Systems biology dissection of PTSD and MDD across brain regions, cell types, and blood

Abstract

The molecular pathology of stress-related disorders remains elusive. Our brain multi-omic, multi-region study of posttraumatic stress disorder (PTSD) and major depressive disorder (MDD) included the central nucleus of the amygdala, hippocampal dentate gyrus, and medial prefrontal cortex (mPFC). Genes and exons within the mPFC carried most disease signals replicated across two independent cohorts. Pathways pointed to immune function, neuronal/synaptic regulation and stress hormones. Multi-omic factor and gene network analyses provided the underlying genomic structure. Single nucleus RNA-sequencing in dorsolateral PFC revealed dysregulated (stress-related) signals. Analyses of brain-blood intersections in >50,000 UK Biobank participants were conducted along with fine-mapping of PTSD and MDD genome-wide association studies’ results. Our data suggest shared and distinct molecular pathology in both disorders and propose potential therapeutic targets and biomarkers.

INTRODUCTION:

Stress-related disorders arise from the interplay between genetic susceptibility and stress exposure, occurring throughout the lifespan. Progressively these interactions lead to epigenetic modifications in the human genome, shaping the expression of genes and proteins. Prior postmortem brain studies have attempted to elucidate the molecular pathology of posttraumatic stress disorder (PTSD) and major depressive disorder (MDD) compared to neurotypical controls (NCs), in a single-omic manner, revealing genomic overlap, sex differences, immune and interneuron signaling involvement. However, without integrative systems approaches, progress in understanding the molecular underpinnings of these prevalent and debilitating disorders is hindered.

RATIONALE:

To tackle this roadblock, we have created a brain multi-omic, multi-region database of individuals with PTSD and MDD and NCs (77/group, n = 231) to describe molecular alterations across 3 brain regions: the central nucleus of the amygdala (CeA), medial prefrontal cortex (mPFC), and hippocampal dentate gyrus (DG) at transcriptomic, methylomic and proteomic levels. By employing this multi-omic strategy that merges information across biological layers and organizational strata and complementing it with single-nucleus RNA sequencing (snRNA-seq), genetics and blood proteomics analyses, we sought to reveal an integrated systems perspective of PTSD and MDD.

RESULTS:

We found molecular differences primarily in the mPFC, with differentially expressed genes (DEGs) and exons carrying most disease signals. However, altered methylation was seen mainly in the DG in PTSD, in contrast to CeA in MDD subjects. Replication analysis substantiated these findings utilizing multi-omic data from two cohorts (n = 114). Moreover, we found a moderate overlap between the disorders, with childhood trauma and suicide being primary drivers of molecular variations in both disorders, and sex-specificity being more notable in MDD. Pathway analyses linked disease-associated molecular signatures to immune mechanisms, metabolism, mitochondria function, neuronal/synaptic regulation, and stress hormone signaling with low concordance across omics. Top upstream regulators and transcription factors included IL1B, GR, STAT3, and TNF. Multi-omic factor and gene network analyses provided an underlying genomic structure of the disorders, suggesting latent factors and modules related to aging, inflammation, vascular processes, and stress.

To complement the multi-omics analyses, our snRNA-seq analyses in the dorsolateral PFC (n = 118) revealed DEGs, dysregulated pathways and upstream regulators in neuronal and non-neuronal cell-types, including stress-related gene signals. Examining the intersection of brain multi-omics with blood proteins (in >50,000 UK Biobank participants) revealed significant correlation, overlap, and directional similarity between brain-to-blood markers. Fine-mapping of PTSD and MDD genome-wide association study results showed a limited overlap between risk and disease processes at the gene and pathway level.

Ultimately, prioritized genes with multi-omic, multi-region or multi-trait disease associations were members of pathways/networks, showed cell-type specificity, had blood biomarker potential or were involved in genetic risk for PTSD and MDD.

CONCLUSION:

Our findings unveil shared and unique brain, multi-omic molecular dysregulations in PTSD and MDD, elucidate the involvement of distinct cell types, pave the way for the development of blood-based bio-markers and to distinguish risk from disease processes. These insights not only implicate established stress-related pathways but also reveal potential novel therapeutic avenues.

Systems biology dissection of PTSD and MDD. The interplay between genetic susceptibility and stress exposure, occurring both early and later in life, contributes to the pathogenesis of stress-related disorders. Our integrative systems approach combines multi-region, multi-omic analyses with single-nucleus transcriptomics, blood plasma proteomics, and GWAS-based fine-mapping to provide deeper insights into molecular mechanisms associated with risk and those involved in the disease process.

The development of stress-related disorders, such as posttraumatic stress disorder (PTSD) and major depressive disorder (MDD), involves complex interactions between genetic susceptibility and exposure to traumatic stress. Over the last two decades, substantial efforts have focused on identifying the underlying risk factors and molecular mechanisms, as well as tackling the high rates of comorbidity and disorder heterogeneity associated with PTSD and MDD.

Genome-wide association studies (GWAS) revealed their heritability, polygenic architecture, and genetic overlap (1–6). Previous work profiling molecular alterations of stress-related mental disorders in blood (7–13) has implicated the innate immune response and regulation by stress hormones such as glucocorticoids (GC), but conclusions have been limited due to lack of direct access to the brain and its cell-types. The availability of a large, well-characterized postmortem brain collection of PTSD and MDD subjects and neurotypical controls (NCs) (14–16) enabled the investigation of brain-based molecular alterations. Genome-wide transcriptomic and methylomic studies of prefrontal cortex (PFC) subregions (16) and amygdala (AMY) nuclei (17, 18) in PTSD and MDD revealed a moderate genomic overlap and sex differences, confirmed immune dysregulations, and implicated both interneuron- and glia-based signaling. These studies converged with parallel functional analyses of GWAS loci (19, 20). Targeted transcriptomic and proteomic postmortem analyses also focused on disease associations with expression of immune- and stress-related genes (21). Finally, single-nucleus RNA-seq (snRNA-seq) suggested some neuronal and non-neuronal cell-types related to risk loci, stress pathways, and/or sex differences (21).

Despite recent progress, reductionist single-omic approaches, while valuable, likely miss a comprehensive picture. Multi-omic approaches combining data from genetics, transcriptomics, epigenomics, and proteomics, among others, may unveil an integrative systems view of stress-related diseases (22–24). Here, we provide a systematic characterization of a large multi-omic multi-region dataset of PTSD and MDD generated by the PTSD Brainomics Project of the PsychENCODE Consortium (PEC) Phase 2 (Fig. 1). The dataset consists of three brain regions [medial PFC (mPFC), dentate gyrus (DG) of the hippocampus (HIP), central nucleus of AMY (CeA)] from 231 subjects with PTSD and/or MDD vs NCs split over 2 cohorts. We interrogated disease-specific molecular changes at four transcriptomic levels [~22k genes, ~335k exons, ~140k exon-exon junctions (jxs), ~198k transcripts (txs)], one mDNA level [~740k cytosine-phosphate-guanine sites (CpGs)] and two proteomic levels [~6.2k proteins, ~60k peptides]. Replication was tested in equivalent datasets from up to 114 additional subjects. Analyses were complemented by a rigorous exploration of downstream pathways, multi-omic latent factors and gene networks. In parallel, we tackled cell-type specificity by conducting snRNA-seq analysis of 118 dlPFC samples (25–27), and evaluated blood-based protein biomarkers in >50,000 UK Biobank (UKBB) participants. Finally, PTSD and MDD risk was captured by the largest currently available GWAS datasets (5, 6), allowing investigation of the overlap of GWAS-based risk genes with our postmortem brain-based disease process genes [defined in (28)]. Our results suggested that multi-region, multi-omic mechanisms underlie shared and distinct brain pathology in PTSD and MDD, and they overlap with cell-type-specific, blood-based and genetically-mediated mechanisms.

Fig. 1. Overall study design.

We generated a large multi-omic postmortem database of PTSD (n=77) and MDD (n=77), compared to NCs (n=77) over two discovery cohorts (Disc.1 and Disc.2). Three brain regions (mPFC, DG, CeA) were assessed for bulk RNA expression (of genes, exons, exon-exon junctions, and transcripts), DNA methylation (CpGs and regions) and protein expression (proteins and peptides). Primary analyses included differential transcriptomic, methylomic, proteomic disease-specific interrogation, followed by pathway, multi-omic factor, and gene co-expression network analyses and identification of top genes. Sub-analyses were performed within traits (i.e., biological sex, childhood trauma, suicide completion), across traits (PTSD-or-MDD vs. NC) and between traits (PTSD vs. MDD) to assess contributing factors, disease-specificity, and overlap. For replication, we (1) generated a new dataset of 73 samples (Rep.1), re-analyzed data from prior studies (Rep.2) consisting of 41 additional samples, and (2) conducted meta-analysis of these two independent cohorts (n_{meta-analysis} = 114). In parallel, we (1) acquired two snRNA-seq datasets (Sc.1, n=47, Sc.2, n=71) from dlPFC to explore disease-associated cell-type-specific transcriptomic signatures by conducting meta-analysis across batches (n_{meta-analysis} = 118), (2) assessed the plasma protein-based biomarker potential in >50,000 subjects of the UKBB, and (3) fine-mapped the PTSD and MDD risk loci using GWAS datasets and investigated the overlap of GWAS-based risk genes and pathways with disease process genes. The extensive generated data enabled us to identify genes significantly involved in both disorders.

Multi-region, multi-omic signatures in PTSD and MDD

Discovery cohort 1 [‘Disc.1’, 50/group (150 total); table S1A–1] and 2 [‘Disc.2’, 27/group (81 total); table S1A–2] were initially analyzed separately. No significant differences were found between disease groups and NCs in cell-type proportions, as estimated from RNA genes or CpGs (table S1B–1/S1B–2). We then meta-analyzed the results from both cohorts and revealed multi-omic alterations for both disorders (PTSD: Fig. 2, table S2A1–21, MDD: Fig. 3, table S2B1–21). Most differential gene expression (DGE) signals passing 5% false discovery rate (FDR) level were found in mPFC for both PTSD (Fig. 2A) and MDD (Fig. 3A) with differentially expressed genes (DEGs) and exons leading, followed by jxs in PTSD, and txs in MDD. In the other brain regions, genes and exons in MDD showed a substantial number of FDR-significant signals (FDR-adjusted P < 0.05). Second, we found many differentially methylated positions (DMPs) in the DG in PTSD (Fig. 2B), and less in MDD (Fig. 3B). Similarly, differentially methylated regions (DMRs) were mainly observed in the DG in PTSD (95; Fig. 2B, table S2C) and the CeA (17) and DG (13) in MDD (Fig. 3B; table S2C). PTSD had slightly less differentially expressed proteins (DEPs) and peptides (Fig. 2C) compared to MDD (Fig. 3C). These findings were confirmed by counting FDR-significant distinct genes in both PTSD (Fig. 2D) and MDD (Fig. 3D).

Fig. 2. Transcriptomic, methylomic, and proteomic analyses of PTSD in three brain regions.

(A-C) Plots of the meta-analysis of 231 subjects in mPFC, DG and CeA. (A, C) Volcano plots of differentially regulated transcriptomic (A) and proteomic features (C). Colored dots denote nominally significant genes (P < 0.05), with the darker ones passing an FDR 5% level. Five features with the lowest P in each direction and the replicated ones are named. (B) Manhattan plots of the CpGs interrogated for differential methylation (x-axis; genomic location, y-axis; −log₁₀P). DMPs passing FDR-adjusted P < 0.05 are denoted in purple. CpGs that belong to a DMR (>2 CpGs, Šidák P < 0.05) are red, and within those the CpGs with FDR-adjusted P < 0.05 are green. (D) Scatterplot denoting the number of distinct genes corresponding to FDR-significant features (size coded) per feature-type per brain region. The percentage of FDR-adjusted features over the N of features is labeled next to the respective point. (E) Boxplot of ρ corresponding to correlations of DGE effect sizes (log₂FC or beta) in discovery meta-analysis (“Disc”) and Rep.1 cohort analysis (“Rep”) across brain regions. Three significance thresholds were used: “genome-wide” (no threshold), considering all features, “nom Disc”, considering only nominally significant (P < 0.05) features in Disc, and “nom Disc+Rep”, considering overlapping nominally significant features in Disc and Rep. (F) Lollipop plots of the number of replicated features per omic with their respective gene/protein annotations and direction of effect (upward black arrowheads: increased; downward black arrowheads: decreased). (G) Left panel: Multi-region boxplots depicting the range of ρ corresponding to correlations of DGE effect sizes between PTSD differential analyses of each brain region for each feature. Middle panel: Boxplots depicting the range of ρ between PTSD differential analyses results with results from sub-analyses including sex-specificity, childhood trauma and suicide completion, across brain regions. Right panel: Boxplots of ρ between PTSD differential analyses results with results from MDD primary analysis, and PTSD-or-MDD and PTSD vs. MDD sub-analyses. In D-G, colors denote different omic features and shape different brain regions. In E and G, horizontal dotted lines denote minimal (ρ<0.1), moderate (0.3<ρ<0.6) and high (ρ>0.6) correlation.

Fig. 3. Transcriptomic, methylomic, and proteomic analyses of MDD in three brain regions.

See legend of Fig. 2 for detailed description. (A, C) Volcano plots of differentially regulated transcriptomic (A) and proteomic features (C). (B) Manhattan plots of CpGs with genomic loci on the x axis and −log₁₀P on the y-axis. red: Sidak P significant DMRs, purple: FDR-significant DMPs, green: DMPs within DMRs. (C) Volcano plots of proteomic feature DE. (D) Scatterplot denoting the number of distinct genes corresponding to FDR-significant features (size coded) per feature-type per brain region, with the respective percentage labeled. (E) Boxplot of correlation coefficient ρ of DGE effect sizes (log₂FC or beta) in discovery meta-analysis (“Disc”) and Rep.1 cohort analysis (“Rep”) across brain regions anf three significance thresholds. (F) Lollipop plots of the number of replicated features per omic with their respective gene/protein annotations and direction of effect (upward arrowheads: increased; downward arrowheads: decreased). (G) Left panel: Boxplots of correlation coefficient ρ of DGE effect sizes between MDD analyses of each brain region for each feature. Middle panel: Boxplots depicting the range of ρ between MDD primary analyses with sub-analyses across brain regions. Right panel: Boxplots of ρ between MDD differential analyses results with results from PTSD primary analysis, and PTSD-or-MDD and PTSD vs. MDD sub-analyses. In D-G, colors denote different omic features and shape different brain regions. In E and G, horizontal dotted lines denote minimal (ρ<0.1), moderate (0.3<ρ<0.6) and high (ρ>0.6) correlation.

We ran mega-analyses and sensitivity analyses focused on multi-ancestry composition (fig. S1) and RNA quality to ensure the robustness of our results. Results were strongly correlating with those from the discovery meta-analysis (fig. S2A, and details in (28).

Replication analyses

We analyzed data from two replication cohorts [“Rep.1”: 24/group (table S3A1–2), “Rep.2”: 14/group (table S3B1–2)] with details described in (28). In summary, correlations of discovery meta-analysis with Rep.1 results were moderate to high (Spearman’s ρ = 0.3–0.6), indicating concordant associations (Fig. 2E/3E, fig. S2B). The correlation drivers were feature type (F-test, P = 7.36E-08; RNA> Protein> mDNA), followed by brain region (F-test, P = 7.91E-07; DG> mPFC> CeA) and statistical threshold (F-test, P = 5.30E-06). FDR-significant features in discovery meta-analyses showed beyond-chance rate of concordant direction and nominal significance (P < 0.05) in Rep.1 results (PTSD: table S3C1–21, MDD: table S3C22–42, enrichment tests: table S3D), which included 43 replicated features (11 PTSD/ 32 MDD) with FDR-adjusted P < 0.05 in Rep.1 mapped to 28 unique genes (table S3E/F). For Rep.2 (fig. S2B), we confirmed stronger correlations at the gene level compared to methylation. Meta-analysis of Rep.1 and Rep.2 revealed concordant replicated DEGs to discovery cohorts in PTSD (binomial P = 2.42E-184) and MDD (binomial P = 4.33E-91), adding three replicated RNA genes for each trait (table S3E/F).

43 out of 49 replicated features had the same direction as the discovery meta-analysis (Fig. 2/3F). 12 out of 14 PTSD replicated features (10 genes) were found in the mPFC, and 29 of 35 MDD replicated features (16 genes) were found in the CeA. PTSD replicated genes were more related to brain cell-types (notable examples: upregulation of EPHA2 and PIRT, ARHGAP24 hypermethylation and increased GLUD1 protein expression; table S3E) compared to replicated MDD genes reflecting elevated cytokine signaling (table S3F).

Between-region correlations

Within the primary PTSD and MDD analyses, we found weak to moderate between-region correlations of effect sizes (ρ range 0.3–0.6) across omics. Moderate correlations were found at the gene, protein, MDD peptide and exon across-region pairs, PTSD exons at the CeA-DG pair, PTSD peptides at the DG-mPFC pair, and MDD CpGs at the CeA-DG pair (Fig. 2/3G: left panel).

Correlations of primary analyses with sub-analyses

We performed sub-analyses distinguishing (1) biological sex, and (2) cases with childhood trauma or death by suicide. We reported moderate to strong (ρ > 0.6) correlations of primary analyses. Female-specific analyses for both disorders demonstrated moderate correlations with their respective primary analyses. In contrast, male-specific analyses in PTSD showed moderate correlations with the primary analyses of PTSD, while those in MDD exhibited strong correlations with the primary analyses of MDD. Additionally, analyses focusing on childhood trauma and suicide within both disorders displayed strong correlations with their respective primary analyses. Proteins and peptides had the lowest correlations among all these analyses (Fig. 2/3G: middle panel). Such observations confirm the role of these factors in the overall omic-disease effects and suggest that distinct multi-omic features may underlie the larger risk for both PTSD and MDD in females.

Correlations with other trait, across-trait, and between-trait analyses

Most cases with primary PTSD diagnosis had secondary depression (table S1A–1/2), whereas none of the MDD cases had secondary PTSD diagnosis. The correlation between PTSD and MDD primary analyses was moderate except for high correlations at the gene, exon and jx-level of mPFC and low correlations of CpGs in the mPFC, which emphasized the importance of epigenetic data in distinguishing the two disorders (Fig. 2/3G: right panel). We also conducted additional analyses comparing (1) all cases (“PTSD-or-MDD”/combined) to NCs, and (2) PTSD cases to MDD cases. The correlation of the primary analysis with combined analysis was high, with mPFC data (especially genes and exons) showing the highest correlations (Fig. 2/3G: right panel). The correlation of PTSD primary analysis with the PTSD vs. MDD analysis was moderate, while the correlation of MDD analysis with the absolute (PTSD vs. MDD) analysis was weak (Fig. 2/3G: right panel).

Functional annotation of multi-omic and multi-region signatures

To identify disease-associated pathways, we performed gene set enrichment analysis (GSEA) based on gene ontology (GO) across omics (table S4A1–18). Ranking pathways based on significance per modality revealed clustering of omic layers in both traits (Fig. 4A). Immune-related biological processes were upregulated at the transcriptome level while adaptive and innate immunity subsets were downregulated at the proteome level. Methylomic signatures mostly related to nervous system, axon and synapse development. Biological processes related to (m)RNA metabolism and transcription were upregulated at the proteome level. Regarding cellular components, ribosomes were upregulated at the transcriptome level and downregulated at the proteome level. Downregulated proteomic signatures were associated with presynaptic cellular components, while methylomic signatures of MDD related to neuron-to-neuron (post)synaptic functions.

Fig. 4. Gene, protein, and methylation set enrichment analysis across brain regions and disorders.

(A) Heatmap depicting −log₁₀ rank of transcriptomic-, proteomic- and methylomic-based pathways in each brain region across disorders (18 analyses). Within each analysis, pathway rank is calculated based on FDR-adjusted P. The 5 most significant pathways per analysis are shown here. GO categories include biological processes (BP), cellular components (CC) and molecular functions (MF). Upward arrows: positive NES, downward arrows: negative NES, circles: methylation-related entries (direction unknown). The full name of third pathway from the top is “Adaptive immune response based on somatic recombination of immune receptors built from immunoglobulin superfamily domains”. (B-D) Boxplots depicting the range of ρ values corresponding to correlations between omics (B), regions (C) or traits (D). (E-F) Bubble plot CP enrichment in the DEGs (blue-green outline) and DEPs (red outline) in PTSD (E) and MDD (F). A pathway can belong to multiple categories. Points are sized based on −log₁₀ (FDR-adjusted P). Shape fill denotes z-scores. The most significant pathways are labeled. In E, an * is used to facilitate the annotation of STAT3 pathway and # for HMGB1 signaling. (G) Heatmap of URs enriched in DEGs per brain region in both traits. Significant URs (FDR-adjusted P < 0.05) were ranked based on z-scores and the first 50 URs for each disorder were selected. The exogenous chemicals and drugs categories were excluded from plotting. Grey color: non-existent data. UR categories are shown on top. PTSD DG did not have URs. Abbreviations of non-gene terms: LPS, lipopolysaccharide; E2, beta-estradiol; Ig, immunoglobulin. (H) Heatmap of the TF binding enrichment (−log₁₀ FET P). Not all modalities showed significant enrichments; # and *: FDR-adjusted P < 0.05.

Within brain regions, pathways showed absent to weak correlations between omics (Fig. 4B). The highest correlations (>0.25) were observed between RNA-protein and RNA-methylation in the DG of PTSD and RNA-methylation in the CeA of MDD. Between region correlations ranged from weak to high (Fig. 4C) with transcriptomic pathways showing higher correlations (>0.60) compared to proteomic (0.2–0.40) and methylomic (0.05–0.45) ones. RNA pathways had the highest correlations between traits (Fig. 4D). This analyses thus suggests that different omic signals are involved in distinct pathways and that, within omics, transcriptomic-based pathways tend to be most conserved between regions (Fig. 4C) and traits (Fig. 4D).

We repeated pathway analyses using Rep.1 results to assess replication of our discovery pathways. Discovery pathways correlated with Rep.1 pathways considerably at the transcriptomic level followed by proteomic and methylomic levels (fig. S3). PTSD pathways replicated more than MDD pathways (573 vs. 91, table S4A–19), most of which were from mPFC RNA (389), DG mDNA (85) and CeA RNA (83). For MDD, replicated pathways were from CeA and DG RNA (35 and 23, respectively) and DG and mPFC mDNA (14 and 12, respectively).

Next, canonical pathway (CP) enrichment in PTSD and MDD DEGs and DEPs across regions (table S4B) revealed shared DEG-driven upregulation of immune pathways related to extracellular matrix (ECM) organization including activin/inhibin, hepatic fibrosis, cytokine storm signaling, and STAT3 signaling pathways, mostly in the mPFC for PTSD (Fig. 4E) and DG for MDD (Fig. 4F). MDD DEGs in CeA and DG pointed to the downregulation of Liver X receptor (LXR) and Retinoid X receptor (RXR) regulatory pathways and PTSD DEPs in DG implicated the complement system.

To identify regulatory changes in PTSD and MDD, we identified upstream regulators (URs) of the RNA signals (table S4C). IL1B, TNF, IFNG, CREB1, TGFB1 and OSM were the most prominently activated URs in both disorders, while IgG, SCD, mir-155, PPARGC1A, DICER1, and glutamine were the most deactivated URs (Fig. 4G, table S4C). We then identified enriched transcription factor (TF) binding on the genes of our disease-associated features (table S4D) with lowest Fisher’s Exact test (FET) P seen in ESR1, ELK3, NR3C1 [glucocorticoid receptor (GR)], RELB and RUNX1 (Fig. 4H).

Spatial registration of signatures

We spatially registered the DEGs and DEPs within cortical layers 1–6 (L1–6) and white matter. We found meningeal and L1 enrichment of DEGs in PTSD and MDD (fig. S4A), but deeper L4-L5 enrichment of DEPs in MDD (fig. S4B). Leptomeninges and L1 contain mostly non-neuronal cells, while deeper cortical layers contain mostly neuronal cells (29), highlighting the importance of deciphering cell-type-specificity.

Prioritizing top genes from bulk-tissue multi-omic data

To identify key genes associated with stress-related disorders, we integrated FDR-significant signals across omic features, brain regions, and traits amounting to 4,469 genes (2,677 PTSD, 2,970 MDD and 1,178 shared, table S5A, fig. S5). Demanding a signal in at least half the features within each omic modality for a given trait, we found 1,690 genes (table S5B); 1,016 were related to PTSD, 1,043 to MDD, and 369 to both. These genes were categorized based on the omic layer of their disease-associated signal (1,355 RNA, 146 mDNA, and 223 protein), with 34 genes having signal in two omic layers. Notably, these 1,690 genes were also distributed across brain regions, (394 in CeA, 441 in DG, 1239 in mPFC, and 292 in all regions). We prioritized top genes from this pool based on at least one of the three criteria: i) multi-region, ii) multi-omic, and iii) multi-trait overlap. This qualified 367 genes, henceforth called top genes (table S5B). Among the top genes, 280 were associated with PTSD (fig. S5A), 360 with MDD (fig. S5B) and 273 were shared. Gene breakdown per criterion can be found in (28) (fig. S5C–D). PTSD and MDD top genes were significantly enriched in replicated genes (FET, P = 2.34E-11 and P = 1.04E-21, respectively).

Following an analogous procedure, we identified 2,330 top pathways (2,040 in PTSD, 2,293 in MDD, 2,003 shared, table S5C). 7 top genes (EDN1, FGF2, IL1B, RAC1, TGFB2, TGFBR1, STAT3) were represented in an outlying number of PTSD top pathways, and 5 (EDN1, FGF2, IL1B, RAC1, TGFB2) in MDD top pathways (table S5D).

PTSD and MDD top genes were enriched in GC-associated DEGs (FET, P = 3.54E-24 and 2.41E-32, respectively) identified in induced pluripotent stem cell (iPSC)-derived neurons after a 4-hour treatment with GR-agonist dexamethasone (DEX) (26). Using data from iPSC-derived cerebral organoids treated 12-hours with DEX (30), we found PTSD and MDD top genes’ enrichment in DEX-associated DEGs of non-neural progenitors (FET, P = 3.10E-09 and 9.67E-11, respectively), neural progenitors (FET, P = 1.40E-08 and 1.19E-10, respectively), and neurons (FET, P = 5.18E-03 and 4.42E-03, respectively; table S5E), from our top genes in GR signaling, shedding light on their roles across cell types.

Multi-omic views

Multi-Omics Factor Analysis [MOFA (31), Fig. 5A] reduced dimensionality by inferring 30 latent factors that captured variance (R²) of the three omic layers from every brain region (9 views). We observed factors focusing on: (i) a single view, namely factor 1 on mPFC methylome and factors 5–7 on CeA, DG, and mPFC proteomes, respectively, (ii) a single-omic view in multiple regions, such as factor 2 capturing variance for transcriptome across brain regions, and (iii) multi-omic views within one region, including factor 15 on CeA and factor 17 on DG (Fig. 5B). Notably, factor 13 was the only factor with high R² values across all nine views and showed strong correlation with age (Fig. 5C). This factor may be a multi-region, multi-omic “clock”. Factor 13 scores differed between diagnoses (F-test, P < 9.88E-05, and Tukey’s post-hoc P < 0.001 for each group vs. NC, Fig. 5D), which suggests a “multi-omic age acceleration” associated with MDD and PTSD. Finally, factor 14 captured the transcriptome of all brain regions with scores higher in both disorders compared to NCs [F-test P < 1.24E-05, and Tukey’s post-hoc P < 0.001 for each group vs. NCs (Fig. 5E)] and with the factor loadings having moderate correlations with DGE effect sizes (fig. S6A).

Fig. 5. Multi-omic integration.

(A) Nine (3×3) views provided by transcriptomic (T-), methylomic (M-), and proteomic (P-) profiles of each brain region were integrated into thirty latent factors using MOFA. We used the same input to additionally create co-expression modules for each omic across brain region using WGCNA. (B) Heatmap of variance explained in each of the nine views by each MOFA factor. (C) Coefficients (ρ) of correlation of MOFA factors scores with age at death. (D) Scatterplot of MOFA factor 13 scores (y-axis) with age at death (x-axis). LOESS-smoothed trendlines are fitted within diagnosis group. (E) Box-and-whisker plots of MOFA factor 14 scores by diagnostic group: data are represented as median ± 1.5 interquartile range (IQR). Individual factor 14 scores are indicated by vertical line markers. In D-E, gray color is used for NCs, blue-green for MDD, and red for PTSD. (F-H) For 3 transcriptomic and/or proteomic features per brain region, the enrichment of the respective differentially expressed (DE) features from the PTSD and MDD analyses as well as of the top PTSD and MDD genes (top genes) is depicted (F: CeA, G: DG, H: mPFC). The x-axis represents the enrichment significance (−log₁₀ P), the size of the point denotes the number of features enriched and the shape the gene set under interrogation in each analysis (DE: circle, top genes: triangle). (I-K) The 10 most significant GO terms (color-coded) associated with the CeA-pink (I), DG-tan (J) and mPFC-red (K) modules. (L) PPI network of the mPFC-red module. The nodes are filled in green if the protein is a PTSD and/or MDD top gene, and the shape of the node indicates whether it is a hub gene (diamond). DEX genes are annotated in purple. Ten GO terms are shown as partially filled donuts around each node according to (K).

Next, we used weighted gene co-expression network analysis (WGCNA, Fig. 5A) to investigate gene network correlates of MOFA factors. In NC samples, we constructed modules at every omic level per region (table S6A). We highlighted 3 modules from each region enriched in disease-associated features and top genes (Fig. 5F–H). The modules (CeA-pink, DG-tan, and mPFC-red) with lowest enrichment P-values also had the strongest association with factor 14 (fig. S6B). Functional annotation of these modules showed shared immune system processes, inflammatory response, and vasculature development/angiogenesis-related pathways (Fig. 5I–K; table S6B). Notably, the mPFC-red module was associated with response to stress and GCs and exhibited the most significant enrichment with GC-responsive genes (33/181, FET, FDR-adjusted P = 1.47E-19; table S6C), including FKBP5. Finally, protein-protein interaction (PPI) network visualization of the three modules (Fig. 5L) emphasized their hub gene (table S6D) enrichments in both PTSD (FET, P = 8.48E-24) and MDD top genes (FET, P = 1.60E-27).

Single-cell transcriptomics

To explore cell-type-specific transcriptional signatures in both disorders, snRNA-seq data from 118 subjects were analyzed [single-cell cohort 1 - “Sc.1” (26) and single-cell cohort 2 - “Sc.2 (27) (Fig. 6A, table S7A). As described (26), Sc.1 batch 1 contained ~363k nuclei post-QC (Fig. 6B), and Sc.1 batch 2 ~137k nuclei (Fig. 6C), clustering in 8 broad cell-types [excitatory (Ex) and inhibitory (In) neurons, astrocytes (Astro), microglia (Micro), oligodendrocytes (Oligo), oligodendrocyte-precursor cells (OPC), endothelial cells (Endo), and pericytes (Per)] and several subtypes (fig. S7A–B). Cell-subtype annotation similarities between the Sc.1 batches were confirmed (fig. S7C–D, table S7B–1). Sc.1 batch1 clustering annotations were projected on Sc.2 (~160k nuclei) and ~35k nuclei were removed as ambiguous neuronal/non-neuronal profiles. Remaining ~126k nuclei were reannotated with the Sc.1 batch 1 cell-subtype labels (Fig. 6D, table S7B–2, fig. S7E–F). Further, we compared cell-type proportions of each disorder across the relevant batches (table S7C–1–2) and reported differences in Micro and OPC in MDD [table S7C–3, supplementary text (28)].

Fig. 6. snRNA-seq study of PTSD, MDD and NCs in dlPFC.

(A) Analytic strategy for snRNA-seq datasets. (B-C) Sc.1 comprised of two batches: Batch 1 with ~363k (B) and Batch 2 with ~137k nuclei (C). Representative tSNE plots of the 8 broad cell-types. (D) tSNE plot of ~126k nuclei from the two integrated batches of Sc.2 (male and female batches included) annotated to match the identity of the clustering from S.c1. (E-F) Volcano plots of the DGE in PTSD (E) and MDD (F) across seven cell-types (color coded). The dots are colored to denote nominally significant genes (P < 0.05) in the respective cell-types and the darker colored dots genes with FDR-adjusted P < 0.05. Up to 5 most significant (FDR-adjusted P) cell-type-specific DEGs er direction are labelled, along with GC-responsive (*) and PTSD-MDD shared DEGs (#). The number of DEGs passing 5% FDR level per cell-type is shown below. (G-H) Correlation of the DGE between cell-types in PTSD (G) and MDD (H). Data are represented as median ± 1.5 IQR, while the individual points denote the correlation for each cell-type. (I) Correlation of the cell-type-specific DGE between PTSD and MDD. Data are represented as median ± 1.5 IQR, while the individual points denote the correlation for each cell-type. The black dot annotates the correlation of PTSD and MDD in the bulk mPFC tissue. (J) Heatmap demonstrating the 5 most significantly enriched GO pathways in each cell-type per disorder. The terms have been clustered based on their NES. The color gradient denotes negative to positive enrichment and the asterisk-annotated pathways have an FDR-adjusted P < 0.05. In G and I, horizontal dotted lines denote minimal (ρ<0.1), moderate (0.3<ρ<0.6) and high (ρ>0.6) correlation.

We performed batch-level cell-type-specific DGE analysis, adjusting for confounders (28), followed by meta-analysis. For PTSD, the meta-analysis of the broad cell-types contained 16 NCs/16 PTSD (table S8A). Based on the batch-specific nuclei contribution to each cell-type in Sc.2 (fig. S7F), we meta-analyzed all batches for Astro, Ex, In and OPC (52MDD/50NCs, table S8B–1–4), and excluded the Sc.2 male batch from the meta-analysis of Endo, Micro, and Oligo (35MDD/34NCs, table S8B–5–7).

In PTSD we reported 52 FDR-significant DEGs (Fig. 6E, table S8A) with 88% in Ex (46), 19% in In (10) and ~3% in Astro (2). 31 out of these 52 DEGs overlapped with PTSD DEGs identified when only analyzing Sc.1 cohort (26) [supplementary text (28)]. In MDD, we reported 779 FDR-significant DEGs across 6 cell-types (Fig. 6F, table S8B). Astro had the most DEGs with 376 (48%), followed by Ex 27% (217) and In 25% (199). The remaining DEGs were found in Oligo 4% (39), 1% OPCs (8) and < 1% Endo (3). Among the MDD DEGs, 18 overlapped with Chatzinakos et al. (27) and 45 with Maitra et al. (27), and 716 were newly reported [supplementary text (28)].

Four genes in the PTSD-associated 17q21.31 locus (6) were prominent DEGs in neurons and Astro in PTSD (ARL17B, LRRC37A2, LINC02210-CRHR1, KANSL1). ARL17B was a DEG in neurons (FDR-adjusted P < 4.6e-09) and Astro (FDR-adjusted P <9.9e-04) with a consistent marked increase of approximately 2 log 2-fold change. In MDD, we reported upregulation of FKBP5, a GC-responsive gene, in neurons and Oligos, expanding previously reported upregulation in In-neurons (26). Further intersecting GC-responsive genes with disease-associated cell-type-specific DEGs revealed upregulation of CDH3, TAF1C and SLC16A6 in Ex in PTSD and, among other 66 genes (table S8B), upregulation of STAT3 in Oligo and downregulation of DDIT4 in MDD. The multi-trait DEGs in same cell-types were limited to downregulation of SRSF6, an alternative splicing regulator and top PTSD gene from the bulk analysis, in Ex and upregulation of TMPRSS9 in In.

In PTSD, Ex and In correlated moderately with each other (ρ = 0.41), but weakly with glia cell-types (ρ = 0.06–0.26). Endo exhibited no correlation with other cell-types (Fig. 6G). In MDD, we detected moderate correlations between neuronal types (ρ = 0.53) and with Oligos (ρ = 0.34), while astrocytes were moderately correlated with OPCs (ρ = 0.34, Fig. 6H). The between-diagnosis cell-type-specific correlation ranged from weak to moderate, but was lower than in bulk (mPFC, ρ = 0.79) highlighting the importance of studying cell-type-specific signals to distinguish pathophysiologically similar disorders (Fig. 6I).

GSEA of cell-type-specific PTSD and MDD profiles revealed downregulated synaptic pathways in neuronal and non-neuronal cell-types (table S9A), particularly pronounced in MDD Ex-neurons, Astro, and Endo (Fig. 6J). Ribosome-related processes were downregulated in Oligo in both disorders. In MDD, these processes were downregulated in In-neurons, Micro, and OPC, while they were upregulated in Endo and Micro but downregulated in PTSD Astro. Metabolic and mitochondrial processes were downregulated in In and Oligo in both disorders and Micro and Oligo/OPC in MDD. Inflammatory pathways were upregulated in PTSD and downregulated in MDD, suggesting differential immune signaling regulation in each disorder. Furthermore, adhesion and extracellular transport pathways were downregulated in Astro and Endo in MDD. Notably, glia-related pathways were downregulated in Micro and Ex in PTSD and In in MDD, while being upregulated in OPCs in MDD.

We further identified activated cell-type-specific URs in MDD (fig. S7G, table S9B–C). STK11, UR of the stress-activating AMPK pathway, was differentially activated in Astro and neurons. PSEN1 was activated in both In and Astro, while APP and TGFB1were de-activated. CPA confirmed strong deactivation of oxidative phosphorylation and ATP processes in neurons in MDD, and mitochondrial dysfunction and Sirtuin signaling pathway activation, especially in In-neurons. Non-neuronal cell-types involved the inhibition of stress-hormone complex in Astro, STAT3, and RAC signaling along with activation of epithelial adherence-related pathways in Oligo and Notch signaling in Endo and OPC (fig. S7H, table S9D–E). Finally, PTSD DEGs of In-neurons were enriched in THOP1 neuroprotection and metabolism of amine-derived hormones (including norepinephrine).

Top genes in live blood plasma

To evaluate blood-based biomarkers of PTSD and MDD, we analyzed 1,463 plasma proteins in 54,219 UKBB subjects (32). PTSD, defined as a binary indicator and a continuous score (~1,300), and MDD, defined broadly and strictly (~ 3,500) were compared to healthy subjects (~15,000). We found more FDR-significant proteins associated with MDD compared to PTSD (fig. S8, table S10A), especially when using the electronic medical record (EMR)-based and help-seeking definitions of MDD. GO pathway analysis of blood DEPs detected ECM organization, response to GCs, various interleukin signaling pathways, neuron projection, and synapse assembly pathways for both disorders (table S10B). FDR-significant CPs and URs for both disorders converged in alterations in transcription regulation, nuclear receptor and cytokine signaling (table S10C–D).

Blood DEPs’ effect sizes had stronger correlations with brain DEGs’ effect sizes from all brain regions compared to brain DEPs, with the strongest detected between MDD (EMR-based) blood DEPs and MDD/PTSD CeA DEGs (fig. S9). Notably, the EMR-based diagnosis is comparable to the definition of MDD and PTSD from postmortem medical records. PTSD and MDD top genes were enriched in PTSD blood DEPs (24 genes, FET P 3.91E-16) and MDD blood DEPs (52 genes, FET P = 1.26E-28), respectively (fig. S10A, table S10E–1), with 23 shared across the two disorders in brain and blood. Out of these 23, 19 were multi-region top genes including EPHA2 and TNFRSF1A/12A. Among the 29 top genes that exclusively overlapped in MDD blood and MDD brain, 22 were multi-region [details in supplementary text (28)].

43 PTSD-associated GO pathways in the brain overlapped with pathways in blood (fig. S10B, table S10E–2), while 121 pathways in the MDD brain overlapped with MDD blood [details in supplementary text (28)]. 22 pathways in brain and blood for both disorders related to lipid processes, immune response, and corticosteroid response. Majority were RNA-derived in brain (95% in PTSD, 69% in MDD). 21 pathways, exclusively overlapping in PTSD brain and PTSD blood were RNA-derived in the brain and included those related to neuroinflammatory response and ECM organization. From 99 pathways that exclusively overlapped in MDD brain and blood, 31 that included synapse process and axon guidance originated from methylation in brain.

We noticed that certain pathways, such as axonal guidance, were detected in both brain and blood, involving one omic layer of brain molecular data, while their gene members shared between the brain and blood could implicate another brain omic layer.

Brain RNA-based CPs showed overlap with blood CPs. Twelve pathways exhibited multi-trait characteristics in both blood protein and brain RNA and protein, encompassing features such as hepatic fibrosis cell activation and LXR/RXR (fig. S10C, table S10E–3). Furthermore, we observed enrichment of brain with blood URs in MDD (fig. S10D, table S10E–4) without substantial overlap of URs in PTSD across brain and blood [details in supplementary text (28)].

Multi-region, multi-omic view of risk genes and pathways

We estimated PTSD and MDD GWAS SNP-based heritability were highly correlated [Rg = 0.887±0.013; Methods (28)]. Polygenic risk scores (PRS) for each disorder were calculated for 304 subjects across Disc.1, Disc.2, and Rep.1 cohorts. Compared to MDD-PRS, the PTSD-PRS had higher associations with both diagnoses in the respective target population (fig. S11) (28).

Fine-mapping of the latest GWAS identified 76 PTSD loci with 68 having 1–3 credible sets each, and 98 MDD loci with 92 having 1–10 credible sets each (table S11A–1/2 and B1–2, respectively). Local heritability calculations revealed loci in PTSD and MDD that exhibited significant heritability for both disorders [table S11A3–4 and S11B3–4]. We observed variability in local genetic correlations as well, particularly with loci involving the 6p22.1/22.2, 17q21.31, and 22q13.1 cytogenetic bands, with the lowest correlations (fig. S12). More details can be found in supplementary text (28).

We used quantitative trait locus (xQTL) panels to conduct transcriptomic, methylomic, and proteomic summary-based mendelian randomization (TSMR, MSMR, and PSMR, respectively; fig. S13, table S11C), and SNP-based multi-omic imputation to conduct transcriptome-, methylome- and proteome-wide association studies (TWAS, MWAS and PWAS, respectively; fig. S14, table S11D). PFC-based TSMR and TWAS revealed more PTSD and MDD risk genes compared to AMY- or HIP-based analyses (PTSD: fig. S13A–B, MDD: fig. S14A–B), consistent with our bulk RNA-seq results. TSMR and TWAS analyses discovered more risk genes than methylomic- and proteomic-based analyses (PTSD: fig. S13C–D, MDD: fig. S14C–D). More details can be found in supplementary text (28).

We detected intersection of 36 TSMR- and TWAS-based PTSD risk genes and the respective DEGs, 10 of which were top genes (Fig. 7A). Similarly, 31 mPFC and 3 DG TSMR-based and TWAS-based MDD risk genes overlapped with respective DEGs (FET, P = 5.22E-06), 6 of which were top genes (Fig. 7B). Only 5 risk genes of each trait with multi-region, multi-omic and/or multi-trait characteristics overlapped with the respective top genes (PTSD: fig. S13E, MDD: fig. S14E).

Fig. 7. Identification of GWAS-based risk genes and pathways for PTSD and MDD.

(A, B) Venn diagram of xSMR/xWAS-based risk genes overlap matched with FDR-significant (FDR-adjusted P < 0.05) disease process genes and with top genes for PTSD (A) and MDD (B). (C, D) Venn diagram of xWAS-based risk pathways overlap matched with FDR-significant disease pathways and top pathways for PTSD (C) and MDD (D). (E) Heatmap of mediation effects of PTSD or MDD risk genes in the association of GWAS SNPs with diagnosis (Dx) through gene expression alterations. SNP-gene pairs qualified by brain TSMR analyses of PTSD and MDD GWAS. The first two columns contain the cytogenetic band and the rsID of the qualified GWAS SNP. The mediating gene along with the tissue can be seen in the third column. The −log₁₀ P of the Average Causal Mediation Effects (ACME) is provided as a measure of the significance of mediation. (F) Heatmap depicting each cell-type-disease association with PTSD and MDD. Numbers indicate FDR-adjusted P of the cell-type-disease associations. Heatmap color denotes the proportion (%) of significantly associated cells (FDR-adjusted P < 0.1) with the trait. *: FDR-adjusted P < 0.05. (G, H) Upset plots of scTSMR-based risk genes for PTSD (G) and MDD (H) at the level of cell-types in the dlPFC.

Using a TWAS pathway method (33), we detected TWAS pathways for all three tissues and PWAS pathways for PFC (table S11E–1–2). Contrary to the xWAS gene analysis, HIP had the most risk pathways for both disorders (292 and 551, respectively) compared to AMY (11 and 14, respectively) and PFC (77 and 102, respectively). Pathways shared between brain regions (PTSD: 23; fig. S13F, MDD: 32; fig. S14F) were mostly neural/neuronal, trafficking, organelle and metabolism-related. Neuronal/synaptic pathways were overlapping in TWAS and PWAS of both disorders (PTSD: 6, MDD: 32).

Overall correlation between risk pathways and respective bulk-tissue-based pathways were low (PTSD: range −0.08–0.06, MDD: range −0.1–0.13). Immune, synaptic and developmental pathways were shared between risk and disease process pathway sets (Fig. 7C–D). Notably, DG RNA gene pathways were significantly enriched in HIP TWAS pathways in both disorders (PTSD: FET P = 3.08E-02, MDD: FET P = 1.99 E-10), with 46/48 found in PTSD and 179/180 in MDD top pathways.

Molecular outcomes of gene-by-environment interactions and molecular mediation of risk

We investigated the molecular impact of SNP-by-childhood trauma interactions on TSMR-identified SNP-gene pairs [Methods (28)]. In PTSD (fig. S15A, table S11–F1), many cis-eQTLs passed an FDR 5% level [61 out of 138 in mPFC, 21 out of 30 in DG, and 19 out of 27 in CeA]. Only rs62060768 (not part of a credible set of 17q21.31) showed an FDR-significant interaction of DG LRRC37A4P expression, and the same interaction nominally affected CeA LRRC37A4P expression. There were two additional 17q21.31 SNPs, both within a credible set, with nominal interaction effects. FDR-significant childhood trauma-only effects were seen for mPFC expression of the top gene LIMK2. In MDD (fig. S15B, table S11–F2), many cis-eQTLs passed an FDR 5% level [45 out of 146 in mPFC, 12 out of 15 in DG, and 8 of 9 in CeA]. We observed only 5 nominal SNP-by-childhood trauma interactions. One of the SNPs in the LRRC37A4P CeA analysis was the same as in the PTSD dataset. FDR-significant childhood trauma-only effects were seen for mPFC LINC00461 and CNPPD1.

In the blood plasma dataset we had power to distinguish the type of abuse and neglect (fig. S16A–B, table S11–F3–4). We observed SNP effects on plasma proteins for many brain-based eQTLs. Within the PTSD and MDD SMR genes analyses, we observed childhood trauma effects for 20 genes (including HLA-E top gene) and 14 genes, respectively.

Finally, we explored the mediation effect of gene expression in the SNP-to-Dx effects (Fig. 7E, table S11G). We found that four SNP-to-PTSD diagnosis effects were mediated though mPFC expression of ATP23, CYP2D6, and ZSCAN29, and CeA expression of LINC02210. We also found two mediation effects for MDD diagnosis through mPFC expression of STAG3L2 and TIPIN.

Multi-cell-type view of PTSD and MDD risk

Using a brain cell-type annotation (34), we detected neuronal enrichment of MDD and PTSD risk genes (fig. S17A–B). We also determined enrichment of risk genes in dlPFC cell-type-specific markers of Sc.1. In- and Ex-neuronal and OPC markers were enriched in risk genes of both disorders (fig. S17C–D), while Oligo markers were enriched only in PTSD. Incorporating gene-level statistics with cell-to-cell heterogeneity (35), we found a stronger signal of risk genes in Ex, In and OPCs for MDD than PTSD (Fig. 7F). We then conducted SMR for PTSD and MDD at the dlPFC cell-type level, finding cell-type-specific risk genes predominantly in Ex-neurons and Oligo in both disorders (Fig. 7G–H, table S11H–1–2). In PTSD, neuronal PTSD risk genes were significantly enriched in neuronal DEGs (LINC02210-CRHR1 and TTC12, FET P = 1.92E-03) and non-neuronal PTSD risk genes in non-neuronal DEGs (ARL17B, FET P = 2.69E-03). In MDD, only ANKRD36 overlapped between non-neuronal MDD DEGs and risk genes.

Integration

For full integration, we aggregated and ranked - in tiers of evidence - signals related to top genes across all levels of analyses (table S12A), including: (i) PTSD and MDD replication, (ii) membership in top pathways, (iii) acting as hub genes in disease-associated networks, (iv) blood associations, (iv) association with childhood trauma in brain or blood, (vi) snRNA-seq analyses, (vii) brain-based genetic analyses, and (viii) GC regulation. Overall, top genes showed high overlap across these dimensions, supported by FDR-significant FET-based enrichments. We visualized the 30 top genes with most convergent evidence (Fig. 8, fig. S18).

Fig. 8. Integration of results.

Top genes were ranked based on accumulating statistical evidence across analyses in tiers (table S12A). Chromosomal location of top genes with the highest amount evidence is visualized. Six genes are localized on chromosome 12, followed by 3 genes on chromosome 6. Red squares depict PTSD top genes while a purple outline reflects membership in PTSD top pathways. Blue-green squares depict MDD top genes, while a purple outline reflects membership in MDD top pathways. Purple squares denote replication in PTSD analysis, while purple circles reflect replication in MDD analysis. Red squares reflect PTSD blood DEPs, red circles reflect MDD blood DEPs. Purple triangles depict association with childhood trauma analysis in the brain analyses, while red triangles reflect association with childhood trauma in the blood analyses. Purple diamonds indicate genes that are module hubs, while yellow diamonds outline reflect GC-regulation (DEX) in iPSC-derived neurons (26). Double squares and circles represent single-cell (sc) findings for PTSD and MDD, respectively; neuronal (Neu) vs non-neuronal (NonNeu) distinction is made by the yellow fill (Neu: not filled, NonNeu: filled). Finally, in relation to genetic analyses, blue shapes represent TSMR and TWAS, and blue shapes with gray fill reflect MSMR and MWAS. The double blue circle with gray fill represents significance in a scTSMR analysis of a non-neuronal cell-type.

We similarly ranked the top pathways (table S12B) to reveal comparable overlaps and enrichments, but with weaker enrichment of plasma pathways and more significant enrichment of cell-type-specific pathways. We visualized the top pathways with most convergent evidence (fig. S19) highlighting the following functional themes: i) neuronal signaling and regulation, ii) immune and inflammatory responses, iii) tissue development and maintenance, and iv) metabolic processes.

Discussion

In this multi-omic, multi-region postmortem brain study of PTSD and MDD conducting discovery and replication analyses, we found both shared and distinct molecular signatures in both disorders. The most robust alterations were mostly captured by genes and exons occurring within the mPFC. PTSD had more regional molecular differences than MDD. mDNA changes for PTSD and MDD were mostly localized in the DG and CeA, respectively. Childhood trauma and suicide were the main drivers of signal in both disorders, while sex differences were more apparent in MDD. Top genes qualified based on multi-omic, multi-region, and/or multi-trait importance and by showing robustness through replication, gene networks, snRNA-seq, blood and/or genetic analyses. Detailed discussions of top genes can be found in supplementary text (28).

Each omic layer implicated distinct biological processes. Notably, we revealed an RNA-based upregulation, but protein-based downregulation of immune-related pathways across regions and disorders. These enriched pathways implicated members of the TNF receptor superfamily, namely, the top gene TNFRSF1A, which is GC-regulated in neural cells (26). IL1β and TNFα, the most prominent URs, are triggered by stress in brain and periphery (36–38) and have been previously associated with PTSD and MDD (8, 14, 39–42). Relatedly, the RNA-based MOFA factor most associated with both diagnoses and immune-related RNA modules.

Our multi-omic analysis revealed distinct involvement of neuronal and non-neuronal cell types in both disorders. DEGs highlighted immune and ECM pathways, predominantly in non-neuronal external cortical layers and leptomeninges. In MDD, protein alterations were more prominent in neuron-rich deeper layers (29), while methylation pathway alterations affected neuronal processes in both disorders. Furthermore, our snRNA-seq analysis demonstrated significant transcriptomic changes in both neuronal and non-neuronal cell types, uncovering previously unreported alterations (25–27). Underlying cell-type-specific pathways showed potential in differentiating between the two disorders. Although both disorders exhibited enrichment in neuronal and non-neuronal cell types, this enrichment appeared more pronounced in MDD. Finally, scTSMR identified risk genes in Ex and Oligo in both disorders.

Dysregulated ECM-related pathways contained top genes such as ICAM1 (multi-omic in PTSD), and COL4A1 (multi-omic and multi-region in MDD), which were replicated in plasma DEPs. The interstice between ECM and neuronal/synaptic plasticity is covered by axonal guidance molecules like EPHA2, an RNA-based top gene, replicated, altered in blood and GC-regulated, which has also been associated with blood brain barrier (BBB) hyperpermeability (43) and suicidality (44). Stress-induced BBB hyperpermeability (45, 46) would allow peripheral inflammation, as detected by our large blood proteomics analysis, to enter brain parenchyma.

The robust correlation between PTSD and MDD in GWAS contrasted with the moderate to weak correlations identified in bulk multi-omic and snRNA-seq data. Additionally, the limited overlap in both disorders between the genes and pathways implicated in GWAS and those identified in brain molecular data, underscored the disparity between disease risk and underlying disease processes. In contrast, greater overlap was seen between disease associate patterns in brain and blood. While our brain measurements were independent from blood measurements, animal models of individual stress differences have supported shared effects for some pathways (47–49). Considering the challenges with accessing antemortem brain tissue to profile the disease in real-time, our findings support the development of brain-informed blood biomarkers.

Integrating results from various analyses related to the stress system revealed distinct genetic loci but shared downstream molecular signatures. For instance, the 17q21.31 locus, which exhibits one of the lowest correlations between PTSD and MDD, has consistently appeared in PTSD GWAS studies (20, 26), with our fine-mapping pointing to CRHR1, among other genes. We discovered indications of mediation involving LINC02210, while the fusion gene of LINC02210 with CRHR1 was identified as a neuronal scTSMR-based risk gene and a DEG for PTSD.

In MDD, GR-encoding NR3C1 did not appear in a locus of low correlation between PTSD and MDD but was identified solely as an MDD risk gene in PFC TWAS. GR also emerged as a prominent TF across omics for both disorders, targeting approximately 20% of the top genes. FKBP5, which encodes an inhibitory protein for GR function (50, 51), exhibited DGE in the mPFC for PTSD, and in Ex/In-neurons and Oligo for MDD. Additionally, it was a gene-member of the most prominently disease-associated module in the mPFC.

Furthermore, the top gene, STAT3, showed DGE in the CeA and mPFC for both traits and in Oligo for MDD, and was involved in several top pathways associated with PTSD related to wound healing, mitochondrial function, inflammation, and synaptic plasticity in neurons (52–54). STAT3 activation in mPFC was also evident for both disorders. Notably, STAT3 is GC-responsive in iPSC-derived neurons (26), acts as a GR co-activator (55), and has recently been linked to depressive-like behavior in animal models (56, 57).

The dynamic interplay between genetic susceptibility and downstream biology evolves across the lifespan. Our investigation unveiled enduring effects of childhood trauma on risk loci linked to both disorders, alongside insights into the influence of aging. MDD and PTSD share mechanistic pathways with neurodegenerative conditions (58). Notably, the PTSD-associated 17q21.31 locus, including MAPT, implicated in neurodegeneration, underscores these connections (59). Moreover, factor 13, a multi-omic ‘clock’ indicative of age acceleration in stress-related disorders, aligns closely with findings from epigenetic-specific clocks based on blood and brain samples (60–63).

Study limitations primarily stem from inherent biases in postmortem brain research around population selection (including ancestry), clinical assessment, comorbidities, and end-of-life state. The current large study is constrained in power at variable levels across various molecular modalities and tissues, and cell-types. We also did not comprehensively characterize the epigenetic landscape and did not fully capture all cell-subtypes and cell states. The description of our results focused primarily on convergent signals across omics or regions or traits, and ancillary studies could explain signal contrasts across the molecular, biological and clinical dimensions. Detailed limitations can be found in supplementary text (28).

In summary, our data suggest that a systems-biology approach is necessary to understand the complexity of molecular alterations in brain circuitry underlying stress-related disorders such as PTSD and MDD. Merging multi-omics from multiple brain regions with other molecular data can result in the identification of specific genes and regulatory mechanisms. Capturing these nuances is critical when aiming to develop informative biomarkers and discover potential therapeutic strategies.

Materials and methods summary

The PTSD Brainomics Project (PEC Phase 2) generated from the mPFC, DG and CeA of 231 LIBD subjects with PTSD and/or MDD and NCs from two cohorts (n_Disc.1 = 150, n_Disc.2 = 81). Samples were i) genotyped using Omni2.5 BeadChip and imputed using TOPMed service, ii) ribo-zero RNA-sequenced using TruSeq v2 and processed using SPEAQeasy (64) to extract transcriptomic features at gene, exon, splice junction and transcript levels, iii) methylation profiled using EPIC BeadChip and processed using minifi package (65), and iv) protein assayed with TMT-based quantification of proteomic features (proteins and peptides) and searched against UniProt database for proteins and peptides. Data were normalized and analyses were adjusted for confounds, cell-type proportions, global ancestry, demographics, and clinical characteristics. We assessed the association of diagnosis across omics and regions and conducted sub-analyses using the limma package (66). Results were meta-analyzed results across discovery cohorts [metafor (67)]. For DMRs were detected with comb-p (68). We replicated our findings by generating Rep.1 (same genomic features as Disc. cohorts, n = 73), and reanalyzing Rep.2 [re-analyzed RNA-seq and mDNA data from a prior study, n = 41, (14, 17)], separately and as a meta-analysis of common features (n = 114).

GSEA was performed with fgsea and methylGSA packages (69, 70). We utilized Ingenuity Pathway Analysis to identify CPs and URs. For transcription factor binding enrichment, we utilized Enrichr (71). For spatial registration of disease signatures we used spatialLIBD (72). Integration of DGE results was performed with MOFA and WGCNA.

We leveraged publicly available dlPFC snRNA-seq datasets from 118 individuals (26), ensuring cell-(sub)type alignment using Seurat (73). We analyzed batches separately and then meta-analyzed. Functional annotations were interrogated with GSEA and CP and UR analyses. Protein-based biomarkers were evaluated in plasma from >50,000 UKBB participants (74).We leveraged multiple dlPFC snRNA-seq datasets from 118 individuals (75), ensuring cell-(sub)type alignment using Seurat (73). We analyzed batches separately and then meta-analyzed using metafor. Functional annotations were interrogated with GSEA and CP and UR analyses. Protein-based biomarkers were evaluated in plasma from >50,000 UKBB participants (74). We performed association testing (limma) and pathway analysis [clusterProfiler (75)].

We used the largest available PTSD and MDD GWAS datasets (5, 6). With LDSC (76), we estimated global SNP-based heritability and genetic correlation between the disorders, while with LAVA (77), we estimated local heritability and correlation. For fine-mapping we used a combination of FUMA (78) with SusieR (79, 80). GWAS-based risk genes were identified with SMR using bulk-tissue cis-xQTL databases matching the omics and brain regions of our bulk-tissue studies (81–83) and cell-type-specific cis-eQTLs matching the cell-types of our snRNA-seq studies (84). Risk genes were also identified with xWAS [implemented by JEPEGMIX2-P using pretrained molecular imputation models (85–87) matching the omics and brain regions of our studies. For cell-type enrichment we used partitioned heritability (88), MAGMA (89) and scDRS (35).(89) and scDRS (35). SNP candidates for gene-by-environment interactions were based on the TSMR results. SNP-gene pairs were considered if the target gene existed in our normalized expression dataset of the respective tissues. For UKBB plasma proteomics, we used the same gene-pairs but further reduced it to SNP-protein pairs if the target protein was expressed in plasma. For environment variables we used childhood trauma. Additive and interaction models were tested. To test mediation effects of gene expression in the SNP-to-Dx effects we used statsmodels (90).

See full methods in (28).

Data and materials availability:

The source data described in this manuscript are available via the PsychENCODE Knowledge Portal (https://psychencode.synapse.org/). The PsychENCODE Knowledge Portal is a platform for accessing data, analyses, and tools generated through grants funded by the National Institute of Mental Health (NIMH) PsychENCODE Consortium. Data are available for general research use according to the following requirements for data access and data attribution: (https://psychencode.synapse.org/DataAccess). For access to content described in this manuscript see: https://www.synapse.org/#!Synapse:syn51217925 All analysis code and scripts are available on GitHub at https://github.com/DaskalakisLab/Daskalakis-Science2024 and Zenodo (91). The GWAS summary statistics are available by the Psychiatric Genomics Consortium (PGC) at pgc/download. The MVP GWAS summary statistics are made available through dbGAP request study_id=phs001672.v11.p1