%20-%20https://sketch.com%20--%3e%3ctitle%3e3ECBBF50-FD08-412A-943D-2CFC5E025A16%3c/title%3e%3cdesc%3eCreated%20with%20sketchtool.%3c/desc%3e%3cg%20id='Stylesheet'%20stroke='none'%20stroke-width='1'%20fill='none'%20fill-rule='evenodd'%3e%3cg%20transform='translate(-926.000000,%20-668.000000)'%20id='BUTTONS'%3e%3cg%20transform='translate(100.000000,%20597.000000)'%3e%3cg%20id='BTN-/-RIGHTS'%20transform='translate(816.000000,%2065.000000)'%3e%3cg%20id='ICON-/-Rights'%20transform='translate(10.000000,%206.000000)'%3e%3cpath%20d='M13.5752841,0%20L1.34943182,0%20C0.59375,0%200,0.586666667%200,1.33333333%20L0,22.6666667%20C0,23.4133333%200.620738636,24%201.34943182,24%20L17.6505682,24%20C18.40625,24%2019,23.3866667%2019,22.6666667%20L19,5.36%20L13.5752841,0%20Z%20M13.5752841,1.89333333%20L17.0838068,5.36%20L13.5752841,5.36%20L13.5752841,1.89333333%20Z%20M1.34943182,22.6666667%20L1.34943182,1.33333333%20L12.1988636,1.33333333%20L12.1988636,5.36%20C12.1988636,6.10666667%2012.7926136,6.69333333%2013.5482955,6.69333333%20L17.6235795,6.69333333%20L17.6505682,22.6666667%20L1.34943182,22.6666667%20Z'%20id='Shape'%20fill='%23595959'%20fill-rule='nonzero'%3e%3c/path%3e%3ccircle%20id='Oval'%20stroke='%23595959'%20stroke-width='1.5'%20fill='%23595959'%20cx='9.5'%20cy='13.5355339'%20r='5.5'%3e%3c/circle%3e%3cpolyline%20id='Path-2'%20stroke='%23FFFFFF'%20stroke-width='1.5'%20transform='translate(9.500000,%2013.035534)%20rotate(45.000000)%20translate(-9.500000,%20-13.035534)%20'%20points='8%2016.0355339%2011%2016.0355339%2011%2010.0355339'%3e%3c/polyline%3e%3c/g%3e%3c/g%3e%3c/g%3e%3c/g%3e%3c/g%3e%3c/svg%3e)
Excessive alcohol use is a significant public health problem, with higher levels of use associated with psychological disorders (e.g., anxiety), motor vehicle accidents, risky sexual behaviors, and physical illnesses (e.g., heart disease) (CDC, 2016). Experimentation with alcohol may occur during adolescence, with over 18% of adolescents nationwide reporting alcohol consumption prior to age 13 (Kann et al., Reference Kann, Kinchen, Shanklin, Flint, Hawkins, Harris and Zaza2014). Although infrequent and low levels of alcohol use among adolescents may have no untoward effects, more frequent and heavy alcohol consumption may result in structural and functional neurocognitive deficits that persist into adulthood (Clark, Thatcher, & Tapert Reference Clark, Thatcher and Tapert2008; Zeigler et al., Reference Zeigler, Wang, Yoast, Dickinson, McCaffree, Robinowitz and Sterling2005). Moreover, while heavier alcohol use during adolescence and young adulthood may be temporally limited for some, other youth who use alcohol during these developmental periods may develop serious alcohol use problems and impairments (Danielsson, Wennberg, Tengström, & Romelsjö, Reference Danielsson, Wennberg, Tengström and Romelsjö2010; Englund, Egeland, Oliva, & Collins, Reference Englund, Egeland, Oliva and Collins2008; Pitkänen, Kokko, Lyyra, & Pulkkinen, Reference Pitkänen, Kokko, Lyyra and Pulkkinen2008). Given that variations in alcohol use likely occur across development, it is important to identify the nature of such variation, along with individual and contextual factors that contribute to the development of heavy use and ultimately abuse and dependence. Such knowledge could be used to inform the development of effective preventive interventions directed at curtailing harmful drinking and its consequences.
Developmental course of alcohol use across adolescence and young adulthood
The use of alcohol during adolescence and young adulthood can be best understood through a developmental lens. Alcohol use shows age-related patterns which can be attributed to sociocultural differences, individual differences in vulnerability to using alcohol, and the developmental tasks (e.g., forming new friendships) and transitions (e.g., entering high school or the work force) associated with adolescence and young adulthood (Masten, Faden, Zucker, & Spear, Reference Masten, Faden, Zucker and Spear2009). Among Western societies, average trajectories of alcohol consumption involve increases in use in adolescence, peak levels of use in the early twenties, and declines in use in later adulthood (Chassin, Sher, Hussong, & Curran, Reference Chassin, Sher, Hussong and Curran2013). However, heterogeneity likely exists in these average trajectories given individual differences in opportunities to use alcohol and the ability to adapt to the numerous changes that occur during adolescence and young adulthood.
Studies examining alcohol use during adolescence and early adulthood have identified between three and eight trajectories (Chassin et al., Reference Chassin, Pitts and Prost2002; Flory, Lynam, Milich, Leukefeld, & Clayton, Reference Flory, Lynam, Milich, Leukefeld and Clayton2004; Nelson, Van Ryzin, & Dishion, Reference Nelson, Van Ryzin and Dishion2015; Su, Supple, Leerkes, & Kuo, Reference Su, Supple, Leerkes and Kuo2018; Tucker, Orlando, & Ellickson, Reference Tucker, Orlando and Ellickson2003; Zucker, Hicks, & Heitzeg, Reference Zucker, Hicks, Heitzeg and Cicchetti2016). Differences in study findings may be due to the baseline assessment of alcohol use, frequency of alcohol use assessments, and the questions used to assess alcohol use (i.e., monthly vs. annual accounts of alcohol use). Across these studies, parallels in the trajectories identified can be observed, with subgroups characterized by (a) low alcohol use during adolescence and young adulthood; (b) moderate alcohol use that is developmentally limited to middle adolescence and the early twenties; (c) alcohol use that begins in early adolescence that persists and intensifies in young adulthood; (d) alcohol use that begins in early adulthood with moderate use over time; and (e) alcohol use that begins in early adulthood characterized by heavier, more frequent alcohol use over time. These trajectories of alcohol use, however, reflect primarily European Americans that are socioeconomically diverse (e.g., Flory et al., Reference Flory, Lynam, Milich, Leukefeld and Clayton2004). Low-income, African American youth may experience different contextual stressors (e.g., higher levels of community disadvantage) that influence their alcohol use behaviors in a different manner than their European American, same-aged peers (Wallace & Muroff, Reference Wallace and Muroff2002; Wallace, Neilands, & Phillips, Reference Wallace, Neilands and Phillips2017), warranting an examination of alcohol use behaviors in this population.
Internalizing and externalizing pathways to alcohol use trajectories
Internalizing symptoms, a broad band category encompassing anxious and depressive symptoms (Achenbach, Reference Achenbach1991), represent an individual-specific factor that has been robustly associated with alcohol use among predominantly European adolescents and young adults (Fite, Colder, & O'Connor, Reference Fite, Colder and O'Connor2006; Hussong, Jones, Stein, Baucom, & Boeding, Reference Hussong, Jones, Stein, Baucom and Boeding2011; Stice, Myers, & Brown, Reference Stice, Myers and Brown1998). Whereas some work has indicated that phenotypic (i.e., observable) internalizing problems are positively related to alcohol and illicit drug use among late adolescents and young adults (e.g., Steele, Forehand, Armistead, & Brody, Reference Steele, Forehand, Armistead and Brody1995; Stice et al., Reference Stice, Myers and Brown1998), other work has linked higher internalizing symptoms to less frequent alcohol use during early and middle adolescence (e.g., Colder et al., Reference Colder, Scalco, Trucco, Read, Lengua, Wieczorek and Hawk2013; Fite et al., Reference Fite, Colder and O'Connor2006). Differences in study findings may be due to the timing in which alcohol use was assessed (e.g., early adolescence vs. late adolescence), as it is possible that internalizing symptoms may confer differential risk for alcohol consumption depending on the developmental stage. For example, youth experiencing internalizing symptoms in early adolescence may begin drinking heavily as a way of self-medicating (Chassin et al., Reference Chassin, Sher, Hussong and Curran2013; Crum Storr, Ialongo, & Anthony, Reference Crum, Storr, Ialongo and Anthony2008; Hussong et al., Reference Hussong, Jones, Stein, Baucom and Boeding2011). However, the untoward consequences of alcohol use in the form of parent/peer disapproval or impaired academic performance may result in declines in alcohol consumption over the late adolescent and young adult years (Crosnoe, Benner, & Schneider, Reference Crosnoe, Benner and Schneider2012; Mrug & McCay, Reference Mrug and McCay2013).
Considerable evidence also indicates that externalizing problems, particularly antisocial behaviors, are associated with alcohol use (Armstrong & Costello, Reference Armstrong and Costello2002; Cook, Pflieger, Connell, & Connell, Reference Cook, Pflieger, Connell and Connell2015; King, Iacono, & McGue, Reference King, Iacono and McGue2004). Higher levels of antisocial behaviors and behavioral disorder symptoms (e.g., conduct disorder, oppositional defiant disorder) predict early adolescent alcohol use (King et al., Reference King, Iacono and McGue2004), and heavy alcohol use and disorders in adulthood (Armstrong & Costello, Reference Armstrong and Costello2002; Lee, Winters, & Wall, Reference Lee, Winters and Wall2010; Trim et al., Reference Trim, Worley, Wall, Hopfer, Crowley, Hewitt and Brown2015). It has been suggested that externalizing disorders and substance use problems are a part of a larger externalizing syndrome, which is supported by work indicating strong, positive correlations between these conditions (King et al., Reference King, Iacono and McGue2004; Krueger, Markon, Patrick, & Iacono, Reference Krueger, Markon, Patrick and Iacono2005).
The role of genetics in alcohol use trajectories
Although internalizing and externalizing phenotypes have been associated with alcohol involvement across the developmental course, less is known about whether genetic variants underpinning these conditions are related to alcohol use behaviors. In samples of predominantly European adult twins, small to moderate positive correlations between internalizing disorders (i.e., depression) and alcohol use frequency and dependence have been observed (r’s ranging from .001 to .60) (Andersen et al., Reference Andersen, Pietrzak, Kranzler, Ma, Zhou, Liu and Han2017; Kendler et al., Reference Kendler1993; Tambs et al., Reference Tambs, Harris and Magnus1997; Torvik et al., Reference Torvik, Rosenström, Ystrom, Tambs, Røysamb, Czajkowski and Reichborn-Kjennerud2017). Other work has indicated that greater genetic propensity for externalizing symptoms is associated with membership in trajectories characterized by heavier alcohol use in adolescence and young adulthood among majority European adult twins (Alati et al., Reference Alati, Baker, Betts, Connor, Little, Sanson and Olsson2014; Kendler et al., Reference Kendler, Prescott, Myers and Neale2003; McAdams, Rowe, Rijsdijk, Maughan, & Eley, Reference McAdams, Rowe, Rijsdijk, Maughan and Eley2012; Wichers, Gillespie, & Kendler, Reference Wichers, Gillespie and Kendler2013). Findings from twin studies thus support the notion that genetic variants linked to internalizing symptoms and antisocial behaviors also underpin alcohol use.
There is increasing evidence, however, that substance use outcomes are influenced by numerous genetic variants that have very small effect sizes (McGue & Carey, Reference McGue, Carey, Tolan and Leventhal2016). One frequently used approach to capturing these polygenic influences is through the use of polygenic risk scores (PRSs). PRS are generated by identifying single nucleotide polymorphisms (SNPs) associated with a phenotype and weighting these SNPs based on the magnitude and strength of their association with a specific phenotype (Maher, Reference Maher2015). Despite empirical evidence that suggests polygenic contributions to the development of problematic alcohol use (Dick et al., Reference Dick, Barr, Guy, Nasim and Scott2017), few studies have investigated whether polygenic influences are associated with alcohol use subgroups among youth in adolescence and early adulthood. In a notable exception, Li et al. (Reference Li, Savage, Kendler, Hickman, Mahedy, Macleod and Dick2017) examined whether a PRS for alcohol abuse and dependence predicted the intercept and slope of alcohol use trajectories among a majority European sample assessed from ages 15 to 28. The authors found that, among males, a higher PRS for alcohol abuse and dependence predicted both greater alcohol use at age 15.5 and greater increases in drinking between the ages of 15.5 and 21.5 relative to males with a lower alcohol PRS (Li et al., Reference Li, Savage, Kendler, Hickman, Mahedy, Macleod and Dick2017).
Similar to the findings above, variations in an internalizing symptoms PRS may be associated with alcohol use over time. Youth higher in polygenic load for internalizing symptoms may experience (a) higher levels of negative affect and stress reactivity, and (b) rejection by mainstream peers due to poorer interpersonal skills, so they may subsequently seek out deviant peers to gain acceptance (Chassin et al. Reference Chassin, Sher, Hussong and Curran2013; Hussong et al. Reference Hussong, Jones, Stein, Baucom and Boeding2011). Higher levels of negative emotional states and/or involvement with deviant peers may increase the likelihood that youths with a higher internalizing symptoms PRS will use alcohol more frequently during early and middle adolescence. However, potential negative ramifications of alcohol use (e.g., parental reprimands, reduced scholastic performance) may result in a reduction in alcohol consumption later in adolescence and young adulthood given that youths with a higher internalizing symptoms PRS may be more affected by negative repercussions. Another possibility is that individuals with a higher internalizing symptoms PRS use alcohol in early adolescence and continue to use alcohol in young adulthood. These youths may reap the affective or interpersonal benefits of using alcohol and thus continue using this substance across developmental periods.
Variations in an antisocial behavior PRS may also be associated with alcohol use behaviors over time. During adolescence, increased autonomy from parents, greater centrality of the peer group, and normative increases in risk taking may set the stage for increased alcohol involvement (Steinberg, Reference Steinberg2007), and this may be particularly true among youths with a greater genetic propensity for antisocial behaviors. Individuals with a higher polygenic load for antisocial behaviors may be more prone to using alcohol during adolescence, as they may be (a) higher in novelty seeking, (b) higher in behavioral disinhibition, and (c) more likely to disregard negative consequences, all of which may exacerbate risk for experimenting with alcohol (Maneiro, Gómez-Fraguela, Cutrín, & Romero, Reference Maneiro, Gómez-Fraguela, Cutrín and Romero2017; Mann et al., Reference Mann, Engelhardt, Briley, Grotzinger, Patterson, Tackett and Harden2017). Higher polygenic load for antisocial behaviors may also confer risk for more frequent alcohol use during young adulthood, though research is insufficient to support this claim.
Contextual factors and alcohol use patterns
Although polygenic markers of internalizing symptoms and antisocial behavior may be associated with alcohol use over time, genetic loading for these problems likely interacts with environmental factors to influence risk for alcohol use. Consistent with bioecological models, the consideration of individual-specific features and proximal contextual influences, such as parental monitoring, is necessary to elucidate patterns of alcohol use behaviors that unfold across development (Bronfenbrenner & Ceci, Reference Bronfenbrenner and Ceci1994). Parental monitoring, defined as parental knowledge and supervision of children's activities (Dishion & McMahon, Reference Dishion and McMahon1998), has a substantial effect on the children's alcohol use across adolescence and young adulthood (Becker et al., Reference Becker, Spirito, Hernandez, Barnett, Eaton, Lewander and Monti2012). For example, lower levels of parental monitoring are associated with alcohol use trajectories characterized by higher levels of alcohol use during early and middle adolescence (Becker et al., Reference Becker, Spirito, Hernandez, Barnett, Eaton, Lewander and Monti2012). Higher levels of parental knowledge and monitoring have been associated with less favorable attitudes towards using drugs and higher self-efficacy in refusing drugs (Chuang, Ennett, Bauman, & Foshee, Reference Chuang, Ennett, Bauman and Foshee2009; Donaldson, Handren, & Crano, Reference Donaldson, Handren and Crano2016; Lac, Alvaro, Crano, & Siegel, Reference Lac, Alvaro, Crano and Siegel2009; Nash, McQueen, & Bray, Reference Nash, McQueen and Bray2005). Parental monitoring may also insulate children from affiliating with deviant peers and/or being exposed to illicit drug use, which may contribute to reduced adolescent alcohol use (Handren, Donaldson, & Crano, Reference Handren, Donaldson and Crano2016; Nash et al., Reference Nash, McQueen and Bray2005). Although parental monitoring is an important contextual factor that may protect against heavy alcohol use among youths, it is unclear whether this parenting behavior influences the alcohol use patterns of individuals varying in polygenic load for internalizing symptoms and antisocial behavior.
The consideration of distal contextual factors (e.g., community disadvantage) in relation to alcohol use over time is also paramount. Community disadvantage has been predictive of alcohol use among adolescents, with higher levels of neighborhood disorder and disadvantage positively predictive of alcohol involvement (Anderson, Sabatelli, & Kosutic, Reference Anderson, Sabatelli and Kosutic2007; Cambron, Kosterman, Catalano, Guttmannova, & Hawkins, Reference Cambron, Kosterman, Rhew, Catalano, Guttmannova and Hawkins2017; Lambert, Brown, Phillips, & Ialongo, Reference Lambert, Brown, Phillips and Ialongo2004). Neighborhood disadvantage may be particularly important to consider during adolescence, given that parents often grant their children more autonomy during this developmental period, resulting in greater exposure to extrafamilial influences such as the neighborhood and peers (Steinberg & Morris, Reference Steinberg and Morris2001). Exposure to neighborhoods characterized by residential segregation, lower community supervision of children's behaviors, and greater access to alcohol and illicit drugs may increase the likelihood of youths using alcohol at an earlier age and continuing to use alcohol into adulthood (Ross & Mirowsky, Reference Ross and Mirowsky2001; Wallace & Muroff, Reference Wallace and Muroff2002). Children residing in more disadvantaged communities may also experience demoralization and hopelessness (Ross & Mirowsky, Reference Ross and Mirowsky2001) and use alcohol during adolescence and young adulthood to reduce their negative cognitive and affective states (Rhodes & Jason, Reference Rhodes and Jason1990).
No studies to our knowledge have considered whether community disadvantage intensifies risk for prolonged, heavy alcohol use among children varying in internalizing symptoms and antisocial behavior polygenic load. However, one study found that the effect of a conduct disorder PRS on marijuana use disorders was only significant when its interaction with community disadvantage was included in the analytic model (Rabinowitz et al., Reference Rabinowitz, Musci, Milam, Benke, Sisto, Ialongo and Maher2018). In particular, individuals with a higher conduct disorder PRS were more likely to have a marijuana use disorder when exposed to higher levels of community disadvantage (Rabinowitz et al., Reference Rabinowitz, Musci, Milam, Benke, Sisto, Ialongo and Maher2018). Given the role of community disadvantage among individuals with higher polygenic load for conduct disorder, it is possible that community disadvantage may similarly confer risk for problematic alcohol use over time among individuals with higher polygenic load for internalizing symptoms and antisocial behavior.
The current study
In the present study, we sought to identify subgroups of alcohol use in adolescence and early adulthood (ages 14–26) among urban African Americans. This population of young people may experience environmental stressors (i.e., higher community disadvantage) that contribute to earlier and more frequent alcohol use over time relative to socioeconomically advantaged peers. Grounded in the theoretical and empirical scholarship regarding the developmental course of alcohol use, we expected to observe five subgroups of alcohol use: (a) an early-onset, increasing class characterized by heavy alcohol use in early adolescence that escalated in young adulthood; (b) an early-onset, decreasing alcohol use group characterized by increases in alcohol use in middle to late adolescence followed by declines in use in the mid-twenties; (c) a late-onset, moderate alcohol use group that exhibited low alcohol use in adolescence and moderate alcohol use in the early- and mid-twenties; (d) a late-onset, heavy alcohol class characterized by heavier and more frequent alcohol use beginning in late adolescence that extended to young adulthood; and (e) a subgroup of individuals that exhibited low alcohol use over time.
We also examined whether antisocial behavior and internalizing symptoms polygenic load (assessed via polygenic risk scores) were associated with alcohol use subgroup membership. We hypothesized that individuals with a greater PRS for antisocial behaviors would be more likely to belong to the early-onset, increasing subgroup relative to the (a) low use subgroup; (b) late-onset, moderate use subgroup; (c) late-onset, heavy use subgroup; and (d) early-onset, decreasing subgroup. These hypotheses are grounded in past research (Crum et al., Reference Crum, Storr, Ialongo and Anthony2008) that has linked higher levels of phenotypic antisocial behaviors to membership in classes characterized by heavy, frequent alcohol use during early adolescence and young adulthood relative to other subgroups (e.g., early-onset, decreasers; low users; late-onset, increasers) (e.g., Alati et al., Reference Alati, Baker, Betts, Connor, Little, Sanson and Olsson2014; Becker et al., Reference Becker, Spirito, Hernandez, Barnett, Eaton, Lewander and Monti2012; Wichers et al., Reference Wichers, Gillespie and Kendler2013). In terms of our expectations regarding the interaction between the antisocial behavior PRS and the contextual factors, we hypothesized that the higher the level of community disadvantage and the lower the level of parental monitoring, the greater the increase in the association between the antisocial behavior PRS and membership in the early onset, increasing class.
Despite the ambiguity in the literature with respect to the direction of the relationship between phenotypic internalizing symptoms and alcohol use, we expected that a higher internalizing symptoms PRS would be positively associated with membership in classes characterized by early alcohol use. These hypotheses are grounded in our past research using an African American sample, which revealed that higher levels of internalizing symptoms in late childhood were associated with increased alcohol use in adolescence (Crum et al., Reference Crum, Storr, Ialongo and Anthony2008), and other research that has linked internalizing symptoms to heavy alcohol and drug use among young adults (e.g., Steele et al., Reference Steele, Forehand, Armistead and Brody1995). We hypothesized that individuals with a higher internalizing symptoms PRS would be more likely to be a member of the early-onset, increasing class relative to the low use subgroup. We expected that exposure to low parental monitoring and high community disadvantage would increase the likelihood of membership in the early-onset, increasing subgroup relative to the low use class among youth with a higher internalizing symptoms PRS.
In addition, we anticipated that a higher internalizing symptoms PRS would increase the likelihood of an individual belonging to the early-onset, decreasing class compared to the low use class. This hypothesis is based on the premise that early adolescent engagement in alcohol use may result in a number of negative consequences that individuals with a higher internalizing symptoms PRS may be more sensitive to. In line with this hypothesis, we expected that when exposed to low parental monitoring or high community disadvantage, individuals with a higher internalizing symptoms PRS would be more likely to be a member of the early-onset decreasing class compared to the low use subgroup.
Method
Participants
The analytic sample for this study consists of 436 African Americans who were originally recruited as first grade students in the fall of 1993 as part of a randomized controlled, universal preventive intervention trial in nine mid-Atlantic urban elementary schools (Ialongo et al., Reference Anderson, Sabatelli and Kosutic1999). These 436 participants provided a successfully assayed DNA sample and completed at least one assessment of alcohol use over the course of annual assessments when participants were 14–26 years of age (see supplementary materials for more information on the alcohol use assessments). Demographic data for this study, including participant sex and free- or reduced-price lunch status (described in more detail below), were drawn from the baseline assessment conducted in the fall of first grade or when the participants were approximately 6 years old. Parental monitoring and community disadvantage data were assessed when participants were in sixth grade (i.e., ≈12 years old). The alcohol use data were derived from annual assessments that began at age 14 and continued to age 26. DNA data was collected from participants when they were approximately 21 years old. Thus, we drew on 15 waves of assessment data. A table regarding the number of alcohol use assessments that participants completed can be found in the supplementary materials.
With respect to the demographic characteristics of the analytic sample (i.e., 436 participants) (Table 1), 50.9% were male, 72.3% received free/reduced priced lunch, and 66.5% were assigned to an intervention condition. The analytic sample generally reflects the characteristics of the larger sample of 585 African American participants recruited in first grade with respect to participant sex (whole sample, 52.8% vs. analytic sample, 50.9%), free/reduced priced lunch status (whole sample, 70.6% vs. analytic sample, 72.5%), and percentage assigned to an intervention condition (whole sample, 67.9% vs. analytic sample, 66.5%).
Sample characteristics
aGED=General Education Degree.
Measures
Frequency of alcohol use
Frequency of alcohol use in the past year was assessed using questions adapted from the Monitoring the Future (MTF) survey (Bachman, Johnston, Lloyd, & O'Malley, Reference Bachman, Johnston and O'Malley1998) when participants were between the ages of 14 and 26. Participants reported on their frequency of alcohol use in the past year on a 0–7 Likert scale: 0 (no use), 1 (once), 2 (twice), 3 (3-4 times), 4 (5-9 times), 5 (10–19 times), 6 (20–39 times), and 7 (40 or more occasions). The MTF survey has been used in a number of studies to model alcohol use over time (e.g., Pokhrel, Unger, Wagner, Ritt-Olson, & Sussman, Reference Pokhrel, Unger, Wagner, Ritt-Olson and Sussman2008; Wagenaar, O'Malley, & LaFond, Reference Wagenaar, O'Malley and LaFond2001).
Parental monitoring
When participants were 12 years old, the Structured Interview of Parent Management Skills and Practices Youth-Version (SIPMSP) was used to assess parental monitoring (Capaldi and Patterson, Reference Capaldi and Patterson1989). Sample items are “How often would your parents or a sitter know if you came home late or on weekends?” and “How often before you go out do you tell your parents when you will be back?” Seven items were rated on a scale from 1 (all of the time) to 5 (never). Items were reverse scored. An average parental monitoring score was calculated with higher scores reflecting more monitoring. Capaldi and Patterson (Reference Capaldi and Patterson1989) report adequate internal consistency and test-retest reliability for the monitoring subscale.
Community disadvantage
Neighborhood disadvantage was assessed by geocoding census data when participants were approximately 12 years old. Community disadvantage as assessed via census data has been extensively used in the literature, with higher levels of disadvantage significantly associated with increased marijuana use offers and heavy marijuana use (Reboussin et al., Reference Reboussin, Green, Milam, Furr-Holden, Johnson and Ialongo2015). In addition, higher levels of community disadvantage have been associated with a greater concentration of alcohol outlets (e.g., Furr Holden et al., Reference Furr-Holden, Nesoff, Nelson, Milam, Smart, Lacey and Leaf2018; Milam et al., Reference Milam, Furr-Holden, Harrell, Ialongo and Leaf2014), which have been linked to higher levels of violent crime, decreased life expectancy, and increased substance use (Furr-Holden et al., Reference Furr-Holden, Nesoff, Nelson, Milam, Smart, Lacey and Leaf2018; Jennings et al., Reference Jennings, Milam, Greiner, Furr-Holden, Curriero and Thornton2013; Reboussin et al., Reference Reboussin, Ialongo, Green, Furr-Holden, Johnson and Milam2018).
Using ArcMap, a spatial join (appends data from two map layers using geographic location) was conducted to determine the census tract for each participant. A community disadvantage score was calculated using census-tract level items from the 1990 and 2000 Decennial census (U.S. Census, 2009). The items used to create the index include the percentages of (a) adults 25 years or older with a college degree, (b) owner-occupied housing, (c) households with incomes below the federal poverty threshold, and (d) female-headed households with children. We used Ross & Mirowsky's (Reference Ross and Mirowsky2001) formula to generate the following index: {[(c / 10 + d / 10) – (a / 10 + b / 10)] / 4}. The score has a possible range of −5 to +5, where -5 reflects low disadvantage and +5 reflects high disadvantage.
DNA and genotyping
At approximately age 21, blood or buccal samples were obtained from participants and DNA was extracted. Genotyping was carried out using the Affymetrix 6.0 microarray, which provides coverage of approximately one million SNPs across the genome (Affymetrix Inc.). Standard quality control steps were implemented to ensure that accurate genotypes were included in subsequent analyses. Subjects with >5% missing genotype data were removed. Single nucleotide polymorphisms were also removed from further analysis when they had a minor allele frequency <.01, missingness >0.05, or departures from Hardy–Weinberg equilibrium at p < .0001 (Anderson et al., Reference Anderson, Pettersson, Clarke, Cardon, Morris and Zondervan2010). These steps were performed using PLINK 2.0 (Chang et al., Reference Chang, Chow, Tellier, Vattikuti, Purcell and Lee2015). Genotypes were imputed to the 1000 Genomes Phase 3 reference panel (1000 Genomes Project Consortium et al., Reference Abecasis, Altshuler, Auton, Brooks, Durbin and McVean2010) using IMPUTE2 (Howie, Donnelly, and Marchini, Reference Howie, Donnelly and Marchini2009) with prephasing performed in SHAPEIT (Delaneau, Zagury, and Marchini, Reference Delaneau, Zagury and Marchini2013). Resulting variants imputed with an INFO (quality) score <0.8 were removed. Uncertainty adjusted dosage data, instead of called alleles, were used to generate the PRSs.
When exploring genetic associations, it is important to identify and control for population stratification or genetic differences between subpopulations so that any significant associations observed are not confounded by ancestry (Hellwege et al., Reference Hellwege, Keaton, Giri, Gao, Edwards and Edwards2018). We used principal components analysis in PLINK 2.0 to create the population stratification control variables (Chang et al., Reference Chang, Chow, Tellier, Vattikuti, Purcell and Lee2015). This process uses an orthogonal transformation to reduce the multidimensional genome-wide SNP data into a smaller number of principal components. All the available measured SNPs (roughly 900,000) were used to generate these components. Although these were not a priori identified ancestry information markers, it has been shown that “randomly” selected SNPs perform equally as well (Pritchard & Rosenberg, Reference Pritchard and Rosenberg1999). We included the first ten principal components in our analyses to sufficiently account for population stratification in the sample.
Discovery sample for internalizing symptoms
The discovery sample results used to generate the internalizing symptoms PRS were provided by a genome-wide meta-analysis of internalizing symptoms among preschool-aged children that was conducted by the Early Genetics and Lifecourse Epidemiology Consortium (EAGLE) (Benke et al., Reference Benke, Nivard, Velders, Walters, Pappa, Scheet and Middledorp2014). To date, this is the only GWAS to our knowledge that has been conducted on internalizing symptoms among children or adolescents. Though the authors of the GWAS examined internalizing symptoms in early childhood, a number of studies have shown that early childhood internalizing symptoms often persist into adolescence (Beyer, Postert, Müller, & Furniss Reference Beyer, Postert, Müller and Furniss2012; Mesman & Koot, Reference Mesman and Koot2001; Pihlakoski et al. Reference Pihlakoski, Sourander, Aromaa, Rautava, Helenius and Sillanpää2006).
The genome-wide meta-analysis of internalizing symptoms was performed using three separate cohorts drawn from the Generation Rotterdam (Generation R) study, Netherlands Twin Registry, and Western Australian Pregnancy (Raine) Cohort study. Internalizing symptoms were measured using 34 items found in common from different versions of the Child Behavior Checklist (Achenbach, Reference Achenbach1991) collected across the three samples. This study evaluated 4,596 total children across >2.4 million imputed SNPs. No genome-wide statistically significant SNPs were identified. The discovery sample SNP names and association results are available for download here: http://www.tweelingenregister.org/EAGLE.
Discovery sample for antisocial behavior
The antisocial behavior PRS was created based on a GWAS for antisocial behavior conducted by Tielbeek et al. (Reference Tielbeek, Johansson, Polderman, Rautiainen, Jansen, Taylor and Posthuma2017). This analysis included participants from five cohorts including the Avon Longitudinal Study of Parents and Children, Collaborative Studies on the Genetics of Alcoholism, Generation R, the Twins Early Development Study, and a population-based study conducted by the Queensland Institute of Medical Research (Tielbeek et al., Reference Tielbeek, Johansson, Polderman, Rautiainen, Jansen, Taylor and Posthuma2017). Participants (N = 16,400) were ethnically diverse children and adults. Antisocial behaviors (e.g., conduct disorder symptoms, rule-breaking) were assessed differently (e.g., questionnaires, interviews) depending on the cohort (Tielbeek et al., Reference Tielbeek, Johansson, Polderman, Rautiainen, Jansen, Taylor and Posthuma2017). The discovery sample SNP names and association results can be downloaded here: http://broadabc.ctglab.nl/documents/p12/BroadABC_METALoutput_Combined.tbl.
PRS generation
Our GWAS panel contained 741,174 (26.30%) directly genotyped SNPs from the internalizing symptoms discovery sample list and 736,488 (7.73%) SNPs from the antisocial behavior discovery sample list. After imputation, 2,554,305 (90.50%) SNPs from the internalizing symptoms and 6,193,047 (65.00%) from the antisocial behavior discovery datasets were available for the current sample. Palindromic (A/T or C/G) SNPs were excluded, as methods for properly orienting strand from discovery to test datasets require precise knowledge of the true minor allele frequency and haplotype structure of the test sample. To account for linkage disequilibrium (LD), two rounds of LD-based results clumping were run in PLINK 2.0 (Chang et al., Reference Chang, Chow, Tellier, Vattikuti, Purcell and Lee2015) against the HapMap Phase III Release 2 Build 36 reference panel (The International HapMap 3 Consortium, 2010), resulting in 219,312 and 168,617 available SNPs for the internalizing symptoms PRS and antisocial behavior PRS, respectively. Based on our chosen p-value threshold of 0.001, the number of SNPs included in the internalizing symptoms PRS was 549 and the number of SNPs included in the antisocial behavior PRS was 393. There was no overlap in the SNPs included in the internalizing symptoms PRS and antisocial behavior PRS. Raw scores were generated in the imputed dosage dataset in PLINK 2.0 (Chang et al., Reference Chang, Chow, Tellier, Vattikuti, Purcell and Lee2015). We used mean imputation for missing genotypes, and alleles were weighted by the effect sizes from the discovery GWAS. The raw PRSs were regressed on the ten ancestry principal components. The z-scored residuals from these regressions were the continuous ancestry-corrected scores used in the primary analyses. A list of the SNPs included in both the internalizing symptoms and antisocial behavior PRSs can be found in the supplementary materials.
Statistical analyses
Bivariate correlations and descriptive statistics were conducted in SPSS Version 25 (IBM, 2016). A longitudinal latent class analysis (LCA) was used to categorize individuals into homogeneous subgroups of past year alcohol use. The frequency of alcohol use variables are considered count data and accordingly, a Poisson model was used to estimate our latent models. To handle missing data, we used Full Information Maximum Likelihood (FIML) estimation in Mplus (Muthén & Muthén, Reference Muthén and Muthén2017), although the count variables were skewed. In simulation studies that have evaluated the effect of non-normality on FIML estimation in latent models, FIML may result in negatively biased standard errors and inflated model rejection rates (Enders, Reference Enders2001). However, FIML has been shown to be superior to other approaches (e.g., multiple imputation) for handling non-normal missing data, and it has generally yielded more accurate estimates relative to other methods (Enders, Reference Enders2001). All predictor variables were z-scored (M = 0, SD = 1).
Consistent with the procedures outlined by Masyn et al. (Reference Masyn2017), we employed a three-step procedure that included class enumeration and involved the examination of the predictors of class membership. The first two steps centered on class enumeration and involved determining the number of latent classes without the covariates included in the model in the first step. LCA begins with a one-class (unconditional) model, and the number of classes is increased until there is no additional improvement in model fit (Nylund, Asparouhov, & Muthén, Reference Nylund, Asparouhov and Muthén2007). Several model fit indices were examined, including the Akaike Information Criterion (AIC) (Akaike, Reference Akaike1987), the Bayesian Information Criterion (BIC) (Schwartz, Reference Schwartz1978), sample-size adjusted BIC (ABIC) (Sclove, Reference Sclove1987), and the Bootstrap Likelihood Ratio Test (BLRT) (Nylund et al., Reference Nylund, Asparouhov and Muthén2007). The size of the smallest class was also considered as small class sizes (less than 5% of the sample) may indicate overfitting of the data and create potential problems with replication and generalizability of the LCA solution. Further, we considered entropy, a measure of classification uncertainty, with higher values (e.g., >.7) indicating better participant classification. Models were also examined to determine whether classes were distinct and conceptually meaningful.
The second step involved assigning individuals into classes based on their most likely class membership. The estimated values of participants’ probabilities of membership in each class were calculated, which were based on the maximum likelihood parameter estimates from the model and participants’ responses on the latent class indicators (Masyn, Reference Masyn2017). These probabilities were saved using the “cprobabilities” command in Mplus and were used in the third step.
The third step involved entering the covariates and predictors while accounting for the classification error rate of participants. The first set of the multinomial logistic regressions involved regressing latent class membership on each of the covariates (participant sex, intervention status, free/reduced priced lunch status), the PRSs (internalizing symptoms, antisocial behavior), and the contextual variables (parental monitoring, community disadvantage). In these regressions, both PRSs were included (i.e., internalizing symptoms PRS and the antisocial behavior PRS). Separate regressions were run for parental monitoring and community disadvantage. We controlled for free/reduced priced lunch status; this variable is often considered a proxy for family income (Hobbs & Vignoles, Reference Hobbs and Vignoles2010; Huang & Barnidge, Reference Huang and Barnidge2016), and it has been robustly associated with psychological impairments and substance use problems among youth (Goodman, Reference Goodman1999; Hanson & Chen, Reference Hanson and Chen2007). Although individuals and families with low incomes may be more likely to live in disadvantaged neighborhoods, there still may be variation in family incomes in these neighborhoods. As such, we controlled for free/reduced priced lunch status to ensure that our results were driven by the neighborhood context and not family income.
The second set of multinomial regressions involved regressing latent class membership on the interaction terms involving the PRS and relevant contextual variables while controlling for the covariates. Significant interactions and slopes were plotted using an automated spreadsheet (Dawson, Reference Dawson2014). Post hoc probing involved creating new moderator variables at the mean and ±1 SD from the z-scored values of the moderator (i.e., parental monitoring or community disadvantage) (Aiken & West, Reference Aiken and West1991; Cohen & Cohen, Reference Cohen and Cohen1983). Interaction terms were created that included these variables. For significant interactions, the post hoc regressions involved entry of the covariates considered in the original regression, the internalizing symptoms and antisocial behavior PRSs, the moderator (at the mean and ±1 SD from the mean of the contextual variable), and the Contextual Variable × PRS interaction (Aiken & West, Reference Aiken and West1991; Cohen & Cohen, Reference Cohen and Cohen1983). In graphing the interactions, we included the unstandardized betas that were at the mean and ±1 SD from the mean of the moderator (Holmbeck, Reference Holmbeck2002).
In total, we planned to carry out 4 sets of planned comparisons and 14 exploratory analyses. For the antisocial behavior PRS, the planned, hypothesis-driven analyses included comparisons between the early-onset, increasing subgroup and the (a) low use group; (b) early-onset, decreasing class; (c) late-onset, moderate use class; and (d) late-onset, heavy use class. The planned, hypothesis-driven analyses for the internalizing symptoms PRS involved comparisons between the low use subgroup and (a) early-onset, increasing class, and (b) early-onset, decreasing class. To test for significance of the exploratory analyses, we employed a Bonferroni correction (.05/14 tests) which yielded a p-value of .003. A list of the planned and exploratory analyses can be found in Table 2.
Expected and actual planned and exploratory analyses
Note: A Bonferroni correction was applied to the exploratory analyses.
aPRS = polygenic risk score.
Results
Bivariate correlations and descriptive statistics of study variables are presented in Table 3. There were no correlations among the predictors that were significant at p < .05. Results from the primary analyses are presented below.
Bivariate correlations, means, standard deviations, ranges, and n's of study variables
Note: AU = frequency of alcohol use; INT PRS = internalizing symptoms polygenic risk score; ASB PRS = antisocial behavior polygenic risk score.
aBivariate correlations and descriptive statistics are presented for the residualized polygenic risk scores.
*p < .05 **p < .01.
Latent class analysis results
Following the three-step procedure described above, a longitudinal LCA was conducted using raw scores for frequency of past year alcohol use from ages 14–26. The AIC, BIC, and ABIC decreased with the addition of each class (Table 4). Entropy remained relatively high across models. Although the ABIC and BLRT decreased in the five-class model, this model included a very small class, potentially indicating model overfitting and a class that was not distinct from another subgroup. Thus, we selected the four-class model.
Fit indices for longitudinal latent class analysis models with 1-5 classes
Note: AIC = Akaike Information Criterion, BIC = Bayesian Information Criterion, ABIC = Sample-size Adjusted BIC, BLRT = Bootstrap Likelihood Ratio Test. BLRT and entropy are not calculated for the 1-class model.
Class 1 was named “early-onset, increasing” (n = 114, 26.2%), as individuals in this class showed greater mean levels of past year alcohol use in early adolescence and higher mean levels of alcohol use in young adulthood relative to the other classes. Class 2 was named “late-onset, moderate use” (n = 93, 21.3%), as individuals in this class exhibited low mean levels of past year alcohol use until age 20 and showed increases in mean levels of past year alcohol use from ages 21 to 26. Class 3 was named “low steady” (n = 136, 31.2%), as participants in this class displayed low mean levels of past year alcohol use from ages 14–26. Class 4 was named “early-onset, decreasing” (n = 93, 21.3%), as individuals in this profile reported moderate to high mean levels of past year alcohol use during early and middle adolescence followed by a decline in mean levels of alcohol use in late adolescence and young adulthood. Given that our Bonferroni correction was based on the exploratory analyses, if a five-class solution was found, we recalculated our Bonferroni correction based on a four-class solution, which yielded a p-value of .007 (.05/7 tests), and then we applied this p-value to the exploratory analyses that we conducted.
Planned comparisons
Main effect results
There was a significant main effect of the internalizing symptoms PRS on class membership. Consistent with our hypotheses, participants with a higher internalizing symptoms PRS were more likely to be in the early-onset, decreasing class compared to the low steady class, OR = 1.34, p = .013, 95% CI [1.06, 1.69]. There was a trend for significance such that youth with a higher internalizing symptoms PRS were less likely to be in the early-onset, increasing class relative to the low steady class, OR = 0.80, p = .057, 95% CI [0.63, 1.01]. None of the antisocial behavior PRS planned comparisons were significant (Tables 5 and 6).
Multinomial logistic regression results involving the PRS and community disadvantage
Note: Step 1 included Step 2 variables.
*p < .10 **p < .05.
Multinomial logistic regression results involving PRS and parental monitoring
Note: Step 1 included Step 2 variables.
*p < .10 **p < .05 ***p < .01.
Interaction results
Consistent with our hypotheses, there was a significant interaction between the internalizing symptoms PRS and community disadvantage when comparing likelihood of class membership between the early-onset, decreasing class and the low steady class, OR = 1.53, p = .013, 95% CI [1.09, 2.14], (Table 5). In the context of higher community disadvantage, individuals with a higher internalizing symptoms PRS were more likely to be in the early-onset, decreasing class (B = 0.65, p = .002) compared to the low steady class (Figure 2). The slopes for average and low community disadvantage were not significant. No interactions between the internalizing symptoms PRS and community disadvantage (or parental monitoring) were found when considering likelihood of membership in the early-onset increasing class vs. the low steady group. However, there was a trend for significance for the internalizing symptoms PRS and parental monitoring interaction when comparing likelihood of class membership in the low steady class versus the early-onset, increasing class (p = .081) (Table 6).
Frequency of past year alcohol use from ages 14 to 26 in the four-class model.
Relative to the low steady class, the log odds of membership in the early-onset, decreasing class based on participant internalizing symptom PRS levels and community disadvantage.
As shown in Table 5, a significant interaction was observed between the antisocial behavior PRS and community disadvantage when comparing likelihood of class membership in the early-onset, increasing class to the early-onset, decreasing class, OR = 0.68, p = .041, 95% CI [0.46, 0.98], in line with our hypotheses (Figure 3a). Individuals with a higher antisocial behavior PRS were more likely to be in the early-onset, increasing class relative to the early-onset, decreasing class when exposed to higher (B = 0.78, p = .003) and average (B = 0.39, p = .038) levels of community disadvantage. The slope for low community disadvantage was not significant.
The log odds of membership in the early-onset, increasing class relative to the (a) early-onset, decreasing class, and (b) late-onset, moderate use class based on participant ASB PRS levels and community disadvantage.
Consistent with our hypotheses, there was also a significant interaction between the antisocial behavior PRS and community disadvantage when comparing likelihood of class membership between the early-onset, increasing class and the late-onset, moderate use class, OR = 0.72, p = .045, 95% CI [0.52, 0.99], (Figure 3b). In the context of higher community disadvantage, individuals with a higher antisocial behavior PRS were more likely to be in the early-onset, increasing class relative to the late-onset, moderate use class (B = 0.48, p = .050). The slopes for average and low community disadvantage were not significant. Paralleling the Internalizing Symptoms PRS × Parental Monitoring interaction results, none of the planned comparisons involving the Antisocial Behavior PRS × Parental Monitoring interactions were significant (Table 6).
Discussion
The present study examined whether subgroups of alcohol use in adolescence and early adulthood could be identified in an urban African American sample. Most of the available work that has identified subgroups of alcohol use during these developmental periods has been conducted among predominantly European youth of varying socioeconomic status (Chassin et al., Reference Chassin, Pitts and Prost2002; Flory et al., Reference Flory, Lynam, Milich, Leukefeld and Clayton2004). Thus, it is uncertain whether the patterns of alcohol use identified in these samples are relevant to low-income African Americans residing in an inner-city. Moreover, it is unclear whether genetic propensity for internalizing symptoms and antisocial behavior, as measured via polygenic risk scores, is associated with alcohol use during adolescence and young adulthood. Consistent with compelling evidence that genes and environmental features interact with each other (Shanahan & Hofer, Reference Shanahan and Hofer2005), we also examined whether alcohol use subgroup membership could be explained by the interplay between contextual factors (i.e., parental monitoring, community disadvantage) and genetic propensity for internalizing symptoms and antisocial behaviors.
Four subgroups of alcohol use were identified, including an early-onset, increasing class characterized by higher mean levels of past year alcohol use in early adolescence and young adulthood. This subgroup mirrors previous work that identified an early-onset, heavy alcohol use group that used alcohol frequently in early adolescence and continued on a trajectory towards increased use in adulthood (Chassin et al., Reference Chassin, Pitts and Prost2002; Flory et al., Reference Flory, Lynam, Milich, Leukefeld and Clayton2004; Su, Supple, Leerkes, & Kuo, Reference Su, Supple, Leerkes and Kuo2018; Zucker et al., Reference Zucker, Hicks, Heitzeg and Cicchetti2016). A late-onset, moderate use class was also identified that displayed low mean levels of past year alcohol use in adolescence followed by moderate mean levels of alcohol use from ages 20 through 26. The individuals identified in this subgroup also parallel the characteristics of subgroups identified in previous studies (Chassin et al., Reference Chassin, Pitts and Prost2002; Nelson et al., Reference Nelson, Van Ryzin and Dishion2015). A low steady group was also observed that reported very low mean levels of past year alcohol use over time, mirroring previous findings regarding low alcohol consumption across developmental periods (Chassin et al., Reference Chassin, Pitts and Prost2002; Nelson et al., Reference Nelson, Van Ryzin and Dishion2015). We also observed an early-onset, decreasing class that exhibited higher mean levels of past year alcohol consumption from early to late adolescence, followed by decreases in mean levels of alcohol use into young adulthood. In contrast to our expectations, a late-onset, heavy alcohol use class was not found in our sample. As noted previously, studies that have examined longitudinal patterns of alcohol use have been conducted among predominantly European American samples, and not all of the classes identified in these samples may be relevant to inner-city African American youth. It is also possible that we may have observed this class in a larger African American sample.
Higher polygenic load for internalizing symptoms was associated with membership in the early-onset, decreasing class relative to the low steady class, consistent with our hypotheses. The contextual (e.g., transition to high school) and neurobiological changes (e.g., puberty) associated with early and middle adolescence may be particularly challenging for individuals with a higher internalizing symptoms PRS given that these individuals may exhibit greater reactivity towards stressors. These youths may subsequently seek out alcohol to reduce tension and negative affective states (Kushner, Sher, Wood, & Wood, Reference Kushner, Sher, Wood and Wood1994). Early use of alcohol during early and middle adolescence may result in negative repercussions (e.g., arrests, parental discipline), to which individuals with a higher internalizing symptoms PRS may be more sensitive to (Fite et al., Reference Fite, Colder and O'Connor2006). As a result, these individuals may attempt to avoid these or other negative consequences by decreasing their consumption of alcohol in later adolescence and early adulthood.
Individuals with a higher internalizing symptoms PRS were also more likely to be in the early-onset, decreasing class compared to the low steady subgroup when exposed to higher levels of community disadvantage. During early and middle adolescence, greater time spent in more socioeconomically deprived communities characterized by prevalent drug use may enable alcohol use among individuals with a higher polygenic load for internalizing symptoms given that they may be more likely to experience negative emotional states (Chassin et al., Reference Chassin, Sher, Hussong and Curran2013; Hussong et al., Reference Hussong, Jones, Stein, Baucom and Boeding2011; Wallace et al., Reference Wallace, Neilands and Phillips2017; Wallace & Muroff, Reference Wallace and Muroff2002). Indeed, lower community supervision of youth behavior, coupled with increased alcohol outlets in these neighborhoods, may facilitate early alcohol use among individuals with a higher internalizing symptoms PRS (Milam et al., Reference Milam, Furr-Holden, Harrell, Ialongo and Leaf2014). As noted above, frequent use of alcohol during early and mid adolescence may result in negative repercussions, and youth with a higher internalizing symptoms PRS may subsequently decrease their consumption of alcohol.
We also found that in the context of higher community disadvantage, individuals with a higher antisocial behavior PRS were more likely to be in the early-onset, increasing class relative to the early-onset, decreasing and late-onset, moderate use classes consistent with our hypotheses. Communities higher in disadvantage characterized by poverty and a greater concentration of alcohol outlets may enable alcohol use among adolescents higher in polygenic load for antisocial behavior, given greater sensation seeking behaviors and impulsivity that these youths may display (Mann et al., Reference Mann, Engelhardt, Briley, Grotzinger, Patterson, Tackett and Harden2017; Maneiro et al., Reference Maneiro, Gómez-Fraguela, Cutrín and Romero2017). Adolescents with greater polygenic propensity for antisocial behaviors may also be lower in fearfulness, affiliate with substance using peers, and may be more inclined to use alcohol in disadvantaged communities where drugs are more available (Mann et al., Reference Mann, Engelhardt, Briley, Grotzinger, Patterson, Tackett and Harden2017; Rosenberg & Anthony, Reference Rosenberg and Anthony2001). Exposure to community disadvantage earlier in development may thus enable alcohol use over time among individuals with greater polygenic load for antisocial behavior.
Although interactive effects were found with regard to the antisocial behavior PRS and community disadvantage, the antisocial behavior PRS was not associated with class membership. This finding is consistent with limited work noted above that showed that genetic liability for conduct disorder was not associated with substance use disorders (i.e., marijuana abuse and dependence) when neighborhood disadvantage was not included in the analytic model (Rabinowitz et al., Reference Rabinowitz, Musci, Milam, Benke, Sisto, Ialongo and Maher2018). It is possible that the predisposition for antisocial behavior only serves as a risk factor for elevated alcohol use in the context of environmental stressors, consistent with the diathesis-stress model (Zuckerman, Reference Zuckerman1999). Future research should examine other contextual factors (e.g., affiliation with deviant peers) that may influence the association between the antisocial behavior PRS and alcohol use patterns over time.
Parental monitoring did not moderate the relationship between polygenic load for internalizing symptoms and antisocial behavior and the alcohol use subgroups identified. However, there was a trend for significance involving the interaction between parental monitoring and the antisocial behavior PRS in predicting likelihood of class membership in the low steady vs. early-onset increasing class. Parental monitoring was assessed via participant self-report on a questionnaire; therefore, the pattern of findings may have been influenced by shared method variance or social desirability. Future work should consider assessing parental monitoring using other informants (e.g., siblings) and other methods (e.g., interviews).
There are limitations of the present study to acknowledge. We used results from a GWAS conducted in samples of largely European ancestry to inform the generation of PRSs in an African American sample. As of 2016, about 16% of individuals included in the GWASs are ethnic minority populations, and most of these individuals are of Asian ancestry (Popejoy & Fullerton, Reference Popejoy and Fullerton2016). Although differences in linkage disequilibrium and ancestral markers have been observed across individuals of different ancestries, recent work has demonstrated using simulated data that PRSs maintain transferability across ancestry groups (Martin et al., Reference Martin, Gignoux, Walters, Wojcik, Neale, Gravel, Daly and Kenny2017). Work across ethnic groups, such as that presented here, is needed to overcome the limitation of the field regarding the lack of representation of African Americans in gene identification efforts and molecular genetics studies. Future research should attempt to replicate these findings in other African American samples and ethnic minority populations. Among the limitations of the study was the use of a single Likert item to capture frequency of alcohol use.
An additional limitation was that the discovery sample GWAS from which the internalizing symptoms PRS was derived included parent reports on their children's internalizing symptoms. While a number of studies have shown that childhood internalizing symptoms often persist into adulthood in both European and African American samples (e.g., Brody et al., Reference Brody, Kim, Murry and Brown2005; Mesman & Koot, Reference Mesman and Koot2001; Musci et al., Reference Musci, Masyn, Benke, Maher, Uhl and Ialongo2015; Pihlakoski et al. Reference Pihlakoski, Sourander, Aromaa, Rautava, Helenius and Sillanpää2006), future GWASs should assess these symptoms during other developmental periods (e.g., adolescence, adulthood) using ethnically diverse samples and attempt to replicate findings from the current study. In addition, we only considered alcohol use in our latent models, as opposed to including other substances (e.g., marijuana, tobacco) that may co-occur with alcohol use over time (Armstrong & Costello, Reference Armstrong and Costello2002; Banks, Rowe, Mpofu, & Zapolski, Reference Banks, Rowe, Mpofu and Zapolski2017). Although this decision enabled us to examine whether internalizing symptoms and antisocial behavior polygenic load were related specifically to patterns of alcohol consumption, polygenic load for these problems may also serve as a liability for other substance use behaviors, something future work should examine. Future research should also consider replicating findings from the current study using other indices of neighborhood disadvantage, such as self-reported neighborhood disadvantage and crime. An additional avenue for future research is to examine phenotypic internalizing symptoms and antisocial behaviors, genetic risk for these problems, and contextual risk and protective factors that may play a role in alcohol use across developmental periods. Such an approach may highlight processes involved in equifinality and multifinality and illuminate why individuals with higher polygenic risk for internalizing and externalizing symptoms may not display these phenotypes.
In terms of future directions, the field could benefit from identifying the pathways through which (a) higher internalizing symptom polygenic load resulted in an attenuation in alcohol use in late adolescence in more disadvantaged communities, and (b) higher antisocial behavior polygenic load resulted in an increase in alcohol use during adolescence and young adulthood in more disadvantaged neighborhoods. An additional avenue for future work is to examine whether spirituality (e.g., prayer, meditation) and specific coping practices (e.g., collective coping or support from extended family) influence alcohol use patterns among African Americans (Krentzman Farkas, & Townsend, Reference Krentzman, Farkas and Townsend2010; Utsey, Bolden, Lanier, & Williams Reference Utsey, Bolden, Lanier and Williams2007). These factors have been associated with sobriety among individuals receiving alcohol abuse treatment and improved psychological and physical health among African Americans (Krentzman et al., Reference Krentzman, Farkas and Townsend2010; Utsey et al., Reference Utsey, Bolden, Lanier and Williams2007).
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0954579419000701.
'/%3e%3cpath%20d='M17%207.971L15.711%206.677%2010.912%2011.495%2010.912%200%209.088%200%209.088%2011.495%204.289%206.677%203%207.971%2010%2015z'%20transform='translate(8%208)'/%3e%3c/g%3e%3c/g%3e%3c/svg%3e)
