. Author manuscript; available in PMC: 2026 Oct 17.

Published in final edited form as: Cell Rep. 2026 Jun 15;45(6):117514. doi: 10.1016/j.celrep.2026.117514

Convergent molecular signatures across eating disorders and obsessive-compulsive disorder in the human brain

Michael S Breen ^1,^2,^3,^9,^10,^11,^*, Ran Tao ^4,⁹, Andy Yang ^1,^2,^3,⁹, Xuran Wang ^1,², Pardis Amini ^1,^2,³, Miguel Rodriguez de los Santos ^1,^2,³, Anna C Brandtjen ⁴, Amy Deep-Soboslay ⁴, Walter H Kaye ⁵, Thomas M Hyde ^4,^6,⁷, Joel E Kleinman ^4,^6,¹⁰, Joseph D Buxbaum ^1,^2,^3,¹⁰, Dorothy E Grice ^1,^2,^8,¹⁰

PMCID: PMC13673927 NIHMSID: NIHMS2190757 PMID: 42295976

SUMMARY

Eating disorder (ED) and obsessive-compulsive disorder (OCD) exhibit clinical and genetic overlap, yet whether they converge at the molecular level in the human brain is unknown. We perform large-scale transcriptomic profiling of the dorsolateral prefrontal cortex (DLPFC) and caudate in postmortem tissue from 86 controls, 57 individuals with ED, and 27 with OCD. ED shows robust, region-specific transcriptional dysregulation (102 differentially expressed genes [DEGs] in DLPFC and 222 in caudate at FDR <1%) that replicates in an independent cohort. OCD shows no single-cohort DEGs, but meta-analysis across three datasets identifies 57 caudate-associated genes. Despite these differences, transcriptome-wide effects strongly correlate between ED and OCD (DLPFC r = 0.67; caudate r = 0.75), indicating shared molecular pathology. Joint ED + OCD analysis identifies 233 DEGs in DLPFC and 815 in caudate, implicating GABAergic signaling, neuroendocrine regulation, mitochondrial metabolism, and CHD8-associated networks. Genetically regulated expression analyses identifies five genes (WDR6, NCKIPSD, P4HTM, DALRD3, and SHISA5) with convergent risk associations across disorders and brain regions, all mapping to a gene-dense region on chromosome 3. These findings define a shared cortico-striatal transcriptional architecture and identify candidate genes for transdiagnostic intervention.

In brief

Breen et al. show that eating disorders and obsessive-compulsive disorder share transcriptional alterations across cortico-striatal circuits. Joint analyses reveal coordinated disruption of CHD8 regulatory networks, mitochondrial energy metabolism, and inhibitory synaptic signaling. Transcriptome-wide association analyses converge on a shared genetic locus on chromsome 3, defining a common molecular substrate for compulsive behavior.

Graphical Abstract

graphic file with name nihms-2190757-f0006.webp

INTRODUCTION

Eating disorders (EDs), particularly anorexia nervosa (AN) and bulimia nervosa, and obsessive-compulsive disorder (OCD) are complex neuropsychiatric conditions, each defined by distinct clinical manifestations yet sharing common behaviors and motivations.^1–6 The genetic overlap between these disorders is increasingly supported by familial patterns,^7,8 twin studies,^9,10 and genome-wide associations,^11–17 suggesting shared biological mechanisms underlying the overlapping obsessive-compulsive symptoms. Moreover, epidemiological evidence indicates heightened comorbid risk between ED and OCD.^10,18 However, despite evidence of convergence at the genetic and clinical levels, it is unknown whether these disorders share molecular alterations in the human brain or whether their overlap reflects distinct neurobiological processes.

Recent transcriptomic studies have begun to define molecular alterations in brain regions implicated in ED and OCD, particularly in frontal and striatal circuits. In OCD, altered gene expression across striatal regions, specifically the caudate, nucleus accumbens, and putamen, points to perturbations in synaptic regulation, immune function, and neurotransmitter homeostasis, each region leaving a distinct transcriptomic signature.^19–22 In ED, preliminary findings have pointed to the dysregulation of genes within the dorsolateral prefrontal cortex (DLPFC), which are implicated in metabolic pathways also disturbed in OCD,²² reinforcing the concept of a functional linkage between the disorders. However, these studies have largely examined single disorders, single regions, or modest sample sizes, limiting direct tests of shared versus disorder-specific transcriptional changes across cortico-striatal circuitry. As a result, whether ED and OCD converge on common molecular pathology at the transcriptomic level remains unresolved.

Clinically, ED and OCD are characterized by anxiety-provoking thoughts and related compulsive behaviors and avoidance, which serve to temporarily reduce distress.^1,3,23 While the themes of the thoughts and behaviors associated with ED and OCD are diagnostically distinct, these disorders share a common thread of anxiety-driven and ritualized response patterns.³ Therapeutically, there is also overlap in the evidence-based treatments for these disorders, including the cognitive-behavioral therapy approaches and pharmacological interventions.^24–28 This overlap has long suggested the possibility of shared circuitry and biology, but that hypothesis has not been tested systematically in postmortem human brain tissue across both disorders and regions.

Here, we present a large-scale transcriptomic analysis of the caudate and DLPFC from individuals with ED or OCD, integrating discovery, replication, and genetic frameworks to directly evaluate cross-disorder molecular convergence. We focused on these regions because they represent key nodes of cortico-striatal-thalamic circuitry, with the DLPFC supporting executive control and cognitive regulation, and the caudate mediating habit formation and action selection, processes central to compulsive behavior in both disorders. By jointly analyzing these regions across both disorders within a unified framework, this study enables direct evaluation of cross-disorder transcriptional convergence within a circuit-level context. To assess replication and diagnostic specificity within the caudate, we employed an independent replication cohort that also includes non-ED major depressive disorder (MDD) psychiatric controls, providing an internal benchmark to minimize false-positive disease signals. Together, this approach advances prior work by integrating cross-disorder, multi-region transcriptomic analyses with complementary concordance and genetic frameworks to define shared and distinct molecular signatures underlying ED and OCD.

RESULTS

Deep RNA sequencing of DLPFC and caudate in ED and OCD

We generated deep bulk RNA sequencing (RNA-seq) data from postmortem DLPFC and caudate tissues of 169 human donors, comprising neurotypical controls (N = 86), individuals with EDs (N = 57), and individuals with OCD (N = 27), with diverse representation of biological sex and ethnicity (Figure 1A). Among the ED cases, 25 were classified as bingeing and purging subtypes (AN-BP, n = 14; bulimia nervosa, n = 11), 8 were diagnosed with restricting-type AN (AN-R), and 24 were designated as unspecified ED. The majority of cases presented with comorbid MDD and a history of substance use (see details in Tables S1A–S1D). Notably, body mass index (BMI) was significantly reduced in the ED group compared to controls (p = 0.009) but did not differ from the OCD group (p = 0.18; Mann-Whitney U test). To investigate disorder-related transcriptomic differences in regions relevant to compulsive behaviors and executive control, we focused on the DLPFC and caudate, two brain regions known to be central in ED and OCD pathophysiology. These regions represent distinct components of cortico-striato-thalamic circuitry implicated in dysregulated decision-making and ritualized behaviors.

Figure 1. — (A) Bulk RNA sequencing (RNA-seq) data were generated from 169 donors examining the dorsolateral prefrontal cortex (DLPFC) and caudate nucleus; sample-level demographics are shown.

(B) Density plots illustrate anticonservative p value distributions from differential expression analyses comparing eating disorder (ED; n = 57) and obsessive-compulsive disorder (OCD; n = 27) cases versus neurotypical controls (n = 86) in the DLPFC (left) and caudate (right); p values were derived from transcriptome-wide moderated t tests with FDR correction (limma-voom).

(C) Venn diagram showing the overlap of all differentially expressed genes (DEGs; FDR <1%) for ED in the DLPFC and caudate. The overlap indicates 11 genes that are down-regulated across both regions.

(D and E) Volcano plots depicting ED gene-level effects in (D) the DLPFC (n = 57 ED, n = 86 controls) and (E) the caudate (n = 57 ED, n = 86 controls); the red dashed line marks FDR <1%, and the gray dashed line marks FDR <5%; p values were derived from transcriptome-wide moderated t tests (limma-voom).

(F and G) Scatterplots comparing log₂ fold change for ED and OCD gene-level effects in (F) the DLPFC and (G) the caudate, with Spearman’s correlation coefficients shown; each point represents one gene (n = 18,601 genes).

(H–K) Replication of ED signatures in the caudate: (H) unsupervised hierarchical clustering of donors unique to the replication cohort (4 ED cases, 10 controls) using the 222 ED DEGs identified in the discovery cohort; VOOM-normalized gene expression was Z scored across samples and clustered using Euclidean distance and Ward’s linkage. (I) Scatterplot comparing log₂ fold changes of the 222 ED DEGs between the discovery cohort (x axis; n = 57 ED, n = 86 controls) and the replication cohort (y axis; n = 38 ED, n = 36 controls), with Spearman’s correlation coefficient shown. (J) Boxplots showing expression of top ED caudate genes, *DIRAS2* and *VGF*, in the discovery cohort (n = 57 ED, n = 86 controls); FDR-corrected p values from transcriptome-wide moderated t tests are shown. (K) Boxplots showing a null MDD effect for the same top ED genes in the replication cohort caudate (n = 37 MDD, n = 36 controls); p values from moderated t tests; n.s., not significant. See also Figures S1–S10 and Table S1.

Regional and cell-type-associated sources of gene expression variation

Rigorous quality control confirmed the integrity of gene expression and allowed us to explore biological and technical sources of variance across samples (Figures S1–S3). As expected, neuronal cell-type proportions estimated from bulk RNA profiles indicated that DLPFC samples were composed of a higher fraction of neurons (~70.8%) compared to caudate samples (~45.5%) (p = 1.9 × 10⁻³⁰; Mann-Whitney U test), consistent with higher neuronal fraction in neocortex relative to striatum²⁹ (Figure S1). Principal-component analysis (PCA) separated DLPFC and caudate samples along PC1 (Figure S1C), consistent with broad anatomical and transcriptional differences between regions. Indeed, ~86% of expressed genes were differentially expressed between the DLPFC and caudate at false discovery rate (FDR) <1% (Figure S2A; Table S2), underscoring the distinct transcriptional landscapes of these regions. Despite these region-specific differences, transcriptome-wide gene expression profiles remained highly correlated between DLPFC and caudate samples from the same individuals (r > 0.9), and this within-individual concordance did not vary by diagnostic group (p = 0.93; Kruskal-Wallis test) (Figure S2B). Together, these findings support analyzing DLPFC and caudate separately while indicating that diagnostic effects are broadly consistent across regions.

To determine which biological and technical factors should be explicitly controlled in disorder comparisons, we next quantified the contribution of known covariates to gene expression variation using a linear mixed model (Figure S3). Collectively, these variables explained ~37.5% and ~54.8% of transcriptome variation in the DLPFC and caudate, respectively. In the DLPFC, variation in neuronal cell-type proportions explained the largest transcriptome-wide variance (median 6.8%), followed by the percentage of intronic reads (5.7%), percentage of rRNA (5.1%), RIN (1.8%), and biological age (0.47%), with remaining factors contributing more modest effects. Expression variation due to clinical diagnosis (i.e., ED or OCD) was detectable in a subset of genes. Importantly, similar covariate-associated variance profiles were observed independently in the caudate (Figures S3C and S3D). These analyses provided a foundation for covariate control in all subsequent differential expression models.

Region-specific differential gene expression in ED and cross-disorder concordance with OCD

We next tested for differentially expressed genes (DEGs) among ED cases versus control donors adjusting for the possible influence of estimated neuronal cell-type proportions, RIN, age, sex, ethnicity, and percentage of intronic and rRNA reads. This analysis identified 102 ED DEGs in the DLPFC and 222 ED DEGs in caudate at strict transcriptome-wide significance (FDR <1%) (Figure 1B). These signatures were largely region-specific, with limited overlap; only 11 genes showed consistent downregulation across both regions (Figure 1C). Among the top-ranked DEGs in the DLPFC (by FDR) were somatostatin (SST) and cortistatin (CORT), genes enriched in a subset of GABAergic inhibitory neurons, alongside reduced expression of VGF and corticotropin-releasing hormone (CRH), highlighting neuroendocrine and inhibitory signaling pathways relevant to stress regulation and behavior control (Figures 1C, 1D, and 1E). A region × diagnosis interaction analysis identified four genes (ADAMTS2, GRB14, NPNT, and PCDH11Y) exhibiting modest region-dependent effects in ED (FDR <1%), while no such effects were observed in OCD.

To further assess robustness and potential sources of heterogeneity, we performed a series of follow-up analyses. First, sensitivity analyses evaluating BMI, MDD comorbidity, substance use history, sex, and ED subtype demonstrated that gene-level effect sizes remained highly concordant across all comparisons (Figure S4), supporting the stability of the observed transcriptional signals. Second, although sex distributions differed across diagnostic groups, consistent with the epidemiology of ED and OCD, no transcriptome-wide significant diagnosis × sex interactions were detected (Table S2). Third, exploratory subtype analyses (AN-R, AN-BP, and unspecified ED) revealed highly similar transcriptional profiles, particularly in the caudate (Figure S5). While not powered to detect subtype-specific differential expression, this concordance suggests that ED-associated transcriptional changes are broadly shared across subtypes rather than driven by specific clinical groups.

Given limited power in the OCD cohort (n = 27), we assessed cross-disorder concordance and observed strong transcriptome-wide correlations in gene-level effect sizes between ED and OCD in both the DLPFC (r = 0.67) and caudate (r = 0.75) (Figures 1F and 1G), despite the absence of OCD-specific DEGs at FDR <1%. This concordance was further supported by significant rank-rank hypergeometric overlap (Figure S6), indicating consistent directional effects across disorders. In line with this, direct differential expression analysis comparing ED and OCD cases identified no significant DEGs (FDR <1%) (Figures S7A and S7B). Bayesian analysis reinforced this interpretation whereby the majority of genes supported the null hypothesis of no differential expression between disorders (BF01 > 1), with a smaller subset providing moderate support (BF01 > 3) and no genes showing strong evidence for differential expression between ED and OCD (Figures S7C and S7D). Together, these results indicate a high degree of transcriptional similarity between disorders.

Replication and specificity of ED caudate transcriptional signatures

Given the robust gene expression effects in the caudate of ED cases, we sought to validate these findings in our independent replication cohort. Replication analyses were designed to assess reproducibility and effect direction, rather than to perform independent discovery (see Methods). This cohort included 38 ED cases (34 re-sequenced samples from donors also in the discovery cohort and 4 donors unique to the replication cohort), 36 neurotypical controls (26 re-sequenced samples from donors also in the discovery cohort and 10 donors unique to the replication cohort), and 37 MDD non-ED neuropsychiatric controls (Tables S1A–S1D). All individuals in this cohort were female and of European ancestry.

We first confirmed that cell-type composition, gene expression variance, and technical covariates were consistent with the discovery cohort (Figure S8). Then, using the subset of 34 ED and 26 control re-sequenced samples from donors shared between the discovery and replication cohorts, we quantified technical reproducibility of the ED caudate effect, observing a strong correlation (r = 0.89) between cohorts (Figure S9), underscoring the robustness of these DEGs to resequencing, library preparation, and related technical sources of variation. Next, to assess replication of the direction and structure of the biological ED signal in independent donors, we tested whether the 222 ED DEGs from the discovery cohort could classify the donors unique to the replication cohort (4 ED cases, 10 controls). Unsupervised hierarchical clustering (Euclidean distance, Ward’s linkage) based on these genes separated ED from control samples (Figure 1H), and expression fold changes were strongly correlated between cohorts (r = 0.73; Figure 1I), supporting the presence of a conserved ED-associated caudate signature. Key ED-associated genes, including DIRAS2 and VGF, showed consistent and significant downregulation across cohorts (Figure 1J), highlighting their potential as ED biomarkers and/or mechanisms. No significant associations between BMI and gene expression were detected, indicating that these transcriptomic ED signatures are not driven by BMI differences. Although independent validation was limited to four ED donors unique to the replication cohort, these analyses support conserved directionality and multigene structure of the ED caudate signature.

A key strength of the replication cohort was the inclusion of an MDD non-ED psychiatric comparison group, enabling direct assessment of specificity. No DEGs were detected between MDD cases (n = 37) and controls (n = 36) at FDR <1% (Figures S10A and S10B), and transcriptome-wide concordance between ED and MDD gene-level effect sizes was low (r = 0.19; Figure S10C). Importantly, core ED-associated genes, including DIRAS2 and VGF, were not dysregulated in MDD (Figure 1K). These findings complement the sensitivity analyses in the discovery cohort (Figure S4) and indicate that ED-associated transcriptomic signatures are not attributable to comorbid depression.

Combined ED + OCD analysis reveals convergent molecular pathology

Given the strong transcriptomic concordance between ED and OCD (Figures 1F and 1G), we performed a joint differential expression analysis comparing combined ED and OCD cases (ED + OCD) against controls in each brain region. This combined analysis substantially increased statistical power, identifying 233 DEGs in the DLPFC and 815 in the caudate at FDR <1% (Figure 2A). The majority of these DEGs overlapped with ED-only signatures (81.3% in DLPFC and 90% in caudate), indicating that OCD contributes a highly concordant molecular signal that becomes detectable when analyzed jointly. Among ED + OCD DEGs, 56 genes were consistently dysregulated across both regions, including key neuroendocrine and synaptic regulators such as CRH, VGF, and tachykinin precursor 1 (TAC1). These convergent signatures remained robust after correction for RNA degradation using quality surrogate variable analysis (qSVA) (Figure S11), supporting their biological validity. To further exclude potential confounding by medication exposure, we performed drug perturbation enrichment analysis using CNS-focused drug signatures (see Methods). No significant overlap was detected between ED + OCD transcriptional signatures and major psychotropic drug profiles in either the DLPFC or caudate (FDR >0.05) (Table S3), arguing against medication exposure as a driver of these molecular alterations.

Figure 2. — (A) Bar graph displaying the count of DEGs (FDR <1%) for ED alone and the combined ED + OCD analysis across DLPFC and caudate (ED: n = 57; OCD: n = 27; controls: n = 86); the inset Venn diagram illustrates ED + OCD DEG overlap between DLPFC and caudate.

(B and C) Volcano plots showing significance (-log₁₀ FDR p value; y axis) versus magnitude of expression change (log₂ fold change; x axis) for ED + OCD DEGs in (B) the DLPFC (n = 84 cases, n = 86 controls) and (C) the caudate (n = 84 cases, n = 86 controls); the red dashed line marks FDR <1%, and the gray dashed line marks FDR <5%; p values from transcriptome-wide moderated t tests (limma-voom).

(D) CAMERA gene-set enrichment heatmap showing enrichment of genes underexpressed in OCD, ED, and ED + OCD within the DLPFC and caudate; enriched categories include mitochondrial translation, oxidative phosphorylation, mRNA stability, genes with postnatal expression bias, and CHD8-interacting proteins; asterisks (*) indicate FDR <1% by competitive gene set enrichment test (CAMERA).

(E) CAMERA barcode plot illustrating gene set enrichment for mitochondrial translation plotted against ED + OCD genes ranked by log₂ fold change in the caudate; this corresponds to the black-outlined cell in (D) (FDR p value = 3.7 × 10⁻²⁸).

(F) Bubble heatmap comparing transcriptome-wide gene-level effect sizes (Spearman correlations) for OCD, ED, and ED + OCD in the current study against those from independent brain transcriptome datasets; bubble size reflects correlation magnitude; OCD¹ (Lisboa et al., 2019), OCD² (Piantadosi et al., 2021), ED/OCD³ (Jaffe et al., 2014), and PTSD³⁰ (Jaffe et al., 2022). See also Figures S11–S12 and Tables S3–S4.

In the DLPFC, ED + OCD cases showed marked downregulation of SST, along with reduced expression of CRHBP and brain-derived neurotrophic factor (BDNF), genes central to inhibitory signaling, neurodevelopment, and synaptic plasticity (Figure 2B). Consistent with these findings, we also observed decreased expression of canonical GABAergic synaptic markers, including GAD1, GAD2, and SLC32A1, genes previously implicated in OCD and related neuropsychiatric disorders.²⁰ In the caudate, prominent downregulated genes included DIRAS2, associated with cognitive control and attention-deficit/hyperactivity disorder (ADHD) susceptibility,^31,32 and DUSP2, linked to neuronal maintenance and neuroinflammatory regulation^33,34 (Figure 2C). To contextualize these gene-level effects, we applied competitive gene set enrichment analysis (CAMERA), which revealed strong overrepresentation of pathways related to mitochondrial translation, oxidative phosphorylation, RNA stability, and developmentally regulated genes that normally increase across cortical maturation among underexpressed genes (Figures 2D and 2E; Table S4). These enrichments define a coordinated disruption of neuronal energy metabolism and translational capacity across cortico-striatal circuits in ED and OCD.

Notably, a substantial fraction of downregulated genes within these pathways are known interaction partners or transcriptional targets of CHD8 (Figure 2D), a chromatin remodeling factor and established genetic risk locus for OCD, autism, and related psychiatric conditions.^35–37 This convergence suggests that CHD8-linked regulatory networks may represent a shared molecular vulnerability underlying compulsive-spectrum disorders, coupling chromatin-mediated transcriptional control to neuronal metabolic and synaptic dysfunction.

Finally, to capture regulatory effects beyond gene-level abundance, we performed differential transcript usage (DTU) analyses across ED, OCD, and ED + OCD cases. DTU analysis identified 917 significant events in the ED + OCD caudate, largely independent of gene-level DEGs (55 overlapping genes in the caudate, 12 in the DLPFC), with isoform switching observed in mitochondrial and synaptic genes including MOSPD2, ATP5F1B, and NDUFB9, suggesting transcript-level dysregulation as a partially orthogonal layer of molecular variation in ED and OCD (Figures S12 and S13).

Cross-disorder transcriptomic concordance with independent postmortem brain datasets

To contextualize our findings within broader neuropsychiatric research, we compared our transcriptomic signatures for ED, OCD, and ED + OCD against transcriptome-wide summary statistics from nine prior postmortem brain tissue studies across a range of psychiatric and inflammatory disorders (Table S5). Two key insights emerged from this cross-disorder comparison (Figure 2F). First, the transcriptomic patterns we identified for ED, OCD, and the combined ED + OCD within the caudate showed strong replication and the highest alignment with transcriptomic data from separate smaller OCD and ED studies. Specifically, correlations with OCD-related transcriptome changes from Lisboa et al., 2019¹⁹ (n = 6 OCD cases) showed substantial agreement in the caudate, putamen, and nucleus accumbens with our ED (r = 0.56, r = 0.52, and r = 0.39), OCD (r = 0.42, r = 0.40, r = 0.37), and combined ED + OCD (r = 0.60, r = 0.57, and r = 0.43) signatures (Figure S14). A similar pattern of high correlation was evident with OCD-related findings from Piantadosi et al., 2021²¹ (n = 7 OCD cases), in the nucleus accumbens, orbitofrontal cortex (OFC), and caudate relative to our ED (r = 0.37, r = 0.38, and r = 0.42, respectively) and ED + OCD (r = 0.32, r = 0.35, and r = 0.40, respectively) signatures in the caudate. We also observed high transcriptome-wide correlations with ED-related findings from Jaffe et al., 2014²² (n = 15 ED cases) in the DLPFC, with ED (r = 0.35), OCD (r = 0.31), and ED + OCD (r = 0.36) signatures in the DLPFC.

Second, independent bipolar disorder (BD) and schizophrenia (SCZ) signatures from the DLPFC^30,38 were more closely aligned with DLPFC than caudate signatures, suggesting a regional specificity in gene expression patterns. The minimal or absent overlap of alcohol abuse disorder (AUD) and inflammatory bowel disease (IBD) signatures³⁸ with our data further implies that the transcriptomic similarities we observed are unlikely to be attributed to comorbid substance abuse, overall health status, or non-specific postmortem brain changes, supporting the validity of our findings.

Cross-cohort meta-analysis expands the OCD caudate transcriptional landscape

Given the convergence of transcriptome-wide OCD gene expression signatures and the modest OCD sample size in our primary dataset, we next performed a gene-based meta-analysis of the caudate transcriptome across three independent studies: our current cohort (N = 27), Lisboa et al., 2019 (N = 6), and Piantadosi et al., 2021 (N = 7). This analysis expanded the OCD sample size (N_total = 40) and improved power for detecting DEGs. Using a restricted maximum likelihood (REML) approach and harmonized gene-level statistics across studies, we identified 57 DEGs in OCD caudate (FDR <1%), including 30 overexpressed and 27 underexpressed (Figure 3; Table S6). Among these, we observed underexpressed DIRAS2 and the overexpressed GLI family zinc finger 2 (GLI2), both of which were prominent and significantly dysregulated in ED, reinforcing the joint ED + OCD analysis findings. We also found overexpression of several cytokine receptor activity genes in OCD caudate, including interleukin-10 receptor subunit beta (IL10RB), interleukin-11 receptor subunit alpha (IL11RA), and interferon lambda receptor 1 (IFNLR1).

Figure 3. — (A) Volcano plot of differential gene expression associated with OCD in the caudate nucleus, plotting significance (-log₁₀ p-value; y axis) against effect size (log₂ fold-change; x axis), aggregating data from 40 OCD cases and 47 controls across three independent studies (current study: n = 27 OCD; Lisboa et al., 2019: n = 6 OCD; Piantadosi et al., 2021: n = 7 OCD); meta-analysis identified 30 overexpressed and 27 underexpressed genes at FDR <1% using a random-effects model (metafor package, R); bold gene symbols are highlighted in (B).

(B) Forest plots showing meta-analysis estimates of log₂ fold change for *DIRAS2*, *GLI2*, and *IL10RB* across three independent studies; markers indicate per-study log₂ fold change; error bars represent 95% confidence intervals (CIs); the pink diamond summarizes the combined random-effects estimate. See also Figure S14 and Tables S5–S6.

Gene co-expression network analysis reveals shared and disorder-specific modules

To identify systems-level transcriptional patterns associated with ED and OCD, we performed weighted gene co-expression network analysis (WGCNA) separately for the DLPFC and caudate across all samples. In total, we identified 13 modules in the DLPFC and 15 modules in the caudate, each representing clusters of genes with highly correlated expression patterns (Figure S15). The integrity of these co-expression modules was largely preserved across both brain regions, with most variances observed in the smaller modules (Figure S15).

Particularly noteworthy were Md1 in the DLPFC and Mc1 in the caudate. Both demonstrated a clear neuronal profile (Figures 4A and 4B; Figure S16), were enriched for DEGs, and were consistently downregulated in ED + OCD (Figure 4C). These modules were enriched for proteins interacting with CHD8 (Figure 4D) and for genes involved in mitochondrial translation and oxidative phosphorylation (Figures 4D–4F), implicating disruptions in neurotransmission and neuronal energy production. A striking 76.1% of genes in Mc1 were also present in Md1, and the high degree of shared protein-protein interactions further supports a coordinated network of synaptic and metabolic dysfunction (Figure S17). The Mc1 module formed a densely interconnected network enriched for mitochondrial and oxidative phosphorylation genes, including core ATP synthase subunits (ATP5F1, ATP5MC1, and ATP5PD) and mitochondrial ribosomal proteins (e.g., MRPL3), pointing to coordinated bioenergetic disruption.

Additional modules offer insights into shared and distinct mechanisms. In the DLPFC, Md2 was downregulated across all conditions (Figure 4C) and enriched for genes related to innate immune responses (Figure 4E). The Md3 module, characterized by a glial cell signature (Figure 4A; Figure S16), was downregulated in ED + OCD (Figure 4C) and implicated in myelination and glial cell development (Figure 4E). In the caudate, Mc2 was distinctly upregulated in ED and ED + OCD but not OCD (Figure 4C), enriched for glial cell function, chromatin remodeling, and RNA processing genes (Figure 4E). The Mc3 module, also upregulated and glial-specific, was enriched for cellular activation and cytokine response pathways (Figure 4E), reinforcing a potential role for glial cells in the pathophysiology of both disorders. Collectively, these findings underscore the roles of Md1 and Mc1 in the shared neuropathology of ED and OCD, while additional modules highlight distinct pathways contributing to each disorder’s molecular underpinnings.

Genetically predicted gene expression changes

To complement differential expression analyses, we performed transcriptome-wide association studies (TWASs) to assess genetically predicted gene expression associated with ED and OCD risk across discovery (DLPFC and caudate) and replication (caudate) cohorts.

For ED, we identified 63 significant associations (FDR <5%), representing 51 unique genes, with five replicating across regions and cohorts (Figures 5A and 5B; Table S7). These genes were not detected as DEGs but were enriched for metabolic pathways (FDR, p = 0.03), including IMPDH2, SLC25A20, GBE1, KYNU, and DPYD. Notably, many associations localized to a gene-dense region on chromosome 3 (47.6–51.4 Mb), overlapping a known AN GWAS locus, consistent with prior TWAS findings in ED³⁹ (Figures S18 and S19). Among replicated signals, WD repeat domain 6 (WDR6) showed strong and consistent upregulation associated with increased ED risk across all regions (Z = 7.11, p = 1.1 × 10⁻¹², DLPFC; Z = 5.54, p = 3.0 × 10⁻⁸, caudate discovery; Z = 7.22, p = 5.3 × 10⁻¹³, caudate replication). Downregulation of NCK interacting protein with SH3 domain (NCKIPSD) was similarly associated with increased ED risk across regions. Other notable downregulated genes in the chromosome 3 cluster associated with increased ED risk include DALRD3, C3orf62, P4HTM, and CTNNB1, as well as MGMT on chromosome 10. Several of these genes have been previously implicated, supporting reproducibility, while others (e.g., LINC02055) represent original associations.

Figure 5. — (A–D) Manhattan-style TWAS plots showing gene-level association Z scores for ED in (A) the DLPFC and (B) the caudate and for OCD in (C) the DLPFC and (D) the caudate; positive Z scores indicate increased genetically predicted expression associated with higher disorder risk, and negative Z scores indicate decreased expression associated with risk; dashed lines mark |Z| = 4; genes with reproducible significant associations across both disorders and brain regions are labeled in bold; TWAS was performed using FUSION with ED GWAS (Watson et al., 2019) and OCD GWAS (Arnold et al., 2018).

(E) Bar plot summarizing the number of significant TWAS hits (FDR <5%) across 13 GTEx brain regions stratified by disorder (ED, pink; OCD, purple); peak signals are observed in cortex, frontal cortex, and hippocampus.

(F) Heatmap of gene rank percentiles (based on FDR significance) for five genes with consistent TWAS associations across ED and OCD in all 13 GTEx brain regions; color intensity reflects relative TWAS significance (0–100th percentile); asterisks indicate FDR significance thresholds (*FDR <5%; **FDR <1%); DALRD3 and NCKIPSD show widespread enrichment across cortical and subcortical regions supporting cross-disorder involvement. See also Figures S18–S19 and Tables S7–S8.

For OCD, TWAS identified 125 significant associations (FDR <5%), representing 106 unique genes, with eight replicating across datasets (Figures 5C and 5D; Figures S18D–S18F; Table S8). Although these genes were not enriched for metabolic pathways, they showed strong overrepresentation within the same chromosome 3 locus previously linked to ED. In the DLPFC, 8 out of 50 significant hits fell within this region (P4HTM, NCKIPSD, DALRD3, WDR6, C3orf62, IMPDH2, AL391244.2, and RBM6), representing a ~9.9-fold enrichment relative to background (Fisher’s exact test, p = 1.2 × 10⁻⁷). Similarly, in the caudate, 6 out of 30 significant hits mapped to this region (P4HTM, NCKIPSD, DALRD3, SHISA5, WDR6, and C3orf62), corresponding to a ~11.7-fold enrichment (Fisher’s exact test, p = 6.4 × 10⁻⁷), emphasizing this locus as a cross-disorder hub for ED and OCD. Consistent with this, overlap between TWAS and DEG results was minimal, suggesting that genetically regulated expression captures complementary aspects of disease biology.

Across both ED and OCD, five genes (WDR6, NCKIPSD, P4HTM, DALRD3, and SHISA5) showed consistent associations with strong effect sizes across regions (Table 1), highlighting a convergent set of loci within the chromosome 3 region. To assess regional specificity, we extended TWAS across 13 GTEx brain regions (Figure 5E). Although genome-wide significance varied, key genes showed consistent cross-regional signals. DALRD3 was robustly enriched across cortex, cerebellar hemisphere, and amygdala (FDR <1%) (Figure 5F), while NCKIPSD and P4HTM were significant in cortex and SHISA5 in frontal cortex (FDR <5%). Even when sub-threshold, these genes consistently ranked among the top TWAS associations, indicating stable directional effects. Notably, DALRD3 exhibited cross-disorder enrichment across both cortical and subcortical structures, including the amygdala, cerebellum, and cortex, reinforcing its putative role in shared disease biology. These results support regional consistency of genetically predicted expression for these genes, particularly within cortical and limbic circuits. Variation in significance across all GTEx regions likely reflect smaller sample sizes in GTEx relative to our postmortem cohorts, emphasizing the value of integrating multiple transcriptomic resources for replication.

Table 1.

Cross-disorder TWAS associations replicating across brain regions in ED and OCD

				ED TWAS DLPFC		ED TWAS caudate		OCD TWAS DLPFC		OCD TWAS caudate		Supporting Literature
Chr	Start	End	Gene symbol	TWAS.Z	FDR.P	TWAS.Z	FDR.P	TWAS.Z	FDR.P	TWAS.Z	FDR.P	Supporting Literature

3	49007062	49015953	WDR6	7.12	1.06E–08	5.54	0	4.85	0.0007	6.39	3.28E–07	[50, 69]
3	48673844	48686364	NCKIPSD	−5.37	1.25E–04	−4.72	0.003	−6.39	3.65E–07	−6.55	2.28E–07	[50, 69]
3	49015488	49022293	DALRD3	−5.40	1.25E–04	−7.06	1.50E−08	−5.89	5.46E–06	−5.16	3.17E–04	[49, 50, 69]
3	48989889	49007153	P4HTM	−5.48	1.02E–04	N/A	N/A	−6.53	2.87E–07	N/A	N/A	[50]
3	48467798	48504826	SHISA5	−2.42	0.423	−6.47	4.18E−07	−2.71	0.172	−4.22	1.47E–02	[50, 69]

Open in a new tab

Results are presented for DLPFC (dorsolateral prefrontal cortex) and caudate.

Definitions: Chr: chromosome number.

Start and End: genomic coordinates (start and end positions) of the gene. Gene symbol: the official gene symbol.

TWAS.Z: Z score from the TWAS analysis, indicating the strength and direction of the association.

FDR.P: false-discovery-rate-corrected p value, controlling for the proportion of false positives among significant results.

Supporting literature cites prior AN-related TWAS evidence.

DISCUSSION

This study provides the first large-scale molecular evidence that ED and OCD share transcriptomic and genetically regulated expression profiles across cortico-striatal circuits. Joint analysis of ED and OCD uncovered coordinated downregulation of genes linked to GABAergic signaling, neuroendocrine dysfunction, mitochondrial translation, oxidative phosphorylation, and mRNA stability. ED caudate signatures replicated independently, supporting reproducibility and resilience to MDD comorbidity. Many of the downregulated genes interact with CHD8, a key OCD risk gene,³⁶ while genetically regulated expression analyses pinpointed WDR6, NCKIPSD, P4HTM, DALRD3, and SHISA5 as convergent loci. Together, these results provide a molecular framework linking compulsivity, metabolism, and neuronal energy regulation across ED and OCD.

The strong transcriptomic concordance between ED and OCD reflects shared pathophysiological mechanisms. Consistent with limited power in the OCD cohort (n = 27), we did not detect OCD DEGs at strict transcriptome-wide thresholds; however, multiple complementary analyses supported shared molecular signal across disorders, including strong cross-disorder correlations and rank-rank overlaps. Among these signals, we observed consistent downregulation of CORT in both the DLPFC and caudate (Figures 1 and 2). Cortistatin, encoded by CORT, is expressed in and modulates GABAergic interneurons, particularly somatostatin-positive (SST⁺) interneuron populations.^40,41 Convergent downregulation of genes including SST, VGF, TAC1, and SLC32A1 further supports disruption of inhibitory neuronal signaling. Importantly, these effects persisted after covariate adjustment for bulk cell composition (including overall neuronal proportion), indicating that the observed case-control differences are not attributable to gross shifts in cellular content (Figure S1B). Prior studies have also reported decreased expression of GAD1 and GAD2 in OCD,²⁰ with GAD2 further implicated as a genetic risk factor,⁴² while reduced GAD1 expression in the striatum has been observed in Tourette syndrome,⁴³ suggesting shared vulnerability of inhibitory circuitry across neuropsychiatric conditions. Co-reduction of BDNF, a key regulator of SST-expressing GABAergic interneurons and VGF, points to a broader neurotrophic deficit affecting inhibitory circuit function, energy balance, and behavior.^44,45 Consistent with these observations, preclinical studies have shown that perturbation of cortico-striatal circuits and disruption of GABAergic interneuron function, particularly SST⁺ populations, can induce compulsive-like behaviors and impair behavioral flexibility.^46–48 These alterations are central to the regulation of compulsive behaviors and stress responses^49,50 and are further supported by genetic links in AN and related metabolic and immunological disorders.^51–53 These findings are consistent with a model in which disruptions in GABAergic interneuron function, alongside neuropeptide signaling, jointly contribute to the shared pathophysiology. Such convergence highlights how altered feedback regulation between prefrontal control and striatal habit systems may mechanistically link anxiety, compulsion, and metabolic stress.

Our work also revealed underexpression of DIRAS2 in OCD and ED in the caudate (Figures 1 and 3). DIRAS2 has been repeatedly implicated in ADHD, impaired cognitive control, and impulsivity through genome-wide association and genetic linkage studies,^32,54–58 underscoring a broader relevance across related neuropsychiatric phenotypes. Located in chromosomal region 9q32.2, DIRAS2 belongs to the Ras superfamily of GTPases,⁵⁹ which regulate cell signaling, differentiation, and growth. DIRAS2 has also been linked to executive functions through its expression in the prefrontal cortex, amygdala, and anterior cingulate cortex.^59,60 Rare variants in DIRAS2 have also been linked to suicidal behaviors,⁶¹ reinforcing its significance in psychiatric risk. Although its specific molecular pathways remain poorly defined, delineating DIRAS2’s cellular function and signaling patterns will be essential for clarifying its role in compulsivity-related circuits.

Beyond gene-level effects, DTU analysis revealed additional, largely independent layers of RNA dysregulation, most pronounced in the caudate, implicating mitochondrial and synaptic genes including MOSPD2, ATP5F1B, and NDUFB9 (Figures S12 and S13). Isoform switching in ATP5F1B and NDUFB9, which encode core subunits of ATP synthase and complex I of the mitochondrial respiratory chain, respectively, suggests that transcript-level dysregulation may fine-tune mitochondrial energy output independently of changes in total gene abundance, representing a partially orthogonal and functionally relevant layer of bioenergetic disruption in ED and OCD.

The marked downregulation of the Md1 and Mc1 modules (Figure 4), enriched for genes supporting neuronal integrity, protein synthesis, and mitochondrial energy production, points to a shared molecular deficit across ED and OCD. Disruption of these modules is likely to compromise synaptic efficacy and plasticity, as impaired bioenergetic capacity constrains translational and metabolic processes essential for neuronal function.^62,63 Protein-protein interaction analysis further revealed that Mc1 and Md1 form densely interconnected networks organized around ATP synthase components and mitochondrial ribosomal genes, reinforcing a central role for mitochondrial dysfunction. Notably, both modules were enriched for CHD8-interacting genes, implicating transcriptional and chromatin-remodeling pathways in the regulation of neuronal energy homeostasis.³⁵ These findings indicate a coordinated disruption of mitochondrial metabolism and chromatin-linked transcriptional control mediated, in part, through CHD8-associated networks.

Although CHD8 is best characterized for its role in early neurodevelopment, the persistence of its downstream network perturbations in adult cortico-striatal tissue suggests a lasting transcriptional vulnerability rather than a purely developmental effect. CHD8 has emerged as a leading OCD risk gene, as identified by us and others,^36,37,64 and its regulatory targets have also been implicated in bipolar disorder,⁶⁵ pointing to molecular convergence across compulsive and affective phenotypes rather than disorder-specific mechanisms. This interpretation is further supported by the elevated risk of comorbid or subsequent bipolar disorder among individuals with OCD.⁶⁶ The downregulation of TUSC2, a top AN candidate identified through transcriptomic imputation studies,⁵¹ adds an additional layer of genetic and transcriptomic overlap. Collectively, these results delineate an interconnected molecular landscape in which disrupted energy metabolism, chromatin regulation, and shared genetic susceptibility intersect in the pathophysiology of ED and OCD.

Network analysis also highlights the importance of immune dysregulation and glial dysfunction in ED and OCD. The downregulation of module Md2 in the DLPFC, enriched in genes associated with innate immune responses, aligns with long-standing evidence suggesting the involvement of immune dysregulation in OCD.^19,20 Similarly, downregulation of Md3, enriched for genes involved in myelination and glial cell development, supports the hypothesis that neuroplasticity disruptions are central to ED and OCD. Together with neuronal energy and inhibitory deficits, these glial and immune alterations point to a multi-cellular basis for network instability in ED and OCD.

The TWAS analyses added a genetic dimension to our transcriptomic findings by integrating GWAS from ED⁶⁷ and OCD,⁶⁸ enabling prioritization of disease-associated genes and inference of directionality. Among the most reproducible signals, WDR6 emerged as a top cross-disorder hit, showing consistent upregulation and strong effect sizes across analyses (Figure 5). WDR6, a conserved gene involved in protein complex assembly and insulin-related pathways, has been previously implicated in ED,^39,50,51 and its convergence across TWAS, conditional analysis, and fine-mapping supports a potential causal role in disease biology. More broadly, TWAS revealed a substantial burden of associations in OCD, with over 100 genes reaching significance and several replicating across brain regions. Beyond WDR6, multiple additional genes overlapped with ED, including NCKIPSD and DALRD3, highlighting shared transcriptional vulnerability. Both disorders independently showed enrichment within a gene-dense region on chromosome 3 (47.6–51.4 Mb), a known AN GWAS locus,^39,50,67 identifying this region as a hub of genetically driven expression in cortico-striatal circuits. Within this locus, a core set of genes (WDR6, DALRD3, NCKIPSD, SHISA5, and P4HTM) exhibited consistent directionality and strong effects across regions (Table 1), reinforcing their role as cross-disorder candidates. Consistent with this, analyses across 13 GTEx brain regions revealed reproducible enrichment for DALRD3 and NCKIPSD, particularly in cortical and limbic structures. Although statistical significance varied across tissues, these genes consistently ranked among the strongest associations. Moreover, the limited overlap between TWAS and DEG results suggests that genetically regulated expression captures upstream, heritable mechanisms distinct from state-dependent transcriptional changes in postmortem tissue. Together, these findings support a model in which shared genetic architecture, centered in part on chromosome 3, drives convergent molecular phenotypes across ED and OCD, linking genetic risk to metabolic and neuronal dysregulation within cortico-striatal networks. Notably, the independent convergence of two disorders with distinct genome-wide architectures onto the same gene-dense chromosomal interval, at ~10-fold enrichment above background in both DLPFC and caudate, is unlikely to reflect chance and instead points to this locus as a shared regulatory hub whose transcriptional output is selectively vulnerable across compulsive-spectrum conditions. Together, these findings converge on a model in which inherited transcriptional vulnerability, mediated through CHD8-linked chromatin regulation and concentrated at a shared chromosome 3 risk locus, manifests as coordinated GABAergic and bioenergetic failure within cortico-striatal circuits, defining a common molecular substrate for the compulsive phenotypes that characterize both ED and OCD and opening a path toward transdiagnostic intervention.

Limitations of the study

This study has several limitations inherent to postmortem brain research. First, although we accounted for multiple clinical and technical factors (substance use, opioid exposure, and psychiatric comorbidities) and quantified their contributions to expression variation using variance partitioning, residual confounding cannot be excluded. Detailed lifetime medication histories were not uniformly available across donors, limiting our ability to directly model medication effects. While sensitivity analyses demonstrated highly concordant results across covariate-adjusted models, unmeasured exposures may still influence observed gene expression patterns. Importantly, the consistency of findings across cohorts, diagnostic groups, and analytical approaches, together with the absence of enrichment for psychotropic drug signatures, supports a predominant disease-related signature and argues against widespread drug-driven confounding. Second, sample size and clinical heterogeneity remain important considerations. Although the ED and OCD cohorts are modest relative to large-scale studies of other neuropsychiatric disorders, they represent, to our knowledge, the largest postmortem brain transcriptomic dataset assembled for these conditions, reflecting the rarity of such cases in medical-examiner-based collections. Nevertheless, heterogeneity within diagnostic categories, particularly across ED subtypes, may limit the detection of subtype-specific effects and reduce generalizability of the findings. Third, our integration of TWAS provides insight into genetically regulated expression but carries inherent limitations. TWAS signals are influenced by linkage disequilibrium structure, are largely restricted to cis-eQTL effects, and depend on reference panels that incompletely capture tissue specificity and ancestral diversity. As such, TWAS associations should be interpreted as putative links between genetically regulated expression and disease risk rather than evidence of direct causality, and these constraints may obscure causal signals or produce false positives. To address this and to further refine candidate genes, future studies should integrate statistical fine-mapping and colocalization approaches with orthogonal datasets, including cell-type-resolved eQTL resources and single-nucleus transcriptomic profiles. Functional validation in disease-relevant cellular systems, such as CRISPR-mediated perturbation or RNA-based modulation in human induced pluripotent stem cell (hiPSC)-derived neuronal models, coupled with transcriptomic and physiological readouts (e.g., synaptic activity, electrophysiology, or calcium imaging), will be required to establish causality and define the underlying mechanisms.

RESOURCE AVAILABILITY

Lead contact

Requests for further information should be directed to and will be fulfilled by the lead contact, Michael S. Breen (michael.breen@mssm.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

All original RNA sequencing data are publicly through the National Center for Biotechnology Information Gene Expression Omnibus https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE262138 (GEO: GSE262138).
All code used for data processing, analysis, and figure generation is publicly available through GitHub: https://github.com/BreenMS/OCD_ED/.
Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Postmortem human brain tissue

Human postmortem brain tissue was obtained from the Lieber Institute for Brain Development (LIBD) Human Brain and Tissue Repository, following rigorous ethical protocols across multiple acquisition sites. Tissue was collected at autopsy from the Office of the Chief Medical Examiner of the State of Maryland (Maryland Department of Health IRB protocol #12–24); the Departments of Pathology at Western Michigan University Homer Stryker MD School of Medicine and the University of North Dakota School of Medicine and Health Sciences; Gift of Life Michigan; and the County of Santa Clara Medical Examiner-Coroner Office (all under WCG IRB protocol #20111080). Additional samples were acquired through the National Institute of Mental Health Intramural Research Program (NIH protocol #90-M-0142) via material transfer agreement with LIBD. Informed consent was obtained via audiotaped and witnessed documentation from the legal next-of-kin for all cases.

Study participants and sample allocation

The discovery cohort comprised 169 donors, allocated to three diagnostic groups: neurotypical controls (N = 86), individuals with eating disorders (ED; N = 57), and individuals with obsessive-compulsive disorder (OCD; N = 27). Donors in the discovery cohort contributed paired DLPFC and caudate tissue samples. The replication cohort comprised 111 donors (post-QC) contributing caudate tissue only, allocated to three groups: neurotypical controls (N = 36), ED cases (N = 38), and non-ED major depressive disorder psychiatric controls (MDD; N = 37). Of the 38 post-QC ED cases in the replication cohort, 34 were donors also present in the discovery cohort whose caudate RNA was re-sequenced; the remaining 4 were donors unique to the replication cohort. Equivalent breakdowns for controls and MDD cases are provided in Table S1A.

Demographic characteristics

In the discovery cohort, donors spanned a range of ages (mean 41.5 ± 14.8 years) and included both male and female donors (103 female, 66 male) with diverse ancestry (European, African American, and other/unknown; Tables S1A–S1D). All donors in the replication cohort were female and of European ancestry (mean age 47.3 ± 17.4 years; Tables S1A–S1D).

Sex and gender reporting

Biological sex was recorded for all donors based on clinical and autopsy records and is reported throughout (Tables S1A–S1D). Gender identity data were not systematically available for this postmortem cohort. To assess the influence of biological sex on gene expression results, sensitivity analyses were conducted in the discovery cohort; no transcriptome-wide significant diagnosis × sex interactions were detected (Table S2). In the replication cohort, all donors were female, which limits assessment of sex effects in that dataset and may reduce generalizability of replication findings to male donors.

Ancestry, race, and ethnicity

Self-reported or proxy-reported ancestry, race, and ethnicity were recorded where available and are summarized in Tables S1A–S1D. Ancestry was included as a covariate in all differential expression models. The replication cohort was restricted to donors of European ancestry, which limits the generalizability of replication findings across ancestral groups.

Clinical characterization

A 39-item LIBD Autopsy telephone screening was conducted at the time of donation with the legal next-of-kin to gather data on the donor’s medical, psychiatric, social, and substance use history. Clinical diagnoses were retrospectively established for every donor using autopsy reports, toxicology testing, forensic investigations, neuropathological examinations, telephone screenings, psychiatric and substance use treatment records, and supplemental family informant interviews. These data were compiled into detailed psychiatric narrative summaries, independently reviewed by two board-certified psychiatrists to establish lifetime psychiatric and substance use diagnoses according to DSM-5 criteria. Non-psychiatric control donors had no lifetime history of psychiatric or substance use disorders and were excluded if they presented with evidence of acute drug or alcohol intoxication at the time of death. All donors were excluded if neuropathological abnormalities, including cerebrovascular accidents, neuritic pathology, or structural brain abnormalities were identified.

Toxicology

Toxicology testing, conducted by medical examiners as part of autopsy and forensic investigations, included assessments for drugs of abuse (ethanol/volatiles, cocaine/metabolites, amphetamines, opiates). Additional postmortem toxicology testing was performed by National Medical Services (www.nmslabs.com), including assessments for nicotine/cotinine and cannabinoids for all cases, and for antidepressants, mood stabilizers, and antipsychotics in psychiatric cases, using blood and/or cerebellar tissue samples. Detailed lifetime medication histories were not uniformly available across donors; however, sensitivity analyses demonstrated concordant results across covariate-adjusted models, and no significant enrichment for psychotropic drug signatures was detected in ED + OCD transcriptional profiles (Table S3).

Clinical and technical metadata

Comprehensive metadata were available for all donors, including age, sex, postmortem interval (PMI), RNA integrity number (RIN), cause of death, comorbid psychiatric diagnoses (e.g., MDD), and substance use history. Technical measures related to RNA quality and sequencing (intronic and rRNA content, library metrics) were also recorded and are summarized in Tables S1A–S1D.

METHOD DETAILS

RNA-sequencing data generation

Discovery cohort:

We assembled a discovery RNA-seq cohort, comprising measurements of the DLPFC and caudate from 86 controls, 57 ED cases, and 27 OCD cases. The ED cases were comprised of eight anorexia nervosa restricting type (AN-R), 25 with bingeing and purging behaviors (AN-BP, n = 14; bulimia nervosa, n = 11), and the remaining were unspecified ED. Individuals in this cohort exhibited diverse biological sex (103 female, 66 male), age (41.5 ± 14.8 years) and ethnicity (Tables S1A–S1D). Sample size was determined by availability of high-quality postmortem tissue meeting inclusion criteria (RIN ≥5, PMI ≤80 h) rather than by formal power calculation, reflecting the rarity of these cases in medical examiner-based collections. Diagnostic group allocation was based on retrospective clinical characterization as described in Experimental Model and Study Participant Details; no randomization or prospective stratification was applied. Total RNA was extracted from all tissue samples using AllPrep DNA/RNA Mini Kit (Qiagen Cat No./ID: 80204) and displayed high RIN values (7.3 ± 1.3). Paired-end strand-specific sequencing libraries were prepared from 300ng total RNA using the TruSeq Stranded Total RNA Library Preparation kit with Ribo-Zero Gold ribosomal RNA depletion which removes rRNA and mtRNA. The libraries were then sequenced on an Illumina HiSeq 3000 at the Promega sequencing facility, producing a mean library depth of 17,299,335 reads across 200-bp paired-end reads per sample. All experimental procedures were conducted using de-identified sample identifiers, and experimenters were blinded to diagnostic status.

Replication cohort:

The replication cohort was derived exclusively from caudate tissue and included 41 ED cases (37 donors also in the discovery cohort, contributing re-sequenced technical replicates, and 4 donors unique to the replication cohort), 42 neurotypical controls (30 donors also in the discovery cohort, contributing re-sequenced technical replicates, and 12 donors unique to the replication cohort), and 44 MDD non-ED neuropsychiatric controls (all unique to the replication cohort). All samples in the replication cohort were obtained from female donors of European ancestry, with a mean age of 47.3 ± 17.4 years (Tables S1A–S1D). RNA extraction and sequencing were performed using protocols identical to those applied in the discovery cohort. RNA quality was high (RIN, 8.4 ± 0.9), and libraries were sequenced using paired-end 200-bp reads to an average depth of 26 million reads per sample at the LIBD Sequencing Facility. The replication cohort was assembled independently of the discovery cohort; all experimental procedures were conducted using de-identified sample identifiers and experimenters were blinded to diagnostic status.

Donors present in both cohorts contributed re-sequenced caudate RNA (technical replicates), enabling quantification of sequencing reproducibility and technical variability across library preparation runs. Donors unique to the replication cohort were not included in the discovery cohort and provide independent validation of the ED caudate signature. The full composition of each cohort, the overlap between cohorts, and the number of re-sequenced samples contributed per donor are detailed in Tables S1A–S1D. Accordingly, analyses accounted for donor-level non-independence, and this cohort was used to assess reproducibility as well as consistency of effect direction. Inclusion of MDD as a psychiatric control further refines diagnostic specificity.

QUANTIFICATION AND STATISTICAL ANALYSIS

RNA-sequencing data processing and outlier removal

RNA-sequencing data were subjected to a rigorous processing protocol using our established pipeline (https://github.com/CommonMindConsortium/RAPiD-nf/). Initially, raw RNA sequences were aligned to the hg38 reference genome from Ensembl via the STAR algorithm.⁶⁹ Quality control checks, including assessments of GC content, duplication rates, gene body coverage, and library complexity were carried out using RSeQ⁸¹ and Picard tools (https://broadinstitute.github.io/picard/). These procedures also included unsupervised clustering and marking of duplicate reads. Gene expression levels were quantified with featureCounts⁸² and the resulting expression matrix was curated to retain genes present in at least one-third of our sample set. This matrix was then normalized through the VOOM method.⁷²

Discovery cohort:

A final normalized data was compiled encompassing 17,218 genes. Principal component analysis (PCA) was applied to identify and exclude gene expression outliers beyond a 95% confidence interval from the grand mean. Although no significant outliers in gene expression were detected, two anomalous samples, one from the caudate and one from the DLPFC, were discovered as misaligned with their respective anatomical regions during PCA clustering. To maintain the robustness of the data, these samples were excluded from all further analyses, resulting in 169 samples per region for downstream analyses.

Replication cohort:

A final normalized data was compiled encompassing 18,601 genes. Similarly, PCA was applied to identify and exclude gene expression outliers beyond a 95% confidence interval from the grand mean. A total of 16 samples were identified as outliers according to this metric (11 from donors unique to the replication cohort and 5 re-sequenced samples from donors also in the discovery cohort), and these samples were excluded from all further analyses. This stringent approach ensures the robustness of the dataset and enhances the reliability of downstream analyses. The resulting final dataset included 38 ED cases (34 re-sequenced samples from donors also in the discovery cohort, and 4 donors unique to the replication cohort), 36 neurotypical controls (26 re-sequenced samples from donors also in the discovery cohort, and 10 donors unique to the replication cohort), and 37 MDD non-ED neuropsychiatric controls.

Cellular deconvolution of bulk tissues

We applied a cellular deconvolution algorithm to our bulk RNA-seq data to discern shifts in cellular composition, using the non-negative least squares (NNLS) approach from the MIND R package⁷⁰ with the Darmanis et al., 2015⁸³ signature matrix. This matrix comprises six primary cell types: mature excitatory neurons, mature inhibitory neurons, astrocytes, oligodendrocytes, endothelial cells, and microglia. The NNLS method, executed via the est_frac function in the limma package,⁷¹ was employed on log2-transformed counts per million (CPM) data. Our analysis targeted these principal cell types to minimize variability and to reflect cell type distributions corroborated by preceding research by us and others.

Computing gene expression variance explained by technical and biological factors

We utilized the variancePartition R package⁷⁴ to construct a linear mixed model (LMM) to guide downstream differential gene expression analyses. Available clinical and technical covariates, including age, sex, postmortem interval (PMI), RNA integrity number (RIN), cause of death, comorbid major depressive disorder (MDD), substance abuse (SA), opioid exposure, estimated neuronal cell type proportions, percentage of intronic and rRNA reads, library depth, clinical diagnosis, and ethnicity were incorporated where appropriate. The LMM quantified the contribution of these technical, biological, and clinical factors to gene expression variability. For each gene within each brain region, variance was partitioned across these covariates, with unexplained variance attributed to residual effects. This framework enabled identification of the primary drivers of transcriptomic variation and informed covariate selection for downstream differential expression models.

Differential gene expression

Differential gene expression analysis was conducted using the limma package.⁷¹ We first performed differential expression analysis between the DLPFC and caudate, given their distinct gene profiles revealed through PCA. For this analysis, we used a VOOM normalized matrix across all samples which were not identified as outliers. Here, differential gene expression was adjusted for the possible influence of clinical diagnosis, percentage of intronic reads and rRNA reads, sex, age, and RIN. Donor as a repeated measure was controlled for using the duplicateCorrelation function in limma when jointly modeling the DLPFC and caudate. Given the profound differences in gene expression between these two regions, we next performed differential expression analysis within each region to reveal ED and OCD effects. For these analyses, we used two gene matrices which were VOOM normalized within each region. These analyses covaried for the estimated neuronal proportions (see Methods below), percentage of intronic reads and rRNA reads, sex, age, and RIN. Differentially expressed genes (DEGs) for each comparison were considered significant if they passed a Benjamini-Hochberg (BH) adjusted p-value threshold of <0.01. A strict threshold of FDR <1% was used to protect against potential false positive findings.

Sensitivity analyses and qSVA

Within each brain region, we performed sensitivity analyses for the following observed covariates sequentially: body mass index (BMI), comorbid major depressive disorder (MDD) and substance abuse defined as opioid use and/or alcoholism and/or polysubstance abuse and/or cannabis use. For each sensitivity analysis, which considered a single confounder from the previous list, we adjusted the original model to include each additional covariate, and then compared the original log2 fold-changes for each diagnosis to the further-adjusted diagnosis coefficient. We also applied the qSVA framework for RNA quality correction in differential expression analysis.⁸⁴ This approach aims to reduce the number of false positive genes which may be impacted by RNA quality differences. Quality surrogate variables (qSVs) were estimated using the qsva() function implemented in the sva package.⁷³ The pairwise Pearson’s correlation between five qSVs, known covariates were investigated using the cor_mat() R function. For comparative case-control analyses in the discovery cohort, within the DLPFC and caudate, we incorporated 15 and 14 qSVs as covariates, respectively.

Drug perturbation enrichment analysis

To assess whether observed transcriptional changes may reflect medication exposure, we performed enrichment analysis against drug-induced gene expression signatures from the Drug Signatures Database.⁸⁵ Gene sets were obtained in GMT format from the Enrichr⁸⁶ resource and filtered to include CNS-active compounds based on curated drug names and mechanism-of-action keywords (e.g., antidepressants, antipsychotics, mood stabilizers, benzodiazepines, and stimulants). Differentially expressed genes (DEGs; FDR <0.05) were mapped from Ensembl identifiers to gene symbols using org.Hs.e.g.,.db, and separate gene sets were defined for up- and downregulated genes. Over-representation analysis was performed using a hypergeometric test to assess overlap between DEG sets and drug-associated gene signatures, with Benjamini-Hochberg correction applied to control the false discovery rate. Analyses were performed independently for DLPFC and caudate.

Differential transcript usage analysis

To investigate transcript-level regulatory changes, we performed differential transcript usage (DTU) analysis using the DRIMSeq (v.1.34.0).⁸⁷ Transcript abundance was quantified using Kallisto⁸⁸ against the Gencode v30 human transcriptome. Gene and transcript annotations were derived from the Gencode v30 GTF file and converted into a transcript database (TxDb) object using the GenomicFeatures R package (v1.58.0). The transcript-to-gene mapping was extracted from the TxDb using the AnnotationDbi R package (v1.68.0). Transcript quantifications from all samples were then imported into R using the tximport (v1.34.0) R package, specifying transcript-level estimates (txIn = TRUE, txOut = TRUE, type = ‘‘kallisto’’). Transcript counts were scaled from TPM values using the countsFromAbundance = ‘‘dtuScaledTPM’’ parameter to account for transcript length and sequencing depth. Prior to DTU analysis, transcripts with zero counts across all samples were removed. Transcript-level count data were reformatted into a gene-transcript count matrix and combined with sample metadata using dmDSdataFromMatrix to construct a dmDSdata object in the DRIMSeq. Genes and transcripts were filtered using dmFilter() with the following default thresholds: genes and transcripts expressed in at least five samples, a minimum of 10 counts per gene or transcript, and transcript proportions ≥1% in at least three samples. Filtered data were modeled using a design matrix that included experimental group, estimated neuronal proportion, and sex as covariates (~0 + condition + neuron.prop + sex). Dispersion estimates were obtained with dmPrecision(), and model fitting was performed with dmFit(). DTU was tested using dmTest(), specifying contrasts between ED vs. Control, OCD vs. Control, and ED + OCD vs. control within each brain region. Gene- and transcript-level p-values from DRIMSeq were adjusted for multiple testing using the Benjamini-Hochberg procedure, and transcripts or genes with FDR<1% were considered significant.

Weighted gene co-expression network analysis

Weighted gene co-expression network analysis (WGCNA)⁸⁹ was used to build two signed co-expression networks using a total of 18,486 genes in the DLPFC and 18,472 genes in the caudate. VOOM normalized gene expression data were used as input. To construct each network, the absolute values of Pearson correlation coefficients were calculated for all possible gene pairs in the DLPFC and caudate, respectively, and resulting values were transformed with an exponential weight (β) so that the final matrices followed an approximate scale-free topology (R²). Thus, for each network we only considered powers of β that lead to a network satisfying scale-free topology (i.e., R² > 0.80), so the mean connectivity is high and the network contains enough information for module detection. A β of 5 was used for the DLPFC and a β of 9 was used for the caudate. The dynamic tree-cut algorithm was used to detect network modules with a minimum module size set to 50 and cut tree height set to 0.999.

After network construction, two follow-up analyses were performed: (1) First, a series of module preservation analyses sought to determine whether underlying gene co-regulatory patterns in the DLPFC were preserved or disrupted when compared to the caudate, and vice versa. For these analyses, module preservation was assessed using a permutation-based preservation statistic, Z_summary, implemented within WGCNA with 500 random permutations of the data.⁷⁵ Z_summary takes into account the overlap in module membership as well as the density and connectivity patterns of genes within modules. A Z_summary score <2 indicates no evidence of preservation, 2<Z_summary<10 implies weak preservation and Z_summary >10 suggests strong preservation. (2) Second, the identified co-expression modules were inspected for association to ED, OCD and ED + OCD as well as all recorded covariates within each region independently. To do so, singular value decomposition of each module expression matrix was performed and the resulting module eigengene (ME), equivalent to the first principal component, was used to represent the overall expression profiles for each module per sample. Modules were evaluated both quantitatively and qualitatively for expression patterns significantly associated with a clinical diagnosis. For quantitative analyses, we computed module significance scores using ME values and clinical diagnoses were treated as main outcomes using a moderated t test through limma. These analyses were conducted to match the modeling and covariate adjustments performed for differential gene expression (described above). For qualitative analyses, we generated gene dendrogram trees for each anatomical region with color bars, indicating gene-to-trait associations. Candidate modules were subjected to protein-protein interactions using the STRING database.⁹⁰ STRING implements a scoring scheme to report the confidence level for each direct protein-protein interaction (low confidence: <0.4; medium: 0.4–0.7; high: >0.7). We used a combined STRING score of >0.4. Hub genes within the protein-protein interaction network is defined as genes with the highest degree of network connections.

Gene ontology and enrichment analysis

Correlation adjusted mean rank (CAMERA) gene set enrichment⁹¹ was performed using the resulting sets of differential gene expression summary statistics between controls with ED, OCD and ED + OCD. We used CAMERA to perform a competitive gene set rank test to assess whether the genes in each gene set were highly ranked in terms of differential gene expression relative to genes that are not in the gene set. For example, CAMERA first ranks gene expression differences in ED DLPFC tissues relative to controls. Next, CAMERA tests whether the user-defined gene sets are over-represented toward the extreme ends of this ranked list. After adjusting the variance of the resulting gene set test statistic by a variance inflation factor that depends on the gene-wise correlation (which we set to default parameters, 0.01) and the size of the set, a p-value is returned and adjusted for multiple testing. We used this function to test for enrichment of gene sets derived from Gene Ontology (GO) biological process, molecular factors and genes known to be dynamically regulated throughout human brain prenatal and postnatal brain development. To annotate co-expression modules, genes for each candidate module were used as input for ToppGene⁷⁸ to test for over-representation for GO: Molecular Functions, GO: Biological Processes, and GO: Cellular Components. SynGo⁷⁹ annotation was also used to describe potential enrichment for synaptic terms.

Secondary analyses of existing OCD transcriptomic datasets

In our secondary analyses, we interrogated 42 raw fastq RNA-sequencing files sourced from Lisboa et al., 2019,¹⁹ comprising data from individuals with OCD (N = 6) and controls (N = 8) sampled across three specific brain regions: the putamen, nucleus accumbens, and caudate. Rigorous data processing and quality control measures were employed, ensuring normalization and testing for differential gene expression were consistent with the methodologies detailed above to reduce technical variance. Adjustments were made in the differential expression analysis to account for potential confounding factors of estimated proportions of neuronal populations, intronic read coverage, sex, and biological age. Complementing this, we incorporated differential gene expression summary statistics from Piantadosi et al., 2021,²¹ involving a comparison between individuals with OCD (N = 7) and controls (N = 8) that focused on the caudate and accumbens. Due to the unavailability of raw RNA-sequencing data, we relied on the summary statistics provided. Lastly, our analysis leveraged differential gene expression data from Jaffe et al., 2014,²² derived via Illumina HumanHT-12 v3 microarrays from the prefrontal cortex of individuals with eating disorders (ED) (N = 15), OCD (N = 16), and a control cohort (N = 102).

A meta-analysis was conducted utilizing the restricted maximum likelihood (REML) method from the metafor R package,⁷⁶ applying the log 2-fold-change alongside standard errors obtained from three distinct OCD comparative transcriptome-wide RNA-sequencing datasets focused on the caudate. These datasets included the current investigation with 27 OCD cases, supplemented by data from Lisboa et al., 2019 (comprising 6 OCD cases), and Piantadosi et al., 2021 (encompassing 7 OCD cases). The analysis was inclusive of genes that were consistently detected across all participant groups, totaling 13,004 genes. To mitigate the risk of false discoveries, the Benjamini-Hochberg (BH) procedure was implemented to adjust the false discovery rate (FDR). Genes were subsequently classified as significantly associated with OCD if they met the BH-adjusted FDR threshold of p < 0.01.

Transcriptome-wide association study (TWAS)

TWAS is a genetic approach that enhances traditional differential expression analyses by integrating estimates of genetically regulated mRNA expression with genetic variant effects on a given phenotype or disorder. Unlike traditional differential expression analyses, which directly measure mRNA levels in cases and controls, TWAS leverages genetically predicted expression by modeling how genetic variants influence mRNA expression. This enables the identification of genes whose predicted expression is associated with disease risk, assigning a directionality (e.g., upregulation or downregulation) based on the estimated genetic effect. In this study, we applied TWAS to prioritize genes potentially involved in ED and OCD. The GWAS summary statistics for both disorders were converted from hg19 to hg38 genome coordinates to ensure consistency. We harmonized SNP names and coordinates across our sample datasets and the GWAS summary statistics. VOOM normalized gene expression matrices were used as input, including18,486 genes from the DLPFC discovery cohort, 18,472 genes from the caudate discovery cohort, and 18,601 genes from the caudate replication cohort. Using the FUSION TWAS framework⁷⁷ (http://gusevlab.org/projects/fusion/), we computed gene-level feature weights for the caudate region and applied these weights to the GWAS summary statistics for both ED⁶⁷ (16,992 AN cases and 55,525 controls) and OCD⁶⁸ (53,660 OCD cases and 2,044,417 controls). This allowed us to calculate functional-GWAS association statistics, effectively linking genetically regulated mRNA expression with disease risk. The FUSION method was chosen for its ability to estimate the association between gene expression and phenotype by leveraging cis heritability of gene expression. This approach is particularly useful when direct measurements of expression are unavailable or incomplete, as it infers the contribution of genetically regulated expression to disease risk.

Statistical significance thresholds and asterisk definitions

Throughout all analyses, differentially expressed genes were considered significant at a Benjamini–Hochberg (BH) FDR-adjusted p-value threshold of <0.01 (FDR <1%) unless otherwise stated. In figure panels, asterisks denote the following significance thresholds: *FDR <0.05; **FDR <0.01 (FDR <1%). In Figure 4C, asterisks reflect FDR-corrected moderated t-tests (limma-voom) assessing module eigengene association with diagnostic status (*FDR <0.05). In Figure 5F, asterisks reflect FDR significance thresholds from FUSION TWAS analyses (*FDR <5%; **FDR <1%). In Figure 2D, asterisks (*) indicate FDR <1% from competitive gene-set enrichment tests (CAMERA). All FDR corrections were performed using the Benjamini–Hochberg procedure unless otherwise noted.

Supplementary Material

NIHMS2190757-supplement-1.pdf^{(7MB, pdf)}

NIHMS2190757-supplement-3.xlsx^{(204.5KB, xlsx)}

NIHMS2190757-supplement-4.xlsx^{(9.8MB, xlsx)}

NIHMS2190757-supplement-5.xlsx^{(14.2KB, xlsx)}

NIHMS2190757-supplement-6.xlsx^{(2.5MB, xlsx)}

NIHMS2190757-supplement-7.xlsx^{(3.3MB, xlsx)}

NIHMS2190757-supplement-8.xlsx^{(1.2MB, xlsx)}

NIHMS2190757-supplement-9.xlsx^{(5MB, xlsx)}

NIHMS2190757-supplement-10.xlsx^{(1.6MB, xlsx)}

SUPPLEMENTAL INFORMATION

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2026.117514.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER

Biological samples

See Table S1 for a complete list of all human samples used in the current study	N/A	N/A

Deposited data

Raw RNA-seq data - DLPFC & Caudate (ED & OCD)	NCBI GEO	GSE262138
Raw RNA-seq data - Caudate (OCD, Lisboa et al. 2019)	NCBI SRA	PRJNA421175
Summary Files, RNA-seq data - Caudate (OCD, Piantadosi et al., 2021)	Piantadosi et al.²⁰ Summary Files	33723209
Summary Files, Microarray - DLPFC (ED & OCD, Jaffe et al. 2014)	Jaffe et al.¹⁴ Summary Files	25180571

Software and algorithms

Code used to process data	This paper	https://github.com/BreenMS/OCD_ED
STAR v2.7.3	Dobin et al.⁶⁹	https://github.com/alexdobin/STAR
Picard v2.22.3	Broad Institute	https://github.com/broadinstitute/picard/
bMIND v0.3.2	Wang et al.⁷⁰	https://github.com/randel/MIND
edgeR v3.28.1	Ritchie et al.⁷¹	https://bioconductor.org/packages/edgeR
limma v3.36.3	Ritchie et al.⁷¹	https://bioconductor.org/packages/limma
voom (limma)	Law et al.⁷²	https://bioconductor.org/packages/limma
duplicateCorrelation (limma)	Ritchie et al.⁷¹	https://bioconductor.org/packages/limma
sva v3.34.0	Leek et al.⁷³	https://bioconductor.org/packages/sva
variancePartition v1.16.1	Hoffman & Schadt⁷⁴	https://bioconductor.org/packages/variancePartition
WGCNA v1.70–3	Langfelder & Horvath⁷⁵	https://cran.r-project.org/package=WGCNA
metafor v3.0–2	Viechtbauer⁷⁶	https://cran.r-project.org/package=metafor
FUSION TWAS framework	Gusev et al.⁷⁷	http://gusevlab.org/projects/fusion/
ToppGene Suite	Chen et al.⁷⁸	https://toppgene.cchmc.org/enrichment.jsp
SynGO	Koopmans et al.⁷⁹	https://www.syngoportal.org
Wakefield approximate Bayes factor analysis (custom implementation)	Wakefield⁸⁰	https://doi.org/10.1002/gepi.20359
R statistical computing environment v4.X.X	R Core Team	https://www.r-project.org

Open in a new tab

Highlights.

ED and OCD converge on shared cortico-striatal transcriptional programs
CHD8 networks, mitochondrial function, and inhibitory signaling disrupted
Coordinated downregulation links neuronal energy and synaptic function
TWAS implicates shared genetic drivers at a chromosome 3 risk locus

ACKNOWLEDGMENTS

We sincerely thank the families of the brain donors for their invaluable contributions to this research. Without their generosity and willingness to support scientific discovery, this work would not have been possible. We also acknowledge the Lieber Institute for Brain Development Human Brain Repository and collaborating institutions for providing access to these precious samples. Their dedication to advancing our understanding of brain disorders is deeply appreciated.

Additionally, we extend our gratitude to the physicians and staff of the Office of the Medical Examiner, Kalamazoo, Michigan; the University of North Dakota Forensic Pathology Center, Grand Forks, North Dakota; the Office of the Medical Examiner-Coroner, County of Santa Clara, San Jose, California; and the Office of the Chief Medical Examiner for the State of Maryland, Baltimore, Maryland, for their time and efforts to advance tissue donation for neuroscientific research into neuropsychiatric disorders. This work was supported by R01 MH124679 (D.E.G.), the Beatrice and Samuel A. Seaver Foundation (M.S.B. and D.E.G.), and the Klarman Family Foundation (J.E.K.). M.S.B. is a Seaver Foundation Faculty Scholar. D.E.G. is the recipient of the IOCDF Innovator Award. J.E.K. is the recipient of a KFF Eating Disorders Research Grants Program award. The supporting organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Footnotes

DECLARATION OF INTERESTS

The authors declare no competing interests.

DECLARATION OF GENERATIVE AI AND AI-ASSISTED TECHNOLOGIES IN THE WRITING PROCESS

During the preparation of this work, the author(s) used Claude to streamline the presentation of manuscript data. After using this tool or service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.

REFERENCES

1.Altman SE, and Shankman SA (2009). What is the association between obsessive-compulsive disorder and eating disorders? Clin. Psychol. Rev. 29, 638–646. [DOI] [PubMed] [Google Scholar]
2.Farstad SM, McGeown LM, and von Ranson KM (2016). Eating disorders and personality, 2004–2016: a systematic review and meta-analysis. Clin. Psychol. Rev. 46, 91–105. [Google Scholar]
3.Swinbourne JM, and Touyz SW (2007). The co-morbidity of eating disorders and anxiety disorders: a review. Eur. Eat. Disord. Rev. 15, 253–274. [DOI] [PubMed] [Google Scholar]
4.Kaye WH, Bulik CM, Thornton L, Barbarich N, and Masters K (2004). Comorbidity of anxiety disorders with anorexia and bulimia nervosa. Am. J. Psychiatry 161, 2215–2221. [DOI] [PubMed] [Google Scholar]
5.Godart NT, Flament MF, Curt F, Perdereau F, Lang F, Venisse JL, Halfon O, Bizouard P, Loas G, Corcos M, et al. (2003). Anxiety disorders in subjects seeking treatment for eating disorders: a DSM-IV controlled study. Psychiatry Res. 117, 245–258. [DOI] [PubMed] [Google Scholar]
6.Drakes DH, Fawcett EJ, Rose JP, Carter-Major JC, and Fawcett JM (2021). Comorbid obsessive-compulsive disorder in individuals with eating disorders: an epidemiological meta-analysis. J. Psychiatr. Res. 141, 176–191. [DOI] [PubMed] [Google Scholar]
7.Lilenfeld LR, Kaye WH, Greeno CG, Merikangas KR, Plotnicov K, Pollice C, Rao R, Strober M, Bulik CM, and Nagy L (1998). A controlled family study of anorexia nervosa and bulimia nervosa: psychiatric disorders in first-degree relatives and effects of proband comorbidity. Arch. Gen. Psychiatry 55, 603–610. [DOI] [PubMed] [Google Scholar]
8.Strober M, Freeman R, Lampert C, and Diamond J (2007). The association of anxiety disorders and obsessive compulsive personality disorder with anorexia nervosa: evidence from a family study with discussion of nosological and neurodevelopmental implications. Int. J. Eat. Disord. 40, S46–S51. [DOI] [PubMed] [Google Scholar]
9.Meier SM, Bulik CM, Thornton LM, Mattheisen M, Mortensen PB, and Petersen L (2015). Diagnosed anxiety disorders and the risk of subsequent anorexia nervosa: a Danish population register study. Eur. Eat. Disord. Rev. 23, 524–530. [DOI] [PubMed] [Google Scholar]
10.Cederlöf M, Thornton LM, Baker J, Lichtenstein P, Larsson H, Rück C, Bulik CM, and Mataix-Cols D (2015). Etiological overlap between obsessive-compulsive disorder and anorexia nervosa: a longitudinal cohort, multigenerational family and twin study. World Psychiatry 14, 333–338. [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Boraska V, Franklin CS, Floyd JAB, Thornton LM, Huckins LM, Southam L, Rayner NW, Tachmazidou I, Klump KL, Treasure J, et al. (2014). A genome-wide association study of anorexia nervosa. Mol. Psychiatry 19, 1085–1094. [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Wang K, Zhang H, Bloss CS, Duvvuri V, Kaye W, Schork NJ, Berrettini W, and Hakonarson H (2011). A genome-wide association study on common SNPs and rare CNVs in anorexia nervosa. Mol. Psychiatry 16, 949–959. [DOI] [PubMed] [Google Scholar]
13.Duncan L, Yilmaz Z, Gaspar H, Walters R, Goldstein J, Anttila V, Bulik-Sullivan B, Ripke S, Thornton L, Hinney A, et al. (2017). Significant locus and metabolic genetic correlations revealed in genome-wide association study of anorexia nervosa. Am. J. Psychiatr. 174, 850–858. [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Stewart SE, Yu D, Scharf JM, Neale BM, Fagerness JA, Mathews CA, Arnold PD, Evans PD, Gamazon ER, Osiecki L, et al. (2013). Genome-wide association study of obsessive-compulsive disorder. Mol. Psychiatry 18, 788–798. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Mattheisen M, Samuels JF, Wang Y, Greenberg BD, Fyer AJ, McCracken JT, Geller DA, Murphy DL, Knowles JA, Grados MA, et al. (2015). Genome-wide association study in obsessive-compulsive disorder: results from the OCGAS. Mol. Psychiatry 20, 337–344. [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Arnold PD, Askland KD, Barlassina C, Bellodi L, Bienvenu OJ, Black D, Bloch M, Brentani H, Burton CL, Camarena B, and Cappi C (2018). Revealing the complex genetic architecture of obsessive-compulsive disorder using meta-analysis. Mol. Psychiatr. 23, 1181. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Yilmaz Z, Halvorsen M, Bryois J, Yu D, Thornton LM, Zerwas S, Micali N, Moessner R, Burton CL, Zai G, et al. (2020). Examination of the shared genetic basis of anorexia nervosa and obsessive-compulsive disorder. Mol. Psychiatry 25, 2036–2046. [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Tyagi H, Patel R, Rughooputh F, Abrahams H, Watson AJ, and Drummond L (2015). Comparative prevalence of eating disorders in obsessive-compulsive disorder and other anxiety disorders. Psychiatry Journal 2015, 186927. [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Lisboa BCG, Oliveira KC, Tahira AC, Barbosa AR, Feltrin AS, Gouveia G, Lima L, Feio dos Santos AC, Martins DC Jr., Puga RD, et al. (2019). Initial findings of striatum tripartite model in OCD brain samples based on transcriptome analysis. Sci. Rep. 9, 3086. [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Piantadosi SC, Chamberlain BL, Glausier JR, Lewis DA, and Ahmari SE (2021). Lower excitatory synaptic gene expression in orbitofrontal cortex and striatum in an initial study of subjects with obsessive compulsive disorder. Mol. Psychiatry 26, 986–998. [DOI] [PubMed] [Google Scholar]
21.Piantadosi SC, McClain LL, Klei L, Wang J, Chamberlain BL, Springer SA, Lewis DA, Devlin B, and Ahmari SE (2021). Transcriptome alterations are enriched for synapse-associated genes in the striatum of subjects with obsessive-compulsive disorder. Transl. Psychiatry 11, 171. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Jaffe AE, Deep-Soboslay A, Tao R, Hauptman DT, Kaye WH, Arango V, Weinberger DR, Hyde TM, and Kleinman JE (2014). Genetic neuropathology of obsessive psychiatric syndromes. Transl. Psychiatry 4, e432. [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Allen A, King A, and Hollander E (2003). Obsessive-compulsive spectrum disorders. Dialogues Clin. Neurosci. 5, 259–271. [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Simpson HB, Foa EB, Liebowitz MR, Ledley DR, Huppert JD, Cahill S, Vermes D, Schmidt AB, Hembree E, Franklin M, et al. (2008). A randomized, controlled trial of cognitive-behavioral therapy for augmenting pharmacotherapy in obsessive-compulsive disorder. Am. J. Psychiatr. 165, 621–630. [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Foa EB, Liebowitz MR, Kozak MJ, Davies S, Campeas R, Franklin ME, Huppert JD, Kjernisted K, Rowan V, Schmidt AB, et al. (2005). Randomized, placebo-controlled trial of exposure and ritual prevention, clomipramine, and their combination in the treatment of obsessive-compulsive disorder. Am. J. Psychiatr. 162, 151–161. [DOI] [PubMed] [Google Scholar]
26.Bulik CM, Sullivan PF, Carter FA, McIntosh VV, and Joyce PR (1998). The role of exposure with response prevention in the cognitive-behavioural therapy for bulimia nervosa. Psychol. Med. 28, 611–623. [DOI] [PubMed] [Google Scholar]
27.Steinglass J, Albano AM, Simpson HB, Carpenter K, Schebendach J, and Attia E (2012). Fear of food as a treatment target: exposure and response prevention for anorexia nervosa in an open series. Int. J. Eat. Disord. 45, 615–621. [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Cumella EJ, Kally Z, and David Wall A (2007). Treatment responses of inpatient eating disorder women with and without co-occurring obsessive-compulsive disorder. Eat. Disord. 15, 111–124. [DOI] [PubMed] [Google Scholar]
29.O’Neill N, Stein TD, Olayinka OA, Empawi JA, Hu J, Tong T, Zhang X, and Farrer LA (2024). Cognitive resilience to Alzheimer’s disease characterized by cell-type abundance. Alzheimer’s Dement. 20, 6910–6921. [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Jaffe AE, Tao R, Page SC, Maynard KR, Pattie EA, Nguyen CV, Deep-Soboslay A, Bharadwaj R, Young KA, Friedman MJ, et al. (2022). Decoding shared versus divergent transcriptomic signatures across cortico-amygdala circuitry in PTSD and depressive disorders. Am. J. Psychiatr. 179, 673–686. [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Reif A, Nguyen TT, Weißflog L, Jacob CP, Romanos M, Renner TJ, Buttenschon HN, Kittel-Schneider S, Gessner A, Weber H, et al. (2011). DIRAS2 is associated with adult ADHD, related traits, and co-morbid disorders. Neuropsychopharmacology 36, 2318–2327. [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Hillmann B, Zuberer A, Obermeyer L, Erb M, Scheffler K, Nieratschker V, and Ethofer T (2022). ADHD patients with DIRAS2 risk allele need more thalamic activation during emotional face-voice recognition. Psychiatry Res. 308, 114355. [DOI] [PubMed] [Google Scholar]
33.Bhore N, Wang BJ, Chen YW, and Liao YF (2017). Critical roles of dual-specificity phosphatases in neuronal proteostasis and neurological diseases. Int. J. Mol. Sci. 18, 1963. [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Pérez-Sen R, Queipo MJ, Gil-Redondo JC, Ortega F, Gómez-Villafuertes R, Miras-Portugal MT, and Delicado EG (2019). Dual-specificity phosphatase regulation in neurons and glial cells. Int. J. Mol. Sci. 20, 1999. [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Cotney J, Muhle RA, Sanders SJ, Liu L, Willsey AJ, Niu W, Liu W, Klei L, Lei J, Yin J, et al. (2015). The autism-associated chromatin modifier CHD8 regulates other autism risk genes during human neurodevelopment. Nat. Commun. 6, 6404. [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Cappi C, Oliphant ME, Péter Z, Zai G, Conceição do Rosário M, Sullivan CAW, Gupta AR, Hoffman EJ, Virdee M, Olfson E, et al. (2020). De novo damaging DNA coding mutations are associated with obsessive-compulsive disorder and overlap with Tourette’s disorder and autism. Biol. Psychiatry 87, 1035–1044. [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Jung S, Halvorson M, Pedersen N, Natividad Avila M, Niarchou M, Devlin B, Roeder K, Crowley JJ, Buxbaum JD, and Grice DE (2025). Rare coding variation in OCD implicates shared genes with other psychiatric disorders. Preprint at medRxiv. 10.64898/2025.12.22.25342827. [DOI] [Google Scholar]
38.Gandal MJ, Haney JR, Parikshak NN, Leppa V, Ramaswami G, Hartl C, Schork AJ, Appadurai V, Buil A, Werge TM, et al. (2018). Shared molecular neuropathology across major psychiatric disorders parallels polygenic overlap. Science 359, 693–697. [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Adams DM, Reay WR, and Cairns MJ (2023). Multiomic prioritisation of risk genes for anorexia nervosa. Psychol. Med. 53, 6754–6762. [DOI] [PMC free article] [PubMed] [Google Scholar]
40.de Lecea L (2008). Cortistatin—functions in the central nervous system. Mol. Cell. Endocrinol. 286, 88–95. [DOI] [PubMed] [Google Scholar]
41.de LL, del Rio JA, Criado JR, Alcantara S, Morales M, Danielson PE, Henriksen SJ, Soriano E, and Sutcliffe JG (1997). Cortistatin is expressed in a distinct subset of cortical interneurons. J. Neurosci. 17, 5868–5880. [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Mas S, Pagerols M, Gassó P, Ortiz A, Rodriguez N, Morer A, Plana MT, Lafuente A, and Lazaro L (2014). Role of GAD2 and HTR1B genes in early-onset obsessive-compulsive disorder: results from transmission disequilibrium study. Gene Brain Behav. 13, 409–417. [DOI] [PubMed] [Google Scholar]
43.Lennington JB, Coppola G, Kataoka-Sasaki Y, Fernandez TV, Palejev D, Li Y, Huttner A, Pletikos M, Sestan N, Leckman JF, and Vaccarino FM (2016). Transcriptome analysis of the human striatum in Tourette syndrome. Biol. Psychiatry 79, 372–382. [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Lin WJ, Jiang C, Sadahiro M, Bozdagi O, Vulchanova L, Alberini CM, and Salton SR (2015). VGF and its C-terminal peptide TLQP-62 regulate memory formation in hippocampus via a BDNF-TrkB-dependent mechanism. J. Neurosci. 35, 10343–10356. [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Thakker-Varia S, Behnke J, Doobin D, Dalal V, Thakkar K, Khadim F, Wilson E, Palmieri A, Antila H, Rantamaki T, and Alder J (2014). VGF (TLQP-62)-induced neurogenesis targets early phase neural progenitor cells in the adult hippocampus and requires glutamate and BDNF signaling. Stem Cell Res. 12, 762–777. [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Castillo-González J, Ruiz JL, Serrano-Martínez I, Forte-Lago I, Ubago-Rodriguez A, Caro M, Pérez-Gómez JM, Benítez-Troncoso A, Andrés-León E, Sánchez-Navarro M, et al. (2023). Cortistatin deficiency reveals a dysfunctional brain endothelium with impaired gene pathways, exacerbated immune activation, and disrupted barrier integrity. J. Neuroinflammation 20, 226. [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Ahmari SE, Spellman T, Douglass NL, Kheirbek MA, Simpson HB, Deisseroth K, Gordon JA, and Hen R (2013). Repeated cortico-striatal stimulation generates persistent OCD-like behavior. Science 340, 1234–1239. [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Burguière E, Monteiro P, Feng G, and Graybiel AM (2013). Optogenetic stimulation of lateral orbitofronto-striatal pathway suppresses compulsive behaviors. Science 340, 1243–1246. [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Berner LA, Brown TA, Lavender JM, Lopez E, Wierenga CE, and Kaye WH (2019). Neuroendocrinology of reward in anorexia nervosa and bulimia nervosa: beyond leptin and ghrelin. Mol. Cell. Endocrinol. 497, 110320. [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Howard D, Negraes P, Voineskos AN, Kaplan AS, Muotri AR, Duvvuri V, and French L (2020). Molecular neuroanatomy of anorexia nervosa. Sci. Rep. 10, 11411. [DOI] [PMC free article] [PubMed] [Google Scholar]
51.Johnson JS, Cote AC, Dobbyn A, Sloofman LG, Xu J, Cotter L, Charney AW, Birgegård A, Jordan J, Kennedy M, et al. (2023). Mapping anorexia nervosa genes to clinical phenotypes. Psychol. Med. 53, 2619–2633. [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Sirufo MM, Magnanimi LM, Ginaldi L, and De Martinis M (2022). Anorexia nervosa and autoimmune comorbidities: A bidirectional route? CNS Neurosci. Ther. 28, 1921–1929. [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Sirufo MM, Ginaldi L, and De Martinis M (2021). Peripheral vascular abnormalities in anorexia nervosa: A psycho-neuro-immune-metabolic connection. Int. J. Mol. Sci. 22, 5043. [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Lesch KP, Timmesfeld N, Renner TJ, Halperin R, Röser C, Nguyen TT, Craig DW, Romanos J, Heine M, Meyer J, et al. (2008). Molecular genetics of adult ADHD: converging evidence from genome-wide association and extended pedigree linkage studies. J. Neural Transm. 115, 1573–1585. [DOI] [PubMed] [Google Scholar]
55.Romanos M, Freitag C, Jacob C, Craig DW, Dempfle A, Nguyen TT, Halperin R, Walitza S, Renner TJ, Seitz C, et al. (2008). Genome-wide linkage analysis of ADHD using high-density SNP arrays: novel loci at 5q13. 1 and 14q12. Mol. Psychiatry 13, 522–530. [DOI] [PubMed] [Google Scholar]
56.Asherson P, Zhou K, Anney RJL, Franke B, Buitelaar J, Ebstein R, Gill M, Altink M, Arnold R, Boer F, et al. (2008). A high-density SNP linkage scan with 142 combined subtype ADHD sib pairs identifies linkage regions on chromosomes 9 and 16. Mol. Psychiatry 13, 514–521. [DOI] [PubMed] [Google Scholar]
57.Seshadri S, DeStefano AL, Au R, Massaro JM, Beiser AS, Kelly-Hayes M, Kase CS, D’Agostino RB, DeCarli C, Atwood LD, and Wolf PA (2007). Genetic correlates of brain aging on MRI and cognitive test measures: a genome-wide association and linkage analysis in the Framingham Study. BMC Med. Genet. 8, S15. [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Grünewald L, Landaas ET, Geissler J, Weber H, Quast C, Röh S, Schartner C, Lesch KP, Romanos M, Kittel-Schneider S, et al. (2016). Functional impact of an ADHD-associated DIRAS2 promoter polymorphism. Neuropsychopharmacology 41, 3025–3031. [DOI] [PMC free article] [PubMed] [Google Scholar]
59.Kontani K, Tada M, Ogawa T, Okai T, Saito K, Araki Y, and Katada T (2002). Di-Ras, a distinct subgroup of ras family GTPases with unique biochemical properties. J. Biol. Chem. 277, 41070–41078. [DOI] [PubMed] [Google Scholar]
60.Grünewald L, Becker N, Camphausen A, O‘Leary A, Lesch KP, Freudenberg F, and Reif A (2018). Expression of the ADHD candidate gene Diras2 in the brain. J. Neural Transm. 125, 913–923. [DOI] [PubMed] [Google Scholar]
61.Wilkerson MD, Hupalo D, Gray JC, Zhang X, Wang J, Girgenti MJ, Alba C, Sukumar G, Lott NM, Naifeh JA, and Aliaga P (2023). Uncommon protein-coding variants associated with suicide attempt in a diverse sample of US Army soldiers. Biol. Psychiatry 96, 15–25. [DOI] [PubMed] [Google Scholar]
62.Li S, and Sheng ZH (2022). Energy matters: presynaptic metabolism and the maintenance of synaptic transmission. Nat. Rev. Neurosci. 23, 4–22. [DOI] [PubMed] [Google Scholar]
63.Faria-Pereira A, and Morais VA (2022). Synapses: the brain’s energy-demanding sites. Int. J. Mol. Sci. 23, 3627. [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Mahjani B, Bey K, Boberg J, and Burton C (2021). Genetics of obsessive-compulsive disorder. Psychol. Med. 51, 2247–2259. [DOI] [PMC free article] [PubMed] [Google Scholar]
65.Palmer DS, Howrigan DP, Chapman SB, Adolfsson R, Bass N, Blackwood D, Boks MPM, Chen CY, Churchhouse C, Corvin AP, et al. (2022). Exome sequencing in bipolar disorder identifies AKAP11 as a risk gene shared with schizophrenia. Nat. Genet. 54, 541–547. [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Cederlöf M, Lichtenstein P, Larsson H, Boman M, Rück C, Landén M, and Mataix-Cols D (2015). Obsessive-compulsive disorder, psychosis, and bipolarity: a longitudinal cohort and multigenerational family study. Schizophr. Bull. 41, 1076–1083. [DOI] [PMC free article] [PubMed] [Google Scholar]
67.Watson HJ, Yilmaz Z, Thornton LM, Hübel C, Coleman JRI, Gaspar HA, Bryois J, Hinney A, Leppä VM, Mattheisen M, et al. (2019). Genome-wide association study identifies eight risk loci and implicates metabo-psychiatric origins for anorexia nervosa. Nat. Genet. 51, 1207–1214. [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Strom NI, Gerring ZF, Galimberti M, Yu D, Halvorsen MW, Abdellaoui A, Rodriguez-Fontenla C, Sealock JM, Bigdeli T, Coleman JR, et al. (2025). Genome-wide analyses identify 30 loci associated with obsessive-compulsive disorder. Nat. Genet. 57, 1389–1401. [DOI] [PMC free article] [PubMed] [Google Scholar]
69.Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M, and Gingeras TR (2013). STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Wang J, Roeder K, and Devlin B (2021). Bayesian estimation of cell type-specific gene expression with prior derived from single-cell data. Genome Res. 31, 1807–1818. [DOI] [PMC free article] [PubMed] [Google Scholar]
71.Ritchie ME, Phipson B, Wu DI, Hu Y, Law CW, Shi W, and Smyth GK (2015). limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 43, e47. [DOI] [PMC free article] [PubMed] [Google Scholar]
72.Law CW, Chen Y, Shi W, and Smyth GK (2014). voom: Precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biol. 15, R29. [DOI] [PMC free article] [PubMed] [Google Scholar]
73.Leek JT, Johnson WE, Parker HS, Jaffe AE, and Storey JD (2012). The sva package for removing batch effects and other unwanted variation in high-throughput experiments. Bioinformatics 28, 882–883. [DOI] [PMC free article] [PubMed] [Google Scholar]
74.Hoffman GE, and Schadt EE (2016). variancePartition: interpreting drivers of variation in complex gene expression studies. BMC Bioinf. 17, 483. [DOI] [PMC free article] [PubMed] [Google Scholar]
75.Langfelder P, Luo R, Oldham MC, and Horvath S (2011). Is my network module preserved and reproducible? PLoS Comput. Biol. 7, e1001057. [DOI] [PMC free article] [PubMed] [Google Scholar]
76.Viechtbauer W (2010). Conducting meta-analyses in R with the metafor package. J. Stat. Softw. 36, 1–48. [Google Scholar]
77.Gusev A, Ko A, Shi H, Bhatia G, Chung W, Penninx BWJH, Jansen R, de Geus EJC, Boomsma DI, Wright FA, et al. (2016). Integrative approaches for large-scale transcriptome-wide association studies. Nat. Genet. 48, 245–252. [DOI] [PMC free article] [PubMed] [Google Scholar]
78.Chen J, Bardes EE, Aronow BJ, and Jegga AG (2009). ToppGene Suite for gene list enrichment analysis and candidate gene prioritization. Nucleic Acids Res. 37, W305–W311. [DOI] [PMC free article] [PubMed] [Google Scholar]
79.Koopmans F, van Nierop P, Andres-Alonso M, Byrnes A, Cijsouw T, Coba MP, Cornelisse LN, Farrell RJ, Goldschmidt HL, Howrigan DP, et al. (2019). SynGO: an evidence-based, expert-curated knowledge base for the synapse. Neuron 103, 217–234.e4. [DOI] [PMC free article] [PubMed] [Google Scholar]
80.Wakefield J (2009). Bayes factors for genome-wide association studies: comparison with P-values. Genet. Epidemiol. 33, 79–86. [DOI] [PubMed] [Google Scholar]
81.Wang L, Wang S, and Li W (2012). RSeQC: quality control of RNA-seq experiments. Bioinformatics 28, 2184–2185. [DOI] [PubMed] [Google Scholar]
82.Liao Y, Smyth GK, and Shi W (2014). featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923–930. [DOI] [PubMed] [Google Scholar]
83.Darmanis S, Sloan SA, Zhang Y, Enge M, Caneda C, Shuer LM, Hayden Gephart MG, Barres BA, and Quake SR (2015). A survey of human brain transcriptome diversity at the single cell level. Proc. Natl. Acad. Sci. USA 112, 7285–7290. [DOI] [PMC free article] [PubMed] [Google Scholar]
84.Jaffe AE, Tao R, Norris AL, Kealhofer M, Nellore A, Shin JH, Kim D, Jia Y, Hyde TM, Kleinman JE, et al. (2017). qSVA framework for RNA quality correction in differential expression analysis. Proc. Natl. Acad. Sci. USA 114, 7130–7135. [DOI] [PMC free article] [PubMed] [Google Scholar]
85.Yoo M, Shin J, Kim J, Ryall KA, Lee K, Lee S, Jeon M, Kang J, and Tan AC (2015). DSigDB: drug signatures database for gene set analysis. Bioinformatics 31, 3069–3071. [DOI] [PMC free article] [PubMed] [Google Scholar]
86.Kuleshov MV, Jones MR, Rouillard AD, Fernandez NF, Duan Q, Wang Z, Koplev S, Jenkins SL, Jagodnik KM, Lachmann A, et al. (2016). Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res. 44, W90–W97. [DOI] [PMC free article] [PubMed] [Google Scholar]
87.Nowicka M, and Robinson MD (2016). DRIMSeq: a Dirichlet-multinomial framework for multivariate count outcomes in genomics. F1000Res. 5, 1356. [DOI] [PMC free article] [PubMed] [Google Scholar]
88.Bray NL, Pimentel H, Melsted P, and Pachter L (2016). Near-optimal probabilistic RNA-seq quantification. Nat. Biotechnol. 34, 525–527. [DOI] [PubMed] [Google Scholar]
89.Zhang B, and Horvath S (2005). A general framework for weighted gene co-expression network analysis. Stat. Appl. Genet. Mol. Biol. 4. [DOI] [PubMed] [Google Scholar]
90.Szklarczyk D, Gable AL, Nastou KC, Lyon D, Kirsch R, Pyysalo S, Doncheva NT, Legeay M, Fang T, Bork P, et al. (2021). The STRING database in 2021: customizable protein-protein networks, and functional characterization of user-uploaded gene/measurement sets. Nucleic Acids Res. 49, D605–D612. [DOI] [PMC free article] [PubMed] [Google Scholar]
91.Wu D, and Smyth GK (2012). Camera: a competitive gene set test accounting for inter-gene correlation. Nucleic Acids Res. 40, e133. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

NIHMS2190757-supplement-1.pdf^{(7MB, pdf)}

NIHMS2190757-supplement-3.xlsx^{(204.5KB, xlsx)}

NIHMS2190757-supplement-4.xlsx^{(9.8MB, xlsx)}

NIHMS2190757-supplement-5.xlsx^{(14.2KB, xlsx)}

NIHMS2190757-supplement-6.xlsx^{(2.5MB, xlsx)}

NIHMS2190757-supplement-7.xlsx^{(3.3MB, xlsx)}

NIHMS2190757-supplement-8.xlsx^{(1.2MB, xlsx)}

NIHMS2190757-supplement-9.xlsx^{(5MB, xlsx)}

NIHMS2190757-supplement-10.xlsx^{(1.6MB, xlsx)}

Data Availability Statement

All original RNA sequencing data are publicly through the National Center for Biotechnology Information Gene Expression Omnibus https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE262138 (GEO: GSE262138).
All code used for data processing, analysis, and figure generation is publicly available through GitHub: https://github.com/BreenMS/OCD_ED/.
Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

PERMALINK

Convergent molecular signatures across eating disorders and obsessive-compulsive disorder in the human brain

Michael S Breen

Ran Tao

Andy Yang

Xuran Wang

Pardis Amini

Miguel Rodriguez de los Santos

Anna C Brandtjen

Amy Deep-Soboslay

Walter H Kaye

Thomas M Hyde

Joel E Kleinman

Joseph D Buxbaum

Dorothy E Grice

SUMMARY

In brief

Graphical Abstract

INTRODUCTION

RESULTS

Deep RNA sequencing of DLPFC and caudate in ED and OCD

Figure 1. Regional and disorder-specific differential expression analysis.

Regional and cell-type-associated sources of gene expression variation

Region-specific differential gene expression in ED and cross-disorder concordance with OCD

Replication and specificity of ED caudate transcriptional signatures

Combined ED + OCD analysis reveals convergent molecular pathology

Figure 2. Joint ED + OCD analysis reveals convergent cortico-striatal transcriptional pathology.

Cross-disorder transcriptomic concordance with independent postmortem brain datasets

Cross-cohort meta-analysis expands the OCD caudate transcriptional landscape

Figure 3. Meta-analysis of OCD gene expression changes in the caudate.

Gene co-expression network analysis reveals shared and disorder-specific modules

Figure 4. Co-expression network analysis implicates shared neuronal bioenergetic and CHD8-linked modules in ED and OCD.

Genetically predicted gene expression changes

Figure 5. Transcriptome-wide association analyses (TWAS) reveal convergent risk loci across ED and OCD.

Table 1.

DISCUSSION

Limitations of the study

RESOURCE AVAILABILITY

Lead contact

Materials availability

Data and code availability

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Postmortem human brain tissue

Study participants and sample allocation

Demographic characteristics

Sex and gender reporting

Ancestry, race, and ethnicity

Clinical characterization

Toxicology

Clinical and technical metadata

METHOD DETAILS

RNA-sequencing data generation

Discovery cohort:

Replication cohort:

QUANTIFICATION AND STATISTICAL ANALYSIS

RNA-sequencing data processing and outlier removal

Discovery cohort:

Replication cohort:

Cellular deconvolution of bulk tissues

Computing gene expression variance explained by technical and biological factors

Differential gene expression

Sensitivity analyses and qSVA

Drug perturbation enrichment analysis

Differential transcript usage analysis

Weighted gene co-expression network analysis

Gene ontology and enrichment analysis

Secondary analyses of existing OCD transcriptomic datasets

Transcriptome-wide association study (TWAS)

Statistical significance thresholds and asterisk definitions

Supplementary Material

KEY RESOURCES TABLE.

Highlights.

ACKNOWLEDGMENTS

Footnotes

REFERENCES

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS