New DNA methylation sites linked to PTSD found in studies of 23 military and civilian groups

Biomarkers, social, and affective predictors of suicidal thoughts and behaviors in adolescents (BEAR) [34] involved a sample of 194 adolescents who had been hospitalized for suicidal thoughts/behaviors. Of those, 163 provided a blood sample, and 162 samples that passed the DNAm quality control (QC) were included in the meta-analysis. PTSD diagnosis was assessed by the Clinician Administered PTSD Scale (CAPS) for Children and Adolescents for DSM-5 [35].

Detroit Neighborhood Health Study (DNHS) [23] involved 1547 participants whose PTSD symptoms were assessed by the PTSD checklist (PCL-C) [36] at the baseline wave. The meta-analysis included 523 participants with available DNAm data that passed the DNAm QC.

Drakenstein Child Health Study (DCHS) [37] is a population-based birth cohort that recruited 1000 pregnant people between 20–28 weeks gestation. DNAm data was available for 98 participants, and 95 samples that passed the DNAm QC were included in the meta-analysis. For the purpose of this study, PTSD was assessed using The Mini International Neuropsychiatric Interview (MINI) [38, 39].

Grady Trauma Project (GTP) is a large-scale ongoing study with > 10,000 participants. The meta-analysis included 744 participants with available DNAm data that passed the DNAm QC. Current PTSD was assessed using the Clinician-Administered PTSD Scale for DSM IV (CAPS-4) [40, 41] or the modified PTSD Symptomatic Scale (PSS) [42].

Northern Illinois University Trauma Study (NIU) involved 812 participants recruited to participate in a federally-funded study (NIH 5R21MH085436-02) to examine risk and protective factors following the NIU campus shooting on February 14th, 2008. Of those, 140 provided blood samples and were included in the meta-analysis. Current PTSD diagnosis at the first post-shooting assessment was assessed by self-report on the Distressing Events Questionnaire (DEQ) [43].

Shared Roots SHRS or “Understanding the SHARED ROOTS of Neuropsychiatric Disorders and Modifiable Risk Factors for Cardiovascular Disease” is a matched case–control study (N = 120) examining the factors that contribute to the comorbidity of metabolic syndrome and neuropsychiatric disorders [44]. The CAPS-5 [45] was administered by clinicians to assess PTSD over the prior month.

Advancing Understanding of RecOvery afteR traumA (AURORA) is a large cohort study of women and men presenting to the ED within 72 h after exposure to psychological trauma [46]. PTSD diagnosis at 6 months was defined using the PTSD Checklist for DSM-5 (PCL-5) [47–49]. The meta-analysis included a subset of the AURORA cohort with available phenotypic, DNA methylation, and RNA sequencing data at the 6-month follow-up after trauma exposure (N = 206).

Human Heredity and Health in Africa, Rwanda (H3A_Rwanda). Data from the H3A_Rwanda cohort were obtained from a subset of participants in a previously published pilot study (n = 50) [50, 51] and newly recruited participants via support from the H3Africa Consortium (n = 40) [52]. All study participants were women of Tutsi ethnicity who were pregnant during the genocide. PTSD was assessed by PCL-17 [53] for the pilot study and by PCL-5 [54] for the H3Africa-affiliated study. The meta-analysis included 73 participants with available phenotype and DNAm data.

World Trade Center Responders (WTC) [55, 56]. PTSD in relation to WTC exposures was assessed by the Structured Clinical Interview for DSM-IV Disorders (SCID) [57]. The sample (N = 180) providing blood samples for the epigenetics assays was restricted to men (the vast majority of the responders) and oversampled for posttraumatic stress disorder (PTSD) [25].

Gallipoli Medical Research Foundation Queensland University of Technology (GMRFQUT) is a large cohort of veterans who have been or are currently being treated for PTSD at the Keith Payne Unit within the Greenslopes Private Hospital in Queensland, Australia [58]. The CAPS-5 [40, 41] was used to assess current PTSD. The meta-analysis included 96 participants with available DNAm data.

Marine Resiliency Study (MRS). In the MRS [59, 60], PTSD was diagnosed by CAPS-5 [40, 41] up to 3 times, once before deployment and 3 and/or 6 months post-deployment. Samples of PTSD cases (N = 127) were selected from the 3- or 6-month post-deployment visits depending on which visit had the highest CAPS score. Combat-exposed controls with low to no PTSD-symptoms (N = 126) were selected from post-deployment visits, matching for age, ancestry, and time of post-deployment visit.

Prospective Research in Stress-related Military Operations (PRISMO) is a large prospective study of Dutch military soldiers [61, 62]. Current PTSD symptoms were assessed using the Self-Report Inventory for PTSD (SRIP) [63], and blood samples were collected approximately 1 month before and at both 1 and 6 months after deployment. PTSD cases (N = 56) were selected for this DNA methylation study from the 1- or 6-month post-deployment visits, depending on which visit had the highest SRIP score. All controls (N = 95) were combat exposed and had low PTSD symptoms [26].

The Army Study to Assess Risk and Resilience in Servicemembers (Army STARRS) is a multicomponent prospective study among US Army personnel [64]. PTSD diagnosis was assigned using multiple imputation methods that relied on PCL and CIDI-SC data [65]. Whole blood for methylation assays was collected approximately 6 weeks before deployment and 1-month post-deployment. PTSD cases (N = 157) were selected based on their PTSD diagnosis at 6 months post-deployment. Controls (N = 161) were participants without PTSD, matched on age, deployment stress, and childhood adversity.

PROlonGed ExpoSure and Sertraline Trial (PROGrESS) is a randomized-controlled trial (RCT; N = 223) designed to examine the comparative effectiveness of multiple treatment strategies across 24 weeks. PTSD is assessed by the CAPS-IV [40, 41]. This study included 112 PTSD cases and 28 trauma-exposed controls selected from the pre-treatment visit.

Boston VA National Center for PTSD/Translational Research Center for TBI and Stress Disorders (NCPTSD/TRACTS) included participants from the NCPTSD study [66], the PTSD and Accelerated Aging study [67], and the TRACTS study [68]. For all three studies, PTSD diagnosis was determined based on the CAPS for DSM-IV [40] or DSM-5 [69]. The NCPTSD and TRACTS cohort DNAm (N = 1028) data were jointly cleaned and analyzed together.

Injury and Traumatic Stress (INTRuST). This cohort included participants in studies of the INTRuST Consortium [70]. PTSD diagnosis was assessed by study-specific measures, including CAPS-IV, CAPS-5 [40, 41], and PCL-17 [53]. The meta-analysis included 303 participants with available phenotype and DNAm data.

Mid-Atlantic Mental Illness Research Education and Clinical Center PTSD Study (VA-M-AA & VA-M-EA) [71–73]. PTSD was diagnosed using the SCID [57]. The meta-analysis included 369 participants from VA-M-AA and 87 participants from VA-M-EA with available phenotype and DNAm data.

The sample is heterogeneous in terms of sex (41% female), ancestry (46% European, 43% African, and 11% of other ancestries), and cohort type (56% military cohort). Civilian cohorts skew towards being more female (67%) and African (70%), whereas military cohorts are predominantly male (80%) and European (64%). In addition, the prevalence of PTSD is higher in military cohorts (54%) than in civilian cohorts (28%).

All participants were exposed to a traumatic event and 42% met the criteria for current PTSD. The current PTSD diagnosis was assessed by each study following the standardized guidelines established by the PGC-PTSD Workgroup [7]. Briefly, current PTSD diagnosis was determined based on the specific criteria set by the principal investigator of each study. Participants without current PTSD but with a history of PTSD (i.e., remitted PTSD), were excluded. Detailed descriptions of cohorts and PTSD assessments are provided in the Additional file 1: eMethods. All participants in these studies gave informed consent. The institutional review boards of each respective institution approved these studies.

Whole blood DNAm was measured using the Illumina MethylationEPIC BeadChip (EPIC array) in 14 studies, and the HumanMethylation450 BeadChip (450 K array) in 9 studies (Table 1). All studies used a standardized consortium-developed QC pipeline that differed somewhat depending on which chip was used. The 450 K array pipeline [29, 31] is described in Additional file 1: eMethods.

The EPIC pipeline (available at https://github.com/PGC-PTSD-EWAS/EPIC_QC↗) [74] was similar to the 450 K array pipeline. Samples with probe detection call rates lower than 90% and average intensity values that were either less than 50% of the overall sample mean or below 2000 arbitrary units (AU) were excluded. Probes with detection p-values > 0.01 were considered low quality and treated as missing. Probes that were missing in > 10% of the samples within the studies and were cross-hybridizing were removed [75]. Data was normalized using single-sample Noob (ssNoob) implemented in R package minfi [76]. ComBat was used to account for batch effects of chip and position while preserving PTSD, age, and sex effects (if applicable) [77]. Blood-cell composition (i.e., the proportion of CD8 + T, CD4 + T, natural killer (NK), B cells, monocytes, and neutrophils) was estimated using the robust partial correlation (RPC) method in Epidish [78] with a reference data specific to EPIC array [79]. For studies without genome-wide genotype data (DNHS, NIU, Shared Roots, AURORA, H3A_Rwanda, GMRFQUT, PROGrESS), we estimated ancestry principal components (PCs) from DNAm data, using the method developed by Barfield et al. [80], as previously described [33]. PCs 2 and 3, which were the components that correlate most with self-reported ancestry, were included as covariates [33, 80]. In cohorts with available genome-wide genotype data, PCs 1–3 from GWAS were used to adjust for ancestry. We used R package bacon to control inflation, only if doing so results in the lambda being closer to 1 [81]. To predict smoking status, a DNAm-based smoking score was calculated, as previously described [27] for cohorts with EPIC array data. A detailed description of DNAm-based ancestry PC and smoking score calculation is provided in Additional file 1: eMethods.

The association between PTSD and DNAm was tested using multivariable linear regression models for cohorts with balanced plate designs. For studies in which plate layouts were not balanced (Shared Roots, H3A_Rwanda, GMRFQUT), we conducted mixed-effect regression models, including chip as a random effect term. The CpGassoc R package was used to fit the models [82]. The models were adjusted for age, sex (if applicable), blood cell composition (i.e., CD8T, CD4T, NK, B cell, and monocyte cell proportions), and ancestry PCs. While not included initially because of concerns about multicoliniarity and collider bias, a post hoc sensitivity analysis was performed including a covariate for smoking: DNAm-based smoking score in studies with EPIC data and current smoking status for studies with 450 K data. Furthermore, we conducted stratified analyses for both sexes, ancestry (European and African ancestry), and cohort type (civilian and military cohorts).

To combine results across studies, we performed inverse-variance weighted (IVW) meta-analysis in meta [83]. Meta-analysis tested 411,786 CpGs common to 450 K and EPIC arrays (23 studies), and 404,794 EPIC array specific CpGs (14 studies). Epigenome-wide significance threshold recommended for the MethylationEPIC BeadChip (p < 9.0e − 08) was used to determine statistical significance [84]. Gene Ontology (GO) enrichment analyses were conducted using the top 1000 CpGs in missMethyl [85]. A false discovery rate (FDR) threshold of 5% was used to identify significant GO terms.

Correlations between PTSD-associated CpG sites’ DNAm levels and expression levels of the corresponding gene (as determined by the EPIC v1 array annotation) were tested in whole-blood RNA-sequencing (RNA-seq) data from participants in the BEAR (n = 127), AURORA (n = 173), NCPTSD Merit (n = 204), and MRS cohorts (n = 128 with multiple visits totaling 357 samples). The results were meta-analyzed using the IVW method using a Bonferroni correction for the CpGs examined. Detailed information about cohort-specific RNA-seq data generation is described in Additional file 1: eMethods.

To evaluate the effect of nearby (< 1 MB) polymorphisms on DNAm levels of CpGs associated with PTSD, we used cis-methylation quantitative trait locus (cis-meQTL) data from GoDMC [86] and meQTL EPIC [87] databases. For both databases, their default multiple testing adjustment was utilized: an FDR threshold of 5% in meQTL EPIC and p < 1e − 08 in GoDMC. We checked the associations between the identified cis-SNPs and PTSD in the recent Freeze 3 GWAS from PGC-PTSD [11]. Finally, we evaluated genetic associations between the genes with PTSD-associated DNAm and PTSD, using the gene-based test results from the recent PGC-PTSD Freeze 3 GWAS [11].

Of the 11 PTSD-associated CpGs, 9 CpGs were annotated to a gene expressed in blood. Meta-analysis across 4 cohorts with RNAseq data identified 5 CpGs whose methylation levels were correlated with their annotated gene expression (p < 0.05), but only 3 CpGs remained significant after multiple test corrections for 9 CpGs (p_Bonferroni = 0.05/9 = 5.5e − 03). Specifically, methylation of cg05575921, cg21161138, and cg23576855 associated with AHRR expression (p < 5.5e − 03, − 16.69 ≤ z ≤ − 4.86; Additional file 1: Table S2). These findings highlight the potential regulatory impact of these CpGs on gene expression in the context of PTSD. Evaluation of the top 1000 CpGs did not result in any significant Gene Ontology enrichments.

When examining the possibility of genetic-epigenetic effects, we noted that 8 CpGs were associated with at least one SNP within 1 MB according to GoDMC [86] and meQTL EPIC [94] databases (Additional file 1: Table S5). In total, we identified 26 lead meQTLs that were tested for association in the latest PGC-PTSD GWAS [11], of which 8 were nominally associated with PTSD (p < 0.05; Additional file 1: Table S5). For instance, lower methylation at the intergenic cg14753356 site is associated with PTSD (z = − 5.72, p = 1.09e − 08, Table 2) and higher methylation at cg14753356 is associated with rs28986310 T allele (beta = 0.31, p < 5e − 324), which increases the risk of PTSD (z = 6.73, p = 1.68e − 11, Additional file 1: Table S5). These data collectively suggest that some CpGs are under environmental influence, some are under genetic influence, and some are influenced by both genes and the environment. It is interesting to note that methylation of some CpGs, such as cg04987734 in CDC42BPB, did not appear to change in response to stress or to associate with underlying genetic variation. However, CDC42BPB did associate with PTSD in the recent PGC-PTSD gene-based analysis (z = 3.20, p = 6.94e − 04, Additional file 1: Table S6), suggesting that there are other mechanisms or perhaps tissue-specific regulation, underlying its association with PTSD.

We also investigated the cell type-specific expression patterns of the 7 genes identified in our study across various blood and brain cell types, using the Blood and Brain Atlas [95, 96]. The genes for which we observed a correlation between CpG methylation and expression in blood (AHRR, TOLLIP, BANP), and where methylation patterns were correlated between blood and brain (AHRR, BANP), were expressed across various immune cell types and brain regions. CHD5, GFI1, ADCY4, and CDC42BPB are expressed, albeit at low levels, across different immune cell types, and at high levels across all brain regions. In our study, the methylation of CpGs in these genes did not correlate with their expression in blood. Additionally, the methylation patterns of CpGs in these genes differed between blood and brain, potentially because these genes are not expressed in many blood cells.

To evaluate whether the PTSD-associated CpGs from the blood-based analysis were also associated with PTSD in the brain regions of dlPFC, vmPFC, amygdala, and DG, we first examined overall patterns of association (p < 0.05) between blood-based PTSD-associated CpGs and those associated with PTSD in each of the respective brain regions using hypergeometric tests in probes with > 10% variability across tissues. The number of nominally significant CpGs associated with PTSD in both blood and the amygdala (p = 0.009) is more than expected by chance (Additional file 1: Table S7).

Next, we evaluated the specific PTSD-associated CpGs from blood-based analyses. Six CpGs were nominally associated with PTSD in at least one brain region (p < 0.05, Fig. 2, Additional file 1: Table S8). Notably, many of the CpG sites that changed in response to that naturalistic stress model associated with PTSD in at least one postmortem brain region (i.e., cg21161138 in AHRR, cg25320328 in GFI, and intergenic cg19719391). However, only cg04987734 (CDC42BPB) in the dlPFC and intergenic cg19719391 in the DG remained associated after applying a Bonferroni correction for the 11 CpGs examined (p < 4.5e − 3, Fig. 2, Additional file 1: Table S8). Specifically, PTSD cases had higher cg04987734 (CDC42BPB) methylation both in the blood (p = 3.26e − 8) and the dlPFC (p = 3.9e − 3), and higher and intergenic cg19719391 methylation both in the blood (p = 2.45e − 8) and the DG (p = 3.04e − 3) compared to trauma-exposed controls, suggesting a robust epigenetic alteration associated with PTSD that is detectable across different tissue types.

Finally, to more specifically examine the location of PTSD-associated CpGs in the brain, we leveraged data from FACS-sorted neuronal nuclei from the OFC of PTSD cases and controls. Of the 5 CpGs that associated with PTSD in any brain region from the bulk tissue, only CpGs in AHRR (cg05575921, cg21161138) appeared to differ in those with PTSD versus controls (Additional file 1: Table S9), suggesting that the other brain-based associations may be driven by other cell types, such as glia.

We identified 3 GO term enrichments in the female-stratified analysis (FDR < 0.05), including nervous system development (Additional file 1: Table S11). We did not identify any significant enrichments in other strata. These findings suggest that PTSD-associated DNAm patterns related to nervous system development may exhibit sex-specific patterns. The absence of significant GO term enrichments in the main analysis or other stratified analyses highlights the potential importance of considering sex-specific factors in epigenomic studies.

To our knowledge, this is the largest EWAS of PTSD to date. Our sample is diverse in terms of sex, ancestry, and cohort type. We leveraged data from postmortem brain samples, a cellular model of prolonged stress, GWAS, and genome-wide expression data to support our findings. However, the study is not without limitations. First, methylation arrays only assess a subset of CpG sites in the genome; therefore, we may not capture all PTSD-associated CpG sites. Second, this is a cross-sectional study of participants with prior exposure to a traumatic event; thus, we were not able to assess whether the differences in DNAm between individuals with and without PTSD are a cause or consequence of PTSD or both. Third, our primary meta-analysis was performed using measures of blood DNAm. While this strategy provided valuable insights for future research on biomarkers of PTSD, it might not accurately represent the DNAm patterns within other tissues that are likely the most relevant to PTSD. However, the majority of the PTSD-associated CpG sites’ methylation levels were correlated between blood and at least one brain region. In addition, most PTSD-associated CpGs in blood were also associated with PTSD in one or more brain regions. Fourth, we do not have cell-type-specific DNAm and gene expression information necessary for further examination of cell-specific gene regulation across blood and brain cell types. Hence, we used bulk tissue and adjusted for cellular heterogeneity, which might have obscured some signals, given the alterations in cell composition in those with PTSD [113, 114]. Additionally, the brain regions examined varied between the online databases used to examine the blood–brain correlation of methylation values and the gene expression differences associated with PTSD, which can complicate interpretation. Finally, most cohorts that participated in the meta-analysis did not have detailed physical or psychiatric information on participants, including detailed information on chronicity, trauma type and timing, PTSD symptom course, and treatment, making it challenging to evaluate and adjust for potential confounders, including substance use, comorbidities, or medication use. The lack of detailed information on the types and timing of traumatic experiences across cohorts prevents us from making definitive conclusions about the impact of different trauma types on DNAm. Future studies are warranted to examine DNAm changes longitudinally, tracking participants both before and after trauma exposure to capture the dynamic epigenetic modifications that occur in response to traumatic events, providing a clearer understanding of how trauma influences DNAm over time.