Developmental science is replete with examples of environmental stress-induced changes to the epigenome that then influence protein expression and downstream physiology, emotion, cognition, and behavior, including mental health symptoms and disorders. Seldom studied are the genetic influences on epigenetic variation, such as histone acetylation and methylation. Some alleles, for example, are more likely to be methylated than others, necessitating the study of genetic influences on epigenetic variation. Epigenetic clocks capture combinations of DNA methylation patterns (1). Biological age indexed by epigenetic clocks are more reliable measures of aging than chronological age (2), and importantly, variation in these clocks predicts mental health problems such as depression (3, 4). Accumulating evidence indicates that epigenetic clocks are heritable (5, 6). The second-generation epigenetic clock GrimAge is currently the best predictor of health span and lifespan (7). A recent Genome Wide Association Study (GWAS) (6) generated summary effect sizes for SNP prediction of GrimAge, providing the statistics needed for forming a polygenic risk score representing genomic influences on GrimAge. Polygenic scores aggregate variation across thousands of genetic variants across the genome, providing a cumulative measure of genetic predisposition (8). In the current study, we present a novel examination of whether polygenic predisposition for GrimAge epigenetic aging is moderated by adverse life events, as a risk factor, and parent acceptance, as a resilience factor, in predicting early adolescent depression.

Developmental psychopathology theory emphasizes the cumulative salience of multiple levels of risk and resilience within the social environment over the course of psychopathology (9, 10). Separate from this, the diathesis-stress theory of depression proposes that preexisting risk factors, such as genetic predisposition, can be exacerbated by risky environments, increasing likelihood of depression (11, 12). Conversely, within a resilience framework, positive environmental influences may buffer genetic predisposition for depression (13). As an extension and integration of these theories, the gene-environment cascade theoretical framework posits that that the interplay between genetic predisposition and the environment can have cascading effects that alter trajectories of mental health over time (14). Leveraging this framework, we expand traditional main effect genetic association models to include risk and resilience processes and developmental outcomes. Furthermore, we account for the known heterogeneity in depression by allowing for developmental subgroups of depression using growth mixture modeling.

Adverse life events refers to experiences or exposure to traumatic or stressful events. A robust literature including a recent meta-analysis indicates that adverse life events are associated with greater incidence of major depressive disorder before the age of 18 (15). The associations between adverse life events and increases in depression and depressive symptoms in childhood and adolescence have also been corroborated in large international and national samples (16, 17). Thus, adverse life events represent a compelling index of environmental risk during childhood. Moreover, there is ample evidence that risk for depression results from the interplay between adverse life events and genetic and epigenetic predispositions.

A recent review concluded that the interplay between genetic predisposition and adverse life events in childhood represents a classic example of gene-environment interaction (GxE) in mental disorder (18). This is supported by an umbrella review indicating that childhood risk for depression is consistent with a diathesis-stress framework resulting, in part, from adverse environmental exposure and genetic and/or epigenetic predispositions (19). Moreover, this review and its cited literature notes that variability in DNA methylation results from both genetic and environmental variation (20). This aligns with another review of biological factors in depression indicating that one route by which adverse life events in childhood contribute to depression is via the interplay between genetic and epigenetic systems (21). Thus, there is theoretical and research evidence indicating that genetic predisposition for epigenetic variation may be moderated by adverse life events in predicting depressive symptoms in early adolescence.

Parent acceptance encompasses warm, involved, and supportive behaviors a parent expresses towards their child. Four meta-analysis support associations between greater parental warmth and lower childhood/adolescent depression (22–25). Compared to GxE research on adverse life events, less research has examined parent acceptance as moderating genetic predisposition in predicting depression. One study found an interaction between polygenic predisposition for genetic sensitivity and parental warmth associated with depressive symptoms, but the polygenic index was based on only a few candidate genes (26). Using a twin model, parent nurturance moderated heritability for psychological resilience (27). More commonly, GxE studies have examined single candidate genes, negative aspects of parenting, and externalizing outcomes (28). This prompts the need for greater examination of positive parenting as a resilience factor operating as a buffer between genetic risk and pathways to early adolescent depression.

The current study examined whether genetic predisposition for GrimAge epigenetic aging was moderated by adverse life events and parent acceptance in predicting depressive symptom trajectories across early adolescence. Due to known heterogeneity in depression development, we used a person-centered approach to uncover depressive symptom classes across early adolescence separately within EA, AA, and LX subgroups. Person-centered approaches identify distinct patterns of behavior for subgroups of individuals whereas variable-centered approaches collapse variance in a behavior across individuals which can miss heterogeneity in the group (29). Growth Mixture Modelling (GMM) is one person-centered approach that identifies subgroups based on patterns of individual-level trajectories on a specific trait or behavior (30). We examined trajectories and genetic effects separately in EA, AA, and LX subgroups because there can be underlying variation in genetic ancestry across race and ethnicity (e.g., allele frequency, linkage disequilibrium patterns) which necessitate examining effects separately within each racial/ethnic group (31).

We hypothesized that genetic predisposition for GrimAge would be associated with greater likelihood of membership in trajectories with higher depressive symptoms. We hypothesized that adverse life events would moderate the effect of polygenic predisposition for GrimAge such that the association between genetic risk and membership in higher vs. lower depressive symptom trajectories would be stronger among youth experiencing more adverse life events. Conversely, we hypothesized that the association between genetic risk and membership in higher depressive symptom vs. lower trajectories would be weaker among youth who report higher levels of parental acceptance.

The Adolescent Brain Cognitive Development (ABCD; N = 11,875) Study is a large longitudinal study of youth in the United States with annual assessments on behavioral, social, and neurocognitive functioning starting at age 9/10 with four waves of data available to age 12/13. Youth in ABCD are 47.8% female and racially/ethnically diverse (52.1% non-Hispanic White, 15.0% non-Hispanic Black, 20.3% Hispanic/Latinx, 2.1% Asian, and 10.5% other [e.g., multiracial]). At age 9/10, median combined family income was $75k to $100k, with approximately 20% of the sample reporting earning $35k or less. The current study included data from the age 10/11, 11/12, and 12/13 assessments. Youth were included if they had genetic, depressive symptom, and environmental data and were White/European American (EA, n = 6,043; 47% female), Black/African American (AA, n = 1,640; 50% female), or Hispanic/Latinx (LX, n = 2,283; 48% female).

Youth in ABCD were primarily recruited using a probability sampling of schools located within 21 national study sites (32). Parents provided written informed consent for their own and their child’s participation and youth provided assent. Baseline data collection for age 9/10 began in September, 2016, with annual follow ups. Youth in ABCD are assessed in an array of domains encompassing psychosocial and family functioning, physical health, contextual and cultural environment, brain imaging, and whole genome genotyping. Data are released through the NIMH Data Archive. Data used in the present study came from ABCD data release 5.0.

Saliva samples were collected from youth at the age 9/10 assessment (33), which were genotyped by the Rutgers University Cell and DNA Repository (RUCDR). DNA from saliva samples was genotyped on the Smokescreen Genotyping Array (34). RUCDR performed DNA quality controls based on calling signals and variant call rates, and the quality-controlled genotyping ABCD data contains 11,099 unique individuals with 516,598 genetic variants. Imputation was performed via the TOPMed imputation server using mixed ancestry and Engle v2.4 phasing. Single nucleotide polymorphisms (SNPs) with a genotyping rate < 0.95 or that violated Hardy–Weinberg equilibrium (p < 10^-6) or with minor allele frequency < 0.01 were excluded from analysis.

There are potential differences in polygenic score functioning across racial and ethnic groups based on genetic ancestry (31). Therefore, we conducted all analyses separately by racial/ethnic subgroup. Using Growth Mixture Modelling (GMM), we examined trajectories of depressive symptoms across ages 10-11, 11-12, and 12-13 separately for EA, AA, and LX subgroups in Mplus v.8.8 including intercept, slope, and quadratic terms. We did not include wave 9/10 depressive symptoms in GMM as youth report of adverse life events was not available until age 10/11. We examined iterative GMM solutions starting with 1 class and increasing to 5 classes. We examined model fit based on Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), adjusted BIC (aBIC), Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (VLMR), Bootstrap Likelihood Ratio Test (BLRT), entropy, class sample proportions, and theoretical interpretability (53). Full information maximum likelihood was used to handle missing data based upon the missing at random (MAR) assumption.

Once we identified the optimal class solution, we examined associations between our predictors and covariates with depression trajectories separately within the EA, AA, and LX groups, using the R3step method (54). The R3step method estimates latent classes based on indicators (e.g., depression), creates a most likely latent class variable based on the posterior distribution, and regresses the predictor(s) on class membership. The R3step method tests for likelihood of trajectory membership across pairwise class comparisons which we examined using the lowest depression trajectory as the reference group.

Within the optimal class solution, we tested two models within each subgroup. The first model examined associations between the Grim-PGS, adverse life events, the Grim-PGS by adverse life events interaction term, and covariates with depression trajectories. The second model examined associations between the Grim-PGS, parent acceptance, the Grim-PGS by parent acceptance interaction term, and covariates with depression trajectories. Based on methodological recommendations, within all models we examined for Grim-PGS by covariate interactions, and non-significant interactions were trimmed from final models (52).

Descriptive statistics and correlations can be found in Table 1 . There were relatively low levels of adverse life events and relatively high levels of parent acceptance across EA, AA, and LX subgroups. Depressive symptoms were also low in all subgroups. There were significant racial/ethnic differences in youth’s experience of adverse life events, with the highest levels in the AA subgroup and lowest levels in the EA subgroup. Parental acceptance was highest in the EA subgroup and lowest in the AA subgroup. Depressive symptoms were lower in the AA subgroup compared to both the EA and LX subgroups. Within the EA subgroup the Grim-PGS was associated with lower parent acceptance and greater depression at age 10/11. Within all subgroups, adverse life events were associated with higher youth depression, and parent acceptance was associated with lower youth depression.

We conducted growth mixture modelling separately within the EA, AA, and LX subgroups, increasing the number of classes iteratively. Initial models had better fit without the quadratic term in all subgroups, so we excluded the quadratic slope from our iterative model testing. Model fit indices can be found in Table 2 . For the EA subgroup, a four-class model was optimal based on lower AIC, BIC, and aBIC values compared to the 3-class model and significant VLMR and BLRT values compared to the 5-class model. Average levels of depressive symptoms can be found in Figure 1 . We classified the four classes as high-decreasing (n = 138; 2.3%), increasing (n = 201; 3.3%), moderate (n = 633; 10.5%), and low (n = 5,071; 83.9%).

For the AA subgroup, a three-class solution was optimal based on lower AIC, BIC, and aBIC values compared to the 2-class solution and significant VLMR and BLRT values compared to the 4-class solution. Average levels of depressive symptoms can be found in Figure 2 . We classified the three classes as high-decreasing (n = 40; 2.4%), moderate (n = 183; 11.2%), and low (n = 1,417; 86.4%).

For the LX subgroup, a two-class solution was optimal based on lower AIC, BIC, and aBIC values compared to the 1-class solution and significant VLMR and BLRT values compared to the 3-class solution. Average levels of depressive symptoms can be found in Figure 3 . We classified the two classes as high-decreasing (n = 140; 6.1%) and low (n = 2143; 93.9%).

Associations between Grim-PGS, adverse life events, and parental acceptance and depressive symptom trajectories can be found in Tables 3 – 5 . Within the EA subgroup (see Table 3 ), adverse life events were associated with greater likelihood of membership in the high-decreasing, increasing, and moderate depression trajectories compared to the low depression trajectory. The Grim-PGS by adverse life events interaction was associated with greater likelihood of being in the moderate vs. low trajectory. Follow up analyses indicated that the association between Grim-PRS and membership in the moderate vs. low trajectory was stronger among youth who reported experiencing greater adverse life events (+.5SD; OR = 1.35, 95% CI [1.05, 1.74]) than among youth who experienced fewer adverse life events (-.5SD, OR = .90, 95% CI [.48, 1.71]). Parental acceptance was associated with lower likelihood of following the high, increasing, and moderate trajectories compared to the low trajectory. There were no significant interaction effects between Grim-PGS and parental acceptance in predicting trajectories of depressive symptoms.

Within the AA subgroup (see Table 4 ), there were no significant main effects of adverse life events or parental acceptance on depression trajectories. However, similar to the finding among EA youth, the Grim-PGS by adverse life event interaction was also associated with greater likelihood of being in the moderate vs. low trajectory. Specifically, the association between Grim-PRS and membership in the moderate vs. low trajectory was stronger among youth who reported experiencing more adverse life events (+.5SD; OR = 1.44, 95% CI [1.05, 1.90]) than among youth who experienced fewer adverse life events (-.5SD; OR = 1.38, 95% CI [.27, 7.17]).

Among LX youth (see Table 5 ), parental acceptance was associated with lower likelihood of membership in the high trajectory of depressive symptoms relative to the low trajectory. No main effect of adverse life events nor GXE effects were detected.