Transient CSF1R inhibition ameliorates behavioral deficits in Cntnap2 knockout and valproic acid-exposed mouse models of autism

Abstract

Microglial abnormality and heterogeneity are observed in autism spectrum disorder (ASD) patients and animal models of ASD. Microglial depletion by colony stimulating factor 1-receptor (CSF1R) inhibition has been proved to improve autism-like behaviors in maternal immune activation mouse offspring. However, it is unclear whether CSF1R inhibition has extensive effectiveness and pharmacological heterogeneity in treating autism models caused by genetic and environmental risk factors. Here, we report pharmacological functions and cellular mechanisms of PLX5622, a small-molecule CSF1R inhibitor, in treating Cntnap2 knockout and valproic acid (VPA)-exposed autism model mice. For the Cntnap2 knockout mice, PLX5622 can improve their social ability and reciprocal social behavior, slow down their hyperactivity in open field and repetitive grooming behavior, and enhance their nesting ability. For the VPA model mice, PLX5622 can enhance their social ability and social novelty, and alleviate their anxiety behavior, repetitive and stereotyped autism-like behaviors such as grooming and marble burying. At the cellular level, PLX5622 restores the morphology and/or number of microglia in the somatosensory cortex, striatum, and hippocampal CA1 regions of the two models. Specially, PLX5622 corrects neurophysiological abnormalities in the striatum of the Cntnap2 knockout mice, and in the somatosensory cortex, striatum, and hippocampal CA1 regions of the VPA model mice. Incidentally, microglial dynamic changes in the VPA model mice are also reported. Our study demonstrates that microglial depletion and repopulation by transient CSF1R inhibition is effective, and however, has differential pharmacological functions and cellular mechanisms in rescuing behavioral deficits in the two autism models.

Introduction

Autism spectrum disorder (ASD) is a range of heterogeneous neurodevelopmental disorders with core clinical symptoms manifested as impaired social interaction and communication, and restricted, repetitive and stereotypical patterns of behavior [1]. Hitherto, there are no drugs approved to treat the core symptoms of ASD [2]. The reason for this severe situation may lie in that the etiology of ASD presents high heterogeneity [3].

As etiological causes of ASD, both genetic and environmental risk factors may lead to immune dysregulation in autistic individuals and animal models [4]. Microglia, a key mediator of immune system and a sensor for pathological conditions in the CNS, can influence postnatal brain development, neuronal plasticity and circuit function [5,6,7]. Accumulating evidence suggests that microglial dysfunction has been linked to the occurrence of ASD. In people with ASD, augmented activation, abnormal morphology or increased density of microglia was observed in the cerebral cortex and subcortical areas, medial frontal gyrus, as well as the cerebellum [8,9,10,11,12]. Noteworthy, microglia in ~ 38% of postmortem cases with autism in one report by Morgan et al. [8] appeared markedly activated, and additional ~ 31% of cases presented marginally activated microglia. At the molecular level, transcriptional analysis and proteomic studies revealed that microglial activation-associated gene expression and complement molecules are up-regulated in people with autism [13, 14]. Compared to typically developing individuals, a significant increase in primed microglia was seen by Lee et al. [15] in grey matter of ASD patients, while average cell number and volume of microglia did not differ in another case observed by Morgan et al. [16]. These studies suggest that a spectrum of microglial pathology in different patients reflects heterogeneity of ASD [15].

Microglial abnormality and heterogeneity are also seen in animal models of ASD. Prenatal valproic acid (VPA) exposure changed microglial phenotypes from resting to activated states in the prefrontal cortex and hippocampus regions of infant and adolescent VPA model rats, but microglial activation was alleviated in adulthood [17]. Microglia in VPA-induced autism model marmosets had abnormal morphology, and their processes were fragmented [18]. A positron emission tomography (PET) imaging showed an increased microglial activation in the prefrontal cortex and hippocampus in mature offspring of prenatal polyinosinic:polycytidylic acid (poly I:C) exposed rats [19], which are considered as an autism model induced by maternal immune activation (MIA) [20]. As a genetic model of ASD, Fmr1 knockout (KO) mice displayed reduced microglia-mediated synaptic pruning in hippocampus [21]. BTBR mouse strain, an idiopathic ASD model, had a higher level of activated microglia [22]. Mouse offspring exposed to CASRP2-antibodies showed marked social interaction deficits and persistent microglial activation in the prefrontal and somatosensory cortices [23]. However, the number or morphology of microglia in Cntnap2 (the gene encoding CASPR2) KO mice, in comparison with that in the wild type (WT) counterparts, was reported to have no difference [24].

Due to the emerging roles of microglia in brain development, plasticity and cognition [6], targeting microglia may act as a promising strategy for the treatment of neurological and psychiatric disorders including ASD [25, 26]. Predominantly expressed on microglia, colony stimulating factor 1-receptor (CSF1R) is necessary for microglia viability [27], and its expression is significantly upregulated in neurodegenerative diseases [28]. Inhibition of CSF1R signaling may lead to microglial depletion and repopulation that can have beneficial effects on behavior in different animal models of diseases, such as traumatic brain injury, Alzheimer's disease, Huntington's disease [28,29,30,31,32,33,34,35,36,37]. Ikezu et al. found that transient depletion of microglia caused by the CSF1R inhibitor PLX5622 can improve social impairment and stereotyped repetitive behavior in autism model mice with poly I:C-simulated maternal immune activation [38]. Interestingly, Elmore et al., who first discovered that CSF1R inhibitors can significantly deplete microglia, did not observe the effects of CSF1R inhibitors on social and other behaviors in healthy adult mice [27, 29]. For wild-type C57BL/6 mice, global depletion of microglia reversibly altered spatial memory, but resulted in no change in sociability behavior [39]. It seems that microglia are dynamic modulators in regulation of behavior effects [39, 40] due to their dynamic property [41]. However, it remains unknown how dynamic change of microglia occurs and whether such a dynamic change reversibly affects behaviors in autism model animals.

In this study, taking into account of etiological heterogeneity and microglial heterogeneity in ASD patients and animal models, we explored if CSF1R inhibition by PLX5622 could incur different pharmacological effects in treating Cntnap2 KO and VPA-exposed autism model mice, which respectively represent genetic and environmental risk factors in etiology of autism [42, 43]. Meanwhile, we investigated dynamic shift of microglial population and morphology in VPA-exposed model mice treated with PLX5622. As a result, we found pharmacological effectiveness and heterogeneity of PLX5622 in treating the two models and mechanistic variance at the cellular level, and observed dynamic changes of microglia alongside behavioral improvement in the VPA model mice.

Methods

Generation of animals

Experimental mice were housed in individually ventilated cages (IVCs) with corncob bedding and maintained on a 12-h light–dark schedule with access to food and water ad libitum. Mice at age of 8 weeks were mated overnight, when the female mice were confirmed in estrus phase by microscopic examination of vaginal epithelial cells. The female mice were checked the next morning for the presence of a vaginal plug that marked the day as embryonic day 0.5 (E0.5). Pregnant female mice were transferred to a single cage and left undisturbed, except for necessary cage cleaning. The day of parturition was designated as postnatal day (P0). On postnatal day 21 (P21), offspring were housed separately according to sex, genotype and drug treatment, in groups of 3–5 mice per cage. Only male offspring were subjected to drug treatment and used for behavioral and cellular studies. Two male and two female Cntnap2 heterozygous KO (Cntnap2^±, Jackson Laboratory (#017482)) mice at the C57BL/6 J background were a gift from Professor Jiada Li’s lab in Central South University (Changsha, China). Cntnap2 homozygous KO (Cntnap2^−/−) and WT and mice were obtained from heterozygous Cntnap2^± mice crossing, and identified by genotyping on P21. As described in Li Jiada’s paper [44], genotyping was performed using the genomic DNA collected from mouse tail, which was amplified using PCR to detect the ~ 39-base pair deletion in Cntnap2^± and Cntnap2^−/− offspring (Fig. S1). For generation of VPA-induced autism model mice, male and female C57BL/6 J mice were obtained from the PKUHSC animal breeding facility. On E12.5, pregnant dams were randomly assigned to two groups, VPA or saline, and intraperitoneally injected with VPA (Sigma; dissolved in physiological saline) at a dose of 600 mg/kg or equivalent volume of physiological saline. After weaning, male litters were separately kept and investigated in this study.

PLX5622 preparation and administration

PLX5622, a CSF1R inhibitor, was synthesized through six steps from commercially available compounds 2-amino-6-fluoropyridine, 5-fluoro-2-methoxypyridine-3-carbaldehyde and 3-iodo-5-methyl-1-(triisopropylsilyl)-1H-pyrrolo [2,3-b] pyridine, following the reference [32]. The optimized PLX5622-fumaric acid salt (PLX5622-FA) was formed by heating to reflux of PLX5622 and fumaric acid in methyl ethyl ketone. The final product PLX5622-FA, 6-fluoro-N-((5-fluoro-2-methoxypyridin-3-yl) methyl)-5-((5-methyl-1H-pyrrolo[2,3-b] pyridin-3-yl) methyl) pyridin-2-amine (fumaric acid salt), was validated using nuclear magnetic resonance spectroscopy (NMR) for chemical structure and high-performance liquid chromatography (HPLC) for purity (Fig. S2–S4). The chow containing PLX5622 at a concentration of 1200 ppm and the normal chow (control) were prepared by Jiangsu Xietong Biological Co., Ltd (Jiangsu, China). Briefly, PLX5622 drug chow was made by homogeneously blending PLX5622-FA powder with AIN-93 diet according to the company’s proprietary procedures. For microglial depletion and repopulation, mice were fed with PLX5622-containing chow from P21 to P31, followed by normal chow for the remainder of the study.

Three-chamber social interaction test

The test apparatus was a rectangular, white acrylic box with three dimensions of 60 cm in length, 40 cm in width and 25 cm in height. The box was divided into three chambers with each one 20 cm long, and the dividing walls have a 10 cm opening in the center which allows test mice to freely access to three chambers. Placed in the center of each side chamber was a metal cylinder (10 cm diameter × 15.5 cm height) with its wall composed of 12 thin metal rods (~ 1 cm apart between two rods), which allows olfactory, visual, auditory contacts between internal and external mice. The apparatus together with the cylinders were washed and cleaned using 70% (v/v) ethanol after each trial. Before the test session, subject mouse was placed in the center chamber of the apparatus and allowed to habituate all chambers for 5 min. In the first phase of test session, an unfamiliar male C57BL/6 J mouse of same age (Stranger 1) was randomly placed in one cylinder, while a black Object with similar size was placed in the other cylinder. The test was started by placing subject mouse in the central chamber, and allowed to explore the chambers freely for 10 min. In the second phase, the Object was replaced with a novel strange male C57BL/6 J mouse (Stranger 2), and the subject mouse was allowed to explore the whole apparatus freely for 10 min. Time spent in sniffing the Object or strange mice was recorded by a computer-linked overhead video camera and analyzed using Smart v3.0 software.

Reciprocal social interaction test

Subjected mice were individually placed in a clean, white acrylic box (40 × 20 × 30 cm³) with an unfamiliar conspecific matched mouse according to genotype, age, sex, and treatment. The mice were allowed to interact with each other freely for 5 min. Time spent in social interactions, including close following, physical touching, nose-to-nose sniffing, nose-to-anus sniffing, and crawling over/under each other, were recorded and analyzed manually in a blinded manner.

Self-grooming test

Tested mice were individually placed in a mouse home cage without bedding for 20 min. Grooming duration of each mouse during the last 10 min was manually analyzed by an observer blind to the treatment.

Nest building

Nest building test was carried out in clean mouse home cages. The cages were individually sterilized and supplemented with fresh, unscented mouse bedding. Subjected mice were individually placed in a cage and allowed to acclimate for 6 h. A corresponding number of commercially available cotton fiber nestlets (5.0 × 5.0 × 0.5 cm³, ~ 2.5 g each, Ancare) were separately weighed on an analytical balance. Next, a nestlet was placed on top of the bedding in the test cage, and the mouse was left undisturbed for further 16 h. After test completion, test mice were moved to a clean cage and the final nest were documented without disturbing. Nest building was scored following a definitive 5-point nest-rating scale [45]. During the whole process of test, mice were allowed to freely access to chow and water.

Marble-burying test

Marble-burying test was performed in non-transparent acrylic cages (20 × 30 × 40 cm³) filled with free, unscented bedding material to a 5-cm depth. Prior to the test, 20 glass toy marbles (assorted colors, 14 mm in diameter) were gently placed on the surface of the bedding in five rows of four marbles, which were evenly set at 5-cm intervals. Subjected mice were individually placed into the corner of the test cage and remained undisturbed for 30 min. After test completion, mice were removed and returned to the home cage. The number of buried marbles was manually calculated in a blind manner. A marble was scored as buried if two-thirds of its surface area is covered by bedding.

Open field locomotion test

The open field test was carried out in a non-transparent square cage (50 × 50 × 50 cm³). Each subjected mouse was placed in the center of the open field arena and allowed to freely explore for 15 min. Mouse movements were recorded by a computer-linked overhead video camera. Data including total distance traveled and time spent in different area were analyzed using Smart v3.0 software.

Immunostaining for microglia

For histology and morphological analysis of microglia, mice were anesthetized and transcardially perfused with physiological saline followed by 4% (w/v) paraformaldehyde (PFA). Next, brains were removed, washed using PBS, and fixed in PFA solution for further 6 h. The fixed brains were cryopreserved in 30% (w/v) sucrose solution for 72 h, and embedded in optimal cutting temperature (OCT) compound (Sakura, Japan). Coronal Sects. (45 μm thick) were cut using a cryostat microtome (Leica CM3050S). Brain sections were allowed to rewarm at room temperature and rinsed three times using PBS. Next, sections were incubated in 0.5% (v/v) Triton X-100 in PBS for 30 min, and then blocked in 5% (w/v) BSA and 0.1% (v/v) Triton X-100 in PBS for 2 h. The blocked sections were then reacted with the rabbit anti-mouse IBA1 primary antibody (1:1000; Wako, 019–19741) or the rabbit anti-PDGFRα primary antibody (1:300; CST, 3174) at 4 °C overnight. Following five PBS-washes, sections were incubated with Alexa Fluor 488-conjugated goat anti-rabbit IgG secondary antibody (1:1000; Jackson, ZF0511) or Alexa Fluor 594-conjugated goat anti-rabbit IgG secondary antibody (1:1000; Jackson, ZF0516) for 2 h at room temperature. Sections were then washed for three times, counterstained with Hoechst 33,342 (Sigma-Aldrich), and mounted with glass coverslips.

Images of the whole brain sections were scanned using Vectra Polaris 1.0 (PerkinElmer), and images of somatosensory cortex, hippocampal CA1 and striatum regions were taken using a confocal microscope (Zeiss LSM 880) with a × 40 objective at multiple z-planes. Both maximum projection of the image stacks and a slice-by-slice view were used to ensure the continuity of microglial branches. Microglial morphology analysis and rendering were performed using Bitplane Imaris 9.9 software. For comparison between the same autism-model mice, all stained images were taken for comparable sections of issue from each mouse, collected with the same laser settings, and processed by automated analysis using the same spot, surface or filament modules from Bitplane Imaris 9.9.

Golgi staining for neurons

Golgi-Cox impregnation of neurons was carried out using the FD Rapid Golgi Stain Kit (FD Neuro Technologies, Columbia, USA) following the manufacturer’s instructions. Briefly, mice were euthanized, and brains were dissected immediately and rinsed using PBS. Next, the brains were immersed in a mixture of FD Solution A and Solution B for 14 d in darkness at room temperature, and then transferred to FD Solution C and kept for 72 h in the dark. Coronal Sects. (150 μm thick) were cut using a Leica cryostat microtome, mounted on gelatin-coated slides, and dried overnight. The sections were then immerged in FD Solution 2 (a mixture of Solution D and Solution E) for 10 min. After three water-washes, the slides were successively dehydrated in 50%, 70%, 95%, and 100% alcohol, and then cleared in Xylene three times. Neurons were imaged using a confocal microscope (Zeiss LSM 880) with a × 63 oil objective at multiple z-planes. Counting and morphometrical analysis of spines was performed using Bitplane Imaris 9.9 software.

Statistical analysis

Statistical analyses, including one-way ANOVA and two-way ANOVA with Sidak’s, Bonferroni’s or Tukey’s post hoc multiple comparisons test, as well as Student’s t test, were performed using the software Graphpad Prism 8.0. Data are presented as raw means ± standard error of the mean (SEM). For all analyses, statistical significance was accepted at p < 0.05, and statistical trend was considered at p < 0.1. The results of statistical analyses are summarized in Table S1.

Results

PLX5622 alleviates behavioral deficits in Cntnap2 KO mice

We first investigated pharmacological effects of PLX5622 on behaviors of Cntnap2 KO mice. As shown in Fig. 1A, the mice in this experiment were divided into four groups: WT, WT + PLX5622, KO and KO + PLX5622. The WT and KO groups belonged to the control chow treatment groups (“control”) that were fed with a control chow from P21-P80; the WT + PLX5622 and KO + PLX5622 were the “microglial depletion and microglial repopulation” groups, which were fed PLX5622-containing chow from P21-P31, followed by control chow for the remainder of the study. Behavioral tests were carried out from P70-P80.

Fig. 1

PLX5622 rescues social behavioral deficits in the Cntnap2 KO mice. A Experimental timeline. After weaning on P21, 3–5 Cntnap2 WT and KO pups were sacrificed for histological test, while others were supplied with PLX5622 (1200 ppm) or normal chow (control) for 10 days, following which mice were returned to normal chow for the remainder of the study. One cohort of mice were killed at P31 or P65, and brain tissues were collected for histological test. Another cohort of mice underwent a battery of behavioral tasks at P70-P80. B, C Three-chamber social interaction test for four groups of Cntnap2 WT and KO mice treated or not treated with PLX5622 (as labeled in the inset of (C). Sniffing time with Stranger 1 and Object in phase 1 (B); sniffing time with Stranger 1 and Stranger 2 in phase 2 (C). n = 8–13 per group; two-way ANOVA for (B) and (C). D Reciprocal social interaction test for all groups. n = 6–8 per group; one-way ANOVA. ***p < 0.001; **p < 0.01; “ns” denotes no significance. Data are expressed as mean ± SEM

As reported by Penagarikano [42], Cntnap2 KO mice show social behavior abnormalities. Firstly, we performed a three-chamber social interaction test to evaluate PLX5622 effect on sociability and social novelty. In the first phase of the test, WT and WT + PLX5622 mice had significantly longer duration of sniffing time spent with Stranger 1 mouse than that with an Object, indicating that they had normal sociability and PLX5622 didn’t alter the sociability of WT mice (Fig. 1B). Meanwhile, Cntnap2 KO mice failed to show good sociability, specifically, there was no significant difference in the sniffing time that KO mice spent with Stranger 1 or the Object. However, PLX5622 significantly increased sniffing time between the KO mice and Stranger 1, indicating that PLX5622 can significantly improve the sociability of Cntnap2 KO mice. In the second phase of the three-chamber test (Fig. 1C), WT mice exhibited a clear preference for interacting with a novel social partner (Stranger 2) rather than with a familiar one, whereas KO mice didn’t display a notable preference for a new partner, supporting that Cntnap2 KO mice can be used as a model for studying social novelty [42, 46]. However, compared with KO mice, KO + PLX5622 mice spent more time interacting with Stranger 2. Although this improvement did not reach a statistically significant difference, PLX5622 did show a tendency to improve social novelty behavior in Cntnap2 KO mice. Secondly, we carried out reciprocal social interaction tests, which have also been utilized for assessment of spontaneous social behaviors in various rodent species [47]. As shown in Fig. 1D, compared with WT mice, WT + PLX5622 mice did not have significantly less time in interacting with intruders, suggesting that PLX5622 can’t affect spontaneous social behaviors of WT mice. Clearly, KO mice had significantly less social interaction time than WT mice. But no significant difference was observed in reciprocal interaction time of WT and KO + PLX5622 groups (Fig. 1D). These results demonstrated that PLX5622 can effectively improve reciprocal social behavior of KO mice.

We next performed a battery of non-social behavioral tests. First, in the open field test, we found that the four groups of mice spent well-matched total time in the periphery or center of the arena (Fig. 2A). In concordance with a previous report [42], Cntnap2 KO mice displayed significantly increased locomotor activity than their WT counterparts (Fig. 2B). PLX5622 treatment effectively reduced the locomotor velocity and alleviated the hyperactive behavior of Cntnap2 KO mice, but had no negative effect on the locomotor activity of WT mice. Second, we explored grooming behaviors of the test mice, which can provide an index of repetitive/stereotypic behavior in rodents [47]. Similar to a previous report [42], KO mice spent more time in self-grooming than WT mice (Fig. 2C). WT + PLX5622 and WT mice spent comparable time in self-grooming, while KO + PLX5622 had significantly less time in this repetitive behavior than KO did. These observations demonstrate that PLX5622 can effectively alleviate repetitive behavior of KO mice, but almost had no effect on WT mice. Third, we performed nest building test, in which the test mice first shredded a tightly packed square nestlet and then arranged it into a nest [45, 48]. As shown in Fig. 2D, E, compared with KO mice, the nesting behavior of Cntnap2 KO mice was significantly impaired, and the nesting score was less than half of WT mice. PLX5622 treatment significantly improved the scores earned by KO mice and did not impair scoring of WT mice (Fig. 2E).

Fig. 2

PLX5622 alleviates hyperactivity and abnormal grooming, and improves nesting behavior of the Cntnap2 KO mice. A, B Open field test. Time spent in center and periphery zone (A) and velocity (B) for four groups of Cntnap2 WT and KO mice treated or not treated by PLX5622. Two-way ANOVA for (A) and one-way ANOVA for (B); n = 8–10 per group. C, Time spent in self-grooming by four groups of mice as labeled in (A). One-way ANOVA; n = 7–11 mice per group. D, E Nesting behavior test. Nestlets after test were documented for all tested animals (D). Scoring was performed using grades ranging from 1 (very poor) to 5 (optimal): grade 1 corresponds to an untouched nestlet, while grade 5 means nestlet fully shredded to form a crater-like, delimited nest (E). One-way ANOVA; n = 8 mice per group. ***p < 0.001; **p < 0.01; “ns” denotes no significance. Data are expressed as mean ± SEM

In summary, PLX5622 not only alleviates autism-like core symptoms (social deficits and repetitive behaviors) in Cntnap2 KO mice, but also mitigates their hyperactivity and improves their nest-building ability. In line with previous findings that global depletion of microglia resulted in no change in sociability behavior of wild-type C57BL/6 mice [27, 29, 39], PLX5622 hardly altered behaviors of Cntnap2 WT mice (Figs. 1, 2). Noteworthy, PLX5622 can successfully correct social and repetitive behavior deficits of MIA model mice without affecting behaviors, microglia and neuronal spine morphology of the saline group [38]. Therefore, in the following experiments, we didn’t set up the group of wild type or control mice treated with PLX5622 so that as less as mice would be sacrificed.

PLX5622 restores microglial number and morphology in Cntnap2 KO mice to the level of wild type

To obtain insights into the cellular mechanisms underlying PLX5622-alliviating autism-like behaviors in Cntnap2 KO mice, we compared microglial number and morphology in different brain regions of Cntnap2 WT, KO and KO + PLX5622 groups of mice on P65 using immunofluorescence for IBA1. Compared with WT mice, the number of microglia in somatosensory cortex of Cntnap2 KO mice significantly increased; correspondingly, microglia in KO mice presented amoeboid-like phenotype with decreased process area, decreased length and number of process branches, and increased cell body (Fig. 3 A-F). PLX5622 administration significantly decreased microglial number, increased process area, length and number, and had tendency in decreasing cell body size in somatosensory cortex of Cntnap2 KO mice, to a level of WT mice (Fig. 3A–F). Different from the case of somatosensory cortex, Cntnap2 KO mice showed no abnormalities in microglial number, cell body volume and branch number in the CA1 region compared with WT mice (Fig. 3G–L). Noteworthy, microglial process area and length in the CA1 of KO mice were significantly smaller than that in WT mice, and this difference was obviated by PLX5622 (Fig. 3G–L). Like the case of the somatosensory cortex region, compared with WT mice, microglial number and cell body volume in the striatum of Cntnap2 KO mice significantly increased, whereas microglial process area, length and branch number significantly decreased (Fig. 3M–R). PLX5622 effectively reversed such changes of microglia in the striatum of Cntnap2 KO mice to the WT level (Fig. 3M–R). In conclusion, PLX5622 administration can reverse the changes in the number and morphology of microglia in somatosensory cortex, hippocampal CA1 and striatum of Cntnap2 KO mice.

Fig. 3

PLX5622 restores microglial number and morphology in the somatosensory cortex (A–F), hippocampal CA1 (G–L) and striatum (M–R) of the Cntnap2 KO mice. A, G, M, Representative immunofluorescence images of IBA1⁺ microglia (top) and representative microglia images reconstructed by Imaris 9.9 (bottom) in the somatosensory cortex (A), hippocampal CA1 (G) and striatum (M) of three groups of Cntnap2 WT, KO or PLX5622-treated KO mice. Scale bar: 50 μm; n = 3–5 mice for each group. B-F, H–L, N-R, Statistical analysis of the number and morphology of microglia in the somatosensory cortex (B–F), hippocampal CA1 (H–L) and striatum (N–R) in the three groups. Comparison of microglia numbers in 1 mm² (B, H, N), microglial process area (C, I, O), microglial process length (D, J, P), number of branches per microglia (E, K, Q), and microglial cell body area (F, L, R) in each group. n = 6–7 cells per group; one-way ANOVA. ***p < 0.001; **p < 0.01; *p < 0.05; “ns” denotes no significance. Data are expressed as mean ± SEM

PLX5622 corrects dendritic spine abnormalities in the striatum region of Cntnap2 KO mice

To investigate whether microglial variation occurs alongside neurophysiological alteration, we detected density and morphology of dendritic spines in the above mentioned three brain regions of Cntnap2 WT, KO and KO + PLX5622 groups of mice on P65 using Golgi staining. As shown in Fig. 4A–H, there was no significant difference in the density and morphology of dendritic spines in the somatosensory cortex and hippocampal CA1 between Cntnap2 KO and WT mice. PLX5622 did not affect dendritic spines in the somatosensory cortex and hippocampal CA1 of Cntnap2 KO mice (Fig. 4A–H). In contrast, the density as well as the proportions of mushroom-shaped and stubby dendritic spines in the striatum of Cntnap2 KO mice significantly decreased in comparison with WT mice (Fig. 4I–L). Such neurophysiological abnormalities in the striatum of KO mice were rescued by PLX5622 to the level of WT (Fig. 4I–L).

Fig. 4

Description and statistical analysis of dendritic spine density and morphology of neurons in the somatosensory cortex (A-D), hippocampal CA1 (E–H), and striatum (I-L) of three groups of mice: Cntnap2 WT, KO mice and PLX5622-treated KO mice. A, E, I, Representative images of dendritic spines of the somatosensory cortex (A), hippocampal CA1 (E), and striatum neurons (I). Scale bar: 10 μm; n = 3–5 mice per group. B, F, J, Representative 3D image of dendritic spines reconstructed by Imaris 9.9 for (A), (E) and (I), respectively. C, G, K, Statistical analysis of the total number of dendritic spines contained in 10 μm dendrites. n = 8–11 cells per group; one-way ANOVA. D, H, L, Statistical analysis of the number of mushroom and stubby dendritic spines contained in 10 μm dendrites. n = 8–11 cells per group; two-way ANOVA. ***p < 0.001, **p < 0.01, *p < 0.05, “ns” denotes no significance. Data are expressed as mean ± SEM

PLX5622 ameliorates behavioral deficits in VPA-model mice

The mice in this experiment were divided into three groups: saline, VPA, VPA + PLX5622 (Fig. 5A). The VPA + PLX5622 mice were provided with PLX5622 (1200 ppm) diet on P21 for 10 days, and then returned to normal chow. At the indicated time points, different cohorts of mice were sacrificed for histological studies or kept alive for a battery of behavioral tasks (Fig. 5A). Here are the results of behavioral test. Briefly, PLX5622 can effectively rescue social deficits and stereotypical repetitive of the VPA model mice.

Fig. 5

PLX5622 corrects social behavioral deficits in VPA model mice. A Experimental timeline. C57BL/6 J female mice received VPA or saline injection on E12.5, and the prenatally saline-treated male offspring were assigned to the “Saline” group. The VPA-treated male offspring were randomly divided into two groups: one group provided with normal diet (“VPA” group); the other group reared with PLX5622 diet at P21 for 10 days and then with normal chow (“VPA + PLX5622” group). One cohort of mice were sacrificed respectively at P21, P31, P38, P45 and P52 for histological analysis, while a separate cohort of mice received a battery of behavioral tests at P70-P80. The “VPA + PLX5622” group of mice that underwent 0, 7, 14 and 21 days of microglia repopulation after withdraw of PLX5622 at P31 were hereinafter categorized as “VPA Rep.0”, “VPA Rep.7”, “VPA Rep.14” and “VPA Rep. 21” group, respectively. B-C, three-chamber social interaction test for the Saline, VPA and VPA + PLX5622 groups of mice. Sniffing time with Stranger 1 and Object in phase 1 (B); sniffing time with Stranger 1 and Stranger 2 in phase 2 (C). n = 8 per group; two-way ANOVA. ***p < 0.001; **p < 0.01; *p < 0.05; ^#p < 0.1, “ns” denotes no statistical significance. Data are expressed as mean ± SEM

In the first phase of three-chamber social interaction test, the saline group of mice had significantly longer time in Stranger 1 over Object, whereas the VPA model mice spent comparable time in sniffing with Stranger 1 and Object (Fig. 5B). Meanwhile, compared with the saline group, VPA model mice spent less time in sniffing with Stranger 1 (Fig. 5B). These results demonstrate that the VPA model mice have impaired sociability. Similar to the saline group, the PLX5622-treated VPA model mice had significantly longer time in Stranger 1 over Object (Fig. 5B), indicating that PLX5622 can ameliorate their deficits in sociability. In the second phase of the test (Fig. 5C), the saline group of mice spent significantly more time in the new comer (Stranger 2), and the VPA model mice did not show statistical difference in the sniffing time spent with Stranger 1 or Stranger 2, indicating that VPA model mice have damaged social novelty. PLX5622 can effectively improve the social novelty of VPA model mice, as showed by a trend of increased sniffing time with the new comer (Fig. 5C).

In open field test (Fig. 6A), compared with the saline group, the VPA model mice spent significantly less time in the center arena and more time in the periphery area, indicating that the VPA model mice have more intensive anxiety. PLX5622 hardly alleviated anxiety of the VPA model mice, since it didn’t alter their time in the center arena (Fig. 6A). In addition, there was no significant difference in the total movement distance among the three groups, indicating that VPA or PLX5622 treatment did not affect locomotor activity of mice (Fig. 6B).

Fig. 6

PLX5622 alleviates anxiety and repetitive behavioral deficits in VPA-model mice. A, B Open field test. Time spent in center and periphery zone (A) and total distance (B) during test for three groups of mice: saline, VPA and VPA + PLX5622. Two-way ANOVA for (A) and one-way ANOVA for (B); n = 8–12 per group. C–E, Repetitive behavior was evaluated by the time spent in self-grooming (C), and number of buried marbles in marble burying test (D, E). One-way ANOVA for (C) and E n = 8–11 per group. ***p < 0.001; **p < 0.01; *p < 0.05; “ns” denotes no statistical significance. Data are expressed as mean ± SEM

In self-grooming test (Fig. 6C), the VPA model mice showed more frequent and continuous grooming behavior, and their total self-grooming time was nearly twice that of the saline group. PLX5622 treatment significantly reduced the self-grooming time of the VPA model mice to the level of the saline group (Fig. 6C). In addition, marble burying test was carried out to evaluate stereotypical repetitive behavior in all groups. Compared with the saline group, the VPA model mice buried significantly more marbles in the test (6.8 ± 3.0 versus 2.7 ± 1.8) (Fig. 6D, E). However, there was no statistically significant difference in the number of marbles buried by the VPA + PLX5622 and saline groups, indicating that PLX5622 significantly alleviated stereotypical marble-burying behavior of the VPA model to the level of the saline group.

PLX5622 induces microglia depletion and repopulation in the VPA model mice

To investigate how PLX5622 affects microglial dynamic change in depletion and repopulation, we detected microglial number and morphology in the above mentioned three brain regions of the ‘Saline’, ‘VPA’ and ‘VPA + PLX5622’ groups at the indicated time points via using immunofluorescence for IBA1 (Fig. 5A). In line with a previous report that prenatal VPA exposure changed microglial phenotypes from resting to activated states in the prefrontal cortex and hippocampus regions of infant VPA model rats [17], we observed that microglia underwent a similar phenotypic change in the somatosensory cortex, hippocampal CA1 and striatum regions of the VPA model mice (Figs. 7, 8, 9). Concretely speaking, in comparison with the Saline P21 group of mice, the VPA model mice on P21 presented decreased process area, length and branch number and increased cell body volume, indicating microglia morphology abnormalities in the three brain regions (Figs. 7, 8, 9). Noteworthy, microglial number increased in the hippocampal CA1 region of the VPA model mice, further indicating microglial abnormality in this brain region (Fig. 8).

Fig. 7

Microglia depletion and repopulation in the somatosensory cortex of PLX5622-treated VPA-model mice. A Representative immunofluorescence images of IBA1⁺ microglia (top) and representative microglia image reconstructed by Imaris 9.9 (bottom) in the somatosensory cortex in the seven groups of mice as indicated. n = 3 mice for each group. B–F, Statistical analysis of the number and morphology of microglia in the somatosensory cortex in each group: number of microglia (B), microglial process area (C), microglial process length (D), number of branches per microglia (E), and microglial cell body area (F). Scale bar: 50 μm; n = 6 cells per group; one-way ANOVA. ***p < 0.001; **p < 0.01; *p < 0.05; “ns” denotes no significance. Data are expressed as mean ± SEM

Fig. 8

Microglia depletion and repopulation in the hippocampal CA1 of PLX5622-treated VPA-model mice. A Representative immunofluorescence images of IBA1⁺ microglia (top) and representative microglia image reconstructed by Imaris 9.9 (bottom) in the hippocampal CA1 in the seven groups of mice as indicated. n = 3 mice for each group. B–F, Statistical analysis of the number and morphology of microglia in the hippocampal CA1 in each group: number of microglia (B), microglia process area (C), microglia process length (D), number of branches per microglia (E), and microglia cell body area (F). Scale bar: 50 μm; n = 6 cells per group; one-way ANOVA. ***p < 0.001; **p < 0.01; *p < 0.05; “ns” denotes no significance. Data are expressed as mean ± SEM

Fig. 9

Microglia depletion and repopulation in the striatum of PLX5622-treated VPA-model mice. A Representative immunofluorescence images of IBA1⁺ microglia (top) and representative microglia image reconstructed by Imaris 9.9 (bottom) in the striatum in the six groups of mice as indicated. n = 3 mice for each group. B–F, Statistical analysis of the number and morphology of microglia in the striatum in each group: number of microglia (B), microglia process area (C), microglia process length (D), number of branches per microglia (E), and microglia cell body area (F). Scale bar: 50 μm; n = 6 cells per group; one-way ANOVA. ***p < 0.001; **p < 0.01; *p < 0.05; “ns” denotes no significance. Data are expressed as mean ± SEM

In general, PLX5622 led to a nearly identical process of microglia depletion in the somatosensory cortex, hippocampal CA1 and striatum regions of the VPA model mice. As shown in Figs. 7, 8, 9, 10 days of PLX5622 treatment eliminated ~ 90% of microglia in the three brain regions of the VPA model mice (VPA Rep.0), and left a small population of microglia exhibiting an enlarged cell body and retracted processes similar with that before PLX5622 treatment (VPA P21). Meanwhile, similar microglial repopulation processes occurred in the somatosensory cortex, hippocampal CA1 and striatum regions of the VPA model mice (Figs. 7, 8, 9). For clarity, the focus here is on the somatosensory cortex region. After a 7 days period of recovery, the total number of microglia in this region of the VPA Rep.7 group increased to a level slightly higher than that before depletion (VPA P21), and microglia recovered to a resting state with significantly larger cell process area and branch number and smaller cell body volume in comparison with the VPA P21 group (Fig. 7). Noteworthy, microglial process length of the VPA Rep.7 mice had no significant difference with that of the VPA Rep.0 group. By the end of 14 days of recovery, the microglial number of the VPA Rep.14 group was close to the level of the VPA Rep.7 group, and the repopulated microglia were ramified (Fig. 7). After 21 days of recovery, microglial number, process parameters (area, length, and branch number) and cell body volume in the VPA Rep.21 group had significant difference with that in the VPA Rep.0 group, and however, didn’t show any significant difference from those in the Saline group on P52 (Fig. 7). In the hippocampal CA1 and striatum regions of the VPA model mice (Figs. 8, 9), microglial number also swiftly restored to and maintained at a high level after 7 days of repopulation, and microglial morphology also underwent a gradual recovery with microglial process area, length and branch number increasing to a high level and cell body volume decreasing to a low level to the saline group on the 21th day after repopulation. Here we admit that we didn’t have mouse in the VPA Rep.14 group for examining microglia in the striatum (Fig. 9). Nevertheless, it is conclusive that microglial repopulation in the striatum share a common dynamic process with those in the somatosensory cortex and hippocampal CA1 regions.

PLX5622 restores the number and morphology of dendrite spine in the VPA model mice

Compared with the saline group, the total number of dendritic spines in the somatosensory cortical region was significantly reduced in the VPA model mice on P21, and the number of mushroom-like dendritic spines was only about a quarter of that in the saline group (Fig. 10A–D). 21 days after removal of the PLX5622 diet (VPA rep.21), the total number of dendritic spines and mushroom-like dendritic spines restored to the level of the saline group. Noteworthy, the density of stubby dendritic spines had no significant difference between the saline and VPA groups. The density of dendritic spines in the hippocampal CA1 was significantly decreased in the VPA model mice on P21, and the number of both mushroom-like and stubby dendritic spines were significantly lower than those in the saline group. PLX5622 treatment effectively restored the number and morphology of dendritic spines in hippocampal CA1 of the VPA model mice (Fig. 10E–H). In the striatum, the density of total dendritic spines or stubby spines in the VPA-model group was also significantly lower than that in the saline group, and however, was higher than that in saline group upon PLX5622 correction (Fig. 10I–L). Expectedly, mushroom spine density in the VPA mice’s striatum was significantly lower than that in the saline group, and rescued to the saline group level by PLX5622 treatment (Fig. 10I–L).

Fig. 10

PLX5622 treatment corrects dendritic spine density and morphology in the somatosensory cortex (A–D), hippocampal CA1 (E–H), and striatum (I–L) of the VPA-model mice. A, E, I, Representative images of dendritic spines in the somatosensory cortex (A), hippocampal CA1 (E), and striatum neurons (I) of four groups of mice: “Saline P21” (the Saline group of mice at P21), “VPA P21” (the VPA group of mice at P21), “Saline P52” (the Saline group of mice at P52) and “VPA rep. 21” (the PLX5622-treated VPA group of mice at P52). Scale bar: 7 μm (A) or 10 μm (E and I); n = 3 for each group. B, F, J, Representative 3D image of dendritic spines reconstructed by Imaris 9.9 for (A), (E) and (I), respectively. C, G, K Total number of dendritic spines contained in 10 μm dendrites. n = 6–9 per group; one-way ANOVA. D, H, L Statistical analysis of the number of mushroom and stubby dendritic spines contained in 10 μm dendrites. n = 6–9 per group; two-way ANOVA. ***p < 0.001, **p < 0.01, *p < 0.05. Data are presented as mean ± SEM

Discussion

Microglial depletion and repopulation through genetic targeting or pharmacological therapies (mainly with CSF1R inhibitors) exhibits great potential as an intervention in many neurological disease states [28, 34, 35, 49]. At present, there is only one study using CSF1R inhibitor-induced microglia depletion to intervene in behavioral abnormalities of MIA-invoked ASD model mice [38]. Considering high heterogeneity of ASD, we studied Cntnap2 KO and prenatal VPA-exposure autism model mice to investigate how CSF1R inhibition may affect autism-like behavioral phenotypes induced by genetic or environmental risk factors, and how microglial depletion and repopulation occurs for behavioral modification.

Retrospective view of biological rationales or proofs for our experimental design

As shown in Figs. 1 and 5, only male offspring of ASD model mice were used in this study, and PLX5622, an CSF1R inhibitor, was used to eliminate microglia on P21 only for 10 days. The biological rationales or proofs underlying such an experimental design are elucidated as follows.

Firstly, why were only male offspring used? Despite striking heterogeneity in manifestation, ASD has a replicated feature that is the male preponderance with the male to female ratio around 4.5:1 [50]. Microglia demonstrate sexual dimorphism in their colonization, structure, function, and transcriptomic and proteomic profiles from early postnatal development onwards of rodents [51,52,53], and can sculpt sex differences in adult social behaviors likely owing to their prominent roles in sculpting neural circuit development [54, 55]. Lipopolysaccharide (LPS)-induced maternal immune activation has been reported to cause male-specific social deficits in Cntnap2 KO mouse model for ASD [56]. Recent study has shown that sexual dimorphism in the social behavior of Cntnap2 KO mice is associated with increased microglial activity and disrupted synaptic connectivity in the anterior cingulated cortex [57]. Meanwhile, prenatal exposure to VPA led to male-specific alteration in synaptic development, morphological development and social interaction, while female offspring showed only marginal deficits [58,59,60]. Therefore, we used only male offspring of the two autism model mice in this study.

Secondly, why were microglia depleted in the time window from P21 to P31? In the rodent brain, microglia numbers peak at postnatal day 14 (P14), undergo a rapid decrease in the third postnatal week, and attain adult levels by P28 [61]. Furthermore, microglia exhibit an activated morphology and present high phagocytic activities during the postnatal three weeks in male rodent offspring [51]. It seems that postnatal three weeks is the critical time window for microglia to exert their functions in male rodents. Given that mice can take diet completely by themselves after weaning, we used CSF1R inhibitor-containing chow to rear male mice at P21 for lunching microglia depletion. The duration of microglia depletion in this study is 10 days from P21 to P31, half the duration of CSF1R inhibition in the MIA model that male and female mice were subjected to CSF1R inhibitor PLX5622 treatment from P21 to P42 [38]. Indeed, in comparison with male counterparts, female rodent offspring have more activated microglia with amoeboid morphology later in development (P30-P60) [51]. This observation likely underlies why longer duration of CSF1R inhibition till juvenile and adolescent stage was needed to deplete and repopulate microglia in female MIA model mice [38]. From this perspective, we think that microglia depletion for male Cntnap2 KO or VPA model mouse doesn’t need longer duration as in the MIA model. As shown in Fig. S5 and Fig. S6A, around 95% of microglia in male Cntnap2 KO mice were depleted after 10 days of PLX5622 treatment from P21. At P21, male Cntnap2 KO mice had significantly increased microglia number and significantly reduced microglia process area, process length and branch number in their somatosensory cortex area compared with male WT (Fig. S6). Unexpectedly, male Cntnap2 KO mice had significantly reduced microglia number in their striatum region (Fig. S6D). Nevertheless, microglia in the striatum area of male Cntnap2 KO exhibited shorter process and less branches (Fig. S6M–O). These data show that microglia in the male Cntnap2 KO during the postnatal three weeks are more amoeboid-like with shorter pseudopodia, which endows microglia higher phagocytic activity [62]. As shown in Figs. 7, 8, 9, male VPA mice at P21 also had more amoeboid-like microglia compared with the saline group. These data also indicate that it is reasonable for us to launch microglia depletion at P21 for male Cntnap2 KO and VPA mice in this study. However, a possibility is that earlier microglia depletion may also take effect, since a recent study shows that microglia in the anterior cingulate cortex of male Cntnap2 KO at P8 and P14 present more activated state than WT mice [57]. A caveat is that microglia depletion at earlier stage and shorter duration may have limited effectiveness for ASD in clinical treatment.

Thirdly, why was PLX5622 selected as a pharmacological tool? Did PLX5622 affect other types of CNS cells like oligodendrocyte progenitor cells (OPCs)? The development, survival and proliferation of microglia depends on signaling of CSF1R [6, 27], a member of the class III receptor tyrosine kinases (RTKs) that also include KIT, PDGFRα, PDGFβ and Flt3 [63]. Among these members, PDGFRα is highly expressed by OPCs and PDGFRα kinase activity is essential for their survival [64]. Different from their diverse extracellular domains, the intracellular kinase domains of the class III RTKs have a highly conserved ATP-binding pocket that is targeted by PLX5622 and PLX3397, two commonly used CSF1R inhibitors for microglia depletion [29]. Nevertheless, resorting to a structure-guided drug design strategy, PLX5622 was created based on PLX3397, and has higher specificity for CSF1R than PLX3397 [32, 65, 66]. As previously reported, 7 days of PLX5622 treatment hardly affected OPC numbers in adult mice and 21 days of treatment only led to a mild OPC number reduction; in contrast, 7 days of PLX3397 treatment caused significant OPC loss [67]. As shown in Fig. S7, 10 days of PLX5622 treatment didn’t result in significant variation of PDGFRα⁺ OPC numbers in the somatosensory cortex, hippocampal CA1 and striatum areas of male Cntnap2 KO. Therefore, we used PLX5622 in this study and persisted that PLX5622 acts through inhibiting CSF1R signaling in microglia. However, we couldn’t exclude a possibility that PLX5622 has side effects inside and outside the CNS, since PLX5622 was observed to affect “the turnover and function of bone marrow-derived, circulating, and tissue-resident macrophages” [68]. Whether such an effect on macrophages contributes to behavioral phenotype of the two autism model mice needs further investigation.

Fourthly, why were the two models of mice not subjected to the same battery of behavioral tests? That is, as reported in Figs. 1 and 5, marble burying test was not displayed in Cntnap2 KO mice, while nest building test was not in VPA model mice. In line with the previous report by Penagarikano et al. [42], we did observe that Cntnap2 KO mice were more focused on nestlet shredding and earned less scores in nest building test than WT (Fig. 2E). However, Cntnap2 KO mice, whether or not treated by PLX5622, hardly buried marbles, whereas WT did (Fig. S8). Coincidentally, to the best of our knowledge, marble-burying test in Cntnap2 KO mice has never been reported. In contrast, we observed that VPA model mice buried more marbles than the saline group (Fig. 6E), which is agreement with a dozen of previous reports [69,70,71,72,73,74,75,76,77,78,79,80]. Likely due to different genetic background in mice, different texture in nestlet and different testing environment employed, we didn’t observe that VPA model mice shredded nestlet (data not shown) as previously reported by the same lab [73, 79, 81]. Both hyper- and hypo-responsiveness to tactile stimulation are recognized as diagnostic criteria for ASD [82], and mice have an innate preference for novel stimuli and can discriminate between the textures of the objects [83]. It is possible that Cntnap2 KO and VPA model mice have different somatosensory preference for tactile stimuli invoked by marbles and nestlets. Therefore, nest-building and grooming tests were used to characterize the effects of PLX5622 on repetitive and stereotyped behaviors of Cntnap2 KO mice (Fig. 2), and marble-burying and grooming tests were used for VPA model mice (Fig. 6).

Pharmacological effects of PLX5622 on Cntnap2 KO and VPA model mice

For the Cntnap2 KO mice, PLX5622 can improve their sociability and reciprocal social behavior, slow down their hyperactivity and repetitive grooming behavior, and enhance their nesting ability. To investigate the role of microglia in Cntnap2 KO behavior alteration, we checked microglial number and morphology. In this study, knockout of Cntnap2 led to increased number and abnormal morphology of microglia in the cortical, hippocampal and striatum regions of male C57BL/6 J mice on P65. Different from our observation, Cope et al. found that the number or morphology of microglia in the hippocampus, medial prefrontal cortex and striatum in 5- to 6-month-old Cntnap2 KO mice do not alter when compared with that in WT mice [24]. It seems that microglial abnormalities in Cntnap2 KO mice can recover over time. Whether behavior deficits of Cntnap2 KO mice may be alleviated at old ages needs further investigation. Nevertheless, when compared with WT, male Cntnap2 KO mice have more activated microglia at P8 and P14 in the anterior cingulate cortex as observed by Dawson et al. [57] and at P21 in the somatosensory cortex and striatum areas as found in this study (Fig. S6). Therefore, we propose that microglial abnormalities at early in postnatal development may lead to autism-like behavioral deficits in male Cntnap2 KO mice. Social cognition deficits in animal models of ASD have been linked to the cortex [84]. From this point of view, increased number and abnormal morphology of microglia in the cortical region of Cntnap2 KO mice is likely a driver event causing weaken sociability, since PLX5622 can reset microglia in this region to the WT level and improve sociability deficits of Cntnap2 KO mice. Meanwhile, it has been reported that striatal dysfunction is a neurological substrate for repetitive behaviors [2, 85]. From this perspective, microglial abnormality in the striatum of Cntnap2 KO mice may lead to striatal dysfunction and therefore constitute pathological basis for repetitive behaviors, since repetitive grooming and nestlet-tearing behaviors of Cntnap2 KO mice can be corrected by PLX5622-mediated microglial repopulation (Figs. 2, 3). As we know, microglia have an established role in modifying synapse and even can mediate changes of dendritic spine morphology [86, 87]. Dendritic spine morphology change was not observed in the cortical and hippocampal regions, but statistically significant in the striatum of the Cntnap2 KO mice when compared with the WT counterparts (Fig. 4). It is reasonable that microglial abnormalities in the cortical and hippocampal regions might modify synapses, rather than dendritic spine morphology, to impair social behaviors of the Cntnap2 KO mice. In comparison, microglial abnormalities in the striatum of the Cntnap2 KO mice coalesce with dendritic spine subtype variation, and PLX5622 treatment resets microglia to a ramified state alongside re-distribution of dendritic spine subtype and improvement of repetitive behaviors (Figs. 2, 3, 4). CASPR2 is present in dendritic spines [88] and enriched in the synaptic plasma membrane to regulate the excitation/inhibition (E/I) balance [89]. Indeed, it has been reported that mouse offspring exposed to CASRP2-antibodies exhibit social interaction deficits, impaired nesting ability, persistent microglial activation and decreased glutamatergic synapses in the prefrontal and somatosensory cortices [23]. It is possible that microglia might affect dendritic spine morphology and mediate the E/I balance of local circuitry for behavioral changes in Cntnap2 KO using a way similar to that of prenatally CASPR2 antibodies-exposed mice.

For the VPA model mice, PLX5622 can significantly improve their social ability and social novelty, anxiety behavior, repetitive grooming and stereotyped marble-burying behaviors. As the most significant difference from the case of Cntnap2 KO mice, PLX5622 effectively improve social novelty of the VPA model mice in the second phase of three chamber social interaction test, with distinctly increased sniffing time with Stranger 2. Unlike the case of the Cntnap2 KO mice, microglial number and morphology, as well as dendritic spine density and subtype, significantly changed in the hippocampal CA1 region of the VPA model mice compared with the saline group. Conceivably, PLX5622-controlled microglia replacement and dendrite spine morphology recovery in hippocampal CA1 likely have influence on the CA2 region, a brain region essential for social memory [90], so that social novelty of the VPA model mice can be improved. In addition, we also observed dynamic changes of number or morphology of microglia in the VPA model mice, and suggested a dynamic process of microglia in autism animal model is similar to that of other neurological and psychiatric disorder models [27, 91]. Noteworthy, to the best of our knowledge, this study provides the first evidence of microglial dynamics in an autism model animal. As an environmental risk factor, prenatal VPA exposure leads to epigenetic processes that regulates gene expressions [43], and therefore likely influences neuronal development, dendritic spine morphology and microglia-neuron interaction in multiple brain regions. PLX5622 alleviates behavioral deficits in the VPA model mice likely through rescuing microglia-neuron crosstalk insulted by VPA epigenetic modification.

Potential mechanisms underlying microglia-mediated neuronal circuits and behavioral modification in Cntnap2 KO and VPA model mice

Overall, short-term suppression of CSF1R signaling by PLX5622 led to microglial depletion and repopulation, and improved social deficits, repetitive and other autism-like behaviors in the Cntnap2 KO and prenatal VPA-exposure autism model mice. As depicted above, the changes of microglia, neuronal dendritic spines in various brain regions and behavioral phenotypes are not exclusively identical in the two models. This reflects shared but differential potential mechanisms underlying how microglia may contribute to neuronal circuits and behavioral phenotypes in Cntnap2 KO and VPA model mice.

As mentioned above, microglia in the rodent brain under normal physiological conditions present highly activated state during the postnatal three weeks, and microglia number reach the maximal level at P21 [51]. Meanwhile, glial cell-dependent synaptic pruning contributes to brain circuit refinement in two phases: the primary phase corresponding to the postnatal three weeks in mice, and the second phase that is referred to adolescence in mammals, for example, the postnatal 3–8 weeks in mice [92,93,94]. The primary phase of synaptic pruning ensures the proper formation of sensory and executive circuits responsible for processing tactile and other sensory information or for regulating memory, while the second phase of synaptic pruning pronouncedly occurs in brain regions that require remodeling for high functioning and goal planning [92, 95, 96]. It seems that P21 is a critical timeline for synaptic pruning and circuit formation in mouse. Therefore, we checked microglia number and morphology in Cntnap2 KO and VPA model mice at P21 and selected this time point for microglia depletion. As shown in Fig. S6, Figs. 7 and 9, microglia in the somatosensory cortex and striatum areas of both Cntnap2 KO and VPA model mice at P21 displayed more activated states than their normal counterparts. Correspondingly, lower dendritic spine density was observed in the somatosensory cortex and striatum of VPA model mice compared with the saline group (Fig. 10). Microglia activation and lower neuronal spine density in VPA model mice at P21 suggest that over synaptic pruning by microglia in the primary phase may result in improper circuits for tactile information processing. This may likely explain why both Cntnap2 KO and VPA model mice exhibited impaired behaviors in nestlet shredding or marble burying (Figs. 2D and 6D). After 10 days of depletion by PLX5622, microglia in both Cntnap2 KO and VPA underwent repopulation and gradually recovered to the normal level at P65 (Figs. 3 and 7, 8, 9). Correspondingly, neuronal spine density and mature spine numbers in the striatum of the two models also restored to the normal level (Figs. 4I–L and 10 I–L), and their ASD-like behaviors were partially or totally corrected (Figs. 1, 2, 5 and 6). These data implicate that repopulated microglia may push the second phase of synaptic pruning back to a correct track so that proper circuits coding for high functions, such as social interaction and nest building, are likely re-formed.

Noteworthy, no significant difference was observed for microglia number and morphology in the hippocampal CA1 area between male Cntnap2 KO and WT at P21, whereas higher microglia density, more activated state of microglia and lower neuronal dendritic spine density were seen in this region of VPA model mice at this time point (Fig. 6S and Fig. 8). This suggests that microglia in VPA model mice may have more profound influence on neural circuit in the hippocampal CA1 than that in Cntnap2 KO. Indeed, a recent study has demonstrated that prenatal VPA exposure invoked microglial activation and neural circuit dysfunction in the hippocampal CA1 region during the early postnatal periods [69]. In contrast, the striatum is the only tested area in Cntnap2 KO at P65 to see synchronized recovery of microglia morphology and neuronal spine density to the WT level (Figs. 3, 4). Intriguingly, the ventral striatum network is an exception to see abnormal connectivity in Cntnap2 KO at P58 [97]. This suggests that microglia may remodel neuronal circuit in the striatum of Cntnap2 KO through directly regulating neuronal spine plasticity in the adolescent period.

As pointed out earlier, although no change of dendritic spine density was observed in the somatosensory and hippocampal CA1 regions of Cntnap2 KO at P65, microglia in the two areas restored to the WT level upon PLX5622 treatment (Figs. 3 and 4). This implicates that microglia may remodel circuits in these areas through other functions, instead of direct synapse phagocytosis. Accumulating evidence shows that microglia can release growth factors and cytokines, such as TGFβ [9], BDNF [98] and Wnt5a [99], to regulate dendritic spine maintenance and alter neural circuitry. Moreover, upon activation by cytokine IL-33, microglia can engulf the extracellular matrix around synapses to promote synaptic plasticity [100]. Indeed, some cytokines, like BDNF [74] and Wnt5a [101], can signal to ERK1/2, a signal molecule on the MAPK pathway that can be upregulated by CSF1R signaling. Moreover, the MAPK cascade is one of signaling pathways related to microglial inflammatory response in VPA model mice [102]. Prenatal exposure to VPA can lead to epigenetic up-regulation of BDNF in VPA model mice[103]. Reasonably, CSF1R inhibition by PLX5622 might suppress improper cytokine-mediated circuits and downregulate microglia-mediated neuroinflammation in VPA model mice. Incidentally, there is no report about cytokine expression or inflammation in Cntnap2 KO model except LPS-induced production of proinflammatory cytokine IL-17a [104]. Nevertheless, upregulation of another proinflammatory cytokine IL-6, which can suppress synapse pruning [105], was observed in the brain and cerebrospinal fluid of individuals with ASD [106], and blocking IL-6 receptor signaling can alleviate CASPR2 autoantibodies-associated syndrome [107], implicating that IL-6 might engage in inflammation or microglia function in Cntnap2 KO. How cytokines change and mediate microglial function in remodeling synapses and circuits in Cntnap2 KO and VPA model mice needs further investigation.

At the molecular level, VPA can promote epigenetic modification of a wide range of gene expressions in multiple brain regions [43]. In comparison, the expression profile of Cntnap2 overlaps tightly with corticostriatothalamic circuitry [108], and CASPR2, a single-pass transmembrane neuronal adhesion molecule encoded by Cntnap2, exerts functions likely through engagement in protein–protein interaction [109]. Therefore, it is understandable that Cntnap2 KO and VPA model mice might have differential mechanisms in remodeling circuits by microglia that reflects the heterogeneity of microglia. It is a limitation of this study that we didn’t further probe microglia heterogeneity at the molecular level. In the future, it will be needed to use scRNA-seq technology to elucidate how microglial subsets are linked to behavioral variations in Cntnap2 KO and VPA models.

In summary, Cntnap2 KO and VPA model mice share commonality and difference in the timeline of microglia overactivation and depletion, the location of neuronal dendritic spine abnormalities, and the potential mechanism underlying circuit modification. Correspondingly, CSF1R inhibition by PLX5622 has pharmacological effectiveness and heterogeneity in correcting behavioral deficits of the two models. In line with our observation in this study, excessive microglia activation occurs in shared but different brain regions of some individuals with ASD [10, 11, 13], and microglia phenotypes present diversity [8, 15]. This implicates that CSF1R inhibition might be an effective strategy for clinical treatment of ASD, and that individual differences should also be considered.