T75 flasks. Astrocytes were cultivated till confluency and subsequently shaken for 18 h (200 rpm). After the medium was removed, astrocytes in flasks were trypsinized and plated onto coverslips or plates for further experiments.
Tissue processing for SB-SEM. Mice were anesthetized and transcardially perfused with 2% PFA and 2.5% glutaraldehyde in 0.15 M cacodylate buffer (pH 7.4). The brains were removed and stored overnight at 4°C in fresh perfusion solution. The brains were sliced in cold 0.15 M cacodylate buffer into 150-µm coronal sections, which were used for dissection of the prelimbic area of the mPFC. Tissue blocks were placed in cacodylate buffer containing 2% OsO4/1.5% potassium ferrocyanide for 1 h. Samples were placed in 1% thiocarbohydrazide (Ted Pella) in ddH 2 O for 20 min and then in 2% aqueous OsO4 for 30 min. Tissues were incubated in 1% uranyl acetate at 4°C overnight and fresh lead aspartate solution at 60°C for 30 min to enhance membrane contrast. On the next day, samples were dehydrated using an ascending series of ethanol (50,70,90, and 100%, 5 min each), followed by ice-cold dry acetone for 10 min. Tissues were then gradually equilibrated with Epon 812 resin (EMS, Hatfield, PA, USA). To enhance specimen conductivity 75 , samples were embedded in Epon 812 resin containing 7% (w/v) carbon black powder (a kind gift from Dr. Nobuhiko Ohno, Jichi Medical University, Tochigi-ken, Japan), mounted on aluminum rivets, and then cured in a dry-oven at 60°C for 2 days.
Acquisition of SB-SEM datasets and analysis. A tissue block was trimmed flat using a glass knife. Silver paste was applied to the specimen to ground the resin to the aluminum pin. The pin was then coated with 10 nm of gold-palladium in a sputter coater to further enhance conductivity. Specimens were imaged with a Merlin VP field emission scanning electron microscope (Carl Zeiss) equipped with 3View2 in-chamber ultramicrotome technology and a backscattered electron detector (Gatan). Serial images were acquired with a 30-μm aperture, high vacuum, acceleration voltage of 2.5 kV, image size of 5000 by 5000 pixels, dwell time of 3.5 µs, and X-Y resolution of 10.8 nm at a nominal thickness of 50 nm.
Eleven stacks (5 stacks for control mice, 6 stacks for MG 4E mice, one stack per mouse) of 500 serial images were obtained for layer 2 of the prelimbic area of the mPFC. The serial images obtained were processed with Image J and Fiji plugins TrakEM2 software (http://fiji.sc/wiki/index.php/Fiji). We randomly selected spiny dendritic segments that could be followed within the stack and whose linear lengths was at least 15 μm. To make sure that our analysis was restricted to pyramidal neurons, we avoided all dendritic segments with few or no spines. A synapse was defined by the presence of a presynaptic bouton with at least three synaptic vesicles within a 50 nm distance from the membrane facing a postsynaptic density (PSD). All dendritic spines were divided into spines with synapses and spines lacking synapses. In total, 70 dendritic branches were segmented in L2. Dendritic branches and all of their protrusions were segmented manually in the reconstruction software (https://synapseweb.clm.utexas.edu/software-0) by three trained annotators who were blind to genotype.
Transmission EM analysis. Both male and female MG 4E and control mice were anesthetized and perfused with 2% PFA and 2.5% glutaraldehyde in 0.15 M NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-15530-3 ARTICLE NATURE COMMUNICATIONS | (2020) 11:1797 | https://doi.org/10.1038/s41467-020-15530-3 | www.nature.com/naturecommunications cacodylate buffer (pH 7.4). The brains were sliced into 150-µm coronal sections using a vibratome and the prelimbic area of the mPFC was dissected out. Tissue blocks were prepared using standard procedures for TEM 76 . Tissues were postfixed in 2% OsO4 for 1 h, en bloc stained with 1% uranyl acetate, dehydrated, and then embedded in Epon 812 resin (EMS). Ultrathin (70-nm-thickness) sections were collected on the 200-mesh grids and post-stained with uranyl acetate followed by lead citrate. For each animal, 30 images were recorded on a Tecnai F20 TEM (FEI) (2500× magnification, 120 kV). Counting frame (6 × 6 μm 2 ) was placed on each image and synapses with a PSD as well as clusters of synaptic vesicles were identified. Synapses were classified into either asymmetric (Gray's type 1) or symmetric (Gray's type 2) based on their distinct PSD shapes. Synaptic density values were averaged to produce animal means. All analyses were performed blind to experimental conditions. Stereotaxic injection of FAM-Aβ. FAM-Aβ  peptides (Anaspec, AS-23525) were dissolved in trifluoroacetic acid (Sigma, T6508), lyophilized in a SpeedVac, and resuspended in PBS/DMSO solution at 37°C overnight to a final concentration of 2 μg/μl. FAM-Aβ aggregates (500 nl) were stereotaxically injected into hippocampi of control and MG 4E mice at P14 using the following coordinates Aβ phagocytosis assay. FAM-Aβ  aggregates were added to primary microglial culture at a concentration of 1.8 μg/ml and incubated for 4 h at 37°C. Extracellular FAM-Aβ  was quenched with 0.2% trypan blue in PBS for 1 min. FAM-Aβ  fluorescence signal intensity was measured at 485 nm excitation and 538 nm emission using a CLARIOstar (BMG LABTECH) microplate reader.
Experimenters were blind to genotype. Signals were acquired with Multiclamp 700B and Digidata 1550A (Molecular Devices, San Jose, CA). Data were low-pass filtered at 2.9 kHz and sampled at 10 kHz. mEPSCs and fEPSPs were analyzed with Clampfit 10.6 software. Detection threshold was set at 5 pA (mEPSC) or 6 pA (mIPSC), and the first 200 events were sampled per neuron. Total charge transfer was calculated by summing the charge transfer of all individual events detected over the first 2-min acquisition period for each neuron.
SUnSET. 400-μm-thick hippocampal slices were obtained using a vibratome and recovered in aCSF for 2 h at 32°C. Puromycin (10 μg/ml) was applied to slices for 1 h to label newly synthesized proteins. To measure protein synthesis in microglia, puromycin (10 μg/ml) was applied to cultured microglia for 1 h. Tissues and cultured microglia were then lysed and used for Western blot. Mouse anti-puromycin antibody (1:2000, Millipore, MABE343) was used to detect puromycinincorporated proteins.
Behavioral tests. Male and female mice at 2-month-old of age were tested during the dark (active) phase of a 12/12 h reversed light-dark cycle. Mice were moved to a holding room in the behavioral testing area at least 1 h ahead of behavioral assays. Apparati were cleaned with 1% Micro-90 between each trial. Automatic scoring was performed using the Ethovision XT video tracking system (Noldus, Netherlands). Mice were tested in batteries with at least 3 days between assays. Details of these paradigms are listed below.
For open field test, a mouse was placed in the open field arena (43.8 cm × 43.8 cm × 32.8 cm) for 30 min. Total distance moved, center zone duration and velocity were recorded automatically.
For elevated plus maze test, a mouse was placed in the center of the maze to start the test. Each trial lasts for 5 min. Total distance moved and durations in open and closed arms were recorded automatically.
For light-dark box test, a mouse was placed in the light chamber and allowed to explore for 5 min. Latency to enter the dark chamber, time in the light chamber and number of crossings between chambers were automatically recorded.
Spontaneous alternation in T-maze was used to assess working memory in mice. Briefly, mice received two successive non-rewarded trials in one day. On trial one mice were allowed to enter one of the two unfamiliar arms. After staying 10 s in the chosen arm, mice were removed from the maze and placed back on the start arm for the second trial. A correct choice was to explore the arm that was not previously explored.
Mice were placed on a rotating rod (ENV-577M, Med Associates Inc.) for rotarod tests. The speed of rotation was gradually increased from 4 to 40 rpm over a 5-min period. Mice received three trials per day for two days, and each trial was spaced at least 1 h apart. The latency to fall was measured.
For fear conditioning test, a mouse was placed into a Phenotyper chamber (29.2 cm × 29 cm × 30.5 cm, Noldus) equipped with an electrified floor and speaker. Training consisted of a 150 s baseline followed by three 0.75-mA foot shocks. Mice were tested for contextual fear memory 24 h after training by returning to their training chambers for 5 min. Freezing behavior was automatically recorded by using the Ethovision XT video tracking system.
For marble burying test, mice were placed individually in a cage filled with 5 cm of corncob bedding and 20 black marbles arranged on top. The test lasted for 30 min, and the number of marbles that were at least two-thirds buried at the end of a trial were counted 38 .
For self-grooming test, mice were placed into a new cage with fresh bedding but no cardboard or nesting material. Self-grooming behavior was video recorded for 30 min and scored manually afterward. Experimenters were blind to genotype during scoring.
A three-chamber arena was used to assess sociability in mice 77 . A test mouse was placed in the central chamber and allowed to explore the empty apparatus for 5 min. Immediately after habituation, a stimulus mouse (same-sex, similar age, and unfamiliar mouse) was introduced into a wire cage located in one of the two side chambers. An identical empty wire cage was placed in the other side chamber. The test mouse was allowed to explore the arena for 10 min. Time spent in each chamber and time spent in sniffing the stimulus mouse (or empty wire cage) was recorded by Ethovision XT video tracking system. In addition to analyzing the time spent in each chamber or investigation time, we also calculated the social preference index, which represents the difference between time spent in investigating the stimulus mouse vs. object, divided by total time spent in investigating both targets.
For social recognition test, a tested mouse was housed alone for 2 h in a home cage-like environment 38 . A same sex, juvenile (3-4 weeks old) stimulus mouse was placed into an acrylic tube (7.25 cm in diameter, 12.5 cm tall) with holes at the bottom. Social recognition test consisted of 5-min stimulus presentations separated by 10-min inter-trial intervals; the first four presentations (H1-H4) were of the same stimulus mouse, with the fifth using a novel juvenile (non-littermate juvenile). Time spent in sniffing the stimulus mouse was manually scored from video. Habituation index is calculated using the formula: (H1 investigation time -H4 investigation time)/total H1 and H4 investigation time.
Novel object recognition was conducted in a three-chamber arena. A test mouse was placed in the middle chamber and allowed to explore the two side chambers with two identical objects located in each side chamber for 10 min. Immediately after this, one object was replaced with a novel object and the test mouse was allowed to explore for another 10 min. Time spent investigating the object (nosepoint tracking) was automatically recorded by Ethovision XT video tracking system. Discrimination index is defined as (novel object timefamiliar object time)/total time on two objects.
RNA-Seq analysis. Total RNA was extracted from dissected hippocampal tissues using a Quick-RNA Miniprep Kit (Zymo Research, R1054). RNA was quantified using the Qubit 2.0 Fluorometer (Invitrogen, Carlsbad, CA) and run on the Agilent 2100 Bioanalyzer RNA nano chip (Agilent Technologies, Santa Clara, CA) for quality assessment. All RNA samples were of excellent quality with RNA Integrity Number (RIN) > 8.0 and were processed for mRNA_seq library preparation. The final mRNA libraries were validated on the bioanalyzer DNA chips, normalized to 1 nM, pooled equally, loaded onto the NextSeq 500 (Illumina) at a final concentration of 1.8 pM, and sequenced using 2 × 40 bp paired-end chemistry. Library preparation and sequencing was performed at the Scripps Florida Genomic Core.
Reads were trimmed off for sequencing adapter and were mapped to the mouse genome using the STAR version 2.5.2a aligner. Differential gene expression analysis was performed with genes with CPM > 0.5 in at least four samples, using Bioconductor packages edgeR 78 with false discovery rate (FDR) threshold <0.05 (http://www.bioconductor.org). Weighted gene co-expression network analysis (WGCNA) was performed to find modules of highly correlated genes via its R package 49 . WGCNA clustering, using the "blockwiseModules" function with the network type "signed hybrid", a power parameter of 5 (as established by scale-free topology network criteria), and minModSize = 50, minKMEtoStay = 0.3, deepSplit = 2, mergeCutHeight = 0.15, dissected the data into 11 modules. The gene set enrichment and pathway analysis were performed using online tools DAVID 79 (https://david.ncifcrf.gov) and Generic GO Term Finder 80 .
Quantitative reverse transcription PCR. Mouse hippocampi were collected at 2-week and 6-week of ages. Total RNA was extracted from Trizol homogenates following the manufacturer's protocol. Following DNaseI (New England Biolabs) treatment, cDNA was synthesized using M-MuLV reverse transcriptase (New England Biolabs) with oligo dT primers. Candidate genes were then quantified by real-time PCR using SYBR Green Mix (Roche). The actin gene ACTB was used as control. The PCR primers for each gene are listed (Supplementary Table 1).
Two-photon imaging of microglial motility and data analysis. We performed two-photon imaging to visualize microglial motility according to a previously described protocol 81 . Briefly, acute brain slices (300 µm thick) were prepared from R26 Eif4e/Eif4e ;CD68-EGFP (control) or Cx3cr1 CreER/+ ;R26 Eif4e/Eif4e ;CD68-EGFP (MG 4E ) mice at P14-P18 using a vibratome (Leica VT 1200S, Germany) and transferred to oxygenated artificial cerebrospinal fluid (aCSF) composed of (mM): 124 NaCl, 3 KCl, 26 NaHCO 3 , 1.25 NaH 2 PO 4 , 1 MgSO 4 , 2 CaCl 2 , and 10 D-glucose. Brain sections were recovered in aCSF at 32°C for 1 h. Microglia were imaged at a depth of 50-100 µm in the brain slice using a Leica TCS SP8 MP multiphoton microscope (25× lens). Stacks of 26 images with 2-µm intervals were acquired every 60 s for 5 min (base line) and for another 10 min after ATP (1 mM) was bathapplied to brain slices. Images were analyzed using ProMoIJ, an ImageJ macros that perform automatic microglial motility analysis 82 . Briefly, microglia were cropped and aligned, subsequently individual microglial processes were selected for automated analysis of their motility. Process motility was analyzed by reconstructing the process 3D skeleton and was calculated as the absolute difference of process length between two consecutive time frames. Specifically, where f is time frame and t corresponds to the time interval between consecutive frames (1 min).
Statistical analysis. Statistical analyses were performed using SPSS. Shapiro-Wilk (n < 10) and D'Agostino and Pearson omnibus (n > 10) normality tests were performed to determine if values fit a Gaussian distribution. Comparisons between two groups/treatments were performed using two-tailed Student's t-test or Chisquare test as applicable. Cumulative curves were compared using Kolmogorov-Smirnov test. For multiple comparisons, two-way ANOVA with Fisher's LSD post hoc test were used. The minimal level of significance was set at p < 0.05. Statistical parameters are detailed in the legend for each figure.