INTRODUCTION

There has been a strong and consistent association between early life economic and social stressors and a range of adverse health outcomes in adulthood, including chronic health conditions, and a greater risk of disease-specific and all-cause mortality (1–4). This patterning suggests that there may be common biological mechanisms that link economic and social stressors with accelerated aging (i.e., increase in biological aging compared with chronological age) and chronic disease development (5–7). Lower socioeconomic status (SES) and greater exposure to stressful experiences may put greater demands and “wear and tear” on biological systems, resulting in accelerated decline of physiological functioning and aging over time (8,9). Telomere length (TL) and DNA methylation (DNAm) are two aging indicators that are useful in examining these biological aging processes (10,11). In addition, both TL and accelerated epigenetic aging estimates have been shown to be independent predictors of mortality risk (12–14).

Epigenetic alterations in DNAm are a potential biological pathway through which social stressors across life domains become physically embodied and get “under the skin” at the molecular level. DNAm, or the addition of methyl groups on specific areas of DNA, plays a role in regulation of gene expression and can modulate a cell’s ability to switch on particular genes. Epigenetic aging is calculated from a selected set of age-related CpG sites depending on the epigenetic clock used (10,15,16). There is no criterion standard for measuring epigenetic age, and each approach uses its own set of assumptions in regard to DNAm aging capturing different aspects of the biological aging process. The Horvath and Hannum clocks, considered first-generation clocks, were developed to predict chronological age. Second-generation clocks (i.e., GrimAge, PhenoAge, Skin, and Blood) incorporate other aging indicators and are deemed to be more predictive of morbidity and mortality than first-generation clocks (17). The Dunedin Pace of Aging (DunedinPACE) clock was recently developed to capture the pace of aging, where DNAm responses to the variation in function decline of multiple organ systems represent the biological processes of aging. Epigenetic “age acceleration” (the difference between epigenetic age and chronological age) has been found to predict all-cause mortality among aging cohorts, independent of chronological age and risk factors (12,18) and has also been associated with economic and social stressors (19–28). For example, experiences of racial discrimination have been associated with accelerated epigenetic aging among African American young adults (29). SES has been associated with epigenetic aging among participants in both the Multi-Ethnic Study of Atherosclerosis and the Health Retirement Study studies (30). Other work examining the relation between traumatic experiences and epigenetic aging has had mixed findings, with stronger associations noted when considering posttraumatic stress disorder and epigenetic aging (31). Thus, a growing body of evidence indicates that DNAm age may be particularly useful and robust in capturing the biological consequences of the accumulation of stressors or “weathering” across the life course (10,12). However, to date, only a few studies have examined childhood stressors in relation to profiles of DNAm in adulthood and altered gene expression, and those studies have been largely among relatively homogenous White populations outside of the United States (32–36). Furthermore, even fewer studies have been able to methodologically or analytically implement a prospective life course approach to examine the accumulation of stressors in both childhood and adulthood in relation to DNAm-based age acceleration (31).

Telomeres, the portions of DNA on the ends of the chromosomes, are key to replicative aging in somatic cells. TL shortens with each cell division, and very short TL results in cellular senescence. TL is similar in cell types within an individual but is highly variable between individuals. Although telomeres shorten with each cell division (and thus with age), shorter TL is associated with increased earlier development of some chronic diseases such as cardiovascular disease (37) and diminished longevity (38). Research specifically examining the association between SES and TL has been mixed, with some studies finding positive associations with SES and others finding no association (39–41). There is evidence that stressful life experiences may be associated with TL (11,42–48), including stress in childhood (49). Racial discrimination has also been found to be associated with shorter TL length among Black populations (50), even after controlling for other forms of stress (51). Potentially contributing to the mixed findings, with a few exceptions, most of the studies in this area have been cross-sectional (40,52,53). Some studies suggest that Black populations have shorter TLs than Whites overall (54,55), whereas other studies have found opposite patterns (56). Furthermore, despite research indicating that there are sex differences in the association between stress and health outcomes (57,58), few studies have examined potential sex differences between stress and TL (41,59).

This work is guided by the life course theory (60–63) as well as weathering of biological systems (9), which proposes that the accumulation of social and economic stressors across the life course accelerates the decline of physical health outcomes. Thus, the accumulation of a particular stressor (e.g., racial discrimination) across the life course or experiencing stressors across multiple domains across the life course may lead to accelerated aging and weathering, particularly among minoritized and disadvantaged populations (9). In addition, its plausible that experiencing stressors at a particular time across the life course (sensitive period) is associated with worse outcomes. We examine how stressors across multiple domains (economic, traumatic, relational, and experiences of discrimination) experienced across two time periods of exposure (e.g., childhood and/or adulthood) are associated with measures of epigenetic aging. We furthermore examine whether associations differ by the timing of experiences (childhood or adulthood) and furthermore examine whether associations differ by race and sex.

METHODS Sample

The DISPAR-Aging sample was drawn from the Child Health and Development Studies Disparities Study (DISPAR), which was designed to evaluate life course predictors of health disparities (64). The adult participants in DISPAR (N = 605) were originally recruited as the offspring of women who participated in the Child Health and Development Studies, a pregnancy cohort that recruited pregnant women receiving prenatal care from the Kaiser Family Foundation Health Care Plan at its clinics in Alameda County, California, between 1959 and 1967. Follow-up examinations and in-person interviews were conducted among mothers (at offspring ages 5, 9–11, and 15–17 years) and their offspring (at offspring ages 5, 9–11, and 15–17 years). Of the 605 adults in DISPAR, 84.3% (N = 510) agreed to participate in a home visit, where serology and anthropometric measurements were taken. Of these 510 participants, 78.4% (N = 400) participated in the serology and 372 agreed to the use of their specimens for future DNA analyses. Although the DNAm and TL assays were based on samples collected at age ~50 years, stressor exposures were collected across the participant’s life course. Of the 605 adults in DISPAR, 513 respondents participated at age 5 years (85.1%), 551 (91.4%) participated at age 9 years, and 447 (78.8%) participated at age 15 years. Overall, 359 respondents (59.3%) had complete data on stressors, completed serology, and agreed to the use of their biospecimens for DNA analyses. All participants provided informed consent. No differences across demographic measures were noted when comparing the full DISPAR sample (N = 605) to the analytic sample (n = 359) (Table S1, Supplemental Digital Content, https://links.lww.com/PSYMED/A999). These analyses were approved by the Institutional Review Board at Emory University. Investigators interested in accessing the data should contact the corresponding author.

Stress Measures

As previously described, we calculated two stressor life period scores based on guidance from the Stress Measurement Network (65) and the work of Epel and colleagues (66) as a guiding framework: a) childhood and b) adulthood. The childhood period score included 11 different items, whereas the adulthood period included 17 (Table S2, Supplemental Digital Content, https://links.lww.com/PSYMED/A999). Composite scores for each life period were created by dichotomizing each item as described hereinafter; items were then summed and divided by the number of items present. To make comparisons across the two life period scores, the scores were standardized to have a mean of 0 and standard deviation of 1. We created a total life stress measure that combines the childhood and adulthood stress measures by adding the composite scores then standardizing to have a mean of 0 and standard deviation of 1.

The life period measures include questions on economic (64,67,68) stressors such as difficulties paying bills and belonging to the lowest quartile of composite SES (occupation, education, and income) at various time points across life course; trauma (69) including stressors such as having experienced the death of a loved one and history of incarceration; relational (67,69) including stressors such as belonging to the highest quartile of adolescent-parental relationship dysfunction reported at age 15 years and belonging to the highest quartile of spousal relationship dysfunction reported at age 50 years. We assessed various forms of discrimination based on the Major Experiences of Discrimination Scale (70) administered at age ~50 years. After reporting a discrimination experience, participants were asked to attribute the reason for the discrimination from a list of possible reasons such as race, skin color, sex, religion, sexual orientation, and body appearance. We included discrimination for any reason, although the main sources in our sample are race and skin color.

The childhood stress score was characterized as any exposure occurring before age 18 years and included exposures such as experience of parental separation and adult report of parental drug/alcohol problems. Adulthood stress score was characterized as any exposure occurring at age 18 years or later and included exposures such as death of a child, experience of marital termination, and experience of physical assault. Because the precise age of a participant when a sibling died was not registered, death of a sibling was characterized as occurring in childhood or adulthood, depending on the age of death of the sibling (younger than 18 years versus 18 years or older). Major discrimination was included in adulthood, as experiences described (i.e., fired from job) pertained to adulthood rather than childhood. There were three events that are included in both adulthood and childhood as participants were asked when the exposure occurred: a) arrest/incarceration history, b) witnessing death or injury, and c) death of a close friend.

DNA Assays

Genomic DNA was extracted from the buffy coat fraction of centrifuged blood, collected at enrollment and preserved at −196°C, with the PureLink Genomic DNA Mini Kit by Invitrogen (Carlsbad, California) according to the vendor’s recommended protocol on an automated purification system (Invitrogen iPrep Purification System).

DNAm Analyses and Epigenetic Aging

Genome-wide DNAm measures were generated on participants’ DNA with high enough quality/concentration (n = 359) using the Illumina Infinium Methylation Epic (850k) BeadChip (Illumina, Inc, San Diego, California). Samples were distributed randomly across the chips. All samples passed quality control procedures. We conducted exploratory analyses using the β value (i.e., the proportion of methylated signal in total signals, ranging from 0 to 1), which has more straightforward biological interpretation. Functional normalization (“noob” for dye-bias and background correction) was then performed to account for technical variation in the data (71). Blood cell proportions were estimated using the Houseman method (72). To maximize the number of sites available for the epigenetic age calculator, probes with detection p values greater than .01 were coded as NA for poor performing samples only and were otherwise retained.

For analysis of epigenetic age, we calculated six different DNAm age estimates. Pan-tissue (10), Hannum (16), Skin and Blood (73), GrimAge (74), and PhenoAge (75) were calculated by uploading the data to the publicly available online calculator (https://dnamage.genetics.ucla.edu/) with the “normalization” and “advanced analysis in blood” options selected from a subset of 30,084 probes with accompanying age, sex, and tissue information. From the results file, we extracted the corresponding residual DNAm age calculations informed by existing epigenetic clocks (AgeAccelerationResidual, DNAmAgeHannumAdjAge, DNAmAgeSkinBloodClockAdjAge, DNAGrimAgeAdjAge, DNAmPhenoAgeAdjAge), representing estimates of DNAm age advancement by adjusting for chronological age at the time of the DNA assessment. We also calculated the DunedinPACE score in R following the procedures described in Belsky et al. (76) using the DunedinPACE R package (https://github.com/danbelsky/DunedinPACE), a new DNAm measure of biological aging.

Telomere Assays

LTL of the 359 DISPAR participants was assessed using the quantitative polymerase chain reaction–based method (77) modified for a high-throughput 384-well format in a reaction volume of 10 μl. Briefly, 5 ng of genomic DNA was assayed in either the telomere or 36B4 (single-copy gene) polymerase chain reaction mixture. Triplicate reactions of telomere and 36B4 reactions were performed for each sample. The relative telomere to single gene (T/S) ratio for each sample was calculated by subtracting the average 36B4 threshold cycle (Ct) value from the average telomere Ct value. The relative T/S ratio was calculated by subtracting the T/S ratio of a calibrator DNA from the T/S ratio of each sample. Quality control samples were interspersed throughout the plates to assess interplate and intraplate variability of Ct values. The coefficients of variation of the Exponential −ddCt values are 8% to 13% in previous projects performed at this facility. The average standard curve R was greater than 99%. The coefficient of variation in the quality control sample was 5.7%.

Covariates

Respondent’s race/ethnicity was based on the self-identified race/ethnicity of the participants in their adult life. Sex and smoking status were also assessed at age 50 years. Smoking status was characterized into two groups—current or former smoker and never smoker. Cell composition was estimated using the method by Houseman et al. (72), using CD8+ T cells, CD4+ T cells, NK cells, B cells, and monocytes.

Data Analyses

We analyzed each of the epigenetic age estimates in relation to total life stress, and childhood and adulthood cumulative stress using multivariable linear regressions. Models are first presented adjusting for sex and race, and subsequently adjusted for sex, race, and smoking status. We tested interactions with life stress period measures and race and separately with sex. No significant interactions between race and stress measures were identified. Statistically significant interactions between sex and stress measures were found for some associations leading us to present sex-stratified analyses. Lastly, regression models were conducted jointly accounting for child and adult cumulative stress in the same model. For determining significance in epigenetic age analyses, we used a Bonferroni threshold of (0.05/6 clocks + 1 TL) = 0.0071 to adjust for multiple hypotheses. All analyses were conducted in R (http://www.R-project.org) using packages from Bioconductor.

In a sensitivity analysis, we tested for associations adjusting for potential heterogeneity in cell proportions for all models and in separate analyses accounting for batch effects. We controlled for batch effects by adjusting for BeadChip number. It is possible that childhood stressors increase the likelihood of experiencing stressors later in life (stress proliferation theory); we thus formally tested adult cumulative stress as a mediator of child cumulative stress on epigenetic aging and TL using the “psych” package in R (78), which uses a Preacher and Hayes (79,80) approach to mediation testing. The indirect paths were bootstrapped, and the default bootstrap number of 5000 was used.

Lastly, we also tested moderation as the impact of child stressors may differ across adult stress experiences. We thus tested for interactions between child and adult cumulative stress on epigenetic aging.

RESULTS

Characteristics of our analytical sample of 359 participants are displayed in Table 1. Slightly more than half were male (53.2%). Among male participants, the mean age was 49.2 years, more than half were White (56.5%), and more than half were nonsmokers (57.1%). Among female participants, the mean age was 49.52 years, more than half were White (56.5%), and slightly more than half were never smokers (51.8%). Some sex differences in relation to stress experiences were also noted (Table 1). For example, women faced greater levels of stress in childhood compared with men (mean [SD] = 1.1 [0.9] for women and 0.9 [0.8] for men), but no differences were noted with regard to adult stressors.

TABLE 1 - Descriptive Statistics, Cumulative Stress and Epigenetic Indices of Aging, and Relative TL of the Sample Stratified by Sex (N = 359) Male (n = 191) Female (n = 168) Age, mean (SD) 49.20 (1.41) 49.52 (1.02) Race, n (%)  White 108 (56.5) 95 (56.5)  Black 68 (35.6) 55 (32.7)  Other race/ethnicity 15 (7.9) 18 (10.7) Smoking status, n (%)  Current/former 82 (42.9) 81 (48.2)  Nonsmoker 109 (57.1) 87 (51.8) Childhood stress  Mean (SD) −0.170 (0.896) −1.089 (0.997)  Range −1.09 to 3.21 −1.09 to 2.92 Adulthood stress  mean (SD) −0.090 (0.944) −0.008 (1.060)  Range −1.70 to 2.74 −1.70 to 2.40 Total life stress  Mean (SD) −0.058 (0.877) 0.090 (1.100)  Range −1.71 to 2.22 −1.714 to 3.062 Horvath age, y  Mean (SD) 59.45 (4.06) 57.91 (4.67)  Range 49.38–70.99 27.33–73.15 Hannum age, y  Mean (SD) 62.89 (4.22) 61.34 (3.99)  Range 50.20–77.67 46.86–71.52 Skin and Blood age, y  Mean (SD) 60.44 (2.84) 60.17 (3.04)  Range 52.12–66.85 50.78–68.24 Grim age, y  Mean (SD) 65.20 (4.31) 63.85 (4.48)  Range 56.99–78.85 52.85–77.63 Pheno age, y  Mean (SD) 43.74 (5.89) 43.11 (6.05)  Range 31.36–66.53 25.45–64.34 DunedinPACE age, y  Mean (SD) 0.963 (0.135) 0.976 (0.136)  Range 0.64–1.47 0.71–1.36 Relative TL  Mean (SD) 0.846 (0.170) 0.845 (0.216)  Range 0.37–1.51 0.32–1.54

SD = standard deviation; TL = telomere length.

Epigenetic Age Analyses

Epigenetic age estimates were all (excluding PhenoAge and DunedinPACE) moderately correlated with chronological age (Table S3, Supplemental Digital Content, https://links.lww.com/PSYMED/A999), which could be due to the limited variability in chronological age in the sample; the Skin and Blood clock showed the highest correlation (r = 0.268), and DunedinPACE showed the lowest (r = −0.034). All clocks and DunedinPACE were also significantly correlated with each other, although none higher than r = 0.640. Horvath’s DNAmAge was most highly correlated with the Skin and Blood clock (r = 0.484) and least with relative TL (r = 0.112; Table S3).

We report distinct patterns of associations between total life stress, child and adult stress scores, and each of the different epigenetic age and relative TL and DunedinPACE estimates, adjusting for chronological age, as well as sex, race, and smoking status (Table 2). In models adjusting for sex, race, and smoking status, increased total life stress score was associated with increased epigenetic age in the GrimAge clock, Skin and Blood clock, and DunedinPACE. When considering time period of exposure, increased levels of childhood and adulthood stress were associated with increased epigenetic age in the GrimAge clock and DunedinPACE. Only adulthood stress was associated with the Skin and Blood clock (β = 0.488 years, SE = 0.156, p = .002) after adjusting for smoking status (Table 2). These associations persist after further adjusting for smoking status. Neither stress time period was associated with Horvath’s clock, Hannum Age, PhenoAge, or relative TL.

TABLE 2 - Associations Between Childhood Stress and Adulthood Stress and Epigenetic Indices of Aging and Relative Telomere Length Horvath Age Accel Hannum Age Accel Skin and Blood Age Accel Grim Age Accel Pheno Age Accel DunedinPACE Relative Telomere Length Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Adjusted for sex and race  Child stress score −0.089 (0.239) −0.321 (0.223) −0.229 (0.158) 0.807 (0.237)* 0.136 (0.333) 0.031 (0.007)* −0.014 (0.011)  Adult stress score 0.295 (0.229) −0.326 (0.214) 0.366 (0.151)** 1.256 (0.221) 0.354 (0.319) 0.041 (0.007)* 0.017 (0.010)  Total life stress 0.179 (0.231) −0.342 (0.215) 0.247 (0.152) 1.423 (0.219)* 0.296 (0.321) 0.045 (0.007)* 0.006 (0.010) Adjusted for sex, race, and smoking status  Child stress score −0.121 (0.243) −0.288 (0.226) −0.186 (0.160) 0.524 (0.220)** 0.053 (0.337) 0.027 (0.007)* −0.015 (0.011)  Adult stress score 0.267 (0.239) −0.279 (0.223) 0.488 (0.156)* 0.808 (0.215)* 0.228 (0.332) 0.037 (0.007)* 0.018 (0.011)  Total life stress 0.186 (0.250) −0.238 (0.235) 0.517 (0.161)* 0.603 (0.208)* 0.117 (0.352) 0.035 (0.007)* 0.008 (0.011) Adjusted for race  Male  Child stress score −0.198 (0.531) −0.071 (0.333) −0.247 (0.218) 0.214 (0.338) −0.136 (0.477) 0.021 (0.010)** −0.014 (0.014)  Adult stress score −0.145 (0.309) −0.378 (0.325) 0.481 (0.211)** 1.407 (0.314)* 0.238 (0.466) 0.049 (0.010)* 0.017 (0.013)  Total life stress −0.125 (0.334) −0.191 (0.352) 0.433 (0.229) 1.560 (0.339)* 0.135 (0.504) 0.055 (0.010)* 0.011 (0.014)  Female  Child stress score −0.012 (0.358) −0.536 (0.297) −0.218 (0.230) 1.339 (0.328)* 0.381 (0.469) 0.039 (0.010)* −0.014 (0.017)  Adult stress score 0.596 (0.335) −0.303 (0.282) 0.249 (0.217) 1.099 (0.314)* 0.447 (0.443) 0.035 (0.009)* 0.018 (0.016)  Total life stress 0.277 (0.324) −0.477 (0.269) 0.095 (0.209) 1.305 (0.295)* 0.398 (0.425) 0.038 (0.009)* 0.004 (0.015) Adjusted for race and smoking status  Male  Child stress score −0.194 (0.317) −0.068 (0.334) −0.224 (0.218) 0.093 (0.310) −0.139 (0.479) 0.020 (0.010)** −0.014 (0.014)  Adult stress score −0.138 (0.313) −0.335 (0.328) 0.542 (0.212)** 1.177 (0.294)* 0.237 (0.472) 0.047 (0.010)* 0.018 (0.014)  Total life stress −0.009 (0.345) 0.013 (0.374) 0.733 (0.232)* 0.900 (0.315)* 0.172 (0.543) 0.048 (0.011)* 0.017 (0.015)  Female  Child stress score −0.102 (0.370) −0.527 (0.307) −0.162 (0.237) 0.936 (0.316)* 0.135 (0.480) 0.033 (0.010)* −0.018 (0.017)  Adult stress score 0.563 (0.371) −0.287 (0.312) 0.431 (0.238) 0.443 (0.325) 0.050 (0.484) 0.027 (0.010)** 0.016 (0.017)  Total life stress 0.115 (0.363) −0.548 (0.303) 0.318 (0.232) 0.326 (0.286) −0.120 (0.470) 0.025 (0.010)** −0.001 (0.017)

Accel = accelerated; SE = standard error.

*Values represent significant associations of p values <.00714.

**Values indicate marginally significant associations with p values <.05.

In models stratified by sex, total life stress was associated with increased epigenetic age in GrimAge clock and DunedinPACE among men and women; however, after adjusting for smoking status, only associations among men are noted with GrimAge, Skin and Blood, and DunedinPACE. When examining time period of exposure, increased levels of adulthood stress were associated with increased epigenetic age in the GrimAge clock and DunedinPACE in male participants, whereas increased levels of childhood and adulthood stress are associated with increased epigenetic age in the GrimAge clock and DunedinPACE in female participants. When adjusting for smoking status, associations in male participants remained; however, only childhood stress associations in female participants remained significant. We have also included estimates with both stress periods in the same model (Table 3), which produced results similar to those in separate models.

TABLE 3 - Joint Associations Between Childhood and Adulthood Stress and Epigenetic Indices of Aging and Relative Telomere Length Horvath Age Accel Hannum Age Accel Skin and Blood Age Accel Grim Age Accel Pheno Age Accel DunedinPACE Relative Telomere Length Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE) Adjusted for sex and race  Child stress score −0.128 (0.241) −0.284 (0.224) −0.278 (0.158) 0.661 (0.230)* 0.093 (0.336) 0.026 (0.007)* −0.016 (0.011)  Adult stress score 0.310 (0.231) −0.293 (0.215) 0.398 (0.151)** 1.181 (0.220)* 0.343 (0.322) 0.038 (0.007)* 0.019 (0.010) Adjusted for sex, race, and smoking status  Child stress score −0.143 (0.243) −0.268 (0.226) −0.278 (0.158) 0.463 (0.217)** 0.036 (0.338) 0.024 (0.007)* −0.016 (0.011)  Adult stress score 0.278 (0.240) −0.259 (0.223) 0.398 (0.151)** 0.773 (0.215)* 0.225 (334) 0.035 (0.007)* 0.019 (0.011) Adjusted for race  Male  Child stress score −0.205 (0.317) −0.107 (0.333) −0.227 (0.216) 0.274 (0.322) −0.126 (0.478) 0.023 (0.010)** −0.013 (0.014)