Cell type-specific DNA methylome signatures reveal epigenetic mechanisms for neuronal diversity and neurodevelopmental disorder

Abstract

DNA methylation plays a critical function in establishing and maintaining cell identity in brain. Disruption of DNA methylation-related processes leads to diverse neurological disorders. However, the role of DNA methylation characteristics in neuronal diversity remains underexplored. Here, we report detailed context-specific DNA methylation maps for GABAergic, glutamatergic (Glu) and Purkinje neurons, together with matched transcriptome profiles. Genome-wide mCH levels are distinguishable, while the mCG levels are similar among the three cell types. Substantial CG-differentially methylated regions (DMRs) are also seen, with Glu neurons experiencing substantial hypomethylation events. The relationship between mCG levels and gene expression displays cell type-specific patterns, while genic CH methylation exhibits a negative effect on transcriptional abundance. We found that cell type-specific CG-DMRs are informative in terms of represented neuronal function. Furthermore, we observed that the identified Glu-specific hypo-DMRs have a high level of consistency with the chromatin accessibility of excitatory neurons and the regions enriched for histone modifications (H3K27ac and H3K4me1) of active enhancers, suggesting their regulatory potential. Hypomethylation regions specific to each cell type are predicted to bind neuron type-specific transcription factors. Finally, we show that the DNA methylation changes in a mouse model of Rett syndrome, a neurodevelopmental disorder caused by the de novo mutations in MECP2, are cell type- and brain region-specific. Our results suggest that cell type-specific DNA methylation signatures are associated with the functional characteristics of the neuronal subtypes. The presented results emphasize the importance of DNA methylation-mediated epigenetic regulation in neuronal diversity and disease.

Introduction

Normal neuronal functions depend on the interaction and balance of signaling across diverse neurons. Learning, memory and distinct cognitive functions are preferentially associated with particular regions and/or neuronal populations in the brain (1). In adult animals, different neuron populations display unique gene expression patterns, broadly establishing their distinct functions (2,3). DNA methylation is regarded as a stable covalent modification in post-mitotic cells to influence spatiotemporal gene expression and to define cell identity. In differentiated mammalian cells, DNA methylation is catalyzed by DNA methyltransferases and mostly established on CpG dinucleotides (mCG). The methylation signal can be recognized by a group of protein ‘readers,’ such as methyl-CpG binding protein 2 (MeCP2), through their methyl-CpG binding domain (4). DNA methylation-based regulation often occurs by altering DNA–protein interactions, such as the binding of transcription factors to DNA.

Identification and characterization of DNA methylation features in different neuronal cell types are important for understanding the functional differences that contribute to neuron diversity and plasticity. In comparison with other tissues or cells, the composition and dynamics of DNA methylation have been shown to be highly distinct in the brain (2). Glutamatergic (Glu; excitatory) and GABAergic (GABA; inhibitory) neurons are two major functional classes of mammalian neocortical neurons, which constitute about 80 and 20% of all cortical neurons, respectively (5). GABA neurons regulate the rate and temporal architecture of the inhibitory output, while the Glu neuronal signaling sustains the long-range excitatory output of the brain (6,7). In adult mouse cortex, widespread DNA methylation differences have been discovered between excitatory and inhibitory neurons (parvalbumin and vasoactive intestinal peptide neurons) (2), which reflect distinct mechanisms of gene regulation in maintaining cellular diversity. DNA methylation landscapes also show significant differences between GABA and Glu neurons in human prefrontal cortex (3). However, despite considerable efforts, features of those specific locations with methylation differences among different cell types remain underexplored. Very little is known about the functional roles of DNA methylation in neuronal diversity due to the lack of detailed context-specific DNA methylation maps. Purkinje neurons, another type of inhibitory neuron, play critical roles in motor control and learning. Purkinje neurons are one of the largest neurons in the brain and are a class of GABAergic inhibitory neurons located in the cerebellum. As the sole output neurons of the cerebellar cortex, Purkinje neurons regulate the activation of excitatory neurons via interactions with their dendrites (8). A previous study found that Purkinje neurons experienced extensive DNA demethylation and subsequent remethylation (including 5-methylcytosine and 5-hydroxylmethylcytosine) in mouse post-mitotic cerebellum during normal development (9). Therefore, the dynamic and distinct methylation signature in Purkinje cells could be critical for maintaining cell stability and specificity.

It is well known that dynamic DNA methylation plays a critical role in learning and memory. DNA methylation-associated gene regulation is important for many neurodevelopmental disorders (10). De novo mutations in MECP2 lead to Rett syndrome (RTT), which is an X-linked, postnatal neurological disorder (11). Mice with a global deletion of Mecp2 show abnormal long-term neural plasticity and cognitive deficits (12). Despite decades of research, the precise mechanism by which MeCP2 drives RTT phenotypes remains unclear. Earlier studies found that MeCP2 interacts specifically with methylated cytosine in the CG context (mCG) (4,13,14). More recent studies have shown that MeCP2 has high affinity for non-CG context (mCH, where H = A, C or T) in adult mouse brain (15–19). Gene expression studies have indicated that MeCP2 is involved in the regulation of gene expression as a repressor or activator at transcriptional or post-transcriptional levels (20,21). However, only subtle alterations of gene expression were detected from transcriptomic studies of brains using either human postmortem RTT samples or Mecp2 knockout (MeCP2 KO) mice, where it is challenging to distinguish direct targets of MeCP2 regulation from secondary gene expression effects (22–24). Another challenge is the diversity of cell types in the brain. Cellular heterogeneity has largely precluded the characterization of cell type-specific features in RTT and many other neurological disorders. Discrete RTT features have been found to associate with dysfunction of discrete neuronal populations. For example, knockout of MeCP2 in GABA neurons can recapitulate most of the RTT-associated symptoms, including repetitive behaviors (18,25), whereas MeCP2 deficiency in Glu neurons leads to early lethality and some anxiety-like behaviors (26). Consequently, the loss of MeCP2 alters gene expression in a cell type-specific manner, which could be regulated by unique alternations of DNA methylation (27). However, the cell type-specific DNA methylation signatures upon the loss of MeCP2 have not been fully characterized.

To address these knowledge gaps, we applied INTACT (isolation of nuclei tagged in specific cell types) to affinity purified labeled nuclei from genetically defined neuron populations. The DNA methylome was characterized in GABA, Glu and Purkinje neurons using WGBS at a single base resolution. We also evaluated the regulatory roles of the identified cell type-specific differentially methylated regions (DMRs). The study was further extended by simultaneously accessing the DNA methylation alterations in GABA and Glu neurons of Mecp2 KO mice. Taken together, our results suggest that cell type-specific DNA methylation signatures are associated with the functional characteristics of the neuronal subtypes. The DNA methylation changes are largely neuronal cell type- and brain region-specific in a mouse model of Rett syndrome. These findings further emphasize the important role of DNA methylation-mediated epigenetic regulation in neuronal diversity and disease.

Results

Distinct DNA methylomes of different neuronal types revealed by WGBS

To capture genetically defined neuron populations, we performed INTACT. We generated three mouse lines by crossing SUN1 mice (SUN1-sfGFP-Myc) with Camk2a-Cre transgenic mice, Gad2-IRES-Cre knock-in mice and Pcp2-Cre transgenic mice, respectively (Fig. 1A). GFP-labeled nuclei were then affinity purified from fresh tissue homogenates using an anti-GFP antibody. The DNA methylation landscape was mapped using WGBS for three mouse neuronal populations: GABA and Glu neurons isolated from cortex, as well as Purkinje neurons purified from the cerebellum (Supplementary Material, Table S2). Principle component analysis (PCA) using whole-genome mCG and mCH sites revealed a clear segregation of neuron types (Supplementary Material, Fig. S1A and B). Consistent with the PCA results, a high correlation between samples for the same cell type was observed by using both mCG and mCH information (Supplementary Material, Fig. S1C and D). We then observed distinct DNA methylome signatures among the three neuronal cell types that we examined. As shown in Fig. 1B, genome-wide mCG levels were consistent with the value of ~78% among all three neuronal populations, while mCH levels were distinguishable. Glu neurons exhibited the highest global mCH level (~2.5%) followed by GABA neurons (~2%), whereas significantly lower mCH levels were detected in Purkinje neurons (~1.3%). We next examined the DNA methylation levels across genomic features, and again, we observed highly variable mCH levels across the three neuronal cell types: mCH levels in Glu neuron were 2-fold higher than that in Purkinje neurons. Such differences extended 10 kb away from both transcription start sites (TSS) and transcription end sites (TES). This observation is consistent with the previous findings that CH methylation is an important indicator of the identity of mammalian neurons (19,28,29). In addition, there was marked reduction of DNA methylation surrounding the TSS, accompanied by slight gains of mCG but loss of mCH throughout the gene body (Fig. 1C).

Figure 1.

Distinct DNA methylomes in different neuronal types. (A) Schematic of INTACT method applied to isolate GABA, Glu and Purkinje neurons from mouse cortex or cerebellum of WT mouse. (B) Bar graph showing genome-wide mCG (left) and mCH (right) levels in each cell type. The methylation level for each cell type was averaged for two replicates. GABA represents GABAergic neuron; Glu represents Glutamatergic neuron; Purkinje represents Purkinje neuron. (C) Metagene plots showing mCG (top) and mCH (bottom) levels across TSS, TES and RefSeq gene bodies. Ten kb upstream and downstream of given genomic features were plotted. TSS, transcriptional start sites; TES, transcriptional end sites. (D) Identification of DMRs between three cell types. DMRs, differential methylation regions. (E) Heatmap visualization of the methylation profiles of DMRs identified between the three cell types. A scale is shown on the right, in which red and blue correspond to a higher and a lower methylation status, respectively. (F) Genomic annotation of DMRs between cell types to their percentage of each genomic feature and enrichment versus expected values.

We next sought to define differences between methylomes in an unbiased manner. The high percent methylation at CG sites in the genome provides sufficient statistical power to perform de novo calling of DMRs. We therefore examined local genomic regions that differ in their CG methylation levels across three neuronal populations. The identified CG-DMRs represent genomic regions ranging in size from 51 bp to 8.3 kb from the three pairwise comparisons by using stringent statistical method (P < 1e−5, methy.diff >0.1) (Fig. 1D; Supplementary Material, Table S3). The smallest number of DMRs (n = 4330) were detected between GABA and Purkinje neurons, with both neuronal populations can release neurotransmitter GABA and regulate the activation of excitatory neurons in cortex and cerebellum, respectively. In contrast, the largest number of DMRs was identified between Glu and Purkinje neurons (n = 22 553), with more than 70% of the DMRs are hypomethylated in Glu neurons. The substantial hypo-DMRs detected in Glu neurons, demonstrating the profiling for comprehensive identification of regulatory regions (2). We also noted that in our study, the GABA and Glu neurons purified from mouse cortex represent heterogeneous pools of inhibitory and excitatory neurons, whereas the purified Purkinje cells are relatively homogenous. Distinct subpopulations of the inhibitory and excitatory neurons might have their specific epigenetic characteristics. The identified DMRs in GABA and Glu neurons could most likely represent common DNA methylation changes occurred in inhibitory and excitatory neurons, respectively. As expected, the three cell types were correctly separated by the methylation profiles of these obtained DMRs (Fig. 1E). Genomic annotation of DMRs reveals biased genomic locations: nearly half of DMRs were in intronic regions, and ~30% of DMRs were found in distal intergenic areas (>5 kb away from known transcription start sites) (Fig. 1F; Supplementary Material, Table S4). Among the annotated genomic features, the DMRs at the 5′UTR showed the highest enrichment versus expected value.

Nuclei RNA-Seq revealed unique gene expression patterns for different neuronal types

In addition to WGBS, transcriptomic profiling was performed using nuclei RNA-Seq for the same INTACT samples to link the differential methylation and altered gene expression. Nuclei RNA-Seq provides a focused snapshot of the nuclear transcriptome, warranting the identification of nascent, neuronal activity-associated mRNAs (30). We observed substantial differentially expressed genes (DEGs, Padj < 0.01 and |log₂ fold-change| > 2) between the gene expression profiles of different cell types (Supplementary Material, Fig. S2A–C and Supplementary Material, Table S5). The majority (77–79%) of identified DEGs belong to protein coding genes, while 12–13% were lncRNAs (Supplementary Material, Fig. S2D–F). We therefore excluded noncoding DEGs from further analyses. Gene Ontology (GO) analysis found significant enrichment of multiple brain-related functions for DEGs (Supplementary Material, Fig. S2G–I). As expected, we observed multiple cell type-specific neuronal markers enriched in the purified nuclei, including Slc17a, Tbr1 and Camk2a for Glu neurons; Gad2, Sst and Lhx6 for GABA neurons, and the Purkinje markers of Pcp2 and Pcp4 (Supplementary Material, Fig. S3A). Pairwise comparisons of the RNA-Seq data enabled the discovery of gene expression differences between two neuronal types. However, many genes had similar expression patterns in the different cell types. We further profiled cell type-specific DEGs (Supplementary Material, Fig. S3B and Supplementary Material, Table S6), which showed unique gene expression patterns in one cell type compared to the other two. Similar numbers of uniquely expressed genes are distributed in GABA and Glu neurons (n = 127 in GABA and n = 136 in Glu). Accordingly, the GABA-specific DEGs function as receptor binding, behavior and forebrain development (Supplementary Material, Fig. S3D), while the Glu-specific DEGs were mainly involved in the development of sensory organs, eye and visual system. Notably, over 1500 genes were specifically regulated in Purkinje neurons, with significant enrichment in synaptic signaling and transmission processes. These results suggest unique gene expression signatures for each cell type. The observed differences were compatible with their biological functions.

Correlation between DNA methylation and gene expression

Among the three comparisons, the smallest gene expression alternations were found between GABA and Glu neurons (Supplementary Material, Fig. S2A). Nearly 21% of these detected protein coding DEGs were differentially methylated and showed significant enrichment in neuronal-related biological functions, such as neuron differentiation and neurogenesis (Fig. 2A and Supplementary Material, Table S7). In contrast, Purkinje neurons displayed remarkable gene expression changes compared to the other two cell types (Supplementary Material, Fig. S2B and C). A sizable number of DEGs [337 (16%) from GABA versus Purkinje, and 1006 (40.3%) from Glu versus Purkinje] were differentially methylated (Fig. 2B and C and Supplementary Material, Table S7). GO analysis identified the DEGs to be associated with neuronal function and synaptic activities (Fig. 2D–F).

Figure 2.

Correlation between DNA methylomes with gene expression. (A–C) Identification of DEGs showing differential methylation between cell types. (A) GABA versus Glu; (B) GABA versus Purkinje; (C) Glu versus Purkinje. (D–F) GO analysis on genes showing differential expression and differential methylation between cell types indicated that these genes were significantly involved in neuronal functions. (D) GABA versus Glu; (E) GABA versus Purkinje; (F) Glu versus Purkinje. (G) Metagene profiles revealed negative relationship between genic mCG levels and differential gene expression in Glu and Purkinje neurons. For each comparison, we grouped genes into three categories based on their differential gene expression between cell types. Cell type-enriched genes indicate genes displayed significant upregulation (log₂ fold change >1 and Padj < 0.01) in this cell type compared to the other cell type. Non-DEGs are for genes stably expressed between cell types with no significant changes (|log₂ fold change| < 0.1 and Padj > 0.05). (H) Metagene profiles revealed clearly negative relationship between genic mCH levels and differential gene expression in Glu and Purkinje neurons but not in GABA neurons. For each comparison, we grouped genes into three categories based on their transcriptional difference between cell types. Cell type-enriched genes indicate genes displayed significant up-regulation (log₂ fold change >1 and Padj < 0.01) in this cell type compared to the other cell type. Non-DEGs are for genes stably expressed between cell types with no significant changes (|log₂ fold change| < 0.1 and Padj > 0.05).

Previous studies revealed that the intragenic DNA methylation is highly informative for gene expression in the mammalian brain (2,28). In this study, as shown in Fig. 2G and H, metagene analysis indeed revealed a negative relationship between gene body DNA methylation (both mCG and mCH) and differential gene expression in Glu and Purkinje neurons. In contrast, the upregulated and downregulated DEGs in GABA neurons displayed similar CG or CH methylation levels. We then extracted mCG and mCH levels from gene bodies for each DEGs and non-DEGs groups in each cell type and conducted a paired t-test based on non-overlapping bins towards non-DEGs. We found that DNA methylation levels (both mCG and mCH) are dramatically different between upregulated and downregulated DEGs in Glu and Purkinje neurons (Supplementary Material, Fig. S4A and B, P < 0.0001). We also found that the inverse relationship between mCH and differential gene expression extended to 10 kb regions away from TSS and TES (Fig. 2H). This observation is consistent with the previous studies that neuronal mCH is reduced in actively expressed genes, with such inverse correlations detected throughout the 5′-upstream, gene body and 3′-downstream regions (2,19,28,29). Nevertheless, Glu and Purkinje neurons also exhibited their own specific patterns. For example, the downregulated genes in Glu neurons exhibited significantly higher CG methylation levels than non-DEGs (Supplementary Material, Fig. S4A and B, P < 0.0001). In Purkinje neurons, similar mCH levels were detected for non-DEGs and actively expressed ones, which are dramatically lower than the repressed genes (Supplementary Material, Fig. S4A and B, P < 0.0001). These results demonstrate substantial differences in the DNA methylomes between different neuronal types. Additional Spearman correlation analysis found a negative but negligible correlation (P < 0.01) between genic DNA methylation and the expression of genes showing upregulation in each cell types compared with the other two, when limiting the gene sets to those with RPKM less or equal than 100 (Supplementary Material, Fig. S4C–E). Over the same range of gene expression levels, CH methylation displayed stronger negative correlation than mCG.

To further evaluate the impact of DNA methylation on overall gene transcription in each cell type, we compared DNA methylation values in-and-around gene bodies across four gene expression quintiles based on FPKM values (Supplementary Material, Fig. S5A–C). Both CG and CH methylation levels around TSS regions displayed a clearly negative correlation with transcriptional abundance. However, the relationship between levels of gene body methylation and gene expression is different between mCG and mCH. Generally, actively expressed genes in the third and fourth quintiles of GABA and Purkinje neurons have the highest mCG levels, while genes in the first quintile showed the lowest DNA methylation percentages, suggesting positive correlation of CG methylation and gene expression. In Glu neurons, the mid-level expressed genes in the second quintile showed higher gene body mCG levels than genes expressed in the third and fourth quintiles, indicating a unique relationship between CG methylation and gene expression. In contrast, we discovered that gene body CH methylation levels have a clear negative correlation with gene expression in all three cell types. The percentages of mCH for highly expressed genes (expressed at third and fourth quintiles) are relatively lower than genes in the first and second quintiles, further supporting a potential repressive role for mCH on gene expression.

Cell type-specific hypo-DMRs occur in candidate regulatory regions

To further characterize the unique methylation signature for each neuronal cell type, we identified cell type-specific DMRs, which show either hyper- or hypomethylation in at least one cell type from the three comparisons. Notably, Glu neurons have the largest number of hypo-DMRs, while Purkinje cells have the highest number of hyper-DMRs (Fig. 3A). These DMRs were restricted to gene introns and intergenic areas. Lister et al. (28) found that intragenic mCG patterns were informative on gene function. We then used GREAT (Genome Regions Enrichment of Annotation Tool) to perform GO analysis to further explore the biological significance of those DMRs. Of note, we observed an inverse correlation between DNA methylation status and gene function (Fig. 3B and C). For instance, the GABA- and Glu-specific hyper-DMRs are associated with genes responsible for cerebellum and hindbrain morphogenesis. In addition, genes related to Purkinje cell development, morphogenesis and differentiation were hypermethylated in GABA neurons. In contrast, hypo-DMRs uniquely identified from Purkinje neurons were related to Purkinje cell activity and cerebellum development. The Glu hypo-DMRs were specifically enriched for genes involved in the processes of long-term synaptic potentiation and regulation of neuronal synaptic plasticity. These results indicate cell type-specific and developmentally regulated activities associated with DNA methylation.

Figure 3.

Cell type-specific hypo-DMRs are enriched at regulatory regions and selective binding motifs. (A) Identification of cell type-specific DMRs. Top, Number of cell type-specific DMRs in each cell type. Glu neurons have the largest number of hypo-DMRs, while Purkinje neurons contain the greatest number of hyper-DMRs. Bottom, Genomic annotation of cell type-specific hyper- and hypo-DMRs in each cell type. GO analysis for cell type-specific hyper-DMRs (B) and hypo-DMRs (C) was examined for their biological significance using GREAT tool. GO biological processes associated with brain functions are highlighted in red or blue words. (D) Glu-specific hypo-DMRs are associated with chromosome accessibility and histone modifications of active enhancers (H3K4me1 and H3K27ac). Venn diagram showing the number of Glu-specific hypo-DMRs overlapped with peaks of ATCT-Seq, H3K27ac and H3K4me1 of excitatory neurons, respectively. IGV browser representation of DNA hypomethylation of the Arpp21 gene in Glu neurons. Peaks in ATAC-Seq read density and marked by histone modifications in excitatory neurons are shown in lower tracks. (E) Motif search on cell type-specific hypo-DMRs revealed significant enrichment of Egr1 binding motif for Glu neurons and the RORα binding motif for Purkinje neuron. Their gene expression profiles across the three cell types are indicated on the right.

Neuron-type classification is supported by the epigenomic state of regulatory sequences (31). Constitutive DNA hypomethylation could indicate regions with active regulatory elements (32) and chromatin accessibility (33). Therefore, we took advantage of the ATAC-Seq and histone modification data (H3K27ac, H3K27me3, H3K4me1 and H3K4me3) of excitatory neurons (2) to address the regulatory potential of the identified cell type-specific DMRs. Most of the ATAC-Seq peaks (93%) of the excitatory neurons were consistent with our Glu-specific hypo-DMRs (Fig. 3D), indicating mCG hypomethylation may confer increased genome accessibility and interaction with DNA-binding factors. We also found that the Glu-specific hypo-DMRs are enriched for histone modifications associated with active enhancers (H3K4me1 and H3K27ac) but not promoters (H3K4me3), which is in line with previous studies that mapped brain region- and cell type-specific DMRs to distal cis-regulatory elements (1,2). These data strongly demonstrate the regulatory roles of these hypo-DMRs for Glu neuron activity.

We then performed a de novo motif search to identify potential regulatory elements that experienced hypomethylation in the three cell types. The analysis revealed highly significant enrichment of the motif 5′-GCGTGGGCGT-3′ for Glu-specific hypo-DMRs (Fig. 3E). This motif is assigned to EGR1, one of the best known immediate early genes, which plays an important function in brain epigenetic programming (34,35). Moreover, RNA-Seq data revealed the highest expression of Egr1 in Glu neurons (Fig. 3E). The other two enriched transcription factor binding motifs were assigned to TAL1_TCF3 and MEF2a (Supplementary Material, Fig. S6A and B), which are involved in neuron development. In Purkinje neurons, the most overrepresented motif was 5′-ATAANTAGGTCA-3′, which is assigned to RORα (Fig. 3E), a member of the nuclear hormone receptor family, and critical for Purkinje cell development (36). Indeed, the Rorα gene exhibited significant induction in Purkinje cells (Fig. 3E), further highlighting its regulatory role in the normal function of Purkinje neurons. Two other TFs regulating neuronal differentiation were also enriched: LIM1 and EBF1 (Supplementary Material, Fig. S6C and D). No significant TF was identified for GABA-specific hypo-DMRs, potentially due to the limited number of DMRs. Taken together, our data indicate that cell type-specific hypo-DMRs are enriched within putative enhancers and accordingly located in regions containing binding sites for specific TFs.

Cell type-specific DNA methylation changes in a mouse model of Rett syndrome

Normal brain function requires fine orchestration of excitatory and inhibitory synaptic development, and abnormal changes could cause neurodevelopmental disorders. Rett syndrome, a neurodevelopmental disorder, is caused by de novo mutations in the MECP2 gene. The loss of functional MeCP2 could lead to altered gene expression and DNA methylation and cause neuronal dysfunction. Both GABA and Glu neurons could play important roles in the disease pathogenesis of Rett syndrome; however, how the loss of MeCP2 alters the DNA methylome in different neuronal types has not been explored previously (37). We performed WGBS of the GABA and Glu neurons isolated from Mecp2 KO mice. We found that DNA methylation was stable in the absence of Mecp2 (Fig. 4A), where the mCG is ~76% and mCH is ~2% across all the INTACT samples. We also noted slightly lower mCG but higher mCH levels in Glu neurons than that in GABA neurons. Differential methylation studies identified 1173 and 1094 CG-DMRs (ranging in size from 51 bp to 3.3 kb) from these two cell types, respectively (P-value <0.05, methy.diff >0.2) (Supplementary Material, Fig. S7C and Supplementary Material, Table S8). We noted that Mecp2 KO caused inverse DNA differential methylation status locally between GABA and Glu neurons. GABA neurons with Mecp2 KO tended to contain more DNA hypomethylation events than hypermethylated ones, while greater hypermethylation was observed in Glu neurons with MeCP2 loss of function (Fig. 4B). Unsupervised hierarchical clustering analysis was conducted to confirm the separation by genotype and reproducibility of DMRs (Supplementary Material, Fig. S7D). Only few DMRs show lack of reproducibility between replicates, which could be caused by the possibility of stochastic effects across individual animals. We also found that DNA methylation differences between KO and WT are subtle compared to cell type differences (Fig. 1D; GABA versus Glu). Genomic annotation found that introns contained the highest proportion of DMRs, followed by distal intergenic regions (Supplementary Material, Fig. S7C).

Figure 4.

Cell type- and brain region-specific DNA methylome alterations associated with the loss of Mecp2. (A) Bar graph showing genome-wide mCG (left) and mCH (right) levels in GABA and Glu neurons with MeCP2 KO, respectively. The methylation levels for MeCP2 KO and WT were averaged for the two replicates. (B) Identification of DMRs between MeCP2 and WT in GABA and Glu neurons, respectively. Differential methylation studies identified 1173 and 1094 CG-DMRs (ranging in size from 51 bp to 3.3 kb) from these two cell types, respectively (P-value <0.05, methy.diff >0.2). (C and D) GO results for genes associated with hyper- and hypo-DMRs in GABA and Glu neurons with MeCP2 KO, respectively. The significantly enriched GO categories differed by cell types and direction of DMRs. (E) GO analysis for DEGs displaying significant downregulation in MeCP2 KO of Glu neurons. The result indicates that MeCP2 KO in Glu neurons suppressed synaptic-related functions. (F) Protein–protein interaction (PPI) analysis reveals that the proteins coding downregulated DEGs in Glu neurons with MeCP2 KO were significantly involved in excitatory postsynaptic potential process. (G) Number of DMRs overlapped between GABA and Glu neurons with MeCP2 loss of function. (H) Number of DMRs overlapped between cortex and striatum for GABA neurons with MeCP2 loss of function.

GO results for the genes associated with DMRs differed by neuron types and directionality of DMRs. For example, the genes associated with GABA hyper-DMRs are significantly enriched for neuronal functions (Fig. 4C), such as neuron differentiation and generation, while the genes harboring hyper-DMRs of Glu neurons are enriched in synaptic related processes (Fig. 4D), including synaptic signaling, organization and chemical transmission. Comparative studies of DMRs only identified 17 genomic regions that show differential methylation in both GABA and Glu (Fig. 4G), indicating cell type-specific changes associated with the loss of Mecp2.

Transcriptomic profiling also revealed opposite gene expression patterns between these two neuron types (Supplementary Material, Fig. S7F and Supplementary Material, Table S9). However, we found that in Glu neurons, the number of Mecp2-activated genes (genes showing downregulation in the Mecp2 KO) is much more than the number of Mecp2-repress genes (genes showing upregulation in the Mecp2 KO), in contrast to the reverse observation of transcriptomic profiling in GABA neurons. Notably, those Mecp2-activated genes in Glu neurons are significantly enriched in the synaptic associated processes (Fig. 4E), which are similar as the enriched GO categories for the Glu hyper-DMR-related genes (Fig. 4D). Additional protein–protein interaction network analysis identified overrepresentation in the regulation of excitatory postsynaptic potential (adjusted P-value = 0.0098) (Fig. 4F) for those Mecp2-activated genes. The results demonstrate that the reduced excitatory synapse activity upon loss of Mecp2 could be affected by the hypermethylation events. However, Wilcoxon tests did not detect significant differences for the distances between DEGs and DMRs and between non-DEGs and DMRs (Supplementary Material, Fig. S7G). Nevertheless, we found 12 genes that display both dysregulation and differential methylation in Glu neurons (Supplementary Material, Table S10), the majority of which (11/12) showed differential methylation at intronic regions. Of these 12 genes, six are brain-related, with critical roles in neuron, synapse, learning and memory. In addition, we plotted gene body mCG and mCH levels in wild-type mice for each category of DEGs (Supplementary Material, Fig. S7H), and we found significant differences of DNA methylation levels between DEGs and non-DEGs, as well as between upregulated and downregulated genes in Glu neurons. Those Mecp2-repress genes have greater mCG and mCH levels than Mecp2-activated genes and non-DEGs. The results suggest that gene body DNA methylation might influence the final transcriptional outcome.

Finally, we processed the WGBS data of GABA neurons from striatum (18) to determine whether DNA methylomes of GABA neurons from different brain regions have similar changes upon the loss of Mecp2. In the striatum GABA neurons, we identified a slightly lower number of DMRs (~954) than the cortex GABA neurons using the same threshold (P-value <0.05, methy.diff > 0.2) (Supplementary Material, Fig. S7A and Supplementary Material, Table S11). However, only one genomic region is overlapped between these two brain regions (Fig. 4H). We further extended the DMRs to 1000 bps and detected 6 overlapped regions, suggesting that GABA neurons exhibited distinct DNA methylation changes in different brain regions in response to the loss of Mecp2. According to limited number of shared DEGs (n = 4) between striatum and cortex, it is likely that there are distinct transcriptional outcomes caused by Mecp2 loss of function in GABA neurons from different brain regions (Supplementary Material, Fig. S7B and C). However, after Fisher’s exact test over genomic background estimates, we found significant shared effects of DEGs identified between the two brain regions (Supplementary Material, Fig. S7D and E). Given the small number of DEGs detected in each study and the heterogeneous nature of GABA neurons purified from brain regions of different genetics mouse models, further studies are required to determine the gene expression dynamics in GABA neurons with Mecp2 KO from different brain regions.

Discussion

DNA methylation plays critical roles in establishing diverse cell types (38). Different brain regions display distinct characteristics of DNA methylation patterns (1,39). Within the same brain region, methylation signatures vary among neuronal cell types (2,40). Our current work aims to elucidate the roles of DNA methylation in neuron diversity and Rett syndrome. We characterized the cell type-specific DNA methylation signatures for GABA, glutamatergic and Purkinje neurons through WGBS analysis. Overall, both CG and CH methylation levels are negatively associated with differential gene expression in Glu and Purkinje neurons but not in GABA neurons. CH methylation exhibits an obvious inverse correlation with transcriptional abundance in all three cell types. We also determined the chromatin states and regulatory potential for hypo-DMRs specific to Glu neurons. We then extended the study by simultaneously exploring DNA methylation alternations in GABA and Glu neurons upon the loss of the Mecp2 gene. Our results suggest that cell type-specific DNA methylation signatures are associated with the functional characteristics of the neuronal subtypes in the brain. Furthermore, we found that DNA methylation alterations are largely specific to neuronal cell types in a mouse model of Rett syndrome. These data suggest a critical role of DNA methylation-mediated epigenetic regulation in neuronal diversity and disease.

The relationship between DNA methylation pattern and the function of cell type-specific CG-DMRs suggests a key role for DNA methylation in shaping neuronal identity. Constitutive hypomethylated regions reflect current transcriptional state of specific neurons (2). Within cortex, Glu neurons are projection neurons that serve to transmit information both between different areas of the cortex and to other brain regions (6). In contrast, GABA neurons are interneurons that play a crucial role in regulating cortical plasticity and reduce excitation in the CNS (41). Purkinje cells are another type of inhibitory neurons located in cerebellum, with a key role in motor control and learning. They release neurotransmitter GABA, and exerts inhibitory actions on certain neurons and thereby reduces the transmission of nerve impulses (8). In this study, we found that Glu-specific hypo-DMRs are uniquely associated with genes involving in regulating excitatory synapse plasticity (Fig. 3C), reflecting persistent increases of neurotransmitter release at excitatory synapses in the cortex. However, genes regulating the processes of Purkinje cell activity and cerebellum development are specifically lost DNA methylation in Purkinje cells, indicating sustainable gene regulation in establishing Purkinje cell identity. Furthermore, localized regions of accessible chromatin and widespread hypomethylation are well-established features of cis-regulatory elements, such as enhancers (42). Here observed that the Glu-specific hypo-DMRs are enriched in open chromatin area of excitatory neurons and are consistent with the regions regulated by histone modification of active enhancers (H3K27ac and H3Kme) (Fig. 3D). However, other two cell types could maintain an inaccessible chromatin state at those regions. The loss of mCG at putative functional regions of the genome may facilitate chromatin accessibility and genome interaction with DNA-binding factors, whereas mCG hypermethylation could cause decreased genome accessibility by direct inhibition of DNA–protein interactions or induction of chromatin compaction (28,43). Our de novo motif analyses further emphasize that cell type-specific genomic regions with DNA hypomethylation could be selectively regulated by distinct transcription factors. Future studies using cell type-specific datasets from different developmental stages are necessary to further investigate the spatial temporal relationships among DNA methylation, chromatin status and gene expression.

DNA methylation is one of the crucial epigenetic mechanisms for mediating gene expression. For many years, methylation in the promoters was believed to play a key role in repressing gene transcription (44). However, some studies suggested that both mCG and mCH in gene body were also inversely correlated with differential gene expression, with mCH being the most discriminating predictor (2,28,40). Here we observed that gene body mCH levels display an inverse relationship with transcriptional abundance for all the three neuronal types (Supplementary Material, Fig. S5), and we discovered negative correlation of genic CH methylation with differential gene expression in both Glu and Purkinje neurons but not GABA (Fig. 2H). The transcriptional repression of CH methylation may be mediated by the interaction between DNA regulatory elements and DNA-binding proteins, such as MeCP2 (45,46). However, the previously described anticorrelation between gene body mCG and gene expression (2,28,32) was not observed in our analyses. Our data suggest a distinct relationship between mCG and gene expression in different neuronal types (Fig. 2G, Supplementary Material, Fig. S4). Therefore, it would be interesting to elucidate the function of intragenic mCG on gene expression in a cell type manner in future studies.

MeCP2 binds to dinucleotides with methylated cytosine and affects transcription. Kinde et al. proposed that MeCP2 regulates expression by binding to mCG and mCH in gene bodies to repress transcription (47). A recent study suggested that MeCP2 reads out mC to repress intragenic enhancer elements and control gene expression, providing a mechanism for how MeCP2 disruption in disease can lead to widespread changes in gene expression (17). Here, we found that the loss of Mecp2 in Glu neurons causes hypermethylation of genes that function in synaptic organization, signaling and transmission processes (Fig. 4D). The expression of some synaptic-related genes in Glu neurons was repressed in response to the loss of Mecp2 (Fig. 4E), supporting the notion that Mecp2 deficiency leads to reduction in excitatory synapse connectivity and plasticity, therefore contributing to a reduction in the spontaneous rate of cortical neurons (48–50). However, we did not find direct relationship of differential methylation and dysregulated gene expression based on the distance between DMRs and DEGs, presuming indirect regulatory effects of the DMRs on the gene expression upon Mecp2 loss of function, such as methylation alternations in enhancer elements regulated by MeCP2 (17). We also found that the majority of genes displaying both differential regulation and differential methylation in Glu neurons showed differential methylation at gene intronic regions, indicating that gene body introns could be the most affected epigenetic sites associated with gene dysregulation in response to Mecp2 KO (Supplementary Material, Table S10). However, the binding of MeCP2 to the genome is very broad and can also be seen at demethylated sites (51,52). Addition research is needed to clearly dissect the association of MeCP2 binding and gene expression. More recently, one study showed MeCP2-mediated differential gene expression in the GABA neurons of striatum has a strong dependence on genic mCH, consistent with mCH playing an important role in the pathogenesis of RTT (53). Here, we found that DEGs in Glu neurons have detected changes in mCG and mCH levels compared with non-DEGs (Supplementary Material, Fig. S7H). Those MeCP2-repressed genes have greater gene body methylation than the other categories of genes, suggesting the contribution of DNA methylation in the final gene expression alternations caused by MeCP2 loss of function. Further studies using additional epigenetic marks and large samples size may help dissect the mechanistic link between neuron-specific epigenetic regulation by MeCP2 and the pathophysiology of RTT.

In summary, we present detailed genome-wide DNA methylation maps for mouse GABA, Glu and Purkinje neurons and identify the distinct DNA methylation patterns among neuronal cell types, which reflect the functional characteristics of the neuronal subtypes. We further reveal the cell type- and brain region-specific DNA methylation changes in a mouse model of RTT. These results will facilitate a deeper understanding of the epigenetic mechanisms for the regulation of neuronal diversity and neurodevelopmental disorder.

Materials and Methods

Mouse

All animal care and manipulation occurred according to protocols approved by Emory University Institutional Animal Care and Use Committee (IACUC). Sun1-tagged mice (R26-CAG-LSL-Sun1-sfGFP-Myc, stock 021039), Camk2a-Cre transgenic mice (stock 003890), Gad2-IRES-Cre knock-in mice (stock 010802) and Pcp2-Cre transgenic mice (stock 004146) were obtained from the Jackson Laboratory (Bar Harbor, ME). The Sun1 mice contain a CAG promoter driving expression of coding sequences for the mouse nuclear membrane protein Sun1 (Sad1 and UNC84 domain containing 1) fused at its C-terminus to two copies of superfolder GFP (sfGFP) followed by six copies of Myc, inserted in the Gt (ROSA)26Sor locus. Cre-dependent removal of a floxed STOP cassette allows expression of this Sun1 fusion protein at the inner nuclear membrane in the targeted cell types. Their nuclei can be immunopurified with antibodies against GFP or MYC. The Camk2a-Cre transgenic mice contain the mouse calcium/calmodulin-dependent protein kinase II alpha (Camk2a) promoter driving Cre recombinase expression. The Gad2-IRES-Cre knock-in mice have Cre recombinase expression directed to GAD2 positive neurons. Pcp2-Cre transgenic mice express Cre recombinase inserted into exon 4 of the mouse Pcp2 gene. Male mice with correct genotypes were aged to 6 weeks for conducting INTACT experiments.

To obtain Mecp2 KO and WT mice, the breeding was as follows. Mecp2 knockout mice (stock 003890) were obtained from The Jackson Laboratory. LoxP sites were inserted around the exon 3 and 4 of the Mecp2 gene on the X chromosome in this mouse model. First, female Mecp2^+/− mice expressing Cre recombinase were generated by mating the female Mecp2^+/− mice with Camk2a-Cre transgenic male mice and Gad2-IRES-Cre knock-in males, respectively. Second, those Mecp2^+/−; Camk2a females were crossed with Sun1-tagged males to obtain Mecp2 deficient males (Mecp2^-/Y) and WT males expressing Sun1 and Camk2a-Cre. In parallel, Mecp2^+/−; Gad2-IRES-Cre females were crossed with Sun1-tagged males to obtain Mecp2^-/Y; Gad2-IRES-Cre; Sun1 and WT males. The resulting progeny were aged to 6 weeks for conducting INTACT experiment.

INTACT to isolate cell type-specific nuclei

The INTACT procedure for mice cortex or cerebellum tissues was followed by the description in (2) with minor modifications. Mouse cortex tissues were collected for extracting GABA and Glu neurons, whereas Purkinje neurons were isolated from cerebellum. The brain tissues were rapidly dissected in ice-cold homogenization buffer (0.25 M sucrose, 25 mM KCL, 5 mM MgCl₂, 20 mM Tricine-NaOH pH 7.8). The tissues were then dounced one at a time in 4 ml supplemented homogenization buffer with 1 mM DTT, 0.15 mM spermine, 0.5 mM spermidine and EDTA-free protease inhibitor (Roche 11836170001) 25 times using a loose pestle. A 5% IGEPAL-630 solution (Sigma-Aldrich l8896) was added to bring the homogenate to 0.3% IGEPAL-630 and further dounced with 15 strokes of the tight pestle. When purifying RNA, RNasin Plus RNase Inhibitor (Promega N2611) was added at 60 U/ml. The sample was filtered through a 40 μm strainer (Fisher Scientific 08-771-1), then mixed with 5 ml of 50% iodixanol density medium (Sigma-Aldrich D1556) to bring the final concentration to 25%. A gradient of 30 and 40% iodixanol was added to each centrifuge tube followed by adding the sample. The samples were centrifuged at 10 000g for 18 min in Beckman ultracentrifuge at 4°C. After centrifuging, nuclei were collected at the 30–40% interface and pre-cleared by incubating with 20 μl of Protein G Dynabeads (Life Technologies 10003D) for 10 min. After removing the beads with a magnet, the mixture was diluted with wash buffer (homogenization buffer plus 0.4% IGEPAL-630) and incubated with 10 μl of 0.2 mg/ml rabbit monoclonal anti-GFP antibody (Life Technologies G10362) for 30 min. Then, 60 μl of Dynabeads were added to the sample mixture followed by incubating for an additional 20 min. To increase yield, the bead-nuclei mixture was placed on a magnet for 30 s to 1 minute, completely resuspended by inversion and placed back on the magnet. This was repeated 5–7 times. Bead-bound nuclei were passed through a 20 μm strainer (Sysmex 04-0042-2315) and then washed with 2 × 10 ml, 1 × 5 ml and 1 × 1 ml wash buffer with 5 min rotation for each wash at 4°C. Bead-bound nuclei were stored at −80°C until DNA/RNA extraction for NGS library preparation.

DNA and RNA Isolation

Bead-bound nuclei were resuspended in 600 μl lysis buffer (100 mM Tris-HCl, pH 8.5, 5 mM EDTA, 0.2% SDS, 200 mM NaCl) for DNA purification, 25 μl Proteinase K (Thermo Fisher EO0491) treatment at 55°C overnight using an end-to-end rotator. The second day, the samples were incubated at 85°C for 45 min to inactive Proteinase K followed by the treatment of 5 μl RNase A for 1 h at room temperature. The entire solution was then transferred to a pre-spun ‘Phase Lock Gel (PLG) 2 ml Heavy’ tube. Equal volume of phenol:chloroform:isoamyl alcohol (25:24:1 saturated with 10 mM Tris, pH 8.0, 1 mM EDTA) (Sigma-Aldrich, P-3803) was added to samples, mixed completely and centrifuged for 10 min at 12 000 rpm. The aqueous layer solution was transferred into a new Eppendorf tube and precipitated with 600 μl isopropanol. The pellet was washed with 75% ethanol, air-dried and eluted with Nuclease-Free Water (Ambion AM9938). Bead-bound nuclei were directly resuspended in TRIzol Reagent (Thermo Fisher 15596026) for RNA extraction following the standard protocol.

WGBS library preparation, sequencing and data analysis

The EZ DNA Methylation Gold kit (Zymo Research D5005) was used for sodium bisulfite conversion prior to library preparation. All WGBS libraries were prepared from 100 ng sonicated nuclei DNA. The Accel-NGS Methyl-Seq (Swift BioSciences 30024) kit was used for library preparation according to the manufacturer’s protocols. Post-library QC was performed with BioAnalyzer DNA 1000 chips (Agilent 5067-1504) and the Qubit dsDNA High Sensitivity fluorometric assay (Invitrogen Q32854). Paired end sequencing (2 × 150) was performed on an Illumina Hiseq 4000 instrument. A PhiX library was spiked in at 30% for each sequencing library.

For all WGBS libraries, 15 bps were trimmed off both end of R1 and R2 to remove bases derived from the adaptor tag introduced in the library preparation procedure using Cutadapt v2.10 (54). The processed reads were aligned to the mouse genome (mm9), deduplicated, examined for coverage and extracted to CG and CH count matrix using Bismark v0.19.0 (55). Briefly, reference sequences were converted to map both cytosine and thymine. The mapping of WGBS libraries was performed with the bismark module followed by duplication removal using deduplicate_bismark. Methylation calling was processed using a module called Methylation Extractor, which classified methylation sites as CG and CH (CHH and CHG contexts). The modules of bismark2bedGraph and coverage2cytosine were used to extract non-CpG sites and estimate their methylation levels. Genome-wide methylation levels for mCG and mCH contexts were estimated as the weighted methylation level of all mCG and mCH sites, respectively. For instance, the average methylation level of each CpG site was weighted by the coverage at each site.

DMR calling, annotation and motif analysis

DMRs were called by utilizing the package DSS (Dispersion Shrinkage for Sequencing) (56). First, the DMLtest function was used to perform statistical tests for all CpG sites with smoothing. Second, the qualified CpG sites were subjected to calling differential methylated loci (DML) using the callDML function. The difference in mean methylation level between the Mecp2 KO and WT was compared with the global methylation difference using a Wald test at each CpG site. DML are those CpG sites with FDR-corrected P-value <0.05. Finally, the callDMR function was used to call DMRs with the following settings: each DMR has minimum length of 50 bp and has at least four DML. The DMRs between cell types were then selected as an absolute difference of methylation level greater than 0.2 relative to the global methylation difference in the comparison, and a detected P-value <1e−5. For the DMRs between Mecp2 KO and WT, an absolute difference of methylation level over 0.2 and P-value <0.05 were selected. DMRs were assigned to a given genomic feature using package annotatr (57) in R. The enrichment of DMRs was calculated by the fold change between the percentage of DMRs located on different genetic features and expected genomic length percentages. The expected values were calculated as the genomic space occupied by the genomic features associated with DMRs by chance. For intergenic features, associated areas are summed up to obtain the length of genomic space. For other genomic features, the length of each feature was summed up based on the obtained annotation information.

Additionally, we used the getMeth function from the package BSseq v1.28.0 (58) in R to calculate the average methylation levels for these obtained DMRs across all the samples. We then drew a heatmap to demonstrate the methylation level change across different sample groups using the pheatmap function from package pheatmap v1.0.10 in R, where the columns correspond to the samples and rows correspond to DMRs. GREAT (59) was used to perform GO analyses for the cell type-specific DMRs. Hypergeometric Optimization of Motif EnRichment (HOMER) (60) was used to perform known and de novo motif searches for DMRs relative to background loci. The findMotifsGenome.pl script was used with corrections for fragment size and normalization CpG content (-chopify –cpg –size given). g:Profiler web server (61) was employed to perform GO analyses for genes associated with DMRs for MeCP2 KO versus WT.

RNA-Seq and data analysis

Biological replicates were subjected to RNA-Seq. One μg of total RNA was used for RNA-Seq library construction following the instructions of the Illumina mRNA sample prep kit. Briefly, total RNA was depleted of ribosomal RNAs, subjected to 5 min of heat fragmentation and converted to strand-specific cDNA libraries using the TruSeq Total RNA library prep kit with Ribo Zero depletion (Illumina 20040529). Libraries were submitted for 150 paired-end sequencing on the Illumina HiSeq 2500 platform. RNA-Seq reads were aligned to mm9 mouse genome using TopHat v2.1.1 (62). The number of reads aligned to each gene was tabulated using FeatureCount v1.5.3 (63). All reads mapped to gene bodies were used for differential gene expression analysis by using R packages edgeR v3.32.1 and DEseq2 v1.30.1 as described in (64). Genes displaying low expression due to a substantially low number of mapped reads and whose edgeR CPM values satisfied the condition ‘rowSums(cpm(data_y)) < 2’ were filtered out from differential gene expression analyses. For each comparison, the results from both edgeR and DESeq2 analyses were merged into a final nonredundant and Padj-controlled list of genes to avoid method-specific biases. The mean fold change and the mean Padj generated from both methods were used for generating plots and heatmaps. For comparisons between different neuronal types, DEGs were restricted to those with |log₂ fold change| > 2 and Padj > 0.01. For Mecp2 KO versus WT comparisons, genes with |log₂ fold change| > 0 and Padj > 0.1 were defined as DEGs. g:Profiler web server was employed to perform GO analyses (61).

Metagene analysis

DNA methylation profiles around gene bodies were calculated and plotted by deepTools2 (65). First, we used WiggleTools (66) to average the DNA methylation percentages of two replicates for each neuronal type. Then, the computeMatrix module from deepTools2 with scale-regions parameter was used to calculate the average scores for a specific input gene set between upstream (−5 kb or −10 kb) TSS and TES downstream (+5 kb or +10 kb) regions. Next, based on the calculated scores, the plotProfile module was used to plot out the DNA methylation profiles around the given gene bodies.

Correlation between gene body DNA methylation and gene expression

We first used RPKM (reads per kilobase million) to normalize gene expression for sequencing depth and averaged the RPKMs of two replicates for each neuronal type. Next, we limited the gene list to those with RPKM less than or equal than 100. DNA methylation located on those regions was extracted and averaged to obtain the gene body DNA methylation percentages. The methylation percentages were further averaged for two replicates. Based on all the RefSeq genes of mm9 genome, we calculated the Spearman correlation between DNA methylation levels with differential gene expression. DEGs for the three pairwise comparisons of different neuronal types were filtered by |log₂ fold change| > 1 and Padj < 0.01. Non-DEGs were also filtered out as a control group using |log₂ fold change| < 0.1 with Padj > 0.05. For Mecp2 KO versus WT comparisons of Glu and GABA neurons, DEGs were defined as |log₂ fold change| > 0 with Padj < 0.1. We also extracted non-DEGs using Padj > 0.1 as a control group. Then, we took the DNA methylation percentage as a function of gene expression on selected gene lists and used Loess regression with span as 0.5 to fit the data. Ninety percent confidence intervals were computed for regression coefficients.

Processing of ATAC-Seq and CHIP-Seq data of excitatory neuron

The ATAC-Seq and CHIP-Seq data of excitatory neurons were downloaded through GEO accession GSE63137 from NCBI. The excitatory neurons were isolated and purified by INTACT procedure from male mice aged 8–11 weeks. We followed the data processing pipeline as described in (2). Briefly, adapter sequences were trimmed from ATAC-seq reads using cutadapt v2.10. The high-quality reads were mapped to mouse genome (mm9) using Bowtie v1.3.0. Peaks were called by HOMER using sub-nucleosomal (<100 bp) fragments, and overlapping peaks were merged. The identified peaks from replicates were merged to generate a peak set for excitatory neurons. CHIP-Seq data sets include four histone modification markers of H3K4me1, H3K4me3, H3K27ac and H3K27me3. The CHIP-Seq and input reads were aligned using Bowtie and duplicated reads were removed by Samtools v1.9. The package of SICER v1.1 was used to identify peaks for each histone modification. SICER parameters with FDR = 0.001 were as follows: redundancy threshold = 1; fragment size = 150; W = 200, G = 200 for H3K4me1, H3K4me3 and H3K27ac; and W = 200, G = 1000 for H3K27me3.

Processing of DNA methylation and RNA-Seq data of GABA neurons from striatum

The MethylC-Seq and RNA-Seq data of MeCP2 KO and WT data were obtained from GEO accession numbers GSE124009 and GSE123941. Data analysis for MethylC-Seq was followed as described in (18). Adaptor sequences were trimmed from sequencing reads using cutadapt v2.10 (54) with parameters -f fastq -q 20 m 50 -a AGATCGGAAGAGCACACGTCTGAAC -A AGATCGGAAGAGCGTCGTGTAGGGA. The sequences were further trimmed 10 bp from both 5′- and 3′-ends of both R1 and R2 reads. Trimmed R1 and R2 reads were mapped to the mm9 reference genome as single-end reads using Bismark v0.19.0 with parameter --bowtie2. --pbat option was used for mapping R1 reads. Mapped reads were sorted using Samtools 1.9 followed by removing duplicate reads using MarkDuplicates module in Picard 1.141 (67). Reads were further filtered by MAPQ >20 using Samtools. DMRs between MeCP2 and KO were called using DSS (56) package as we did for MeCP2 KO versus WT in GABA and Glu neurons from cortex.

The RNA-Seq reads were trimmed for adaptor sequences using cutadapt v2.10 (54). The clean reads were mapped to mouse reference genome (mm9) using STAR aligner with default parameters. The number of reads aligned within the gene body (from TSS to TES) of each gene was tabulated using FeatureCount v1.5.3 (63). DEG analyses using the read counts were performed using package DEseq2 v1.30.1 in R. Genes with total read counts less than 10 were filtered out from analysis.

Data Availability

Accession numbers for all the datasets applied in this paper are summarized in Supplementary Material, Table S1. WGBS and RNA-Seq data are available from NCBI’s Gene Expression Omnibus (GSE191211). The accession number for all the datasets reported in this paper is GSE191211. The ATAC-Seq and CHIP-Seq data of excitatory neuron used in this study were sourced from (2) with accession number of GSE63137 in NCBI. The DNA methylation and RNA-Seq data of MeCP2 KO and WT for GABA neuron from striatum (18) were obtained from GEO accession numbers GSE124009 and GSE123941.

Funding

This work was supported, in part, by the National Institutes of Health (MH116441, HD104458 and NS111602 to P.J. and E.G.A.).

Conflict of Interest statement. The authors declare no competing interests.

Authors’ contributions

P.J. conceived the study. Y.J. and P.J. wrote the manuscript. H.E.K. and Z.W. performed mouse breeding and INTACT procedure. Y.L., R.L. and Y.J. established WGBS library preparation for this study. Y.J. and K.S. performed most of the data analysis. W.M. and H.W. contributed to data analysis. All the authors read and commented on the manuscript. The author(s) read and approved the final manuscript.