Associations of Early-Life Adversity With Later-Life Epigenetic Aging Profiles in the Multi-Ethnic Study of Atherosclerosis

Abstract

Epigenetic biomarkers of accelerated aging have been widely used to predict disease risk and may enhance our understanding of biological mechanisms between early-life adversity and disparities in aging. With respect to childhood adversity, most studies have used parental education or childhood disadvantage and/or have not examined the role played by socioemotional or physical abuse and trauma in epigenetic profiles at older ages. This study leveraged data from the Multi-Ethnic Study of Atherosclerosis (MESA) on experiences of threat and deprivation in participants’ early lives (i.e., before the age of 18 years) to examine whether exposure to specific dimensions of early-life adversity is associated with epigenetic profiles at older ages that are indicative of accelerated biological aging. The sample included 842 MESA respondents with DNA methylation data collected between 2010 and 2012 who answered questions on early-life adversities in a 2018–2019 telephone follow-up. We found that experiences of deprivation, but not threat, were associated with later-life GrimAge epigenetic aging signatures that were developed to predict mortality risk. Results indicated that smoking behavior partially mediates this association, which suggests that lifestyle behaviors may act as downstream mechanisms between parental deprivation in early life and accelerated epigenetic aging in later life.

Abbreviations

ACEs

adverse childhood experiences

CpG

cytosine-phosphate-guanine

CVD

cardiovascular disease

DNAm

DNA methylation

EAA

epigenetic age acceleration

ELA

early-life adversity

MESA

Multi-Ethnic Study of Atherosclerosis

SD

standard deviation

SE

standard error

SES

socioeconomic status

TILDA

Irish Longitudinal Study on Aging

The Centers for Disease Control and Prevention has estimated the annual economic burden of child maltreatment in the United States to be in the trillions of dollars, in part because exposures early in life continue to impact health and well-being throughout the life span (1). Early-life adversity (ELA), or experiences of socioemotional or physical abuse and trauma in childhood, have been linked to accelerated pubertal development, cellular aging, and compromised brain development, which may contribute to poorer mental and physical outcomes across the life course (2–6). Research has begun to document the negative impacts of ELA on adult health and well-being, including inflammation, poorer metabolic functioning, phenotypic aging (7), psychological disorders (8, 9), and mortality (10, 11). Theoretically, relationships between childhood adversity and later-life health may arise due to latent independent effects that disrupt critical or sensitive periods of development in early life (i.e., “biological embedding” or “programming” of adult disease), and/or via the role of early environment in subsequent life trajectories that have cumulative or “weathering” effects on health (12–16).

In this study, we contribute to the growing literature on the long reach of childhood adversity by examining whether ELA is correlated with accelerated biological aging in older adults. Biological aging is the progressive loss of physiological integrity that occurs with advancing chronological age, leading to impaired function and increased vulnerability to death (17). We measure biological aging using epigenetic aging measures that leverage age-related changes in DNA methylation (DNAm) across the genome. DNAm is a type of epigenetic modification that refers to the addition of a methyl group to a cytosine nucleotide at a cytosine-phosphate-guanine (CpG) site. DNAm can affect gene expression by altering DNA accessibility and can change over time as a function of genes and the environment.

Epigenetic aging measures, often referred to as epigenetic clocks or DNAm age, are estimated by taking the genomewide weighted average of DNAm at CpG sites that are highly associated with chronological age or phenotypic hallmarks of aging. DNAm age accurately predicts chronological age (18–23), and numerous studies have linked deviations between DNAm age and chronological age—that is, epigenetic age acceleration (EAA)—with age-related diseases and mortality (24–35), suggesting EAA measures are molecular biomarkers of aging that reflect both resilience and vulnerability during the aging process. EAA measures tend to outperform other biomarkers of aging in predicting life spans (36–38), and their correlations with age-related conditions make them useful in a variety of contexts, including anti-aging interventions (39).

A growing number of studies have investigated whether ELA is associated with accelerated epigenetic aging in youths and adults (2, 4–6, 40–46), but to date only a handful of studies have examined whether epigenetic alterations persist at older ages (≥50 years) and, if so, the potential mechanisms that may be driving these associations across the life course (43, 45, 47, 48). Research to date has also primarily used small and/or single-sex samples (5, 6, 41, 49) and has only examined associations with Horvath et al. (19) and Hannum et al. (22) epigenetic clocks that were trained on chronological age. These clocks have shown inconsistent associations with socioeconomic status (SES) across studies (50), whereas next-generation clocks trained on phenotypic aging and mortality risk (DNAm PhenoAge and DNAm GrimAge) and pace-of-aging metrics trained on the rate of decline in system integrity (DunedinPoAm and DunedinPACE) have shown more consistent associations with SES (50–53). A handful of recent studies have broadened the examination of ELA and EAA to include next-generation epigenetic aging measures. In a study that most closely parallels the aims of this study, childhood poverty, but not ELA (as captured by self-reports of physical or sexual abuse, death of a parent, or parental alcohol or drug abuse), was associated with significant GrimAge EAA and DunedinPoAm among 490 participants aged 50–87 years in the Irish Longitudinal Study on Aging (TILDA) (46). In a study of 183 premenopausal women aged 25–51 years, childhood abuse, but not neglect, predicted faster GrimAge EAA (4). In the Health and Retirement Study (n = 2,672), adverse childhood experiences (ACEs) (i.e., low parental education, parental physical abuse and alcohol/drug abuse, separation from either parent, death of a parent, living in a foster home or orphanage, parental separation or divorce, and childhood poverty) were associated with GrimAge EAA and DunedinPoAm at older ages and partially mediated the association between ACEs and depressive symptoms (47).

Notably, research has highlighted the importance of distinguishing between subtypes of ELA that may differentially affect accelerated development (5). Specifically, the dimensional model of adversity and psychopathology posits that a wide range of ELA experiences can be organized into 2 underlying dimensions: threat, which involves physical, emotional, or sexual harm or the threat of harm to the child, and deprivation, or the absence of appropriate physical or emotional stimulation or nurturance in childhood (54, 55). Distinct types of ELA may influence the pace of development differently, and ELA types that accelerate development may not necessarily influence biological aging later in the life course. For example, studies in children and adolescents have found that experiences of threat are linked to advanced pubertal timing, cortical thinning, and shorter telomere length, whereas deprivation was unrelated to accelerated development, perhaps because it does not pose an immediate threat to survival and reproductive fitness (56). Studies on the ELA-EAA relationship in children support the existence of such a pattern, finding a relationship between EAA and threat or trauma but not deprivation or neglect (4, 5, 57). However, to our knowledge, less attention has been given to the question of whether differential ELA-EAA patterns by ELA type (threat vs. deprivation)  persist or diverge in their trajectories across the life course, producing weaker associations among older adults. For example, a recent study provides some initial evidence on this point, suggesting that poverty, which is similar to our dimension of deprivation though more specifically related to financial disadvantage as opposed to a lack of socioemotional attention from caregivers, is perhaps more important than their measure of abuse (46).

In this study, we leverage newly collected ELA data on adults aged ≥50 years in the Multi-Ethnic Study of Atherosclerosis (MESA) (n = 842) to examine relationships between ELA in childhood and 6 epigenetic aging biomarkers at older ages. We extend prior work by 1) evaluating whether associations at older ages vary by ELA type, distinguishing between experiences characterized by threat and those characterized by deprivation; 2) evaluating whether ELA associations are global or specific to how epigenetic aging is measured; and 3) assessing whether these associations vary according to sociodemographic characteristics, including age, sex, race/ethnicity, and childhood SES. Finally, we examine health or lifestyle behaviors that may be driving associations at older ages, including current/former smoking, alcohol use, and obesity.

METHODS

Multi-Ethnic Study of Atherosclerosis

MESA is a population-based longitudinal study that was designed to identify risk factors for the progression of subclinical cardiovascular disease (CVD) (58). A racially and ethnically diverse sample of 6,814 men and women aged 45–84 years without clinically apparent CVD was recruited between July 2000 and August 2002 from the following 6 regions in the United States: Forsyth County, North Carolina; northern Manhattan and the Bronx, New York, New York; Baltimore City and Baltimore County, Maryland; St. Paul, Minnesota; Chicago, Illinois; and Los Angeles County, California. Five additional examinations have since been completed. DNAm data were collected at examination 5, conducted between April 2010 and February 2012. DNAm was assessed in monocytes in a subsample of 1,264 non-Hispanic White, non-Hispanic Black, and Hispanic MESA participants aged 55–94 years from the Baltimore, Forsyth County, New York, and St. Paul field centers who agreed to participate in an ancillary study examining the effects of methylation on CVD. ELA data were collected at MESA telephone follow-up 20, conducted from 2018 to 2019. We excluded 422 respondents with DNAm data who had missing ELA exposure data. When compared with the baseline examination 1 MESA sample, the DNAm-ELA sample contained fewer Black participants, current smokers, and individuals without a high school diploma (see Web Table 1, available at https://doi.org/10.1093/aje/kwad172).

Our final study sample included 842 participants with both DNAm and ELA data. This study was approved by the institutional review boards of all MESA field centers, the MESA Coordinating Center, the University of Wisconsin–Madison, the University of Michigan, and the University of California, Los Angeles.

DNAm data

A detailed description of the data extraction and processing procedures used to profile DNAm data in MESA participants can be found elsewhere (59). Briefly, blood was drawn in the morning after a 12-hour fast. Monocytes were isolated using autoMACs automated magnetic separation units (Miltenyi Biotec B.V. & Co. KG, Bergisch Gladbach, Germany) and were consistently more than 90% pure. Samples were plated using a stratified random sampling technique to reduce bias from batch, chip, and position effects. Methylation was measured using the Illumina HumanMethylation450 BeadChip kit (Illumina, Inc., San Diego, California). Quantile normalization was performed using the “lumi” package in R with default settings (60). Quality control measures included checks for sex and race/ethnicity mismatches and outlier identification. Criteria for elimination of probes included: “detected” methylation levels in less than 90% of MESA samples (using a detection P value threshold of 0.05), overlap with a nonunique region, and probes that assay single nucleotide polymorphisms (61). We adjusted for chip and position effects in all analyses.

Measures

Epigenetic aging outcomes.

We analyzed 6 epigenetic aging biomarkers, including the Horvath, Hannum, PhenoAge, and GrimAge epigenetic clocks, the epiTOC “mitotic clock,” and the DunedinPACE pace-of-aging measure (50, 62–65). First-generation clocks, including the Horvath and Hannum clocks, use elastic net to train an epigenetic predictor of chronological age in whole blood (Hannum) or across several tissue and cell types (Horvath) (19, 22). Second-generation clocks, including PhenoAge and GrimAge, incorporate more complex training phenotypes based on clinical, phenotypic aging measures and mortality risk to more closely capture underlying biological features of accelerated aging (36, 38). All first- and second-generation clocks incorporate DNAm levels from selected CpG sites into an age estimate that can be compared with chronological age to assess whether individuals are aging faster or slower biologically than they are chronologically. The epiTOC clock was developed to approximate a “mitotic clock” that estimates the average level of methylation across 385 CpG sites associated with age-related increases in cell division or cell-replication errors (66). More recently, pace-of-aging measures that capture individual variation in the rate of aging over time using longitudinal DNAm data have been developed. The DunedinPACE measure was developed by tracking DNAm patterns associated with the rate of change in 19 system-integrity biomarkers over the course of 20 years in Dunedin Study participants (67). DunedinPACE values over 1 indicate faster biological aging than expected, and values below 1 indicate slowed aging.

We constructed epigenetic aging biomarkers in MESA from the CpG-level data according to author-specific algorithms using weights estimated in outside reference samples that did not include MESA participants (Web Table 2 provides further details on construction and interpretation). In regression analyses, we adjusted for age so that the dependent variable represented epigenetic age or pace of aging net of chronological age.

Measures of exposure to ELA.

Retrospective reports of ELA were measured at MESA phone follow-up 20 using 6 items adapted from the scale of Felitti et al. (9) (Table 1). Participants were asked to rate aspects of their family environment during childhood or before age 18 years, on 5-point Likert scales that ranged from 1 (“never”) to 5 (“very often”). Responses to these questions were classified into 2 primary dimensions: threat and deprivation. Threat is the mean of the Likert scores for questions relating to emotional and physical abuse, and deprivation is the mean of the reversed Likert score across all 4 deprivation questions relating to parental monitoring and parental warmth. Parental monitoring and parental warmth were also analyzed separately using the mean Likert score of the 2 questions pertaining to each dimension. Finally, we created a count index for the total number of early-life adversities (called the 6-item ACE questionnaire) by taking the sum of all 6 ELA questions after they were converted to binary indicators; this latter index most closely parallels earlier work using the original summated Felitti index of adverse childhood experiences (68). Responses to threat questions were assigned a value of 1 if individuals reported that adverse experiences happened “sometimes,” “often,” or “very often” and 0 otherwise. Responses to deprivation questions were assigned a value of 1 if experiences occurred “almost never” or “never” and 0 otherwise. In total, we analyzed 5 ELA outcomes that capture global or specific dimensions of ELA within the family environment: 6-item ACE questionnaire score, threat, deprivation, parental warmth, and parental monitoring. We acknowledge that these measures were not meant to be a definitive assessment of all possible ELA exposures in childhood and were designed to reflect physical/emotional threat of harm or the absence of physical/emotional nurturance within the home specifically.

Table 1.

Questions on Exposure to Early-Life Adversity Fielded in the Multi-Ethnic Study of Atherosclerosis, 2018–2019

6-Item ACE Questionnaire Item	Exposure Question
Threat	How often did a parent/other adult in the household swear/insult/put you down, or make you feel threatened?
	How often did a parent/other adult in the household push/grab/shove/hit you so hard you had marks/injuries?
Deprivation
Parental monitoring	How often did your family know what you were up to?
	How often was the household where you grew up well-organized/well-managed?
Parental warmth	How often did a parent/other adult in the household make you feel loved, supported, and cared for?
	How often did a parent/other adult in the household express physical affection for you (hugging or other physical gesture of warmth/affection)?

Abbreviation: ACE, adverse childhood experience.

Covariates.

Potential confounders include age, sex, race/ethnicity, childhood SES, adult SES, and self-reports of depressive symptoms and stressful life events. Chronological age and all other time-varying covariates were taken from examination 5 when DNAm was assayed, except for reports of stressful life events, which were taken from examination 3. Race and ethnicity were based on self-report and were categorized as non-Hispanic White, non-Hispanic Black, or Hispanic. Marital status was categorized as married or not married. Parental education and the participant’s own education were measured using the highest level of education completed and were categorized as no high school diploma, high school diploma or General Educational Development certificate, or any college certificate or degree. Total gross family income was divided into 4 categories: annual income <$20,000, $20,000–$39,999, $40,000–$74,999, or ≥$75,000.

Depressive symptoms were measured using the Center for Epidemiologic Studies Depression Scale, a 20-item self-report questionnaire that asks respondents to rate how often in the past week they have experienced depressive symptoms on a 4-point scale (0–3). The stress score was constructed from 5 questions pertaining to stressful life events. For each of 5 stressors (personal health problems, health problems of someone close to you, ongoing difficulties with your job, ongoing financial strain, and ongoing relationship difficulties), respondents were asked whether these stressors were present, whether they had lasted for 6 or more months, and how stressful they were (1 = not very stressful, 2 = moderately stressful, 3 = very stressful). The stress score was the average rating of the stressors (0 = did not last for ≥6 months and 1–3 otherwise). If information on more than 2 of the 5 items was missing, the stress score was set to “missing.”

Mediators.

Health behaviors were explored as potential mediating factors between ELA and accelerated epigenetic aging. Tobacco smoking was categorized as never smoker, former smoker (>1 year), or current smoker. Alcohol consumption was categorized as nondrinker, ≤2 alcoholic drinks/day, or >2 drinks/day. Obesity was indexed according to the World Health Organization classification of body mass index (weight (kg)/height (m)²): underweight or normal-weight (<25.0), overweight (25.0–29.9), or obese (≥30.0) (69). Exposure to secondhand smoke was set equal to 1 if the participant reported living with a smoker and 0 otherwise.

Analytical approach

Linear mixed models were used to assess associations between the 5 ELA measures and the 6 epigenetic aging biomarkers with SAS software (SAS Institute, Inc., Cary, North Carolina) (70). Since ELA measures were correlated with one another (r = 0.29–0.89, P < 0.0001; Web Table 3), to adjust for multiple-hypothesis testing, we used a Bonferroni-adjusted significance threshold for 6 independent tests and a familywise error rate of 0.05 (Bonferroni-corrected P = 0.008). Additionally, because GrimAge is a composite biomarker of 8 DNAm-based surrogate biomarkers, if an ELA measure was significantly associated with GrimAge we examined separate associations with each composite biomarker, including the DNAm-based estimator of smoking pack-years and DNAm-based estimators of plasma proteins (i.e., adrenomedullin levels, β₂ microglobulin, cystatin C, growth differentiation factor 15, leptin, plasminogen activation inhibitor 1, and tissue inhibitor metalloproteinase 1).

Model 1 regressions included controls for sex, race/ethnicity, age, and marital status, random effects for methylation chip and position, and adjustments for residual sample contamination with nonmonocytes (enrichment scores for neutrophils, B cells, T cells, and natural killer cells). Model 2 regressions added adjustments for childhood and adult SES (parental education, own education, and household income). Model 3 added adjustments for self-reports of depressive symptoms and stressful life events. Model 4 added controls for health behaviors (smoking status, secondhand smoke exposure, alcohol consumption, and obesity). To avoid losing observations with missing information on a covariate that was not our primary exposure of interest, we set missing values equal to 0 and included an additional dichotomous variable set equal to 1 if an observation was missing. Finally, we examined the extent to which attenuation in any significant ELA–epigenetic aging biomarker relationship was attributed to lifestyle behaviors in adulthood, using Karlson et al. (71) mediation analysis with R software.

RESULTS

Table 2 gives the characteristics of the sample. The mean age at blood collection was 67.56 (standard deviation (SD), 8.53) years. Average epigenetic age varied between 62.2 and 71.5 years, depending on the clock algorithm (for clocks that approximate biological age). The epiTOC clock showed an average 9% (SD, 0.01) increase in DNAm due to putative cell-replication errors. DunedinPACE, which measures how rapidly a person has aged over the recent past per year of calendar time, estimated that these individuals were on average accruing 1.24 (SD, 0.10) biological years of aging per chronological year. Participants reported more instances of deprivation (mean = 1.85 (SD, 0.77)) than threat (mean = 1.53 (SD, 0.85)) and were more likely to report deprivation due to lack of parental warmth (mean = 2.04 (SD, 1.00)) than lack of monitoring (mean = 1.66 (SD, 0.78)).

Table 2.

Characteristics of the Early-Life Adversity Sample (n = 842) at Examination 5 in the Multi-Ethnic Study of Atherosclerosis, 2010–2012

Variable	No.	%	Mean (SD)	Minimum	Maximum
Epigenetic aging biomarkersa
Horvath clock			62.21 (7.76)	37.41	91.70
Hannum clock			71.33 (7.86)	51.54	96.32
PhenoAge clock			69.93 (8.42)	42.10	99.90
epiTOC clock			0.09 (0.01)	0.08	0.16
GrimAge clock			71.50 (7.42)	56.02	91.86
DunedinPACE clock			1.24 (0.10)	0.95	1.65
Measures of ELAa,b
6-item ACE questionnaire			0.66 (1.13)	0	6
Threat			1.53 (0.85)	1	5
Deprivation			1.85 (0.77)	1	4.75
Parental warmth			2.04 (1.00)	1	5
Parental monitoring			1.66 (0.78)	1	5
Sociodemographic characteristicsa
Age, years			67.56 (8.53)	54	91
Sex
Female	450	53.44		0	1
Male	392	46.56		0	1
Race/ethnicity
White	417	49.52		0	1
Black	160	19.00		0	1
Hispanic	265	31.47		0	1
Highest level of education completed
No high school diploma	108	12.84		0	1
High school diploma/GED	301	35.79		0	1
Any college certificate/degree or more	432	51.37		0	1
Highest level of parental education completed
No high school diploma	349	40.02		0	1
High school diploma/GED	341	41.08		0	1
Any college certificate/degree or more	140	16.87		0	1
Married
Yes	496	59.98		0	1
No	331	40.02		0	1
Income, dollars/year
<20,000	121	14.81		0	1
20,000–39,999	219	26.81		0	1
40,000–74,999	260	31.82		0	1
≥75,000	217	26.56		0	1
Health behaviors/physical and mental health
Tobacco smoking
Never smoker	402	48.43		0	1
Former smoker (>1 year)	348	41.93		0	1
Current smoker	80	9.64		0	1
Alcohol drinking
Nondrinker	540	64.21		0	1
≤2 drinks/day	249	29.61		0	1
>2 drinks/day	52	6.18		0	1
Obesity
Underweight or normal-weight (BMIc <25.0)	150	17.84		0	1
Overweight (BMI 25.0–29.9)	330	39.24		0	1
Obese (BMI ≥30.0)	361	42.93		0	1
CES-D score	8.46	7.48		0	51
Stressful life eventsd	0.48	0.54		0	3

Abbreviations: ACE, adverse childhood experience; BMI, body mass index; CES-D, Center for Epidemiologic Studies Depression Scale; ELA, early-life adversity; GED, General Educational Development; SD, standard deviation.

^a Values are expressed as mean (standard deviation). For dichotomous variables, values are expressed as number (percent). Percentages for indicator variables may not sum to 100 if information was missing for some individuals.

^b Questions on ELA were posed in 2018–2019.

^c Weight (kg)/height (m)².

^d Data on stressful life events were taken from examination 3, which was conducted in 2004–2005.

Over half (53.44%) of the participants were female, and approximately half (49.52%) of participants reported being non-Hispanic White. Fifty-one percent of participants had obtained a college certificate or degree as compared with 17% of participants’ parents. Income categories were approximate quartiles, slightly skewed to the right, with only 15% of participants reporting income under $20,000 per year. Forty-eight percent of participants reported never smoking, and approximately 10% reported being current smokers. Ninety-four percent reported not drinking or consuming 2 or fewer alcoholic drinks per day, and 82% reported a body mass index of 25.0 or more, putting them in the overweight (39%) or obese (43%) category.

Regression results presented in Table 3 show a positive association between deprivation and one of its subcomponents (parental warmth) for the GrimAge clock. Associations with the 6-item ACE questionnaire and threat dimensions were null for all epigenetic aging outcomes. In baseline (model 1) estimates, a 1-unit increase in self-reports of deprivation during childhood was associated with a 0.46-year increase in GrimAge epigenetic age acceleration (standard error (SE), 0.17; P = 0.007). These findings appeared to be driven in magnitude and significance by lack of parental warmth (β = 0.37 (SE, 0.13); P = 0.004). PhenoAge was marginally associated with parental warmth, and the association was similar in magnitude to the GrimAge association (β = 0.36 (SE, 0.19); P = 0.067).

Table 3.

Associations Between Measures of Epigenetic Aging and Exposure to Early-Life Adversity (n = 842) in the Multi-Ethnic Study of Atherosclerosis, 2010–2012^a

	Model
	Model 1 b		Model 2 c		Model 3 d		Model 4 e
Epigenetic Aging Measure   and ELA Outcome	β (SE)	P Value	β (SE)	P Value	β (SE)	P Value	β (SE)	P Value
Horvath clock
6-item ACE	0.009 (0.152)	0.953	0.045 (0.152)	0.767	0.023 (0.154)	0.879	0.033 (0.156)	0.831
Threat	−0.198 (0.203)	0.330	−0.146 (0.204)	0.474	−0.216 (0.207)	0.298	−0.191 (0.210)	0.362
Deprivation	0.074 (0.221)	0.737	0.087 (0.221)	0.694	0.054 (0.224)	0.810	0.053 (0.227)	0.817
Parental warmth	0.091 (0.167)	0.592	0.115 (0.170)	0.497	0.098 (0.170)	0.566	0.086 (0.173)	0.620
Parental monitoring	−0.021 (0.217)	0.922	−0.045 (0.217)	0.836	−0.074 (0.219)	0.737	−0.052 (0.222)	0.816
Hannum clock
6-item ACE	−0.056 (0.128)	0.661	−0.045 (0.129)	0.728	−0.053 (0.131)	0.686	−0.064 (0.132)	0.629
Threat	−0.124 (0.171)	0.467	−0.097 (0.172)	0.574	−0.126 (0.175)	0.473	−0.126 (0.177)	0.478
Deprivation	−0.096 (0.185)	0.605	−0.096 (0.187)	0.608	−0.122 (0.189)	0.518	−0.153 (0.191)	0.425
Parental warmth	0.017 (0.142)	0.904	0.019 (0.144)	0.893	0.012 (0.145)	0.935	−0.019 (0.147)	0.896
Parental monitoring	−0.18 (0.181)	0.321	−0.188 (0.183)	0.304	−0.218 (0.185)	0.239	−0.221 (0.187)	0.237
PhenoAge clock
6-item ACE	0.195 (0.179)	0.275	0.201 (0.179)	0.264	0.165 (0.181)	0.363	0.193 (0.182)	0.291
Threat	−0.251 (0.239)	0.293	−0.228 (0.240)	0.342	−0.332 (0.244)	0.174	−0.302 (0.246)	0.220
Deprivation	0.343 (0.260)	0.188	0.322 (0.261)	0.217	0.271 (0.263)	0.303	0.288 (0.266)	0.279
Parental warmth	0.364 (0.199)	0.067	0.350 (0.200)	0.080	0.322 (0.201)	0.109	0.321 (0.203)	0.114
Parental monitoring	0.080 (0.255)	0.755	0.054 (0.256)	0.831	0.005 (0.258)	0.983	0.049 (0.260)	0.850
epiTOC clock
6-item ACE	0.0001 (0.0001)	0.505	0.0001 (0.0001)	0.631	0.0001 (0.0001)	0.319	0.0001 (0.0001)	0.301
Threat	0.0001 (0.0001)	0.656	0.00004 (0.0001)	0.747	0.0001 (0.0001)	0.381	0.0002 (0.0001)	0.248
Deprivation	0.0001 (0.0001)	0.657	0.00004 (0.0001)	0.770	0.0001 (0.0001)	0.427	0.0001 (0.0001)	0.468
Parental warmth	0.0001 (0.0001)	0.447	0.0001 (0.0001)	0.581	0.0001 (0.0001)	0.336	0.0001 (0.0001)	0.393
Parental monitoring	−0.00004 (0.0001)	0.806	−0.00004 (0.0001)	0.787	0.00003 (0.0001)	0.851	0.00003 (0.0001)	0.850
GrimAge clock
6-item ACE	0.18 (0.117)	0.124	0.117 (0.116)	0.314	0.095 (0.117)	0.418	0.040 (0.098)	0.682
Threat	0.012 (0.157)	0.940	−0.062 (0.155)	0.688	−0.100 (0.158)	0.526	−0.130 (0.132)	0.325
Deprivation	0.459 (0.170)	0.007f	0.399 (0.169)	0.018	0.370 (0.170)	0.030	0.206 (0.142)	0.147
Parental warmth	0.374 (0.130)	0.004f	0.316 (0.129)	0.015	0.304 (0.130)	0.019	0.208 (0.109)	0.055
Parental monitoring	0.256 (0.167)	0.126	0.237 (0.166)	0.153	0.201 (0.167)	0.230	0.055 (0.139)	0.694
DunedinPACE clock
6-item ACE	0.003 (0.003)	0.312	0.002 (0.003)	0.575	0.001 (0.003)	0.797	0.001 (0.003)	0.606
Threat	−0.001 (0.004)	0.907	−0.002 (0.004)	0.616	−0.004 (0.004)	0.372	−0.004 (0.004)	0.341
Deprivation	0.002 (0.004)	0.660	0.0004 (0.004)	0.935	−0.001 (0.004)	0.850	−0.001 (0.004)	0.830
Parental warmth	0.001 (0.003)	0.661	−0.0001 (0.003)	0.988	−0.001 (0.003)	0.849	−0.001 (0.003)	0.800
Parental monitoring	0.001 (0.004)	0.766	0.001 (0.004)	0.877	−0.001 (0.004)	0.904	−0.00002 (0.004)	0.995

Abbreviations: ACE, adverse childhood experiences; CES-D, Center for Epidemiologic Studies Depression Scale; ELA, early-life adversity; SE, standard error.

^a Questions on ELA were posed in 2018–2019. All models were fitted using linear mixed models.

^b Model 1 included controls for age, sex, race/ethnicity, and marital status, random effects for methylation chip and position, and adjustments for residual sample contamination with nonmonocytes (enrichment scores for neutrophils, B cells, T cells, and natural killer cells).

^c Model 2 added additional controls for parental educational attainment, own educational attainment, and income.

^d Model 3 added additional controls for stressful life events and CES-D score.

^e Model 4 added controls for health behaviors (smoking, alcohol consumption, body mass index, and exposure to secondhand smoke).

^f Statistically significant P value after Bonferroni correction for 6 independent tests at a familywise error rate of 0.05 (P < 0.0083).

The statistical significance of GrimAge model 1 estimates for deprivation and parental warmth declined slightly after adjustment for parental and adult SES (model 2) and again after adjustment for stressful life events and depressive symptoms (model 3); however, the magnitudes of the effect sizes were not statistically different from model 1 estimates (95% confidence intervals overlapped). Conversely, after adjustment for health behaviors (model 4), estimates declined in magnitude and were no longer statistically significant at P < 0.05 (see Web Tables 4 and 5 for full results).

Associations of deprivation and parental warmth with GrimAge appeared to be driven by its DNAm-based composites for pack-years of smoking (DNAm pack-years) and cystatin C (DNAm cystatin C) (Table 4). Notably, the association between parental warmth and DNAm cystatin C remained significant even after controlling for smoking (P = 0.03). In mediation analyses, the proportion of the total effect mediated by lifestyle behaviors was largest for current smoking behavior (49%), but the estimate was not statistically significant (P = 0.06). Other lifestyle behaviors had effect estimates less than 15%, and none were statistically significant (Table 5). These results provide suggestive evidence that smoking behavior in particular may act as a downstream mechanism between parental deprivation in early life and accelerated epigenetic aging in later life.

Table 4.

Associations Between DNAm GrimAge Components and Childhood Deprivation or Parental Warmth (n = 842) in the Multi-Ethnic Study of Atherosclerosis, 2010–2012^a

	Model
	Model 1		Model 2		Model 3		Model 4
GrimAge Component	β (SE)	P Value	β (SE)	P Value	β (SE)	P Value	β (SE)	P Value
Deprivation
GrimAge totalb	0.459 (0.170)	0.007	0.399 (0.169)	0.018	0.370 (0.170)	0.030	0.21 (0.142)	0.147
DNAm ADM	0.542 (0.673)	0.421	0.478 (0.677)	0.480	0.505 (0.685)	0.461	0.74 (0.687)	0.283
DNAm B2M	3,642.430 (3,875.380)	0.348	3,835.940 (3,889.580)	0.324	3,286.280 (3,941.180)	0.405	3,020.670 (4,000.290)	0.450
DNAm cystatin C	1,086.210 (831.965)	0.192	1,074.07 (834.264)	0.198	946.251 (844.750)	0.263	1,069.52 (847.919)	0.208
DNAm GDF-15	4.642 (3.833)	0.226	4.208 (3.832)	0.273	3.537 (3.888)	0.363	1.82 (3.864)	0.638
DNAm leptin	−96.710 (112.551)	0.390	−102.260 (113.195)	0.367	−93.226 (114.949)	0.418	−43.53 (116.267)	0.708
DNAm pack-years	1.280 (0.479)	0.008	1.116 (0.476)	0.019	1.047 (0.481)	0.030	0.26 (0.365)	0.479
DNAm PAI-1	116.423 (96.122)	0.226	81.413 (96.217)	0.398	76.756 (97.490)	0.431	81.74 (93.236)	0.381
DNAm TIMP-1	12.634 (24.959)	0.613	7.928 (24.968)	0.751	7.755 (25.308)	0.759	19.68 (25.112)	0.433
Parental Warmth
GrimAge total	0.374 (0.130)	0.004	0.316 (0.129)	0.015	0.304 (0.130)	0.019	0.208 (0.109)	0.055
DNAm ADM	0.489 (0.515)	0.343	0.409 (0.519)	0.431	0.434 (0.523)	0.406	0.597 (0.525)	0.255
DNAm B2M	3,119.230 (2,965.510)	0.293	3,317.720 (2,985.940)	0.267	3,077.760 (3,007.160)	0.306	2,925.190 (3,053.790)	0.338
DNAm cystatin C	1,297.461 (637.059)	0.042	1,277.719 (641.018)	0.047	1,236.894 (645.166)	0.056	1,392.388 (647.475)	0.032
DNAm GDF-15	3.564 (2.939)	0.226	3.312 (2.949)	0.262	2.985 (2.974)	0.316	2.228 (2.956)	0.451
DNAm leptin	−77.921 (86.473)	0.368	86.030 (87.258)	0.325	−79.559 (88.114)	0.367	−45.501 (89.123)	0.610
DNAm pack-years	0.888 (0.367)	0.016	0.733 (0.365)	0.045	0.705 (0.367)	0.055	0.202 (0.278)	0.468
DNAm PAI-1	136.098 (73.673)	0.065	104.619 (74.018)	0.158	102.739 (74.571)	0.169	104.000 (71.368)	0.146
DNAm TIMP-1	23.878 (19.123)	0.212	19.127 (19.199)	0.319	19.808 (19.345)	0.306	29.496 (19.190)	0.125

Abbreviations: ADM, adrenomedullin; B2M, β₂ microglobulin; DNAm, DNA methylation; GDF-15, growth differentiation factor 15; PAI-1, plasminogen activation inhibitor 1; SE, standard error; TIMP-1, tissue inhibitor metalloproteinase 1.

^a Questions on early-life adversity were posed in 2018–2019. All models were fitted using linear mixed models. See Table 3 footnotes for model details.

^b DNAm GrimAge components include adrenomedullin levels (DNAm ADM), β₂ microglobulin (DNAm B2M), cystatin C (DNAm cystatin C), growth differentiation factor 15 (DNAm GDF-15), leptin (DNAm leptin), pack-years of smoking (DNAm pack-years), plasminogen activation inhibitor 1 (DNAm PAI-1), and tissue inhibitor metalloproteinase 1 (DNAm TIMP-1).

Table 5.

Proportion of the Association Between GrimAge and Childhood Deprivation or Parental Warmth Mediated by Health Behaviors (n = 842) in the Multi-Ethnic Study of Atherosclerosis, 2010–2012^a

Outcome	β (SE)	Proportion   Mediated, %	Z Score	P Value
Deprivation
Total effect	0.365 (0.141)		2.59	0.009
Indirect effectsb
Tobacco smoking
Former smokerc	0.010 (0.032)	2.63	0.30	0.764
Current smoker	0.182 (0.098)	49.78	1.86	0.063
Missing data	0.008 (0.011)	2.28	0.79	0.431
Alcohol consumption
≤2 drinks/dayd	−0.001 (0.005)	−0.26	−0.21	0.838
>2 drinks/day	0.008 (0.009)	2.17	0.89	0.374
Missing data	−0.001 (0.007)	−0.27	−0.14	0.892
Obesity
Overweight (BMIe,f 25.0–29.9)	−0.003 (0.011)	−0.77	−0.26	0.793
Obese (BMI ≥30.0)	−0.050 (0.026)	−13.61	−1.89	0.059
Missing data	−0.004 (0.005)	−1.10	−0.76	0.446
Secondhand smoke exposure
Lives with a smoker	0.023 (0.021)	6.42	1.13	0.260
Missing data	0.008 (0.010)	2.09	0.76	0.456
Parental Warmth
Total effect	0.302 (0.107)		2.81	0.005
Indirect effects
Tobacco smoking
Former smoker	−0.003 (0.024)	1.09	−0.14	0.892
Current smoker	0.119 (0.076)	39.38	1.56	0.118
Missing data	0.007 (0.008)	2.27	0.82	0.412
Alcohol consumption
≤2 drinks/day	−0.002 (0.004)	−0.78	−0.54	0.591
>2 drinks/day	0.007 (0.007)	2.28	0.92	0.359
Missing data	−0.001 (0.005)	−0.24	−0.14	0.886
Obesity
Overweight (BMI 25.0–29.9)	−0.005 (0.014)	−1.58	−0.34	0.735
Obese (BMI ≥30.0)	−0.041 (0.021)	−13.55	−1.93	0.054
Missing data	−0.003 (0.003)	−0.83	−0.76	0.447
Secondhand smoke exposure
Lives with a smoker	0.024 (0.016)	7.96	1.46	0.144
Missing data	0.006 (0.009)	2.00	0.71	0.478

Abbreviations: BMI, body mass index; CES-D, Center for Epidemiologic Studies Depression Scale; SE, standard error.

^a Questions on early-life adversity were posed in 2018–2019. All models adjusted for model 3 covariates (i.e., age, sex, race/ethnicity, marital status, white blood cell proportions, parental education, own education, income, CES-D score, and stressful life events).

^b Indirect effects were calculated using the method of Karlson et al. (71) after simultaneously adjusting for all health behaviors.

^c Omitted category for smoking: never smoker.

^d Omitted category for alcohol drinking: nondrinker.

^e Weight (kg)/height (m)².

^f Omitted category for BMI: underweight or normal-weight (BMI <25.0).

DISCUSSION

A growing body of research examining the relationship between ELA and biological aging is increasing our understanding of how early-life experiences may accelerate aging at a cellular level. In this study, we found suggestive evidence that experiences of deprivation, but not threat, are associated with accelerated DNAm GrimAge. According to a recent study that used a larger sample in MESA to examine associations of SES with EAA (n = 1,211), the magnitudes of these effects are similar to GrimAge associations with current occupational or neighborhood disadvantage at older ages, but are considerably smaller in magnitude than GrimAge associations with education or current household income (50). Although results from mediation analyses were statistically imprecise, the estimated proportion of the total effect mediated by current smoking behavior was large (49%). Moreover, both deprivation and parental warmth were associated with the DNAm pack-years GrimAge component. Together, these results suggest that smoking may be a key factor on the mechanistic pathway between deprivation and accelerated epigenetic aging. However, parental warmth was also significantly associated with higher DNAm cystatin C, even after conditioning on smoking and other lifestyle behaviors, which suggests the existence of other pathways that may be independent of smoking. Specifically, as a biomarker, higher levels of cystatin C have been associated with a multitude of age-related factors, including oxidative stress, kidney dysfunction, cognitive decline, CVD, and mortality (72–74).

Given the co-occurrence of threat and deprivation, in exploratory analyses we evaluated whether significant associations between parental warmth and GrimAge persisted after controlling for levels of monitoring or threat and/or whether parental warmth associations varied by levels of monitoring or threat (Web Table 6). Model 1 effect sizes for parental warmth were unchanged for GrimAge when adjusting for monitoring but larger when adjusting for threat. There were no significant interactions between parental warmth and monitoring or threat for any of the epigenetic aging measures. Further, we looked for sociodemographic-specific results by comparing the magnitudes of GrimAge model 1 effect-size estimates in analyses stratified by age, sex, race/ethnicity, and parental education (Web Table 7 and Web Figure 1) (75). Although we probably lacked sufficient statistical power to assess statistical significance, ELA-GrimAge associations for deprivation and parental warmth were larger in magnitude for White participants and participants whose parents obtained no high school diploma or at most a high school diploma or General Educational Development certificate, as opposed to the null results we observed for participants from higher-SES families.

Our findings are consistent with research on older adults in the Health and Retirement Study which found associations between ACEs and GrimAge EAA (47), but they differed from research on older adults in TILDA which found that childhood poverty, but not adverse experiences related to childhood abuse, was associated with GrimAge EAA and DunedinPoAm (46). In MESA, self-reports of childhood SES are limited to parental education, whereas respondents in TILDA were asked more explicitly to rate their financial circumstances between birth and 14 years of age. Thus, the null relationship between childhood SES and EAA in MESA may reflect differences in how childhood SES was measured. Notably, in TILDA, MESA, and the Health and Retirement Study, smoking behavior was identified as a primary mediator between ELS and later-life EAA. This suggests that earlier mental health interventions for at-risk youth may not only curb substance-use disorders but also prevent more rapid aging later in life. Indeed, mental health is increasingly being recognized as a primary risk factor for accelerated aging (76, 77).

Additionally, our findings differ from research in children and adolescents that found experiences of threat, but not deprivation, were associated with Horvath EAA (associations in second-generation clocks or pace-of-aging measures were not assessed) (4, 5, 42). In part, deprivation-EAA relationships may be more apparent in epigenetic algorithms that were trained on time to mortality or the pace of aging, which have been linked to socioeconomic disadvantage in children at multiple time points (78). However, our study also failed to identify associations between threat and Horvath EAA. Divergent findings in children and older adults may exist for several reasons. First, if experiences of deprivation primarily affect accelerated aging via midlife mental health or substance abuse pathways, deprivation-EAA associations may not be apparent until later in the life course. Second, the impact of threat on EAA may be more acute during childhood or adolescence when the abuse is occurring. Third, because this study relied on retrospective reports of ELA, respondents may have misreported, downplayed, or suppressed reports of childhood abuse due to embarrassment, distress, memory issues, or anchoring of retrospective reports to current life circumstances (79). Additionally, due to strict limits on the number of items that could be added to the ELA questionnaire in MESA, the measures used in this study are not comprehensive indicators of ELA and do not include, for example, sexual abuse or exposures outside the home that may have affected long-term health and well-being. Finally, because MESA is a sample of older adults that were recruited on the basis of their lack of clinically apparent CVD, the null threat-EAA relationships we report may stem from survivor bias or ascertainment bias. In particular, studies have shown that childhood maltreatment and exposure to interpersonal violence are related to accelerated pubertal timing and increased risk of obesity and central adiposity—that is, risk factors for CVD in midlife that would preclude individuals from appearing in our sample because of premature mortality or MESA sample selection criteria (56, 80, 81). Furthermore, in recent research that utilized a larger, population-representative study of older adults in Canada, Mian et al. (82) found that biological aging calculated from blood-chemistry data was associated with neglect and abuse, although effect sizes were considerably larger for experiences of neglect.

Our study had additional limitations. First, we cannot draw conclusions regarding the causal relationship between ELA in childhood and EAA in adulthood, and our modeling strategy may have omitted important confounders. Due to data limitations, we also cannot assess whether the timing or severity of the exposure affected epigenetic aging patterns. Further, DNAm in MESA was profiled in monocytes only, which on the one hand minimizes any potential confounding due to cell type but on the other hand limits our ability to observe epigenetic aging in other types of cells or tissues. Finally, due to sample selection and attrition, our analytical sample was more educated, more White, and less likely to smoke than the original MESA sample, which may have biased our estimates downwards. Thus, our results should be interpreted with caution in terms of their broader generalizability.

Strengths of this study include 1) examination of the ELA-EAA relationship in one of the largest samples of older adults to date, 2) analysis of distinct dimensions of ELA to further elucidate the consequences of ELA for aging, 3) examination of relationships with first- and next-generation epigenetic aging measures, and 4) exploration of potential midlife factors that may affect the ELA-EAA relationship. Overall, further research is needed to validate connections between ELA and EAA across the life course. Because the epigenome is highly sensitive to disease and various environmental exposures that could affect DNAm at the time of blood collection, longitudinal DNAm data coupled with more detailed exposure data would minimize errors in reporting and measurement.