Long-term DNA changes linked to PTSD in three groups of male soldiers

Post-traumatic stress disorder (PTSD) is a debilitating psychiatric disorder that can develop following direct or indirect exposure to a potentially life-threatening traumatic incident. Symptoms include persistent re-experiencing of the trauma, avoidance behavior, hyperarousal, and negative mood [1]. Although most individuals have the potential to withstand negative effects of trauma exposure on long-term mental health and to recover promptly, some are more vulnerable and at increased risk of developing PTSD. Understanding the molecular and neurobiological underpinnings of this differential susceptibility is currently receiving considerable attention, and epigenetic mediation of environmental influences has been proposed as a potential key mechanism [2–4].

Several epigenome-wide association studies (EWAS) have aimed to identify differentially methylated CpGs in PTSD [5–8]. However, most of these studies are based on association analyses where methylation was assessed at a single time point (cross-sectional), with limited ability to adjust for confounding variables. Only one PTSD study to date reported longitudinal changes in methylation profiles across a period of combat exposure in order to capture changes in DNA methylation over time in relation to phenotypic changes [7]. We made use of the Prospective Research In Stress-related Military Operations (PRISMO) study, which in the present study was used as a replication cohort.

Here, we followed a previously published two-stage design [9] where we first meta-analyzed two longitudinal, USA-based military cohorts in order to identify associations between changes in methylation levels from pre-deployment to post-deployment and the development of PTSD. We then sought replication of our significant findings in PRISMO. In the second stage, we combined all three cohorts in a meta-analysis. This two-stage approach allows us to investigate the robustness of our findings through replication in stage 1, while increasing power for gene discovery by combining all three studies in stage 2. For all three cohorts, DNA methylation data and phenotypic data were collected prior to and following a 4–7-month deployment to an active ware zone in Iraq or Afghanistan. All studies selected PTSD cases and controls at post-deployment and only included subjects without PTSD at pre-deployment. Of the significant CpGs in the second analysis stage, we assessed associations with nearby single nucleotide polymorphisms (SNPs) and gene expression data, and examined correlations between blood and brain methylation status. To the best of our knowledge, this is the largest study aimed at detecting methylation changes associated with the development of PTSD. This prospective and longitudinal analysis permits us to more accurately capture dynamic changes in DNA methylation in relation to PTSD development while minimizing confounding due to intra-individual variability.

Exposure to trauma is a prerequisite for the development of PTSD, yet not all individuals develop PTSD following trauma [12]. The underlying biological mechanisms of this differential susceptibility have not yet been fully identified, and even the largest genome-wide association studies (GWAS) to date explain only a small proportion of the disease liability [13, 14]. Epigenetic changes have been studied as one potential mechanism, but most association studies have used cross-sectional designs which render it impossible to establish causality. Here, we use a more powerful longitudinal design to investigate associations of DNA methylation with post-deployment PTSD status across very similar military cohorts deployed to combat in Iraq and Afghanistan. We started with combining the USA-based MRS and Army STARRS cohorts and sought replication using the previously published Dutch PRISMO study [7]. To maximize power for new discoveries, we also performed a meta-analysis across all three cohorts. The first analysis stage revealed 4 genome-wide significant DMPs and 19 DMRs which were linked to post-deployment PTSD status. One of these DMPs replicated in PRISMO. In the second stage, a meta-analysis of all 3 studies revealed that the replicating DMP and 7 DMRs remained significant, and 2 additional DMPs and 12 DMRs were identified.

The replicating DMP cg05656210 remained the top-ranked significant marker in the second analysis stage. This CpG site is an intergenic site annotated near SPRY4. SPRY4 is a member of the Sprouty proteins which are mainly involved in inhibiting receptor tyrosine kinase (RTK) signaling [15]. Upon activation by growth factor ligands, RTK signaling has a wide variety of downstream effects ranging from the regulation of cell proliferation and differentiation to the modulation of cellular metabolism [16]. One particular receptor involved in RTK signaling, i.e., receptor tyrosine kinase B (TrkB), and its main ligand, brain-derived neurotrophic factor (BDNF), have repeatedly been shown to be affected in stress-related disorders such as depression [17]. Consistently, two independent studies reported decreased mRNA levels of BDNF and TrkB in the prefrontal cortex and hippocampus of individuals who committed suicide as compared to healthy control subjects [18, 19]. Moreover, SPRY4-IT1, a long non-coding RNA derived from the second intron of SPRY4 [20], has been shown to interact with SKA2 [21], a gene that was suggested to be a promising biomarker for suicidal behavior [22, 23], stress susceptibility, and stress-related disorders such as PTSD [23–25]. Finally, SPRY4 was previously found differentially methylated in the blood of patients diagnosed with schizophrenia [26]. Together, these results suggest that alterations within SPRY4 could contribute to psychiatric disorders such as depression, PTSD, and schizophrenia and potentially play a role in suicidal behavior. The question as to whether and how the identified DMP influences the expression of SPRY4 and is involved in these phenotypes still remains to be answered.

The second top significant probe, cg12169700, and one DMR are located within MAD1L1. MAD1L1 is part of the mitotic spindle-assembly checkpoint (SAC) which monitors the proper attachment of chromosomes to the microtubule spindle apparatus, delays the start of anaphase until all chromosomes are properly attached, and in doing so, ensures correct chromosome separation [27, 28]. Malfunctions of the MAD1L1 protein could therefore contribute to aneuploidy and chromosomal instability. Specific SNPs within this gene have previously been associated with bipolar disorder [29, 30], schizophrenia [30–32], and depression [33]. Interestingly, MAD1L1 was recently identified in a PTSD GWAS of the Million Veteran Program (MVP) [34]. The SNP that underlies cg12169700, rs11761270, is located in the same large linkage disequilibrium (LD) block as the MVP finding. In the MVP, carriers of the minor allele of rs11761270 showed decreased levels of methylation and were at increased risk of having PTSD. This corresponds to our own findings in which PTSD cases show a reduction in methylation from pre- to post-deployment. Moreover, using expression data from MRS, we found that methylation at this site was positively associated with gene expression of MAD1L1. This also aligns with previous findings that showed that blood levels of MAD1L1 were decreased in highly stress-susceptible individuals [35]. Together, these findings suggest that specific methylation profiles within MAD1L1 may be regarded as a risk factor for PTSD in addition to several other psychiatric disorders [36]. However, the underlying mechanisms through which such disturbances could contribute to psychiatric disorders warrant further research.

The third CpG site is located in HEXDC which to date has no known implications in any psychiatric disease. Although the biological functions of its product, hexosaminidase D, remain largely unknown, it is believed to be a glycosidase and previous studies found associations with rheumatoid arthritis [37, 38]. Interestingly, recent studies found significant pleiotropy between rheumatoid arthritis and PTSD [39], and rheumatoid arthritis and schizophrenia [40]. Other studies found associations between PTSD symptoms and rheumatoid arthritis in a twin population [41] and in an epidemiological study of military veterans [42]. This link between mental disorders and immune-related processes is discussed more in detail below. The DMP of HEXDC was located directly adjacent to rs4789774, a known expression quantitative trait locus (eQTL) that regulates the expression of HEXDC in the human brain cortex and of NARF and NARF-IT1 in a number of tissue types including blood (http://genome.ucsc.edu/↗). Moreover, a modest negative correlation was found between methylation of this site and gene expression of HEXDC, both at baseline and at post-deployment.

Twelve significant DMRs were found in the second phase of the analysis. Our strongest finding was in the HLA region, a gene-dense region which contains over 200 genes that encode human leukocyte antigen complex proteins in charge of presenting peptide antigens to trigger immune reactions, among other (non-)immune functions [43]. Their non-immunological roles include processes such as neurodevelopment, synaptic plasticity, learning, memory, and stress reactivity [44, 45]. It is therefore not entirely surprising that several epigenetic modifications and genetic variants within this region have repeatedly been found implicated in neuropsychiatric disorders such as schizophrenia and bipolar disorder (recently reviewed in [46]). Along with our HEXDC finding, these observations further enhance the existing notion that immune factors play an important role and should continue to be studied in relation to mental disorders such as PTSD [47]. Although it is now clear that immune imbalances are present in PTSD, questions related to causality and further implications for prevention strategies and treatment options largely remain to be answered.

Follow-up analyses were done using the significant DMPs from the stage 2 meta-analysis only. The discovery that methylation levels at the top three PTSD-associated CpGs were highly associated with the genotype of the nearby SNPs led us to question whether the associations between methylation and PTSD status were mainly driven by genotype. However, direct adjustment for genotype in a sensitivity analysis did not attenuate the associations between DNA methylation and PTSD status and our current sample size limits our ability to conduct analyses specific to genotype strata to further investigate interaction effects between SNPs and methylation.

Since our methylation data were based on DNA from peripheral blood, we further examined correlations between blood and several brain regions, i.e., the prefrontal cortex, entorhinal cortex, superior temporal gyrus, and cerebellum. The results indicate that blood-brain correlations of all top DMPs were strong for all four brain regions suggesting that these findings could potentially also be relevant for tissues other than blood. Assessing these correlations is relevant when dealing with disorders such as PTSD which are characterized by functional and structural alterations within the brain but for which the accessibility to human brain tissue is limited. However, these and similar findings will need to be confirmed using postmortem brain tissue and their precise role in PTSD development will need to be investigated further.

The main limitation of the present study is its small sample size which likely captures only a fraction of all implicated CpGs and renders additional analyses such as pathway and network analyses underpowered. It further needs to be emphasized that this study used data generated with Illumina’s 450K arrays which only assess a subset of all CpG sites. Next, although examining blood-derived DNA methylation is informative when seeking relatively easily accessible biomarkers, follow-up studies are needed in order to assess these methylation patterns within the tissue of interest, i.e., the brain. Another important limitation of the study is the fact that subjects that developed PTSD were exposed to significantly more traumatic events than control subjects in MRS and Army STARRS. One way to address these differences would be to include trauma exposure as a covariate. However, given the high correlation between trauma exposure and PTSD, this would likely diminish the association of methylation changes and PTSD. Therefore, we acknowledge that our analyses may include associations with trauma exposure in addition to PTSD. Next, the limited replication of findings from the US cohorts in the Dutch PRISMO study, and vice versa [7], may point to type I errors, or be partially due to small sample sizes and/or heterogeneity in study designs, study environments, and potential confounders such as immune status or medication use. For example, findings reported from PRISMO [7] point towards the involvement of DMPs and DMRs located within ZFP57, RNF39, and HIST1H2APS2, which were not implicated in our analyses. However, these genes are located within the HLA region, a region which is indeed implicated in our analyses. Furthermore, whereas the PRISMO study is entirely based on subjects of European ancestry, the US MRS and Army STARRS studies include more ancestral diversity. Genes in the HLA region tend to be highly polymorphic, particularly with respect to ancestral background [48]. Thus, it is possible that this heterogeneity may contribute to the lack of replication. Further analyses should be ancestry-specific, once sample sizes are adequate. Furthermore, at this stage, it is unclear whether the identified differential methylation patterns in PTSD cases have any functional consequences. Although they may influence gene expression, the current dataset has limited power to establish causality. In order to make claims regarding causation, performing functional studies in vitro and animal studies will be needed in order to unravel precise biological mechanisms. Finally, to maximize power for discovery, the present cohorts were chosen to be highly similar in regard to demographics, type of trauma, and time since trauma exposure. Thus, the degree to which these findings on active duty, predominantly European-ancestry military men, may generalize to females, civilians, or other ancestries is unclear.

In summary, this is the largest study on methylation changes associated with the development of combat-related PTSD to date. Our observations point towards the implication of biologically interesting genes such as HLA region which is involved in immune-related processes, HEXDC which also has been suggested to play a role in immunity, and MAD1L1, a PTSD-related gene recently identified in the large MVP. These findings strengthen the notion that specific DNA methylation profiles are involved in the development of combat-related PTSD. Larger longitudinal studies and integrative efforts are now needed to build upon these preliminary findings in order to understand their functional consequences and integrate them more broadly into our current understanding of the (epi)genomic basis of PTSD.