Genetics of educational resilience after adverse childhood experiences

Summary

Background

Adverse childhood experiences (ACEs) are associated with poorer mental health and may hinder educational attainment. Yet, some ACEs survivors still achieve higher education, suggesting the presence of educational resilience, whilst its underlying mechanisms remain unclear. This study aimed to evaluate the cross-cultural consistency of the ACEs-education association and to identify the underlying genetic determinants of educational resilience.

Methods

This multi-cohort study included 127,571 UK Biobank participants, 27,128 from the Icelandic Stress-and-Gene-Analysis (SAGA) Cohort, and 4910 from the China Severe Trauma Cohort (CSTC). Associations between ACEs and educational attainment (i.e., achieving a college education or years of education) were examined using multivariable regression models. In the UK Biobank and CSTC, we performed stratification analyses by polygenic score (PGS) for educational attainment to evaluate the modifying role of genetic predisposition. We then conducted a genome-wide association study (GWAS) in the UK Biobank to identify genetic variants associated with educational resilience. Resilient cases were defined as individuals exposed to ACEs in the top 10^th percentile of residuals (i.e., those with the lowest 10^th percentile of predicted probability of completing college who nonetheless achieved a college degree), while the remaining ACE-exposed participant served as controls. Gene–ACEs interaction analyses among all UK Biobank participants studied were used to verify the modification effects of identified variants.

Findings

ACEs were associated with lower odds of obtaining a college degree across all three cohorts (odds ratio [95% confidence interval] = 0.75 [0.72–0.77] for the UK Biobank, 0.69 [0.66–0.73] for SAGA cohort, and 0.46 [0.39–0.54] for CSTC). These associations were comparable across different PGS strata both in the UK Biobank and CSTC. GWAS in the UK Biobank identified a potential novel locus (rs2743239) associated with educational resilience, mapped to several neuroprotective and immune-related pathways. We further detected a significant gene-ACEs interaction (p for interaction <0.0001) in the UK biobank, indicating that among ACEs exposed individuals, TT carriers of rs2743239 had higher odds of college completion than AA carriers (1.44 [1.21–1.71]); this trend was not observed among individuals without ACEs (1.01 [0.95–1.09]).

Interpretation

ACEs were associated with lower educational attainment, irrespective of genetic predisposition to education. A potential novel variant, rs2743239, was linked to educational resilience specifically among ACEs-exposed individuals. While further validation is required, these findings offer suggestive evidence for a genetic basis of educational resilience that may inform future intervention strategies for risk management among ACEs survivors.

Funding

National Natural Science Foundation of China.

Research in context.

Evidence before this study

We conducted a systematic search of literature to characterise the genetic architecture of resilience. We searched PubMed for studies published up to July 4, 2025, with no language restrictions, using the following terms: (“Resilience” [Title]) AND (GWAS [Title/Abstract] OR “genome wide association study” [Title/Abstract] OR “genetics” [Title/Abstract]). The search yielded 213 records, of which 17 were considered eligible after manual screening. These studies measured resilience using highly variable approaches, including self-report scales (three studies), the absence of adverse health outcomes (six studies), and residuals derived from models of self-reported psychological scales, neuroimaging-based cognitive function, or late-life neurodegenerative disorders (nine studies). Although nine studies reported genome-wide significant associations, the identified genetic variants were largely inconsistent, with no common loci verified across studies. This genetic heterogeneity underscores a critical limitation in the field: the absence of an objective and easily accessible definition of resilience. To address this gap, our study adopted educational attainment following exposure to adverse childhood experiences (ACEs) as a proxy for resilience-an approach that reflects functional adaptation despite early-life adversity and enables the identification of genetic determinants in a contextually meaningful and reproducible manner.

Added value of this study

To the best of our knowledge, this is the first study to evaluate educational resilience following ACEs and to investigate its underlying genetic determinants. Leveraging large-scale, multi-ancestry data from trauma-exposed individuals in the UK, Iceland, and China, we demonstrated that ACEs were consistently associated with lower odds of college attainment, independent of polygenic risk for educational attainment. Our genome-wide association study analyses for educational resilience in the UK Biobank identified a novel locus (rs2743239) that modified the effect of ACEs on education, with TT carriers showing significantly higher odds of college completion after ACEs exposure. Functional annotation further mapped this locus to genes and pathways involved in neuroprotective and immune regulation.

Implications of all the available evidence

Our findings support the use of educational attainment as a meaningful and scalable proxy of resilience to ACEs. Moreover, the identification of educational resilience-related genetic variants, particularly those linked to neuroprotective and immune-related pathways, highlights potential biological targets for interventions aimed at promoting adaptive outcomes among individuals exposed to early-life trauma.

Introduction

Adverse childhood experiences (ACEs) encompass a spectrum of traumatic events or adverse circumstances that occur during childhood and adolescence, including maltreatment (e.g., sexual, physical, and emotional abuse and neglect) and other traumatic experiences.¹ Extensive research has established that ACEs exert a pervasive and long-lasting burden on both social development and adult health, often exhibiting a clear dose–response relationship with a broad spectrum of negative outcomes.2, 3, 4 These impacts range from lower educational attainment⁵ and economic prosperity⁶ to increased risks for psychiatric disorders,⁷ health-harming behaviours,⁸ and various chronic non-communicable diseases such as respiratory illnesses,² cardiovascular disease,⁹ and endocrine diseases.¹⁰ Furthermore, recent estimates highlight that ACEs contribute significantly to the global burden of disease and social inequality, imposing staggering economic costs on healthcare systems.^3,4 However, many individuals exposed to ACEs show minimal signs of disruptions in social development or adverse health outcomes, which has sparked research interests in resilience.^11,12

Resilience is conceptualised as the ability to adapt well in the face of adversity, trauma, and other hardships.¹³ Prior research has operationalised resilience as a capacity, process, or favourable outcome in various ways,¹² including self-assessed questionnaires (e.g., the Connor-Davidson Resilience Scale¹⁴), the interaction effect of genomics (e.g., polygenic risk for depression) with adversity,¹⁵ the absence of negative health consequences (e.g., psychiatric morbidity¹⁶ and cognitive impairment¹⁷) after adversities, or their residuals¹⁸ as an indicator. The varying definitions and measurement approaches have resulted in notable heterogeneity in the results of previous studies.

Despite challenges in defining resilience, twin studies have revealed that self-assessed resilience is heritable, with genetics accounting for 25%–77% of the variance.19, 20, 21 Yet, to date, only a handful of genome-wide association studies (GWASs) have focused on resilience,^12,22 yielding inconsistent results. For instance, prior GWASs have identified rs4260523 for self-assessed resilience (measured by questionnaire as a capacity),²³ alongside loci for outcome-based resilience identified using residual approach, such as rs2571244 for better-than-predicted cognitive performance in Alzheimer's disease,¹⁷ whereas some studies failed to identify any significant loci.24, 25, 26 Beyond small sample sizes^27,28 and non-representative study populations,^17,23,27,29 existing measures vary widely across studies-ranging from subjective scales to disorder specific outcomes-making it difficult to establish reproducible genetic findings.²³ Ultimately, the primary challenge remains the substantial conceptual heterogeneity and the lack of an objective, standardised operational definition of resilience.

Educational attainment could serve as a robust indicator of resilience in diverse populations, given the consistently demonstrated inverse association between ACEs and educational attainment,^5,30,31 as well as its standing as an objective, reliably measured, and cross-culturally standardised life-course milestone.³² Unlike self-reported psychological scales, educational attainment is less susceptible to recall bias or transient mood states. It captures sustained functional adaptation across childhood and adolescence, integrating cognitive, motivational, behavioural, and social competencies in the context of adversity.^33,34 From a life-course perspective, this metric reflects long-term adaptive functioning in a socially meaningful domain with well-established links to future health outcomes.³⁵ To address prior limitations, we operationalise achieving better-than-predicted educational attainment following ACEs (i.e., educational resilience) as a novel outcome-based resilience indicator. Leveraging data from the UK Biobank, the Icelandic Stress-and-Gene-Analysis (SAGA) Cohort, and the China Severe Trauma Cohort (CSTC) including a large sample of traumatised individuals from the UK, Iceland, and China, respectively, we first aimed to evaluate the association between ACEs and educational attainment across three diverse populations, and to examine the potential modifying role of polygenic score (PGS) for educational attainment on this association. Employing a residual approach to measure educational resilience, we further attempted to identify genetic determinants of educational resilience after ACEs.

Methods

This study consisted of three main parts (Fig. 1). First, to determine the role of ACEs in educational attainment across populations, we constructed three cohorts for assessing the association between ACEs and educational attainment based on data from the UK Biobank (n = 127,571), SAGA cohort (n = 27,128), and CSTC (n = 4910). We then tested if PGS for educational attainment modified the association between ACEs and educational attainment in the UK Biobank and CSTC. Second, among ACEs-exposed participants of the UK Biobank (n = 17,219), we employed a residual approach to measure educational resilience after ACEs and performed GWAS to identify the genetic determinants associated with this educational resilience. Third, we verified specificity of the identified genetic variants of educational resilience subsequent to ACEs, through studying the association between ACEs and educational attainment by the identified variants among individuals exposed or not exposed to ACEs, in the UK Biobank (n = 127,571).

Fig. 1.

Components of the study design. ^a Adverse childhood experiences. ^b Residuals were derived from the regression model of educational attainment among ACEs exposed individuals, adjusting for number of ACEs, birth year, sex, number of full siblings, and polygenic score for educational attainment. Individuals with notable model residuals were defined as those who were with the lowest 10^th percentile predicted probability of obtaining a college degree but had achieved a college degree after ACEs.

Study design–association between ACEs and educational attainment

Cohort construction

Analytical sample based on the UK Biobank

The UK Biobank is a large community-based prospective cohort study that enrolled more than 500,000 individuals (aged 40–69 years) between 2006 and 2010 from England, Scotland, and Wales.³⁶ At recruitment, detailed information on demographic (e.g., birth year, sex, and number of full siblings) and socioeconomic (e.g., Townsend deprivation index [TDI]) factors was collected using touchscreen questionnaires. Genotyping data, derived from blood samples collected at recruitment, were released for approximate 500,000 participants using two closely related arrays (95% shared marker content).³⁷ Between July 2016 and July 2017, a proportion of the participants (n = 339,229) were invited to complete an online mental health survey, of whom 157,349 (46.4%) completed questions about traumatic events. Adverse events in childhood were assessed using Childhood Trauma Screener-5 item (CTS-5),³⁸ with 5 items related to emotional neglect, physical neglect, physical abuse, emotional abuse, and sexual abuse, respectively.

In the present study, the cohort based on the UK Biobank included 157,349 individuals who participated in the online mental health survey. We excluded participants who withdrew (n = 42) or had incomplete information on CTS-5 (n = 42) or educational attainment (n = 1180). To harmonise phenotypic and genetic data, we additionally removed individuals with non-white British ancestry (n = 24,390), related individuals (third degree or greater: kinship coefficient >0.044, n = 3744), and individuals failed to pass GWAS QC steps (n = 380), leaving 127,571 eligible participants for the final analyses (Supplementary Figure S1A).

Analytical sample based on the SAGA cohort

The SAGA cohort is an ongoing population-based study on the impact of trauma on women's health in Iceland, which recruited 30,403 women at the age of 18–69 from March 1st 2018 through July 1st 2019.¹⁶ Participants were invited to complete an online survey to collect data on demographics (e.g., age at recruitment and educational attainment), socioeconomic factors (e.g., childhood deprivation and region of residence), and traumatic events. ACEs were measured by the Adverse Childhood Experiences International Questionnaire (ACE-IQ)³⁹ which includes 39 items on 13 ACEs during the first 18 years of a person's life.

Among the 30,403 participating women of the SAGA cohort, we excluded those with missing data on ACEs (n = 857) or educational attainment (n = 126) and participants aged 23 and younger (n = 2292), leaving 27,128 eligible participants in the final analysis (Supplementary Figure S1B).

Analytical sample based on the CSTC

CSTC, launched on June 1, 2020, is an ongoing prospective cohort study in China.⁴⁰ In brief, CSTC recruited patients (aged 12–80 years) who were admitted to the Trauma Center of West China Hospital for severe physical trauma experienced within three months. Information on sociodemographic characteristics (e.g., age at recruitment, sex, educational attainment, and region of residence) and adverse life experiences was collected through face-to-face interviews at recruitment (conducted by trained data collectors). ACEs were assessed with the 28-item Childhood Trauma Questionnaire-Short Form (CTQ-SF)⁴¹ in CSTC.

In the present study, we included all CSTC participants with available data on ACEs (n = 5624) after excluding individuals who withdrew (n = 18), had no information on educational attainment (n = 127), or were under 23 years old (n = 569), leaving 4910 participants in the analyses of the CSTC (Supplementary Figure S1C).

Ascertainment of ACEs

To harmonise the ACEs measurement in the three cohorts, we focused on five types of ACEs, including emotional neglect, physical neglect, physical abuse, emotional abuse, and sexual abuse, commonly measured in all cohorts. We then studied the presence of any ACE (yes or no) and number of ACEs (0, 1, 2, or ≥ 3). The specific questions and cutoff points are shown in Supplementary Table S1.

Ascertainment of educational attainment

Educational attainment was assessed by the self-reported highest education level in all cohorts (Supplementary Table S2). To enhance comparability across cohorts, we defined educational attainment both as a binary variable (college degree [yes or no]) and a continuous variable (years of education) according to the International Standard Classification of Education (ISCED) category^32,42 (Supplementary Table S2). For sensitivity analyses, secondary education was further defined as ISCED level ≥2.

Covariates

Information on potential confounders, including birth year (in UK Biobank) or age at recruitment (in SAGA cohort and CSTC), sex (in UK Biobank and CSTC), number of full siblings (in UK Biobank), and childhood deprivation (in SAGA cohort) was collected at recruitment. PGS for educational attainment were computed in the UK Biobank and CSTC, based on the independent summary statistics of GWAS released from 69 European cohorts (n = 324,162, excluding UK Biobank and 23andMe)³² and 176,400 East Asian population,⁴² using penalised regression (LASSO) approach. As a validation, we found a significant association between the calculated PGS and educational attainment in the UK Biobank (college degree: odds ratios [OR] = 1.62 [1.60–1.63]; education years: beta = 1.13 [95% CI 1.10–1.16]) and CSTC (college degree: OR = 1.23 [1.13–1.34]; educational years: beta = 0.36 [0.23–0.49]).

Study design–genetic determinants of educational resilience after ACEs

Definition of educational resilience

Among the 17,219 UK Biobank participants who were exposed to ACEs and had available genotyping data, we established a logistic regression model (model 1), including number of ACEs (n = 1, 2, or ≥ 3), birth year, sex, number of full siblings, and PGS for educational attainment, to predict obtaining a college degree after ACEs exposure. To take into account the aforementioned important confounders related to educational attainment, we defined educational resilience based on the residual (i.e., the difference between the observed probability [0 or 1] and the predicted probability [0–1] of obtaining a college degree) derived from model 1 (model 2). The distribution of residual is shown in Supplementary Figure S2. The residuals were calculated from linear regression models when education attainment was defined as years of education.

$$logit\left(P\left(Y=1\right)\right)=\beta 0+{\beta }_{1}ACEs+{\beta }_{2}YOB+{\beta }_{3}Sex+{\beta }_4{N}_{siblings}+{\beta }_{5}PG{S}_{education}+\epsilon$$ (1)

$$e_i=Y_i-{\hat{Y}}_i$$ (2)

where logit (P(Y = 1)) is the natural logarithm of the odds of obtaining a college degree; YOB is birth year and $\epsilon$ is the error term. $e_i$ (i.e., residual) is the difference between the observed probability (0 or 1) and the predicted probability of obtaining a college degree (0–1) for an individual i.

GWAS

In the GWAS analysis, we operationalised educational resilience as a binary phenotype. Cases were defined as individuals with the top 10^th percentile of residuals (i.e., those who had the lowest 10^th percentile predicted probability of obtaining a college degree but had achieved a college degree after ACEs, n = 1722), while the remaining ACEs-exposed individuals were considered as controls (n = 15,497).

Study design–gene–environment interaction

To further verify the modification effect of the identified genetic variants from the GWAS, we assessed the association between ACEs and educational attainment by variations of the identified variants among individuals exposed and not exposed to ACEs in the UK Biobank sample (n = 127,571).

Statistical analysis

Association between ACEs and educational attainment

In the three cohorts, we employed multivariate logistic regression models to estimate the association between ACEs (any ACE or numbers of ACEs) and obtaining a college degree, adjusted for birth year (in UK Biobank) or age at recruitment (in SAGA cohort and CSTC), sex (in UK Biobank and CSTC), number of full siblings (in UK Biobank), childhood deprivation (in SAGA cohort), and PGS for educational attainment (in UK Biobank and CSTC). To assess the effect modifying role of genetic predisposition to educational attainment on the studied association, we stratified the analyses by levels of PGS for college degree in the UK Biobank and CSTC (low: <first tertile of PGS; intermediate: between first and second tertile; high: >second tertile). Further, we analysed the associations for five specific types of ACEs (i.e., emotional neglect, physical neglect, physical abuse, emotional abuse, and sexual abuse). Finally, we repeated the analyses using years of education as an alternative measure of educational attainment through linear regression.

Other subgroup and sensitivity analyses are shown in Supplementary Material. A 2-sided p < 0.05 was considered statistically significant. Number of ACEs was treated as a continuous variable to calculate P for trend. The statistical significance of difference between odds ratios (ORs) was assessed by Wald test or including an interaction term in the regression model.

Genetic determinants of educational resilience after ACEs

The quality control procedures for the UK Biobank and CSTC genotype data are detailed in Supplementary Material. Based on 1722 cases and 15,497 controls of the UK Biobank, we conducted GWAS analysis to identify SNPs associated with educational resilience after ACEs, using the mixed linear model (MLM)-based approach (fastGWA)⁴³ after adjustment for genotyping batch and the top 20 principal components (PCs). To evaluate the statistical detectability of the genetic findings, we performed a formal power analysis using the genpwr R package, assuming an additive genetic model at a genome-wide significance level (p < 5 × 10⁻⁸). To test the robustness of our findings to the definition of educational resilience after ACEs, we repeated the GWAS analyses using an alternative definition for cases (i.e., using the lowest 20^th, instead of the lowest 10^th predicted possibility) and residuals (calculated from years of education rather than college degree). Finally, we estimated the SNP heritability (h²) of the operationalised educational resilience phenotype using LD Score Regression (LDSC).⁴⁴

Genomic loci identified with significant SNPs of the GWAS (p < 5 × 10⁻⁸), together with surrounding genomic loci identified based on linkage disequilibrium (LD) structure (i.e., r² ≥ 0.6), were gene-mapped and functionally annotated, using the online Functional Mapping and Annotation of GWAS (FUMA) tool⁴⁵ (see Supplementary Material for details). All mapped genes were subsequently included in the gene-set enrichment analysis using the webtool Metascape⁴⁶ to identify potential biological pathways.

Gene-environment interaction

Among the 127,571 participants of the UK Biobank, we evaluated the association of ACEs and educational attainment using logistic regression models by different genotypes (AA, TA, and TT) of the genetic variants identified in the GWAS, adjusted for birth year, sex, number of full siblings, PGS for educational attainment, and the top 20 PCs. This analysis was first performed in the total sample and then separately for individuals exposed (n = 17,219) or not exposed (n = 110,352) to ACEs. The statistical significance of the difference in genotype-specific ORs between individuals with different ACEs exposure statuses was assessed by introducing an interaction term between ACEs and the variations of genetic variants in the regression model.

Ethics

All participants in the above three cohorts provided written informed consent before data collection. The UK Biobank has full ethical approval from the NHS National Research Ethics Service (16/NW/0274). The SAGA cohort and CSTC was approved by the National Bioethics Committee of Iceland (NBC 17-238) and the ethics committee of West China Hospital, Sichuan University (2020.243), respectively. The study was performed in accordance with the principles of the Declaration of Helsinki and approved by the biomedical research ethics committee of West China Hospital (2019.1171).

Role of funders

The funders had no role in the study design, data collection, data analysis, data interpretation, or writing of the manuscript.

Results

Association between ACEs and educational attainment

A total of 127,571, 27,128, and 4910 individuals were included in the analytical samples of UK Biobank, SAGA cohort, and CSTC, respectively (Supplementary Figure S1). Characteristics of the three cohorts were largely comparable (Table 1), except that the SAGA cohort included exclusively females, participants in the UK Biobank were more likely to be female (56.01%), whilst participants of the CSTC were more likely to be male (58.82%). Regardless, in all cohorts, individuals exposed to ACEs were at middle age at recruitment (mean age = 55.90, 45.60, and 54.30 years, respectively) and were more likely to have more than three full siblings (32.67% vs 22.37% in the UK Biobank) and childhood deprivation (10.19% vs 1.27% in SAGA cohort), compared to the non-exposed individuals.

Table 1.

Characteristics of the study populations.

	Individuals without ACEsa	Individuals with ACEsa	Overall
UK Biobank
Sample size	110,352	17,219	127,571
Birth year, Mean (SD)	1950 (7.62)	1950 (7.79)	1950 (7.65)
Age at recruitment, Mean (SD)	56.30 (7.64)	55.90 (7.79)	56.20 (7.66)
Sex, No. (%)
Female	60,704 (55.01%)	10,745 (62.40%)	71,449 (56.01%)
Male	49,648 (44.99%)	6474 (37.60%)	56,122 (43.99%)
Number of full siblings, No. (%)
0	15,597 (14.13%)	2095 (12.17%)	17,692 (13.87%)
1	41,155 (37.29%)	5065 (29.42%)	46,220 (36.23%)
2	28,738 (26.04%)	4384 (25.46%)	33,122 (25.96%)
≥3	24,687 (22.37%)	5626 (32.67%)	30,313 (23.76%)
Unknown	175 (0.16%)	49 (0.28%)	224 (0.18%)
Education polygenic score, by tertile, No. (%)
Low	36,001 (32.62%)	6519 (37.86%)	42,520 (33.33%)
Intermediate	36,761 (33.31%)	5758 (33.44%)	42,519 (33.33%)
High	37,590 (34.06%)	4942 (28.70%)	42,532 (33.34%)
The Icelandic Stress-and-Gene-Analysis (SAGA) Cohort
Sample size	17,259	9869	27,128
Age at recruitment, Mean (SD)	45.60 (12.70)	45.60 (12.00)	45.60 (12.50)
Sex, No. (%)
Female	17,259 (100%)	9869 (100%)	27,128 (100%)
Male	–	–	–
Childhood deprivation, No. (%)
Often	219 (1.27%)	1006 (10.19%)	1225 (4.52%)
Sometimes	876 (5.08%)	1549 (15.70%)	2425 (8.94%)
Seldom	1529 (8.86%)	1530 (15.50%)	3059 (11.28%)
Never	14,594 (84.56%)	5743 (58.19%)	20,337 (74.97%)
Can/Will not answer	41 (0.24%)	41 (0.42%)	82 (0.30%)
The China Severe Trauma Cohort
Sample size	2785	2125	4910
Age at recruitment, Mean (SD)	46.60 (14.00)	54.30 (13.90)	49.90 (14.50)
Sex, No. (%)
Female	1208 (43.38%)	814 (38.31%)	2022 (41.18%)
Male	1577 (56.62%)	1311 (61.69%)	2888 (58.82%)
Education polygenic score, by tertile, No. (%)
Low	572 (20.54%)	457 (21.51%)	1029 (20.96%)
Intermediate	571 (20.50%)	457 (21.51%)	1028 (20.94%)
High	599 (21.51%)	428 (20.14%)	1027 (20.92%)
Unknown	1043 (37.45%)	783 (36.85%)	1826 (37.19%)

^aAdverse childhood experiences.

In all cohorts, ACEs exposure was associated with a reduced odds of obtaining a college degree (OR [95% confidence interval (CI)] for any ACE = 0.75 [0.72–0.77], 0.69 [0.66–0.73], and 0.46 [0.39–0.54] in UK Biobank, SAGA cohort, and CSTC, respectively) (Fig. 2A), yielding a significant pooled estimate of 0.63 (0.47–0.83) (Supplementary Figure S3). There was a dose–response association between number of ACEs and educational attainment (e.g., OR [95% CI] for more than three ACEs = 0.70 [0.63–0.78], 0.54 [0.48, 0.60], and 0.46 [0.30, 0.69] in UK Biobank, SAGA cohort, and CSTC, respectively) (Fig. 2B). We obtained largely comparable estimates for all five types of ACEs (Supplementary Figure S4) and among individuals with different levels of education PGS in the UK Biobank and CSTC (Fig. 2A). Similar results were obtained when using education years (β [95% CI] for any ACE = −0.76 [−0.83, −0.69], −0.88 [−1.01, −0.76], and −1.90 [−2.17, −1.64] in the UK Biobank, SAGA cohort, and CSTC, respectively) (Supplementary Table S3) or attainment of secondary education (OR [95% CI] for any ACE = 0.67 [0.63, 0.70], 0.70 [0.65, 0.76], and 0.44 [0.38, 0.51] in the UK Biobank, SAGA cohort, and CSTC, respectively) (Supplementary Table S4) as the outcome.

Fig. 2.

The associations between (A) any adverse childhood experiences (ACEs) or (B) number of ACEs and college degree in the UK Biobank, theIcelandicStress-and-Gene-Analysis (SAGA)Cohort, and the China Severe Trauma Cohort (CSTC). ^a In the UK Biobank, odds ratio (OR) was derived from logistic regression, adjusted for birth year, sex, number of full siblings, and polygenic score (PGS) for educational attainment. In SAGA cohort, OR was derived from logistic regression, adjusted for age at recruitment and childhood deprivation. In CSTC, OR was derived from logistic regression, adjusted for age at recruitment, sex, and PGS for educational attainment. ^b PGS for educational attainment was calculated based on the independent GWAS summary statistics with European ancestries released by Okbay A et al. (2022), using penalised regression (LASSO) approach. ^c PGS for educational attainment was calculated based on the independent GWAS summary statistics with East Asian population released by Chen TT et al. (2024), using penalised regression (LASSO) approach.

The observed associations remained robust even after additional adjustment for health status and adult socioeconomic status (Supplementary Table S5). Subgroup analyses revealed no effect modification by socioeconomic factors, such as number of siblings or TDI in the UK Biobank, childhood deprivation or region of residence in the SAGA cohort, or region of residence in the CSTC (Supplementary Tables S6–S8). Additionally, we observed largely similar results among individuals with different birth years in the UK Biobank (Supplementary Table S6), and across age groups or sexes in the CSTC (Supplementary Table S7), although the estimates were somewhat stronger among younger individuals in the SAGA cohort (Supplementary Table S8) and among males in the UK Biobank (Supplementary Table S6).

Restricting the analysis to individuals who were not students at recruitment in the SAGA cohort and CSTC or using all items to define ACEs in the SAGA cohort (Supplementary Tables S9 and S10) rendered similar results as well.

Genetic determinants of educational resilience after ACEs

In the GWAS analysis, we found one SNP (i.e., rs2743239, β = 0.26, p = 1.93 × 10⁻⁸, INFO = 0.963, MAF = 0.191), located on chromosome 20, which was statistically significantly associated with educational resilience after ACEs (Fig. 3A). The regional association and LD patterns around this locus are visualised in Fig. 3C. The genomic location and functional annotation of rs2743239 were verified using the NCBI dbSNP database⁴⁷ and the Ensembl Genome Browser (GRCh38.p14 release),⁴⁸ identifying it as an intron variant within the SYS1-DBNDD2 read-through transcript. The Q–Q plot (Fig. 3B) and an LDSC intercept of 0.996 (SE = 0.007) indicated no noticeable genomic inflation. A formal power analysis estimated a 68.00% probability of detecting rs2743239 based on the observed effect size (OR = 1.30) and MAF (0.191), suggesting that the locus is unlikely to be an artifact of limited power. Furthermore, the SNP heritability (h²) was estimated at 7.37% (SE = 2.74) on the observed scale, corresponding to 21.55% (SE = 8.00%) on the liability scale (assuming a 10% population prevalence; p = 0.007), which is indicative of a significant polygenic architecture of educational resilience. A total of 60 SNPs were highly correlated with this SNP based on LD structure (i.e., r² ≥ 0.6; Fig. 3C and Supplementary Table S11). This locus was mapped to 32 genes (Supplementary Table S12), including the host gene SYS1-DBNDD2 (annotated as SYS1 in Fig. 3C), where the most enriched pathways were related to antibacterial humoural response (e.g., Gene Ontology [GO]:0019731) and biological processes involved in interspecies interaction between organisms (GO:0044419) and immune system (GO:0002376, Supplementary Figure S5).

Fig. 3.

Summary of the Genome-Wide Association Study (GWAS) findings of educational resilience among 17,219 UK Biobank participants with adverse childhood experiences. (A) Manhattan plot for educational resilience associations. The x axis is chromosomal position, and the y axis is the significance on a −log₁₀ scale (two-tailed test). The red dashed line shows the genome-wide significance level (5 × 10⁻⁸). (B) Q–Q plot showing the distributions of the observed to expected p-values. (C) LocusZoom plot of rs2743239 in educational resilience GWAS based on the European reference panel.

When using the lowest 20^th (instead of the lowest 10^th) predicted probability to define cases of educational resilience after ACEs, no genome-wide significant variant was identified at p < 5 × 10⁻⁸ but the same variant (i.e., rs2743239) reached a suggestive significance (β = 0.16, p = 4.79 × 10⁻⁶, Supplementary Figure S6). The results of GWAS analysis using education years, instead of college degree, as a measurement of educational resilience after ACEs indicated the same SNP as the top signal (i.e., rs2743239, β = 0.26, p = 3.06 × 10⁻⁸, Supplementary Figure S7). This variant has not been identified in previous GWAS of educational attainment in either the UK Biobank or other populations (Supplementary Figure S8).

Gene-environment interaction

The association between ACEs and educational attainment was significantly modified by variation in rs2743239 (p for interaction <0.0001, Supplementary Table S13). Specifically, the odds of obtaining a college degree after ACEs was 44% higher among TT carriers of rs2743239 compared to AA carriers (OR [95% CI] = 1.44 [1.21–1.71]) among individuals exposed to any ACE (Table 2). Conversely, no such difference was observed among individuals unexposed to ACEs (1.01 [0.95–1.09]).

Table 2.

rs2743239-ACEs interaction analysis for college degree in UK Biobank among individuals with and without adverse childhood experiences (ACEs).

Genotypes	Individuals without ACEs (n = 110,352)		No. of outcome (%)	Individuals with ACEs (n = 17,219)	p for interaction
	No. of outcome (%)	ORa (95% CI)		ORa (95% CI)
AA	31,604/68,538 (46.11)	Ref	3877/10,722 (36.16)	Ref
TA	14,989/32,608 (45.97)	0.99 (0.97, 1.02)	1902/5025 (37.85)	1.08 (1.01, 1.16)
TT	1787/3831 (46.65)	1.01 (0.95, 1.09)	270/614 (43.97)	1.44 (1.21, 1.71)	<0.0001

^aOdds ratio (OR) derived from logistic regression, adjusted for birth year, sex, number of full siblings, polygenic score for educational attainment, and the top 20 principal components.

Discussion

Based on three cohorts with participants from the UK, Iceland, and China, we observed a consistent inverse association between ACEs and educational attainment, which was independent of genetic predisposition to educational attainment according to analyses performed in the UK Biobank and CSTC. Furthermore, we constructed an algorithm to define educational resilience after exposure to ACEs, using a residual approach and performed GWAS analyses in the UK Biobank. Such analyses identified a potential novel locus (rs2743239) on chromosome 20 for educational resilience after ACEs, which was mapped to genetic components related to neuroprotective and immune-related pathways. Notably, we found that ACEs exposed TT carriers of rs2743239 showed a higher likelihood of obtaining a college degree than AA carriers; however, no such effect was observed among individuals without ACEs exposure. These findings further motivate risk assessment and support for individuals exposed to ACEs and, if validated, may facilitate further mechanistic investigations of educational resilience.

In prior studies, resilience has been measured in diverse ways, often through self-assessed questionnaires or indirectly inferred from the absence of vulnerability.¹² There are limitations related to the subjective reporting^14,23,49 and inconsistent definitions between studies.^17,28 Further, some studies have failed to account for other confounding factors such as genetic predispositions.^16,17 As a result, previous investigations have yielded largely inconsistent findings. In the present study, we first consistently demonstrated an inverse association between ACEs and educational attainment in three cohorts from different countries, providing evidence for using educational attainment as an objective and easily measured indicator of resilience after ACEs. Further, we addressed potential modification by genetic predisposition to educational attainment and socioeconomic factors without demonstrating substantial influence of these factors on the reported association. We then operationalised educational resilience using the residuals obtained from a prediction model adjusting for both the PGS for educational attainment and other environmental factors. Notably, the genetic variant (rs2743239) identified in our GWAS analysis has not been previously associated with educational attainment or intelligence. This finding helps alleviate a common concern regarding residual-based approaches¹²-that the observed residuals or their genetic variations might largely reflect general educational attainment rather than specifically capturing educational resilience in the context of ACEs.

Despite limited understanding of resilience mechanisms, emerging evidence underscores the importance of neuroimmune interactions in mediating stress resilience. Several biological pathways, such as the serotonergic, dopaminergic and noradrenergic systems, along with hypothalamic–pituitary–adrenal axis, have been proposed to contribute to resilience against stress and psychiatric disorders.²² Several neurotransmitters (e.g., reducing brain d-serine⁵⁰) and neurotransmitter receptors (e.g., restoring of N-methyl-d-aspartate receptor subunit 2B⁵¹) have been identified to potentially improve stress resilience through modulating neuroinflammation. Alternated immune responses have also been associated with stress resilience.^52,53 The identified novel locus of our study (i.e., rs2743239) has not been previously reported in relation to resilience or with educational attainment or intelligence. However, its biological plausibility is supported by its location as an intron variant within the SYS1-DBNDD2 read-through transcript.^47,54 According to GeneCards⁵⁵ and GWAS Catalogue data,⁵⁶ this locus acts as a GeneHancer for the adjacent SDC4 gene cluster and is linked to memory performance⁵⁷ as well as the regulation of key neurotrophic proteins, including Midkine,⁵⁸ secreted frizzled-related protein 1 (sFRP1),⁵⁸ and PHF-tau levels,⁵⁹ all of which play critical roles in neuroprotection and neural structural stability.60, 61, 62 Collectively, these findings support a potential neurobiological basis for educational resilience, suggesting that the maintenance of neurobiological integrity may help mitigate the deleterious effects of early-life adversity. Beyond these mechanisms, the rs2743239 locus was also mapped to genes (e.g., WFDC2, SPINT3, and SPINT4) and pathways (e.g., GO:0019731: antibacterial humoural response and GO:0002376: immune system process) related to immunity. For instance, results of human and animal studies have indicated that WFDC2, mainly involved in neuroimmune-related signalling pathways,⁶³ is associated with the latent stress responses related to reward circuitry in nucleus accumbens,⁶⁴ anti-inflammatory functions in inflammatory bowel disease,⁶⁵ and cognitive decline in neurological disorders.66, 67, 68 Serine protease inhibitors SPINT3 and SPINT4 are involved in several critical biological processes⁶⁹ regulating homoeostasis and inflammation with impact in the pathogenesis of many diseases such as glioma,⁷⁰ epilepsy,⁷¹ and schizophrenia.⁷² The identified pathways (i.e., GO:0019731: antibacterial humoural response) and biological process (GO:0002376: immune system process) are also linked to infectious and inflammatory diseases (e.g., bacterial meningitis,⁷³ sepsis,⁷⁴ and asthma after allergic rhinitis⁷⁵). Together with the noted protective role of rs2743239 TT genotype for educational attainment after ACEs, our work adds to the growing literature by highlighting the contribution of neuroprotective and immunity-related pathways in the individual vulnerability to ACEs. Nevertheless, these interpretations are based primarily on bioinformatic annotation and indirect evidence and should be considered hypothesis-generating pending independent replication and functional validation.

Major strengths of the study include the use of multiple cohorts with participants from the UK, Iceland, and China. The diversity of study samples demonstrates the generalisability of our findings, although genetic analyses were only performed in two of the cohorts. Notably, a pivotal strength of this study lies in the operationalisation of resilience through educational attainment. The consistency of our findings across three socioeconomically and culturally distinct populations—and across alternative outcome definitions—supports educational attainment as an objective and internationally harmonisable indicator of functional adaptation. By examining the genetic determinants of educational resilience, we extended the conceptualisation of resilience beyond traditional symptom-based metrics (e.g., absence of disorders) toward a broader framework centred on sustained functional performance in real-world settings. Moreover, the attenuated association in cohorts with greater educational accessibility and stronger social welfare systems (e.g., SAGA and UK Biobank cohorts) suggests that universal social welfare systems may partially buffer the long-term educational consequences of childhood adversity by promoting more equitable access to educational resources. Accordingly, academic attainment may function as both a key indicator of and a potential pathway toward resilience at the population level. These insights underscore the necessity of policies that integrate psychosocial risk screening with educational support to mitigate the long-term consequences of adversity in vulnerable populations.

Limitations of our study are as follows: first, all ACEs were retrospectively self-reported at recruitment, which may give rise to recall bias. Nevertheless, prior studies have indicated a reasonable validity of retrospective assessment of ACEs,^76,77 showing a moderate agreement between retrospective and prospective measures (correlation coefficient = 0.47, p < 0.001).⁷⁶ Second, the prevalence of childhood sexual abuse reported in the CSTC (1.5%) was significantly lower than established estimates for the Chinese population (∼9%).⁷⁸ The potential for underreporting might partly explain the different association observed for this ACEs subtype in the CSTC compared with the other two cohorts. Nevertheless, as this subgroup accounted for a very small proportion of the total sample, its influence on the overall ACE-education association remains minimal. Third, we must acknowledge that educational attainment is a complex and multifaceted phenotype shaped not only by individual adaptive capacities but also by socioeconomic conditions, educational systems, cultural expectations, and policy environments. As such, it does not represent a comprehensive measure of resilience across all domains (e.g., emotional, interpersonal, or occupational functioning), but rather a specific, outcome-based indicator of long-term functional adaptation within the educational domain. Fourth, we acknowledge that comprehensive data on early-life socioeconomic and health characteristics were limited. Although we adjusted for important confounders in the three cohorts (e.g., family environmental factors and PGS for educational attainment in the UK Biobank), we cannot rule out the possibility that other unmeasured confounders, such as social class and mental health before age 18,⁵ may have biased the estimates. The absence of these granular longitudinal data also precluded a formal quantitative exploration of the potential pathways between ACEs and educational attainment. Fifth, we acknowledge the inherent limitations of the residual-based GWAS approach, including the possibility that residuals remain partially correlated with the underlying educational attainment phenotype. To mitigate this concern, we adjusted for polygenic predisposition to educational attainment. Further, the identified lead variant is not previously associated with general educational attainment, suggesting a resilience-specific rather than baseline educational effect mechanism. Crucially, the signal was validated through a formal gene–environment interaction test (p for interaction <0.0001), demonstrating a context-dependent effect and reinforcing that the identified locus is unlikely to be an artifact of the residualization procedure. Last, although we cross-validated the phenotypic association between ACEs and educational attainment across three cohorts, this study lacks external replication of the GWAS findings. The limited availability of large, independent cohorts with both genomic data and detailed, harmonised ACE measures comparable to those used here precluded a conventional replication analysis. Accordingly, our genetic results should be interpreted with caution and require confirmation in independent samples to establish their generalisability.

In conclusion, we noted a strong association between ACEs and educational attainment in sample of multiple ethnicities, independent of individual genetic propensity to educational attainment. A potential novel locus, rs2743239, was identified for educational resilience after ACEs, which was mapped to genes and pathways related to neuroprotective and immune responses. Notably, the TT genotype of rs2743239 was associated with educational resilience specifically after ACEs. If confirmed in independent samples, our findings could facilitate further mechanistic explorations aimed at improving risk management of ACEs survivors.

Contributors

HS, UAV and HY were responsible for the study's concept and design. AH, EBT, GT, JJ, TA, HY, and ZY did the data and project management. HY, YZ, and CH did the data cleaning and analysis. HY, HBD, HW, JS, FF, UAV and HS interpreted the data. HY, JS, FF, UAV and HS draughted the manuscript. All the authors read and approved the final manuscript and agree to be accountable for all aspects of the work. YZ have accessed and verified the underlying data in the UK Biobank and CSTC and HBD have accessed and verified the underlying data in the SAGA cohort.

Data sharing statement

Data from the UK Biobank (http://www.ukbiobank.ac.uk/) is available to all researchers upon making an application. This research was conducted using the UK Biobank Resource under Application 54,803 (approved on October 29, 2019). Data from the SAGA cohort is not publicly available due to stringent Icelandic data protection laws and the specific ethical approvals granted by the National Bioethics Committee (NBC) of Iceland. Given the sensitive nature of the information, data access is restricted to authorised scientific research and is subject to NBC approval (vsn@vsn.is). Qualified researchers may request access to de-identified data by submitting a formal research proposal to the SAGA cohort data management board (afallasaga@hi.is), which will facilitate the submission of the necessary ethics amendment to the NBC. The CSTC database is not accessible publicly because of privacy policy and data protection regulation of participants. However, access to de-identified data for legitimate scientific research may be granted upon reasonable request. Interested investigators should contact the organisation committee (via wchukb@wchscu.cn or songhuan@wchscu.cn) and submit a formal research proposal outlining their study objectives. GWAS summary statistics for educational resilience are publicly available at https://doi.org/10.5281/zenodo.18464967.

Declaration of interests

The authors declare no competing interests.