Alcohol consumption and alcohol use disorder – exposing an increasingly shared genetic architecture

In their work, “Genetic underpinnings of the transition from alcohol consumption to alcohol use disorder: shared and unique genetic architectures in a cross-ancestry sample,” Kember et al. report an impressive undertaking to identify genetic contributors to both alcohol consumption and alcohol use disorder in the largest study to date, both in terms of sample size and in terms of inclusion of non-European population groups (1). The authors also capitalize on the unique construction of the Million Veterans Program (MVP), with longitudinal data from alcohol consumption screenings and alcohol use disorder diagnoses in health records, to better refine phenotypes and analyses, something few other studies can accomplish. In these analyses, performed within and across ancestries, they report a notable 24 independent variants (19 loci) associated with alcohol consumption, quantified by Alcohol Use Disorder Identification Test-Consumption (AUDIT-C) scores (2) and 26 independent variants (21 loci) associated with alcohol use disorder. The authors also perform gene-based associations, mediation analyses, calculate genetic correlations, construct polygenic risk scores, and perform a phenome-wide association study in two external datasets: the Vanderbilt Biobank (BioVU) and UK Biobank (UKB). Through these analyses, they conclude that differences in the associated loci, differences in genetic and phenotypic correlations, and non-mediating genetic variation support a conclusion that alcohol consumption and AUD have distinct underlying genetic architectures. We submit that, considered with other recent publications on the genetics of both alcohol use and disorder, the results presented by Kember et al., highlight the necessity of minimizing trait heterogeneity by reducing the misclassification of individuals who now abstain from alcohol but who also have with a lifetime history of alcohol use disorder. When trait heterogeneity is minimized, the overall genetic underpinnings of alcohol consumption and use disorder are, in composition and pattern, quite similar.

Given the heavy focus on genetic correlations as an indication of shared vs. distinct genetic architectures for alcohol consumption and AUD, we think it prudent to give a brief overview of genetic correlations reported for these traits. Genetic correlations reported from twin studies suggested moderate to high genetic correlations (r_g range = 0.45–0.99) among several alcohol consumption traits, and high genetic correlations between alcohol consumption traits and problematic alcohol use (3, 4). Using cross-trait linkage disequilibrium score regression, reported genetic correlations between alcohol consumption traits and problematic alcohol use or alcohol use disorder have ranged widely (Table 1). In one of the earliest studies of individuals of European ancestry, the reported genetic correlation between AUDIT and alcohol use disorder was negligible (r_g = 0.08) (5). Notice that several of the comparisons in Table 1 have used the same cohorts or have used the same trait (e.g., AUDIT-C) in different cohorts, illustrating that seemingly innocuous differences in trait derivation and sample makeup can lead to large differences in estimated genetic correlations. Importantly, the study by Kember et al. compares alcohol consumption and alcohol use disorder measured in the same individuals, which eliminates biases in the estimated genetic correlations based on sample selection and comorbid illness. When the authors focus analyses on those who report current drinking and remove those who abstain from alcohol, 15% of whom have a lifetime history of alcohol use disorder, the genetic correlation between alcohol consumption and alcohol use disorder increases (r_g = 0.86–1) and the genetic correlations for individuals of African ancestry are very high (r_g = 0.98–1). These findings highlight that the genetic architecture of alcohol consumption and alcohol use disorder is primarily shared.

Table 1.

Reported genetic correlations between alcohol consumption traits (in rows) and problematic alcohol use (in columns). Unless otherwise noted, all reported correlations are from genetic studies using individuals of European ancestry.

Problem Use →Consumption ↓	ICD-based AUD (stringent, amongAUDIT-C > 0) (1)	ICD-based AUD (stringent) in MVP (1, 6)	ICD-based AUD (less stringent) in MVP (1)	DSM-IV AUD in PGC (7)	AUD in MVP, PGC (8)	PAU in MVP, UKB, and PGC (8, 9)	DSM-IV AUD symptom count (10)	AUDIT-P in UKB (11)
AUDIT-C (among AUDIT-C > 0) in MVP (1)	EUR: 0.86 AFR: 1	EUR: 0.87 AFR: 1	EUR: 0.90 AFR: 0.98
AUDIT-C in MVP (1)	EUR: 0.71 AFR: 0.97	EUR: 0.76 AFR: 0.96	EUR: 0.78 AFR: 0.93
AUDIT-C in MVP (6)		EUR: 0.52 AFR: 0.93
Maximum drinks in one day in a typical month in MVP (12)					0.76	EUR: 0.79 AFR: 0.67		0.73
AUDIT-C in UKB (11)				0.33			0.41	0.70
AUDIT in UKB (11)				0.39			0.48	0.81
Drinking quantity among those who drink at least once or twice a week in UKB (13)				0.75
Drinking frequency in UKB (13)				NS
Derived average intake per week in UKB (14)				0.37				0.76
AUDIT in 23andMe (5)				0.08				0.64
Average number of drinks per week in GSCAN (15)						0.77
Average number of drinks per week in GSCAN (16)						0.76
Derived grams per day in AlcGen and CHARGE+ (17)				0.70				0.76

Abbreviations:

AlcGen: Alcohol Genome-Wide Association Consortium

AUDIT: Alcohol Use Disorder Identification Test consisting of all 10 questions

AUDIT-C: AUDIT-Consumption, consisting of the first three questions of the AUDIT

AUDIT-P: AUDIT-Problems, consisting of the last seven questions of the AUDIT

CHARGE+: Cohorts for Heart and Aging Research in Genomic Epidemiology Plus

DSM: Diagnostic and Statistical Manual of Mental Disorders

GSCAN: GWAS & Sequencing Consortium of Alcohol and Nicotine Use

ICD-based AUD (less stringent): least one ICD-9/10 code for AUD (1).

ICD-based AUD (stringent): at least one inpatient or two outpatient codes for AUD (1, 6).

PAU: Problematic alcohol use

PGC: Psychiatric Genomics Consortium

UKB: UK Biobank

Several previous studies have reported seemingly divergent patterns of genetic correlation between consumption and use disorder when compared to non-alcohol related traits. The AUDIT-C was reported to have puzzling positive genetic associations with variables related to socioeconomic status (e.g., educational attainment) and some health outcomes (e.g., HDL cholesterol), and negative associations with other health outcomes (e.g., obesity, triglycerides) and some forms of psychopathology (e.g., major depression diagnosis, attention-deficit/hyperactivity disorder) (5, 6, 11, 14). Opposite patterns of correlation were observed for AUDIT-P or alcohol use disorder (6, 11). Explanations for these divergent patterns have included true biological differences, confounding by selection bias, genetic heterogeneity, and measurement error. Kember and colleagues again help clarify most of these discrepancies between these genetic correlations between alcohol consumption and use disorder with other traits. When examining only those who report consuming alcohol and eliminating those who abstain, fewer differences in the genetic correlations between consumption and alcohol use disorder were still evident, and most were different in magnitude, not direction of effect, as reported in Supplemental Table 16 and visualized in Figure 2c of Kember et al. (1).

We are at the crux of a revolution in which increasingly data-driven approaches are used to increase phenotypic precision and mitigate biases in genetic analyses of an inherently heterogeneous behavior and disease (18–20). With recent evidence that Black and Hispanic veterans from the MVP dataset were more likely to have an AUD diagnosis than White veterans, despite similar AUDIT-C scores, this revolution is needed not only to improve our power in genetic association studies, but also mitigate racial and ethnic bias in diagnosis (21). We applaud Kember et al., for engaging with this revolution by examining different ancestral groups and exploring variations on the trait definitions under study: namely by modifying the AUD-case threshold and removing those who abstain from alcohol (AUDIT-C > 0), which were previously shown to impact genetic associations (1, 20). In removing those who abstain, the authors only “lost” one to two genome-wide significant loci, despite losing almost a quarter of their original sample, suggesting greater power from higher phenotypic precision and reduced misclassification. Kember et al., also observed an increase in the SNP heritability of both AUDIT-C and AUD, further demonstrating the positive impact of reducing bias and improving phenotypic precision on understanding both the genetic architecture of consumption and of AUD (18, 20).

Like all behavior and neuropsychiatric studies, we are at the mercy of phenotype definitions that boil a lifetime of intersecting factors: genetic, social, environmental, and institutional, into a single scale (in the case of AUDIT scores) or code (in the case of diagnoses). This heterogeneity of individual experience means that we simply cannot make the mistake of assuming that individuals sharing a diagnosis are homogeneous in trait presentation or genetics. This may seem a pessimistic view, but to the contrary, the underlying variation is ripe for interesting studies in which geneticists, epidemiologists, and medical professionals can work together. By using refined phenotype definitions and careful studies, we can find meaningful associations to translate into clinical knowledge and therapeutics. For example, are there methods to mitigate bias in diagnoses of AUD, especially in individuals attesting moderate to high AUDIT-C scores? Or, are there social and genetic factors distinctive of individuals with similarly moderate consumption but disparate AUD diagnoses? As we work to answer these questions, we posit that, in the effort to understand the genetic etiology of alcohol consumption and alcohol use disorder, there is more is in common than there is different. We look forward to other creative work from this group and others to move the field further along the path to understanding alcohol consumption and treating problematic alcohol use and alcohol use disorder.