Socioeconomic Disadvantage and the Pace of Biological Aging in Children

Participants included 600 (285 girls) children and adolescents from 328 unique families aged 8 to 18 years (mean = 12.68, SD = 3.02) from the Texas Twin Project.³¹  The Texas Twin Project is an ongoing longitudinal study that includes the collection of salivary samples. Saliva samples were selected to be assayed for DNA methylation by using EPIC arrays if participants self-identified their race/ethnicity as White and/or Latinx and had contributed cortisol data (not reported here). After we excluded 8 participants during DNA methylation preprocessing, there were n = 457 participants who identified as only White, n = 77 who identified as only Latinx, and n = 61 who identified as Latinx and White. We capitalize these terms to highlight that racial and/or ethnic identities are social constructions that are not based on innate biosocial boundaries but may have biosocial effects through people’s lived experiences.³²  The University of Texas Institutional Review Board granted ethical approval.

Saliva samples were collected during a laboratory visit by using Oragene kits (DNA Genotek, Ottawa, Ontario, Canada). DNA extraction and methylation profiling were conducted by the Edinburgh Clinical Research Facility (Edinburgh, United Kingdom). The Infinium MethylationEPIC BeadChip Kit (Illumina, Inc, San Diego, CA) was used to assess methylation levels at 850 000 methylation sites. Methylation profiles were residualized for cell composition, array, and slide; all samples came from the same batch (see Supplemental Information for details). Comparison of DunedinPoAm-measured pace of aging in blood and saliva by using out-of-sample data sets is reported in Supplemental Fig 3. The analysis indicated good correspondence of DunedinPoAm measurements across blood and saliva.

See Table 1 for a description of DunedinPoAm-measured pace of aging calculation and the Supplemental Information for a description of epigenetic clocks calculation, which were converted to age-acceleration residuals for analysis by regressing participants’ computed epigenetic-clock age values on their chronological ages and predicting residual values.

We measured children’s socioeconomic disadvantage at the family and neighborhood levels of analysis (Table 1).

Smoking and obesity are socially patterned health behavior exposures that are more common in children from families and neighborhoods of lower SES.^7,33  They are also associated with differential DNA methylation patterns across the genome.^34–36  We therefore considered tobacco exposure and BMI in our analysis (Table 1).

Puberty tends to begin at younger ages in children growing up in conditions of early-life disadvantage, and meta-analytic evidence suggests that this is driven by experiences of threat but not deprivation or socioeconomic disadvantage.^23,37,38  In our sample, associations of socioeconomic disadvantage and puberty were positive but not statistically different from 0 (family level: r = 0.08, 95% confidence interval [CI] −0.02 to 0.18, P = .10; neighborhood level: r = 0.05, 95% CI −0.05 to 0.16, P = .29). Puberty is also associated with a range of DNA methylation changes.^39,40  We therefore considered children’s pubertal development in our analysis (see Table 1).

We tested associations of childhood socioeconomic disadvantage and racial and/or ethnic identity with the pace of biological aging using regression analysis. To account for nonindependence of data on siblings, we fitted regressions using linear mixed models implemented with the lme4 R package. To ensure that our primary mixed-effects models appropriately corrected for nonindependence of the sample, we reran models with 1 randomly selected twin per pair and also reran models with clustered SEs in Mplus software. The coefficients were similar across all analytic strategies, increasing confidence in the robustness of our results (see Results section of the Supplemental Information).

We report parameter estimates with bootstrapped 95% CIs (computed with 500 simulations by using lme4’s confint.merMod function). Continuous measures were standardized for analysis to mean = 0.00 and SD = 1.00, allowing for interpretation of effect sizes in the metrics of Pearson’s r in the case of continuously distributed exposure variables and Cohen’s d in the case of nominal variables (ie, race/ethnicity). All models were adjusted for sex and children’s chronological age.

To test if associations between socioeconomic disadvantage and pace of biological aging were accounted for by social gradients in tobacco exposure and/or obesity, we conducted covariate-adjusted regressions to evaluate sensitivity of results. We conducted a parallel analysis to evaluate independence of associations from pubertal development.

For comparison, we repeated the pace-of-aging analysis using age-acceleration residuals from 5 published epigenetic clocks.

DunedinPoAm-measured pace of aging from salivary DNA was approximately normally distributed in children and adolescents in the Texas Twin Project (before correction for the cell composition of saliva samples, the mean DunedinPoAm-measured pace of aging was 0.79 [SD = 0.05]). DunedinPoAm-measured pace of aging was similar in boys and girls (d = −0.03, 95% CI −0.11 to 0.06, P = .48). DunedinPoAm-measured pace of aging was similar for younger and older children (r = 0.02, 95% CI −0.08 to 0.11, P = .75). The relatively young and restricted age range of our sample may also explain why DunedinPoAm-measured pace of aging did not differ by age. In previous research, investigators have found that the pace of aging does increase with age but that this acceleration happens later in the life course.^24,41 

We first tested if children growing up in more socioeconomically disadvantaged circumstances exhibited faster methylation pace of aging. We conducted a separate analysis of socioeconomic disadvantage measured for children’s families and their neighborhoods. At both levels of analysis, children growing up under conditions of greater socioeconomic disadvantage exhibited a faster pace of aging, as measured by DunedinPoAm (family level: r = 0.18, 95% CI 0.08 to 0.27, P = .001; neighborhood level: r = 0.18, 95% CI 0.07 to 0.28, P = .001; Fig 1).

We next tested if children identifying as Latinx (12.9% of sample) and Latinx-White (10.3% of sample) exhibited a faster DunedinPoAm-measured pace of aging as compared with children identifying solely as White (76.8% of sample). Latinx-identifying children exhibited a faster DunedinPoAm-measured pace of aging compared with both White-identifying children (d = −0.21, 95% CI −0.32 to −0.09, P = .001) and Latinx White–identifying children (d = −0.16, 95% CI −0.27 to −0.04, P = .008; Fig 2).

Latinx children tended to be exposed to higher rates of family- and neighborhood-level cumulative socioeconomic disadvantage compared with White-identifying children (family level: d = −0.33, 95% CI −0.47 to −0.18, P < .001; neighborhood level: d = −0.50, 95% CI −0.55 to −0.30, P < .001) and Latinx White–identifying children (family level: d = −0.21, 95% CI −0.35 to −0.07, P = .004; neighborhood level: d = −0.29, 95% CI = −0.35 to −0.11, P < .001). We therefore tested if racial/ethnic group differences in DunedinPoAm-measured pace of aging were statistically explained by socioeconomic differences between the groups. Adjusting for group differences in family- and neighborhood-level disadvantage largely accounted for differences in methylation pace of aging between Latinx-identifying children and children identifying as White only (d = −0.07, 95% CI −0.22 to 0.08, P = .41) or Latinx-White (d = −0.09, 95% CI −0.23 to 0.05, P = .23; Fig 2); associations were no longer statistically different from 0. Statistical adjustment for racial and/or ethnic identity only modestly attenuated associations of socioeconomic disadvantage with DunedinPoAm-measured pace of aging (family level: r = 0.16, 95% CI 0.05 to 0.25, P = .003; neighborhood level: r = 0.14, 95% CI 0.03 to 0.26, P = .02).

To test confounding of socioeconomic disadvantage associations with pace of aging by smoking, we conducted 2 sets of analyses. First, we repeated the regression analysis, excluding self-reported smokers. There were few self-reported smokers in the sample (n = 11 children [1.8% of the sample]). Results were unchanged after we excluded these participants. Second, we repeated the regression analysis, adding covariate adjustment for DNA methylation measures of tobacco exposure (genome-wide DNA methylation scores [poly-DNAm] and AHRR smoking scores). Poly-DNAm smoking profiles were positively associated with DunedinPoAm-measured pace of aging (r = 0.19, 95% CI 0.11 to 0.28, P < .001). Covariate adjustment for DNA methylation measures of tobacco exposure only modestly attenuated associations at family and neighborhood levels (family level: r = 0.16, 95% CI 0.06 to 0.25, P = .001; neighborhood level: r = 0.15, 95% CI 0.05 to 0.25, P = .003).

To test confounding of socioeconomic disadvantage associations with pace of aging by obesity and/or overweight, we repeated the regression analysis, adding covariate adjustment for BMI. Children with higher BMI had a faster DunedinPoAm-measured pace of aging (r = 0.26, 95% CI 0.18 to 0.35, P < .001). Covariate adjustment for BMI modestly attenuated associations of DunedinPoAm-measured pace of aging with socioeconomic disadvantage (family level: r = 0.14, 95% CI 0.04 to 0.23, P = .005; neighborhood level: r = 0.13, 95% CI 0.02 to 0.25, P = .03).

To test confounding of socioeconomic disadvantage associations with pace of aging by accelerated pubertal development, we repeated the regression analysis, adding covariate adjustment for pubertal development. We considered 2 measures of puberty, the Pubertal Development Scale and, in girls, menarcheal status. DunedinPoAm-measured pace of aging was weakly associated with self-reported pubertal development; however, the effect size was not statistically different from 0 (Pubertal Development Scale: r = 0.07, 95% CI −0.02 to 0.15, P = .08). DunedinPoAm indicated somewhat faster aging in girls who had experienced their first menses compared with those who had not; the effect size was small but statistically different from 0 (d = 0.19, 95% CI 0.00 to 0.38, P = .04).

Covariate adjustment for the Pubertal Development Scale modestly attenuated associations of socioeconomic disadvantage with DunedinPoAm-measured pace of aging (family level: r = 0.17, 95% CI 0.07 to 0.27, P = .001; neighborhood level: r = 0.16, 95% CI 0.05 to 0.27, P = .003). Results were similar for covariate adjustment for menarcheal status among girls (unadjusted family level: r = 0.13, 95% CI 0.00 to 0.26, P = .06; unadjusted neighborhood level: r = 0.18, 95% CI 0.06 to 0.34, P = .01; adjusted family level: r = 0.14, 95% CI −0.01 to 0.27, P = .07; adjusted neighborhood level: r = 0.17, 95% CI 0.04 to 0.33, P = .02).

In contrast to results for DunedinPoAm, epigenetic clocks were not associated with socioeconomic disadvantage and, in 2 cases, associations were in the opposite direction expected (range of r = −0.03 to 0.05; Table 2). See the Supplemental Information for full results and correlations between epigenetic clocks (Supplemental Fig 4).

We analyzed saliva DNA methylation data from children and adolescents participating in the Texas Twin Project to test associations between childhood socioeconomic disadvantage and the pace of biological aging using a novel DNA methylation measure, DunedinPoAm. We found that children and adolescents growing up in more disadvantaged families and neighborhoods exhibited a faster pace of aging as measured by DunedinPoAm. Children with higher BMI and more advanced pubertal development for their age tended to have a faster DunedinPoAm-measured pace of aging, but these covariates did not account for observed socioeconomic disparities. Our results suggest that the DunedinPoAm DNA methylation measure of the pace of biological aging, originally validated in adults,²⁴  could potentially serve as a salivary biomarker sensitive in real time to social determinants of health experienced during childhood.

Questions remain about when in the life course biological processes of aging begin.⁴²  Observations of variation in cellular-level measurements of aging in pediatric samples suggest that quantification of biological aging in children may be possible.^16,30,43  In our study, children from socioeconomically disadvantaged families and neighborhoods showed a faster pace of aging on the basis of a DNA methylation measure that tracks faster declines in organ-system integrity in midlife adults and forecasts mortality. This result could reflect at least 2 phenomena. First is that socially disadvantaged children in our study showed patterns of biological wear and tear similar to adults who have faster pace of aging. Second is that the DunedinPoAm algorithm detected molecular signatures of exposures, including early-life adversities, that were causes of faster pace of aging in the adults in the original study. Despite this ambiguity, our results reveal a molecular continuity between social inequalities in childhood and health disparities in adulthood that can inform future studies of etiology and intervention.

In our analysis of racial and/or ethnic group differences, we found that Latinx-identifying children exhibited a faster DunedinPoAm-measured pace of aging compared with children identifying as White and Latinx-White. These differences were statistically accounted for by the greater socioeconomic disadvantage experienced by the Latinx children. Thus, our findings are consistent with observations that racial and/or ethnic socioeconomic disparities are an important contributor to racial and/or ethnic disparities in health.⁴⁴  Importantly, racial and/or ethnic disparities in adult health typically persist, although reduced, across all levels of SES, for example, because of race-based discrimination.^32,45,46 

In contrast to findings for DunedinPoAm-measured pace of aging, an analysis of several published DNA methylation clocks yielded null associations with socioeconomic disadvantage. These clocks revealed the expected associations with children’s chronological age (r = 0.66 to 0.81 in an age range of 8–18 years). However, the difference between participants’ epigenetic ages and their chronological ages (ie, epigenetic-age acceleration) did not vary with socioeconomic disadvantage, consistent with results from a recent meta-analysis revealing that threat, but not deprivation, was associated with epigenetic clocks in childhood samples.²³  Measures of epigenetic-age acceleration also appear to be less sensitive to racial and/or ethnic group differences in childhood and adolescence.⁴⁷  These results suggest that pace-of-aging measures, such as DunedinPoAm, may prove more sensitive to health-damaging effects of socioeconomic deprivation and of racialized disparities in SES, particularly in studies focused on the early life course.

We acknowledge limitations. First, we measured methylation in saliva DNA, which comes from a mixture of buccal cells and leukocytes, whereas the measures we analyzed were developed from blood or other tissues. We used DNA methylation algorithms to make statistical adjustment for the cellular composition of the saliva samples. Increasing confidence in our findings, they replicate results for DunedinPoAm and epigenetic clocks from blood DNA methylation measured in 18-year-olds.²⁴  Additionally, our analysis indicated good correspondence of DunedinPoAm measurements across blood and saliva (Supplemental Information).

Second, our observational design included DNA methylation data from a single time point, precluding a test of how change in socioeconomic circumstances might affect DNA methylation measures. Ultimately, not only longitudinal repeated-measures studies but also natural experiment studies and randomized controlled trials of social programs are needed to establish causal effects of social disadvantage on DunedinPoAm-measured pace of aging and to establish DunedinPoAm as a mediator of the process through which childhood disadvantage leads to aging-related health conditions.^48,49 

Finally, the biology that causes variation in DunedinPoAm-measured pace of aging and the epigenetic clocks remains poorly understood. Epigenetic changes are understood to be core features of the biological process of aging.^12,50  Yet, the methylation measurements we studied are only correlates of the unobserved processes of biological aging, not direct observations of it.⁵¹ 

Within the bounds of these limitations, our findings have implications for theory and future research. Theory and evidence from animal models suggest epigenetic changes are a mediator of early-life adversity’s effects on aging-related health decline.⁵²  However, human studies following-up specific mechanisms identified in animals have yielded equivocal results.⁵³  Our results contribute evidence toward proof of concept for the hypothesis that early adversity increases risk for adult disease through acceleration of the pace of biological aging.

In future research, investigators can use DunedinPoAm to test how aging processes may be accelerated in at-risk young people and if and how such accelerated aging may be modified, in particular, by program and policies that are aimed at reducing social and material disadvantage. Childhood interventions to improve equitable access to nutritious food, lower family stress, enhance neighborhood safety, and increase greenspace have the potential to improve concurrent and lifelong health.⁶  However, childhood interventions antedate the onset of adult disease by decades. This long gap has motivated interest in biological measures that can serve as surrogate end points for assessing the effectiveness of programs and policies aimed at improving lifelong health by promoting positive child development. Our results suggest that salivary DNA methylation measures of pace of aging may provide a surrogate or intermediate end point for understanding the health impacts of such interventions. Such applications may prove particularly useful for evaluating the effectiveness of health-promoting interventions in at-risk groups.

We thank all participants of the Texas Twin Project.

CI

confidence interval

poly-DNAm

genome-wide DNA methylation

SES

socioeconomic status