Cumulative psychosocial factors and epigenetic age acceleration in the Hispanic Community Health Study/Study of Latinos

ABSTRACT

Background

Hispanics/Latinos in the United States experience disproportionately high psychosocial factors compared to non-Hispanic/Latino Whites. Psychosocial factors may accelerate biological aging, measured by epigenetic age acceleration (EAA), a DNA methylation biomarker predictive of morbidity and mortality.

Methods

We investigated the cumulative impact of psychosocial factors on EAA over time in 922 adults from the Hispanic Community Health Study/Study of Latinos (HCHS/SOL). Psychosocial exposure profiles were derived using self-organizing maps (SOM), an unsupervised clustering method. We calculated EAA from whole blood DNA methylation at two timepoints using GrimAge and DunedinPACE.

Results

SOM identified four clusters: Cluster 1 (n = 196; 21.3%) had high levels of all psychosocial factors; Cluster 2 (n = 250; 27.1%) exhibited chronic, traumatic, and childhood stress; Cluster 3 (n = 250; 27.1%) showed mental health symptoms, low social support, and high perceived stress; and Cluster 4 (n = 238; 24.5%) had relatively low psychosocial stress. Adjusted weighted linear mixed models exhibited increased GrimAge in Cluster 1 (1.27 years, 95% CI: 0.57,1.97) and Cluster 2 (0.62 years, 95% CI: 0.01,1.23) compared to Cluster 4. DunedinPACE increased 3% (95% CI: 0.01,0.05) and 2% (95% CI: 0.001,0.04) in Clusters 1 and 3, respectively.

Conclusions

These findings highlight the cumulative impact of psychosocial factors on EAA and how stressors can get “under the skin” and contribute to health disparities.

Plain Language Summary

Psychosocial stressors, such as childhood trauma, discrimination, stressful life experiences, and chronic stress, can affect how quickly the body ages. In this study, we looked at whether being experiencing different types of stress at the same time is linked to faster biological aging. We used data from Hispanic/Latino adults in a large US study, the Hispanic Community Health Study/Study of Latinos. We grouped people based on how much and what kinds of stress they had experienced. Then, we used blood sample to measure biological aging using markers in the DNA. We found that people with more long-term and traumatic stress showed signs of faster aging compared to those with fewer stressors. These findings suggest that stress can “get under the skin” and speed up the body’s aging process. Understanding how psychosocial stress impacts health over time can help us identify groups of people most at risk for poor health outcomes. It can also help us develop programs to improve health and reduce disparities in Hispanic/Latino communities.

1. Introduction

Hispanics/Latinos in the United States experience higher rates of psychosocial and socioeconomic stressors than their non-Hispanic/Latino White counterparts. It is well established that psychosocial factors such as traumatic stress, chronic stress, and depression symptoms, are risk factors for adverse health outcomes including cardiovascular disease, metabolic syndrome, and obesity in Hispanic/Latino populations [1–3]. One potential biological mechanism underlying the association between psychosocial factors and health outcomes is DNA methylation and particularly epigenetic aging. DNA methylation is an epigenetic modification that regulates gene expression by the addition of a methyl group at a cytosine-phosphate-guanine (CpG) site. Epigenetic modifications are often dynamic and highly sensitive to external environmental factors. Epigenetic age is a biomarker of biological aging based on a composite of numerous CpG sites scattered throughout the genome that are predictive of morbidity and mortality phenotypes. It is hypothesized that exposures such as psychosocial factors and low socioeconomic status are associated with accelerated epigenetic aging in which a person’s predicted biological age exceeds their chronological age [4–6]. A few previous studies have discovered evidence of a relationship between psychosocial factors and accelerated epigenetic age across the life course [7–11].

While there is a growing body of evidence linking psychosocial factors and epigenetic aging, the vast majority of studies have been conducted in populations of primarily European ancestry. While minoritized populations disproportionately carry the burden of stressors, studies in underrepresented populations are sparse, and associations among racially and ethnically diverse populations are still largely unknown [12]. Additionally, many studies are limited by their cross-sectional design, likely due to the cost and feasibility of acquiring multiple DNAm samples from participants. As DNAm assays became more widely available, some recent studies have had the opportunity to investigate longitudinal changes in epigenetic age in relation to exposures such as adverse childhood experiences (ACEs) and neighborhood deprivation, but more evidence is needed to determine if and how epigenetic age acceleration can change across the life course in response to cumulative psychosocial factors [13,14]. Finally, most studies only look at one exposure at a time. However, exposure to multiple stressors (e.g., acculturative stress, adverse childhood experiences, discrimination, mental health, etc.) typically co-occur and it is important to assess the cumulative and joint effects of several exposures to gain a more holistic understanding and improve generalizability of results. Stressors are often highly correlated, and many statistical methods cannot handle multicollinearity. Additionally, it is computationally inefficient to assess k-order interactions in a standard generalized linear model. Exposure mixture methods that are commonly used in the field of environmental epidemiology to assess multiple environmental exposures are designed to handle these common issues, and the use of these methods in social epidemiology is novel but promising [15].

In this study, we investigate whether exposure to multiple psychosocial factors is associated with a change in epigenetic age acceleration (EAA) across two time points among 922 adults in the Hispanic Community Health Study/Study of Latinos (HCHS/SOL). We utilized self-organizing maps (SOM), a mixture clustering method, to categorize participant exposure profiles and assess the joint effects of psychosocial factors on EAA across four domains: acculturation and discrimination, mental health, stress, and social support. Finally, we investigate whether sex modifies the association between psychosocial factors and EAA.

2. Methods

2.1. Study population

The Hispanic Community Health Study/Study of Latinos (HCHS/SOL) is a multi-center, population-based, prospective cohort study of 16,145 Hispanic/Latino adults aged 18–74 in the United States. Participants resided in one of the four study sites (the Bronx, NY; Chicago, IL; Miami, FL; and San Diego, CA) and were recruited between 2008 and 2011. The Sociocultural Ancillary Study (SCAS) is a subset of the HCHS/SOL (N = 5313) with the goal of further exploring psychosocial, sociocultural, and socioeconomic factors, conducted approximately 1 year after the baseline assessment. Information on the study design and sampling for HCHS/SOL and SCAS are described elsewhere [16,17]. A subset of participants that were part of the SCAS study and had consented to blood sampling and DNA analysis at both time points approximately 6 years apart were randomly sampled (N = 983). Methylation data was obtained from these participants for analysis. Our final analytic sample as described below was 922 adults aged 18–73.

Written consent for participation was obtained from all participants, and ethical approval for human subjects research was given by the Institutional Review Boards at the University of North Carolina at Chapel Hill, Albert Einstein College of Medicine, Feinberg School of Medicine at Northwestern University, Miller School of Medicine at the University of Miami, and San Diego State University. The current study was approved by the Institutional Review Board at Emory University.

2.2. DNA methylation measurements

2.2.1. Probe filtering and normalization

DNA methylation was measured from whole blood at two time points approximately 6 years apart using the well-established Illumina Infinium HumanMethylationEPIC BeadChip (850K; Illumina, Inc., San Diego, CA, USA). To minimize batch effects, repeated measurements from the same individual were run on the same chip. Additionally, principal component analysis further confirmed no clustering of individuals by sample plate. Preprocessing and normalization were performed on raw IDAT files using the minfi package [18] in the R environment (version 4.2.1). We removed any samples with a non-significant mean probe intensity detection p-value (p > 0.01), which resulted in the removal of one failed sample; the matched timepoint for this sample was also removed after our final normalization step prior to downstream analyses. Background correction and dye-bias normalization were then performed utilizing the within-array normalization single sample noob (normal-exponential out-of-band) method, which is suitable for larger datasets that may be analyzed in multiple batches while allowing for dye-bias normalization without the need for a reference sample. Both batches of normalized datasets were then combined, resulting in 2,049 total samples with 866,238 CpGs. Normalized samples were then filtered for unreliable CpG probe signals in one or more samples (probe intensity detection p < 0.01; removed 41,541 CpGs), probes on sex chromosomes (removed 17,605 CpGs), probes whose sequence covered known SNPs (removed 2,488 CpGs), and cross-reactive probes (removed 40,074), which resulted in 2,049 samples with reliable data across 764,530 CpGs. To correct for probe design type biases, the Regression on Correlated Probes (RCP) method in the Enmix package was used [19]. Next, we estimated blood immune cell-type proportions (natural killer cells, B cells, monocytes, CD8 T cells, CD4 T cells, and granulocytes) using the estimateLC function in the ewastools package [20] with reference dataset specific for peripheral blood cell types provided by Salas et al. [21]. Finally, methylation β-values were obtained for each CpG locus, which were based on the ratio of the methylated CpG probe intensity and total methylation probe intensity (methylated + unmethylated probe intensity), resulting in DNAm proportions ranging from 0.00 to 1.00 (corresponding to unmethylated to fully methylated).

2.2.2. Sample filtering

Sex was predicted based on the median total methylation intensity of X- and Y-mapped CpG probes, which uses the signal intensities (unmethylated and methylated) of the sex chromosomes to estimate the copy number of each sex chromosome; based on the expectation that males have one X and Y chromosome; and females having two X chromosomes. Samples whose methylation predicted sex did not match the reported sex were removed from downstream analyses (n = 2 from timepoint 1); likely as a result of technical error in sample processing. Outlier detection was conducted using the lumi [22] and watermelon [23] R packages and outlier samples were removed if detected by either method (n = 4). Fourty-eight technical duplicate pairs and 21 internal control probes were assessed for batch effects, and no obvious batch effects were seen, therefore one of each technical duplicate pair was removed at random (n = 48) and all control samples were removed for downstream analysis (n = 42). After sample and probe filtering, 973 samples at timepoint 1, 974 samples at timepoint 2, and 764,530 probes remained.

2.2.3. Epigenetic age calculation

We calculated epigenetic age acceleration (EAA) using the GrimAge (version 1) [24] and DunedinPACE [25] algorithms, both second- and third-generation markers of biological aging trained on phenotypic measures of mortality and health. These clocks were selected over first-generation clocks (e.g., Horvath or Hannum), which are trained solely on chronological age. In contrast, later generation clocks are designed to capture DNAm changes associated with aging phenotypes, morbidity, and mortality. Both GrimAge and DunedinPACE have been previously associated with a variety of morbidity and mortality phenotypes such as cardiovascular disease, metabolic syndromes, inflammation, and all-cause mortality [24,25].

GrimAge was calculated using Horvath’s online calculator (https://dnamage.clockfoundation.org/) and DunedinPACE was calculated using the DunedinPACE GitHub repository (https://github.com/danbelsky/DunedinPACE). A measure of GrimAge acceleration (GrimAA) was calculated from GrimAge by regressing the epigenetic age measure on chronological age. A participant with GrimAA greater than 0 is considered to have an epigenetic age older than their chronological age, while a participant with GrimAA less than 0 is considered to have an epigenetic age younger than their chronological age. GrimAge and chronological age were highly correlated in our sample (r = 0.93 at both timepoints) (Figure S1). DunedinPace effect estimates are unitless and represent the percentage increase or decrease in the rate of epigenetic aging rather than a biological age in years, with values greater than 1.0 indicating an accelerated rate of biological aging. GrimAA and DunedinPACE were moderately correlated in our sample (r = 0.50 at visit 1 and r = 0.52 at visit 2), consistent with what was reported by Belsky et al. [25] (Figure S1).

Briefly, GrimAge was developed by first identifying DNAm surrogates of protein plasma levels that are indicators of physiological health, i.e., C-reactive protein, plasminogen activation inhibitor 1 (PAI-1), growth differentiation factor 15 (GDF15) and smoking pack-years. Second, time-to-death from all-cause mortality was regressed on chronological age, sex, and the surrogate DNAm biomarkers to develop an estimate of biological age and mortality risk. The algorithm was validated using 485k CpG sites available on both the Illumina EPIC and 450K arrays from the Framingham Heart Study (N = 2356) [24]. DunedinPACE is distinct from traditional epigenetic clocks, which are typically trained on cross-sectional data with large chronological age ranges, in that it was trained on longitudinal data from the Dunedin Study, a cohort of participants who are all the same chronological age (N = 1037). This allows for the creation of a biomarker of biological aging that can estimate the rate of aging over time, in which any differences in epigenetic age over time may be more confidently attributed to “deterioration in organ systems” and less likely due to confounding and survival bias [25].

In sum, the GrimAge and DunedinPACE clocks were selected as they are best suited to capture DNAm changes associated with both aging phenotypes and external factors driving morbidity and mortality [26]. Additionally, previous studies have shown that psychosocial factors and adversity are more consistently associated with second- and third-generation clocks than with first-generation clocks, which further supports their use when assessing psychological, social, or stress-related exposures [27–30].

2.3. Psychosocial exposures

We assessed psychosocial factors from four domains: Acculturative and Discrimination Stress, Social Support, Mental Health, and Stress. All psychosocial factors were collected in the SCAS, approximately concurrent with HCHS/SOL Visit 1.

2.3.1. Acculturative and discrimination stress

Acculturative stress was measured by the abbreviated 17-item Hispanic Stress Inventory (HSI) which is used to assess lifetime exposure to stress considering cultural, political, and social factors across five subscales including occupational/economic stress, parental stress, marital stress, immigration stress, and familial/cultural stress [31,32]. Each question was assessed using a 5-point Likert scale, with 1 meaning “not at all worried/tense” and 5 meaning “extremely worried/tense.” All responses were summed to compute a total score where a higher score meant a higher presence of acculturation stress. Interpersonal discrimination was assessed using the Perceived Ethnic Discrimination Questionnaire – Community Version Brief (PEDQ-CVB), a 17-item scale examining the frequency of perceived racial or ethnic discrimination across four domains: exclusion/rejection, stigmatization/devaluation, discrimination at work/school, and threat/aggression. Questions were asked on a 5-point scale where 0 meant “never happened” and 5 meant “happened very often” [33]. All responses were summed to compute a total score where a higher score indicated greater lifetime discrimination. This scale has been validated in the current cohort for use in Hispanic/Latino populations [34].

2.3.2. Social support

Social support was assessed using a 12-item version of the Interpersonal Support Evaluation List (ISEL), which examined the availability of different types of social support across three domains: appraisal, belonging, and tangible support [35]. Questions were asked on a 4-point scale where 0 meant “definitely false” and 3 meant “definitely true.” All responses were summed to compute a total score where a higher score meant more overall social support. This scale has been validated in the current sample for use in Hispanic/Latino populations [36].

2.3.3. Mental Health

Mental health was assessed using the Center for Epidemiologic Studies Depression Scale (CES-D) and the Spielberger Trait Anxiety Inventory (STAI). The CES-D is a 10-item scale that assesses depression symptoms in the last 7 days [37]. Questions were asked on a 4-point Likert scale, where 0 meant “rarely or none of the time” and 3 meant “all of the time.” Responses were summed to compute a total depression score, where higher scores reflected more depression symptomatology. The scale has been validated in the current sample for use in Hispanic/Latino populations [38]. The STAI is a 10-item scale that examines the tendency to experience features of trait anxiety across a range of situations and events. Questions were asked on a 4-point Likert scale, where 0 meant “Almost Never” and 3 meant “Almost Always.” All responses were summed to compute a total score and a higher score indicated greater anxiety.

2.3.4. Stress

Stress exposure assessments encompassed childhood stress, adult traumatic stress, ongoing chronic stress, and perceived stress. Childhood stress was measured using the Adverse Childhood Experience (ACE) scale and traumatic stress was measured using the Traumatic Stress Schedule (TSS). The 10-item ACE scale assesses childhood stress exposure by measuring the number of adverse experiences such as abuse, neglect, divorce, and imprisonment of a household member, before the age of 18 [39]. Questions were asked in a Yes/No format. All responses were summed to compute a total score and a higher score indicated greater childhood adversity. The TSS is a 10-item scale that assesses lifetime exposure to traumatic stress by measuring the occurrence, frequency, and recency of traumatic events such as assault, death of a close family member or friend, natural disaster, and exposure to war [40]. Responses were summed to compute a lifetime exposure to stress score, where a higher score indicated a higher burden of lifetime stress. Perceived stress was evaluated using the 10-item Perceived Stress Scale (PSS), a widely used instrument for measuring psychological stress in the past 30 days based on respondents’ impressions of stressful, uncontrollable, and unpredictable situations that occurred during the last month [41]. Questions were asked on a 4-point Likert scale to assess how often participants felt stressed during situations, where 0 meant “Never” and 4 meant “Very Often.” All responses were summed to compute a total score and higher scores meant higher perceived stress. This scale has been validated in the current sample for use in Hispanic/Latino populations [42]. Chronic stress was evaluated using the 8-item Chronic Burden Scale, which examines the presence of ongoing problems (such as health problems, ability to work, financial strain, and relationship difficulties) [43]. Questions were asked in Yes/No format, with the first question inquiring about the occurrence of a stressor and the second question inquiring whether the problem has been ongoing for at least 6 months. A chronic stressor was defined as to have occurred and lasted at least 6 months. Responses were summed to compute a total count of chronic stressors.

2.4. Statistical analysis

First, we modeled the joint effects of psychosocial factors using self-organizing maps (SOM) to create exposure cluster profiles. Second, we estimated the association of the SOM exposure cluster with EAA at each time point separately using linear regression models. Third, we estimated the effect of SOM exposure cluster on epigenetic age over time using linear mixed effect models, which we consider our primary analysis. We used a directed acyclic graph (DAG) for a priori covariate selection (Figure S2). All models were adjusted for sex and chronological age at baseline. All models also accounted for sampling weights and therefore we employed robust standard errors. For each of these models, we assessed effect measure modification by sex and conducted several sensitivity analyses to assess the robustness of our results when additionally adjusting for additional covariates, as described later.

2.4.1. Modeling joint effects using self-organizing maps

We utilized self-organizing maps (SOM), to model the joint effects of psychosocial factors on EAA [44]. SOM is an unsupervised machine learning algorithm that reduces the dimensionality of the data by clustering participants with similar exposure profiles using a hybrid method of k-means clustering and multi-dimensional scaling. It is useful when multiple exposures are highly correlated and is easily interpretable, unlike other clustering approaches such as principal component analysis. SOM can help us identify vulnerable subgroups and identify harmful exposure patterns. The algorithm works by iteratively presenting each input vector (i.e., a psychosocial exposure) to the map, identifying the best matching unit, which is defined as the cluster with the closest weight vector, and updating the best matching unit and its neighbors to more closely resemble the input exposure. As the learning rate increases, the magnitude of this adjustment decreases over time and with distance from the best matching unit. As a result, similar clusters are mapped closer together and clusters that are different are placed further apart. Clusters are identified by maximizing intra-cluster homogeneity and inter-cluster heterogeneity [45]. The number of clusters chosen was decided based on several statistical measures identifying group structure including the adjusted R², the Akaike information criterion (AIC), the distribution of participants across clusters, as well as visual inspection of clusters for interpretability. The SOM framework results in a two-dimensional projection (Kohonen map) of exposure class profiles or clusters. Further details on the SOM algorithm have been described elsewhere [46]. Before inputting the exposures into the SOM algorithm, all exposure variables were scaled and centered with a mean of 0 and a standard deviation of 1. Skewed exposures were log-transformed on the log base 10 scale, and a constant of 1 was added to survey scales that had zero-values. The ISEL social support scale was reversed so that all exposures were consistent in direction. The SOM clusters were assessed as the main exposure in all statistical analyses using indicator variables, with one cluster with the lowest average psychosocial factor levels chosen as the reference category.

2.4.2. Modeling epigenetic aging over time

We utilized linear mixed-effect models (LMEM) to estimate the effect of exposure cluster (${\beta }_1$) on EAA over the two time points. $\mathit{EA}{A}_{\mathit{ij}}$ is epigenetic age acceleration for individual $\mathit{i}$ at time $\mathit{j}$. We included a random slope for time since visit 1 $u_{1\mathit{i}}\sim \mathit{N}\left(0,{\sigma }_{u_1}^2\right)$ to allow for between-subject variation in the rate of change between the exposure cluster and EAA and a random intercept $u_i\sim \mathit{N}\left(0,{\sigma }_u^2\right)$ to account for between-subject variation in EAA. ${\epsilon }_{\mathit{ij}}$ is the residual error term weighted by sampling weights. We accounted for heterogeneity across visit dates by adjusting for time since visit 1 between the two visit dates (${\beta }_2$). We also included an interaction term between time since visit 1 and the exposure cluster to account for potential increases in epigenetic age acceleration over time (${\beta }_5$). We adjusted for sex (${\beta }_4$) and chronological age at baseline (${\beta }_3$). Because this was a weighted analysis, we employed robust standard errors (${\epsilon }_{\mathit{ij}}$) using the sandwich R package.

$$\mathit{EA}{A}_{\mathit{ij}}={\beta }_0+{\beta }_1\left(\mathit{SOM}\mathit{Exposure}\mathit{Cluste}{r}_{\mathit{ij}}\right)+{\beta }_2\left(\mathit{tim}{e}_{\mathit{ij}}\right)+{\beta }_3\left(\mathit{ag}{e}_{\mathit{i}1}\right)+{\beta }_4\left(\mathit{se}{x}_{\mathit{ij}}\right)+{\beta }_5\left(\mathit{SOM}\mathit{Exposure}\mathit{Cluste}{r}_{\mathit{ij}}\ast \mathit{tim}{e}_{\mathit{ij}}\right)+u_i+u_{1\mathit{i}}\left(\mathit{tim}{e}_{\mathit{ij}}\right)+{\epsilon }_{\mathit{ij}}$$

2.4.3. Sensitivity analyses

We conducted five sensitivity analyses to assess the robustness of our results. First, we adjusted for education as a proxy for socioeconomic status (SES). Since SES can have a bidirectional relationship with some of the psychosocial factors, we did not consider it a traditional confounder. Additionally, the SOM clusters were well distributed by education status. Education was defined as a three-level variable: no high school diploma/GED; at most high school diploma/GED; greater than high school diploma/GED. Health-related behaviors such as cigarette smoking and body mass index (BMI) likely lie on the causal pathway between psychosocial factors and EAA and are likely acting as mediators than true confounders (Figure S2). Adjusting for these covariates would block part of the total effect of psychosocial factors on EAA we aim to estimate. Additionally, the GrimAge algorithm incorporates DNAm surrogates of smoking exposure and therefore adjusting for smoking may lead to overadjustment. Given this complexity and that there is currently no gold standard for covariate adjustment for smoking in GrimAge models, we opted not to adjust for these potential mediators in our main models and instead, additionally adjusted for cigarette smoking or BMI in a sensitivity analysis. Cigarette smoking was defined as a three-level variable: never smoker; former smoker; current smoker. A continuous measure for BMI, which was defined as weight in kilograms divided by height in meters squared was used. Third, we adjusted for genetic ancestry using the first five genetic principal components to assess for confounding by population stratification in the subset of participants that had this data available (N = 740). Finally, we adjusted for cell-type proportions to assess whether the relationship between the exposure clusters and EAA is partly explained by changes in blood immune cell composition.

2.4.4. Effect measure modification by sex

To assess whether the effect of exposure cluster on epigenetic aging differs by sex, we included an interaction term between the exposure cluster and sex. At visit 1, sex was defined as a binary variable, and participants were identified as male or female.

3. Results

3.1. Study sample characteristics

This analysis sample consisted of 922 participants with exposure data, DNAm data and all relevant covariates available. This sample consisted of 61.1% females with an average age of 45 years old at the first visit. Most participants were not born in the US 50 states (81.7%), with the largest fraction of participants identifying as Mexican (41.1%), followed by Cuban (14.0%) and Puerto Rican (13.7%) (Table 1). Education status varied as well, with 36.7% of the population having no high school diploma/GED, 27.1% having at most a high school diploma/GED, and 36.0% having greater than a high school diploma/GED. The majority of participants identified as never smoking cigarettes (62.5%), followed by 17.9% identifying as former smokers and 19.5% as current smokers.

Table 1.

Study population characteristics of the Hispanic Community Health Study/Study of Latinos sociocultural Ancillary study (HCHS/SOL SCAS) analytic sample stratified by self-organizing map (SOM) cluster.

	Cluster 1	Cluster 2	Cluster 3	Cluster 4 (Ref)	Overall
	(N = 196)	(N = 250)	(N = 250)	(N = 226)	(N = 922)
A. Study Population Characteristics
Age
Mean (SD)	45.6 (11.1)	43.9 (12.4)	45.1 (13.8)	44.3 (14.0)	44.7 (13.0)
Sex (%)
Female	127 (64.8%)	136 (54.4%)	169 (67.6%)	131 (58.0%)	563 (61.1%)
Male	69 (35.2%)	114 (45.6%)	81 (32.4%)	95 (42.0%)	359 (38.9%)
Education (%)
No high school diploma/GED	91 (46.4%)	65 (26.0%)	110 (44.0%)	72 (31.9%)	338 (36.7%)
At most high school diploma/GED	50 (25.5%)	73 (29.2%)	74 (29.6%)	53 (23.5%)	250 (27.1%)
Greater than high school diploma/GED	55 (28.1%)	111 (44.4%)	65 (26.0%)	101 (44.7%)	332 (36.0%)
Missing	0 (0%)	1 (0.4%)	1 (0.4%)	0 (0%)	2 (0.2%)
Center (%)
Bronx	55 (28.1%)	67 (26.8%)	53 (21.2%)	51 (22.6%)	226 (24.5%)
San Diego	41 (20.9%)	69 (27.6%)	46 (18.4%)	76 (33.6%)	232 (25.2%)
Chicago	53 (27.0%)	46 (18.4%)	86 (34.4%)	46 (20.4%)	231 (25.1%)
Miami	47 (24.0%)	68 (27.2%)	65 (26.0%)	53 (23.5%)	233 (25.3%)
Hispanic/Latino Background (%)
Dominican	17 (8.7%)	26 (10.4%)	25 (10.0%)	32 (14.2%)	100 (10.8%)
Central American	15 (7.7%)	37 (14.8%)	22 (8.8%)	24 (10.6%)	98 (10.6%)
Cuban	28 (14.3%)	30 (12.0%)	42 (16.8%)	29 (12.8%)	129 (14.0%)
Mexican	70 (35.7%)	96 (38.4%)	108 (43.2%)	105 (46.5%)	379 (41.1%)
Puerto Rican	43 (21.9%)	39 (15.6%)	28 (11.2%)	16 (7.1%)	126 (13.7%)
South American	14 (7.1%)	16 (6.4%)	21 (8.4%)	14 (6.2%)	65 (7.0%)
More than one/Other heritage	9 (4.6%)	6 (2.4%)	2 (0.8%)	6 (2.7%)	23 (2.5%)
Missing	0 (0%)	0 (0%)	2 (0.8%)	0 (0%)	2 (0.2%)
Nativity Status (%)
Not Born in US 50 states	152 (77.6%)	185 (74.0%)	222 (88.8%)	194 (85.8%)	753 (81.7%)
Born in US 50 states	44 (22.4%)	65 (26.0%)	27 (10.8%)	32 (14.2%)	168 (18.2%)
Missing	0 (0%)	0 (0%)	1 (0.4%)	0 (0%)	1 (0.1%)
Cigarette Use (%)
Never smoker	93 (47.4%)	146 (58.4%)	146 (58.9%)	160 (70.8%)	576 (62.5%)
Former smoker	48 (24.5%)	53 (21.2%)	51 (20.6%)	33 (14.6%)	165 (17.9%)
Current smoker	55 (28.1%)	51 (20.4%)	51 (20.6%)	33 (14.6%)	180 (19.5%)
Missing	0 (0%)	0 (0%)	0 (0%)	0 (0%)	1 (0.1%)
Body Mass Index
Median [IQR]	29.3 [6.7]	29.5 [6.9]	28.7 [6.6]	28.4 [7.0]	29.1 [6.9]
Missing	2	1	0	0	3
B. Psychosocial Measures
Hispanic Stress Inventory
Median [Min, Max]	24.0 [0, 70.0]	10.0 [0, 54.2]	12.0 [0, 65.0]	3.00 [0, 33.0]	10.0 [0, 70.0]
Racism Scale
Median [Min, Max]	33.0 [17.0, 63.0]	25.0 [17.0, 59.0]	21.0 [17.0, 47.0]	19.0 [17.0, 42.0]	23.0 [17.0, 63.0]
Depression Scale
Median [Min, Max]	13.0 [2.00, 29.0]	5.00 [0, 23.0]	9.00 [0, 24.0]	2.00 [0, 11.0]	7.00 [0, 29.0]
Anxiety Scale
Median [Min, Max]	23.2 [12.0, 35.0]	15.0 [10.0, 29.0]	19.0 [12.0, 32.0]	13.0 [10.0, 22.0]	17.0 [10.0, 35.0]
Chronic Stress Scale
Median [Min, Max]	3.00 [0, 8.00]	2.00 [0, 6.00]	1.00 [0, 5.00]	0 [0, 4.00]	2.00 [0, 8.00]
Perceived Stress Scale
Median [Min, Max]	22.0 [8.00, 38.0]	13.7 [0, 28.0]	17.0 [4.00, 30.0]	8.00 [0, 23.0]	15.0 [0, 38.0]
Traumatic Stress Schedule
Median [Min, Max]	3.00 [0, 9.00]	3.00 [0, 9.00]	1.00 [0, 5.00]	1.00 [0, 6.00]	2.00 [0, 9.00]
Adverse Childhood Experiences
Median [Min, Max]	4.00 [0, 10.0]	3.00 [0, 10.0]	1.00 [0, 7.00]	1.00 [0, 7.00]	2.00 [0, 10.0]
Social Support Scale
Median [Min, Max]	20.0 [0, 36.0]	30.0 [17.0, 36.0]	22.0 [6.00, 36.0]	30.0 [12.0, 36.0]	26.0 [0, 36.0]
C. Epigenetic Age Measures
GrimAge
Visit 1 Mean (SD)	55.3 (9.6)	53.6 (10.3)	54.1 (11.5)	53.1 (11.6)	54.0 (10.8)
Visit 2 Mean (SD)	59.5 (9.2)	57.5 (10.1)	58.0 (11.3)	57.4 (11.5)	58.0 (10.6)
GrimAge Acceleration
Visit 1 Mean (SD)	0.58 (4.2)	0.22 (3.7)	−0.25 (4.0)	−0.49 (3.6)	−0.004 (3.9)
Visit 2 Mean (SD)	0.66 (4.2)	0.15 (3.7)	−0.37 (4.0)	−0.40 (3.7)	−0.020 (3.9)
DunedinPACE
Visit 1 Mean (SD)	1.08 (0.12)	1.06 (0.11)	1.07 (0.11)	1.05 (0.11)	1.06 (0.11)
Visit 2 Mean (SD)	1.09 (0.12)	1.08 (0.11)	1.08 (0.11)	1.06 (0.12)	1.08 (0.12)

3.2. Self-organizing maps

We used the SOM algorithm to assess the effect of observed psychosocial exposure patterns on epigenetic aging. The SOM algorithm assigns individuals to clusters based on similar psychosocial exposure profiles and as such variability in the distribution of the psychosocial factors across clusters is expected by design. The SOM algorithm identified four clusters and participants were evenly distributed across clusters (Table 1; Figure 1). Cluster 1 participants (n = 196 (21.3%)) are characterized as experiencing on average the highest levels of all psychosocial risk factors and sources of stress. Cluster 2 (n = 250 (27.1%)) participants are characterized as experiencing on average high levels of chronic, traumatic, and childhood stress (ACEs). Cluster 3 (n = 250 (27.1%)) participants are more likely to experience more mental health symptoms, low social support, and higher perceived stress. Cluster 4 (n = 226 (24.5%)) participants are characterized as experiencing lower levels of all psychosocial risk factors and sources of stress relative to the other clusters and therefore this cluster was used as the reference group in all statistical analyses. A larger percentage of participants were born in the US 50 states in Cluster 2 (26.0% US-born) in comparison to Cluster 3 (10.8% US-born) which had more immigrants relative to the total study sample (18.2% US-born). Clusters 1 and 2 had a larger percentage of Puerto Ricans (21.9% and 15.6%, respectively) relative to the study sample (13.7%). Additionally, Cluster 4 participants had a larger percentage of Mexicans (46.5%) relative to the study sample (41.1%). Cluster 1 participants had a higher percentage of participants that did not complete high school/GED (46.4%), whereas Cluster 4 participants had a higher percentage of participants that completed more than high school/GED (44.7%).

Figure 1.

Self-organizing map (SOM) clusters star plot, where each circle is an SOM cluster, and each slice of a circle represents the median value of a mixture component. Cluster 1 participants (n = 196 (21.3%)) are characterized as experiencing on average the highest amounts of all psychosocial factors and sources of stress. Cluster 2 (n = 250 (27.1%)) participants are characterized as experiencing on average high levels of chronic, traumatic, and childhood stress (ACEs). Cluster 3 (n = 250 (27.1%)) participants are more likely to experience more mental health symptoms, low social support, and higher perceived stress. Cluster 4 (n = 238 (24.5%)) participants are characterized as experiencing lower levels of all psychosocial factors and sources of stress relative to the other clusters and therefore this cluster was used as the reference group in statistical analyses.

3.3. The association between exposure cluster and epigenetic age acceleration at visit 1 and visit 2

Participants in Cluster 1 (high levels of all psychosocial factors) and Cluster 2 (high levels of chronic, traumatic, and childhood stress) exhibited accelerated epigenetic aging with GrimAA and DunedinPACE in comparison to Cluster 4 (lowest levels of all psychosocial factors) at both timepoints (Table 2). The magnitudes of effect for GrimAA, interpreted as mean increase in GrimAA, were stronger among Cluster 1 (Visit1: β = 1.32 years; 95% CI (0.34,2.30); Visit 2: β = 1.62 years; 95% CI (0.61,2.63) in comparison to Cluster 2 (Visit 1: β = 0.93 years; 95% CI (0.10,1.76); Visit 2: β = 0.89 years; 95% CI (−0.05,1.83)). However, the DunedinPACE magnitudes of effect, interpreted as percent increase in DunedinPACE of aging, were similar for both Cluster 1 (Visit1: β = 0.03; 95% CI (0.001,0.06); Visit 2: β = 0.03; 95% CI (0.002,0.06)) and Cluster 2 (Visit 1: β = 0.03; 95% CI (1.30E–04,0.06); Visit 2: β = 0.04; 95% CI (0.007,0.07)). Additionally, magnitudes of effect were generally consistent across both time points. There was no evidence that participants in Cluster 3 (more mental health symptoms, low social support, and high perceived stress) experienced accelerated epigenetic aging in comparison to Cluster 4.

Table 2.

Association between self-organizing map exposure cluster and epigenetic age acceleration at visit 1 and visit 2. Models were adjusted for age at baseline and sex and were weighted by sampling weights. Robust standard errors are presented.

EAA algorithm	GrimAA		DunedinPACE
Visit 1 (N = 922)	Beta (95% CI)	P-value	Beta (95% CI)	P-value
Cluster 1	1.32 (0.34,2.30)	0.008	0.03 (0.001,0.06)	0.043
Cluster 2	0.93 (0.10,1.76)	0.028	0.03 (1.30E–04,0.06)	0.049
Cluster 3	0.36 (−0.48,1.20)	0.404	0.02 (−0.01,0.04)	0.262
Cluster 4	Reference		Reference
Visit 2 (N = 922)	Beta (95% CI)	P-value	Beta (95% CI)	P-value
Cluster 1	1.62 (0.61,2.63)	0.002	0.03 (0.002, 0.06)	0.038
Cluster 2	0.89 (−0.05,1.83)	0.063	0.04 (0.007,0.07)	0.016
Cluster 3	0.13 (−0.81,1.06)	0.789	0.01 (−0.02,0.04)	0.497
Cluster 4	Reference		Reference

Abbreviations: EAA, epigenetic age acceleration; GrimAA, GrimAge Acceleration.

3.4. The association between exposure cluster and epigenetic age acceleration over time

We found evidence of accelerated epigenetic aging across all exposure clusters (Table 3), with Cluster 1 exhibiting the strongest effect estimate. Participants in Cluster 1 had a mean increase in GrimAA of 1.27 years (β = 1.27; 95% CI (0.57,1.97)), and 3% higher DunedinPACE of aging (β = 0.03; 95% CI (0.01,0.05)) in comparison to Cluster 4. Participants in Cluster 2 had a mean increase in GrimAA of 0.62 years (β = 0.62; 95% CI (0.01,1.23) and 1% higher DunedinPACE of aging (β = 0.01; 95% CI (−0.005, 0.03)) in comparison to Cluster 4. Cluster 3 participants exhibited a mean increase in GrimAA of 0.51 years (β = 0.51; 95% CI (−0.13,1.15)) and 2% higher DunedinPACE of aging (β = 0.02; 95% CI (0.001,0.04)). Additionally, there was no evidence that the rate of EAA was changing over time, as the interaction terms between all exposure clusters and the change in chronological age were null (Table S1).

Table 3.

Association between self-organizing map exposure cluster and epigenetic age acceleration over time. The model was adjusted for age at baseline and sex and weighted by sampling weights. Robust standard errors are presented.

	Linear mixed-effects model (N = 922)
	GrimAA		DunedinPACE
EAA algorithm	Beta (95% CI)	P-value	Beta (95% CI)	P-value
Cluster 1	1.27 (0.57,1.97)	<0.001	0.03 (0.01,0.05)	0.003
Cluster 2	0.62 (0.01,1.23)	0.045	0.01 (−0.005,0.03)	0.155
Cluster 3	0.51 (−0.13,1.15)	0.11	0.02 (0.001,0.04)	0.042
Cluster 4	Reference		Reference

Abbreviations: EAA, epigenetic age acceleration; GrimAA, GrimAge Acceleration.

The estimated standard deviation for the random intercept was 3.61 (variance = 13.06), indicating substantial between-individual differences in baseline EAA. The standard deviation for the random slope of time was 0.32 (variance = 0.10), suggesting some variability in the rate of EAA change across individuals, though less pronounced than variation in baseline levels. The residual standard deviation was 0.54 (variance = 0.29), reflecting relatively low within-individual variance and low residual error. Additionally, the random intercept and slope were moderately negatively correlated (r = −0.26), suggesting that individuals with higher baseline EAA tended to experience slower rates of age acceleration over time.

3.5. Sensitivity analyses

We conducted five sensitivity analyses in which we additionally adjusted for education, cigarette use, BMI, cell-type proportions, or genetic ancestry in the longitudinal analysis in which we investigated the association between exposure cluster on EAA over time. The sensitivity analyses additionally controlling for either education or cell-type proportions were consistent with the primary model, with similar effect estimates over time (Table S2; Table S3). When additionally adjusting for cigarette smoking, the association between exposure clusters and GrimAA became largely null and the magnitudes of effect were attenuated, however, the associations remained robust with the main analysis for DunedinPACE (Cluster 1 (β = 0.02; 95% CI (3.8E–04,0.04)) and Cluster 3 (β = 0.02; 95% CI (7.5–04,0.04)) (Table S4). When additionally adjusting for genetic ancestry using genetic principal components, the associations persisted for both Cluster 1 and 2 and GrimAA and for Cluster 1 and DunedinPACE. However, the magnitude of effect was smaller for Cluster 1 and GrimAA than in the main analysis (β = 0.85; 95% CI (0.10,1.59)) (Table S5). Additionally, the association between Cluster 3 and DunedinPACE was no longer seen although the magnitude of effect changed little (β = 0.01; 95% CI (−0.007,0.03)) (Table S5). This can be expected given the loss of statistical power in this subset analysis (N = 746). Finally, when additionally adjusting for BMI, the associations between the exposure cluster and EAA remained consistent with the primary model, with the exception for Cluster 2 and GrimAA, which was no longer statistically significant (Table S6).

3.6. Effect modification

There was no evidence of effect measure modification by sex of the association between SOM exposure cluster and epigenetic age acceleration over time using either clock (Table S7).

4. Discussion

In this prospective cohort study of Hispanic/Latino adults, we found evidence that exposure to cumulative psychosocial factors is associated with epigenetic age acceleration at both visits and over time. Using self-organizing maps, we identified four clusters of participants with different exposure profiles. Cluster 1 participants are characterized as experiencing high levels of all psychosocial factors and stressors. Cluster 2 participants are characterized as experiencing high levels of chronic, traumatic, and childhood (ACEs) stress but are less likely to report certain mental health outcomes, such as depression and anxiety, and tend to report adequate social support. Conversely, Cluster 3 participants remote more likely to have poorer mental health symptoms, low social support, and higher perceived stress. Cluster 4 participants are characterized as having the lowest level of all psychosocial factors and sources of stress. At visit 1 and visit 2, we found evidence of EAA with both DunedinPACE and GrimAA among Cluster 1 participants, characterized as experiencing on average high levels of psychosocial factors across all domains (acculturative stress, discrimination stress, low social support, mental health, and other stress) and among Cluster 2 participants, characterized as experiencing on average high levels of stress, particularly traumatic stress, chronic stress, lifetime stress, and adverse childhood experiences. Using linear mixed effect models to assess EAA over time, the associations among Cluster 1 and 2 remained for both clocks, though less precise for Cluster 2 and DunedinPACE. In addition, we identified an association among Cluster 3 participants and DunedinPACE, characterized as experiencing on average high levels of mental health symptoms including depression, anxiety, and perceived stress, and low levels of social support.

A handful of prior studies have investigated the effects of psychosocial factors on epigenetic age acceleration in adults, but longitudinal studies have yet to be conducted and not many have assessed cumulative stressors. Two prior studies have investigated the effects of cumulative stressors on EAA and our findings are comparable. A 2015 study identified an association between cumulative lifetime stress on EAA at one time point, measured using the Horvath clock, among a cohort of African Americans (n = 392) [9]. Another study identified independent associations between cumulative lifetime stress, childhood stress, and adult stress on EAA (DunedinPACE and GrimAge) among a cohort of Non-Hispanic Black and White individuals (N = 359) [11]. Several studies have identified associations between individual stressors and EAA, including socioeconomic status [7], chronic economic hardship [8], major depressive disorder [10], posttraumatic stress disorder [47], and adverse childhood experiences [48].

The magnitudes of effect were strongest among Cluster 1 and 2 participants, which both exhibited high levels of stress, particularly traumatic stress, adverse childhood experiences, and chronic stress. This is in line with the hypothesis that exposure to chronic stress [49], early-life adversity [50,51], and trauma [52] activates the stress response pathways and contributes to biological weathering and premature aging. Such stress response pathways include chronic, low-grade inflammation [53,54], metabolic dysfunction of glucocorticoids including glucocorticoid resistance [49,55], and dysregulation of the hypothalamic–pituitary–adrenal (HPA) axis [51,56,57]. More research should be done to further disentangle the biological pathways that lead to premature aging and EAA.

The magnitudes of effect were smallest, and the confidence intervals were least precise among Cluster 3 participants, which were characterized by higher levels of mental health symptoms including anxiety, depression, and perceived stress. The questionnaires that measured these asked about the participants’ symptoms in the short-term and were not diagnostic measures of mental health, whereas the questionnaires related to stress and trauma asked about experiences from childhood and across the life-course. This difference may suggest that chronic, consistent, and less time-invariant stressors (as characterized by Cluster 2 participants) may impact physiological dysregulation and consequently EAA in a more systemic manner than short-term, inconsistent psychosocial factors such as symptoms of depression and perceived stress (as characterized by Cluster 3 participants). Previous evidence has shown that chronic stressors are associated with systemic physiological changes such as low-grade inflammation, while acute stressors are associated with brief but strong inflammatory responses [58]. Further research is needed to disentangle the biological and epigenetic mechanisms between acute/symptomatic and chronic stressors and psychosocial factors.

Both GrimAge and DunedinPACE exhibited consistent directions of effect across clusters, with slight differences in precision. GrimAge is trained to predict time-to-death, while DunedinPACE estimates the current pace of physiological aging. The associations in the primary longitudinal analysis were more precise for GrimAge, which may suggest that cumulative psychosocial factors have a greater impact on long-term aging trajectories than on short-term pace of aging. Interestingly, Cluster 3, which consisted of more short-term psychosocial symptoms, was associated with DunedinPACE over time but not with GrimAge, which may further support the notion that GrimAge is more sensitive to long-term and chronic exposures while DunedinPACE may also capture more acute, time-varying physiological changes reflecting current pace of aging. A couple previous studies have suggested that DunedinPACE is more sensitive to detecting short-term changes than traditional epigenetic clocks [59,60].

In sensitivity analyses, we assessed the impact of additionally adjusting for education, cell-type proportions, cigarette smoking, or BMI; variables that may lie along the causal pathway between psychosocial factors and EAA. Results remained consistent with the primary analysis when additionally adjusting for cell-type proportions, education, and BMI. However, when additionally adjusting cigarette smoking, associations with GrimAA were attenuated and became largely, while associations with DunedinPACE remained robust. This attenuation could likely reflect the role of smoking as a health-related behavioral mediator in which adjusting for it blocks part of the total effect from psychosocial factors to EAA. Another possibility is that since smoking is already incorporated into the GrimAge algorithm, which includes DNAm surrogates for smoking pack-years and lifetime exposure, adjusting for smoking in the GrimAA models could potentially lead to overadjustment.

Our study has a few limitations. First, we were only able to assess EAA during middle age, and while we were able to detect an association, we were unable to pinpoint when the accelerated aging process began. While we did find evidence of EAA longitudinally, the magnitude of effect was very similar to those of Visit 1 and Visit 2 separately. This suggests that while the effect of psychosocial factors on EAA persists over time, the rate of acceleration does not change over 6 years. This further supports the hypothesis that EAA changes may begin earlier in the life course [61]. Evidence shows that EAA can begin as soon as birth and that maternal stressors during pregnancy can cause DNAm changes in the fetus [62,63]. The developmental origins of health and disease hypothesis (DOHaD) suggest that children are most vulnerable to epigenetic changes during important developmental changes and that these changes can influence disease risk later in life [64]. Future research should further investigate at what stages of the life course people are most vulnerable to epigenetic changes and EAA. Second, psychosocial factors were ascertained only at the first timepoint, and thus we were unable to assess whether changes in psychosocial factors contribute to the changes in EAA over time. Additionally, certain measures (e.g., depression symptoms and recent perceived stress) captured short-term exposure and may not be stable over time which can introduce non-differential exposure misclassification, biasing effect estimates toward the null. Next, while our primary research question was to identify potentially harmful patterns of cumulative psychosocial factors using a clustering approach, we did not assess the associations between individual psychosocial factors and EAA. Future work should explore which specific psychosocial domains contribute most to EAA. Finally, our sample size was relatively small which limited our ability to assess effect measure modification by potentially important subgroups such as nativity status and age.

This study has several strengths. First, we were able to assess epigenetic age acceleration longitudinally at two time points. Most epigenetic studies have assessed epigenetic changes and EAA at a single time point, but due to the natural variation of epigenetic modifications, it is important to assess how DNAm patterns are changing over time. Second, we employed a novel mixture method, SOM, to assess the joint effects of multiple, correlated, psychosocial factors. Finally, our analysis was conducted on a well-characterized, underrepresented cohort of Hispanics/Latinos, the largest minority group in the United States. Hispanics/Latinos experience higher rates of psychosocial factors, which enabled us to detect strong associations with EAA. This highlights the importance of conducting epigenetic studies among marginalized populations, who disproportionately carry the burden of stress and disease [65].

5. Conclusion

Overall, these findings suggest that exposure to multiple psychosocial factors contributes to epigenetic age acceleration over time and is particularly heightened among clusters with high exposures to multiple stressors across several domains, supporting the weathering hypothesis among minority and immigrant populations. Our findings provide motivation to assess how EAA potentially mediates the association between psychosocial factors and aging-related health outcomes such as cardiovascular disease. This study will help advance our understanding of how stressors influence EAA and gain insight into how biological processes may contribute to health disparities among Hispanic/Latino populations. Additionally, these results highlight the importance of addressing structural and early-life determinants of health through public health interventions aimed at reducing structural inequities and expanding culturally responsive mental health services. Such interventions may help mitigate the adverse health effects of chronic and traumatic stress and promote resilience in Hispanic/Latino populations. Finally, our findings support the potential use of epigenetic aging biomarkers as tools to evaluate how psychosocial stressors and social determinants of health become biologically embedded across the life course and to quantify the impact of prevention strategies on adverse health outcomes.

Funding Statement

The Hispanic Community Health Study/Study of Latinos was carried out as a collaborative study supported by contracts from the National Heart, Lung, and Blood Institute (NHLBI) to the University of North Carolina [N01-HC65233], University of Miami [N01-HC65234], Albert Einstein College of Medicine [N01-HC65235], Northwestern University [N01-HC65236], and San Diego State University [N01-HC65237]. The following Institutes/Centers/Offices contribute to the HCHS/SOL through a transfer of funds to the NHLBI: National Center on Minority Health and Health Disparities, the National Institute of Deafness and Other Communications Disorders, the National Institute of Dental and Craniofacial Research, the National Institute of Diabetes and Digestive and Kidney Diseases, the National Institute of Neurological Disorders and Stroke, and the Office of Dietary Supplements. Additional support was provided by the National Institute on Minority Health and Health Disparities [award R01MD013320] (Suglia). The project described was supported by Grant Number [2T32ES012870] from the National Institute of Environmental Health Sciences (Abrishamcar). The contents of this publication are solely the responsibility of the authors and do not necessarily represent the official views of the NIH. The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Article highlights

• This study investigated the cumulative impact of psychosocial factors on epigenetic age acceleration (EAA) at two time points in 922 adults from the Hispanic Community Health Study/Study of Latinos (HCHS/SOL) Sociocultural Ancillary Study (SCAS).

• An unsupervised clustering approach, self-organizing maps (SOM), was used to cluster participants based on their psychosocial exposure profiles, resulting in four clusters.

• Cluster 1 (n = 196; 21.3%), characterized by high levels of all psychosocial factors, and Cluster 2 (n = 250; 27.1%), characterized by adverse childhood experiences, traumatic stress, and chronic stress showed the greatest GrimAge acceleration and DunedinPACE over time.

• Smaller effect estimates were observed in Cluster 3 (n = 250; 27.1%), characterized by high levels of acute stressors (e.g., depression symptoms, perceived stress), which exhibited increased DunedinPACE but not GrimAge acceleration.

• Chronic, consistent, and less time-varying stressors may exert a greater influence on biological aging than short-term, fluctuating psychosocial factors.

• These findings highlight the cumulative impact of psychosocial factors on EAA and how stressors can get “under the skin” and contribute to health disparities.

Author contributions

Sarina Abrishamcar: Conceptualization; investigation; methodology; formal analysis; visualization; writing-original draft; writing-review and editing

Christian C. Dye: Data curation; writing-review and editing

Jasmine K. Aqua, Rebecca Jones-Antwi, Yinxian Chen, Anke Hüls, Andrea Baccarelli, Karen Conneely: Methodology; writing-review and editing

Linda C. Gallo, Maria M. Llabre, Krista M. Perreira, Martha L. Daviglus, Maria Argos, Bharat Thyagarajan, Jianwen Cai, Carmen R. Isasi, Robert C. Kaplan: Data curation; funding acquisition; resources; writing-review and editing

Shakira F. Suglia: Supervision; conceptualization; funding acquisition; data curation; methodology; resources; writing-review and editing

Disclosure statement

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

No writing assistance was utilized in the production of this manuscript.

Reviewer disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Ethical declaration

Written consent for participation was obtained from all participants, and ethical approval for human subjects research was given by the Institutional Review Boards at the University of North Carolina at Chapel Hill, Albert Einstein College of Medicine, Feinberg School of Medicine at Northwestern University, Miller School of Medicine at the University of Miami, and San Diego State University. The current study was approved by the Institutional Review Board at Emory University.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/17501911.2025.2540260