Single-cell transcriptome analysis of NSCs with conditional Arid1b deletion

In order to determine the causal relationship between adult NSCs and the development of autism in the postnatal brain, we asked whether genetic ablation of autism risk genes exclusively in adult NSCs in either juvenile or adult brains could lead to autism. ARID1B is part of the BAF complex involved in chromatin remodeling, and loss of function mutations in ARID1B have been linked to an increased risk of autism spectrum disorders [22,23,24]. We conditionally deleted Arid1b specifically in NSCs using Nestin-CreERT2;Arid1b (Arid1b) mice at 3 weeks of age through tamoxifen treatment. Subsequent to tamoxifen administration, we confirmed the reduction of ARID1B levels in the dentate gyrus of conditional knockout Arid1b mice (Supplementary Fig. 1A). To effectively monitor NSCs, specifically referred to as radial glia (RGs) in the postnatal brain, we crossed a fluorescent reporter for Cre-mediated recombination, R26-LSL-tdTomato (tdT), into the Nestin-CreERT2 mice (Supplementary Fig. 1B). Also, we confirmed that Arid1b is mainly expressed in the dentate gyrus but not in the SVZ region (Supplementary Fig. 1C, D).

First, to explore the effects of Arid1b ablation on RGs and neurogenesis in the postnatal brain, we conducted single-cell RNA sequencing (scRNA-seq) in the dentate gyrus of control and Arid1b mice after tamoxifen induction. After conducting integration analysis with both the public dataset of the mouse dentate gyrus and our dataset (Fig. 1A and Supplementary Fig. 2A), we discovered that neural cell types in our dataset were aligned with their counterparts in the public dataset [27] (Fig. 1B). Based on this observation, we focused on four key neurogenic cell types: astrocytes, RGs 1, RGs 2, and intermediate progenitor cells (IPCs) (Fig. 1C-E and Supplementary Fig. 2B, C) [28, 29]. These four cell types were selected for further analysis due to the condition knockout of NSC populations by Nestin-Cre in Arid1b mice. Additionally, we confirmed significant changes in the differentially expressed genes (DEGs) of NSCs, while other cell types exhibited minimal alterations in the mutation condition (Supplementary Fig. 2D-F).

The majority of qNSCs exist in a condition known as reversible cell cycle arrest, or quiescence state. This state is divided into deep quiescent and shallow quiescent states, deep quiescent state means complete state of cell cycle arrest, or quiescence state of RG but shallow state have more flexible capacity in RG which can differentiate into active NSCs [30]. We observed enrichment of the RGs 1 cluster in deep quiescence, the RGs 2 cluster in shallow quiescence, and the IPC cluster in active NSCs within the dentate gyrus (Fig. 1F). Notably, we observed indications of increased deep quiescence in the RGs 1 cluster and decreased shallow quiescence in the RGs 2 cluster in Arid1b mice (Fig. 1G, H and Supplementary Fig. 3A). Subsequently, pseudotemporal ordering showing the transcriptional dynamics of RG clusters showed a potential trend towards an increased frequency of less differentiated NSCs in the RGs 1 and 2 clusters of Arid1b samples (Fig. 1I and Supplementary Fig. 3B, C), while the astrogenic lineage was unaffected (Fig. 1J). Consistent with this result, we further observed that down-regulated genes in RGs 1 and RGs 2 of Arid1b mice were mostly associated with neurogenesis, NSC maintenance, and neuronal differentiation (Fig. 1K, L). These observations indicate a possible retardation in neurogenic differentiation from undifferentiated qNSCs.

The aberrant qNSC activity in the Arid1b mice was further confirmed by clonal analysis in vivo by targeting a few individual RGs. Treatment with low-titer tamoxifen in Arid1b mice, carrying the tdT reporter, led to limited clonal labeling of isolated RGs [31, 32]. This enabled us to closely monitor their activity and progression within the lineage on a clone-by-clone basis. To be more specific, the analysis of cell composition within isolated clones provided insights into whether recombined RG remained in a quiescent state, underwent division, either symmetrically or asymmetrically, and, in the end, if they were depleted from the clone within a one-month timeframe (Fig. 2A and Supplementary Fig. 4A, B).

Our examination of cellular composition indicated a higher occurrence of quiescent clones, as indicated by single radial glia (R) in Arid1b mice, while the ratios of amplification clones, characterized by two radial glia (R+R), and maintenance clones, represented by radial glia and neurons (R+N), showed a decrease (R = RGs, N = Neurons, A = Astrocytes, X = N or A) (Fig. 2B and Supplementary Fig. 4C). Consistent with these results, we observed that the number of Id4+, tdT+ quiescent RGs was significantly increased, whereas the number of Ki67+ tdT+, Tbr2+ tdT+, EdU+ tdT+ active, proliferating IPCs and Prox1+ tdT+, Dcx+ tdT+ neuroblasts significantly declined during neurogenesis in Arid1b mice (Fig. 2C-L and Supplementary Fig. 4D). In contrast, the total number of Prox1+ neurons in Arid1b juvenile mice remained unchanged (Supplementary Fig. 4E). Additionally, the TMZ assay, designed to confirm the restoration potential of quiescent RGs through the elimination of active IPCs exhibit decreased EdU+ cells at 7 days after TMZ treatment in Arid1b mice (Fig. 2M, N), implies a delay in neurogenesis due to increased quiescent-like state in these cells.

Next, we sought to investigate whether loss of Arid1b in adult RGs might lead to aberrant neurogenesis. We administered tamoxifen to 3 months old Nestin-CreERT2; Arid1b mice and observed a complete loss of Arid1b expression in nestin-positive radial glia, indicating non-mosaic deletion in these cells (Supplementary Fig. 4F). Three weeks after tamoxifen treatment, we observed a significant increase in the number of Id4+ tdT+ quiescent radial glia (qRGs) (Fig. 2O, P). Morever, the total number of Prox1+ neurons, EdU+ tdT+ actively proliferating IPCs, and Prox1+ tdT+ neuroblasts in adult Arid1b mice was consistent with our findings in juvenile Arid1b mice (Fig. 2Q, R and Supplementary Fig. 4G, H).

Next, to investigate whether the selective loss of Arid1b in RGs induces autistic-like phenotypes, we conducted a series of behavioral tests. In the three-chamber assay, both Arid1b control and Arid1b male mice explored the lateral chambers equally during habituation. However, Arid1b mice exhibited reduced sociability with a novel mouse compared to an empty cage (Fig. 3A, B). To assess the impact of Arid1b loss on short-term social memory acquisition, we employed a three-trial paradigm (Fig. 3C), [33]. We observed that the initial sniffing response to stranger 1 in Arid1b mice was reduced (Fig. 3D, E), consistent with the diminished sociability shown in Fig. 3B.

Moreover, Arid1b mice displayed a shorter duration of sniffing when interacting with a stranger 2 mouse (Fig. 3D, E). However, we found that both groups exhibited similar interaction times with a familiar mouse (Fig. 3D, E). These observations suggest that Arid1b mice exhibited a deficit in short-term social memory acquisition, which manifested certain behavioral traits commonly seen in autism. However, during the open field test, Arid1b mice exhibited exploratory behavior that did not significantly differ from their littermates (Fig. 3F-H). Subsequently, in marble burying and self-grooming tests, no noteworthy behavioral changes in Arid1b mice were observed (Fig. 3I-K).

Moreover, adult Arid1b mice also exhibited reduced sociability when interacting with a novel mouse compared to an empty cage in the three-chamber assay (Supplementary Fig. 5A, B). They showed a shorter duration of sniffing when interacting with a stranger 2 mouse in the three-trial paradigm assay (Supplementary Fig. 5C, D), demonstrating certain autistic-like behaviors, including social preference and memory, consistent with our findings in juvenile Arid1b mice.

Previous studies have demonstrated that social interactions induce neural activation in hippocampal regions, particularly the dentate gyrus (DG), including the granule cell layer (GCL), and the CA3 regions, as measured by c-Fos expression [34]. To link impaired neurogenesis caused by Arid1b-deficient radial glia to functional circuit changes relevant to social behavior, we investigated c-Fos-dependent neural activity in hippocampal regions, including the DG, CA1, and CA3, after social interaction. Upon initial examination of c-Fos activity in the hippocampus to the social interaction, we observed an increase in c-Fos+ cells within the DG and CA3 regions of Arid1b mice, whereas the CA1 region exhibited no significant activation (Fig. 3L-N and Supplementary Fig. 5E). To understand how delayed neurogenesis may disrupt neuronal activity and synaptic transmission in the autistic brain, we examined the structural attributes of granule cells in Arid1b mice carrying the tdT reporter. There was a significant reduction in the width of the calbindin-labeled main bundle and tdT+ axons in Arid1b mice after tamoxifen treatment at both post-injection day (dpi) 21 and dpi 60 (Fig. 3O, P, Supplementary Fig. 5F). However, there was no discernible effect on the length of the main bundle and tdT+ axons (Fig. 3Q). These data indicate that disrupted neuronal activity and synaptic transmission in the DG-CA3 result from delayed neurogenesis and suggest a potential link to impairments in social behaviors.

ARID1B is a component of the BAF (SWI/SNF) chromatin remodeling complex, which plays a crucial role in governing gene expression through the control of chromatin dynamics [22,23,24]. Thus, to explore how Arid1b gene disruption in radial glia of the dentate gyrus triggers autistic behavior, we sought to uncover the epigenetic modifications governed by Arid1b in RGs. Interestingly, gene set enrichment analysis (GSEA) revealed that H3K27me3 target genes were negatively enriched in the RG clusters of Arid1b mice, including gene sets marked by H3K27me3 but not H3K4me3, and also gene sets co-marked by H3K27me3 and H3K4me3 (Fig. 4A, B and Supplementary Fig. 6A). Moreover, gene ontology analysis (FDR = 0.05) showed that the gene signatures exhibiting negative correlation were linked to processes of neuronal differentiation, proliferation, and development (Fig. 4C). These results indicate that a deficiency in Arid1b led to adverse effects on the development and differentiation of RGs. Consistent with this idea, we confirmed an increase in the levels of the heterochromatin-associated H3K27me3 mark within the dentate gyrus area of Arid1b mice, whereas there was no such change observed in the prefrontal cortex (Fig. 4D, E). In contrast, no difference in the levels of H3K27ac and H3K9me3 was detected between control and Arid1b mice (Fig. 4D, E and Supplementary Fig. 6B).

Thus, to more accurately assess the impact of Arid1b deficiency on the epigenetic landscape of RGs, we performed Cleavage Under Targets & Release Using Nuclease (CUT&RUN) sequencing for H3K27me3 and Ezh2, key enzyme responsible for directly catalyzing H3K27me3, on RGs of the dentate gyrus. Consistently, Ezh2 and H3K27me3 occupancy across transcriptional start sites (TSS) was generally increased in the absence of Arid1b in RGs of the dentate gyrus (Fig. 4F, G). Furthermore, Ezh2 occupancy at enhancers was also increased in Arid1b RGs, whereas no differences were observed in H3K27me3 levels at these regions (Supplementary Fig. 6C). Moreover, we identified an overlap of 2160 genes showing elevated levels of both Ezh2 and H3K27me3 in the context of Arid1b (Fig. 4H). Within this group, we observed a subset of genes related to neurogenesis and neuron projection, including Brd4, Mki67, Ctnnb1, Wnt2b, and Homer2 (Fig. 4I, J).

We examined the impact of suppressing H3K27me3 accumulation by delivering lentiviruses expressing shRNA targeting Ezh2 into the SGZ of the dentate gyrus in Arid1b mice. Ezh2 intensity was reduced in dentate gyrus of Arid1b mice treated with shEzh2 (Supplementary Fig. 7A, B). After 21 days, there was a marked increase in EdU+ tdT+ proliferating RGs and Prox1+ tdT+ immature neurons in the dentate gyrus in response to Ezh2 knockdown in Arid1b mice (Fig. 5A-D). Next, we sought to determine whether Arid1b-mediated social behavioral abnormalities also could be rescued by Ezh2 knockdown. Importantly, we found a profound improvement in social interaction and short-term social memory acquisition among Arid1b mice treated with Ezh2 shRNA in comparison to the Arid1b mice (Fig. 5E-H). Arid1b mice did not exhibit any significant behavioral changes in open field tests following Ezh2 knockdown, consistent with these behaviors being unaffected by Arid1b deficiency (Supplementary Fig. 7C). Importantly, c-Fos+ neuronal activity in the DG and CA3 regions was significantly restored upon encountering a novel mouse in Arid1b mice treated with shEzh2 (Fig. 5I, J). Additionally, the width of the calbindin-labeled main bundle and the tdT+ axon were also rescued in Arid1b mice treated with shEzh2 (Supplementary Fig. 7D, E).

We next asked whether pharmacological inhibition of EZH2 could rescue autistic phenotype using GSK-126, a highly selective inhibitor of EZH2 methyltransferase [35, 36], administering it by intraperitoneal injection into 3-week-old Arid1b mice after tamoxifen induction. We firstly observed a significant increase in Ki67+ tdT+ proliferating RGs and Prox1+ tdT+ immature neurons among the total tdT+ cells in the dentate gyrus of Arid1b mice treated with GSK-126 (Supplementary Fig. 8A-C). Notably, the elevated H3K27me3 levels observed in Arid1b mice were reduced in the dentate gyrus of Arid1b mice treated with GSK-126 (Supplementary Fig. 8D, E). We further observed that a subset of genes related to neurogenesis and neuron projection, including Brd4, Mki67, Ctnnb1, Wnt2b, and Homer2, in GSK-126 treated Arid1b mice was slightly increased compared to Arid1b mice (Supplementary Fig. 8F). Following the GSK-126 treatment, we monitored autistic phenotypes by conducting the social preference test from day 0 to day 21. We confirmed that Arid1b mice exhibited a distinct autistic phenotype starting from 14 days after tamoxifen treatment initiation. Remarkably, GSK-126 treatment prevented the development of any social deficit (Fig. 5K and Supplementary Fig. 8G). In conclusion, these findings demonstrate the potential for enhancing neurogenesis from RGs by preventing aberrant H3K27me3 accumulation in order to alleviate autism-like social deficits associated with Arid1b-related ASD.

We next proceeded to examine the molecular phenotypes associated with autism in human cells lacking ARID1B. We first generated human induced pluripotent stem cell (iPSC) lines with ARID1B mutation using CRISPR/Cas9 targeting exon 5 of Arid1b. Reductions in ARID1B protein was confirmed in independently targeted ARID1B mutant iPSCs (Supplementary Fig. 9A), and these cells were then further differentiated into neural stem cells, including RGs in 3D brain organoids. We initially investigated whether ARID1B mutant could induce abnormal human RG activity, consequently leading to aberrant neurogenesis. Notably, we observed a substantial increase in the number of NESTIN+,TNC+, KI67- and NESTIN+, SOX2+, ID4+ quiescent RGs within these NSC cultures (Fig. 6A, B and Supplementary Fig. 9B, C). Subsequently, we quantified the population of actively proliferating NSCs in response to the introduction of EGF and bFGF. After 7 days of culturing hRGs with proliferative stimuli EGF and bFGF, there was a significant reduction in the number of NESTIN+, TNC+, and EdU+ active NSCs within the ARID1B-mutant NSCs relative to their isogenic control cultures (Fig. 6C, D). Thus, these findings imply a linkage between aberrantly quiescent hRGs and ARID1B mutation.

To further investigate whether EZH2 functionally contributes to ARID1B mutant autism phenotypes, we transduced ARID1B mutant human RG with lentivirus encoding EZH2 shRNA. The proportion of NESTIN+, TNC+, and EdU- quiescent RGs was markedly diminished in ARID1B mutant NSCs upon EZH2 knockdown (Supplementary Fig. 9D, E). Consistently, we observed a significant increase in the proportion of NESTIN+, TNC+, and EdU+ active NSCs in EZH2 shRNA-treated ARID1B mutant NSCs (Supplementary Fig. 9D, F). Remarkably, we identified a significant increase in DCX+ neuroblasts and KI67+ proliferating cells in ARID1B- mutant NSCs upon EZH2 knockdown (Supplementary Fig. 9G, H). Additionally, we observed a reduction in NESTIN+, KI67+ NSCs and NESTIN-, KI67+ NSCs in the human brain organoids with ARID1B+/- mutation (Supplementary Fig. 10A, B). Consistent with previous result, after 21 days of culture with GSK-126 treatment, the number of NESTIN+, KI67+ NSCs and NESTIN-, KI67+ NSCs was significantly increased in the ARID1B+/- mutant brain organoids ((Supplementary Fig. 10A, B). Notably, the elevated H3K27me3 levels observed under ARID1B mutation were reduced in ARID1B+/- mutant brain organoids treated with GSK-126 (Supplementary Fig. 10C).

To further validate abnormal qNSC activity in autism patient NSCs, we conducted re-analysis of scRNA-seq data obtained from cortical organoids with ARID1B mutation (Fig. 6E and Supplementary Fig. 11A, B). We observed that mouse NSC types at the juvenile stage exhibited a high degree of transcriptional similarity when compared to previous data from human cortical organoids (Supplementary Fig. 11C). By comparing the ratios of different cell types within the entire cell population of the mutant brain organoids to that of the control organoids, we observed a consistent increase in the cluster of apical radial glia (aRG), which represents qNSCs (Fig. 6F). Additionally, we noted a decrease in the clusters of proliferative active RGs and IPCs in the ARID1B human cerebral organoids (Fig. 6F). Moreover, cell-cycle scoring analysis confirmed a significant increase in the G1 phase and a decrease in the G2/M phase within the aRG and proliferating aRG clusters, suggesting a decrease in cell cycle and proliferation of NSCs in organoids harboring the ARID1B mutation (Fig. 6G). Additionally, based on the precisely defined origin of RGs within the human cortical germinal zone, as demonstrated by [37] (Fig. 6H), we further investigated transcriptional alterations in ARID1B-mutant RGs, specifically in relation to the expression of cell cycle genes. Remarkably, the G1 phase signature of human RGs, exhibited an increase in aRG clusters in ARID1B mutant organoids (Fig. 6I). However, the G1/S and G2/M phase signature of human RGs decreased in the aRG cluster, indicating an enhanced quiescent-like state in human ARID1B-mutant RGs (Fig. 6I and Supplementary Fig. 12A-D).

We further explored the transcriptional properties of human RGs harboring mutations in other BAF complex members, BCL11 and SMARCC2, to assess whether they exhibit similar transcriptional landscapes to the ARID1B mutant RGs [38]. After clustering, we sub-sorted outerRGs (refer to oRGs) expressing HOPX, RGs expressed VIM and GFAP, cycling RG (refer to ccRGs) expressing TOP2A and intermediate progenitor cells (IPC) expressing EOMES (Fig. 6J-L and Supplementary Fig. 12E, F). Next, we confirmed the cell density pattern of RGs and the pseudotime trajectory from RGs to IPCs in these mutant organoids. Importantly, we observed increased quiescence in the least undifferentiated RG clusters such as oRG or RGs and a decrease in the IPC cluster in cultures with these BAF mutations (Fig. 6M), indicating that aberrant quiescent NSC states potentially serve as a key mechanism underlying the pathogenesis of autism driven by mutations in the BAF complex.

Finally, we aimed to determine whether aberrant activity of quiescent RGs could be observed in sporadic autism patients. Upon analyzing published single-cell transcriptome datasets from the brains of healthy control and autistic male patients aged 19-21 years [39], we identified cluster 3 that exhibited a transcriptional signature similar to both our mouse RGs and the other mouse NSCs of the dentate gyrus regions [27] using module scoring analysis (Fig. 6N, Supplementary Fig. 12G). To investigate the cell-cycle-dependent transcriptome of human RG cells (RGs), we identified sub-sorted 3 clusters of RGs, hRG1, hRG2, and hRG3 (Fig. 6O and Supplementary Fig. 12H). Among these, hRGs 2 showed a significantly increased G1 phase signature in ASD samples, when inter-individual variability was accounted for using a linear mixed-effects model (Fig. 6P). While other clusters and phases did not reach statistical significance, the direction of change was largely consistent, suggesting a potential trend toward altered cell cycle dynamics across multiple hRG subtypes (Supplementary Fig. 13). However, since these clusters were derived from multiple patients, additional future studies are necessary for patient-specific analyses to examine the cell cycle signatures within each individual patient's samples separately. In addition, we identified that the hRG1 cluster exhibited human oRG markers [37] compared to hRG2 and hRG3 (Supplementary Fig. 12I). In summary, these results suggest that the aberrant activity of qNSCs can be observed more broadly in the brains of individuals with autism. Targeting the activity of qNSCs in autism patients could serve as a promising therapeutic strategy for the treatment of autism.