Alcohol consumption

Using the AUDADIS alcohol consumption measures, we created a variable indicating maximum number of drinks in a 24 hour period (Maxdrinks;(Saccone et al., 2000; Shmulewitz et al., 2012). AUDADIS consumption items have good psychometric properties and good to excellent inter-rater reliability (ICC=0.59-0.88 for quantity consumed) (Hasin et al., 1997a; Grant et al., 2003). Weekly at-risk drinking (risk drinking) was defined as ≥5 drinks in a day for males or ≥4 in a day for females at least once a week during period of heaviest drinking, and assessed with questions querying the frequency of drinking specific quantities, as described previously (Keyes et al., 2009). Reliability of AUDADIS lifetime drinking frequency was very good to excellent (ICCs 0.72-0.76), and fair-to-good for ≥5 drinks (ICCs 0.47-0.69) (Hasin et al., 1997b; Grant et al., 2003).

Diplotype analysis

As variants in both ADH1C-rs698 and ADH1B-rs1229984 showed similar relationships with alcohol phenotypes, we constructed haplotypes with these two SNPs (SAS 9.2, HAPLOTYPE procedure, www.sas.com). Three haplotypes with frequencies >1% where found (rs698-rs1229984): A-A, A-G, and G-G. After removing 11 individuals with the rare haplotype (G-A), haplotypes were assigned with 100% certainty to all remaining individuals (N=1,118). Because haplotypes were assigned instead of estimated, deviations from HWE due to association (such as HW disequilibrium among AUD cases for rs1229984) should not lead to inaccurate haplotype assignment. To ensure this, we assigned haplotypes separately in cases and in controls, and found no differences. Haplotypes were combined into six diplotypes: A-A, A-A; A-A, A-G; A-A, G-G; A-G, A-G; A-G, G-G; G-G, G-G. These diplotypes were categorized to match the allelic risk model described above according to the level of risk they were hypothesized to confer. The diplotype hypothesized to confer high risk included the absence of the protective allele for both rs698 and rs1229984: G-G, G-G. Diplotypes hypothesized to confer intermediate risk included the absence of the protective allele for rs1229984 only: A-G, A-G (rs698=AA, rs1229984=GG) and A-G, G-G (rs698=AG, rs1229984=GG). Of those without the protective allele for rs698, 95.7% also lacked the protective allele for rs1229984; thus, the diplotype categories can be viewed as dividing up the rs1229984 group without the protective allele into those with the protective allele for rs698 (intermediate risk group) and those without the protective allele for rs698 (high risk group). (Note that there is no category for absence of protective allele for rs698 with presence of protective allele for rs1229984 due to the haplotype structure. Accordingly, we did not formally test interaction between ADH1B-rs1229984 and ADH1C-rs698.) Diplotypes hypothesized to confer low risk had protective alleles for both rs698 and rs1229984. To investigate the association of the diplotype variable with the alcohol phenotypes, adjusted logistic or Poisson regression analyses were carried out, and ORs/RRs were calculated for all comparisons, to determine if both ADH1B and ADH1C played a role in increasing risk.

Association of ADH1C-rs698/ADH1B-rs1229984 diplotypes and alcohol-related traits

Mean values were significantly different from each other for AD, any AUD, and AD and AUD severity, with increased ORs or RRs when comparing diplotype groups (Table 5), suggesting that ADH1B and ADH1C both affect these phenotypes. For example, the intermediate risk group (high risk for ADH1B but not ADH1C) had 2.06 times increased odds of AD compared to the low risk group (low risk for both ADH1B and ADH1C), while the high risk group (high risk for both ADH1B and ADH1C) had 4.31 times increased odds compared to the low risk group and 2.09 times increased odds compared to the intermediate group (Table 5). For risk drinking and maxdrinks, the intermediate-risk and high-risk diplotype groups were not significantly different, suggesting that in this sample, only ADH1B plays a role.

ADH1B and ADH1C

We found that while ADH1B-rs1229984 and ADH1C-rs698 were correlated (r²=0.32; D′=0.44), they did not represent a single signal; each variant contributed to risk independently. Due to the unique haplotype structure in this population, we were able to directly examine the joint effects of ADH1B-rs1229984 and ADH1C-rs698 on alcohol phenotypes by identifying ADH1B-rs1229984 and ADH1C-rs698 diplotypes with certainty, categorizing them into three risk groups, and testing the specific combinations of risk variants for alcohol phenotypes. For both the binary and count measures of AD and AUD, significant differences were observed across each of the three diplotype risk groups, indicating that both SNPs play a role in these outcomes. These results agree with the single SNP analyses, which indicated that both ADH1B-rs1229984 and ADH1C-rs698 were associated with AUDs, while ADH1C-rs698 was not significantly associated with consumption. While the diplotypes explained roughly the same amount of the variance in these traits (e.g. ~19% of the variance in AD) as either ADH1B-rs1229984 (~18% of the variance in AD) or ADH1C-rs698 (~17% of the variance in AD) individually, considering these SNPs together provides a model that more accurately describes the polygenic architecture of these traits. Thus, the higher effect sizes (ORs or RRs) in the diplotype analysis (for the high vs. low risk comparison) than in the single marker analyses was most likely due to better characterization of the gene effect due to partitioning the ADH1B-rs1229984 high risk group into two groups based on the presence or absence of the ADH1C-rs698 high risk group.

While prior studies have examined gene-gene interactions between ADH genes (Wu et al., 2012), only three studies explicitly examined the joint effect of these specific ADH SNPs for alcohol outcomes (Neumark et al., 2004; Tolstrup et al., 2008; Toth et al., 2011). Results from these studies were mixed; the Neumark study found no evidence of interaction between these two SNPs, while the more recent studies indicated that the joint effects of these ADH risk alleles (using haplotypes/diplotypes) increased risk for excessive consumption and AUDs (Tolstrup et al., 2008; Toth et al., 2011). Results from our study can potentially clarify these discrepancies in that we found the joint effect (of both ADH1B and ADH1C) is related to AUDs, but not significantly to measures of consumption in this large population of Israeli Jews. While a genetic variant could plausibly be consistently associated with all related phenotypes (consumption and AUDs), that is not necessarily the case if different measures tap into different aspects of a related phenotype which may have unique etiological risk factors (Meyers et al., 2012). Further studies are needed to examine the association of AUDs and consumption with ADH1C.

Although many gene mapping studies are carried out in Ashkenazi Jewish (AJ) samples (i.e., all four grandparents born in Europe; (Guha et al., 2012), due to the unique genetic structure of that population (Guha et al., 2012), we decided to also include non-Ashkenazi Jews (NAJ; mainly of Middle Eastern origin), to determine if the same genetic effects were found in both potential subpopulations (AJ/NAJ). The AJ/NAJ distinction also connotes cultural differences, possibly encompassing differences in drinking behavior. Although we asked respondents to self-identify as “Ashkenaz” or “non-Ashkenaz”, close to 20% of the sample chose “both” or “other”, reflecting the blurred distinctions between these categories, both in terms of cultural practices and genetic origins (i.e. individuals with grandparents from both groups), in modern-day Israel. Therefore the traditional AJ/NAJ distinction was not useful in this sample. Rather, we used AIMs, which produced a continuum of genetic ancestry measuring the Northern/Southern European cline, with higher proportion of “Northern” ancestry indicating more Ashkenazi origin, and higher proportion “Southern” ancestry indicating more non-Ashkenazi origin (Listman et al., 2010). We analyzed the whole sample (instead of AJ/NAJ separately), including ancestry proportions in regression analyses to ensure that even if there were cultural differences in drinking related to ancestral origins, the association results were not due to genetic admixture. Similar results with and without the ancestry proportions suggest that the ADH gene effects were similar across subpopulations, and support utility of gene association studies in mixed Jewish populations.

Study limitations are noted. First, we do not claim to have identified all the causative genetic variants; we may be detecting significant signals where there are contributions from LD with an additional untested, “true” variant (or variants), not only in ADH1B but across the ADH gene cluster. However, enzymatic studies show that the ADH1B*2 and ADH1C*2 variants are functional, affecting enzyme kinetics (Bosron et al., 1983; Bosron & Li, 1986). Next, the unique haplotype structure in this population was useful in enabling us to identify diplotypes with certainty and to categorize them into three risk groups. However, to more fully understand the joint or related interaction effects, samples with different LD patterns and greater prevalence of the rare haplotypes/diplotypes (e.g., lack of ADH1C protective allele with the presence of the ADH1B protective allele) should be studied. In addition, this study examined two genes in relation to six related alcohol outcomes in an effort to systematically assess the association. Prior evidence for genetic association, along with the high degree of correlation between the alcohol phenotypes (r² range from 0.35-0.96) and SNPs (r² from 0.01-0.75), mitigates the concern that results are merely a consequence of multiple testing. Nevertheless, when we applied the Benjamini-Hochberg procedure to control the false discovery rate at 0.05 (Benjamini & Hochberg, 1995), all results remained significant except that ADH1B-rs1229982 was not associated with AUD severity. Last, in Israel, consuming several drinks on one occasion is common in religious practice (e.g., Purim and the Passover Seder), and might occur for cultural/environmental reasons even among individuals who would otherwise drink much less. As a sensitivity analysis, we conducted post-hoc analyses of another measure, “usual drinks,” which indicates how many drinks a respondent usually had in a single day. Association results were similar in that ADH1B was significantly related to usual drinks, and ADH1C, when analyzed jointly with ADH1B, was not related to usual drinks.

In conclusion, this study adds support for the role of ADH1B and ADH1C in alcohol phenotypes in a large Israeli household population. Further, the results suggest that considering these SNPs together provides a model that more accurately depicts the polygenic architecture of AUD traits. Studies of the association of alcohol phenotypes and either ADH1B or ADH1C alone are employing an oversimplified model, which may mask relevant information regarding the genetic influence of these ADH variants on behavioral alcohol phenotypes.