Posttraumatic Stress Disorder Epigenome-Wide Association Studies in the Million Veteran Program

Abstract

Genomic studies have improved our understanding of posttraumatic stress disorder (PTSD) biology. Epigenetic differences in DNA methylation can reflect environmental influences, which are critical in PTSD etiology, and may additionally differentiate between PTSD cases and controls or distinguish severity of PTSD symptoms. We conducted PTSD epigenome-wide association studies (EWAS) in a large cohort, n = 22,141 subjects from the United States V.A. Million Veteran Program (MVP), with the PTSD Checklist (PCL-17), including 17,674 European ancestry (EUR), 3,430 African ancestry (AFR), and 1,037 admixed American descent (AMR). We evaluated the 17-item PCL-Total as well as symptom subdomains (re-experiencing, avoidance, and hyperarousal), and used lasso regression on electronic health records to define lifetime PTSD diagnoses. There were 9,413 total cases of Lifetime PTSD and 27,257 controls, with 5,184 cases and 19,382 controls in EUR, 3,095 cases and 6,240 controls in AFR, and 1,134 cases and 1,635 controls in AMR. We identified a total of 241 epigenome-wide-significant associated CpG sites across all PTSD traits. We replicated 5 out of 11 CpG sites reported in a prior (Psychiatric Genomics Consortium) PTSD EWAS. Overlap with differentially methylated CpG sites in subregions of the amygdala and hippocampus of human postmortem brains of individuals with PTSD vs. controls was evaluated at the CpG and gene levels. Ten genes of the 195 identified in the brain study overlapped with the 192 identified in our EWAS. There was significant enrichment of genes downregulated in somatostatin interneurons and brain endothelial cells. Regression analyses showed that PTSD case status was significantly associated with accelerated DNA methylation aging. Contrary to previous PTSD EWAS findings, smoking status stratification suggested that the association between PTSD and methylation of the AHRR gene may be confounded by smoking status. Our findings showcase the largest-scale PTSD EWAS in multiple ancestries and the first large PTSD EWAS on symptom severity. We replicated several findings from previous EWAS and greatly extended current knowledge of the relationship between epigenetics and PTSD.

Introduction

Posttraumatic stress disorder (PTSD) can occur after experiencing or witnessing a traumatic event and, per DSM-IV diagnostic criteria as implemented in the Million Veteran Program (MVP) sample, is marked by re-experiencing, avoidance, and hyperarousal symptoms [1]. These symptoms are associated with lower quality of life, higher likelihood of substance use and suicidality, and higher risk for numerous other psychiatric and medical disorders [2, 3]. Approximately 70% of people will experience serious trauma in their lifetimes, but only approximately 1 in 20 will develop PTSD [4, 5], suggesting that there are factors that modify risk. Previous genome-wide association studies (GWAS) have identified numerous genetic risk variants for PTSD and its symptom subdomains [6–8].

Trauma exposure is integral to PTSD diagnosis. The environmental factors that influence PTSD risk may do so through mechanisms that affect gene regulation. The best-studied mechanism is DNA methylation (DNAm) at cytosine-guanine dinucleotides (CpG sites), which can modulate gene expression and is stably maintained through successive cell divisions [9]. This maintenance of DNA methylation at regulatory sites ensures that epigenetic information, once established, can persist long after the initial environmental exposure, providing a molecular record of past biological or environmental influences.

Exposure to trauma has been shown to affect epigenetic marks [10], and there have been several prior epigenome-wide association studies (EWAS) and meta-EWAS of PTSD [11–13], the largest of which included 5,077 subjects from 23 individual studies. These studies [11–13] included individuals of European (EUR) and African (AFR) ancestry. Previous PTSD EWAS identified significant CpGs in genes related to the immune system and neurotransmission. The most recent EWAS meta-analyses included 2,156 PTSD cases and 2,921 controls from 23 civilian and military studies [11]. They identified 11 CpG sites, nine of which were novel, associated with PTSD and replicated two findings of a prior epigenome-wide meta-analysis [13] that included 10 civilian and military studies and reported lower AHRR methylation in individuals with PTSD. AHRR methylation is a notable characteristic of smoking traits [14], to the extent that it can be a more reliable indicator of smoking than self-report [15]. Accordingly, it is very hard to adjust EWAS for smoking exposure to account for AHRR signals, and this may have been an issue in previous studies that highlighted PTSD associations with this locus.

Sample size and heterogeneity were limitations in prior studies. Here, we address these limitations by conducting the largest EWAS of PTSD to date within a single well-characterized cohort, increasing statistical power and reducing cross-study heterogeneity. This study includes many more subjects than have been studied previously in a well-characterized sample. The MVP is based on a military veteran sample, includes many individuals affected with PTSD, and has been a major contributor to prior GWAS studies [6, 7]. The MVP includes several kinds of PTSD-relevant data, including electronic health record (EHR) diagnoses and the self-reported PCL-17 [16]. The PCL-17 allowed us to evaluate PTSD as a quantitative trait and to investigate DSM-IV symptom subdomains, as we did previously in a GWAS [7]. Also, our study extends epigenetic analyses of PTSD to more-informative samples of non-EUR participants; we include data from EUR, AFR, and AMR. Finally, we quantified methylation-based biological aging using a discrepancy-based approach, regressing Horvath DNA methylation age (DNAmAge) onto chronological age to derive age acceleration metrics and evaluate their association with PTSD case status.

Results

PTSD epigenome-wide associations

In the primary analyses, we identified the following epigenome-wide significant sites for EUR: 139 CpG sites associated with total PCL score (PCL-Total), 30 CpG sites associated with re-experiencing (REX), 230 CpG sites associated with avoidance (AVOID), 62 CpG sites associated with hyperarousal (HYP), and 4 CpG sites associated with lifetime PTSD diagnosis (PTSD-Dx) (Supplementary Table 1). Because hypomethylation of AHRR CpG sites is associated with cigarette smoking—this is one of the best-replicated findings in epigenetics—we performed sensitivity analyses adjusting for smoking status. When we adjusted for smoking status, 7 CpG sites were associated with PCL-Total, all of which were captured in the primary analysis. No CpG sites were associated with REX when we adjusted for smoking status. Twenty-three CpG sites were associated with AVOID, all of which were captured in the primary analysis except cg00994936 (DAZAP1). Two CpG sites were associated with HYP when adjusted for smoking status; cg20761853 (TIMP2) was not significant in the primary analysis of HYP but was significant in the primary analysis of PCL-Total. No CpG sites were associated with PTSD-Dx when adjusted for smoking status.

In AFR, there were no significant sites for PCL-Total, HYP, AVOID, or REX in the primary analysis nor when adjusted for smoking. PTSD-Dx had significant associations cg25322382 (DDI2) and cg10213873 in the primary analyses, neither of which was significant after smoking status adjustment. In AMR, PCL-Total and REX were associated with cg22036458 in the primary analyses, but this did not survive the smoking adjustment.

Overall, in the primary analyses, we identified 241 unique CpG sites associated with PTSD and 192 genes total across all ancestries. In almost all cases, genomic inflation factors were lower after adjustment for smoking status, indicating that these test statistics were less inflated by systemic biases (Supplementary Table 2). Slight increases in genomic inflation factors in AFR REX, AFR AVOID, and AMR PTSD-Dx can be attributed to noise introduced by small sample sizes. Early GWAS showed marked inflation in lambda and QQ plots as sample sizes increased, reflecting pervasive polygenicity rather than solely confounding, a distinction enabled by methods such as LD score regression (LDSC) [17]. EWAS presents additional challenges because inflation may arise from correlated methylation patterns driven by latent causal exposures (e.g., smoking or cell composition), which are currently more difficult to disentangle from true biological signal using existing methodologies.

Replication of previous work

In a recent PTSD PGC EWAS paper [11], there were 11 PTSD-associated CpGs that passed the epigenome-wide significance threshold; all 11 of these CpGs passed probe quality control in our sample. We replicated 5 of these 11 PTSD-associated CpGs (Supplementary Table 3). The 5 replications, which occurred in primary analyses, are mainly from AHRR: cg05575921 (AHRR), cg21161138 (AHRR), cg23576855 (AHRR), cg14753356 (Intergenic), and cg19719391 (Intergenic). There were no replications when adjusted for smoking status.

Postmortem human brain overlap

To assess whether PTSD-associated methylation signals identified in peripheral blood show cross-tissue convergence with molecular alterations observed in a directly disease-relevant tissue, we compared the significant blood EWAS findings with differentially methylated CpG sites identified in postmortem PTSD brain tissue (subregions of the hippocampus and amygdala; 117 cases, 54 controls; 11.6% female; 3,648 significant CpGs out of 1,136,867 tested). Although no CpGs precisely overlapped between the two cohorts, we identified four pairs of blood EWAS CpGs and brain CpGs located within 1 kb of one another, mapping near SLC25A34, EXOC2, LIF, and NFE2L3. The closest of these was observed at the LIF locus, where the brain CpG (chr22:30,639,828) and blood EWAS locus (chr22:30,639,979) were separated by only 151 bp and both localized to the same annotated CpG island (chr22:30,639,729–30,639,994) (Figure 3). At the gene level, 10 of 192 blood EWAS genes were also present among the 1,208 unique genes annotated to significant PTSD brain CpGs (SLC25A34, NOTCH1, MYO1C, LIF, RIN3, PIP5K1C, CREBBP, NFE2L3, LINC00299, TULP4). These overlapping genes have diverse biological functions but are predominantly involved in transcriptional and chromatin regulation, neurodevelopmental and immune signaling, and intracellular trafficking and signaling.

Figure 3.

Brain and blood CpG sites localizing to a single CpG island at the LIF locus.

Enrichment analyses

According to Enrichr [19–21], 342 gene sets from 38 libraries, representing publicly available data sources as well as purpose-built libraries from data from the Ma’ayan lab [19–21], were significantly enriched, with FDR < 0.05 for the term associated with the gene set (Supplementary Table 4). From the Allen Brain Atlas, the set of genes downregulated in GABAergic inhibitory interneurons expressing somatostatin (SST) and P4HA3 was enriched (Human Inh L1 SST P4HA3 down, adjusted p-value = 2.77e-03, OR = 14.8, overall score = 175.6) [22]. Five of the six overlapping genes had associated CpG sites with positive effect directions in the EWAS meta-analyses, indicating that a higher PCL-Total score was associated with hypermethylation. (WDPCP had a negative direction of effect.) Similarly, genes downregulated in interneurons expressing somatostatin and neuropeptide Y in Huntington’s disease (HD) patients vs. control were enriched (adjusted p-value = 3.81e-03, OR = 1.89, overall score = 19.5); this gene set is from HDsigDB, a database of molecular signatures in HD [23–25]. The GABAergic interneuron gene module, also from HDsigDB, was also enriched (adjusted p-value = 3.62e-02, OR = 5.78, overall score = 40.5). HDsigDB yielded 57 Enrichr hits out of 342 total significant hits, more than any other gene set library; by contrast, our Enrichr analysis of an EWAS of height we completed in this sample as a comparator [26] yielded two HDsigDB hits. Other Allen Brain Atlas enrichments include genes downregulated in brain endothelial cells (BECs) expressing NOSTRIN and SRGN (Human Endo L2–5 NOSTRIN SRGN down, adjusted p-value = 0.030, OR = 2.89, overall score = 25.3) [22] and genes downregulated in astrocytes expressing FGFR3 and PLCG1 (Human Astro L1–6 FGFR3 PLCG1 down, adjusted p-value = 0.041, OR = 2.56, overall score = 20.6) [22]. HDsigDB yields an enrichment for BECs of HD patients vs. control (genes with a change in expression, adjusted p-value = 1.48e-06, OR = 2.90, overall score = 58.5) [23] and astrocytes of HD patients vs. control (genes upregulated, adjusted p-value = 6.16e-03, OR = 2.97, overall score = 28.2) [23].

Other notable significant CpG sites include cg21253130, in FRY, FRY microtubule-binding protein, which is associated with neuron projection development. This CpG site is also significant in EWAS of NO₂ exposure (a respiratory irritant and potential neurotoxin) [27], obesity [28], and type 2 diabetes [28]. Its p-value in the EWAS meta-analysis of PCL-Total, with smoking included as a covariate, is 7.90e-09, with a positive direction of effect in all three ancestries. Its p-value in an EWAS of never smokers in EUR is 1.48e-03, suggesting that its significance is not solely due to residual smoking confound.

DNMT3A, DNA methyltransferase 3 alpha, responsible for de novo DNA methylation, has a significant CpG site, cg21277502, in the 5' UTR in the EWAS meta-analysis of PCL-Total, with p-value = 4.90e-09 and a negative direction of effect in all three ancestries. In EUR never smokers, p-value = 0.00563. DNMT3A is associated with GO terms “response to ionizing radiation” (GO:0010212), “response to lead ion” (GO:0010288), “response to Vitamin A” (hepatotoxic, GO:0033189), “response to cocaine” (GO:0042220), and “response to ethanol” (GO:0071361).

CREBBP, CREB-binding protein, a histone acetyltransferase that acetylates H3K27, has a significant CpG site, cg05194552, in the EWAS meta-analysis of HYP, with p-value = 8.62e-09 and a positive effect direction in all three ancestries. In EUR never smokers, p = 8.73e-04. Among other roles, CREBBP is associated with “long-term potentiation” (hsa04720). The cg05194552 probe is significant in EWAS of asthma [29] and B-cell acute lymphoblastic leukemia [30].

NR3C1, the glucocorticoid receptor, also has a significant CpG site in the EWAS meta-analysis of PCL-Total, cg22233604 (p-value = 7.89e-09, positive direction of effect in all three ancestries, p-value in EUR never smokers = 0.0498). In addition to its endocrine roles, NR3C1 is associated with “synaptic transmission, glutamatergic” (GO:0035249), “positive regulation of neuron apoptotic process” (GO:0043525), and “astrocyte differentiation” (GO:0048708). The cg22233604 probe is significant in an EWAS of myalgic encephalitis/chronic fatigue syndrome [31].

Smoking status stratification

Within PTSD cases in EUR, there were 30.1% current smokers, 47.1% former smokers, and 22.9% never smokers, for a total of 5,044 individuals (vs. 5,184 when smoking status was not included). Within PTSD controls in EUR, there were 19.6% current smokers, 53.3% former smokers, and 27.1% never smokers, for a total of 18,041 individuals (vs. 19,382 when smoking status was not included). Within PTSD cases in AFR, there were 32.0% current smokers, 40.7% former smokers, and 27.3% never smokers, for a total of 3,024 individuals (vs. 3,095 when smoking status was not included). Within PTSD controls in AFR, there were 29.7% current smokers, 42.5% former smokers, and 27.7% never smokers, for a total of 6,015 individuals (vs. 6,240 when smoking status was not included). Within PTSD cases in AMR, there were 23.6% current smokers, 42.6% former smokers, and 33.8% never smokers, for a total of 1,094 individuals (vs. 1,134 when smoking status was not considered). Within PTSD controls in AMR, there were 18.3% current smokers, 44.7% former smokers, and 37.0% never smokers, for a total of 1,513 individuals (vs. 1,635 when smoking status was not included). Since lower methylation of CpG sites at AHRR has been reliably and very strongly associated with smoking [32], a previous PTSD EWAS study [13] examined the differences of effect between PTSD and smoking status due to a higher rate of smoking within their cohort. We stratified all the PTSD EWAS analyses in EUR, AFR, and AMR which had significant AHRR CpG sites by smoking status to see the differences in effect in the corresponding PTSD trait.

In all the stratified analyses, we adjusted for the number of tests performed and assessed the association between AHRR CpGs and PTSD. CpGs in AHRR that were significant in the pooled analysis with smoking status as a covariate all consistently failed to show an association with PTSD in separate EWAS of current, former, and never smokers.

DNAm Adjusted Age

Age at initial blood draw and Horvath DNAmAge for cases and controls in EUR and AFR are provided in Supplementary Table 5. Horvath DNAmAge was significantly correlated with chronological age in both groups (EUR: β = 0.893, SE = 0.002, p < 2e-16, $R_{adj}^2$ = 0.889; AFR: β = 0.927, SE = 0.004, p < 2e-16, $R_{adj}^2$ = 0.876). The residuals from this model were retained, representing age adjustment. In the full regression model predicting age adjustment, including covariates of sex, seven cell types, and smoking status, PTSD case status was significantly associated with accelerated aging (age adjustment > 0) in EUR (β = 0.266, SE = 0.0658, p = 5.46e-05; Supplementary Tables 6, 7). There was no significant association in AFR. Sex was also a significant predictor of age adjustment. Female sex was associated with decelerated aging (age adjustment < 0) in both EUR (β = −0.750, SE = 0.132, p = 1.38e-08) and AFR (β = −0.437, SE = 0.148, p = 3.22e-03). Each of the seven cell types was significantly associated with age adjustment in EUR and AFR. In the full regression models (case status and covariates [sex, cell composition, and smoking status]), the was 0.0408 for EUR and 0.0570 for AFR. DNAmAge and age adjustment were not analyzed in AMR because of small sample size.

Methods

Cohorts and PTSD definitions

We considered five PTSD traits—two diagnostic traits (PCL-Total and Lifetime PTSD) and three symptom subdomain traits (REX, AVOID, and HYP). PCL-Total is the PCL-17 for DSM-IV used to assess PTSD symptoms over the past month by self-report; scores range from 17 to 85 [33]. The other three scores are subscales of the PCL-17. REX is the sum of questions 1–5; the range is from 5 to 25. AVOID is the sum of questions 6–12; the range is from 7 to 35. HYP is the sum of questions 13–17; the range is from 5 to 25. The Lifetime PTSD variable used the case and control outputs from lasso regression with cross-validation based on manual chart review [34]. The lasso algorithm classified Lifetime PTSD by selecting the category with the highest probability from the following: case = PTSD case, possible = possible PTSD case, and control = PTSD control. We only used the cases and controls due to their certainty and did not use the “possible” category. Ancestry was aggregated into population groups—EUR, AFR, and AMR—based on genetic similarity to the 1000 Genomes Project reference panel [35]. For PCL-Total, REX, AVOID, and HYP, there were a total of 22,141 samples each, 17,674 for EUR, 3,430 for AFR, and 1,037 for AMR. For Lifetime PTSD, there were 9,413 cases and 27,257 controls, with 5,184 cases and 19,382 controls for EUR, 3,095 cases and 6,240 controls for AFR, and 1,134 cases and 1,635 controls for AMR.

DNA methylation quality control

The following quality control was done by the MVP epigenetics group [36]. Samples for the MVP methylation data were selected based on enrichment of phenotype and advanced age due to higher prevalence of diseases relevant to veterans. Raw fluorescence intensity data (IDAT files) were converted to methylation beta values. SeSAMe was used for processing the methylation data and for quality control with their P-value with Out Of Band probes for Array Hybridization (pOOBAH) [37]. In the probe quality control, a given probe was removed if it had a pOOBAH detection p-value ≥ 0.05 in > 10% of samples. There were 865,918 total probes on the Illumina Infinium Methylation EPIC v1 array, which was run on 45,500 whole-blood samples from the MVP cohort to generate DNA methylation profiles. Next, 862,927 probes were retained after removing non-CpG probes. Probes with pOOBAH failure rate in > 10% of samples were removed, leaving 836,196 probes. Technically unreliable probes were removed due to poor mapping quality, poor titration correlation, or color channel switching, leaving 768,569 probes. The final number of CpG probes in the MVP data release was 768,569[36], yielding a p-value threshold of 6.51e-08 (0.05/768,569) for epigenome-wide significance.

Epigenome-wide association analysis

The meffil R package [38] was used to fit the EWAS analyses, with methylation betas as outcome, using as covariates age, methylation age (from Horvath’s clock) [39], sex, cell type proportion (Houseman) [40], 20 genotype principal components, 20 technical principal components, scanner ID, and duration of sample storage. Cell types included B cells, CD4+ T cells, CD8+ T cells, eosinophils, monocytes, neutrophils, and natural killer cells. In the smoking status sensitivity analyses performed on all traits, smoking status was added as a covariate. Smoking status was defined using the lasso algorithm, which returns the probability of being a never, former, or current smoker; the category with the highest probability was selected as the smoking value for each individual [41]. The R package Bacon [18] was used to control inflation. Bacon works by constructing an empirical null distribution using a Gibbs sampling algorithm, fitting a three-component normal mixture on z-scores. The genomic inflation factors were closer to 1 after doing so (Supplementary Table 2). Epigenome-wide significance was defined as a Bonferroni-corrected empirical p-value < 6.51e-08 (0.05/768,569 tests). Using METAL [42], we performed sample-size-weighted multi-ancestry meta-analyses of our EWAS results with smoking included as a covariate for all five PTSD traits; effective sample size was used for Lifetime PTSD. Further, smoking status stratification was applied to all the PTSD EWAS in EUR and AFR to see the association between AHRR methylation and PTSD using the significant AHRR CpG sites in each of the EWASs. Further, within cases of Lifetime PTSD, we regressed Horvath age (DNAmAge) on chronological age, with covariates of sex, cell type proportion, and smoking status, and found that the mean epigenetic age acceleration was significantly > 0 (p-value = 0.042) based on applying a one-sample t-test of the residuals of the model for PTSD cases (see supplementary materials for model fit).

DNAm Adjusted Age

Horvath DNA methylation age (DNAmAge) estimates were calculated using 353 CpG sites in 8,000 samples, across 51 healthy tissues and cell types [39]. We then regressed DNAmAge onto chronological age at time of initial blood draw and retained the residuals to reflect age adjustment. To investigate the relationship between algorithmically defined PTSD case status (EUR: 5,184 cases and 19,382 controls; AFR: 3,095 cases and 6,240 controls) and age adjustment, we included case status and covariates of genetically defined sex, seven cell types, and smoking status in a linear regression.

Brain CpG and blood EWAS overlap

Differentially methylated CpG sites (DMCs) associated with PTSD were identified using targeted methyl-seq data from human postmortem brain, spanning subregions of the amygdala (medial, basolateral, central) and hippocampus (subiculum, dentate gyrus, CA subfields) as reported previously [43]. DMCs were called with the DSS package from Bioconductor [44], fitting a beta-binomial model with an arc-sine link function, with diagnosis group as the variable of interest and brain subregion, age, sex, postmortem interval, tissue source, smoking status, neuronal proportion, and ancestry included as covariates. By adjusting for brain subregion, we conducted a cross-region analysis designed to identify PTSD-associated methylation differences while accounting for regional variation in DNA methylation. This approach increased statistical power by leveraging data across all six brain regions and prioritized signals with evidence of association across PTSD-relevant subregions of the amygdala and hippocampus. CpGs with a Benjamini-Hochberg-adjusted FDR < 0.05 were considered significant and annotated to the nearest gene. Based on the current blood-based analyses and the previously reported brain-based analyses [43], overlap with blood EWAS results was evaluated at both the CpG and gene level. At the CpG level, brain CpG coordinates (hg38) were compared with blood EWAS CpG coordinates to identify exact matches and loci within ±1 kb, using the findOverlaps() function from the GenomicRanges package[45]. At the gene level, genes annotated to significant brain CpGs were intersected with the blood EWAS gene list by gene-symbol matching.

Enrichment Analyses

We performed gene set enrichment analysis with Enrichr [19–21]. Enrichr uses Fisher’s exact test, calculates the odds ratio (OR) to determine whether genes from the input list appear in a gene set more often than would be expected by chance, and applies the Benjamini-Hochberg correction to account for multiple testing across all tests performed in the run. It then calculates an overall enrichment score equal to the negative log of the p-value multiplied by the OR [19–21]. The input gene list of 186 genes included 1 gene from a trans-ancestry meta-analysis of lifetime PTSD, 135 additional genes from a trans-ancestry meta-analysis of PCL-Total, 8 additional genes from a trans-ancestry meta-analysis of REX, 41 additional genes from a trans-ancestry meta-analysis of AVOID, and 1 additional gene from a trans-ancestry meta-analysis of HYP. All meta-analyzed EWAS included smoking status as a covariate. Significant CpG sites from the meta-analyses were mapped to genes using the R package IlluminaHumanMethylationEPICanno.ilm10b4.hg19 [46]. Significant CpG sites of interest, those mapping to genes found both in our input gene list and in brain- or PTSD-relevant gene sets, were annotated using CpG Atlas [47].

Discussion

In this study, we performed EWAS of PTSD across multiple ancestries within what is, to our knowledge, the largest single cohort available for DNAm data. We include 9,413 cases with a total of 36,670 participants for Lifetime PTSD Values; in the previous largest PTSD EWAS, there were 2,156 PTSD cases with a total of 5,077 participants. This U.S. military sample is also comparatively homogeneous. Furthermore, we include 22,141 individuals who completed the PCL-17 to give scores for total PCL, re-experiencing, avoidance, and hyperarousal, which enables us to have greater power for EWAS of PTSD based on their total severity score and symptom domain subscales. In the PTSD EWAS of blood DNAm levels, we identified a total of 241 unique CpG sites associated with PTSD and 192 genes, replicating 5 CpG sites that were identified in a prior meta-analysis of PTSD EWAS. The 5 replicated CpG sites all map to AHRR or are intergenic. The identified PTSD-associated CpGs had epigenome-wide significance within 2.92e-34 ≤ p < 6.51e-08. The most significant CpGs were mapped to AHRR (cg05575921, p = 2.92e-34 and cg21161138, p = 9.88e-21), F2RL3 (cg03636183, p = 1.50e-21) and RARA (cg17739917, p = 2.56e-19), or were intergenic (cg21566642, p = 2.67e-21, cg01940273, p = 1.52e-19. AHRR is well known to be affected by smoking [32, 48] and all of our effect sizes were reduced when smoking status was added as a covariate to the EWAS analyses. The genes have been implicated in smoking (AHRR, F2RL3), inflammation and immune disorders (GPR15, RARA, ITGAL), regulation of hematopoietic cell differentiation (ASB2), myocardial infarction and platelet function (F2RL3), as potential targets for gastric or breast cancer therapy, tumor suppression (MAN1C1, DENND2D), cancer/tumor progression and drug resistance (PRSS23, RARA, SLC1A5, ITGAL), and metabolic and bone disorders (SLC1A5).

Findings differed by ancestral population. The total number of unique associated CpGs was 238 in EUR, 2 in AFR, and 1 in AMR; the large discrepancy is likely due to much smaller sample size in AFR and AMR. In AFR, cg25322382 (p = 2.96e-08 for PTSD-Dx) was significant and mapped to DDI2, and cg10213873 (p = 9.15e-09 for PTSD-Dx) was significant and intergenic. The significant CpG uniquely identified in AMR was cg22036458 (intergenic; p = 6.30e-08 for PCL-Total and p=3.83e-08 for REX). There was no overlap of significant CpGs between AFR and AMR.

Compared to the prior largest PGC PTSD EWAS [11], we also identified significant results for 5 of 11 CpGs reported as significant in the latest investigation: cg19719391 (Intergenic), cg14753356 (Intergenic), cg05575921 (AHRR), cg21161138 (AHRR), and cg23576855 (AHRR). We also identified cg26703534 (AHRR) in PCL-Total, REX, AVOID, and HYP and cg25648203 (AHRR) in PCL-Total and AVOID, not identified in the latest PTSD PGC paper [11] but significant in the earlier PTSD PGC paper [13].

A challenge in the interpretation of EWAS results is relating findings in peripheral blood to methylation patterns in the brain, the preferred target tissue. However, there is often correspondence in epigenetic findings between brain and peripheral blood [49]. Genes associated with significant CpGs in both the blood and the previously reported brain analyses [43] were SLC25A34, EXOC2, LIF, and NFE2L3. The protein products of these genes affect aspects of metabolic regulation—including mitochondrial substrate transport (SLC25A34), vesicle exocytosis (EXOC2), cytokine signaling (LIF), and stress-responsive transcription (NFE2L3). In a CpG island intronic to LIF, leukemia inhibitory factor interleukin 6 family cytokine, we found a brain CpG and a blood CpG (cg23635663) 151 bp apart. LIF is neuroprotective, anti-inflammatory, and promotes neural stem cell maintenance and tissue repair [50]. At EXOC2, the blood and brain DMCs were also separated by less than 1 kb, providing a second example of local cross-tissue convergence. Although these genes do not constitute a recognized molecular pathway, their identification in EWAS of two biologically distinct tissues provides evidence of cross-tissue convergence and supports their prioritization for further investigation in PTSD.

Enrichment analyses revealed enrichments in multiple brain cell types, including somatostatin interneurons and brain endothelial cells, which accords with recent scRNAseq analyses of brain tissue of individuals with PTSD [51]. Further research is needed to determine whether genes downregulated in astrocytes, also found to be enriched here, are similarly enriched in PTSD brain tissue, although there is some evidence of astrocyte involvement in animal models of PTSD [52]. The involvement of the glucocorticoid receptor points to the role of the stress response in PTSD [53–55], as does the behavior of other differentially methylated genes in response to cell stressors such as UV radiation, NO₂, and alcohol. The CpG site we identify in NR3C1, cg22233604, lies in the body of the gene, not in the promoter region; it is therefore not clear that the hypermethylation of this site in our results necessarily means deactivation of the gene. A positive EWAS regression coefficient at this site was previously observed in a study of Vietnam veterans from Australia and New Zealand [56]. Finally, the association of PTSD and DNMT3A, which is hypomethylated and thus possibly more expressed in patients with PTSD, requires further research to characterize the link between PTSD and the methylation machinery.

In the smoking status stratification between former, current, and never smokers for EUR, AFR, and AMR, we observed no significant associations between PTSD and any of the previously identified AHRR CpG sites within any smoking stratum. This contrasts with findings in prior PGC PTSD EWAS analyses [13] that associations between AHRR methylation and PTSD were most prominent among non-smokers and absent among smokers, leading to the interpretation that AHRR methylation reflected PTSD-related biology in addition to smoking exposure; in that prior report, the AHRR association was interpreted to reflect PTSD in addition to smoking. In contrast, our results suggest that observed associations between PTSD and AHRR DNA methylation reflects smoking-related effects rather than PTSD-specific epigenetic changes. Cohort characteristics differed between studies: former smokers comprised the majority of both PTSD cases and controls in the MVP, whereas the PGC PTSD sample had a higher prevalence of current smoking among PTSD cases. Our sample sizes were also much larger, allowing us to stratify three smoking statuses of “current,” “former,” and “never,” whereas they had two categories of smokers and non-smokers. To summarize, our conclusions regarding AHRR and PTSD differ substantially from previous reports; we conclude that the observed association most likely reflects smoking, not PTSD.

We examined algorithmically defined PTSD case status as a predictor of residuals derived from regressing chronological age onto Horvath DNAmAge, or age adjustment. PTSD case status was significantly associated with age adjustment > 0, or accelerated aging (on average, 0.266 years) in EUR ancestry when controlling for sex, smoking status, and cell type proportion. Although the same direction of effect was identified in AFR ancestry, it was not statistically significant. These results are consistent with prior reports demonstrating significant age acceleration associated with PTSD [57, 58].

Our study has limitations. Our U.S. military veteran sample was mostly male (95.3% in EUR, 89.6 in AFR, 90.6 in AMR); findings in females may differ. While our study has comparatively good representation of non-EUR individuals (67.0% EUR, 25.5% AFR, and 7.6% AMR for Lifetime PTSD; 79.8% EUR, 15.5% AFR, and 4.7% AMR for PCL scores), the lower number of non-EUR limited discovery in those populations. Study of blood rather than brain is a limitation for linking epigenetic variants to regulation of functionally relevant genes in the brain, but an advantage for possible future clinical applications. For any clinical use of these data in living patients for, e.g., treatment personalization, results are useful only if they can be obtained in readily-accessible tissues. Further, this study was cross-sectional, with a mean age of blood draw at approximately 65.4 years of age. Most participants were diagnosed with PTSD many years before the measurement of methylation levels by array, i.e., we likely detected chronic rather than acute methylation differences. The methylation age analyses focused on Horvath DNAmAge, providing initial insight into the relationship between DNAmAge and PTSD; however, future research should compare across alternative methylation age clocks and take into consideration other factors that may impact DNAmAge particularly within the context of PTSD (e.g., comorbid conditions or impact of PTSD treatment on DNAm age acceleration).

PTSD is essentially linked to environmental exposure, so epigenetic mechanisms may be particularly relevant because they can capture persistent molecular correlates of trauma and stress. Epigenetic variation may help explain why individuals with similar trauma exposure differ in PTSD susceptibility and symptom severity beyond differences attributable to genetic risk alone. Our work represents the largest-to-date epigenetic analysis of PTSD, with the identification of many novel epigenetically-modified loci associated with this trauma-related trait and the discovery of new biology. Our sample was less heterogenous, across multiple dimensions, than prior meta-analyzed samples. In the largest single-cell genomics analysis of human brain detailing PTSD [51], genes and pathways associated with PTSD included immune and neuroinflammatory mechanisms, which overlap with our EWAS study’s interpretations as well as those of the prior literature [11–13]. We identify convergence between blood and postmortem brain methylation findings and implicate brain-relevant cellular programs together with accelerated DNA methylation aging. These findings substantially extend the molecular characterization of PTSD while illustrating the importance of distinguishing PTSD-associated epigenetic variation from correlated environmental exposures, e.g., smoking. This work may eventually have clinical utility, by allowing the identification of novel pharmacological targets for PTSD, a disorder for which there presently are few pharmacological treatments—and those that exist are not highly effective.

Figure 1:

Miami plot for PCL-Total in EUR with positive associations on top and negative on the bottom. Statistics were adjusted for inflation and bias using Bacon [18].

Figure 2:

Miami plot for PCL-Total in EUR with smoking status as a covariate. Positive associations are on top and negative on the bottom. Statistics were adjusted for inflation and bias using Bacon [18].

Table 1:

Demographics of PTSD traits

	Sample Size	Score (1st quartile, 2nd quartile, 3rd quartile) or numbers of cases and mean + SD	Average Age	Males/Females (percentage)
	EUR
PCL-Total	17,674	(19, 24.8, 37)	69.3	95.3% males, 4.7% females
Re-experiencing	17,674	(5, 6, 10)	69.3	95.3% males, 4.7% females
Avoidance	17,674	(7, 10, 15.2)	69.3	95.3% males, 4.7% females
Hyperarousal	17,674	(6, 8, 12)	69.3	95.3% males, 4.7% females
Lifetime PTSD positive	24,566	21.1% cases and 78.9% controls	68.1	95.4% males, 4.6% female
	AFR
PCL-Total	3,430	(21, 30, 48)	63.3	89.6% males, 10.4% females
Re-experiencing	3,430	(5, 8, 14)	63.3	89.6% males, 10.4% females
Avoidance	3,430	(8, 12, 20)	63.3	89.6% males, 10.4% females
Hyperarousal	3,430	(6, 9, 15)	63.3	89.6% males, 10.4% females
Lifetime PTSD positive	9,335	33.2% cases and 66.8% controls	60.8	89.8% males, 10.2% females
	AMR
PCL-Total	1,037	(21, 33, 52)	63.1	90.6% males, 9.5% females
Re-experiencing	1,037	(6, 9, 15)	63.1	90.6% males, 9.5% females
Avoidance	1,037	(8, 13, 22)	63.1	90.6% males, 9.5% females
Hyperarousal	1,037	(7, 10, 16)	63.1	90.6% males, 9.5% females
Lifetime PTSD positive	2,769	41.0% cases and 59.0% controls	58.1	91.6% males, 8.4% females

Data availability

All MVP and meta-analysis summary statistics are made available through dbGAP, https://dbgap.ncbi.nlm.nih.gov, under accession number phs001672.

Code availability

Code for all software packages used in this analysis is publicly available via the citations for each method.