Effects of adverse childhood experiences on behavior in adolescence: Insights from brain imaging and genetic risk

Abstract

Adolescence is a critical developmental stage, and adverse childhood experiences (ACEs) are important environmental factors influencing behavior. However, the relationship between ACEs, individual behavior, brain mechanisms, and the role of genetic risk remains underexplored. In this study, data from 9796 youths aged 9–13 years in the Adolescent Brain Cognitive Development (ABCD) study were analyzed, including 67 ACE-related variables and 30 behavioral outcomes. Using sparse canonical correlation analysis, we identified two distinct patterns: a parental mental health–child psychopathology, linking parental psychiatric status to child mental health (r = 0.576, P_perm < 0.001) and a socioeconomic–cognitive development, in which higher family socioeconomic status and a more supportive family atmosphere correlated with better child cognition (r = -0.272, P_perm < 0.008). Neuroimaging revealed these associations were mediated by structural alterations in brain regions including the temporal and parietal cortex, hippocampus, and cerebellar cortex, along with changes in white matter tracts and functional networks. Genetic risk for psychiatric disorders significantly moderated these ACEs–brain–behavior pathways, with effects varying across ancestry groups. Longitudinal cross-lagged panel models further demonstrated that parental psychiatric conditions, particularly depression and attention deficit hyperactivity disorder, predicted child psychiatric symptoms two years later. Our findings delineate distinct ACEs-related pathways influencing adolescent behavior through neurostructural and functional mediators, moderated by genetic risk. This integrated framework advances the understanding of ACEs mechanisms and highlights targets for early intervention.

Highlights

• •Sparse canonical correlation analysis of a large adolescent cohort (ABCD, n = 9796) revealed two distinct ACEs–behavior covariation patterns: one linking parental mental health to child psychopathology, and another connecting family socioeconomic status to child cognitive function.
• •Multi-sCCA identified underlying neuroimaging phenotypes that mediate ACEs–behavior pathways, involving structural and functional alterations in key brain regions (e.g., hippocampus, cerebellum) and related networks. These brain-mediated pathways were further moderated by psychiatric polygenic risk, with ancestry-dependent variation.
• •Longitudinal cross-lagged penal models demonstrated that parental psychiatric conditions, particularly depression and ADHD, prospectively predicted child psychiatric symptoms two years later.
• •The findings advocate for integrated intervention strategies that jointly address genetic susceptibility and adverse environmental exposures.

1. Introduction

Early-life exposure to adverse experiences has profound and long-lasting effects on physical and mental health (Hughes et al., 2017, Kim and Royle, 2024, Luby et al., 2017, Norman et al., 2012). The adverse childhood experiences (ACEs) instrument is a widely utilized tool for assessing cumulative exposure to various negative life events or circumstances (e.g., childhood abuse, household dysfunction) (Merrick, Ford, Ports, and Guinn, 2018). The accessibility of this approach has contributed to increased recognition of the pernicious and enduring effects of early life adversity, particularly in relation to mental health (Copeland et al., 2018, Grummitt et al., 2022, Köhler-Forsberg et al., 2024, Reiss, 2013) and cognitive functioning (Copeland et al., 2018, Lupien et al., 2009, Osler et al., 2013, Sheridan et al., 2017, Wade et al., 2019). Notably, cumulative ACEs scores often demonstrate stronger predictive validity for mental health and cognitive outcomes than individual indicators, supporting the utility of a composite approach (Felitti et al., 1998, Hughes et al., 2017). However, reliance on cumulative indices alone may obscure the distinct mechanisms through which different forms of adversity influence behavioral outcomes (LaNoue, George, Helitzer, and Keith, 2020).

Childhood represents a critical period for psychological and behavioral development. Numerous studies have focused on the mechanisms through which childhood adversity experiences may be biologically embedded in the developmental processes, ultimately leading to negative emotional and health outcomes. More specifically, various ACEs exposures, including abuse, mental illness in household members, emotional neglect and bullying have been linked to depression and anxiety (Kim and Royle, 2024, Köhler-Forsberg et al., 2024, Madigan et al., 2023). Evidence also suggests that deprivation, characterized by factors such as low parental education and child neglect, is linked to poor executive function (Sheridan et al., 2017). Cumulative childhood trauma and persistent low SES are jointly associated with both heightened mental health problems (anxiety, depression, substance use) and cognitive impairment (Copeland et al., 2018, Osler et al., 2013, Reiss, 2013). Thus, different types of adversity may differentially influence emotional versus cognitive domains.

To better account for this heterogeneity, contemporary theoretical frameworks distinguish between cumulative risk models and dimensional models of adversity, particularly the threat–deprivation framework (Evans et al., 2013, McLaughlin et al., 2014, Sheridan and McLaughlin, 2014). Threat-related experiences, such as violence exposure, maltreatment, and severe family conflict, have been hypothesized to preferentially affective and stress–response systems, whereas deprivation-related experiences, such as socioeconomic disadvantage, neglect, and reduced cognitive stimulation, may be more closely linked to cognitive and language development. Consistent with dimensional models of adversity, threat-related experiences have been associated with internalizing and externalizing psychopathology in children and adolescents. For example, Schafer et al. found that threat, but not deprivation, was concurrently associated with both internalizing and externalizing symptoms in a large youth cohort (Schäfer et al., 2023). However, the distinction between threat and deprivation is relative rather than exclusive. Deprivation has also been linked to psychopathology through potentially distinct developmental mechanisms; for example, deprivation was prospectively associated with externalizing problems through lower verbal abilities (Miller et al., 2018). Caregiver psychopathology, an important form of family stress, has likewise been consistently associated with internalizing and externalizing problems in children (Connell and Goodman, 2002). Thus, although threat-related and family-stress experiences may show particularly direct associations with emotional and behavioral symptoms, deprivation and socioeconomic disadvantage may also contribute to psychopathology in addition to their established associations with cognitive and language development. Guided by these frameworks, the present study selected ACE indicators spanning caregiver psychopathology, family environment, maltreatment/violence-related experiences, neighborhood safety, socioeconomic disadvantage, and physical trauma exposure. This design allowed us to sample multiple theoretically relevant domains of adversity while using data-driven multivariate models to identify dominant ACE–behavior patterns.

A growing body of research has explored the neurobiological mechanisms underlying the strong association between early adversity and individual behaviors (Rakesh et al., 2021, Sheridan et al., 2017). For instance, family-based maltreatment has been correlated with larger volumes in fronto-limbic regions, while socioeconomic disadvantage has been linked to smaller volumes in temporal-limbic regions during childhood (Vannucci et al., 2023). Notably, smaller volumes in the inferior frontal gyrus and impaired emotional awareness have been identified as sequential mediators of the relationship between early ACEs, poor general health and increased depression severity during adolescence (Rakesh et al., 2021). Parenting-related factors, including caregiver psychopathology, maltreatment, family conflict, and disrupted caregiving, represent proximal family-context influences related to stress regulation, and emotion-related neurodevelopment, partly overlapping with threat-related adversity and fronto-limbic circuitry (McLaughlin et al., 2014).In contrast, SES-related factors, such as household income, parental education, and neighborhood disadvantage, reflect broader resource-related contexts that shape cognitive stimulation, language exposure, and educational opportunity, aligning more closely with deprivation-related pathways and frontoparietal-temporal systems (Sheridan and McLaughlin, 2014, Vogel et al., 2021). This distinction motivated our multivariate approach to identify separable ACE-related behavioral and neurobiological profiles. Although previous neuroimaging studies have identified associations between adversity and individual brain regions or imaging measures, relatively few have investigated distributed patterns spanning multiple imaging modalities or examined such patterns as potential mediators of adversity-behavior associations. The present study therefore uses a multivariate neuroimaging framework to identify latent components across structural MRI, diffusion MRI, and resting-state functional connectivity and to determine whether these distributed neuroimaging patterns partly mediate the associations between ACE profiles and behavioral outcomes.

Previous studies have demonstrated interactions between exposure to ACEs and genetic risk for mental disorders in shaping individual mental health outcomes (Crouse et al., 2024, Kwong et al., 2021). These findings are consistent with gene–environment interaction frameworks, including diathesis-stress and differential susceptibility models, which propose that genetic liability may alter individuals’ sensitivity to adverse environments (Belsky and Pluess, 2009, Caspi et al., 2003). Empirical evidence further suggests that higher psychiatric polygenic risk scores (PRS) are associated with greater vulnerability to adversity-related psychopathology (Kwong et al., 2021, Peyrot et al., 2018). In the present study, we examined PRS for major psychiatric and neurodevelopmental disorders because they index genetic liability relevant to the behavioral domains assessed here, including internalizing, externalizing problems and psychosis-related symptoms. Previous studies of genetic liability and childhood adversity have primarily focused on behavioral or psychiatric outcomes, with comparatively limited investigation of whether genetic liability modifies the neurobiological pathways linking adversity to behavior. Integrating PRS into the present framework allows us to examine whether genetic liability modulates the underlying neurobiological pathways linking ACEs to behavioral outcomes. This approach provides a more comprehensive understanding of how genetic and environmental factors jointly shape individual differences in mental health and cognition.

Although cross-sectional studies have established robust associations of ACEs and behavioral outcomes, evidence regarding their temporal directionality remains limited. Prior longitudinal work has reported domain-specific or bidirectional temporal associations among related constructs, including externalizing symptoms predicting later family conflict (Aaron et al., 2026, Neville et al., 2025), socioeconomic disadvantage predicting lower cognitive abilities (Rosen et al., 2020), and child externalizing problems contributing to increases in parental depressive symptoms over time (Wiggins, Mitchell, Stringaris, and Leibenluft, 2014). However, these findings are primarily based on single-domain exposures or outcomes, and less is known about the temporal dynamics of ACEs in the context of multivariate, data-driven adversity profiles. This motivates the use of cross-lagged panel models (CLPM) to further examine the temporal stability and directional consistency of ACE–behavior associations in the present study.

However, adversity itself exhibits substantial heterogeneity, further complicated by its co-occurrence and chronicity (Gee, 2021), making it challenging to accurately capture the variability within ACEs. This complexity hinders a comprehensive understanding of how, why, and when ACEs impact brain development and behavior across the lifespan. Research on ACEs has predominantly focused on isolated adversities, often overlooking their interrelated effects. Most studies have examined a single or a limited number of exposures (Lupien et al., 2009, Tomasi and Volkow, 2021, Vannucci et al., 2023), frequently relying on simple sum scores or cumulative frequency measures to assess cumulative effects (Ansell et al., 2012, Gee, 2021, Hughes et al., 2017, Köhler-Forsberg et al., 2024, Luby et al., 2017, Thorp et al., 2023). However, such approaches risk obscuring the unique influence of different childhood adversity on mental health and cognitive development. Given the richness and complexity of ACEs data, data-driven methodologies provide a promising avenue to develop conceptual models that offer a more comprehensive understanding of these intricate relationships.

The present study has three interconnected aims. First, we used sparse canonical correlation analysis (sCCA) and multi-modal sCCA (msCCA) to identify multivariate ACEs–behavior profiles and to characterize the associated neuroimaging patterns across structural, diffusion, and functional MRI. Second, we examined whether psychiatric PRS moderates the identified ACEs–brain–behavior pathways. Third, to assess the temporal robustness of key cross-sectional associations, we conducted targeted longitudinal analysis using representative ACE and behavioral indicators selected based on their prominence in the multivariate results. CLPM was then applied to examine temporal stability and potential directional associations between ACE indicators and behavioral outcomes.

Guided by the theoretical framework outlined above, we formulated four hypotheses: First, adversity-related profiles would show differential associations with behavioral outcomes, such that threat- or family-stress-related profiles would be more strongly associated with internalizing and externalizing symptoms, whereas deprivation- or socioeconomic disadvantage-related profiles would be more strongly associated with cognitive performance. Second, distributed neuroimaging components derived from structural MRI, diffusion MRI, and resting-state functional connectivity would partially mediate ACE–behavior associations, reflecting distributed neurodevelopmental pathways, without a priori predictions regarding the specific components involved. Third, psychiatric PRS would moderate ACE–brain–behavior pathways. Given the heterogeneity and limited specificity of prior gene–environment interaction evidence, we did not make a priori predictions regarding which specific PRS would moderate particular pathways or the direction of moderation; these differences were examined exploratorily. Finally, longitudinal analyses would provide evidence for temporal consistency and potential directional or bidirectional associations between representative ACE indicators and behavioral outcomes. A schematic summary of the proposed framework is presented in Fig. 1.

Fig. 1.

Study design overview. The study included 9796 participants to identified ACEs profiles associated with distinct behavioral groups using sCCA within a train–test dataset split design. Subsequently, msCCA was employed to analyze the relationships among ACEs profiles, neuroimaging measures and behavioral groups also within a train–test dataset split design. Mediation analysis was conducted to assess whether neuroimaging components mediated the relationships between ACEs profiles and behavioral outcomes. Additionally, PRS for mental disorders, including MDD, ADHD, SCZ, BIP, and ASD, were calculated to examine their moderating effects on neuroimaging components linked to ACEs profiles and behavioral groups. Finally, a cross-lagged panel model was applied to investigate the longitudinal associations between adverse childhood experiences and behavioral outcomes. Abbreviation: ACEs, adverse childhood experiences; ADHD, attention-deficit/hyperactivity disorder; ASD, autism spectrum disorder; BIP, bipolar disorder; CLPM, cross-lagged panel model; GWAS, genome-wide association studies; MDD, major depressive disorder; msCCA, multi-modal sCCA; PRS, polygenic risk scores; sCCA, sparse canonical correlation analysis; SCZ, schizophrenia.

2. Materials and Methods

2.1. Participants

The dataset was selected from the Annual Curated Data Release 5.1 from the ABCD consortium (https://abcdstudy.org/index.html) which includes over 11,000 children aged 9–11 years, recruited from 22 sites across the United States with a diverse range of geographic, socioeconomic, ethnic, and health backgrounds (Hagler et al., 2019). The study protocols were approved by the Institutional Review Boards (IRBs) at each participating site, with written informed consent obtained from youth and caregivers. Comprehensive details about the participant cohort, data collection procedures, and preprocessing protocols are provided at the ABCD website (https://abcdstudy.org/scientists/protocols/) and are described elsewhere (Barch et al., 2021, Hagler et al., 2019).

2.2. Adverse Childhood Experiences and Behavioral Measures

ACEs variables were identified from the range of ABCD baseline measures (Barch et al., 2021, Brieant et al., 2023, Hoffman et al., 2019). A total of 139 potential item-level variables were selected based on their relevance to ACEs constructs. To enhance conceptual clarity, the selection and organization of these variables were guided by established ACEs frameworks, including the Adverse Childhood Experiences framework (Felitti et al., 1998) and related extensions that incorporate broader contextual and environmental stressors. Specifically, these variables were mapped onto several core adversity domains, including caregiving disruption, caregiver psychopathology, maltreatment, neighborhood safety/violence, family/community support, socioeconomic disadvantage, and physical trauma exposure. Given the potential conceptual overlap among certain variables (e.g., internalizing and depression-related symptoms), these measures were treated as interrelated components within broader adversity domains rather than as independent indicators. These variables were reported by children, caregivers, or assessed and derived by trained research personnel. The variables utilized in the current study were derived from validated scales that are widely used in the literature, though some of these scales were modified by the ABCD Study (Barch et al., 2021). The final variables included binary, polytomous, and continuous variables. To facilitate interpretation of results, all variables were coded such that higher scores reflected greater adversity, with the exception of parental highest educational level and total household income, which higher values represented higher socioeconomic advantage.

We analyzed data from the selected individual behavioral assessments, yielding 10 cognitive and 20 mental health measures. Cognitive measures included vocabulary, attention, working memory, executive function, processing speed, episodic memory, reading, fluid cognition, crystallized cognition, overall cognition. The mental health measures encompassed anxious depressed, withdrawn depressed, somatic complaints, social problems, thought problems, attention problems, rule-breaking behavior, aggressive behavior, internalizing problem, externalizing problem, total psychosis symptoms, depressive problem, anxiety problem, somatic problem, ADHD problem, oppositional defiant problem, conduct problem, sluggish cognitive problem, obsessive-compulsive problem, and stress problem. Higher scores on mental health measures indicated greater dimensional psychopathology, while higher scores on cognitive measures reflected better cognitive performance. A detailed description of all ACEs and behavioral variables is provided in Supplementary Material [Table S1].

For ACE-related variables, an initial quality control procedure was applied prior to analysis. Variables with a missing rate exceeding 50% were excluded, resulting in the removal of 31 variables. No variables were excluded based on low endorsement rates (< 0.05%). To further eliminate variables with extremely skewed distributions, the nearZeroVar function from the caret R package was applied (Kuhn et al., 2020), leading to the exclusion of an additional 29 variables. Two variables with redundant content were manually removed. Following these steps, 77 ACEs variables were retained for subsequent processing. Continuous environmental variables were screened for outliers, defined as values exceeding four times the median absolute deviation (MAD). Identified outliers were treated as missing values. To minimize multicollinearity arising from familial relatedness, twins, triplets, and—when applicable—one sibling per family were excluded (n = 2054). After excluding related individuals, 10 ACEs variables exhibited missing rates exceeding 50% and were therefore removed, yielding a final set of 67 ACEs variables for analysis.

For individual behavioral measures, all variables were retained, as none exceeded the 50% missingness threshold and no variables were excluded by the nearZeroVar function. Additionally, eight participants lacking age information were excluded from the analysis.

An overview of the variable selection and data-cleaning procedures is presented in Fig. 2(A), Supplementary Material [Figure S1] and Supplementary Material [Table S2]. Missing values were addressed using multiple imputation implemented in the mice R package (v3.16.0) (Van Buuren and Groothuis-Oudshoorn, 2011). Imputation was performed using predictive mean matching for continuous variables and logistic regression for binary and categorical variables, with 10 iterations generated for each imputed dataset.

Fig. 2.

Distinct adverse childhood experiences profiles correlated with specific individual behavioral groups. (A) The total number of variables included for adverse childhood experiences exposures and the final subset of variables included for analysis. (B) The covariance and variance explained by the canonical variates. Two canonical models were statistically significant at an FDR-corrected P value < 0.05 (*). (C & E) The correlation between ACEs profiles and individual behavioral groups for the first significant canonical mode (red) and the same for the second significant mode (blue). Results are presented for both the discovery dataset (dark points) and the test dataset (light points). (D & F) The contributing variables to the ACEs profiles (bar plots) and behavioral groups (radar plots). Positive weights are indicated in orange, while negative weights are shown in green, reflecting the directionality of the associations. The details of the variables are provided in Supplementary Material [Table S1].

2.3. Neuroimaging Measures

Preprocessed structural MRI (sMRI), diffusion MRI (dMRI), and resting-state functional MRI (rsfMRI) data were obtained from the ABCD Study, with all preprocessing procedures performed centrally by the ABCD consortium using standardized pipelines, as described in detail elsewhere (Hagler et al., 2019) and in the Supplementary Methods. Neuroimaging measures included 88 regional brain volume metrics (68 cortical regions and 20 subcortical regions), 37 regional fractional anisotropy (FA) measures derived from dMRI, and 78 between-network functional connectivity metrics based on the 13 Gordon networks. Details of these metrics are provided in Supplementary Material [Table S3].

Although ABCD Release 5.1 includes repeated neuroimaging assessments, the present study focused on baseline neuroimaging measures because baseline provided the largest and most complete harmonized dataset for high-dimensional multivariate analyses. At baseline, imaging data were available for 11,067 participants. Quality control of processed sMRI data, dMRI-derived FA metrics, and rsfMRI-derived functional connectivity networks (FCNs) were conducted by the ABCD Study. In total, 873 participants failed quality control for sMRI data, 1291 for dMRI data, and 4314 for rsfMRI data (see Supplementary Methods for details). After quality control, the remaining participants were merged with those included in the sCCA analysis. The final analytic samples comprised 8953 participants with sMRI morphometric data, 8082 participants with dMRI-derived FA data, and 5622 participants with rsfMRI-derived FCN data, all of whom also had available ACEs profiles and behavioral measures. Demographic characteristics of these participants are summarized in Table 1.

Table 1.

Demographics of participants included in the specific statistical analysis.

Demographics	sCCA (N = 9796)	msCCA (sMRI) (N = 8953)	msCCA (dMRI) (N = 8082)	msCCA (rsfMRI) (N = 5622)	CLPM (N = 7817)
Sex (%Male)	5159 (52.6)	4688 (52.3)	4204 (52.0)	2776 (49.0)	4139 (52.9)
Race (%White)	4957 (50.7)	4547 (50.7)	4173 (51.6)	3076 (54.3)	4209 (53.8)
Age (years), Mean (SD)	9.47 (0.51)	9.48 (0.51)	9.48 (0.51)	9.51 (0.51)	9.47 (0.51)

Abbreviation: CLPM, cross-lagged panel model; msCCA, multi-modal sCCA; sCCA, canonical correlation analysis.

2.4. Potential Confounders

To minimize the effects of potential confounders, covariate adjustment was performed before the main multivariate analyses. Specifically, each variable of interest was regressed on relevant covariates using linear regression models, and the resulting residuals were retained for subsequent analyses. Continuous covariates included age, body mass index, and puberty score, whereas categorical covariates, including sex, race, and handedness, were dummy-coded before inclusion in the regression models. When multiple imputation was applied, residuals were derived within each imputed dataset and then averaged across imputed datasets to obtain a single dataset for downstream analyses. In analysis involving regional gray matter volume (GMV), total intracranial volume was additionally regressed out. For diffusion metrics, the mean framewise displacement computed from the DTI data was regressed out for FA, whereas for functional metrics, the mean framewise displacement computed from the fMRI data was regressed out for FCNs.

2.5. Statistical analysis

2.5.1. Identification of Multivariate Relationships between Adverse Childhood Experiences and Individual Behaviors

To identify multivariate associations between ACEs and individual behavioral outcomes, sCCA was performed in R using the sgcca.wrapper function from the mixOmics package (Modabbernia et al., 2021). The analysis was performed as follows: 1) dataset division: The original dataset was randomly split into a discovery subsample (80% of the original dataset) and a test subsample (20% of the original dataset) to ensure sufficient statistical power for model estimation while retaining an independent subset for evaluating generalizability; 2) normalization and parameter selection: Prior to analysis, ACEs exposure variables and behavioral variables were z-normalized. Optimal sparsity parameters were selected using ten-fold cross-validation by systematically testing a range of candidate values spanning from 1/√p to 1 (in ten evenly spaced increments), where p denotes the number of variables in each dataset. Models were fitted across candidate parameter combinations, and the optimal configuration was selected based on maximal canonical correlation. 3) application of sCCA: Using the optimized parameters, sCCA was applied to estimate canonical correlations (r-values) between ACEs profiles and behavioral groups. Canonical weights were obtained for both ACEs and behavioral variables, and canonical loadings were additionally calculated to quantify the contribution of each variable to the identified canonical modes in both the discovery and test datasets; 4) significance testing: Statistical significance of the sCCA models was evaluated using a permutation-based framework. Specifically, 1000 random permutations were generated within the discovery dataset, and sCCA was re-estimated for each permutation. Canonical weights derived from the permuted discovery sets were subsequently projected onto the unpermuted test dataset to obtain null distributions of canonical correlations. Permutation procedures are illustrated in Supplementary Material [Figure S2]. One-tailed permutation P values (P_perm) were used to assess statistical significance. To account for multiple comparisons across canonical modes, false discovery rate (FDR) correction was applied using the Benjamini-Hochberg approach (Benjamini and Hochberg, 1995), with an FDR-adjusted threshold of P_FDR < 0.05.

2.5.2. Reliability and Stability Assessments

To evaluate the reliability and reproducibility of the sCCA findings, multiple complementary validation analyses were conducted, including: 1) stability analysis across varying sample sizes; 2) random resampling validation; 3) a leave-one-site-out cross-validation framework; 4) assessment of sex bias; 5) impact of data imputation. The details of the analyses are described in the Supplementary Methods.

2.5.3. Neuroimaging Correlates of ACEs-Behavioral Patterns

To explore the neurobiological correlates underlying the associations between ACEs exposure and individual behaviors, we performed msCCA using the mixOmics R package. The analysis included ACEs profiles, neuroimaging measures, and individual behavioral groups. To maximize statistical power and capture modality-specific associations, each neuroimaging modality was analyzed separately, including sMRI of regional brain volumes, dMRI-derived FA, and rsfMRI-derived FCNs. Similarly, the dataset was randomly divided into a discovery subset comprising 80% of the sample and an independent test subset comprising the remaining 20%. Statistical significance was evaluated using permutation testing with 1000 iterations. For each imaging modality, canonical variates representing neuroimaging components (hereafter referred to as brain components) were derived. Canonical correlations (r) were computed between ACEs profiles and brain components, as well as between brain components and individual behavioral groups. In addition, canonical weights for each neuroimaging variable were extracted to quantify their relative contributions to the identified brain components.

2.5.4. Polygenic Risk Scores Calculation

Polygenic risk scores (PRS) were calculated to assess the genetic susceptibility to five major psychiatric disorders: major depressive disorder (MDD) (Als et al., 2023), attention deficit hyperactivity disorder (ADHD) (Demontis et al., 2023), schizophrenia (SCZ) (Trubetskoy et al., 2022), bipolar disorder (BIP) (Mullins et al., 2021) and autism spectrum disorder (ASD) (Grove et al., 2019). Summary statistics from large-scale GWAS in European ancestry populations were used as base datasets for PRS derivation. PRS were computed using PRSice-2 software (version 2.3.5) (Choi and O'Reilly, 2019) with clumping and thresholding parameters, and multiple significance thresholds were applied. To minimize population stratification, primary analyses were conducted in participants of European ancestry, followed by trans-ancestry validation to assess generalizability. Detailed quality control procedures and PRS computation steps are provided in the Supplementary Methods.

2.5.5. Mediation and Moderated Mediation Analyses

Mediation analyses were conducted to explore the potential pathway linking ACEs profiles, brain components, and individual behavioral outcomes, testing whether the associations between ACEs exposure and behaviors were mediated by alterations in brain structure and function. Model-based mediation analyses were performed using a quasi-Bayesian Monte Carlo simulation with 1000 iterations implemented in the R package mediation (version 4.5.0) (Tingley, Yamamoto, Hirose, Keele, and Imai, 2014). Canonical variates derived from each analysis were used to represent the corresponding ACEs, brain, and behavioral components. To control for multiple comparisons, FDR correction was applied, with statistical significance defined as an adjusted P < 0.05.

To further investigate whether the mediation effects of brain components were moderated by genetic susceptibility, we conducted moderated mediation analyses in participants of mixed and European ancestry using Model 59 in PROCESS for R (version 4.3) (Hayes, 2017). Specifically, we tested whether the association between ACEs profiles (independent variable, X) and individual behaviors (dependent variable, Y) was mediated by brain components (mediator, M) and moderated by PRS for specific psychiatric disorder (moderator, W). The moderator (disorder-specific PRS) was categorized into low, medium, and high genetic risk groups based on the 16th, 50th, and 84th percentiles, following the default settings of the PROCESS for R package. Bias-corrected bootstrapping method with 1000 iterations was applied to estimate the indirect effects and their 95% confidence intervals (CIs).

To control for multiple comparisons, FDR correction was applied to the p-values of all indirect effects in both the mediation and moderated mediation analyses. The correction was performed separately within each set of tests using the Benjamini–Hochberg procedure.

2.5.6. Longitudinal Data Analysis

To examine the longitudinal associations between ACEs exposure and behavioral outcomes, a longitudinal analysis was conducted in the 7817 participants with available follow-up data after merging with the baseline sCCA sample. Variables considered for the longitudinal analyses were prioritized based on their canonical weights in the primary multivariate analyses, which reflect each variable’s contribution to the construction of the corresponding canonical variate. Variables with relatively large absolute canonical weights were considered as candidates. Among these, we further selected variables that had clear theoretical interpretability and comparable measures available at both assessment points. ACEs variables (including parental depression, ADHD, household income) and individual behavioral variables (including total psychiatric problems, crystallized composite score) were obtained at the 2-year follow-up.

A classic two-wave cross-lagged panel model (CLPM) was implemented using structural equation modeling in Mplus (version 8.3) (Muthen et al., 2003), leveraging longitudinal data from the ABCD Study (Ronald C Kessler, 2014). Model fit was evaluated using the Tucker–Lewis Index (TLI), Comparative Fit Index (CFI), and root mean square error of approximation (RMSEA). TLI and CFI values greater than 0.90, along with RMSEA values close to or below 0.05, were considered indicative of acceptable model fit (Bentler, 1990, Hu and Bentler, 1999). Relevant confounding variables were included in the CLPM to obtain robust and unbiased estimates. Missing data was handled using the Mplus missing value code (999).

3. Results

3.1. Demographics

A total of 9796 participants were included in the analysis, comprising 5159 males (52.6%) and 4637 females (47.4%). The mean age of the participant was 9.47 years (standard deviation [SD] = 0.51; range = 9–11 years). Participants were identified as Asian (2.3%), Black (15.2%), Hispanic (21.3%), White (50.7%), and other races (10.5%). Detailed demographic characteristics of the participants are summarized in Table 1 and Supplementary Material [Figure S3(A–C)].

3.2. Correlation of Adverse Childhood Experiences with Individual Behaviors

Model 1 (parental mental health–child psychopathology) demonstrated a significant association between parental mental health and children’s mental health in both the discovery dataset (r = 0.603, P_perm < 0.001; covariance explained = 18.314%) and the test dataset (r = 0.576, P_perm < 0.001; covariance explained = 17.328%) (Fig. 2(B–C)). The canonical weights for the ACEs variables and individual behavioral variables are presented in Fig. 2(D) and Supplementary Material [Table S6]. Children exposed to greater parental mental health problems (e.g., ADHD, depression, hyperactivity, anxiety) exhibited more pronounced mental health difficulties, including higher total psychosis symptom scores.

Model 2 (socioeconomic–cognitive development) identified a significant association between SES and cognition in both the discovery dataset (r = -0.247, P_perm < 0.001; covariance explained = 8.397%) and the test dataset (r = -0.272, P_perm < 0.008; covariance explained = 5.552%) (Fig. 2(B, E)). The canonical weights for the ACEs variables and individual behavioral variables are shown in Fig. 2(F) and Supplementary Material [Table S6]. Children from families with higher SES (e.g., higher parental education levels, greater household income, and lower neighborhood deprivation) exhibited better cognitive performance, including higher scores in total composite cognition, crystallized cognition, and reading skills.

The canonical loadings and cross-loadings of ACEs variables and individual behavioral variables are shown in Supplementary Material [Figure S4(A–D)].

3.3. Reliability and Stability Assessments

We assessed the reliability and stability of the sCCA results using multiple complementary approaches. Across sensitivity analyses, the two primary ACE–behavior modes showed consistent canonical correlation patterns and similar variable-loading structures, supporting the stability of the main findings. These results indicate that the identified ACE–behavior profiles were unlikely to be driven by a particular subsample, site-specific effects, or imputation procedures. In sex-stratified analyses, the overall patterns of canonical weights and loadings were highly consistent between males and females (all r > 0.9). However, both modes were significant in males, whereas only Model 1 remained significant in females, whereas Model 2 did not reach significance. This suggests that the structure of the ACE–behavior patterns was largely preserved across sexes, although the strength or statistical detectability of the socioeconomic–cognitive development mode may vary by sex. Detailed results are provided in the Supplementary Results.

3.4. Neuroimaging Correlations Underlying ACEs–Behavioral patterns

To investigate neuroimaging correlations underlying ACEs profiles and individual behaviors, we conducted an msCCA incorporating ACEs profiles, neuroimaging measures, and behavioral groups. We identified 9 regional brain volumes significantly associated with the parental mental health group (Model 1), showing significant correlations with the first ACEs profiles (r_discovery = 0.093, r_test = 0.099, P_perm < 0.002), and the mental health group (r_discovery = 0.044, r_test = 0.073, P_perm < 0.002) (Supplementary Material [Figure S7(A)]). These regions included the right superior and middle temporal gyri, precuneus, precentral gyrus, the left superior parietal gyrus, and the bilateral postcentral gyri and supramarginal gyri (Fig. 3(A); Supplementary Material [Table S8]).

Fig. 3.

Neuroimaging correlations underlying ACEs profiles and individual behavioral groups. (A) The regional brain volumes associated with parental mental health group (Model 1). (B) The regional brain volumes associated with socioeconomic cognitive development group (Model 2). (C) The intra- and inter- functional connectivity networks associated with parental mental health group (Model 1). The different colors represent different functional connectivity networks. (D) White matter FA associated with parental mental health group (top) and the socioeconomic cognitive development group (bottom). Abbreviation: atr, anterior thalamic radiations; cc, corpus callosum; cgc, cingulate cingulum; cgc, cingulo_opercular network; cgh, parahippocampal cingulum; cst, corticospinal tract (pyramidal tract); dla, dorsal_attention network; dt, default network; fmaj, forceps major; fmin, forceps minor; fo, fronto_parietal network; fscs, superior corticostriate-frontal cortex only; fxcut, fornix, excluding fimbria; fx, fornix; ifo, inferior frontal occipital fasciculus; ifsfc, inferior frontal to superior frontal cortical tract; ilf, inferior longitudinal fasciculus; lh, left hemisphere; pscs, parietal superior corticostriate; pslf, parietal superior longitudinal fasciculus; rh, right hemisphere; rspltp, retrosplenial_temporal network; scs, superior corticostriate; sifc, striatal inferior frontal cortex tract; slf, superior longitudinal fasciculus; smh, sensorimotor_hand network; smm, sensorimotor_mouth network; tslf, temporal superior longitudinal fasciculus (arcuate fasciculus); unc, uncinate; vta, ventral_attention network.

We also identified 13 regional brain volumes significantly correlated with socioeconomic cognitive development group (Model 2), showing significant correlations with the second ACEs profiles (r_discovery = −0.287, r_test = −0.288, P_perm < 0.001) and the cognition group (r_discovery = 0.139, r_test = 0.190, P_perm < 0.001) (Supplementary Material [Figure S7(B)]). These regions included the right superior temporal gyrus, the left precentral gyrus, postcentral gyrus, hippocampus, and lateral orbitofrontal gyrus, as well as the bilateral cerebellar cortex, fusiform gyrus, and middle and inferior temporal gyri (Fig. 3(B); Supplementary Material [Table S8]).

For FCNs, significant intra- and inter- network connectivity was observed between the second ACEs profiles (r_discovery = −0.112, r_test = −0.091, P_perm < 0.002) and the mental health group (r_discovery = −0.102, r_test = −0.058, P_perm < 0.035) (Supplementary Material [Figure S7(C)]). These findings involved connectivity between the default network and the dorsal attention network, between the sensorimotor hand network and the sensorimotor mouth network, between the dorsal attention network and the ventral attention network, between the cingulo-opercular network and the default network. Additionally, significant inter-network connectivity was observed within the dorsal attention network and the default network (Fig. 3(C); Supplementary Material [Table S9]). However, no significant results were observed for Model 2.

For FA, we identified several white matter tracts significantly associated with the first ACEs profiles (r_discovery = −0.060, r_test = −0.059, P_perm < 0.016) and the mental health group (r_discovery = −0.055, r_test = −0.065, P_perm < 0.009) (Supplementary Material [Figure S7(D)]). These tracts included the left corticospinal (pyramidal) tract and bilateral superior corticostriate fibers, particularly those connecting the frontal and parietal cortices (Fig. 3(D); Supplementary Material [Table S10]).

We also observed significant associations between white matter tracts and the second ACEs profile (r_discovery = 0.142, r_test = 0.149, P_perm < 0.001), as well as the cognitive group (r_discovery = −0.109, r_test = −0.074, P_perm < 0.003) (Supplementary Material [Figure S7(E)]). The implicated tracts included the left inferior longitudinal fasciculus and inferior fronto-occipital fasciculus, along with the bilateral cingulum (parahippocampal) and superior longitudinal fasciculus, particularly involving parietal and temporal cortices (Fig. 3(D); Supplementary Material [Table S10]).

Overall, the parental mental health–child psychopathology profile was associated with lower regional brain volumes in selected cortical regions, higher FA in specific white matter tracts, lower intra- and inter-network functional connectivity, and worse mental health outcomes. In contrast, the socioeconomic–cognitive development profile was associated with larger regional brain volumes, lower FA in specific association pathways, and better cognitive performance.

3.5. Mediation and Moderated Mediation Analyses

To investigate the potential pathway linking ACEs profiles, brain components, and individual behaviors, mediation analyses were conducted using canonical variates as latent representations of the three variable sets (ACEs profiles → brain components → individual behavioral groups).

For regional brain volumes, brain components partially mediated the association between ACEs profiles and individual behavioral groups (Model 2: proportion mediated = 9.7%, β = −0.026, 95% CI = −0.032 to −0.020; Fig. 4(A)); Supplementary Material [Table S11], whereas no significant mediation effect was observed in Model 1. For FA, significant partial mediation effects were identified only in Model 2 (proportion mediated = 4.0%, β = −0.010, 95% CI = −0.013 to −0.007; Fig. 4(A); Supplementary Material [Table S11]). For FCNs, a significant partial mediation effect was observed in Model 1 (proportion mediated = 0.5%, β = 0.003, 95% CI = 0.001–0.005; Fig. 4(B); Supplementary Material [Table S11]), whereas Model 2 did not show a significant indirect effect.

Fig. 4.

Mediation and moderated mediation analysis for significant models. (A) Mediation role of regional FA and regional brain volumes in linking the ACEs profile to the behavioral group in socioeconomic cognitive development group (Model 2). (B) The mediation role of FCN in linking the ACEs profile to the behavioral group in parental mental health group (Model 1). (C) The moderated mediation role of ASD’ PRS in the ACEs profile → brain component → behavioral group pathway among individuals of European ancestry.

Moderated mediation analyses were subsequently performed for models exhibiting significant mediation effects. For regional brain volumes in Model 2, among individuals of European and mixed ancestry, all examined psychiatric PRS significantly moderated the indirect pathways linking ACEs profiles → brain component → behavioral outcomes, across different levels of genetic risk (e.g., ASD PRS as a moderator in the European ancestry group: proportion mediated = 7.8–10.5%; Fig. 4(C); Supplementary Material [Figure S8(A–B)] and Supplementary Material [Tables S12–S13]). These findings indicate statistically significant but modest PRS-related variation in the indirect pathway linking ACEs, brain structure, and behavioral outcomes. For FA in Model 2, all examined psychiatric PRS significantly moderated the indirect pathways across different genetic risk levels in both European and mixed ancestry groups (e.g., ASD PRS in European ancestry group: proportion mediated = 4.9–6.2%; Fig. 4(C); Supplementary Material [Figure S8(C–D)] and Supplementary Material [Tables S12–S13]). These results indicate that PRS-related moderation of FA-mediated pathways varied across ancestry groups and ACE–behavior modes. For FCNs in Model 1, among individuals of European ancestry, a subset of psychiatric PRS significantly moderated the indirect pathways at moderate levels of genetic risk (e.g., ASD PRS: proportion mediated = 0.5–0.6%; Fig. 4(C); Supplementary Material [Figure S8(E–F)] and Supplementary Material [Tables S12–S13]), with a greater number of significant moderation effects observed in individuals of mixed ancestry. This pattern suggests that genetic susceptibility may differentially relate to functional network pathways across levels of risk.

Overall, the moderated mediation results indicated that PRS-related variation in ACE–brain–behavior pathways was present but modest. These findings suggest that genetic liability may be associated with variation in statistical indirect effects, but the relatively small differences across genetic risk levels indicate that these results should be interpreted cautiously.

3.6. The Longitudinal Association between ACEs and Behavioral Outcomes

Structural equation modeling with CLPM was used to examine longitudinal associations between baseline and the 2-year follow-up. Results indicated that the parental psychiatric problems at baseline, including depression (β = 0.060, SE = 0.009, P < 1 ×10⁻⁴) and ADHD (β = 0.055, SE = 0.009, P < 1 ×10⁻⁴), were significantly associated with children’s total psychiatric problem score at the 2-year follow-up. Significant reciprocal associations were also observed, whereby children’s baseline psychiatric problems predicted parental depression (β = 0.106, SE = 0.010, P < 1 ×10⁻⁴) and ADHD symptoms (β = 0.078, SE = 0.009, P < 1 ×10⁻⁴) at follow-up (Fig. 5(A–B); Supplementary Material [Table S14]). Additionally, higher family income at baseline was significantly associated with higher crystallized composite scores at the 2-year follow-up (β = 0.082, SE = 0.011, P = 1 ×10⁻⁴), and a significant reverse cross-lagged association was also detected, with baseline crystallized cognition predicting family income at follow-up (β = 0.015, SE = 0.006, P < 0.009). However, the global model fit indices indicated suboptimal overall model fit (Supplementary Material [Figure S9] and Supplementary Material [Table S14]).

Fig. 5.

The longitudinal relationship between ACEs variables and individual behavioral variables. The longitudinal association between the parental psychiatric problems, such as depression (A) and ADHD (B), and Children’s psychiatric problems total score. Abbreviation: ADHD, attention-deficit/hyperactivity disorder.

4. Discussion

In this study, we analyzed data from 9796 participants from the ABCD Study, encompassing 67 adverse childhood experiences (ACEs) exposures and 30 individual behavioral variables, using a multivariate framework to investigate patterns of covariation. Two distinct ACEs–behavior covariation profiles were identified, along with neuroimaging alterations that partially mediated the associations between these ACEs profiles and specific behavioral groups. In addition, genetic risk for several psychiatric disorders was found to moderate multiple mediation pathways. Longitudinal analyses further supported the associations between ACEs exposure and subsequent behavioral outcomes. Together, these findings provide a comprehensive framework for understanding the complex interplay among ACEs exposure, candidate neurobiological mechanisms, and behavioral outcomes, shedding light on the biological pathways through which ACEs shape distinct behavioral profiles.

The first ACEs profile was predominantly characterized by parental psychopathology, including ADHD-related symptoms, depression, and hyperactivity/impulsivity, and was strongly associated with poorer mental health outcomes in children. This finding is consistent with prior research linking parental psychopathology to increased risk of mood and behavioral disorders in offspring (Cooke et al., 2021, Esteves et al., 2020; Kessler, Davis, and Kendler, 1997; Oladeji, Makanjuola, and Gureje, 2010). In the present study, we found that this association was related to neuroimaging measures, including reduced cortical volumes, higher FA in specific white matter tracts, and decreased functional connectivity. Previous studies have suggested that greater exposure to early-life adversity is associated with alterations in white matter development, although the direction and functional significance of these associations vary across tracts and developmental periods (Chahal, Kirshenbaum, Ho, Mastrovito, and Gotlib, 2021). Pollmann et al. found that greater family conflict was associated with higher overall FA levels across early adolescence (Pollmann, Sasso, Bates, and Fuhrmann, 2024). This finding provides relevant context for the higher FA associated with the parental mental health-child psychopathology profile, because both family conflict and parental psychopathology represent proximal family-context stressors. Compared with prior studies that have emphasized subcortical structures (Mannie et al., 2014, Mattoni et al., 2023, Pagliaccio et al., 2020), our results suggest a broader pattern of cortical involvement. At the functional level, alterations in large-scale brain networks, including the default mode network, the control network, and the salience network were linked to parental mental health group. These networks are essential for self-referential thinking, executive functioning, and emotion regulation, domains frequently disrupted in psychiatric disorders (Menon, 2011). Overall, these findings suggest that neurobiological variation may partly account for the association between parental psychopathology and offspring mental health. However, given the relatively small indirect effects in FCNs pathway and the cross-sectional nature of imaging data, these associations should be interpreted as correlational rather than causal.

The second ACEs profile was predominantly characterized by SES indicators, including family income, parental education, and derived residential index. Higher SES was positively correlated with better cognitive performance in children, consistent with previous findings (Gur et al., 2019, Hair et al., 2015, Osler et al., 2013, Tomasi and Volkow, 2021). In the present study, this association was partially mediated by temporal gyrus, precentral gyrus, fusiform gyrus, hippocampus, and cerebellum. These findings align with prior evidence linking SES to widespread differences in brain structure across cortical and subcortical regions (Noble et al., 2015), particularly those supporting language, memory, and executive function (Dehaene et al., 2010, Jednoróg et al., 2012, Romeo et al., 2018, Taylor et al., 2020). In addition, FA differences in frontoparietal pathways, particularly the superior longitudinal fasciculus, were also associated with SES-related differences in cognitive performance. Prior studies have suggested that these pathways support working memory and language-related processes, which may contribute to SES-related cognitive disparities (Schrempft et al., 2024, Ursache and Noble, 2016). White matter microstructure continues to develop throughout childhood and adolescence, with substantial variation in developmental timing across fiber pathways and relatively prolonged maturation of association tracts (Lebel and Beaulieu, 2011). Socioeconomic conditions may also be related to variation in these developmental trajectories; for example, Pollmann et al. found that lower socioeconomic status and neighborhood safety were associated with decreasing FA trajectories across early adolescence. However, these longitudinal changes are not directly equivalent to the lower cross-sectional FA associated with higher SES and better cognitive performance in the present multivariate profile. Because FA is influenced by multiple aspects of tissue microstructure and fiber architecture, lower FA should not necessarily be interpreted as poorer white matter development. Rather, these findings identify a distinct pattern of covariation among socioeconomic conditions, association-pathway diffusion properties, and cognitive performance, the developmental basis of which requires further longitudinal investigation.

Taken together, these findings suggest that parental mental health and SES-related adversity may be associated with partially distinct neurobehavioral pathways. The parental psychopathology profile may reflect a proximal family-context pathway, in which caregiver mental health difficulties, family stress, and disrupted caregiving environments are linked to children’s emotional and behavioral problems. Consistent with this interpretation, this profile was primarily associated with children’s mental health outcomes and neuroimaging patterns involving cortical regions, white matter pathways, and functional networks implicated in emotion regulation, self-referential processing, and behavioral control (Hanson et al., 2015, Marusak et al., 2015, Whittle et al., 2014). In contrast, the SES-related profile may reflect a broader resource-related pathway, in which household income, parental education, and neighborhood disadvantage influence cognitive development through differences in material resources, cognitive stimulation, language exposure, educational opportunity, and neighborhood context. This profile was more strongly associated with cognitive performance and brain systems supporting language, memory, executive function, and learning-related processes, including temporal, hippocampal, orbitofrontal, cerebellar, and frontoparietal systems (Michael et al., 2024, Taylor et al., 2020, Tomasi and Volkow, 2023). These differentiated patterns support the rationale for distinguishing ACE profiles rather than relying only on a global cumulative ACE index. At the same time, parental mental health and SES-related adversity may co-occur and interact, and the present cross-sectional imaging findings should be interpreted as evidence of partially separable statistical associations rather than fully independent causal pathways (McLaughlin et al., 2014, Sheridan and McLaughlin, 2014).

Previous studies demonstrated interactions between ACEs exposures and genetic risk for mental in shaping individual mental health outcomes (Crouse et al., 2024, Park et al., 2020, Thorp et al., 2023). In the present study, psychiatric PRS were associated with statistically significant variation in several indirect pathways linking ACEs, brain measures, and behavioral outcomes across genetic risk levels and ancestry groups. However, conditional indirect effects were relatively similar across risk levels, suggesting modest rather than strong moderation effects of PRS. This pattern differs from prior studies reporting stronger gene–environment interactions at higher levels of genetic risk (Belsky and Pluess, 2009, Caspi et al., 2003), and may reflect differences in sample characteristics, analytic approaches, or the multivariate framework adopted in the present study. Across imaging modalities, moderation patterns varied, with more consistent effects observed for structural and white matter measures, and weaker effects for functional connectivity. These findings suggest that genetic influences may relate more strongly to structural neurobiological variation than to functional connectivity in ACE-related pathways, although further work is needed to clarify these relationships. Overall, these results indicate that genetic risk is associated with subtle modulation of ACE-related neurobiological pathways rather than large shifts in effect magnitude. Collectively, the observed ancestry-related differences in moderation patterns underscore the importance of examining gene–environment interactions across diverse populations.

Longitudinal analyses revealed that parental psychiatric problems, particularly MDD and ADHD, were significantly associated with children’s total psychiatric problems scores measured two years later. Notably, this association was bidirectional, such that children’s psychiatric problems also predicted parental psychiatric problems at the two-year follow-up. This finding is consistent with prior work highlighting dynamic family processes, including parental stress, socioeconomic adversity, and maladaptive interaction patterns (Breaux and Harvey, 2019, Cree et al., 2018, Maher et al., 2024). In addition, family income was positively associated with later cognitive performance, consistent with prior longitudinal evidence (Hair et al., 2015). These longitudinal findings should be interpreted in the context of the study design, which focused on a targeted subset of variables to evaluate the temporal stability and directional relationships of key associations identified at baseline, rather than to exhaustively model all available repeated measures.

Sex-stratified sensitivity analyses suggested potential sex-related variation in the socioeconomic–cognitive development profile. Specifically, Model 2 significant only in males, while Model 1 was significant in both sexes. However, the canonical weight and loading patterns were highly consistent across males and females, indicating that the overall multivariate association structure was largely similar across sexes. Therefore, this finding may reflect sex-related variation in the magnitude or statistical detectability of SES–cognition associations rather than a qualitatively distinct sex-specific pathway. This interpretation is consistent with developmental literature showing that adolescence is characterized by sex differences in pubertal timing, cortical maturation, white matter development, and sensitivity to environmental context (Blakemore, 2012, Kaczkurkin et al., 2019, Lenroot and Giedd, 2010). Such developmental differences may influence how socioeconomic exposures relate to cognitive outcomes during early adolescence. Nevertheless, because these analyses were secondary and exploratory, and because sex-by-profile interactions were not tested as primary hypotheses, these findings should be interpreted cautiously and require replication.

Overall, the present findings provide partial support for the hypotheses proposed in this study. First, the sCCA results identified two distinct ACE–behavior profiles: a parental mental health–child psychopathology profile and a socioeconomic–cognitive development profile. These findings are broadly consistent with the hypothesis that different adversity-related domains are linked to distinct behavioral outcomes. Second, modality-specific neuroimaging components partially mediated ACE–behavior associations, although the indirect effects were modest, indicating that brain-based pathways represent one of multiple mechanisms linking adversity to behavioral development. Third, psychiatric PRS showed modest moderation of several ACE–brain–behavior pathways, but conditional indirect effects were relatively similar across genetic risk level, indicating subtle rather than strong genetic modulation. Finally, longitudinal analyses supported temporal and bidirectional associations between selected parental and child behavioral measures, although these analyses were limited to representative variables and should be interpreted cautiously. Collectively, these findings refine the proposed framework by showing that ACE-related developmental pathways are multidimensional, partially brain-mediated, and modestly shaped by genetic liability.

4.1. Limitations

Several limitations should be acknowledged. First, covariation does not imply causality, and although many of the observed associations are biologically plausible, causal inferences cannot be drawn from the present analysis. Second, brain development is influenced by multiple factors operating prior to the enrollment age of ABCD participants, and some findings may reflect the effects of early-life exposures that were not fully captured in this study. Third, although follow-up data were used to validate environmental and behavioral associations, not all variables were included, and neuroimaging data were unavailable beyond the two-year follow-up period, limiting our ability to assess longitudinal changes in brain structure and connectivity. Finally, our analyses were limited to linear relationships, which may overlook complex nonlinear dynamics contributing to the observed outcomes. In addition, the cross-sectional design limits our ability to infer temporal or causal relationships among ACEs, behavioral outcomes, and neurobiological measures. Future large-scale studies incorporating nonlinear modeling and extended longitudinal neuroimaging are warranted to further validate these findings and to provide a more comprehensive understanding of the impact of ACEs on brain and behavioral development.

5. Conclusions

In conclusion, this study demonstrates that distinct ACEs profiles are associated with specific behavioral groups through differentiated neurobiological mechanisms. Moreover, genetic factors, particularly psychiatric PRS, modulate these associations, highlighting the complex interplay between genetic predisposition and environmental exposures. Finally, longitudinal analyses provide additional support for the associations between ACEs exposure and behavioral outcomes. Together, these findings underscore the importance of early intervention and personalized strategies to mitigate the adverse effects of ACEs on mental health and cognitive development.

CRediT authorship contribution statement

Huaigui Liu: Writing – review & editing, Supervision, Project administration, Methodology, Investigation, Funding acquisition, Conceptualization. Jie Tang: Writing – review & editing, Investigation, Methodology, Conceptualization. Zhen Zhao: Writing – review & editing, Investigation, Methodology, Visualization. Yujing Zhou: Writing – review & editing, Data curation. Hao Ding: Writing – review & editing, Investigation, Visualization. Nana Liu: Writing – review & editing, Investigation. Yuan Ji: Writing – review & editing, Funding acquisition, Data curation. Piaoran Wang: Writing – review & editing, Formal analysis. Xiangzheng Wu: Writing – review & editing, Formal analysis. Feng Liu: Writing – review & editing. Wen Qin: Writing – review & editing. Vince D. Calhoun: Writing – review & editing, Supervision. Jiayuan Xu: Writing – review & editing, Supervision, Methodology, Conceptualization. Yurong Jiang: Writing – original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Data curation, Conceptualization.

Ethics declaration

Written informed consent to take part in the study and to publish the article has been obtained from all participants or their legal representatives. The privacy rights of participants have been observed.

This study was performed in compliance with relevant laws, regulatory frameworks and guidelines where the research took place. Ethics committee approval was not required under relevant laws and institutional guidelines. The present study involved secondary analysis of de-identified data from the Adolescent Brain Cognitive Development (ABCD) Study and did not involve new participant recruitment or data collection. The original ABCD Study protocol was approved by the Central Institutional Review Board at the University of California, San Diego (IRB #160091), with local IRB approval at participating sites as applicable.

Ethics approval and consent to participate

The study was approved by the Institutional Review Board of the University of California, San Diego, and other relevant institutional review boards. Most ABCD research sites rely on a central Institutional Review Board (IRB) at the University of California, San Diego for the ethical review and approval of the research protocol, with a few sites obtaining local IRB approval (e.g., Washington University in St. Louis) (https://www.sciencedirect.com/science/article/pii/S1878929317300622). Parents or guardians provided written informed consent after the procedures were fully explained, and children assented prior to participation in the study.

Consent for publication

Not applicable.

Funding

This study was supported by the Tianjin Natural Science Foundation (24ZXGQSY00040, 25JCQNJC00210), the Natural Science Foundation of China (81971599, 81771818, 82030053, 82001797, 82072001), the Tianjin Science and Technology Program (25JCQNJC00210), the Special Research Fund for Central Universities, Peking Union Medical College (3332025055) and China Postdoctoral Science Foundation (2025M772058), and Tianjin Key Medical Discipline (Specialty) Construction Project (TJYXZDXK-3-008C).

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

Supplementary material

Data availability

The data used in this article stem from the ABCD study. The ABCD Study releases curated, anonymized data annually, beginning in 2018, to the research community. Data not available/The data that has been used is confidential.