Mounting evidence suggests that stress-related psychopathology, such as posttraumatic stress disorder (PTSD), is associated with accelerated cellular aging in the epigenome. Specifically, DNA methylation (DNAm) data can be used to estimate cellular age from established epigenetic age algorithms developed by Horvath (Reference Horvath2013) and Hannum et al. (Reference Hannum, Guinney, Zhao, Zhang, Hughes, Sadda, Klotzle, Bibikova, Fan, Gao, Deconde, Chen, Rajapakse, Friend, Ideker and Zhang2013). When estimated from whole blood, these estimates of epigenetic age (i.e. ‘DNAm age’) show strong and equivalent patterns of association with chronological age, despite being based on different methods and tissues and despite the two estimates sharing very few DNAm loci. DNAm age estimates that exceed chronological age (i.e. advanced DNAm age) have been associated with a variety of important clinical phenotypes including shortened time until death (Marioni et al., Reference Marioni, Shah, McRae, Chen, Colicino, Harris, Gibson, Henders, Redmond, Cox, Pattie, Corley, Murphy, Martin, Montgomery, Feinberg, Fallin, Multhaup, Jaffe, Joehanes, Schwartz, Just, Lunetta, Murabito, Starr, Horvath, Baccarelli, Levy, Visscher, Wray and Deary2015; Chen et al., Reference Chen, Marioni, Colicino, Peters, Ward-Caviness, Tsai, Roetker, Just, Demerath, Guan, Bressler, Fornage, Studenski, Vandiver, Moore, Tanaka, Kiel, Liang, Vokonas, Schwartz, Lunetta, Murabito, Bandinelli, Hernandez, Melzer, Nalls, Pilling, Price, Singleton, Gieger, Holle, Kretschmer, Kronenberg, Kunze, Linseisen, Meisinger, Rathmann, Waldenberger, Visscher, Shah, Wray, McRae, Franco, Hofman, Uitterlinden, Absher, Assimes, Levine, Lu, Tsao, Hou, Manson, Carty, LaCroix, Reiner, Spector, Feinberg, Levy, Baccarelli, van Meurs, Bell, Peters, Deary, Pankow, Ferrucci and Horvath2016; Perna et al., Reference Perna, Zhang, Mons, Holleczek, Saum and Brenner2016), cancer (Zheng et al., Reference Zheng, Joyce, Colicino, Liu, Zhang, Dai, Shrubsole, Kibbe, Gao, Zhang, Jafari, Vokonas, Schwartz, Baccarelli and Hou2016), and neurocognitive decline among those with Alzheimer's disease (Levine et al., Reference Levine, Lu, Bennett and Horvath2015) and PTSD (Wolf et al., Reference Wolf, Logue, Hayes, Sadeh, Schichman, Stone, Salat, Milberg, McGlinchey and Miller2016). Thus advanced DNAm age may be an important part of understanding the link between PTSD and premature health and neurocognitive decline (see Wolf and Morrison, Reference Wolf and Morrison2017, for a review).
Using these age calculators, a number of studies have suggested that PTSD symptoms are cross-sectionally associated with advanced epigenetic age relative to chronological age, albeit the effect tends to be small in magnitude and not uniform across studies. Specifically, Wolf et al. (Reference Wolf, Logue, Hayes, Sadeh, Schichman, Stone, Salat, Milberg, McGlinchey and Miller2016) showed that an index of PTSD symptom severity across the lifespan was associated with advanced epigenetic age (relative to chronological age) using the Hannum, but not the Horvath, age algorithm in a sample of 281 young veterans returning from the wars in Iraq and Afghanistan (an overlapping cohort is featured in this study). Following this, Wolf et al. (Reference Wolf, Logue, Stoop, Schichman, Stone, Sadeh, Hayes and Miller2017) demonstrated that PTSD hyperarousal symptoms were more specifically associated with advanced Hannum epigenetic age in a middle-aged sample of 339 mixed war era veterans. In contrast, Zannas et al. (Reference Zannas, Arloth, Carrillo-Roa, Iurato, Roh, Ressler, Nemeroff, Smith, Bradley, Heim, Menke, Lange, Bruckl, Ising, Wray, Erhardt, Binder and Mehta2015) found no association between PTSD and advanced epigenetic age using the Horvath index (the Hannum index was not evaluated in that study) in a sample of 392 predominately female civilians with substantial exposure to community violence. However, Zannas et al., did report an association between the cumulative burden of daily life stressors and advanced epigenetic age per the Horvath algorithm. Similarly, in a sample of 101 African American children living in an urban environment, Jovanovic et al. (Reference Jovanovic, Vance, Cross, Knight, Kilaru, Michopoulos, Klengel and Smith2017) showed that exposure to violence was associated with advanced Horvath DNAm age (the Hannum index was not evaluated). In the only longitudinal PTSD and epigenetic age study to date, Boks et al. (Reference Boks, van Mierlo, Rutten, Radstake, De Witte, Geuze, Horvath, Schalkwyk, Vinkers, Broen and Vermetten2015) reported that in a sample of 96 Dutch male veterans, increased PTSD symptoms from immediately pre-to-post warzone exposure (i.e. difference scores) were associated with decreased epigenetic age over the same interval while combat exposure was associated with increased epigenetic age per the Horvath algorithm (the Hannum algorithm was not examined).
In an effort to clarify differential patterns of association, Wolf et al. (Reference Wolf, Maniates, Nugent, Maihofer, Armstrong, Ratanatharathorn, Ashley-Koch, Garrett, Kimbrel, Lori, Va Mid-Atlantic Mirecc, Aiello, Baker, Beckham, Boks, Galea, Geuze, Hauser, Kessler, Koenen, Miller, Ressler, Risbrough, Rutten, Stein, Ursano, Vermetten, Vinkers, Uddin, Smith, Nievergelt and Logue2018) conducted a cross-sectional meta-analysis of the associations between trauma exposure, PTSD diagnosis and symptom severity, and epigenetic age (per both the Horvath and Hannum algorithms) relative to chronological age using data from nine cohorts contributing to the Psychiatric Genomics Consortium PTSD Epigenetics Workgroup (Ratanatharathorn et al., Reference Ratanatharathorn, Boks, Maihofer, Aiello, Amstadter, Ashley-Koch, Baker, Beckham, Bromet, Dennis, Garrett, Geuze, Guffanti, Hauser, Kilaru, Kimbrel, Koenen, Kuan, Logue, Luft, Miller, Mitchell, Nugent, Ressler, Rutten, Stein, Vermetten, Vinkers, Youssef, Workgroup, Workgroup, Uddin, Nievergelt and Smith2017) which included over 2000 participants. Results suggested that lifetime PTSD symptom severity was associated with advanced epigenetic age per the Hannum, but not the Horvath, index and that childhood trauma exposure, when assessed with a uniform measure, was also associated with advanced epigenetic age per the Hannum metric. The meta-analytic effect for PTSD was estimated at unstandardized β = 0.01, implying that the difference between individuals with no PTSD symptoms v. substantial symptoms (i.e. severity scores of 0 v. 100 on the Clinician-Administered PTSD Scale) was about 1.1 years, though the cumulative burden of such symptoms over time on cellular age was unknown.
There are two primary limitations of the foregoing studies. The first is that, with few exceptions (reviewed below), the psychopathology/DNAm age literature has largely focused on trauma and PTSD, as opposed to other forms of psychopathology, as predictors of cellular aging. The second limitation is that all but one study employed a cross-sectional design. There is no reason to suspect that accelerated DNAm age is unique to PTSD. Rather, we hypothesize that accelerated epigenetic age may be a common biological correlate of a variety of stress-related psychiatric disorders. Support for this comes from multiple lines of converging research. This includes evidence that shortened telomere length is a common correlate of many forms of stress-related psychopathology (Darrow et al., Reference Darrow, Verhoeven, Revesz, Lindqvist, Penninx, Delucchi, Wolkowitz and Mathews2016) and that sleep disturbance, a transdiagnostic symptom across psychological disorders, was associated with advanced age per a Hannum-based metric of epigenetic age that also incorporated age-related components of the immune system (though it was not associated with advanced Horvath age; Carroll et al., Reference Carroll, Irwin, Levine, Seeman, Absher, Assimes and Horvath2017). Moreover, PTSD is highly comorbid with a range of other psychiatric conditions including those spanning internalizing (e.g. depression and anxiety) and externalizing (e.g. substance use disorders) psychopathology dimensions (Kessler et al., Reference Kessler, Sonnega, Bromet, Hughes and Nelson1995; Miller et al., Reference Miller, Lin, Wolf and Miller2012) and this raises the possibility that PTSD comorbidity may also contribute to accelerated aging. In addition, accelerated DNAm age has been associated with a variety of other conditions including Alzheimer's disease (Levine et al., Reference Levine, Lu, Bennett and Horvath2015), cancer (Durso et al., Reference Durso, Bacalini, Sala, Pirazzini, Marasco, Bonafé, do Valle, Gentilini, Castellani and Faria2017), and obesity (Quach et al., Reference Quach, Levine, Tanaka, Lu, Chen, Ferrucci, Ritz, Bandinelli, Neuhouser and Beasley2017), which would argue against the specificity of accelerated DNAm age to trauma and PTSD. That said, it is important to note that the few studies which did examine epigenetic age and other forms of psychopathology found null or negative associations. One post-mortem study found reduced Horvath epigenetic age (relative to chronological age) in neurons in the orbitofrontal cortex among 37 heroin users who overdosed in comparison with both controls and non-overdose suicide cases (Kozlenkov et al., Reference Kozlenkov, Jaffe, Timashpolsky, Apontes, Rudchenko, Barbu, Byne, Hurd, Horvath and Dracheva2017). Two studies reported null effects between schizophrenia and Horvath DNAm age residuals assessed in temporal (McKinney et al., Reference McKinney, Lin, Ding, Lewis and Sweet2017) and frontal (Voisey et al., Reference Voisey, Lawford, Morris, Wockner, Noble, Young and Mehta2017) post-mortem brain tissue.
The general focus on cross-sectional studies has precluded evaluation of the predictive strength of psychopathology on changes in DNAm age acceleration. This concern is accentuated in light of preliminary evidence from epidemiological cohorts suggesting that accelerated Horvath epigenetic age is largely stable over time (Kananen et al., Reference Kananen, Marttila, Nevalainen, Kummola, Junttila, Mononen, Kähönen, Raitakari, Hervonen and Jylhä2016), raising the possibility that observed relationships with PTSD may reflect co-occurring phenomena, as opposed to a causal association. It is important to test this by examining potential pathogenic environmental factors that might impact the stability of accelerated aging. The primary aim of this study was to examine the predictive strength of posttraumatic psychopathology [PTSD, major depressive disorder (MDD), generalized anxiety disorder (GAD), and alcohol-use disorders] in association with accelerated epigenetic aging over time. A second aim was to test the utility of a novel approach to quantifying accelerated DNAm age over time using a metric that captured changes in the pace of epigenetic aging relative to intervening time. This was accomplished in a longitudinal sample of 179 veterans post-deployment to the wars in Iraq and/or Afghanistan, who had DNAm data available at two-time points, spanning, on average, 1.89 years.
Methods
Participants
Participants were US military veterans of the post 9/11 conflicts who were enrolled in the ongoing research protocol, the Translational Research Center for TBI and Stress Disorders (TRACTS), a US Department of Veterans Affairs (VA) Rehabilitation Research and Development Traumatic Brain Injury Center of Excellence at VA Boston Healthcare System. TRACTS (McGlinchey et al., Reference McGlinchey, Milberg, Fonda and Fortier2017) is a longitudinal investigation of traumatic brain injury (TBI), traumatic stress, health, and neural and cognitive factors among returning veterans. Study exclusion criteria were neurological illness, history of seizures unrelated to head injury, acute psychotic or bipolar disorder, severe or unstable psychological diagnosis that prevented participation, acute homicidal and/or suicidal ideation with intent to act, and cognitive disorder due to general medical condition other than TBI. Analyses for this study were based on a subset of TRACTS participants with Time 1 (T1) and Time 2 (T2) DNAm data at the time of our second data freeze in which DNA was sent for processing. We had previously obtained DNAm data using the Illumina HumanMethylation 450k beadchip on 281 participants at their T1 visit and reported on cross-sectional associations between PTSD and accelerated DNAm age in that initial cohort in Wolf et al. (Reference Wolf, Logue, Hayes, Sadeh, Schichman, Stone, Salat, Milberg, McGlinchey and Miller2016). Of that group, 153 returned for a second visit and an additional 26 new T1 veterans enrolled in the protocol and returned for a T2 visit prior to the data freeze, yielding a total T1/T2 sample size of 179. DNAm values were simultaneously obtained on the Illumina EPIC beadchip for both T1 and T2, including for those with existing T1 DNAm data from the 450k beadchip.
Of the longitudinal sample of n = 179, 88.3% were male veterans and the mean age at T1 was 32.84 years (range: 19–65; s.d. = 9.28; Table 1). The majority of the sample (75.7%) self-reported as white, 12.4% as Hispanic or Latino/a, 9.6% as black, 1.7% as Asian, and 0.6% as American Indian (Table 1).
Demographic and psychiatric characteristics of the longitudinal sample (n = 179)
T1, time 1; T2, time 2; PTSD, posttraumatic stress disorder; MDD, major depressive disorder; GAD, generalized anxiety disorder.
Procedure
After providing written informed consent, participants completed a comprehensive interview- and self-report-based psychological assessment, with diagnostic interviews administered by doctoral-level psychology professionals. Each interview was reviewed by a team of psychologists to inform consensus ratings of presence/absence of psychological diagnoses. Blood was drawn at each time point for DNA extraction and metabolic assays. The two assessment points occurred an average of 1.89 years apart (Table 1). The study was approved by the VA Boston Healthcare System IRB.
Measures
Traumatic Life Events Questionnaire (TLEQ; Kubany et al., Reference Kubany, Haynes, Leisen, Owens, Kaplan, Watson and Burns2000), a self-report assessment of exposure to 21 trauma types, was used to quantify the number of different types of traumatic experiences reported at T1. PTSD was assessed with the Clinician-Administered PTSD Scale for DSM-IV (CAPS; Blake et al., Reference Blake, Weathers, Nagy, Kaloupek, Gusman, Charney and Keane1995). This well-validated diagnostic interview was administered to capture current (past month) symptoms, the worst period of symptoms post-military discharge, and any pre-military symptoms. Follow-up analyses focused on PTSD symptom clusters using factor scores representing latent lifespan severity of PTSD symptoms at T1. Other psychiatric disorders (mood, anxiety, and substance use) were measured with the well-established Structured Clinical Interview for DSM-IV Disorders (SCID; First et al., Reference First, Gibbon, Spitzer and Benjamin1997; Spitzer et al., Reference Spitzer, Gibbon and Williams1998).
DNA data
Genotypes
DNA was extracted from peripheral blood samples and genotyping obtained via the Illumina HumanOmni2.5-8 beadchip. Genotypes are relevant as they pertain to the development of principal components (PCs) to model ancestry (Supplementary Materials).
Methylation
DNA was hybridized to the Infinium MethylationEPIC beadchip according to the manufacturer's protocol. To reduce the possibility of systematic bias, T1 and T2 samples were run together on the same chip using a balanced chip design. We followed the processing pipeline established by the PGC PTSD Epigenetics workgroup (Ratanatharathorn et al., Reference Ratanatharathorn, Boks, Maihofer, Aiello, Amstadter, Ashley-Koch, Baker, Beckham, Bromet, Dennis, Garrett, Geuze, Guffanti, Hauser, Kilaru, Kimbrel, Koenen, Kuan, Logue, Luft, Miller, Mitchell, Nugent, Ressler, Rutten, Stein, Vermetten, Vinkers, Youssef, Workgroup, Workgroup, Uddin, Nievergelt and Smith2017) for the Illumina HumanMethylation450 BeadChip, but applied here to the EPIC chip (Supplementary Materials).
Calculation of epigenetic age
Horvath DNAm age estimates were computed with an R script based on the 335 probes that were assessed on the EPIC chip and passed quality control (QC). For the Hannum algorithm, data were normalized using the beta mixture quantile dilation (BMIQ), as in prior publications (Wolf et al., Reference Wolf, Logue, Hayes, Sadeh, Schichman, Stone, Salat, Milberg, McGlinchey and Miller2016; Reference Wolf, Logue, Stoop, Schichman, Stone, Sadeh, Hayes and Miller2017; Reference Wolf, Maniates, Nugent, Maihofer, Armstrong, Ratanatharathorn, Ashley-Koch, Garrett, Kimbrel, Lori, Va Mid-Atlantic Mirecc, Aiello, Baker, Beckham, Boks, Galea, Geuze, Hauser, Kessler, Koenen, Miller, Ressler, Risbrough, Rutten, Stein, Ursano, Vermetten, Vinkers, Uddin, Smith, Nievergelt and Logue2018; Supplementary Materials). We have previously demonstrated (Logue et al., Reference Logue, Smith, Wolf, Maniates, Stone, Schichman, McGlinchey, Milberg and Miller2017) that the correlations between EPIC chip-derived Horvath and Hannum age estimates with chronological age (rs = 0.90 and 0.87, respectively) were very similar to those reported for Infinium-derived Horvath and Hannum estimates (rs = 0.88; Wolf et al., Reference Wolf, Logue, Hayes, Sadeh, Schichman, Stone, Salat, Milberg, McGlinchey and Miller2016).
White blood cell (WBC) count estimation
Proportional WBC counts at T1 and T2 were estimated from the methylation data (Supplementary Materials).
Data analyses
Operational definition of accelerated aging in DNAm over time
We examined a new metric of accelerated aging over time focused on the pace of change in the raw DNAm age estimates (i.e. the rate at which the epigenetic clock advances). This was defined as the difference in raw DNAm age estimates (T2–T1) over time as a function of the time between the two assessments:
$$ {\hbox {Rate of raw DNAm age change per calendar year}} = {\displaystyle {\hbox {T2 DNAm age - T1 DNAm age}} \over \#\;{\hbox {of years between}} \, {\rm T}1\; \&\;{\rm T}2 \, {\rm assessments}}$$A score of 1.0 indicates that for each year between the two assessments, DNAm age also increased by 1 year. Values greater than 1.0 suggest an accelerated pace of DNAm age change relative to the number of intervening years, while values less than 1.0 are indicative of slower pace of the epigenetic clock relative to time between assessments.
Analytic plan
We first calculated the correlation between the DNAm age estimates and chronological age at each time point. We then extracted DNAm age residuals for T1 and T2 data (defined by the residuals from an equation in which T1/T2 chronological age was regressed out of the T1/T2 Hannum and Horvath DNAm age estimates, respectively) and compared DNAm and DNAm residual correlations across the Horvath and Hannum algorithms. Positive residuals reflect over-predicted DNAm age (i.e. advanced epigenetic age) at a given time point. We examined the stability of DNAm age and DNAm age residuals over time both within and across algorithms via correlation and by examining the descriptive statistics pertaining to the average rate change in raw DNAm age estimates per calendar year per the operational definition defined above. We also examined the stability of the estimated WBCs using correlation.
We conducted two multiple regressions apiece for the two (Horvath/Hannum) DNAm age rate variables: the first simultaneously evaluated the contribution of T1 current diagnoses of PTSD, MDD, GAD, and alcohol-use disordersFootnote †Footnote 1 to the pace of epigenetic aging and the second evaluated the contribution of T1 lifespan PTSD symptom cluster severity to this phenomenon. T1 lifespan PTSD symptom cluster severity was represented by factor scores from a confirmatory factor analysis (CFA) of scores on pre-military, post-military, and current (T1) PTSD symptom clusters. Each regression controlled for sex, the top two ancestry PCs, and T1 total trauma exposure. We did not include WBCs as covariates because the dependent variable captured change across T1 and T2 DNAm age and it was not evident which set of WBCs (if any) should be included. Rather, we examined this question in preliminary regressions (detailed in Supplementary Materials) which separately tested T1 and T2 WBCs as predictors of the pace of DNAm change. In follow-up analyses detailed in Supplementary Materials, we evaluated interactions between psychopathology and demographic variables and examined potential confounds (cigarette use, body mass index, traumatic brain injury, psychotropic medication use) of our primary associations of interest. In additional follow-up analyses detailed in Supplementary Materials, we also evaluated if a change in psychopathology variables over the two assessment points covaried with the rate of DNAm age change. As each analysis was executed twice (once for Horvath and once for Hannum-based indices), we performed a statistical procedure to correct the p values using a Monte Carlo null simulation with 1000 replicates that also took into account the correlational structure of the dependent variables.
Results
Estimated DNAm age
DNAm age estimates were strongly associated with chronological age. Specifically, correlations between Horvath DNAm age and chronological age were r = 0.90 (p < 0.001) and r = 0.91 (p < 0.001) at T1 and T2, respectively. Hannum DNAm age correlations with chronological age were r = 0.88 (p < 0.001) and r = 0.85 (p < 0.001) at T1 and T2, respectively. At T1, the Horvath and Hannum estimates correlated with each other at r = 0.88 (p < 0.001) and at T2, at r = 0.87 (p < 0.001). At both T1 and T2, the correlation between the Horvath and Hannum age residuals was r = 0.44 (p < 0.001).
Stability of DNAm age estimates and WBCs over time
The raw DNAm age estimates were highly correlated over time: r Horvath = 0.95, p < 0.001; r Hannum = 0.93, p < 0.001. The DNAm age residuals were also highly correlated over time (Table 2). Horvath DNAm age estimates increased at a rate of, on average, 0.76 years per calendar year (s.d.: 1.75, range: −3.62 to 7.11 years) while Hannum DNAm age estimates increased by an average rate of 1.10 years per calendar year (s.d.: 1.68, range: −5.35 to 5.86). WBC proportional estimates were highly correlated (across the same cell type) over time (range: rs = 0.67 to 0.75; Table 3). The rate of change in the raw DNAm age estimates was weakly correlated across the Horvath and Hannum algorithms (r = 0.26, p < 0.001). Neither T1 nor T2 WBCs was associated with epigenetic age acceleration and thus WBCs were not included in subsequent models (Supplementary Materials).
Descriptive statistics (and correlations) pertaining to DNAm age residuals and the rate of DNAm age change over time
DNAm, DNA methylation; T1, time 1; T2, time 2; diff, difference.
Note. All correlations were statistically significant at the p < 0.01 or better level.
Stability (Pearson correlations) of white blood cell proportional estimates over time
*p < 0.05. **p < 0.01. ***p < 0.001.
PTSD, MDD, GAD, and alcohol-use disorders as predictors of the rate of DNAm age acceleration
We examined if T1 PTSD, MDD, GAD, or alcohol-use disorders predicted the rate of epigenetic aging.Footnote 2 Alcohol use diagnoses at T1 were associated with the rate of epigenetic aging estimated via the Horvath algorithm (β = 0.26, p = 0.001, corrected p = 0.002; Table 4). The mean rate of epigenetic aging as a function of T1 alcohol-use diagnosis was 1.58 years, indicating that individuals with a T1 alcohol use disorder evidenced a 1.58 year DNAm age increase per calendar year (compared with 0.60 among those without a T1 alcohol use disorder; p = 0.006; Fig. 1). None of the mental health diagnoses predicted the rate of epigenetic aging estimated with the Hannum method (Table 4). The association between alcohol-use disorders and the rate of change in DNAm age was not moderated by demographic characteristics (Supplementary Materials).
The figure shows the mean differential rate of change in the Horvath DNAm age estimate per year as a function of alcohol-use disorder diagnoses at time 1. Error bars represent standard errors. DNAm, DNA methylation; T1, time 1.
Regression results predicting the rate of epigenetic aging over time
DNAm, DNA methylation; PTSD, posttraumatic stress disorder; PC, principal component; T1, time 1; sx, symptom; dx, diagnosis; reexp, re-experiencing; avd, avoidance; numb, numbing; MDD, major depressive disorder; GAD, generalized anxiety disorder; Alc, alcohol.
Note. All PTSD severity variables were factor scores derived from a confirmatory factor analysis of pre-military, post-military, and current PTSD symptoms as assessed at T1.
PTSD symptom cluster severity as a predictor of the rate of DNAm age acceleration
In the PTSD symptom cluster factor scoreFootnote 3 models, T1 avoidance and numbing symptoms were associated with an increased pace of epigenetic aging per the Horvath index (β = 0.43, p = 0.016, corrected p = 0.029; Table 4, Fig. 2).Footnote 4, Footnote 5 There were no significant predictors of the rate of epigenetic aging as estimated by the Hannum method (Table 4). There was no evidence that demographic variables moderated the associations between PTSD symptom clusters and epigenetic aging and no evidence that change in psychiatric diagnoses or symptoms across the two assessment points was associated with the rate of DNAm age acceleration (Supplementary Materials).
The figure shows the relationship between time 1 avoidance and numbing PTSD latent variable factor scores in association with the rate of change in estimated DNAm age each year per the Horvath algorithm (controlling for all other covariates in the model as per Table 4, Model 2). DNAm, DNA methylation.
Discussion
Longitudinal study is critical for distinguishing the neurobiological correlates of trauma- and stress-related psychopathology from the neurobiological consequences of such symptoms. In this first longitudinal study of the associations between a range of posttraumatic psychiatric disorders and change in epigenetic age, we found that PTSD avoidance and numbing symptoms and alcohol-use disorders predicted increases in the pace of the epigenetic clock. This study represents an important first step in extending results of prior cross-sectional research and advancing our understanding of the neurobiological consequences of psychopathology. A quickened pace of the epigenetic clock could contribute to early onset of age-related health decline, such as cardiometabolic disorders (Perna et al., Reference Perna, Zhang, Mons, Holleczek, Saum and Brenner2016), cognitive decline (Levine et al., Reference Levine, Lu, Bennett and Horvath2015; Marioni et al., Reference Marioni, Shah, McRae, Chen, Colicino, Harris, Gibson, Henders, Redmond, Cox, Pattie, Corley, Murphy, Martin, Montgomery, Feinberg, Fallin, Multhaup, Jaffe, Joehanes, Schwartz, Just, Lunetta, Murabito, Starr, Horvath, Baccarelli, Levy, Visscher, Wray and Deary2015), and mortality (Marioni et al., Reference Marioni, Shah, McRae, Chen, Colicino, Harris, Gibson, Henders, Redmond, Cox, Pattie, Corley, Murphy, Martin, Montgomery, Feinberg, Fallin, Multhaup, Jaffe, Joehanes, Schwartz, Just, Lunetta, Murabito, Starr, Horvath, Baccarelli, Levy, Visscher, Wray and Deary2015; Chen et al., Reference Chen, Marioni, Colicino, Peters, Ward-Caviness, Tsai, Roetker, Just, Demerath, Guan, Bressler, Fornage, Studenski, Vandiver, Moore, Tanaka, Kiel, Liang, Vokonas, Schwartz, Lunetta, Murabito, Bandinelli, Hernandez, Melzer, Nalls, Pilling, Price, Singleton, Gieger, Holle, Kretschmer, Kronenberg, Kunze, Linseisen, Meisinger, Rathmann, Waldenberger, Visscher, Shah, Wray, McRae, Franco, Hofman, Uitterlinden, Absher, Assimes, Levine, Lu, Tsao, Hou, Manson, Carty, LaCroix, Reiner, Spector, Feinberg, Levy, Baccarelli, van Meurs, Bell, Peters, Deary, Pankow, Ferrucci and Horvath2016; Wolf et al., Reference Wolf, Logue, Stoop, Schichman, Stone, Sadeh, Hayes and Miller2017). Results also highlight the utility of a novel metric to represent the changing pace of the epigenetic clock that was directly interpretable. We suggest that prior cross-sectional studies, including our own, should be conceptualized as examining associations between psychiatric phenotypes and advanced but not necessarily accelerated, DNAm age. In contrast, by modeling the changing speed of the epigenetic clock compared with the passage of time, we were able to more definitively quantify acceleration of cellular age.
Though alcohol-use disorders per se have not previously been investigated in association with a changing pace of the epigenetic clock, our results are broadly consistent with one prior cross-sectional study which suggested that problematic alcohol use was associated with advanced DNAm age. Specifically, in a population study that included DNAm data from 105 individuals, heavy alcohol use was related to advanced epigenetic age using a DNAm age index developed by Weidner and Wagner (Reference Weidner and Wagner2014), based on just three loci obtained via pyrosequencing. We did not evaluate that index because it did not evidence as strong an association with chronological age in the validation samples (Weidner and Wagner, Reference Weidner and Wagner2014) compared with the Horvath (Reference Horvath2013) and Hannum et al. (Reference Hannum, Guinney, Zhao, Zhang, Hughes, Sadda, Klotzle, Bibikova, Fan, Gao, Deconde, Chen, Rajapakse, Friend, Ideker and Zhang2013) metrics and because we did not have DNAm data on loci that were not on the Illumina chip. In contrast to Weidner and Wagner (Reference Weidner and Wagner2014), a recent cross-sectional study suggested that moderate alcohol use was not related to DNAm age residuals per the Horvath metric, but it was negatively associated with a related index hypothesized to reflect immune system age more generally (Quach et al., Reference Quach, Levine, Tanaka, Lu, Chen, Ferrucci, Ritz, Bandinelli, Neuhouser and Beasley2017). That index was referred to as ‘extrinsic epigenetic age acceleration’ and it was operationalized as the residuals obtained from an equation in which chronological age was regressed from Hannum DNAm age estimates combined with age-weighted WBC estimates (Quach et al., Reference Quach, Levine, Tanaka, Lu, Chen, Ferrucci, Ritz, Bandinelli, Neuhouser and Beasley2017). We did not examine the extrinsic epigenetic age index in this study because: (a) we wanted to extend our prior work on advanced DNAm age specifically; (b) doing so would complicate interpretation of the rate variable as WBCs are not intrinsically linked to the passage of time in the way DNAm age is; and (c) it is not evident how DNAm v. WBCs contribute to effects observed for the extrinsic epigenetic age index. Despite differential quantification of DNAm age indices across our study and the aforementioned ones, results collectively raise the possibility that alcohol-use disorders may accelerate epigenetic aging while moderate alcohol use may be protective; additional research is needed to test this directly.
Prior epigenetic studies suggest that alcohol-related phenotypes are associated with differential DNAm profiles (Zhang and Gelernter, Reference Zhang and Gelernter2017). A number of epigenome-wide association studies (EWAS) have together implicated alcohol-related phenotypes in association with methylation of genes involved in immune response, signal transduction, metabolism, and apoptosis-related gene networks (Zhang et al., Reference Zhang, Miao, Wang, Zhao, Li, Haile, Hao and Zhang2013; Zhao et al., Reference Zhao, Zhang, Li, Liao, Tang, Miao and Hao2013; Harlaar et al., Reference Harlaar, Bryan, Thayer, Karoly, Oien and Hutchison2014; Philibert et al., Reference Philibert, Penaluna, White, Shires, Gunter, Liesveld, Erwin, Hollenbeck and Osborn2014; Liu et al., Reference Liu, Marioni, Hedman, Pfeiffer, Tsai, Reynolds, Just, Duan, Boer, Tanaka, Elks, Aslibekyan, Brody, Kuhnel, Herder, Almli, Zhi, Wang, Huan, Yao, Mendelson, Joehanes, Liang, Love, Guan, Shah, McRae, Kretschmer, Prokisch, Strauch, Peters, Visscher, Wray, Guo, Wiggins, Smith, Binder, Ressler, Irvin, Absher, Hernandez, Ferrucci, Bandinelli, Lohman, Ding, Trevisi, Gustafsson, Sandling, Stolk, Uitterlinden, Yet, Castillo-Fernandez, Spector, Schwartz, Vokonas, Lind, Li, Fornage, Arnett, Wareham, Sotoodehnia, Ong, van Meurs, Conneely, Baccarelli, Deary, Bell, North, Liu, Waldenberger, London, Ingelsson and Levy2016). There are also numerous candidate gene DNAm studies of alcohol-use, two of which are notable because the genes are relevant to the Horvath algorithm. Specifically, differential methylation of the gene promoter region of ANP (which encodes for the atrial natriuretic peptide, a vasodilator secreted by heart muscle cells) was reported in association with alcohol withdrawal (Glahn et al., Reference Glahn, Riera Knorrenschild, Rhein, Haschemi Nassab, Groschl, Heberlein, Muschler, Frieling, Bleich and Hillemacher2014). A probe in the gene encoding the corresponding receptor (ANPRC: atrial natriuretic peptide receptor) is part of the Horvath DNAm age algorithm, which could potentially contribute to the sensitivity of DNAm age to alcohol-related phenotypes. ANP has strong anxiolytic effects, and attenuates the stress response by inhibiting the hypothalamic-pituitary-adrenal (HPA) axis, via moderation of adrenocorticotropin (ACTH) and cortisol (Antoni et al., Reference Antoni, Hunter, Lowry, Noble and Seckl1992; Wiedemann et al., Reference Wiedemann, Jahn, Yassouridis and Kellner2001; Strohle et al., Reference Strohle, Feller, Strasburger, Heinz and Dimeo2006). Likewise, methylation of the 5′ upstream promoter of the hormone precursor polypeptide pro-opiomelanocortin (POMC) gene evidenced marginal associations with alcohol-use disorders and was specifically associated with alcohol cravings among those with alcohol-related disorders (Muschler et al., Reference Muschler, Hillemacher, Kraus, Kornhuber, Bleich and Frieling2010). POMC is also involved in regulation of the HPA axis and a probe in POMC is included in the Horvath DNAm age algorithm. These findings are broadly consistent with evidence that the probes that comprise the Horvath DNAm age index are responsive to glucocorticoid modulation (Zannas et al., Reference Zannas, Arloth, Carrillo-Roa, Iurato, Roh, Ressler, Nemeroff, Smith, Bradley, Heim, Menke, Lange, Bruckl, Ising, Wray, Erhardt, Binder and Mehta2015) and that advanced Horvath DNAm age is associated with greater diurnal cortisol (Davis et al., Reference Davis, Humphreys, McEwen, Sacchet, Camacho, MacIsaac, Lin, Kobor and Gotlib2017).
In addition to shared sensitivity to glucocorticoids, there is evidence for a common role of inflammation and immune system dysregulation across PTSD, alcohol-use disorders, and cellular aging. Specifically, heavy alcohol use is associated with increases in memory T-cells and reductions in naïve T-cells (Cook et al., Reference Cook, Ballas, Waldschmidt, Vandersteen, LaBrecque and Cook1994; Reference Cook, Waldschmidt, Ballas, Cook, Booth, Stewart and Garvey1995), which could contribute to the association between alcohol use and infections (Szabo and Mandrekar, Reference Szabo and Mandrekar2009). Consistent with this, a gene expression study conducted in a small sample found that alcohol dependence was associated with enrichment in T-cell receptor genetic pathways (Beech et al., Reference Beech, Qu, Leffert, Lin, Hong, Hansen, Umlauf, Mane, Zhao and Sinha2012). PTSD has been similarly associated with changes in adaptive immunity, including increases in late-stage differentiated T-cells and decreases in naïve T-cell populations (Sommershof et al., Reference Sommershof, Aichinger, Engler, Adenauer, Catani, Boneberg, Elbert, Groettrup and Kolassa2009; Aiello et al., Reference Aiello, Dowd, Jayabalasingham, Feinstein, Uddin, Simanek, Cheng, Galea, Wildman and Koenen2016), as well as gene expression profiles enriched for innate immune system signaling (Breen et al., Reference Breen, Maihofer, Glatt, Tylee, Chandler, Tsuang, Risbrough, Baker, O'Connor, Nievergelt and Woelk2015). Shifts in T-cell populations from naïve to differentiated types is often conceptualized as part of an immunosenescent profile that is itself a biological indicator of aging (Macaulay et al., Reference Macaulay, Akbar and Henson2013; Fülöp et al., Reference Fülöp, Dupuis, Witkowski and Larbi2016), and one that can contribute to increased inflammation via pro-inflammatory cytokines (Macaulay et al., Reference Macaulay, Akbar and Henson2013). Both alcohol-use disorders (Achur et al., Reference Achur, Freeman and Vrana2010; Leclercq et al., Reference Leclercq, Cani, Neyrinck, Starkel, Jamar, Mikolajczak, Delzenne and de Timary2012) and PTSD (Passos et al., Reference Passos, Vasconcelos-Moreno, Costa, Kunz, Brietzke, Quevedo, Salum, Magalhães, Kapczinski and Kauer-Sant'Anna2015; Miller et al., Reference Miller, Wolf, Reardon, Greene, Ofrat and McInerney2017) are also associated with increased inflammatory parameters including C-reactive protein, IL-6, and TNFα; inflammation may be a common pathway across the two disorders. Finally, accelerated DNAm age has also been associated with both increased inflammation (Quach et al., Reference Quach, Levine, Tanaka, Lu, Chen, Ferrucci, Ritz, Bandinelli, Neuhouser and Beasley2017) and reductions in CD4-T cell proportions (Marioni et al., Reference Marioni, Shah, McRae, Chen, Colicino, Harris, Gibson, Henders, Redmond, Cox, Pattie, Corley, Murphy, Martin, Montgomery, Feinberg, Fallin, Multhaup, Jaffe, Joehanes, Schwartz, Just, Lunetta, Murabito, Starr, Horvath, Baccarelli, Levy, Visscher, Wray and Deary2015; Chen et al., Reference Chen, Marioni, Colicino, Peters, Ward-Caviness, Tsai, Roetker, Just, Demerath, Guan, Bressler, Fornage, Studenski, Vandiver, Moore, Tanaka, Kiel, Liang, Vokonas, Schwartz, Lunetta, Murabito, Bandinelli, Hernandez, Melzer, Nalls, Pilling, Price, Singleton, Gieger, Holle, Kretschmer, Kronenberg, Kunze, Linseisen, Meisinger, Rathmann, Waldenberger, Visscher, Shah, Wray, McRae, Franco, Hofman, Uitterlinden, Absher, Assimes, Levine, Lu, Tsao, Hou, Manson, Carty, LaCroix, Reiner, Spector, Feinberg, Levy, Baccarelli, van Meurs, Bell, Peters, Deary, Pankow, Ferrucci and Horvath2016; Wolf et al., Reference Wolf, Maniates, Nugent, Maihofer, Armstrong, Ratanatharathorn, Ashley-Koch, Garrett, Kimbrel, Lori, Va Mid-Atlantic Mirecc, Aiello, Baker, Beckham, Boks, Galea, Geuze, Hauser, Kessler, Koenen, Miller, Ressler, Risbrough, Rutten, Stein, Ursano, Vermetten, Vinkers, Uddin, Smith, Nievergelt and Logue2018). Collectively, this suggests that one possible mechanism for the common association between PTSD and alcohol-use disorders and an increased pace of the epigenetic clock is alterations in key immune and inflammation pathways.
Differential findings relative to prior work
We have previously reported (Wolf et al., Reference Wolf, Logue, Hayes, Sadeh, Schichman, Stone, Salat, Milberg, McGlinchey and Miller2016; Reference Wolf, Logue, Stoop, Schichman, Stone, Sadeh, Hayes and Miller2017; Reference Wolf, Maniates, Nugent, Maihofer, Armstrong, Ratanatharathorn, Ashley-Koch, Garrett, Kimbrel, Lori, Va Mid-Atlantic Mirecc, Aiello, Baker, Beckham, Boks, Galea, Geuze, Hauser, Kessler, Koenen, Miller, Ressler, Risbrough, Rutten, Stein, Ursano, Vermetten, Vinkers, Uddin, Smith, Nievergelt and Logue2018) cross-sectional associations between PTSD and the Hannum, but not Horvath, index of DNAm age residuals. The only prior study to report an association between PTSD and the Horvath index of epigenetic age suggested that increased PTSD symptoms were associated with decreased epigenetic age, but the relative relationship with chronological age was not modeled (Boks et al., Reference Boks, van Mierlo, Rutten, Radstake, De Witte, Geuze, Horvath, Schalkwyk, Vinkers, Broen and Vermetten2015). In contrast, stress (Zannas et al., Reference Zannas, Arloth, Carrillo-Roa, Iurato, Roh, Ressler, Nemeroff, Smith, Bradley, Heim, Menke, Lange, Bruckl, Ising, Wray, Erhardt, Binder and Mehta2015), trauma (Boks et al., Reference Boks, van Mierlo, Rutten, Radstake, De Witte, Geuze, Horvath, Schalkwyk, Vinkers, Broen and Vermetten2015), and violence exposure (Jovanovic et al., Reference Jovanovic, Vance, Cross, Knight, Kilaru, Michopoulos, Klengel and Smith2017) have previously shown positive associations with Horvath advanced age. One possible explanation for these different patterns of association is that the effects of stress, trauma, and psychopathology on Horvath-defined epigenetic aging may be slow to manifest and cumulative, and therefore only evident in longitudinal studies, such as this one, and studies that examine the effects of prior trauma exposure over the course of many years. The Horvath estimate is a reliable index of chronological age across many tissue types (Horvath, Reference Horvath2013) and thus may be fairly robust to the acute or transient effects of biological processes in any one particular tissue type (e.g. metabolic aberrations found in blood). Additional research is needed to examine this possibility and account for differences in the two metrics of accelerated epigenetic age.
Another difference between the results of this study and that of Wolf et al. (Reference Wolf, Logue, Stoop, Schichman, Stone, Sadeh, Hayes and Miller2017) is that we previously demonstrated cross-sectional associations between the hyperarousal PTSD symptom cluster and advanced Hannum DNAm age whereas these longitudinal results supported an effect for the avoidance and numbing symptom cluster. This could be related to differences in the two age algorithms or to differences in the study methodology (cross-sectional v. longitudinal, use of different Illumina beadchips) and sample composition (e.g. young v. middle-aged veterans; mixed ancestry compared with a white, non-Hispanic sample) across the two studies. Replication in other studies is needed to determine if a given PTSD symptom cluster is more strongly associated with accelerated cellular aging.
Limitations and conclusions
Results of this study should be considered in light of a number of limitations. The sample comprised primarily of white, non-Hispanic men who were military veterans thus the generalizability of the results is limited and additional research in more heterogeneous samples is necessary. This is all the more important given evidence from prior work that age acceleration may differ by sex (Hannum et al., Reference Hannum, Guinney, Zhao, Zhang, Hughes, Sadda, Klotzle, Bibikova, Fan, Gao, Deconde, Chen, Rajapakse, Friend, Ideker and Zhang2013; Marioni et al., Reference Marioni, Shah, McRae, Chen, Colicino, Harris, Gibson, Henders, Redmond, Cox, Pattie, Corley, Murphy, Martin, Montgomery, Feinberg, Fallin, Multhaup, Jaffe, Joehanes, Schwartz, Just, Lunetta, Murabito, Starr, Horvath, Baccarelli, Levy, Visscher, Wray and Deary2015; Wolf et al., Reference Wolf, Logue, Stoop, Schichman, Stone, Sadeh, Hayes and Miller2017) and possibly ethnicity (Horvath et al., Reference Horvath, Gurven, Levine, Trumble, Kaplan, Allayee, Ritz, Chen, Lu, Rickabaugh, Jamieson, Sun, Li, Chen, Quintana-Murci, Fagny, Kobor, Tsao, Reiner, Edlefsen, Absher and Assimes2016). As well, as all participants were trauma-exposed, we could not fully evaluate potential effects attributable to trauma exposure (but did include a metric of trauma load in the analyses). The stability of DNAm age estimates over time yields little variability in the rate of change over time and is likely one reason that effect sizes are relatively small in magnitude, consistent with prior work (Wolf et al., Reference Wolf, Logue, Hayes, Sadeh, Schichman, Stone, Salat, Milberg, McGlinchey and Miller2016, Reference Wolf, Logue, Stoop, Schichman, Stone, Sadeh, Hayes and Miller2017, Reference Wolf, Maniates, Nugent, Maihofer, Armstrong, Ratanatharathorn, Ashley-Koch, Garrett, Kimbrel, Lori, Va Mid-Atlantic Mirecc, Aiello, Baker, Beckham, Boks, Galea, Geuze, Hauser, Kessler, Koenen, Miller, Ressler, Risbrough, Rutten, Stein, Ursano, Vermetten, Vinkers, Uddin, Smith, Nievergelt and Logue2018). Thus, caution is needed in interpreting results as the implications on gene expression are currently unknown (Breton et al., Reference Breton, Marsit, Faustman, Nadeau, Goodrich, Dolinoy, Herbstman, Holland, LaSalle, Schmidt, Yousefi, Perera, Joubert, Wiemels, Taylor, Yang, Chen, Hew, Freeland, Miller and Murphy2017). Related to this, although the longitudinal design is a strength of the study, the interval between assessments was fairly small (on average, just under 2 years), and it is possible that greater time between assessments is necessary to fully understand the cumulative effects of psychological stress on accelerated aging. As with any study, we cannot rule out possible confounds that could account for the primary findings of the study and there is much that is not yet known about the mechanisms underlying accelerated epigenetic aging that could be important for understanding and contextualizing these results. All of these limitations highlight the importance of future replication efforts in large, independent cohorts.
In conclusion, this study is the first to suggest that PTSD symptoms and alcohol-use disorders were associated with a quickening pace of the epigenetic clock, per the Horvath-derived estimate of DNAm age. Findings raise the possibility of overlap in the biological consequences of the two disorders. Results also demonstrated the applicability of the epigenetic age algorithms to DNAm data obtained from the latest Illumina methylation beadchip, the Infinium MethylationEPIC array, and highlight a novel approach to quantifying change in age acceleration. The study addresses a number of critical gaps in the literature to date, through use of a longitudinal design and the focus on the predictive strength of an array of posttraumatic psychiatric conditions that may have a shared pathophysiology. Findings from this study highlight the need to take the critical next steps in understanding accelerated epigenetic age and its contribution to health and disease. In particular, it is important to identify the biological mechanisms underlying accelerated aging and determine if intervention, including pharmacological approaches, meaningfully alter or improve the pace of epigenetic aging and thereby reduce the risk of premature health decline.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291718001411
Acknowledgements
This work was supported by Merit Review Award Number I01 CX-001276-01 from the United States (U.S.) Department of Veterans Affairs Clinical Sciences R&D (CSRD) Service and by a Presidential Early Career Award for Scientists and Engineers (PECASE 2013A) to EJW, as administered by U.S. Department of Veterans Affairs Office of Research and Development. Funding for this work is also supported by NIMH grant R21MH102834 to MWM and the Translational Research Center for TBI and Stress Disorders (TRACTS), a VA Rehabilitation Research and Development Traumatic Brain Injury Center of Excellence (B9254-C), and the Cooperative Studies Program, Department of Veterans Affairs. FGM's contribution to this work was supported by National Institute of Mental Health award # 5T32MH019836-16. This research is the result of work supported with resources and the use of facilities at the Pharmacogenomics Analysis Laboratory, Research and Development Service, Central Arkansas Veterans Healthcare System, Little Rock, Arkansas. The contents of this article do not represent the views of the US Department of Veterans Affairs, the National Institutes of Health, or the US Government.
Conflict of interest
All authors report no financial or other conflicts of interest in relationship to the contents of this article.
