Alcohol use is increasingly common in the United States. Nearly 86% of individuals 18 and older report using alcohol at some point in their lives, with approximately 55% reporting use in the previous month (Substance Abuse and Mental Health Services Administration, 2020). By the end of high school, 59% of adolescents have reported consuming at least one alcoholic drink (Johnston et al., Reference Johnston, Meich, O’Malley, Bachman, Schulenberg and Patrick2020). Alcohol use and problems are also heritable, with genetic variation accounting for approximately 50% of the variance in alcohol dependence (Verhulst et al., Reference Verhulst, Neale and Kendler2015). Similar heritability estimates (h ² = 30 to 51%) have been found for other alcohol use behaviors like alcoholic drinks per week and maximum number of drinks (Grant et al., Reference Grant, Agrawal, Bucholz, Madden, Pergadia, Nelson, Lynskey, Todd, Todorov, Hansell, Whitfield, Martin and Heath2009). The influence of genetic factors on alcohol use behaviors, estimated in biometric studies of twin resemblance, appears to strengthen over time, increasing in magnitude from early adolescence to young adulthood before stabilizing (Kendler et al., Reference Kendler, Schmitt, Aggen and Prescott2008). This mirrors a decline in the influence of shared environmental factors (i.e., those environmental factors that are shared by individuals reared in the same home), from explaining approximately 40% of the variance in alcohol use at age 14 to near zero by the mid-20’s.

Alcohol consumption also often co-occurs with use of other substances, like tobacco (Agrawal & Lynskey, Reference Agrawal and Lynskey2009; Conway et al., Reference Conway, Green, Kasza, Silveira, Borek, Kimmel, Sargent, Stanton, Lambert, Hilmi, Reissig, Jackson, Tanski, Maklan, Hyland and Compton2017), marijuana (Vrieze et al., Reference Vrieze, Hicks, Iacono and McGue2012), and other illegal drugs (Kessler et al., Reference Kessler, Chiu, Demler and Walters2005; Krueger, Reference Krueger2002). Phenotypic correlations between symptom counts of alcohol, marijuana, and tobacco use disorders in a longitudinal study from ages 14 to 29 ranged from r = 0.24 to r = 0.61 (Vrieze et al., Reference Vrieze, Hicks, Iacono and McGue2012). Correlations between measures of consumption are also similarly high across ages 14 to 40 (Zellers et al., Reference Zellers, Iacono, McGue and Vrieze2022). These correlations declined in magnitude across developmental age, suggesting that young individuals tend to use substances broadly while older individuals show relatively more preferences for one substance over another.

There are several proposed theories to account for substance use comorbidity. For example, co-occurrence among multiple substances may be explained by an underlying externalizing factor that accounts for the associations between these substances, often termed the common liability model. In this case, a latent factor represents a general vulnerability to risk of substance use and development of substance use disorders (Iacono et al., Reference Iacono, Malone and McGue2008; Krueger, Reference Krueger2002). Indeed, several studies have provided evidence that a latent factor explains the majority of covariation between externalizing behaviors like substance dependence, conduct disorder, and ADHD (Hicks et al., Reference Hicks, Krueger, Iacono, McGue and Patrick2004, Reference Hicks, Foster, Iacono and McGue2013; Krueger, Reference Krueger2002). Moreover, this general vulnerability to externalizing behaviors was found to be highly heritable (h ² = 0.80) (Hicks et al., Reference Hicks, Krueger, Iacono, McGue and Patrick2004). Substance use co-occurrence is also consistent with the idea that use of one substance causally influences risk for use of other substances (i.e., the gateway hypothesis) (Kandel, Reference Kandel1975; Kandel & Kandel, Reference Kandel and Kandel2015). In this case, it may be that after an individual uses one drug, they seek out other drugs from which to get a stronger high. The gateway hypothesis explains both the co-occurrence of use of different substances and developmental sequencing of drug use, where users of “hard” drugs (e.g., cocaine) almost always have first used other, legal drugs (e.g., alcohol or tobacco) (Degenhardt et al., Reference Degenhardt, Dierker, Chiu, Medina-Mora, Neumark, Sampson, Alonso, Angermeyer, Anthony, Bruffaerts, de Girolamo, de Graaf, Gureje, Karam, Kostyuchenko, Lee, Lépine, Levinson, Nakamura and Kessler2010). In general, evidence for the gateway hypothesis has been limited and is often subject to methodological criticisms (Peele & Brodsky, Reference Peele and Brodsky1997; Vanyukov et al., Reference Vanyukov, Tarter, Kirillova, Kirisci, Reynolds, Kreek, Conway, Maher, Iacono, Bierut, Neale, Clark and Ridenour2012). The more recent literature largely supports the common liability model, with consistent findings across multiple studies and age ranges (Hicks et al., Reference Hicks, Blonigen, Kramer, Krueger, Patrick, Iacono and McGue2007; Hicks BM et al., Reference Hicks, Krueger, Iacono, McGue and Patrick2004; Krueger, Reference Krueger2002; van Leeuwen et al., Reference van Leeuwen, Verhulst, Reijneveld, Vollebergh, Ormel and Huizink2011; Vrieze et al., Reference Vrieze, McGue, Miller, Hicks and Iacono2013).

The gateway hypothesis is based on the premise that use of one drug causally influences use of others. One method for evaluating evidence for causal effects in observational studies is through Mendelian randomization. Mendelian randomization (MR) is a method frequently employed in epidemiology that uses a genetic variant(s) that is robustly associated with an exposure to evaluate the association with outcome (Lawlor et al., Reference Lawlor, Harbord, Sterne, Timpson and Davey Smith2008). In other words, MR evaluates the relationship between genetic variants that are of known function and strongly associated with an exposure (e.g., alcohol use), and an outcome of interest (e.g., illegal drug use). MR takes advantage of the random assortment of genes during meiosis which is analogous to the randomization process in randomized control trials. In this way, MR offers a relatively strong test of causal effects in observational data. Using genetic data as a proxy for an environmental modifiable exposure like alcohol use minimizes the likelihood that the observed exposure–outcome association is due to unmeasured confounding and removes the possibility of reverse causality (Smith & Ebrahim, Reference Smith, Ebrahim, Weinstein, Vaupel and Wachter2008; Smith & Hemani, Reference Smith and Hemani2014). Because genotypes are assigned randomly before birth, they are unlikely to be associated with factors that may confound the relationship between alcohol and other drug use (e.g., socioeconomic status) and because genotype is unchangeable, reverse causality is ruled out. If the gateway hypothesis is correct, we would expect the genetic variant to be associated with both alcohol use exposure and the use of other substances. This would provide support for a causal effect of alcohol use on risk for other substance use.

There are two genetic variants that have shown robust associations with alcohol use, and on which we focus in the present paper: ALDH2 (rs671) and ADH1B (rs1229984) (Edenberg, Reference Edenberg2007). Both variants influence alcohol consumption by affecting the way alcohol is metabolized in the body. After alcohol is consumed, alcohol dehydrogenase converts it into acetaldehyde. Aldehyde dehydrogenase then oxidizes acetaldehyde into acetate, which is ultimately cleared from the body. High concentrations of acetaldehyde in the body lead to significant physical discomfort. ADH1B acts to accelerate the conversion of alcohol to acetaldehyde, while ALDH2 slows the conversion into acetate. Both genes cause a build-up of acetaldehyde in the body leading to severe discomfort and have been robustly associated with reduced alcohol consumption and dependence. The ALDH2*2 allele, which is common only in populations of East/North-East Asian ancestry and is well tagged by a single variant, has the largest effect on alcohol consumption. A single copy of this variant can affect the breakdown of alcohol enough to cause severe symptoms. ADH1B*2, which is also common in these populations, has a weaker, but still robust, association with alcohol consumption. The structure of the ADH1B gene may be more complex than ALDH2, being defined by more than one genetic variant with weaker effects.

Much of the existing literature on the associations between ALDH2 and ADH1B status and alcohol use are based on cross-sectional designs, including several genome-wide association studies (GWAS) of alcohol use and dependence. In a sample of Japanese individuals, the effect size of the ALDH2 protective allele on drinks per week was estimated at −0.43 standard deviation (SD) units (Matoba et al., Reference Matoba, Akiyama, Ishigaki, Kanai, Takahashi, Momozawa, Ikegawa, Ikeda, Iwata, Hirata, Matsuda, Murakami, Kubo, Kamatani and Okada2020). Other GWAS have estimated the effect of variants within ADH1B on drinks per week between −0.07 and −0.25 SD units (Brazel et al., Reference Brazel, Jiang, Hughey, Turcot, Zhan, Gong, Batini, Weissenkampen, Liu, Barnes, Bertelsen, Chou, Erzurumluoglu, Faul, Haessler, Hammerschlag, Hsu and Vrieze2019; Liu et al., Reference Liu, Jiang, Wedow, Li, Brazel, Chen, Datta, Davila-Velderrain, McGuire, Tian, Zhan, Choquet, Docherty, Faul, Foerster, Fritsche, Gabrielsen, Gordon, Haessler and Vrieze2019; Matoba et al., Reference Matoba, Akiyama, Ishigaki, Kanai, Takahashi, Momozawa, Ikegawa, Ikeda, Iwata, Hirata, Matsuda, Murakami, Kubo, Kamatani and Okada2020). The few studies of longitudinal associations between these genetic variants and alcohol use patterns have found that the protective effect of ALDH2*2 increases over developmental age (Irons et al., Reference Irons, Iacono, Oetting and McGue2012; Kim et al., Reference Kim, Lee, Shin, Son and Ju2010). Irons et al. (Irons et al., Reference Irons, Iacono, Oetting and McGue2012), using an earlier data freeze from the present study, found that the association between ALDH2*2 and alcohol consumption was negligible around age 15 but increased to moderate in effect size by age 22. To our knowledge, there is one study of the longitudinal effect of ADH1B status on alcohol use, finding an interaction between gene status and age from early to mid-adolescence such that those with ADH1B*2 increased their alcohol use less than those without (Cleveland et al., Reference Cleveland, Schlomer, Vandenbergh, Wolf, Feinberg, Greenberg, Spoth and Redmond2018). Increasing magnitudes of association between ALDH2 and ADH1B variants and alcohol use over time would be consistent with the observed pattern of strengthening influence of genetic factors from adolescence to adulthood (Kendler et al., Reference Kendler, Schmitt, Aggen and Prescott2008). This may be explained by a reduction in the influence of the shared environment as offspring leave the rearing, parental home and begin to select their own environments, which itself may be partially influenced by genetic factors. It is not yet clear whether the strength of association between ALDH2 and ADH1B variants and alcohol use continues to increase beyond early adulthood.

The effect of ALDH2 and ADH1B status on alcohol consumption has also been found to vary in magnitude by environmental or cultural factors. For example, among those with the ALDH2 and ADH1B polymorphisms there is substantial variability in mean alcohol intake levels within China (Millwood et al., Reference Millwood, Walters, Mei, Guo, Yang, Bian, Bennett, Chen, Dong, Hu, Zhou, Yu, Jia, Parish, Clarke, Smith, Collins, Holmes, Li and Chen2019). Despite having two copies of the ALDH2 protective allele (i.e., the same genotype; ALDH2*2), the mean alcohol intake differed by a factor of 24 across geographic areas of China, illustrating cultural effects on consumption patterns. The estimated magnitude of association between ALDH2 status and alcoholic drinks per week in East Asian populations has also been shown to vary by country. Genetic effect sizes of ALDH2 are 2.5 times larger when estimated in samples of East Asian individuals living in Japan (Matoba et al., Reference Matoba, Akiyama, Ishigaki, Kanai, Takahashi, Momozawa, Ikegawa, Ikeda, Iwata, Hirata, Matsuda, Murakami, Kubo, Kamatani and Okada2020) compared to samples of East Asian individuals living in the United States (effect size of −0.43 in Japan versus −0.17 in the United States) (Jorgenson et al., Reference Jorgenson, Thai, Hoffmann, Sakoda, Kvale, Banda, Schaefer, Risch, Mertens, Weisner and Choquet2017). Because the ALDH2 variant is only polymorphic in East Asian populations, the vast majority of individuals in the United States, who are not of East Asian ancestry, do not have this genotype. Thus, mean alcohol consumption is higher in the United States compared to many East Asian countries, where the ALDH2 deficiency is common, though this difference is decreasing. It may be that the protective effect of ALDH2*2 (and potentially ADH1B*2) is weaker within cultures or environments like the United States with higher rates of alcohol consumption. Indeed, Higuchi et al. (Higuchi et al., Reference Higuchi, Matsushita, Imazeki, Kinoshita, Takagi and Kono1994) found that the percentage of Japanese individuals suffering from alcoholism grew from 2.5% to 13% from 1979 to 1992, mirroring cultural shifts in alcohol consumption in Japan at that time. Environmental influences within families and amongst peers may also moderate the protective effects of ALDH2*2 and ADH1B*2. Previous research has found that parent, sibling, and peer alcohol consumption patterns predict continued use in those with the ALDH2 deficiency (Irons et al., Reference Irons, McGue, Iacono and Oetting2007, Reference Irons, Iacono, Oetting and McGue2012; Luk et al., Reference Luk, Liang and Wall2017; O’Shea et al., Reference O’Shea, Thomas, Webb, Dick, Kendler and Chartier2017).

The current study makes use of a unique longitudinal adoption sample to extend the literature on the associations between ALDH2/ADH1B and alcohol consumption from adolescence to adulthood (ages 14 to 40) and to evaluate the causal relationship between alcohol exposure and other substance use outcomes. The sample consists of individuals who were adopted as infants from South Korea, primarily by White American adoptive parents living in Minnesota. Thus, they are reared in homes with genetically unrelated adoptive parents and siblings, allowing for exploration of important environmental influences on alcohol use behaviors, as well as adult outcomes hypothesized to be caused by alcohol exposure. The adoption and MR design provide quasi-experimental controls for genetic confounding in typical observational and family studies of closely related individuals. Our specific hypotheses are

1. 1. Individuals with ALDH2/ADH1B protective alleles will drink less than those without and that this protective effect will increase in magnitude with age.

2. 2. Sibling and parent alcohol use will moderate the protective effect of ALDH2*2/ADH1B*2 and that this moderating effect, if present, will diminish with age. We made no explicit hypotheses about the environmental moderation of peer alcohol use as this was included based on reviewer feedback.

3. 3. Alcohol use will not be associated with tobacco, marijuana, or other drug use in Mendelian randomization tests with ALDH2/ADH1B variants as instruments, inconsistent with the gateway hypothesis.

Methods

Sample

The current sample was taken from the Minnesota Center for Twin and Family Research (Wilson et al., Reference Wilson, Haroian, Iacono, Krueger, Lee, Luciana, Malone, McGue, Roisman and Vrieze2019) Sibling Interaction and Behavior Study (SIBS) which includes 409 adoptive families. Adoptive families consisted of two genetically unrelated siblings and their rearing parents. Families were ascertained through infant placements at large, private adoption agencies in Minnesota. Eligible adoptive families included those in which the adopted adolescent was between the ages of 11 and 21 at the intake assessment, was placed in the adoptive home before 2 years of age, and had an adolescent sibling to which they were unrelated. While the sibling could not be genetically related to the adopted child, they could be genetically related to one or both of the parents, or could have also been adopted and placed before the age of 2 years. A more complete description of the SIBS sample is given elsewhere (McGue et al., Reference McGue, Keyes, Sharma, Elkins, Legrand, Johnson and Iacono2007). SIBS participants have been followed longitudinally with up to four assessments. At the intake assessment, both parents and offspring were assessed. The first follow-up, approximately 3.5 years after intake, included an assessment and collection of DNA samples. The second and third follow-up assessments occurred approximately 7 and 17 years after the intake assessment, respectively.

The sample used in the current analysis was restricted to families with at least one offspring adopted from East/North-East Asia who had also provided genetic data. This results in N = 412 East Asian adopted participants, all of whom were born in South Korea, from N = 259 families. Adopted offspring had a mean age of 15.1 years (SD = 1.9 years) at intake, 18.3 years (SD = 2.1) at the first follow-up, 22.3 years (SD = 1.8) at the second follow-up, and 32.2 years (SD = 2.6) at the third follow-up. Offspring data used in the present study were taken from the first, second, and third follow-up assessments. Table 1 shows demographic information for adopted offspring. At the intake assessment N = 239 mothers and N = 241 fathers reported their own alcohol use. All adult participants gave informed consent, and minor participants gave assent with their parents providing consent.

Table 1. Sample characteristics of East/North-East Asian adoptive offspring

Note. Both ALDH2 and ADH1B groups include those with at least one copy of the protective A allele. Comparisons of means (or proportions) were made within genotype (e.g., comparison of means between those with at least one ALDH2 protective allele and those without) and were based on a t statistic for continuous variables or χ² statistic for binary variables. Parents reported lifetime DSM-IV alcohol dependence symptoms at the intake assessment. Sibling alcohol quantity and frequency was defined as the mean across all available assessments. Peer use was based on a single question of how many friends use alcohol on an ordinal scale from 1 = none to 4 = all. M = mean; SD = standard deviation.

Phenotypic measures

All substance use measures were collected using either the Substance Abuse Module (SAM) (Robins et al., Reference Robins, Babor and Cottler1987) of the Composite International Diagnostic Interview (CIDI) (Robins LN et al., Reference Robins, Wing, Wittchen, Helzer, Babor, Burke, Farmer, Jablenski, Pickens and Regier1988) or the Computerized Substance Use Questionnaire (CSU), depending on the age of the participant.

Alcohol-related outcomes

Quantitative alcohol-related measures, covering the time since the prior assessment, included frequency of alcohol use (scored from 6 = every day to 1 = less than once per month), usual quantity of alcohol use (measured as number of drinks), and maximum number of drinks in a 24-hour period. A single lifetime alcohol-related measure included the age of first alcoholic drink without parental permission. The physical effects of ALDH2 and ADH1B deficiency are largely dependent on the amount of alcohol consumed, so the protective effect of these genes may be greater for measures of consumption than for frequency of use. For this reason, we included multiple measures of alcohol-related behaviors that index different aspects of consumption instead of creating a composite measure of alcohol use. Sibling frequency and quantity of use were measured in the same way as adopted offspring. At the intake assessment, parents reported lifetime number of DSM-IV alcohol dependence symptoms. We used the mid-parent mean symptom count as the measure of parental alcohol problems. Peer alcohol use was measured at the first and second follow-up assessments and was based on a single question of how many friends use alcohol (scored from 1 = none of my friends to 4 = all of my friends).

Other substance use outcomes

Lifetime binary measure of substance use included ever used tobacco regularly, ever used marijuana before the age 22 assessment, and ever used illegal drugs before the age 22 assessment. Quantitative outcomes, covering the time since the prior assessment, included number of cigarettes per day amongst regular smokers, number of marijuana uses amongst ever users, and a count of the number of illegal drug types used (e.g., cocaine, opiates, amphetamines, etc.). Tobacco, marijuana, and drug use data were taken only from the age 18 and age 22 assessments (follow-ups 1 and 2).

Genotyping

Genotype data (obtained from offspring at the first follow-up assessment) was used to identify relevant variants within the ALDH2 (rs671; chromosome 12) and ADH1B (rs1229984; chromosome 4) genes in adoptees. Participants were genotyped on 527,829 single nucleotide polymorphism (SNP) markers using Illumina’s Human660W-Quad array (Miller et al., Reference Miller, Basu, Cunningham, Eskin, Malone, Oetting, Schork, Sul, Iacono and McGue2012). Genotypes were then imputed to 1000 Genomes using the Michigan imputation server (Das et al., Reference Das, Forer, Schönherr, Sidore, Locke, Kwong, Vrieze, Chew, Levy, McGue, Schlessinger, Stambolian, Loh, Iacono, Swaroop, Scott, Cucca, Kronenberg, Boehnke and Fuchsberger2016). Both ALDH2 and ADH1B were imputed with imputation qualities (expected squared correlations) of r ² = 0.48 and r ² = 0.82, respectively, in N = 365 individuals. Despite the modest expected quality of the rs671 variant, the ALDH2 gene was also directly assayed in N = 354 individuals of East Asian ancestry. Here, we use the directly assayed genotype whenever available. The correlation between the directly assayed ALDH2 genotype and the imputed ALDH2 genotype was .89, suggesting the expected imputation quality calculated from the imputation procedure was downwardly biased. Further details on how ALDH2 was assayed can be found in Irons et al. (Irons et al., Reference Irons, McGue, Iacono and Oetting2007).

The ALDH2 and ADH1B gene variants have three possible genotypes: A/A, A/G, and G/G. In both, the A allele is associated with the strongest protection against alcohol use problems. Because just a single copy of each allele has been shown to be protective against alcohol use (Luczak et al., Reference Luczak, Glatt and Wall2006), genotypes for each variant were dichotomized and coded as 1 for A/A homozygotes and A/G heterozygotes (e.g., ALDH2*1/*2 and ALDH2*2/*2) and 0 for G/G homozygotes (e.g. ALDH2*1/*1).

Statistical analysis

To address the first hypothesis, regression models were used to examine the associations between ALDH2 and ADH1B polymorphisms and alcohol-related outcomes with a focus on how the effect sizes change from ages 14 to 40. Given the longitudinal nature of the data, mixed effects models were fit that included ALDH2 genotype, ADH1B genotype, sex, and age at assessment (linear and quadratic terms) as fixed effects. These models also included an interaction between genotype and age to test whether the magnitude of the protective effects of ALDH2*2 and ADH1B*2 changed significantly over development. All models included a random intercept for each individual, to account for the correlated family structure, and random slope for age.

To address the second hypothesis, we fit similar mixed effects models adding predictors of parent, sibling, and peer alcohol use and their interaction with ALDH2/ADH1B status. In this way we evaluated whether parent, sibling, or peer alcohol use is associated with an adoptee’s alcohol use and whether the magnitude of the effects differs by genotype. Finally, to test whether ALDH2/ADH1B status predicted tobacco, marijuana, and illegal drug use (hypothesis 3), we used mixed effects models (with sex, age at assessment, and genotype as fixed effects and random effects of age and individual), with each of these substances as the outcome. Given the multiple hypothesis tests performed here with correlated outcome variables, we calculated the effective number of tests using the correlations between substance-related outcomes (Derringer, Reference Derringer2018; Nyholt, Reference Nyholt2004). This results in a corrected statistical significance threshold of p < 0.005.

Results

Descriptive statistics

Among 412 East/North-East Asian adopted participants, 301 (73.1%) were G/G homozygotes for ALDH2, 9 (2.2%) were A/A homozygotes, and 102 (24.8%) were heterozygotes. For ADH1B, 196 (53.7%) were G/G homozygotes, 26 (7.1%) were A/A homozygotes, and 143 (39.2%) were heterozygotes. For both variants, we combined those with at least one protective A allele (i.e., genotypes of A/A and A/G) into a single group which was then compared to those with G/G genotypes. This results in 111 individuals (26.9%) with at least one copy of the ALDH2 A allele (i.e., ALDH2*2/*2 or ALDH2*1/*2) and 169 (46%) with at least one ADH1B A allele (i.e., ADH1B*2/*2 or ADH1B *1/*2).

Table 1 reports descriptive statistics of the full sample including comparisons between those with and without ALDH2 and ADH1B protective alleles and indicates that those with the protective genotypes did not differ from those without in gender, age at follow-up, or family member alcohol use. Approximately 62% of the sample is female with no difference in proportions by genotype group. The large proportion of females reflects the gender imbalance of children adopted from East Asia at the time. The effects of sample attrition were evaluated by comparing participants and nonparticipants at a given assessment with their responses at the previous assessment. In general, we found no differences in alcohol or other substance-related outcomes between follow-up 2 participants and nonparticipants at their prior assessment (follow-up 1) or between follow-up 3 participants and nonparticipants at the prior follow-up 2 assessment. The single exception to this was for differences in usual alcohol quantity at both follow-up 1 and follow-up 2 assessments. Those who completed the follow-up 1 assessment, but not the second follow-up, had higher levels of usual alcohol quantity than those who participated at both assessments (standardized mean difference of 0.82, p = 0.04). While this is a large mean difference, we note that there were only N = 13 individuals who participated in the first, but not second, follow-up assessment. Those who completed the follow-up 2 assessment, but not the third follow-up, had higher levels of usual alcohol quantity than those who participated at both assessments (standardized mean difference of 0.28, p = 0.02).

ALDH2/ADH1B associations with alcohol use

Figure 1 shows the observed patterns of alcohol frequency, usual quantity, and maximum drinks from ages 14 to 40 for the full sample and split by ALDH2 status. In general, alcohol consumption and frequency increased across adolescence, peaking in the early 20’s, before stabilizing or declining slightly through age 40. We did not find an association with age at first alcohol drink and either ALDH2 or ADH1B status (p-values of 0.76 and 0.72, respectively). This is expected given that an individual may not know their ALDH2/ADH1B status prior to first initiating alcohol use. Longitudinal models showed significant associations between ALDH2*2, but not ADH1B*2, with alcohol-related outcomes (shown in Table 2). Interactions between ALDH2 genotype and age were all in the expected direction, though this effect was only significant for maximum drinks (p-value = 0.0047). These results indicate that while we find evidence for increasing effect size estimates across age the magnitude of this change is too small to detect at conventional levels of statistical significance in this sample.

Figure 1. Plot of alcohol-related outcomes across age for the full sample and split by ALDH2 status. Loess curves were added to graphically display the relationship between alcohol behaviors and age by genotype. Confidence intervals around the loess lines were removed for clarity. Points are colored by genotype where orange indicates those with G/G genotype (ALDH2*1/*1) and blue indicates those with the protective A/A or A/G genotype (ALDH2*1/*2 or ALDH2*2/*2).

Table 2. Associations between ALDH2/ADH1B status and alcohol-related outcomes

Note. All models included genotype, sex, age at assessment (linear and quadratic term), and a genotype × age interaction as fixed effects along with random intercept per individual and slope for age. For both ALDH2 and ADH1B, genotypes of A/A or A/G (i.e., at least one protective allele) are coded as 1. β = standardized coefficient; SE = standard error;

^a denotes variables that were log transformed prior to analysis;

* denotes p-values below the corrected threshold of 0.005.

Environmental influence of parent/sibling/peer alcohol use

Next, we evaluated the environmental influences of family and peer alcohol use on adoptee’s own drinking behaviors and whether this effect differed by genotype. Because we did not find evidence for associations between ADH1B and alcohol-related outcomes, we focus here only on ALDH2. For each alcohol outcome, we fit separate models with sibling frequency of alcohol use, sibling usual quantity of alcohol use, mid-parent number of alcohol dependence symptoms, and peer use as predictors, along with an interaction between each predictor and adoptee ALDH2 status. Sibling alcohol frequency and quantity were defined as the mean across all available follow-up assessments, while parental alcohol dependence symptoms were reported lifetime. Peer alcohol use was indexed by a single question at the adoptee’s age 18 and 22 assessments. Full results are reported in Table 3. While nearly all estimated main effects of parent and sibling use were in the expected positive direction, only one estimate (the association between sibling usual alcohol quantity and adoptee’s usual alcohol quantity) was significantly different from zero based on the corrected significance threshold of 0.005. Peer alcohol use was significantly associated with all measures of adoptee alcohol use, showing the largest effect sizes. None of the interactions between parent, sibling, or peer use and adoptee ALDH2 status were significantly different from zero. This suggests evidence for environmental moderation of alcohol use through peer, but not familial, alcohol use and that these environmental effects do not differ as a function of ALDH2 status.

Table 3. Associations between parent, sibling, and peer alcohol use and alcohol-related outcomes.

Note. All models included ALDH2 status, sex, age at assessment (linear and quadratic term), and an ALDH2 status × sibling, parent, or peer alcohol use interaction as fixed effects along with random intercept per individual and slope for age. β = standardized coefficient; SE = standard error;

^a denotes variables that were log transformed prior to analysis;

* denotes p-values below the corrected threshold of 0.005.

ALDH2 associations with other substance use

The associations between ALDH2 status and other substance use outcomes are shown in Table 4. We again focus only on ALDH2 given the lack of association between ADH1B and alcohol use measures in the current sample. There is minimal evidence for an association between ALDH2 status and any tobacco, marijuana, or illegal drug use measures. Associations with marijuana and illegal drug use were in the expected direction but effect sizes were small in magnitude, and all p-values were greater than 0.13. We note that sample sizes for the binary measures of substance use are substantially lower than for continuous measures because they are lifetime and not repeated measures.

Table 4. Associations between ALDH2 status and other substance-related outcomes.*

Note. All models included ALDH2 genotype, sex, age at assessment (linear and quadratic term; or year of birth for lifetime measures) as fixed effects along with random intercept per individual and slope for age. ALDH2 genotypes of A/A or A/G (i.e., at least one protective allele) are coded as 1. β = standardized coefficient; OR = odds ratios; SE = standard error;

^a denotes variables that were log transformed prior to analysis;

* denotes p-values below the corrected threshold of 0.005.