Alcohol misuse costs the United States hundreds of billions of dollars annually (Sacks et al., Reference Sacks, Gonzales, Bouchery, Tomedi and Brewer2015) and it is estimated that 95,000 people die each year from alcohol-related causes (Centers for Disease Control and Prevention, 2019). These rates of mortality make alcohol misuse the third leading preventable cause of death in the US, following tobacco use and poor diet/physical inactivity (Mokdad et al., Reference Mokdad, Marks, Stroup and Gerberding2005). Beyond these outcomes, alcohol use is associated with multiple types of cancer, other chronic health conditions (Yoon, Reference Yoon and Chen2018), and a range of mental health issues (Burns & Teesson, Reference Burns and Teesson2002), including alcohol use disorder (Moos et al., Reference Moos, Schutte, Brennan and Moos2004). Developing a better understanding of the etiology of alcohol misuse is key in improving effectiveness of prevention and intervention efforts.

To improve understanding of the etiology of alcohol misuse one must think about risk for alcohol misuse in the context of development. Generally, alcohol use begins in early to mid-adolescence, approximately 14–17 years old (Chen & Jacobson, Reference Chen and Jacobson2012), becoming more normative in adolescence and early adulthood, approximately 18–20 years old. These general trends of increased alcohol use across adolescence are impacted by early adolescent traits that can forecast propensity for use/misuse during adolescence and into adulthood. Extant theory and literature have found that externalizing behavior, including aggression, delinquency, and behavioral disinhibition, is a developmental precursor to substance use in adolescence (e.g., Brook et al., Reference Brook, Whiteman, Finch and Cohen1996; Iacono et al., Reference Iacono, Malone and McGue2008; Sitnick et al., Reference Sitnick, Shaw and Hyde2014). Specifically, some individuals may inherit a liability for externalizing behavior, which manifests prior to the onset of alcohol use. Early in life, the expression of this propensity for externalizing behavior is broad and non-specific, but over the course of adolescence it can result in liability for alcohol use as a result of previous externalizing behavior or due to exposure to certain environmental factors (e.g., parents, peers) as a result of this broad liability (Iacono et al., Reference Iacono, Malone and McGue2008). Further, there is heterotypic continuity in these externalizing facets over time, that is, changing manifestation of individual externalizing behaviors across development based on an underlying shared etiology. In particular, externalizing behaviors such as aggression, delinquency, and behavioral disinhibition are most common in early to mid-adolescence and substance use becomes more common in adolescence into early adulthood as externalizing behaviors normatively subside (Samek et al., Reference Samek, Hicks, Keyes, Iacono and McGue2017; Tielbeek et al., Reference Tielbeek, Vink, Polderman, Popma, Posthuma and Verweij2018; Zellers et al., Reference Zellers, Corley, Thibodeau, Kirkpatrick, Elkins, Iacono, Hopfer, Hewitt, McGue and Vrieze2020). Thus, externalizing behavior in early adolescence is predictive of alcohol use in adolescence once drinking becomes more normative and both behaviors may emerge due to underlying etiological factors. One factor that may underlie both externalizing behavior in adolescence and later alcohol use is genetic predisposition.

Liability for both externalizing behavior and alcohol use can be due to genetic predisposition. There is a robust literature indicating a largely shared genetic etiology underlying alcohol use, substance use, aggression, antisocial behavior, and disinhibition (Barr et al., Reference Barr, Salvatore, Wetherill, Anokhin, Chan, Edenberg, Kuperman, Meyers, Nurnberger, Porjesz, Schuckit and Dick2020; Derringer et al., Reference Derringer, Corley, Haberstick, Young, Demmitt, Howrigan, Kirkpatrick, Iacono, McGue and Keller2015; Gizer et al., Reference Gizer, Otto, Ellingson, Beauchaine and Hinshaw2016; Kendler et al., Reference Kendler, Prescott, Myers and Neale2003; Krueger et al., Reference Krueger, Hicks, Patrick, Carlson, Iacono and McGue2002; McGue et al., Reference McGue, Zhang, Miller, Basu, Vrieze, Hicks, Malone, Oetting and Iacono2013; Vrieze et al., Reference Vrieze, McGue, Miller, Hicks and Iacono2013; Waldman et al., Reference Waldman, Rhee, LoParo, Park and Patrick2018; Young et al., Reference Young, Friedman, Miyake, Willcutt, Corley, Haberstick and Hewitt2009). This broad “externalizing” factor has been found to have a large heritable component (e.g., heritability = .80–.85; Hicks et al., Reference Hicks, Schalet, Malone, Iacono and McGue2011; Krueger et al., Reference Krueger, Hicks, Patrick, Carlson, Iacono and McGue2002). The high heritability of the broad externalizing factor, on which each of the individual behaviors loads, has led many to conceptualize these lower order factors as “facets” of this broader construct. Further, both broad and specific genetic predisposition for externalizing behaviors increases risk for developing externalizing behaviors. For instance, one study found general genetic liability across conduct disorder, adult antisocial behavior, alcohol dependence, and drug dependence increased one’s risk for developing any/all of these externalizing phenotypes in parent-to-child transmission (Hicks et al., Reference Hicks, Krueger, Iacono, McGue and Patrick2004). Conversely, specific liability for individual disorders, such as alcohol use disorder, increased transmission of distinct behaviors in siblings. Moreover, these genetic effects can vary by age.

Twin and polygenic studies indicate that genetic effects on facets of externalizing behavior and alcohol use can vary with age (Elam et al., Reference Elam, Clifford, Shaw, Wilson and Lemery-Chalfant2019, Reference Elam, Ha, Neale, Aliev, Dick and Lemery-Chalfant2021; Kendler et al., Reference Kendler, Gardner and Dick2011; van der Laan et al., Reference van der Laan, Morosoli-García, van de Weijer, Colodro-Conde, Lupton, Mitchell, McAloney, Parker, Burns, Hickie and Pool2021). Collectively, this work illustrates that genetic influences on externalizing behavior peak in early to mid-adolescence then decline over time, whereas genetic influences on alcohol use increase to a peak in early adulthood and attenuate thereafter. Further, recent findings indicate that polygenic scores explain greater variance in alcohol use in early adulthood compared to adolescence as incidence of alcohol use increases (Elam et al., Reference Elam, Ha, Neale, Aliev, Dick and Lemery-Chalfant2021; Kandaswamy et al., Reference Kandaswamy, Allegrini, Plomin and von Stumm2021). However, it is unclear whether this is due to increases in genetic effects or increased prevalence of alcohol use due to greater exposure and access to alcohol across adolescence so genetic effects are more easily detected, or both. What is apparent is that genetic effects over time mirror normative patterns of externalizing behavior and alcohol use when examined within construct. Less research has examined associations across these constructs over time, that is, whether genetic predisposition for externalizing behaviors is associated with substance use, and vice versa, but associations are plausible given overlap in liability for these behaviors.

Early genetics research has demonstrated the GABRA2 gene is associated with alcohol use in adulthood (e.g., Agrawal & Bierut, Reference Agrawal and Bierut2012). Extending this finding to younger ages and different externalizing constructs, GABAergic genes and related polymorphisms have been found to be associated with lower behavioral control in late childhood and greater rule breaking from mid- to late-adolescence, which both mediated effects on later substance use in adolescence (Trucco, Villafuerte, et al., Reference Trucco, Colder, Wieczorek, Lengua and Hawk2014; Trucco et al., Reference Trucco, Hicks, Villafuerte, Nigg, Burmeister and Zucker2016). Using a twin design, common genetic influences contributed to externalizing behavior in adolescence and substance use disorder in early adulthood (Samek et al., Reference Samek, Hicks, Keyes, Iacono and McGue2017). When examining polygenic scores, genetic predisposition for smoking has been associated with externalizing at ages 11, 14, and 17 (Hicks et al., Reference Hicks, Clark, Deak, Liu, Durbin, Schaefer, Wilson, Iacono and Vrieze2021). In a related study, polygenic scores for alcohol, cannabis, and smoking were broadly associated with alcohol use, cannabis use, nicotine use, and a broader substance use factor, cumulatively measured from ages 14 to 24 (Schaefer et al., Reference Schaefer, Jang, Clark, Deak, Hicks, Iacono, Liu, McGue, Vrieze and Wilson2021). When examining this issue in a longitudinal model, cannabis use disorder and regular smoking polygenic scores were associated with behavioral disinhibition at age 11, which predicted latent substance use across age 14 to 24. Problematic alcohol use and regular smoking polygenic scores directly predicted latent substance use. Finally, using polygenic scores for risky behavior, associations were found with alcohol and cannabis use in early adulthood, but not with antisocial behaviors (Ksinan et al., Reference Ksinan, Smith, Barr and Vazsonyi2022).

It is important to note that little genetic research has examined associations with externalizing behavior and alcohol use in diverse populations. Twin and molecular genetic research have found converging and unique influences on externalizing behavior and alcohol use in African Americans and European Americans (Brick et al., Reference Brick, Keller, Knopik, McGeary and Palmer2019; Sartor et al., Reference Sartor, Nelson, Lynskey, Madden, Heath and Bucholz2013; Webb et al., Reference Webb, Edwards, Wolen, Salvatore, Aliev, Riley, Sun, Williamson, Kitchens, Pedersen and Adkins2017). Most recently, different time-varying patterns of polygenic influence on alcohol use were found for African Americans and European Americans from adolescence to adulthood (Elam et al., Reference Elam, Ha, Neale, Aliev, Dick and Lemery-Chalfant2021). Specifically, effects emerged earlier in adulthood and declined more sharply for African Americans, whereas effects in European Americans emerged slightly later and remained elevated. This is supported by evidence that there is variation in genetic effects across genetic ancestry (Martin et al., Reference Martin, Gignoux, Walters, Wojcik, Neale, Gravel, Daly, Bustamante and Kenny2017, Reference Martin, Kanai, Kamatani, Okada, Neale and Daly2019). Thus, it is important to examine genetic effects on externalizing behavior and alcohol use separately in African Americans and European Americans.

Collectively, accumulating evidence appears to indicate that genetic predisposition for externalizing behavior and substance use are primarily associated with externalizing behaviors in early- to mid-adolescence and substance use in adolescence and early adulthood. We add to this literature by examining polygenic predisposition for alcohol use disorder as a predictor of (1) externalizing behavior in early adolescence and alcohol use in adolescence, and (2) externalizing behavior in adolescence and alcohol use in early adulthood.

In the current study, we created polygenic scores for alcohol use disorder (AUD-PGS) using summary statistics from two large genome-wide association studies (GWAS) on alcohol use disorder in African Americans (AAs) and European Americans (EAs) (Kranzler et al., Reference Kranzler, Zhou, Kember, Vickers Smith, Justice, Damrauer, Tsao, Klarin, Baras, Reid, Overton, Rader, Cheng, Tate, Becker, Concato, Xu, Polimanti, Zhao and Gelernter2019), allowing for investigation of developmental associations in distinct ethnic subgroups. We chose to examine polygenic scores for alcohol use disorder given previous evidence that polygenic scores for various substance use disorders and problematic substance use predict subclinical substance use, externalizing and impulsive behavior, and sensation seeking in adolescence and early adulthood, possibly because they better capture genetic signals across constructs and developmental periods (Johnson et al., Reference Johnson, Sanchez-Roige, Acion, Adams, Bucholz, Chan, Chao, Chorlian, Dick, Edenberg, Foroud, Hayward, Heron, Hesselbrock, Hickman, Kendler, Kinreich, Kramer, Kuo and Edwards2021; Li et al., Reference Li, Savage, Kendler, Hickman, Mahedy, Macleod, Kaprio, Rose and Dick2017; Salvatore et al., Reference Salvatore, Aliev, Bucholz, Agrawal, Hesselbrock, Hesselbrock, Bauer, Kuperman, Schuckit, Kramer, Edenberg, Foroud and Dick2015; Schaefer et al., Reference Schaefer, Jang, Clark, Deak, Hicks, Iacono, Liu, McGue, Vrieze and Wilson2021).

We examined this research question using two mediation models in which the AUD-PGS predicted externalizing behavior (“a” path) and alcohol use (“c” path), and externalizing behavior predicted alcohol use (“b” path). This conceptual framework is in-line with past research examining polygenic prediction of substance use via earlier behavior and allows us to examine genetic prediction of both externalizing behavior and alcohol use while examining for genetically mediated effects (e.g., Li et al., Reference Li, Savage, Kendler, Hickman, Mahedy, Macleod, Kaprio, Rose and Dick2017; Schaefer et al., Reference Schaefer, Jang, Clark, Deak, Hicks, Iacono, Liu, McGue, Vrieze and Wilson2021; Trucco, Villafuerte, et al., Reference Trucco, Villafuerte, Heitzeg, Burmeister and Zucker2014, Trucco et al., Reference Trucco, Hicks, Villafuerte, Nigg, Burmeister and Zucker2016). In Model 1, the AUD-PGS was considered a predictor of externalizing behavior in early adolescence and alcohol use in adolescence, and externalizing behavior a predictor of alcohol use. We examined this in the Early Steps Multisite sample (referred to as Early Steps hereafter) which is a randomized control trial of a family centered intervention, the Family Check-Up (FCU), that has assessed individuals from age 2 to 16. In Model 2, the AUD-PGS was considered a predictor of externalizing behavior in adolescence and alcohol use in early adulthood, and externalizing behavior a predictor of alcohol use. We examined this in the Project Alliance 1 (PAL1) sample which is a randomized control trial of the same FCU intervention and has assessed individuals from ages 12 to 31. We examined both models separately in AA and EA participants. Intervention condition did not have direct effects on externalizing behavior or alcohol use, thus, intervention effects were controlled for but not a focus of the study based on the relatively small sample sizes.

We hypothesized that the AUD-PGS would predict externalizing behavior in early adolescence with attenuated or absent associations with alcohol in adolescence, and that externalizing behavior would predict alcohol use. We also hypothesized that the AUD-PGS would predict alcohol use in early adulthood, but associations with externalizing behavior in adolescence would be attenuated or absent, and that externalizing behavior would predict alcohol use.

Method

Participants and procedures

Early Steps sample

The Early Steps sample is a longitudinal randomized trial of 731 ethnically and racially diverse, low-income families with 2-year-old children. Families were recruited between 2002 and 2003 from Women, Infants, and Children Nutritional Supplement Programs at three sites in metropolitan Pittsburgh, Pennsylvania (urban), Eugene, Oregon (suburban), and within and outside Charlottesville, Virginia (rural). Given this recruitment strategy this sample is considered at-risk. Screening procedures were used to recruit families of toddlers at high risk for conduct problems, based on socio-demographic risk, primary caregiver risk, and toddler behavior problems. Participation rates of those families invited to participate who qualified by risk status were high across the three sites [83.2% total (49% female); 84% in Eugene (n = 271), 76% in Charlottesville (n = 188), and 88% in Pittsburgh (n = 272)]. Primary caregivers (96% mothers) self-identified as belonging to the following ethnic groups: 11% Latino, 28% African American, 54% European American, 4% biracial, and 3% other groups (e.g., Native American, Asian American, Pacific Islander). For more information about sample characteristics see Dishion et al. (Reference Dishion, Shaw, Connell, Gardner, Weaver and Wilson2008).

Similar to the PAL1 sample, families were randomly assigned to control or intervention conditions of the FCU after the baseline assessment at child age 2. All families were re-contacted at child ages 3, 4, 5, 7.5, 8.5, 9.5, 10.5, 14, and 16 years (81% of the sample participated at age 16) for home-based assessments with primary caregivers (96% biological mothers at age 2) and children. Primary caregivers completed questionnaires regarding the physical and socio-cultural environment and children’s behavior. Of the adolescents who participated at 14 years, 515 were genotyped (86.7% of the sample who participated in home visits at age 14) and were used in the present study. The final analytic sample consisted of 477 individuals after accounting for missing demographic covariates and within this sample there were 223 EA participants and 138 AA participants available. Selective attrition analyses revealed no significant differences between members of the initial sample and those who were genotyped with respect to ethnicity, gender, age, study site, aggressive/disinhibited behavior, or alcohol use.

PAL1 sample

The PAL1 sample is a longitudinal randomized trial of 999 adolescents and their families recruited in 1996-1997 in Portland, Oregon. Participants were randomized to intervention and control conditions of the FCU which is designed to reduce adolescent problem behavior by improving parenting and family functioning. All adolescents in 6th grade at three middle schools were invited to participate (of which 90% consented). Children were initially assessed at 11–12 years old, after which the intervention was administered, followed by four annual assessments (waves 1–5). Additional assessments were administered at 16–17, 18–19, 23–24, 24–25, 26–27, 28–30 years old (referred to as waves 6–11). At waves 1–5 a stratified sampling technique based on teacher assessments of risk status on externalizing behavior and substance use was used to classify participants as either no risk, at-risk, or high risk, and only those at high risk were subsequently assessed on externalizing measures. All individuals were assessed on all measures in waves 6–11. Retention at the last completed wave of data collection was excellent (83%). The sample was later genotyped as part of an age 26–27 assessment. Participants self-reported as 51% male, 44% European American, 30% African American, 13% multiracial, 6% Latino, 4% Asian American, and 4% other groups (Native American, Pacific Islander). A representative subsample of the genotyped individuals with data available on externalizing behavior and alcohol use comprise the sample for the current study (n = 631). For more information see Dishion & Kavanagh (Reference Dishion and Kavanagh2003).

All PAL1 participants were assessed at ages 17 and 19 (waves 6 and 7) so the adolescent-early adult model (Model 2) represented the full continuum of risk. The final analytic sample for the adolescent-early adult model consisted of 551 individuals after accounting for missing demographic covariates and within this sample there were 240 EA participants and 169 AA participants available.

Across both samples, selective attrition analyses revealed no significant differences between members of the initial sample with no genetic data and those who were genotyped with respect to ethnicity, gender, risk indices, or aggressive/disinhibited behaviors. In the PAL1 sample there was evidence of a greater proportion of females (p < .001) and some evidence of lower alcohol use on some indices (alcohol use frequency in early adulthood [p = .007]), but not quantity or 5 drinks in a row in early adulthood (ps > .29). Intervention condition did not have direct effects on externalizing behavior or alcohol use, thus, intervention effects were controlled for but not a focus of the study based on the relatively small sample size.

All study protocols for both the Early Steps and PAL1 samples were approved by the University’s Institutional Review board. Parent or guardian consent was obtained for all minors and adolescents provided assent for participation in the study, while adult participants provided their own consent. Families were compensated for their time at each age.

Genotyping procedures

For both Early Steps and PAL1 samples, DNA was collected using the Oragene saliva collection kits and extracted according to Oragene’s recommended procedures. Genotyping was performed at Rutgers University Cell and DNA Repository using the Affymetrix BioBank Array. Single nucleotide polymorphisms (SNPs) that are palindromic with ambiguous effect directions (A/T or C/G), SNPs with a genotyping rate of <.95, SNPs that did not pass Hardy-Weinberg equilibrium (p < 10–6), or SNPs with a minor allele frequency <.01 were excluded. Given our relatively small sample sizes, we opted to use these quality control thresholds to maximize the number of SNPs available for polygenic score formation as the PRS-CSx method (see below) significantly drops the number of SNPs used in the final score through the use of external reference panels.

Measures

AUD-PGS in Early Steps and PAL1 samples

We leveraged discovery GWAS summary statistics for alcohol use disorder from both AA and EA ancestries from Kranzler et al. (Reference Kranzler, Zhou, Kember, Vickers Smith, Justice, Damrauer, Tsao, Klarin, Baras, Reid, Overton, Rader, Cheng, Tate, Becker, Concato, Xu, Polimanti, Zhao and Gelernter2019). From these GWAS, we used information for those of European Ancestry (n = 202K) as well as those of African Ancestry (n = 57K) as the discovery GWAS. To create polygenic scores, we used PRS-CSx (Ruan et al., Reference Ruan, Lin, Feng, Chen, Lam, Guo, He, Sawa, Martin, Qin and Huang2022), a Python based command line tool that integrates GWAS summary statistics and external linkage disequilibrium (LD) reference panels from multiple populations to improve cross-population polygenic prediction. Posterior SNP effect sizes are inferred under coupled continuous shrinkage (CS) priors across populations. PRS-CSx is an extension of the Bayesian polygenic prediction method PRS-CS (Ge et al., Reference Ge, Chen, Ni, Feng and Smoller2019) method and uses LD information from 1000 Genomes Project European and African reference panels and estimates the posterior effect sizes for SNPs in a given set of GWAS summary statistics. Empirical tests and simulations have shown improved PRS-CS and PRS-CSx methods improve predictive power beyond traditional methods of polygenic construction (Ge et al., Reference Ge, Chen, Ni, Feng and Smoller2019; Ruan et al., Reference Ruan, Lin, Feng, Chen, Lam, Guo, He, Sawa, Martin, Qin and Huang2022). Because PRS-CS uses LD information from an external reference panel, we matched ancestries between the discovery samples and the ancestry reference provided by PRS-CS. Summary statistics for AAs and EAs AUD-PGS were drawn from each respective alcohol use disorder GWAS. AUD-PGS were calculated using joint modeling across GWAS summary statistics via coupled shrinkage priors (Ruan et al., Reference Ruan, Lin, Feng, Chen, Lam, Guo, He, Sawa, Martin, Qin and Huang2022). Final AUD-PGS were based on posterior PRS-CSx weights and created using the score procedure in PLINK 1.9 (Chang et al., Reference Chang, Chow, Tellier, Vattikuti, Purcell and Lee2015).

Population stratification and genetic admixture

Principal Components Analysis were conducted to represent population admixture separately in the Early Steps sample, using PLINK (Purcell et al., Reference Purcell, Neale, Todd-Brown, Thomas, Ferreira, Bender, Maller, Sklar, de Bakker, Daly and Sham2007), and in the PAL1 sample, using snpgdsPCA function from R SNPRelate package (Zheng et al., Reference Zheng, Levine, Shen, Gogarten, Laurie and Weir2012). For both samples, the first 20 principal components (PCs) were extracted and residualized from the AUD-PGS in each respective sample.

Externalizing behavior

Early Steps sample

At age 14 externalizing behavior was assessed using the Child Behavior Checklist (Achenbach et al., Reference Achenbach, Dumenci and Rescorla2001). The 10-item attention, 18-item aggression, and 17-item rule breaking subscales were used to index externalizing behavior (0 = Not True to 2 = Often True; “Destroys own things.” “Disobedient at home”; α range .82–.93). An externalizing latent variable was formed using these subscales as indicators.

PAL1 sample

At age 17 externalizing behavior was assessed using adolescent self-report on the Early Adolescent Temperament Questionnaire (Rothbart et al., Reference Rothbart, Ahadi, Hershey and Fisher2001). 16 items from the activation control, attention, and inhibitory control subscales were conceptualized as a single effortful control index (α = .72) and along with the 6-item aggression subscale (α = .75) were used as two indicators of externalizing behavior (1 = Almost always untrue to 5 = Almost always true; “When I get angry I throw or break things,” “When someone tells me to stop doing something, it is easy for me to stop”). Adolescents also reported on their behavioral responses to stressful situations using the life events coping scale inventory (Dise-Lewis, Reference Dise-Lewis1988), which assesses behavioral responses to stressful situations and the delinquency/aggression subscale was used in the present study (1 = I would definitely do this to 5 = I would definitely not do this; “When I am stressed I get in a fight with someone,” α = .87). An externalizing latent variable was formed using these three scales as indicators.

Alcohol use

Early Steps sample

At 16 years old alcohol use was assessed by separately asking about quantity and frequency of consumption of alcohol (e.g., “How often did you drink beer in the last 3 months?”; 0 = Never to 7 = 2–3 times a day or more; “How much beer did you drink in the past 3 months?”; 0 = Less than one can to 5 = More than five cans). A separate item assessed the largest ever number of alcoholic drinks a participant had consumed in a 24 hour period in the last 3 months, an indicator of problematic drinking. These three measures, alcohol quantity, frequency, and largest number of drinks, were used as indicators of an alcohol use latent variable. There were low levels of alcohol use at age 16 with 84% reporting “less than one can” for quantity of alcohol use, 88% reporting “never” for frequency of alcohol use, and 86% reporting zero for the largest number of drinks. Means and standard deviations for these scales can be found in Table 1.

Table 1.

Means, standard deviations, and correlations among primary constructs in the Early Steps sample for African Americans (below the diagonal) and European Americans (above the diagonal)

Note. AUD-PGS = Alcohol use disorder polygenic risk score. First 20 ancestry principal components were residualized from the AUD-PGS.

⁺ p < .10; *p < .05; **p < .01; ***p < .001.

PAL1 sample

At 19 years old alcohol use was assessed asking about quantity and frequency of consumption of alcohol (e.g., “How often did you drink beer in the last 3 months?”; 0 = Never to 7 = 2–3 times a day or more; “How much beer did you drink in the past 3 months?”; 0 = Less than one can to 5 = More than five cans). A separate item assessed number of times in the past 3 months the participant had consumed five or more drinks in a row, an indicator of problematic drinking. These three measures, alcohol quantity, frequency, and five in a row, were used as indicators of an alcohol use latent variable. There were moderate levels of alcohol use with no individuals reporting “less than one can” for quantity of alcohol use, 18% reporting “never” for frequency of alcohol use, and 33% reporting zero for having five drinks in a row. Means and standard deviations for these scales can be found in Table 2.

Table 2.

Means, standard deviations, and correlations among primary constructs in the PAL1 sample for African Americans (below the diagonal) and European Americans (above the diagonal)

Note. AUD-PGS = Alcohol use disorder polygenic risk score. First 20 ancestry principal components were residualized from the AUD-PGS.

⁺ p < .10; *p < .05; **p < .01; ***p < .001.

Covariates

Participant gender, age, and intervention group status were included as covariates in all analyses, as was location site in Model 1 with the Early Steps sample. Given both samples participated in the FCU, we also included an AUD-PGS by intervention interaction term to account for possible interaction effects.

Statistical Analyses

Given the recruitment strategy in Early Steps sample, Model 1 constituted tests in high-risk participants. Conversely, Model 2 using PAL1 participants represented those individuals across the continuum of risk. We performed sensitivity analyses by also investigating Model 2 in at-risk and high-risk individuals.

After examining descriptive statistics and bivariate correlations, study variables were examined in a multigroup path model in Mplus. Where indicated by significant AUD-PGS to externalizing behavior and externalizing behavior to alcohol use effects, indirect effects were estimated using Rmediation (Tofighi & MacKinnon, Reference Tofighi and MacKinnon2011) which computes asymmetric confidence limits for the distribution of the product of alpha and beta paths, and has greater statistical power and better adjustment of Type I error when compared to the Sobel test. Model 1 and Model 2 were run in the AA and EA ethnic subgroups within each sample. We refrained from testing for genetic differences across AA and EA ethnic subgroups given known differences in genetic ancestry (e.g., allele frequency and LD patterns).

Full information maximum likelihood was used to handle missing data. Descriptive statistics were conducted in SPSS and all other analyses were conducted in Mplus Version 8.3.

Results

Means, standard deviations, and correlations are presented for the Early Steps sample in Table 1 separately for AAs (below the diagonal) and EAs (above the diagonal), and for the PAL1 sample in Table 2 separately for AAs (below the diagonal) and EAs (above the diagonal). In Early Steps, the AUD-PGS was positively associated with alcohol frequency in adolescence in EAs. In PAL1, the AUD-PGS was positively associated with alcohol frequency in EAs. Broadly, as indicated in Tables 1 and 2, in AAs and EAs in both samples the indices of externalizing and alcohol use were correlated within construct, with some evidence of associations across constructs and over time.

Results for Model 1 in early adolescence to adolescence are presented in Table 3. Model Fit was adequate (X ² (71) = 144.11, p < .001, RMSEA = .07, CFI = .95, TLI = .93). In the Early Steps sample loadings for the alcohol and externalizing latent factors were significant and adequate in both AA and EA participants (.59–.97). In the EA subgroup, the AUD-PGS was associated with externalizing behavior. In the AA subgroup, externalizing behavior was associated with alcohol use. In neither AAs nor EAs was there evidence of associations from the AUD-PGS to externalizing behavior and externalizing behavior to alcohol use so indirect tests were not conducted. The variance explained in EA externalizing behavior and alcohol use were small (R ² = .05; R ² = .05).

Table 3.

Standardized coefficients of polygenic scores for alcohol use disorder predicting externalizing and alcohol use in Early Steps and PAL1 samples

Note. AUD-PGS = Alcohol use disorder polygenic risk score. First 20 ancestry principal components were residualized from the AUD-PGS. Associations with significance at p < .05 are bolded.

Results for Model 2 in adolescence to early adulthood can be found in Table 3. Model fit was adequate (X ² (78) = 118.37, p = .002, RMSEA = .04, CFI = .91, TLI = .90). In the PAL1 sample, loadings for the alcohol and externalizing latent factors were significant and adequate in both AA and EA participants (.41–.83). In the EA subgroup, the AUD-PGS was associated with alcohol use. No effects were detected in the AA subgroup. In neither AAs nor EAs was there evidence of associations from AUD-PGS to externalizing behavior and externalizing behavior to alcohol use so indirect tests were not conducted. The variance explained in EA externalizing behavior and alcohol use were small (R ² = .18; R ² = .03).

As a sensitivity test, we ran Model 2 in those designated as at-risk and high risk in the initial wave of the study to mirror the sample in Model 1, results can be found in supplemental Table 1. In the EA subgroup, the AUD-PGS was associated with alcohol use. In neither AAs nor EAs was there evidence of associations from AUD-PGS to externalizing behavior and externalizing behavior to alcohol use so indirect tests were not conducted.