Introduction

Externalizing psychological problemsi in youth refer to oppositional behavior, conduct problems, attention-deficit/hyperactivity disorder (ADHD), and substance misuse (Krueger et al., 2021; Lahey et al., 2004), which are associated with adverse consequences for the individual and society (Costello, Egger, & Angold, 2005; Lahey, 2021). We examine potential biological factors in externalizing problems using integrated genetic, neuroimaging, and behavioral methods. Although social factors undoubtedly play a role in externalizing problems, twin studies reveal that externalizing problems are substantially heritable and share their genetic influences with other problems (Allegrini et al., 2020; Beauchaine, Hinshaw, & Pang, 2010; Cosgrove et al., 2011; Du Rietz et al., 2021; Pettersson et al., 2019). To assess heritable influences at the individual level, polygenic scores have been developed that quantify aspects of genetic risk for psychological problems captured by the weighted sum of many single nucleotide polymorphisms (SNPs; Wray et al., 2021). A polygenic score for externalizing problems (extPGS) has been developed from data on approximately 1.5 million individuals (Karlsson Linner et al., 2021) using genomic structural equation modeling to identify genetic variance common to externalizing phenotypes, including ADHD, alcohol misuse, tobacco and cannabis use, and risky sexual behavior. In independent tests, the extPGS explained 10.5% of the variance in a broad externalizing factor, 1.65% of the variance in ADHD, 3.10% of the variance in conduct disorder, and 1.96% of the variance in oppositional defiant disorder (Karlsson Linner et al., 2021).

Previous studies have found inverse associations between concurrently measured externalizing problems and brain structure (Cao et al., 2022; Durham et al., 2021; Fairchild et al., 2019; Hoogman et al., 2017, 2019; Mackey et al., 2019; Mooney et al., 2021; Patel, Parker, Salum, Pausova, & Paus, 2022; Zhang et al., 2022). A meta-analysis by the ENIGMA ADHD Working Group identified widespread cortical and subcortical structural differences in those with ADHD compared with controls (Hoogman et al., 2017, 2019). Critically, gray matter volume and other metrics are substantially heritable (Albaugh et al., 2019; Mooney et al., 2021; Pol, Peper, Boomsma, & Kahn, 2007; Zhao et al., 2019), raising the possibility that genetic variants that influence risk for externalizing problems may do so partly by influencing individual differences in brain structure related to risk for externalizing problems.

Note that previous cross-sectional analyses have attempted to map hypothesized mediated biological risk pathways from genetic polymorphisms to individual differences in brain structure and to psychological problems (Forbes et al., 2021; Hagler et al., 2019). These initial tests ofmediation were severely limited by the use of variables measured at the same points in time, however, because concurrently measured variables bias estimates of mediation and do not rule out reverse causation (Maxwell, Cole, & Mitchell, 2011; Shrout, 2011). Therefore, the present analyses use prospective data to test statistical mediation from the extPGS through brain structure to psychological problems measured 1 and 2 years later.

Methods

Sample

Data were from Release 4.0 of the longitudinal Adolescent Brain Cognitive DevelopmentSM Study (ABCD Study
) conducted at 21 sites on 11,876 from 9- to 10-year-old participants at baseline.

Ethical considerations

Data were collected at each site under institutional review board approval from each participating university for sharing de-identified data. Waivers were obtained for analyses of the public use de-identified data from the Institutional Review Boards of the University of Chicago and Vanderbilt University.

Measures

The Child Behavior Checklist (CBCL; Achenbach, 2009) is a parent rating scale describing child psychological problems on a scale of 0 = not true (as far as you know), 1 = somewhat or sometimes true, or 2 = very true or often true. We used a subset of CBCL items based on factor analyses that eliminated CBCL items that did not address psychological problems, such as biting fingernails, constipation, and wishing to be the opposite sex, as well as items with low endorsements in late childhood (Moore et al., 2020). Confirmatory correlated factors (CF) models defined conduct problems, ADHD, and internalizing factors (Table S1) and confirmatory bifactor models defined a general factor and specific conduct problems, ADHD, and internalizing factors (Table S2). The 66 items used in the present analyses and their factor loadings during each annual assessment are in these tables.

Polygenic score (PGS)

Saliva was collected from children as per the ABCD Study biospecimen protocol. Genomic DNA was genotyped at Rutgers University using the NIDA Affymetrics Smokescreen array (Baurley, Edlund, Pardamean, Conti, & Bergen, 2016). Following quality control (https://nda.nih.gov/study.html?id=634), genotyping data consisted of 10,627 samples and 517,724 genotyped SNPs. Samples were removed for problematic genotyping plates (n = 56); low call rates (n = 126); poorly called or monomorphic SNPs (<90%); and reported versus DNA sex mismatches (n = 10). Distribution of SNP and sample call rates were inspected; 155 samples with call rates <97% were excluded. Furthermore, 2,195 SNPs were excluded due to violation of the Hardy–Weinberg equilibrium with p < 10
10. Quality control used PLINK (1.9; Purcell et al., 2007), after which 8,326 individuals and 462,767 SNPs remained for imputation. The Michigan imputation server pipeline v1.2.4 (https:// imputationserver.sph.umich.edu/index.html#!) conducted genome-wide imputation using Minimac 4 with the Haplotype Reference Consortium panel r1.1 as the reference population. Subsequently, SNPs with minor allele frequency <1% or low imputation quality (R2 < .8) were removed. Additionally, 12 samples were identified as outliers in a population structure plot (StataCorp, 2019) using principal component analysis in the R 4.1.0 prcomp package (RCoreTeam, 2021) and were removed from analysis, leaving 8,314 samples and 8,574,976 SNPs for PGS analysis. The extPGS was calculated using effect sizes from summary statistics for the general externalizing factor using PRS-CS (Karlsson Linner et al., 2021). PRS-CS uses a Bayesian regression and continuous shrinkage method to correct for the nonindependence among nearby SNPs. SNPs in the extPGS were limited to those from HapMap3 that overlapped between the original GWAS summary statistics and the LD reference panel (1,000 Genomes EUR reference panel).

Brain image acquisition, processing, and quality assurance

The ABCD Study structural imaging protocol was described previously (Casey et al., 2018). Protocols were harmonized for all scanner platforms. Scanning included 3D T1- and 3D T2- weighted images of brain structure, taking place over one or two scanning sessions. The 21 data collection sites across the United States used several models of 3 tesla (3 T) Siemens, General Electric, or Philips scanners; specific scanner models used for data collection were General Electric Discovery MR750, Siemens Prisma, Siemens Prisma Fit, Philips Achieva dStream, and Philips Ingenia. Imaging parameters were repetition time 2,400 to 2,500 ms; echo time 2 to 2.9 ms; field of view (FOV) 256 9 240 to 256; FOV phase of 93.75% to 100%; matrix 256 9 256; 176 to 225 slices; inversion delay 1,060 ms; flip angle of 8°; voxel resolution of 1 9 1 9 1 mm; total acquisition time from 5 min 38 s to 7 min 12 s.

The ABCD Data Analysis and Informatics Center performed centralized processing and analysis of structural data using a collection of processing steps via the Multi-Modal Processing Stream, a software package developed and maintained at the Center for Multimodal Imaging and Genetics at the University of California, San Diego (UCSD). This pipeline consisted of: (a) preprocessing [correction for gradient nonlinearity distortions, intensity scaling and inhomogeneity correction, registration to an averaged reference brain in standard space, and manual quality control (QC)]; (b) brain segmentation (cortical surface reconstruction and subcortical segmentation performed based on automated, atlas-based, segmentation procedures in FreeSurfer v5.3); (c) derivation of morphometric measures [calculation of average volume in each cortical parcel of the standard FreeSurfer Desikan parcellation scheme (Desikan et al., 2006) and in each subcortical region (Fischl et al., 2002)]; and finally, (d) postprocessing QC (manual review for motion, intensity inhomogeneity, white matter underestimation, pial overestimation, and magnetic susceptibility artifact). Detailed descriptions of these methods have been published (Casey et al., 2018; Hagler et al., 2019).

Statistical analyses

Because the extPGS was derived from data on only persons of European ancestries, we included only genotyped children who self-identified as non-Hispanic white, were identified as European ancestry based on principal components from the 1,000 Genomes Project, and who passed quality assurance for both genotyping and structural magnetic resonance imaging. Demographic characteristics of the analyzed subsample are in Table S3. The analyses used longitudinal data from three annual assessments of 4,475 children of European ancestry at baseline, 4,343 in the first follow-up, and 4,092 in the second follow-up. The extPGS, brain volumes, and child problem factors were standardized by z-transformation on children of European ancestry prior to analyses.

We examined possible predictive relationships from genetic risk to individual differences in brain to psychological problems using two complementary measurement models to define the dimensions of psychological problems (Forbes, Greene, et al., 2021; Moore et al., 2020). One of the most important properties of psychological problems is that they are positively correlated (Forbes et al., 2021; Krueger et al., 2021; Lahey, 2021; Watson et al., 2022). Therefore, CF models usefully define dimensions of psychological problems while preserving such correlations. In contrast, bifactor measurement models are valuable for a different purpose. Rather than defining clinical phenotypes, bifactor models are a statistical tool for testing the hierarchical hypothesis that some risk factors are specific to each particular dimension of problems, whereas other risks are nonspecifically shared by all psychological problems (Caspi et al., 2014; Lahey, Krueger, Rathouz, Waldman, & Zald, 2017; Neumann et al., 2022; Zald & Lahey, 2017). To accomplish this, bifactor models partition the variance common to all psychological problems from the variance that is unique to particular subsets of problems to define orthogonal general and specific factors. Thus, bifactor measurement models are well suited for the present analyses of genetic and neural correlates of psychological problems. That is, they allow valid tests to determine whether polygenic risk for externalizing problems is prospectively associated with variance that is specific only to the dimension of ADHD and/or to the dimensions of conduct problems, or is associated with the variance shared by all psychological problems, or both (Laceulle, Chung, Vollebergh, & Ormel, 2020; Lahey et al., 2017; Lahey, Van Hulle, Singh, Waldman, & Rathouz, 2011; Waszczuk et al., 2020). Because some researchers prefer only to use CF models (Watts, Poore, & Waldman, 2019), however, results based on both measurement models were used.

Associations between the extPGS and measured GMVs were tested using Statistical Analysis Software (SAS; https://www. sas.com) procedures for complex surveys. Generalized linear regression models took clustering within families and stratification by sites into account and incorporated poststratification weights (Heeringa & Berglund, 2018). Response variables were 68 cortical regions, which were derived from a surface-based atlas procedure previously developed (Desikan et al., 2006), and 19 automatically labeled subcortical regions (Fischl et al., 2002). Associations of the extPGS and brain volume measures with latent dimensions of psychological problems in each assessment wave were estimated in Mplus (Muthen &
Muthen,
1998-2017), taking clustering and stratification into account. These adjustments were needed to calculate appropriate standard errors because children within the same families and data collection sites are likely to be more similar to one another than randomly selected children in the population. Poststratification weights (Heeringa & Berglund, 2018) based on the known demographics of the US population to adjust estimates of parameters to be more representative of the population.

Parent-reported sex, age in months, and the first 10 principal components for ancestry were included in all analyses as covariates and MRI model, and manufacturer was a covariate in analyses of brain volumes. Associations of the extPGS with total cortical gray matter volume, total subcortical gray matter volume, and total cortical volume were tested separately. Subsequent tests of associations between the extPGS and the 68 cortical regions and 19 subcortical regions were conducted with and without total intracranial volume (TICV) as a covariate to evaluate the possibility that any associations with volumes were global rather than regional.

As shown in Figure 1, all direct and indirect paths between extPRS and brain volumes measured at baseline, and problem behavior dimensions in each of the two annual follow-up assessments were tested using structural equation models in Mplus (Muthen & Muth
en,
1998-2017). Direct and indirect paths from the extPGS to each latent problem dimension in each of the three annual assessments through TGMV were estimated in Mplus using 99% bootstrapped confidence intervals. Paths were estimated in separate analyses for problem factors defined using either CF or bifactor models.

Longitudinal growth models were estimated in Mplus (Muthen & Muth
en,
1998-2017) using data from each annual assessment to test the associations of the extPGS with the intercept and slope of the general factor and the specific conduct problems, ADHD, and internalizing factors. Factor scores were extracted from bifactor models of parent problem ratings in Mplus separately for each wave then converted to age-based data. Longitudinal data were available on 2,321 (9 years), 4,068 (10 years), 3,629 (11 years), 1,866 (12 years), and 310 (13 years) participants.

As stated in each table of results, family-wise error rates (Benjamini & Hochberg, 1995; Pena, Habiger, & Wu, ~ 2011) were adjusted for false discovery in six families of analyses of related tests of significance. Interactions between sex and the extPRS were tested in exploratory analyses but were not significant for any outcome in any model.

Results

Model psychometrics

CF and bifactor models each fit the CBCL data well in all waves (Tables S2 and S3). Table S4 shows that the general and specific factors defined in bifactor models in all three waves had acceptable H indices >0.70 (Hancock & Mueller, 2001), and each specific factor was reliable according to omega statistics. Explained common variance and omegaH indicated that the robust general factor explained more than half of the estimated variance in each specific factor. Factor determinacies showed that all factors were well defined (Rodriguez, Reise, & Haviland, 2016). Note that the factor loadings on the ADHD factors and the H index for ADHD in both models in every wave were lower than other factors, suggesting that the measurement of ADHD was less strong than other factors.

Polygenic score prediction of behavior

Wave-by-wave analyses. When correlated dimensions of psychological problems were defined in CF models, the extPGS concurrently and prospectively predicted conduct problems (R2 = 2.0%–2.9%) and ADHD problems (R2 = 1.4%–1.7%), but not internalizing problems in each of the three annual assessments (Table 1, upper rows). Because the variance in each factor defined in CF models reflects both the variance specific to that factor and the variance that it shares with the other CF, estimates of association with extPRS were repeated for problem dimensions defined in bifactor models. These analyses reveal a more complex picture. When problem dimensions were defined using bifactor models, extPGS positively predicted both the general (R2 = .8%–1.4%) and specific conduct problems factors (R2 = 1.7%– 2.1%), and inversely predicted the specific internalizing factor (R2 = .2%–.6%), in each wave. The extPGS only inconsistently predicted the specific ADHD factor defined in bifactor modeling across waves (Table 1, lower rows).

Longitudinal analyses. Longitudinal growth models revealed that the extPGS predicted the levels (intercepts) of the general factor and the specific conduct problems, ADHD, and internalizing factors defined in bifactor models across 9 through 13 years of age, but not the rates of change (slopes) of any problem factor with increasing age (Figure 2; Table S5). Coefficients for the significant associations of extPGS with each factor were positive, except for the inverse associations with specific internalizing problems.

Polygenic score prediction of brain structure

Significant associations were found in separate analyses between the extPGS and total cortical GMV [b =
.06 (SE = .02), p < .0001], total subcortical GMV [b =
.04 (SE = .02), p < .0115], and total GMV in the entire brain [b =
.06 (SE = .02), p < .0001]. The extPGS accounted for 0.36% of variance in TGMV. The upper rows of Table 2 present the results of separate complementary analyses at the regional level. The extPGS was significantly inversely associated with GMV in 37 cortical and subcortical regions after FDR correction. When TICV was added as a covariate, however, the extPGS was significantly associated with only one regional GMV after FDR correction (lower rows of Table 2), suggesting that the association of the extPGS with gray matter volume is more global than regional.

Prediction of psychological problems by total gray matter volume

Table 3 shows that TGMV was inversely associated concurrently and prospectively with conduct problems (R2 = 3.1%–4.7%), ADHD (R2 = 1.7%–3.0%), and internalizing problems (R2 = <1.0%–1.0%) defined in CF models in all annual waves, adjusting for multiple testing. Using bifactor models, TGMV was associated inversely with both the general factor (R2 = 1.6%–3.1%) and the specific conduct problems factor (R2 = 1.7%–2.1%) in all waves, but not with specific ADHD. In addition, specific internalizing problems were modestly associated in the opposite direction with TGMV in each wave.

Tests of statistical mediation

Because the extPGS was found to be associated with both TGMV and conduct problems, and TGMV was found to be associated with conduct problems, formal tests were conducted of indirect statistical paths from the extPGS through TGMV to each problem dimension in each wave (see Figure 1). In all three annual waves, the direct and indirect paths from extPGS to both conduct problems and ADHD defined in CF models were ‘significant’. That is, the 99% confidence intervals for both the direct and the indirect paths from extPGS to conduct problems and to ADHD did not include zero in any wave (top rows of Table 4). When bifactor modeling was used (lower rows of Table 4), the 99% confidence intervals for both the direct and the indirect paths from extPGS through TGMV to specific conduct problems similarly did not include zero in any wave. The proportion of the association between extPRS and a problem dimension mediated by the indirect path can be calculated as the indirect effect divided by the total effect (sum of the direct and indirect paths). Thus, the proportions of the total observed association between extPGS and specific conduct problems that were mediated by TGMV in the three annual assessments were 7.3%, 9.8%, and 6.6%, respectively.

Sensitivity tests

To check for potential inflation of R2 values due to overfitting, we performed 10,000 random split-half (50%/50%) cross-validation estimations of the direct effects within the mediation model, in which the model was estimated in one half and used in the second half to predict the bifactor-defined problem scores in the first follow-up. Correlations between true and predicted values of problem dimensions were compared between training and testing sets. As illustrated in the histograms in Figure S1, the generalizability of prediction between training and testing sets was nearly perfect, indicating little or no inflation. Prediction of the specific ADHD factor did produce some anomalous bimodal results toward the lower end of the distribution, but this was limited to the direct effect because the R2 values for the indirect effect were near-zero and there was not a significant predictive effect to cross-validate.

Because polygenic scores capture only part of the genetic influences on a phenotype, we also evaluated potential bias in the estimated mediated path due to genetic confounding using Gsens (Pingault et al., 2022). Two complementary external estimates of h2 based on SNP heritability (Choi et al., 2022), and twin studies (Hicks, Krueger, Iacono, McGue, & Patrick, 2004) were used in these tests of bias. As shown in Table S6, there was significant genetic confounding of the mediated relationship between extPGS and specific conduct problems in all three annual assessments based on both estimates of h2 . Nonetheless, all estimates of the mediated associations remained statistically significant (p range .005 to 6.56e
97) after adjustment for genetic confounding.

Discussion

The present findings confirm and extend evidence from earlier twin studies of genetic influences on child externalizing problems (Rhee & Waldman, 2002) using longitudinal analyses and polygenic score methods, which are based on different assumptions than cross-sectional twin studies. In particular, the present findings support the utility of the extPGS as a measure of part of the genetic influences on conduct problems quantifiable from common SNPs (Wray et al., 2021) by independently replicating its association with childhood conduct problems in a new sample (Karlsson Linner et al., 2021). Furthermore, the present longitudinal analyses showed that prospective associations of the extPGS with conduct problems endured over 3 years in the same youth.

Figure 2 Growth modeling revealed significant associations between the externalizing polygenic score (extPGS) and the intercepts (levels), but not the slopes of change across age, for each general and specific latent factor defined in bifactor measurement models (inversely for internalizing). These findings are illustrated using standardized scores for repeated measures within children on the (a) general factor, (b) specific conduct problems, (c) attention-deficit disorder (ADHD), and (d) internalizing factors for parent-rated psychological problems across 9 through 13 years of age in four groups defined by the child’s externalizing extPGS (
1: ≤ 1 SD below mean; below 0: >
1 SD and <0; above 0: ≥ 0 and <1 SD; +1: ≥1 SD above mean). Standard error bars are presented for the
1 and +1 SD polygenic score groups

The present findings of a significant association between the extPGS and TGMV also confirm twin studies (Pol et al., 2007; Zhao et al., 2019), and previous studies using other polygenic scores (Alemany et al., 2019; Alnaes et al., 2019), that brain volume is substantially heritable. Crucially, our findings expand on previous cross-sectional findings that smaller TGMV is associated with greater child behavior problems (Kaczkurkin et al., 2020; Snyder, Hankin, Sandman, Head, & Davis, 2017) by revealing robust prospective associations of TGMV with both the variance specific to conduct problems and variance shared by all problem dimensions when they are measured at different points in time during the adolescent transition.

Both the extPGS and TGMV were associated with conduct problems and ADHD when these dimensions were defined in CF models (Tables 1 and 3). This is important, but it does not reveal if these variables were separately associated with each dimension for different reasons or were associated with the variance shared by the two dimensions. Results based on bifactor models resolved these alternative explanations. The extPGS was associated with the general factor of psychological problems over time (Figure 2; Table 1; Table S5), indicating that the extPGS predicted the variance shared by all measured problems. Nonetheless, the extPGS also was associated with the orthogonal specific conduct problems factor over time. Similar results were found for associations with TGMV (Table 3). Thus, both extPGS and TGMV were associated with conduct problems through both the variance specific to conduct problems and through the variance that conduct problems shared with other dimensions. In contrast, associations of extPGS and TGMV with ADHD were primarily through the variance shared with all problem dimensions. Alternatively, this finding could reflect the weaker measurement of ADHD than the other dimensions.

Notably, a comparison of the results of analyses based on CF and bifactor measurement models revealed potentially important findings on internalizing problems. Consistent with previous hypotheses that low levels of anxiety are associated with conduct problems (Frick, Lilienfeld, Ellis, Loney, & Silverthorn, 1999), the extPGS polygenic score was associated inversely with variance uniquely associated with specific internalizing problems, but only in bifactor models that partitioned the variance that is not shared by internalizing with other problems.

Tests of mediated pathways

The present analyses were the first to combine polygenic scores, brain structure, and longitudinal data on child psychological problems in the same models to examine potential temporally ordered risk pathways from genome to brain to behavior. Formal tests of statistical mediation found evidence that the association of the extPGS with conduct problems was significantly mediated by TGMV in all waves when conduct problems were defined in either CF or bifactor models. In contrast, no significant mediation of the association between the extPGS and the general factor was found (Table 4). Rather, most of the concurrent and prospective associations between the extPGS and psychological problems were found to be direct, rather than being mediated by TGMV. That is, these findings suggest that TGMV statistically mediates the association of extPGS with only the variance that is specific to conduct problems and not the variance that conduct problems share with other problem dimensions. The genetic sensitivity analyses suggested that the observed mediation by TGMV of the association between extPGS and specific conduct problems in all assessment waves was not an artifact of bias. Confidence in this finding is substantial because the twin-based estimate used was quite high (h2 = 0.81), providing a conservative test of genetic confounding.

Limitations and future directions

Because the current analyses were based only on children of European ancestry, the findings are not generalizable to other populations. Individuals of non-European ancestry were excluded because the extPGS was derived from samples of individuals of only European descent (Karlsson Linner et al., 2021). Polygenic scores generally underperform when there is mismatch in ancestries between the discovery and target samples, which could result in misrepresentations of the role of polygenic risk in other populations (Duncan et al., 2019; Martin et al., 2019). It is imperative, however, that future molecular genetic studies address the role of genetic risk for psychological problems in non-European populations using appropriate methods.

Across the age span of 9–13 years when many children are entering puberty, the extPGS was associated with levels of problem behaviors over time, but not the rate of change in these problems with increasing age (Figure 2). Because greater changes in psychological problems with age are expected later in adolescence, however, it is imperative to address this issue again when additional waves of the ABCD Study data are available.

Although the present findings are revealing, they cannot support the inference that variations in the SNPs that define the extPGS contribute causally to variations in brain, which causally influence risk for conduct problems. This is because differences in SNPs and brain structure between children from different families are confounded by many potential environmental differences between families, such as maltreatment and economic deprivation (D’Onofrio, Sjolander, Lahey, Lichtenstein, & Oberg, 2020; Lahey & D’Onofrio, 2010). To infer causation, it will be necessary to move from case–control and other designs that compare different children from different families to within-family designs that can break the confounding of genes and brain with unmeasured environmental variables. Nonetheless, it is encouraging that Karlsson Linner et al (Karlsson Linner et al., 2021) conducted tests of the associations of the extPGS with externalizing phenotypes in very large samples of adults of European ancestry that contain many siblings and found few differences in the magnitudes of associations of the extPGS with adult externalizing behavior when estimated in comparisons conducted within families versus between families. Additionally, a polygenic score for educational attainment was found to significantly predict impairing alcohol, nicotine, and marijuana use in a large sample of young adults using less confounded sibling comparisons (Salvatore et al., 2020). There are insufficient numbers of siblings in the ABCD Study sample to conduct adequately powered within-family tests, but the results of the present analyses encourage future investments in research designs that can map causal paths from genetic variation through brain to significant psychological problems. This would advance understanding of the biological nature of psychological problems and could ultimately lead to improved strategies of prevention and intervention.

Summary

This study investigated the potential biological factors involved in externalizing problems in youth, specifically focusing on the association between genetic risk, brain structure, and psychological problems. The research utilized longitudinal data from the Adolescent Brain Cognitive Development (ABCD) Study, a large-scale, multi-site investigation of brain development and behavior in adolescents.

Introduction

Externalizing problems, including oppositional behavior, conduct problems, attention-deficit/hyperactivity disorder (ADHD), and substance misuse, are prevalent in youth and associated with adverse consequences for individuals and society. While social factors contribute to these issues, twin studies indicate substantial heritability and shared genetic influences among externalizing problems. Polygenic scores (PGSs), which quantify genetic risk for complex traits, have been developed to assess heritable influences at the individual level. A polygenic score for externalizing problems (extPGS) was derived from a large dataset of approximately 1.5 million individuals and has been shown to explain a portion of the variance in externalizing phenotypes.

Previous research has demonstrated inverse associations between externalizing problems and brain structure, specifically reduced gray matter volume (GMV). Given the heritability of both externalizing problems and brain structure, researchers hypothesized that genetic variants influencing externalizing problems might act through their impact on brain structure.

Previous cross-sectional studies have investigated the hypothesized mediated biological pathway from genetic polymorphisms to brain structure to psychological problems. However, these studies were limited by the use of concurrently measured variables, which can bias mediation estimates and introduce the possibility of reverse causation. This study addressed these limitations by utilizing prospective data to test statistical mediation from the extPGS through brain structure to psychological problems measured at one and two years later.

Methods

Sample

The study utilized data from the ABCD Study, which included 11,876 participants aged 9-10 years at baseline.

Ethical considerations

Data collection at each site was approved by the relevant institutional review board, and waivers were obtained for analyses of the public use de-identified data from the Institutional Review Boards of the University of Chicago and Vanderbilt University.

Measures

The Child Behavior Checklist (CBCL) was used to assess child psychological problems based on parent ratings. Confirmatory correlated factor (CF) models were used to define dimensions of conduct problems, ADHD, and internalizing factors. Confirmatory bifactor models were used to define a general factor and specific conduct problems, ADHD, and internalizing factors.

Polygenic score (PGS)

Genomic DNA was collected from participants and genotyped using the NIDA Affymetrics Smokescreen array. After quality control, data from 8,314 individuals and 8,574,976 SNPs remained for PGS analysis. The extPGS was calculated using effect sizes from summary statistics for the general externalizing factor using PRS-CS, a method that uses a Bayesian regression and continuous shrinkage approach to account for non-independence among SNPs.

Brain image acquisition, processing, and quality assurance

The ABCD Study structural imaging protocol involved acquiring 3D T1- and 3D T2-weighted images of brain structure at multiple sites using 3 tesla (3T) Siemens, General Electric, or Philips scanners. The ABCD Data Analysis and Informatics Center performed centralized processing and analysis of structural data using a multi-modal processing stream pipeline that included preprocessing, brain segmentation, derivation of morphometric measures, and post-processing quality control.

Statistical analyses

The analyses included only genotyped children who self-identified as non-Hispanic white, were identified as European ancestry based on principal components from the 1,000 Genomes Project, and who passed quality assurance for both genotyping and structural magnetic resonance imaging. Longitudinal data from three annual assessments were analyzed for 4,475 children of European ancestry at baseline, 4,343 in the first follow-up, and 4,092 in the second follow-up.

Associations between the extPGS and measured GMVs were tested using generalized linear regression models that accounted for clustering within families and stratification by sites, incorporating poststratification weights. Associations of the extPGS and brain volume measures with latent dimensions of psychological problems were estimated in Mplus, taking clustering and stratification into account. Poststratification weights were used to adjust estimates to be more representative of the population.

The direct and indirect paths between extPGS and brain volumes measured at baseline and problem behavior dimensions in each of the two annual follow-up assessments were tested using structural equation models in Mplus.

Longitudinal growth models were estimated in Mplus using data from each annual assessment to test the associations of the extPGS with the intercept and slope of the general factor and the specific conduct problems, ADHD, and internalizing factors.

Family-wise error rates were adjusted for false discovery in families of analyses of related tests of significance. Interactions between sex and the extPRS were tested in exploratory analyses.

Results

Model psychometrics

CF and bifactor models fit the CBCL data well in all waves. The general and specific factors defined in bifactor models had acceptable H indices and reliability according to omega statistics, indicating that the factors were well defined. The factor loadings on the ADHD factors and the H index for ADHD were lower than other factors, suggesting that the measurement of ADHD was less strong than other factors.

Polygenic score prediction of behavior

Wave-by-wave analyses. The extPGS concurrently and prospectively predicted conduct problems and ADHD problems when dimensions of psychological problems were defined in CF models. The extPGS predicted both the general and specific conduct problems factors, and inversely predicted the specific internalizing factor, in each wave when dimensions were defined in bifactor models.

Longitudinal analyses. Longitudinal growth models revealed that the extPGS predicted the levels (intercepts) of the general factor and the specific conduct problems, ADHD, and internalizing factors defined in bifactor models across 9 through 13 years of age.

Polygenic score prediction of brain structure

The extPGS was significantly inversely associated with total cortical GMV, total subcortical GMV, and total GMV in the entire brain. The extPGS was significantly inversely associated with GMV in 37 cortical and subcortical regions after FDR correction. When TICV was added as a covariate, however, the extPGS was significantly associated with only one regional GMV after FDR correction, suggesting that the association of the extPGS with gray matter volume is more global than regional.

Prediction of psychological problems by total gray matter volume

TGMV was inversely associated concurrently and prospectively with conduct problems, ADHD, and internalizing problems defined in CF models in all annual waves. Using bifactor models, TGMV was inversely associated with both the general factor and the specific conduct problems factor in all waves.

Tests of statistical mediation

The association of the extPGS with conduct problems was significantly mediated by TGMV in all waves when conduct problems were defined in either CF or bifactor models. No significant mediation of the association between the extPGS and the general factor was found. The genetic sensitivity analyses suggested that the observed mediation by TGMV of the association between extPGS and specific conduct problems was not an artifact of bias.

Limitations and future directions

The study was limited to individuals of European ancestry and did not address the role of genetic risk for psychological problems in other populations. Future research should investigate these issues with appropriate methods.

Further limitations include the reliance on parent reports for assessing psychological problems, the focus on a specific age range, and the inability to infer causation due to confounding factors in the study design.

Future research should address these limitations by incorporating diverse populations, utilizing multiple methods of assessing psychological problems, examining a broader age range, and employing more robust research designs.

Discussion

The present findings provide evidence for the role of genetic influences in childhood externalizing problems, specifically conduct problems, and support the utility of the extPGS as a measure of genetic risk. The findings also highlight the importance of TGMV in mediating the association between the extPGS and specific conduct problems.

The study underscores the need for further research into the complex interplay between genetic risk, brain structure, and psychological problems, with the goal of advancing understanding of the biological underpinnings of these conditions and informing prevention and intervention strategies.

Summary

This research examines the influence of genetic predisposition on externalizing problems in youth, specifically focusing on conduct problems, attention-deficit/hyperactivity disorder (ADHD), and internalizing problems. Utilizing the longitudinal Adolescent Brain Cognitive DevelopmentSM Study (ABCD Study), researchers investigated the relationship between a polygenic score for externalizing problems (extPGS), brain structure, and psychological problems measured over three years.

Methods

The study included 4,475 children of European ancestry from the ABCD Study. The extPGS, a measure of genetic risk for externalizing problems, was calculated using data from approximately 1.5 million individuals. Brain volumes, including total cortical gray matter volume (TGMV), were assessed using structural magnetic resonance imaging (MRI). Parent-reported psychological problems were measured using the Child Behavior Checklist (CBCL).

Statistical Analyses

The study employed various statistical methods, including confirmatory factor (CF) models, bifactor models, and longitudinal growth models. Associations between the extPGS, TGMV, and psychological problems were analyzed, accounting for factors such as clustering within families and site stratification.

Results

The extPGS significantly predicted conduct problems and ADHD, both concurrently and prospectively. The extPGS also showed a positive association with the general factor of psychological problems, indicating a shared influence across problem dimensions. TGMV was inversely associated with conduct problems, ADHD, and internalizing problems, suggesting that smaller brain volume may be linked to greater risk for these issues. Notably, mediation analyses indicated that TGMV significantly mediated the association between the extPGS and conduct problems, but not the general factor of psychological problems.

Discussion

The findings support previous research demonstrating the heritability of externalizing problems and brain structure. The study highlights the potential influence of genetic factors on brain development and subsequently on behavioral outcomes. While the results do not establish causality, they provide evidence for a potential pathway from genetic predisposition to brain structure and ultimately to psychological problems. Further research is needed to investigate the underlying mechanisms and explore potential intervention strategies.

Summary

This study explores the connection between genetic predisposition for externalizing problems (like conduct issues, ADHD, and substance abuse), brain structure, and subsequent psychological problems in adolescents.

Introduction

Researchers are increasingly interested in understanding the biological factors that contribute to externalizing problems in young people. While social factors undoubtedly play a role, twin studies have shown that these behaviors have a strong genetic component. Polygenic scores (PGS) are tools that measure genetic risk by analyzing variations in DNA. A specific PGS for externalizing problems (extPGS) has been developed and found to explain a significant portion of the variation in externalizing behaviors like ADHD, conduct disorder, and substance use.

Previous studies have shown an inverse relationship between externalizing problems and brain structure, specifically gray matter volume (GMV). Since GMV is also significantly influenced by genetics, researchers believe that genetic variants associated with externalizing problems might work through their effects on brain structure.

This study builds upon earlier work by using longitudinal data, meaning they track the same individuals over time, to investigate whether the extPGS influences brain structure, which in turn predicts later externalizing problems. This design allows for stronger inferences about cause-and-effect relationships.

Methods

The study used data from the Adolescent Brain Cognitive Development (ABCD) Study, a large-scale longitudinal study of nearly 12,000 participants. Data included behavioral measures (parent reports of psychological problems using the Child Behavior Checklist), genetic information (extPGS calculated from DNA samples), and brain imaging data (magnetic resonance imaging, MRI, scans).

Measures

Parent-reported information about their child's behavior using the CBCL was analyzed to assess conduct problems, ADHD, and internalizing problems (like anxiety and depression).

Polygenic Score (PGS)

The extPGS was calculated based on individual genetic data. The PGS was designed to capture genetic risk factors common to externalizing behaviors.

Brain Image Acquisition, Processing, and Quality Assurance

MRI scans were collected at various sites across the United States. Advanced techniques were used to process and analyze the brain images to measure specific brain regions and structures, including GMV.

Statistical Analyses

The study utilized various statistical methods to explore the relationships between the extPGS, brain structure, and psychological problems. These methods controlled for various factors like age, sex, and genetic ancestry.

Results

The study found significant associations between the extPGS and GMV. Specifically, higher extPGS scores were associated with lower GMV. Furthermore, lower GMV was consistently associated with higher levels of conduct problems, ADHD, and internalizing problems.

Tests of Statistical Mediation

The most important finding was that the association between the extPGS and conduct problems was significantly mediated by GMV. This means that the extPGS indirectly influences conduct problems through its effect on GMV. This mediation effect was consistent across all three annual assessments.

Discussion

The study provides further evidence for the role of genetic factors in externalizing problems in adolescents. The findings also demonstrate that the extPGS is a valuable tool for predicting future conduct problems. The study suggests that part of the influence of genetic predisposition on externalizing problems might be through changes in brain structure, particularly GMV.

Limitations and Future Directions

The study's findings are based on a sample of children of European ancestry. It is essential to replicate these findings in diverse populations. While the study found significant mediation effects, it is important to understand the underlying biological mechanisms involved. Future research should focus on identifying specific genes associated with these brain structural changes and their impact on behavior.

Summary

This study looked at the relationship between genes, brain structure, and behavioral problems in children. It used a large group of children and followed them over time to see how these factors changed.

Introduction

Scientists are interested in understanding the reasons why some children have more behavior problems than others. They know that both genes and environment play a role. This study focused on genes and brain structure.

Scientists created a "polygenic score" (PGS) for externalizing problems. This score estimates the risk of a child having problems like aggression or difficulty paying attention. They found that the PGS was related to the size of certain parts of the brain.

The study also found that the size of a part of the brain called the "gray matter" was related to the severity of behavior problems.

This study is important because it helps scientists understand how genes and brain structure might influence behavior problems in children. This knowledge could help them develop better ways to prevent and treat these problems.