Sex differences in brain cell-type specific chromatin accessibility in schizophrenia

Our understanding of the sex-specific role of the non-coding genome in serious mental illness remains largely incomplete. To address this gap, we explored sex differences in 1,393 chromatin accessibility profiles, derived from neuronal and non-neuronal nuclei of two distinct cortical regions from 234 cases with serious mental illness and 235 controls. We identified sex-specific enhancer-promoter interactions and showed that they regulate genes involved in X-chromosome inactivation (XCI). Examining chromosomal conformation allowed us to identify sex-specific cis- and trans-regulatory domains (CRDs and TRDs). Co-localization of sex-specific TRDs with schizophrenia common risk variants pinpointed male-specific regulatory regions controlling a number of metabolic pathways. Additionally, enhancers from female-specific TRDs were found to regulate two genes known to escape XCI, (XIST and JPX), underlying the importance of TRDs in deciphering sex differences in schizophrenia. Overall, these findings provide extensive characterization of sex differences in the brain epigenome and disease-associated regulomes.


Introduction
In schizophrenia (SCZ), multiple clinical features, including age of onset, symptomatology, prevalence, the course of illness, and response to treatment often differ by sex [1][2][3] .Men are slightly more likely to develop SCZ than women and women tend to be older when they rst experience symptoms.These differences have been ascribed to different gene contexts within the sex chromosomes, roles of hormones, cultural and environmental differences, and differences in behavior 4 .However, the molecular mechanisms and underlying biology remain largely unclear.A previous large-scale transcriptome study by the Genotype-Tissue Expression (GTEx) project 5 characterized widespread differences in gene expression levels and identi ed 13,294 sex-differentially expressed genes across 44 human tissues 6 .
Studies focusing speci cally on the brain have identi ed differences in brain region-speci city 7 , developmental stages 8 , as well as the interaction between sex and complex traits 9,10 .In the context of disease, investigations into sex-speci c effects on SCZ have primarily been limited to studies of the transcriptome 9,11,12 .Despite these valuable resources, the representation of sex-speci c regulatory regions, particularly enhancers and promoters, is lacking and, as such, the interaction effects of sex and disease are under-characterized.While evidence from studies of both humans 13 and mice 14 indicate that sex-biased gene expression may arise from differences in chromatin accessibility, neither of these studies examined the brain.
In this study, we focused on sex-and disease-speci c alterations in chromatin accessibility and interrogated the higher-order chromatin structure spanned by cisand trans-regulatory domains (CRDs and TRDs).We generated and analyzed ATAC-seq data from 469 individuals (172 females, 297 males), in two broad cell types (neurons and non-neurons).We used this dataset to quantify and characterize sex differences in chromatin accessibility (sex-speci c chromatin accessibility) and to identify sex-speci c open chromatin regions (OCRs) associated with the regulation of genes involved in X-chromosome inactivation (XCI).Integrating these functional epigenomic data with common genetic risk variation allowed us to identify genetic effects on the epigenome that vary between sexes and colocalize with complex trait associations.Finally, by integrating disease-speci c chromatin accessibility, we identi ed SCZ-associated OCRs within sex-speci c TRDs, particularly those on the X chromosome involved in regulating XCI in the human brain.By combining our ndings with functional gene sets and transcription factor binding annotation, we characterized both cell-type-speci c and universal drivers and mechanisms of sex differences in the brain epigenome.

Sex effects on chromatin accessibility
We previously studied cell-type speci c chromatin structure and genetic regulation of the brain regulome as part of the CommonMind Consortium (CMC) 15,16 .Here, we present an extensive characterization of sex differences in the human brain epigenome from 469 individuals (172 females, 297 males), consisting of 157 SCZ cases, 77 bipolar disorder (BD) cases, and 235 unaffected controls from the CMC dataset (Fig. 1a and Table S1).We performed the assay for transposase-accessible chromatin followed by sequencing (ATAC-seq) to quantify and characterize sex differences in chromatin accessibility in neuronal (NeuN+) and non-neuronal (NeuN-) nuclei isolated by uorescence-activated nuclear sorting (FANS) from two brain regions, i.e. the anterior cingulate cortex (ACC) and dorsolateral prefrontal cortex (DLPFC).After a comprehensive quality control analysis of ATAC-seq libraries (Methods), the nal dataset consisted of a total of 1,393 samples comprising over 54.8 billion unique reads (39.4 million reads per library) (Table S1).Due to the large chromatin accessibility differences between the two cell types (Fig. 1b), we separated neuronal and non-neuronal samples for downstream analysis.A total of 391,420 and 260,431 neuronal and non-neuronal open chromatin regions (OCRs) were then identi ed, respectively.
We next quanti ed sex-speci c chromatin accessibility in each cell type.For each cell type, we rst t a linear mixed model using dream 17,18 to account for covariates, including diagnosis, cell type, brain region, brain bank, as well as other technical and biological factors (Methods).Consequently, we discovered a total of 8,984 and 5,327 differentially accessible OCRs between males and females (sex-speci c OCRs) in neuronal and non-neuronal cells, respectively (FDR < 0.05) (Fig. 1c and Table S2).We identi ed more female-speci c OCRs than male-speci c OCRs in both neuronal and non-neuronal cells, the majority of which mapped to the sex chromosomes (Fig. 1c, d).In neurons, 6,072 differentially accessible OCRs (68%) were X-linked, 132 (1%) were Y-linked, and 2,780 (31%) were autosomal; while in non-neurons, 4,305 OCRs (81%) were X-linked, 84 (1%) were Y-linked, and 938 (18%) were autosomal.As expected, the analysis of sex differences also revealed that all Y-linked regulatory elements with increased chromatin accessibility were speci c to males, while nearly all X-linked regulatory elements with increased chromatin accessibility were speci c to females, potentially due to escape from X-chromosome inactivation (XCI) 19 (Table S4).By validating our differential chromatin accessibility analysis against another epigenome study of postmortem brain samples from DLPFC region 20 , we achieved a high concordance (Spearman correlation coe cients (ρ) of 0.877 and 0.869, p-value < 2.2e-16 for both) in neuronal and non-neuronal samples (Fig. 1e).
Following the observed enrichment of female-speci c OCRs on the X chromosome, we next investigated sex-speci c changes in chromatin accessibility patterns associated with XCI genes.Out of 277 genes known to escape XCI 6,19,[21][22][23][24][25] , a substantial number -205 genes in neurons and 206 genes in nonneurons (181 genes shared between the two cell types) -appear to be regulated by sex-speci c OCRs (Table S3).Notably, XIST, a non-coding RNA essential for the initiation of XCI 26 , was found to be upregulated by female-speci c OCRs in neurons and non-neurons (Fig. 1f).FIRRE, an X-linked lncRNA that escapes XCI 27 , also demonstrated up-regulation in both cell types (Fig. 1f).This complex regulation of XCI highlights the sex-speci c interplay between chromatin accessibility and gene expression.

Functions of sex-speci c chromatin accessibility
To gain insights into the biological pathways regulated by genes associated with sex-speci c OCRs, we performed gene set enrichment analysis (GSEA) in each cell type.We identi ed enrichment in a wide range of biological pathways, including those involved in response to hormones, reproductive processes, metabolism, immune responses, histone modi cation, and others (Table S4).Speci cally, we discovered that genes associated with sex-speci c OCRs were involved in synaptic signaling, neuron projection development, and regulation of axonogenesis, which might contribute to the functional sex differences in brain activity (Fig. 2a).Although signi cantly associated, sex determination, dosage compensation, and regulation of dosage compensation by inactivation of the X chromosome (Table S4) were not ranked among the most enriched pathways.However, we identi ed X-linked disease terms that were top ranked and signi cantly associated with female-speci c OCRs (Fig. 2b, Table S4).We also identi ed heparan sulfate (Hs) glycosaminoglycan (GAG) degradation and metabolism in female-speci c OCRs, a previously reported sex difference in the hippocampus known to affect neuronal plasticity and brain development 28 .A set of pathways involved in histone modi cations, regulated by female-speci c OCRs, has been reported to result in sex-speci c gene expression (Table S4).Trimethylation of Histone H3 at lysine-36 (H3K36me3), driven by female-speci c OCRs, is also observed to be less abundant on the male X chromosome 29 and is known to induce gene silencing and XCI 30 .Taken together, these results suggest that sex-speci c chromatin accessibility plays a critical role in a range of biological pathways, including some that have not previously been linked to sex.
To further explore the importance of sex-speci c regulatory mechanisms and their role in disease, we quanti ed the overlap of sex-speci c OCRs with disease risk variants using the LD-score partitioned heritability (LDsc) approach.We note that LDsc does not support analysis on sex chromosomes; therefore, we conducted partitioned heritability analysis using only non-sex chromosomal OCRs.Enrichment of SCZ risk variants was only observed in male-speci c neuronal OCRs (Fig. 2c, Table S5).Furthermore, we observed that ADHD, cognitive performance, and reaction time variants were signi cantly colocalized with male-speci c OCRs in neurons.This nding highlights the critical role of sex-speci c regulatory mechanisms of neurons in psychiatric disorders and brain functions, speci cally in males.
Because transcription factors (TFs) contribute to sex regulating network structures in various human tissues 31 , we hypothesized that TFs can affect patterns of differential chromatin accessibility by differential binding of TF motifs at sex-speci c OCRs.We employed TF motif enrichment analysis 32 in sex-speci c OCRs to identify the enrichment of known motifs and to discover novel motifs (Methods).We identi ed a total of 54 TF motifs enriched for male-and female-speci c OCRs (Fig. 2d, Table S6), including 39 TFs enriched for female-speci c OCRs and 11 TFs enriched for male-speci c OCRs in neurons, as well as 7 TFs enriched for female-speci c OCRs and 4 TFs enriched for male-speci c OCRs in non-neurons.Of these, 2 neuronal TFs (RFX and RFX2) were shared between males and females, while 5 female-speci c TFs (CTCF, CTCFL, MAZ, NF1, and ZFP281) were shared between neurons and nonneurons.We also identi ed 4 TFs, ELF1, ELK1, KLF5, and YY1, which have been reported to exhibit consistent male-speci c enrichment in human tissues 6 .Among these, ELF1, KLF5, and YY1 were enriched in the brain.The impact of sex on the majority of the remaining TFs has not been fully characterized.Overall, these results suggest that a wide range of TFs, some evidently unrelated to sex hormones, play crucial roles in driving sex-speci c regulatory programs in the human brain.

Sex-speci c enhancer-promoter regulatory landscape
We next sought to determine the concordance between sex-speci c transcriptional activity and genomewide chromatin accessibility.First, to reconstruct functional enhancer-promoter (E-P) pairs responsible for the regulation of sex-speci c genes, we applied the activity-by-contact (ABC) 34 model, combining our open chromatin data, histone acetylation data 35 and 3D interaction frequency maps 36 .Using separate models for each sex and cell type, we identi ed 42,119 − 42,229 E-P links (E-P ABC ) in neurons and 37,772 − 38,261 E-P ABC in non-neurons (Table S7).On average, 64% of E-P ABC were shared between sex in neurons, with a moderate correlation of ABC scores (ρ = 0.52; Fig. 3a left panel), whereas 82% of E-P ABC were shared in non-neurons, with a high correlation score (ρ = 0.79; Fig. 3a right panel).Additionally, when examining these links on autosomal chromosomes versus the X chromosome, we found a signi cant drop in the correlation between males and females for X-linked genes (Extended Data Fig. 1).This suggests that X-linked gene activity is controlled by sex-speci c enhancer activity, aligning with our initial expectations.
Although a majority of enhancers were predicted to interact with a single gene, 38% were linked to two or more genes (Extended Data Fig. 2a).We found that over 70% of the regulated genes were linked to multiple OCRs (OCR ABC ) (Extended Data Fig. 2b).Of these, only 22% were linked to the nearest gene (Extended Data Fig. 2c).This nding is consistent with our previous analyses [36][37][38] , demonstrating the limitations of proximity-based annotation of regulatory elements.Still, the frequency of E-P ABC links sharply decreased with distance to the transcription start site (TSS), with 75% of interactions identi ed within 100 kb, upstream and downstream, of the TSS (Extended Data Fig. 2d).
To evaluate the sex speci city of the observed E-P ABC interactions, we compared sex-speci c, but not shared, E-P ABC pairs identi ed in neurons (9,447 female and 9,332 male pairs) and non-neurons (3,459 female and 3,998 male pairs).On average, 96% of OCR ABC were autosomal and 4% were X-linked in neurons; 90% of OCR ABC were autosomal and 10% were X-linked in non-neurons (Fig. 3b, Table S7).
Comparative analysis of sex-speci c changes between autosomal and X-linked OCR ABC revealed substantially higher effect sizes for X-linked OCR ABC in both neurons and non-neurons, respectively (Fig. 3c), further con rming that sex-speci c effects are predominantly exhibited on the X chromosome on the enhancer level.When compared to adult human brain chromatin states 35 , X-linked OCR ABC regions showed a relative enrichment in polycomb (H3K27me3) repressed chromatin states as opposed to autosomal OCR ABC regions (Fig. 3d, Extended Data Fig. 2e).This observation suggests that X-linked OCR ABC play a role in sex-speci c silencing of gene expression.
Moreover, our analysis revealed sex-speci c distal regulatory landscapes of XCI in the human brain.We identi ed 313 E-P ABC regulating 126 genes known to escape XCI (Table S7).For instance, FIRRE is an Xlinked lncRNA that escapes XCI and is involved in chromosome topological organization 13,27,39 .FIRRE harbors a series of putative intronic enhancers demonstrating female-speci c chromatin accessibility through differentially regulated E-P ABC (Fig. 3e.Intriguingly, our ndings indicate that two enhancers within intron 2 of FIRRE are active in male-speci c non-neuronal cells, whereas six enhancers spanning introns 2 to 12 are active only in female-speci c cells.Notably, there was no observed activity of FIRRE enhancers in male-speci c neurons.NHSL2, a reported XCI gene 6 , is another illustrative example showing that cell-type speci c enhancers can regulate sex-speci c activity via XCI (Fig. 3e).Taken together, our results indicate that FIRRE plays a role in maintaining enrichment of a repressive histone mark (H3K27me3) to mediate gene silencing on the X chromosome.Differential accessibility of putative FIRRE's enhancers between neurons and non-neurons further highlights its contribution to cell-typespeci c patterns.

Sex-speci c CRDs and TRDs associated with schizophrenia
Cis-regulatory domains (CRDs) are physically interacting regulatory elements that contain multiple locally correlated OCRs and play a crucial role in gene regulation contributing to SCZ risk.Using an in-house analytical framework (Fig. 4a) 15,40 , we showed that 37% of neuronal OCRs were assembled into 6,706 CRDs and 33% of non-neuronal OCRs were assembled into 4,612 CRDs (Extended Data Fig. 3a).Next, we identi ed 155 neuronal and 48 non-neuronal CRDs that exhibit signi cant differential accessibility in their mean between males and females (sex-speci c CRDs) (Fig. 4b, Table S8).We observed sex-speci c OCRs are more likely to be found inside, rather than outside, sex-speci c CRDs in both neurons (odds ratio (OR) = 4.39, p-value < 2.2e-16) and non-neurons (OR = 7.63, p-value < 2.2e-16) (Extended Data Fig. 3a).We previously identi ed CRDs that exhibit signi cantly differential accessibility between individuals with SCZ and controls (SCZ CRDs) 15 .Because OCRs located within neuronal CRDs exhibit signi cant heritability for SCZ (Extended Data Fig. 3b), we compared the relationship among sex-speci c OCRs, SCZ OCRs, sexspeci c CRDs and SCZ CRDs in neurons.We found a signi cant correlation between sex-speci c OCRs and SCZ OCRs in neurons, evident in a genome-wide analysis (ρ = 0.59, p-value < 2.2e-16, Extended Data Fig. 4a) and also when focusing on a subset of OCRs inside sex-speci c CRDs (ρ = 0.58, p-value < 2.2e-16, Fig. 4c, Extended Data Fig. 4b).The chromatin accessibility of OCRs from male-speci c CRDs showed a signi cant correlation with a SCZ case/control comparison (ρ = 0.52, p-value = 6.2e-5,Fig. 4d, Extended Data Fig. 3c).Conversely, a markedly lower correlation of OCRs from female-speci c CRDs (ρ = 0.18, pvalue = 0.13) may indicate a different or less direct involvement of these regions with SCZ (Extended Data Fig. 4c, Extended Data Fig. 3c).
We next examined whether the interactions between sex-speci c CRDs could identify trans-regulatory domains (TRDs).Using a correlation matrix of expression of sex-speci c CRDs across all neuronal samples, we applied hierarchical clustering and Gamma statistics 41 to identify a total of 14 neuronal sexspeci c TRDs (Fig. 5a).While most TRDs exhibit a mix of female-and male-speci c CRDs, there is one particular TRD (TRD1) that stands out with a predominant role in the regulation of male-speci c CRDs in neurons.TRD1 also showed the highest proportion of male-speci c OCRs compared to other TRDs (Fig. 5b).Conversely, TRD2, TRD4, TRD6, and TRD13 displayed a higher proportion of female-speci c OCRs (Fig. 5a and b).For TRD1, TRD2, and TRD13, the majority of sex-speci c OCRs showed higher accessibility in SCZ cases compared to controls, whereas for TRD4 and TRD6, the majority of sex-speci c OCRs showed lower accessibility (Fig. 5c-d).While we observed a moderate correlation of male-speci cwith SCZ-changes (ρ = 0.37; p-value = 0.012) in TRD1, we did not observe a signi cant correlation of female-speci c changes with SCZ-changes in any TRD (Extended Data Fig. 5).Thus, this measurement implies sex-speci c chromatin regions are likely to be more accessible in SCZ compared to controls.This suggests a complex relationship between chromatin accessibility and the expression of genes associated with SCZ in females, potentially in uenced by other regulatory mechanisms or factors [42][43][44] .
Functional pathway analysis in male-speci c OCRs that are upregulated in SCZ and TRD1 identi ed biological processes related to ketone body metabolism and acid thiol ligase activity (Fig. 5e).These ndings are further supported by a recent study elucidating sex-dependent modulation of metabolic hormone 45 , indicating that SCZ may involve sex-speci c epigenetic regulation in males, particularly affecting metabolic and enzymatic pathways.Functional pathway analysis in female-speci c OCRs that are upregulated in SCZ and TRD13 identi ed 36 signi cant associations with biological processes, including regulation of dosage compensation by inactivation of the X chromosome and negative regulation of neural precursor cell proliferation (Fig. 5e; Table S9).This highlights the complexity of female-speci c epigenetic regulation in SCZ, particularly emphasizing the role of XCI in neurons 46 .To evaluate the sex speci city of E-P ABC interactions in TRDs, we identi ed 34 male-speci c OCR ABC and 95 female-speci c OCR ABC associated with dysregulation of SCZ in neurons (Table S8).For instance, a malespeci c OCR ABC (Peak_83969) in TRD1 can regulate four protein-coding genes via E-P ABC interactions, including TMEM132B, AACS, BRI3BP, and DHX37 (Fig. 5f).Similarly, a female-speci c OCR ABC (Peak_391105) in TRD13 can regulate two lncRNA genes, XIST and JPX (Fig. 5f), both of which are known XCI genes 21 , suggesting that sex-speci c enhancers can regulate activity through XCI within SCZ associated TRDs.This highlights the potential of TRDs as key to understanding the regulatory mechanisms driving sex differences in SCZ.

Discussion
We have created the largest resource to date, consisting of 1,393 samples derived from neurons and nonneurons, to examine effects of biological sex on chromatin accessibility in the human brain.We rst determined differentially accessible regions between males and females (sex-speci c OCRs) in both cell types.We found high concordance with sex-speci c OCRs from another epigenome study on postmortem brains, con rming the robustness of our ndings.As expected, our analysis found the strongest sex bias from regions on the X-chromosome, while the majority of male-speci c OCRs regulated autosomal genes, suggesting the impact of sex on regulatory programs throughout the genome.Male-but not femalespeci c neuronal OCRs were signi cantly co-localized with common SCZ risk variants and other brainrelated traits.Our analysis was restricted to autosomal OCRs due to the inherent limitations of the LD score model.This restriction potentially narrowed our exploration, particularly since sex chromosomes, as documented in the latest SCZ GWAS 47 , are also known to host SCZ loci.
Sex-speci c OCRs are involved in a highly diverse set of biological functions.In addition to expected ndings, such as dosage compensation by XCI and response to sex hormones, we also observed enrichment for synaptic organization.This is consistent with the role of synapses observed in sexspeci c transcriptome studies of SCZ 9 and depression 48 , and is indicative of their contribution to sexspeci c differences in brain development and function.The enrichment of the epigenetic modi cation H3K36me3 was driven by female-speci c OCRs, which is consistent with an established role for this mark in inducing gene silencing involved in XCI 30 .It is important to note that sex-speci c H3K36me3 methylation has previously been reported to balance the transcriptional output of the male X chromosome and the fourth chromosome in Drosophila 29 .H3K27me3 has also been reported to result in sex-biased gene expression in mammalian placenta 49 and liver 50 .Conversely, H3K27me3 did not show enrichment within sex-speci c OCRs in this study and further investigation integrating histone marks may provide additional mechanistic insights into sex-biased patterns.The large difference in sex-speci c chromatin accessibility and enrichment of sex-speci c TF motifs between neurons and non-neurons suggests a high degree of cell-type speci city in sex-biased regulatory mechanisms in the human brain.
To date, studies of sex effects on psychiatric and neurologic diseases have largely been limited to the brain transcriptome 6,9,12,48 .In building on this knowledge, our study maps the sex-speci c E-P landscape and connects sex-speci c OCRs in distal intergenic areas to their target genes.Interestingly, we show that the repressive histone marker, H3K27me3, dependent on the activity of Polycomb repressive complex 51 , exhibited higher levels of X-linked E-P ABC links compared to autosomal E-P ABC links.Our approach enabled us to uncover distal regulatory E-P connections that play a role in governing XCI.To explore the in uence of sex on higher-order chromatin structure, we identi ed cis-regulatory domains (CRDs) and trans-regulatory domains (TRDs).Notably, one male-speci c TRD (TRD1) showed a signi cant correlation to SCZ-related changes at the chromatin accessibility level, a pattern not mirrored in the female-speci c TRDs (TRD2, TRD4, TRD6, and TRD13).This indicates complex regulatory programs underlying sex differences in SCZ.Beyond gene expression, factors like sex-biased genetic regulation 6 , protein expression 52 , and polygenic risk 43 may also contribute to phenotypic sex differences in the context of disease.Mapping sex-speci c OCRs to E-P ABC links allowed us to pinpoint sex-speci c enhancers in TRDs.Intriguingly, among these, certain enhancers were identi ed that appear to regulate sex-speci c activity through XCI in SCZ.
Cumulatively, this study produced maps of sex-speci c alterations in the epigenome of both neurons and non-neurons, focusing on changes in chromatin accessibility as well as other higher-order chromosomal conformations.These maps provide valuable insights into the mechanisms underlying sex-speci c aspects of SCZ, particularly in relation to XCI.Finally, we provide this unique resource of human brain sexspeci c OCRs, E-P links, CRDs and TRDs, enabling the scienti c community to further study sexdifferentiated mechanisms underlying brain-related disorders.

Methods Study Population
To systematically examine the sex-speci c epigenome, we utilized our recently completed atlas of open chromatin accessibility generated using postmortem tissue collected by the CommonMind Consortium (CMC) 53 , as previously described 15,16 .This cohort of SCZ, BD and unaffected controls was assembled after applying stringent inclusion/exclusion criteria.All subjects had to meet the appropriate diagnostic DSM-IV criteria, as determined after review of medical records, direct clinical assessments, and interviews with family members or care providers.Frozen brain tissue derived from ACC (anterior cingulate cortex/Brodmann area 10) and DLPFC (prefrontal cortex/Brodmann area 9 and 46) were obtained from three separate brain banks, including the Icahn School of Medicine at Mount Sinai (MSSM), University of Pittsburgh (PITT) and NIMH Human Brain Collection Core (HBCC).The complete demographic and clinical information of the present study population is provided at the Synapse platform (syn52264219).

ATAC-Seq Data Generation and Processing
The assay for transposase-accessible chromatin with sequencing (ATAC-seq) was performed using an established protocol 54 with minor modi cations.Throughout the library generation process, randomization was applied with respect to diagnosis status, brain region, and cell type at multiple steps, including indexing and pooling prior to sequencing.ATAC-seq libraries were sequenced by Hi-Seq 2500 and Novaseq 6000 (Illumina) obtaining 50bp paired-end reads.We implemented our in-house pipeline for processing and quality control of ATAC-seq data as described previously 15,16 .The raw reads were trimmed with Trimmomatic 55 and then mapped to human reference genome GRCh38 with the pseudoautosomal region masked on chromosome Y with the STAR aligner 56 .For each sample, this yielded a BAM le of mapped paired-end reads sorted by genomic coordinates.From these les, reads that mapped to multiple loci, or to the mitochondrial genome, were removed using samtools 57 and duplicate reads removed with PICARD.Quality control metrics were reported with ataqv 58 , phantomtools 59 and Picard.

Differential Chromatin Accessibility
To identify open chromatin regions (OCRs) showing differential accessibility between males and females, we evaluated them statistically.First, we ltered out OCRs that were lowly accessible in most of the samples.Then, we applied an approach based on the Bayesian information criterion to select technical covariates as described in previous publications 15,16 .The nal formula for statistical modeling using voomWithDreamWeights 17,18 incorporated both biological covariates, such as sex (i.e., male, female), brain region (i.e., DLPFC, ACC), age at death and diagnosis (i.e., SCZ, BD, and Controls), as well as technical covariates selected by BIC (i.e.fraction of reads in peaks (FRiP) and GC content).The resulting models, based on weighted least-squares linear regression, estimated the effect of the these variables on the chromatin accessibility of each OCR: Because we keep multiple samples per specimen, we used the statistical software package dream 17,18 to properly model correlation structure and thus keep the false discovery rate (FDR) low.Multiple hypothesis testing was adjusted for using an FDR of < 5%.

Disease Heritability Analysis
To examine the role that identi ed OCRs may play in various diseases and traits, we tested whether they were enriched in common trait-associated genetic variants from a selection of genome-wide association studies (GWASs) 47,[60][61][62][63][64][65][66][67][68][69] .For this, linkage disequilibrium (LD) score-partitioned heritability 70 was used, whereby common genetic variants located in genomic regions of interest are tested to see if they explain more of the heritability than variants not in the regions of interest, while correcting for the number of variants in either category.Additionally, the LD-score approach corrects for potential biases from the general genetic context of the regions of interest, e.g.coding regions and conserved regions.From this regression, a P-value as well as a regression coe cient is outputted.For the AD GWAS, we removed the APOE effect in the model by excluding SNPs around the APOE gene.For MDD and PD GWAS, we used summary statistics generated without 23andMe individuals.The broad MHC-region (hg19:chr6:25-35MB) was excluded due to its extensive and complex LD structure 71 but, otherwise, default parameters were used.

Prediction of Enhancer-promoter Interactions
We used the Activity-by-contact model (ABC) 34 to construct a comprehensive regulatory map of enhancerpromoter (E-P) interactions in neuronal and non-neuronal cell types.This model requires (i) contact frequency between putative enhancers and promoters of regulated genes; and (ii) enhancer activity data.
The contact frequency matrices were created from previously generated neuronal and non-neuronal Hi-C datasets composed of eight post-mortem human brains 35 .For enhancer activity data, we used cell type and brain region speci c ATAC-seq signal (current study) and H3K27ac ChIP-seq signal 36 .We used the default threshold of ABC score (minimum score of 0.02), the default screening window (5MB around TSS of each gene) and we removed E-P links to ubiquitously expressed genes.Additional information on computational validation of E-P predictions using overlap with chromatin states from the human adult brain can be found in Dong et.al 35 .

Cis regulatory landscape analysis
Cell-type speci c covariates corrected expression matrices from four datasets (i.e. two cell types and brain regions) were utilized for CRD analysis.The CRD calling pipeline followed the methodology described in Girdhar et.al 72 .In total, 6,706 neuronal CRDs and 4,612 non-neuronal CRDs were obtained.
We employed a two-stage testing approach using the stageR package 73 to identify signi cant CRDs.The process involved two distinct stages: the screening stage (stage 1) and the con rmation stage (stage 2).
A CRD was considered sex-speci c if it obtained a p-value < 0.05 in stage 2, indicating signi cant differences in OCR accessibility between males and females.In order to summarize the directionality (or sex speci city) of CRDs, we calculated the mean log 2 FC (males vs females) of OCRs within each CRD.An upregulated (male-speci c) CRD had a mean log2FC (males vs females) > 0, while a downregulated (female-speci c) CRD had a mean log2FC (males vs females) < 0.

Clustering of sex-speci c CRDs into TRDs
We evaluated whether the interaction between sex-speci c CRDs across samples is strati ed into domains called trans-regulatory domains (TRDs) that can inform us about speci c underlying molecular mechanisms.We limited the identi cation of TRDs to sex-speci c neuronal CRDs because few sexspeci c CRDs were observed in non-neurons.The optimal number of clusters was found by evaluating Baker-Hubert GAMMA index 41 while varying the cluster size (from k = 2:20).GAMMA index is a measure of compactness (how similar are the objects within the same cluster), separation (how distinct are objects from different clusters), and robustness (how reproducible are the clusters in other datasets).
Using the GAMMA index, we found k = 14 optimal clusters for neuronal CRDs.Also, we curated two lists, one with all OCRs and the second with only sex-speci c OCRs, within each TRD for all downstream analysis.We measured the Spearman correlation coe cient with SCZ-associated changes detected at FDR < 0.05 in Girdhar et.al 72 .

Gene Set Enrichment and TF Analyses
To explore the function of a gene set, we collected functional gene set annotations from i) MSigDB 7.0 gene sets 74 of sizes 10 to 1000 genes, ii) SynGO 75 , and iii) DisGeNET database 33 .Fisher exact tests were used to test the enrichment and signi cance.Metascape was used to perform clustering analysis of signi cant functional gene sets 76 .To prioritize TFs that are likely responsible for observed changes in

Figures Figure 1
Figures

Figure 4 Sex
Figure 4

Figure 5 Sex
Figure 5