Early life trauma patterns and adult epigenome-wide and NR3C1-specific DNA methylation in the Sister Study

Abstract

Patterns of co-occurring early life trauma (ELT), unlike cumulative trauma, are understudied as potential stress-related health risks. DNA methylation (DNAm) is a posited mechanism linking socioenvironmental stress and disease pathogenesis. We assess whether ELT patterns differentially affect adult DNAm, both epigenome-wide and specifically stress-related genes (eg, glucocorticoid receptor gene, NR3C1). Data represent a case-cohort of non-Hispanic white participants (N = 2566) from the Sister Study, a cohort of US women (ages: 35-74; enrollment: 2003-2009; N = 50 884). Early life trauma measures included a count score and 4 latent ELT classes: low ELT (referent); sexual and emotional; high betrayal; and high ELT. We evaluated epigenome-wide DNAm, differentially methylated regions (DMRs), pathway enrichment, and NR3C1-specific methylation from whole blood. Twenty-two differentially methylated Cytosine–phosphate–Guanine (CpG) sites were associated with ELT classes and none with ELT score. Furthermore, 108 DMRs were associated with ELT score (n = 5) and ELT classes: sexual and emotional (n = 7), high betrayal (n = 37), and high ELT (n = 61). Cardiovascular signaling and leptin signaling pathways were both associated with high betrayal and high ELT classes. In NR3C1-specific analyses, 11 CpGs were associated with ELT score (n = 3) and specific ELT classes (n = 9). Results suggest that specific patterns of co-occurring trauma may contribute to meaningful variability in peripheral blood DNAm in adulthood.

Introduction

Early life trauma (ELT), including interpersonal maltreatment, social and financial hardships, and traumatic events, has been shown to have a lasting impact on health,¹ but the mechanisms through which ELT may affect downstream health remain unclear. DNA methylation (DNAm), a form of epigenetic modification, occurs when a methyl group is added to a cytosine base at sites followed by a guanine base (Cytosine–phosphate–Guanine, CpG). Gene-specific and epigenome-wide patterns of DNAm have been associated with increased risk of poor health. Unlike DNA, DNAm can be modified by external factors, making it a pathway of interest through which the physical and psychosocial environment may influence alterations in gene expression and downstream health outcomes.

Prior studies have assessed the relationship between cumulative trauma (ie, summed number of traumas experienced),² distinct traumatic events (eg, loss of a parent),³ or specific trauma domains⁴ over the life course or during sensitive periods (eg, childhood) and DNAm with varied results due in part to differences in study design, exposure measures, targeted CpG sites, and sample populations. Certain types of trauma tend to co-occur, however, and may differentially affect the extent to which ELT impacts downstream health consequences. Few studies have assessed the association between patterns of co-occurring trauma and epigenome-wide or targeted gene methylation; thus, understanding of the role of specific ELT patterns on later life DNAm remains limited.

DNA methylation of genes involved in stress response via the hypothalamus–pituitary–adrenal (HPA) axis has been studied as a potential pathway linking ELT and disease risk.⁵ One such gene is NR3C1, which controls tissue-wide glucocorticoid receptor expression and subsequently regulates negative feedback of the HPA stress response, thus moderating the impact of socioenvironmental stress on biologic pathways.⁶ Differential NR3C1 methylation (NR3C1m) has been associated with risk for psychopathologies, immune dysfunction, and tumorigenesis⁵; however, findings have been mixed based on the studied health outcome and ELT measure.⁷ Therefore, further assessment of the role of NR3C1m in this pathway, in addition to employing an epigenome-wide approach to identify additional genes of interest, is warranted. Using data from the Sister Study, we investigated the relationship between both count trauma score and co-occurring early life (ie, childhood and adolescence) traumas and epigenome-wide and NR3C1-specific adult leukocyte DNAm.

Methods

Study population

The Sister Study is a prospective study of environmental and genetic risk factors for breast cancer and other health conditions.⁸ Participants are self-identified females residing in the United States, including Puerto Rico (aged 35-74 years; n = 50 884) with a sister diagnosed with breast cancer but who had no personal breast cancer history at enrollment (2003-2009).⁸ Participants completed a baseline computer-assisted telephone interview, and blood samples were collected at a home visit. Health updates are completed annually, and detailed follow-up questionnaires were completed at 2 years and every 3 years thereafter. All participants provided written informed consent. The study was approved by the Institutional Review Board of the National Institutes of Health.

Participants in this analysis were previously sampled from The Sister Study cohort using a case-cohort design to assess the association between DNAm and incident breast cancer (February 2014).^9,10 The subsample consisted of 1336 non-Hispanic white participants who were randomly sampled from the whole cohort (including 75 participants who were later diagnosed with breast cancer) and an additional 1539 non-Hispanic white participants who developed breast cancer between blood sample collection and time of sampling.¹¹ We excluded 102 participants whose blood sample did not meet methylation-related quality control thresholds including >5% low-quality methylation values, low average bisulfite intensity, or implausible distributions of methylation values^9,10 and an additional 204 who were missing all ELT data (n = 194) and/or data on key covariates (missing was <1% for each covariate), leaving 2566 participants for this analysis. Data are from Sister Study Data release 10.1.

Trauma assessment

Data on prior trauma were collected during the first detailed follow-up interview (approximately 2 years after baseline blood sample collection). Questions were adapted from the 14-item Brief Betrayal Trauma Survey¹² and included 26 trauma items (eg, natural disasters, death of a parent, and physical abuse). The Brief Betrayal Trauma Survey is a self-report measure that was designed to capture interpersonal and non-interpersonal traumatic events over the life course, as well as to assess the degree of betrayal attributed to interpersonal events (ie, whether interpersonal events of mistreatment were perpetrated by someone close or not close to the reporter). Construct validity is high, and initial test–retest reliability was approximately 82% for childhood measures (before age 18) among women who completed the survey 3 years apart.^12,13 We retained 20 items pertinent to early life—which are listed in Appendix S1, excluding items related to a participant’s spouse or child. Participants were asked whether they experienced each trauma type (yes, no, don’t know, refused to answer) and whether the trauma occurred before age 13 (childhood) and ages 13-17 (adolescence), as well as during adulthood (age 18-12 months prior to completing the questionnaire). We do not account for adult trauma in analytic models as we are unable to determine whether the exposure occurred after blood draw collection.

Early life trauma was assessed as (1) the number of different trauma types experienced during childhood and/or adolescence (ELT score; individual trauma types that were missing or reported as don’t know or refused to answer were excluded from the score) since counts of individual traumatic incidents were not collected and (2) patterns of co-occurring trauma types (ELT classes)—identified using latent class analysis (LCA; Mplus version 8.6, Muthén & Muthén). Among the analytic sample, we estimated LCA models with 2-5 latent classes using the 20 dichotomous ELT types as indicators with no a priori assumptions regarding the number of latent classes required to best capture the co-occurrence of ELT (Table S1). Model fit was evaluated using adjusted Bayesian information criterion (aBIC) and Lo–Mendell–Rubin adjusted likelihood ratio test (P < .05), and the LCA solution was selected based on entropy—the likelihood of participants being correctly placed in a given class (ideally > 0.8)—and substantive groupings of traumatic experiences.¹⁴ A 4-class LCA model was selected based on model fit while maintaining a moderate level of certainty that participants were correctly grouped (entropy = 0.76). We identified 4 latent classes based on patterns of conditional response probabilities: (1) low ELT (mean response probability across all trauma types = 1.9%, range 0.0%-5.7%; n = 1924), (2) sexual and emotional trauma (n = 531), (3) high betrayal trauma (ie, the perpetrator was someone close to the participant; n = 70), and (4) high ELT (mean response probability across all trauma types = 35%, range 3.3%-92%; n = 41) (Figure S1). Class memberships were exported as a 4-category nominal variable for subsequent analyses.

DNA methylation

Genomic DNA was extracted from whole blood samples using an automated system (Autopure LS, Gentra Systems) housed at the Molecular Genetics Core Facility at the National Institute of Environmental Health Sciences or using DNAQuick at BioServe Biotechnologies Ltd.⁸ DNA was analyzed using the Illumina Infinium HumanMethylation450 BeadChip following the manufacturer’s protocol. Preprocessing and quality control of methylation data were conducted using the ENmix R package,¹⁵ which resulted in 423 500 CpGs that passed quality control protocols, including 40 CpGs within NR3C1. Full details of DNAm data preprocessing have been described by Xu et al.⁹

DNAm levels were estimated as β values (the ratio of methylated probes to the sum of methylated and unmethylated probes at a given locus), which can range from 0 (no copies methylated) to 1 (all copies methylated).¹⁶ To avoid issues of heteroskedasticity and non-uniform distribution of P-values when using raw methylation β values, we use logit-transformed β values, or M values, in analyses, with positive values indicating a greater number of methylated copies than unmethylated copies and negative values representing the inverse.¹⁶ In NR3C1m analyses, we used mean M values to assess differential methylation overall (ie, across all 40 NR3C1 CpGs) and by subsets of CpGs based on genomic function (ie, promoter-associated [n = 21] or within the gene body [n = 7]).

Covariates

Covariates adjusted for in analyses included age at enrollment blood draw (continuous) and measures of early life socioeconomic position (SEP): highest level of educational attainment in the household at age 13 (less than high school, high school graduate, some college, and college degree or higher), family income while growing up (poor, low income, middle income, and well off), and whether there were times as a child when the family did not have enough food to eat (yes, no). We also adjusted for proportions of white blood cell types (CD8 T cells, CD4 T cells, natural killer cells, B cells, monocytes, or granulocytes), which were estimated using the Houseman method.¹⁷ Although potential mediators of the ELT-DNAm pathway, we included body mass index (BMI; continuous; kg/m²) and smoking history–number of pack years smoked (continuous, including never smokers) as covariates to facilitate comparability with existing DNAm literature.^18,19

Statistical analysis

We used sampling weights based on the case-cohort sampling protocol (calculated weights: cases = 1.31; non-cases: 31.81).¹¹ For epigenome-wide association study (EWAS) analyses, we used survey-weighted multivariable linear regression to assess the association between ELT (ELT score and ELT classes) and DNAm at each CpG site using survglm in R, version 4.2.1 (R Foundation for Statistical Computing). Differentially methylated regions (DMRs), or larger areas of DNAm across the genome, were identified from EWAS results using combp in R package Enmix.²⁰ For ELT classes with at least 20 DMRs, we conducted ingenuity pathway analysis to predict downstream biologic mechanisms potentially influenced by genes linked to identified DMRs. Genes were assigned based on the locations of the majority of the DMR’s CpGs according to the UCSC Genome Browser.²¹ We then used the random permutation method (n = 10 000) to calculate enrichment P-values. To correct for multiple testing, a false discovery rate (FDR) of 0.05 using the q-value method²² was applied for individual CpG sites and Šidák-corrected P-values for DMRs.²³ For NR3C1-specific analyses, we used weighted multivariable robust linear regression to evaluate the association between ELT and NR3C1m (overall and by genomic function) using PROC ROBUSTREG in SAS, version 9.4 (SAS Institute Inc.). Models were adjusted for age and cell type proportions, with fully adjusted models additionally controlling for measures of early life SEP, smoking history, and BMI. We report results for both the minimally and fully adjusted models since controlling for smoking history and BMI potential mediators may remove part of the total estimated effect. In analyses with ELT score, estimates represent the change in mean DNAm associated with each additional trauma type. In analyses with ELT classes, estimates are interpreted as the mean difference in DNAm between the specific ELT class and the low ELT class (referent).

Results

Sample characteristics

Characteristics of Sister Study participants included in the analytic sample are shown in Table 1. Approximately 46% reported at least 1 ELT, which is slightly lower than the rate of ELT in the full Sister Study cohort (50%).²⁴ The median number of ELT types reported was 0 (range: 0-15). Participants who reported at least one ELT more often reported lower parental educational attainment, growing up in a poor or low-income household, and periods of food scarcity as a child compared to those who reported no ELT.

Table 1.

Characteristics of non-Hispanic White Sister Study participants included in DNA methylation studies based on experience of early life traumas before age 18 (N = 2566).

	No early life trauma (n = 1381)		Any early life trauma (n = 1185)
	n	% a	n	% a
Breast cancer status at time of sampling
No breast cancer	574	95.7	540	96.3
Invasive or in situ breast cancer	807	4.3	645	3.7
Age at baseline, mean (SD)	56.6	34.2	55.1	32.9
Highest level of education in the household at age 13, %
Less than high school	193	14.9	200	16.0
High school graduate	509	35.5	421	37.8
Some college	276	20.0	246	19.4
College degree or higher	403	29.6	318	26.9
Family income while growing up, %
Poor	52	3.5	112	10.1
Low income	321	23.3	348	30.7
Middle income	924	64.8	655	53.5
Well off	84	8.4	70	5.8
As a child, were there times when your family did not have enough food to eat, %
Yes	60	4.0	137	12.5
No	1321	96.0	1048	87.5
Body mass index
Underweight	16	1.1	10	0.6
Normal weight	532	39.5	454	40.6
Overweight	448	32.3	389	31.2
Obese	385	27.2	332	27.6
Smoking status
Never smoked	759	57.0	587	48.9
Past smoker	532	36.3	513	43.5
Current smoker	90	6.8	85	7.6

Missing values not included in the table: methylation quality control (n = 102); early life trauma data (n = 202); household education at age 14 (n = 26); family income while growing up (n = 3); family did not have enough food to eat (n = 1); body mass index (n = 1); smoking history (n = 10).

^aWeighted percentages.

Epigenome-wide association study results

ELT score was not associated with DNAm of any CpGs after FDR correction in minimally and fully adjusted models. Conversely, in minimally adjusted models using the low ELT class as the comparator, we identified 432 unique CpGs with differential DNAm associated with ELT classes (sexual and emotional trauma, n = 1; high betrayal trauma, n = 20; high ELT, n = 411; Table S2). After additional model adjustment for childhood SEP and adult health behaviors, 22 CpGs remained differentially methylated (sexual and emotional trauma, n = 1; high betrayal trauma, n = 7; high ELT, n = 14; Table 2). Compared to the low ELT class, 4 of 7 CpGs (57%) for high betrayal trauma and 9 of 14 CpGs (64%) for the high ELT class had decreased DNAm. Conversely, membership in the sexual and emotional trauma class was only associated with greater methylation at CpG site cg08695113, located in P4HA1—a collagen synthesis gene (estimate = 0.075, FDR = 0.03). Of the CpGs with the smallest FDR values associated with the high betrayal trauma class compared to the low ELT class, 2 had lower methylation and were not associated with known genes (cg01552748 estimate = −0.237, FDR = 0.01 and cg02802309 estimate = −0.086, FDR = 0.01), whereas cg15655138 (PIM1, a proto-oncogene) had higher methylation (estimate = 0.256, FDR = 0.01). Finally, the 3 CpGs with the lowest FDR scores associated with the high ELT class had either higher methylation (cg00834536 [ADCY9] estimate = 0.266, FDR = 0.003) or lower methylation (cg00454734 [MYL7] estimate = −0.203, FDR = 0.01 and cg00719767 [ZNF680] estimate = −0.154, FDR = 0.02) in fully adjusted models.

Table 2.

CpGs surviving false discovery rate (FDR) correction in epigenome-wide analysis by latent class of early life trauma compared to low early life trauma among Sister Study participants.

Latent class of early life trauma	ProbeID	Gene	Position	Model estimate a	P	FDR
Sexual and emotional trauma	cg08695113	P4HA1	chr10: 74842425	0.075	6.50E-08	0.027
High betrayal trauma	cg01552748		chr13: 112225526	−0.237	2.01E-08	0.008
	cg02802309		chr12: 49628458	−0.086	1.64E-07	0.014
	cg15655138	PIM1	chr6: 37139981	0.256	1.42E-07	0.014
	cg02588283		chr7: 93954346	0.151	2.80E-07	0.019
	cg01275006	GGN	chr19: 38876250	0.099	7.19E-07	0.033
	cg05348776	CLMN	chr14: 95693784	−0.186	1.24E-06	0.043
	cg02276665	CTNNA1	chr5: 138088442	0.230	8.89E-01	0.043
High early life trauma	cg00834536	ADCY9	chr16: 4013537	0.266	1.51E-07	0.003
	cg00454734	MYL7	chr7: 44183410	−0.203	1.14E-06	0.012
	cg00719767	ZNF680	chr7: 64027686	−0.154	2.30E-06	0.018
	cg00653179		chr1: 240157611	−0.165	3.82E-06	0.023
	cg00562151	SEC14L2	chr22: 30792888	0.208	1.02E-05	0.040
	cg00684125		chr7: 21209165	0.296	1.24E-05	0.045
	cg03403719	NXT1	chr20: 23335483	−0.184	1.27E-05	0.045
	cg05718015	HIVEP3	chr1: 42237913	−0.178	1.27E-05	0.045
	cg15880016	ARHGAP42	chr11: 100661648	−0.359	1.26E-05	0.045
	cg02335141	CCR6	chr6: 167551083	−0.227	1.28E-05	0.045
	cg25219329		chr22: 42679444	0.252	1.38E-05	0.046
	cg26795346	MMP2	chr16: 55516959	−0.135	1.36E-05	0.046
	cg03110165	GPC6	chr13: 93884312	−0.283	1.37E-05	0.046
	cg00555456	ACN9	chr7: 96745696	0.198	1.34E-05	0.046

^aAdjusted for age, childhood socioeconomic position, smoking history, body mass index, and cell proportions.

We identified 108 DMRs associated with ELT (ELT score: n = 5; ELT classes: sexual and emotional trauma, n = 7; high betrayal trauma, n = 37; high ELT, n = 61) in fully adjusted models (Šidák correction < 0.05; Figure 1). Details for each DMR, including location, gene, gene function (if known), and direction of methylation at individual CpGs are available in Table S3. Three DMRs contained differentially methylated CpGs identified during analysis of individual CpG sites and were associated with sexual and emotional trauma (chr10:74842425-74842426, cg08695113; P4HA1), high betrayal trauma (chr5:138088442-138088663; cg02276665; CTNNA1), and high ELT class (chr16:4013337-4013660; cg00834536; ADCY9) compared to the low ELT class. We also identified 2 DMRs that were associated with ELT score and either the sexual and emotional trauma class (chr7:27143235-27143586; HOXA2) or the high ELT class (chr16:633722-634010; PIGQ).

Figure 1.

Differentially methylated regions associated with early life trauma among Sister Study participants (N = 2566). Created in BioRender. Woo, J. (2025) https://BioRender.com/eax20vk.

Pathway enrichment analyses were conducted for genes attributed to DMRs associated with the latent classes for high betrayal trauma and high ELT. We identified 18 pathways enriched at P < .05; 8 associated with high betrayal trauma and 10 associated with the high ELT class, including disease-specific pathways, cardiovascular signaling, and intracellular and second messenger signaling (Table 3). Two overlapping pathways were associated with both high betrayal trauma and high ELT classes: (1) cellular effects of sildenafil, a cardiovascular signaling pathway, and (2) leptin signaling in obesity.

Table 3.

Pathway analysis of differentially methylated regions associated with latent classes of early life trauma among sister study participants (N = 2566).

Ingenuity canonical pathway	Size	P sim	Genes	Pathway category
High betrayal trauma
Cellular effects of sildenafil (Viagra)	84	.04	MYH4; ADCY4	Cardiovascular signaling, disease-specific pathways
Gi signaling	98	.04	STAT3; ADCY4	Intracellular and second messenger signaling
Hepatic fibrosis/hepatic stellate cell activation	122	.03	LEP; FGF2	Disease-specific pathways, ingenuity toxicity list pathways
Leptin signaling in obesity	62	.01	STAT3; ADCY4	Disease-specific pathways
Osteoarthritis pathway	135	.04	CTNNA1; FGF2	Cellular stress and injury, disease-specific pathways
Renin-angiotensin signaling	85	.04	STAT3; ADCY4	Cardiovascular signaling, growth factor signaling
Serotonin receptor signaling	35	<.01	HTR2A; ADCY4	Neurotransmitters and other nervous system signaling
Tight junction signaling	127	.03	MYH4; CTNNA1	Cell cycle regulation, apoptosis
High early life trauma
Atherosclerosis signaling	92	<.01	APOB; ALOX12	Cardiovascular signaling, disease-specific pathways
Caveolar-mediated endocytosis signaling	48	.01	ITGB2; EGFR	Cellular immune response, organismal growth and development, pathogen-influenced signaling
Cellular effects of sildenafil (Viagra)	84	.04	ADCY9; PLCD3	Cardiovascular signaling, disease-specific pathways
GPCR-mediated integration of enteroendocrine signaling   exemplified by an L cell	57	<.01	ADCY9; CALCA; PLCD3	Intracellular and second messenger signaling, neurotransmitters and other nervous system signaling
GPCR-mediated nutrient sensing in enteroendocrine cells	76	.04	ADCY9; PLCD3	Intracellular and second messenger signaling, neurotransmitters and other nervous system signaling
Leptin signaling in obesity	62	.03	ADCY9; PLCD3	Disease-specific pathways
Reelin signaling in neurons	62	.03	ITGB2; ARHGEF3	Neurotransmitters and other nervous system signaling, organismal growth and development
Sphingosine-1-phosphate signaling	109	.04	ADCY9; PLCD3	Intracellular and second messenger signaling
STAT3 pathway	62	.03	DDR1; EGFR	Transcriptional regulation, cellular growth, proliferation and development
Thrombin signaling	183	.02	ADCY9; EGFR; PLCD3; ARHGEF3	Cardiovascular signaling

Candidate gene results: NR3C1-specific methylation

In our a priori candidate gene analysis of NR3C1, we identified 11 differentially methylated CpGs associated with ELT in fully adjusted models (ELT score, n = 3; ELT classes compared to low ELT class: sexual and emotional trauma, n = 2; high betrayal trauma, n = 2; high ELT, n = 5; P < .05; Figure S2). Lower DNAm of CpG cg08845721 was associated with both the high ELT class (compared to low ELT class) and ELT score. Mean DNAm across NR3C1 CpGs was 35%, with approximately 10% methylation of promoter-associated CpGs and 94% methylation of CpGs in the gene body (Table 4). Early life trauma score was not associated with changes in NR3C1m overall, across promoter-associated CpGs, nor across CpGs in the gene body (Table 4). Conversely, using ELT classes, high betrayal trauma was associated with slightly higher NR3C1m in fully adjusted models (estimate = 0.016; 95% CI: 0.002, 0.03), whereas sexual and emotional trauma was associated with slightly lower NR3C1m of promoter-associated CpGs compared to the low ELT class (estimate = −0.008; 95% CI: −0.02, −0.001; Table 4).

Table 4.

Weighted association between early life trauma and DNA methylation of NR3C1 CpG sites as measured by difference in m-value across relevant CpG sites; overall and by CpG location, Sister Study, United States, 2004-2015 (N = 2566).

		Overall		Promoter-associated		Gene body
	n	Model 1 β (95% CI) a	Model 2 β (95% CI) b	Model 1 β (95% CI) a	Model 2 β (95% CI) b	Model 1 β (95% CI) a	Model 2 β (95% CI) b
% Methylation, mean (SD)			34.8 (0.1)		10.1 (0.02)		93.5 (0.02)
Early life trauma score	2566	0.000 (−0.002, 0.002)	0.001 (−0.001, 0.002)	0.003 (−0.002, 0.002)	0.001 (−0.001, 0.003)	−0.003 (−0.006, 0.000)	−0.003 (−0.006, 0.001)
Early life trauma Latent Classes
Low early life trauma	1924	Reference	Reference	Reference	Reference	Reference	Reference
Sexual and emotional trauma	531	−0.006 (−0.012, 0.000)	−0.005 (−0.011, 0.002)	−0.009 (−0.017, −0.002)	−0.008 (−0.015, 0.000)	−0.011 (−0.023, 0.001)	−0.010 (−0.022, 0.002)
High betrayal trauma	70	0.014 (−0.001, 0.028)	0.015 (0.001, 0.030)	0.013 (−0.005, 0.030)	0.013 (−0.005, 0.031)	0.006 (−0.023, 0.034)	0.007 (−0.021, 0.036)
High early life trauma	41	−0.010 (−0.027, 0.007)	−0.006 (−0.023, 0.011)	0.000 (−0.021, 0.020)	0.005 (−0.016, 0.026)	−0.023 (−0.056, 0.010)	−0.021 (−0.055, 0.012)

^aAdjusted for age and cell proportions.

^bAdjusted for age, childhood SEP, smoking history, body mass index, and cell proportions.

Discussion

This study is one of the largest to date to assess the relationship between ELT and adulthood leukocyte DNAm using EWAS and candidate gene (NR3C1) approaches and the first to assess these relationships using ELT latent classes. Previous studies have relied on trauma scores, individual trauma types, or singular events to assess the ELT and adult DNAm association. Because specific trauma types may co-occur, we used LCA to capture ELT patterns and present these results along with those using a traditional trauma score. Compared to the low ELT class, ELT classes were associated with differential DNAm at 22 unique CpGs and 108 DMRs. Of interest, we observed overlap of 2 DMRs identified using ELT score and ELT classes, with no overlap in DMRs from different ELT classes. We further identified 2 enriched pathways related to cardiovascular signaling and leptin signaling in obesity associated with the ELT classes for high betrayal trauma and high ELT. In a candidate gene approach, 11 CpGs were associated with ELT measures, with only 1 overlapping between ELT score and the high ELT class, although none reached genome-wide significance. These findings highlight the importance of considering the nuanced effects of trauma profiles when assessing the impact of ELT on adult DNAm.

In this study, we include results for a trauma count score, of which similar scores are commonly used in studies of early life trauma, as well as a 4 latent class model of early life trauma that we classified as low early life trauma, sexual and emotional trauma, high betrayal trauma, and high overall trauma. While these latent classes capture potentially unique patterns of experienced early life trauma in the Sister Study cohort, there is some overlap with trauma domains that are commonly utilized, including sexual abuse and emotional trauma, which are usually assessed as separate exposures, but may not be easily disentangled. Finally, our overall low and high trauma groups may approximate dichotomous cutoffs that are derived from continuous scores, such as the adverse childhood experiences measure,²⁵ to delineate high and low experiences of trauma. Despite the potential overlap in these categories, our findings were mostly novel from previous EWAS studies of ELT.

While the ELT score was not associated with methylation of individual CpGs in EWAS analysis, similar to the results from 2 UK cohort studies² that identified unique DMRs associated with different adverse childhood experiences, we observed differential DNAm associated with specific ELT classes at individual CpGs and DMRs. These findings suggest that patterns of experienced trauma may be more relevant for DNAm change than incremental change in the sum of traumatic experiences. Furthermore, there was minimal overlap in differentially methylated CpGs and DMRs across ELT classes, which may indicate that biological pathways may be differentially affected by types and patterns of ELT. Variations in the estimated association different types or patterns of trauma may be due to differences in chronicity, duration, and perceived severity, as well as the possibility of distinct biological mechanisms, which have yet to be elucidated.

Despite the high level of heterogeneity of results across ELT classes, we identified a DMR in C15orf26 (ch15; unknown function) containing lower methylated CpGs associated with the high betrayal trauma class that overlaps with a DMR replicated in 2 UK cohorts, albeit with higher methylation, associated with childhood experience of having a parent with physical or mental illness.² The remaining DMRs did not overlap with genes previously associated with ELT in other studies. In pathway analyses, both high betrayal trauma and high ELT classes were associated with chronic disease-specific pathways related to cardiovascular signaling and leptin signaling in obesity. Early life trauma has been associated with both cardiovascular disease and obesity in adulthood,^26,27 which support DNAm as a potential link between ELT and these chronic diseases.

We included a candidate gene approach to study NR3C1m to compare our findings with previous literature in which NR3C1 has been linked to post-traumatic stress disorder and other psychological disorders. Prior studies of the ELT and NR3C1m association have been inconsistent, with studies reporting increased NR3C1m, decreased methylation, or no association.^3,4 This variation and lack of replication are likely attributable to prior studies focusing on CpGs within the NR3C1 1F promoter region, utilizing targeted pyrosequencing techniques versus EWAS, and evaluating the ELT-NR3C1m association in targeted patient groups (eg, chronic fatigue syndrome and psychiatric disorders). Similar to our results, differential NR3C1m is rarely detected in EWAS studies of ELT despite noticeable findings using candidate gene approaches, which suggests that NR3C1m may be masked by stronger DNAm signals in epigenome-wide studies.²⁸ These findings support the shift in the field from candidate gene methylation approaches to more unbiased EWAS approaches to study differential psychological stress–related DNAm.²⁹

The variability between our NR3C1-specific and EWAS results and those of previous studies may be attributable to heterogeneity in ELT measures, age at which DNAm was assessed, inconsistent use of multiple testing corrections, inappropriate adjustment of potential mediators of the ELT-DNAm pathway, and/or disparate study populations.³⁰ A recent review by Parade et al. on the association between childhood maltreatment and DNAm highlights the low comparability between study methods and the lack of longitudinal DNAm data as key factors contributing to varied results.³⁰ While we lack longitudinal DNAm measures, we contribute to the literature by supporting comparability across studies by including both ELT score results and LCA models of experienced ELT patterns. This work also has potential implications for clinical practice and trauma-informed care approaches,³¹ emphasizing the need to consider life course exposures to trauma, patterns of trauma that may have a greater impact on downstream health, and epigenetic risk factors that may mediate the association between experienced trauma and negative health outcomes.

A primary limitation of our study is the report and measurement of ELT. Although ELT occurred prior to measurement of adult DNAm, ELT data were collected approximately 2 years after baseline blood collection. However, studies have demonstrated stable recall for report of ELT and adversity,³² supporting our ELT measure. In addition, we exported ELT LCA results for analysis, which does not account for error related to probability-based class assignment.³³ Furthermore, although we collect a wide range of trauma types, we are unable to distinguish frequency, chronicity, and severity of each trauma type, which may differentially affect epigenetic pathways. While current or persistent trauma may also contribute to patterns of DNAm in adulthood,³⁴ we were unable to isolate trauma specifically experienced around the time of sample collection, as the Stress and Coping Questionnaire was administered approximately 2 years later and the period of adult trauma was captured from age 18 to 1 year prior to survey completion. Therefore, we do not adjust for trauma experienced in adulthood. In addition, due to the limited sample size and high overlap between trauma experienced during childhood and adolescence, models that attempted to isolate the effects of trauma experienced during childhood or adolescence were underpowered. Longitudinal studies could help elucidate the association between experienced trauma, potential critical periods of exposure, and temporal changes in DNAm. Similarly, while we include age as a potential confounder of the ELT and DNAm association, future larger studies may also consider the impact of age as an effect measure modifier of this relationship. Variation in age (or birth cohort) may impact the report of ELT, as well as affect the accumulation of additional psychosocial stressors and traumatic events over the life course and the development of resilience mechanisms and other stress-coping mechanisms (ie, smoking) that may impact DNAm.

Finally, our study takes advantage of an existing DNAm dataset that included samples processed using the Illumina Infinium HumanMethylation450 BeadChip. While newer chips, including the EPIC array, cover a greater number of CpG sites, we were limited to available data on this sample of non-Hispanic white women with a sister with breast cancer sampled from the larger Sister Study cohort. The Sister Study cohort is also generally older, of higher educational attainment, and more likely to be healthy and economically advantaged than the general US population, which may underestimate ELT, compared to that experienced among more vulnerable populations. Although our results remain internally valid,^8,35 these differences may limit generalizability to more diverse populations. Future epigenetic work in more diverse samples of Sister Study participants that span a greater number of CpG sites will help to enhance the current findings.

In this large study of ELT and adult DNAm, we use LCA to model co-occurring traumas and observed that ELT classes were more strongly associated with differential DNAm from peripheral blood cells compared to a trauma count score. Our results suggest that patterns of co-occurring traumas may be more relevant in shaping later life DNAm than experiencing a greater number of trauma events overall.

Funding

This work was funded by the Intramural Program of the NIH, National Institute of Environmental Health Sciences (Z01 ES044005 [D.P.S.]).

Conflict of interest

The authors declare no conflicts of interest.

Data availability

Requests for Sister Study data can be submitted by following the directions at http://sisterstudy.niehs.nih.gov/English/coll-data.htm.