Alcohol use is ubiquitous across most societies, and many alcohol use traits, from regular drinking to alcohol use disorder (AUD), are common, or at least common enough to cause problems to individuals and to society that are widely recognized. People use alcohol for many reasons, and they experience its effects and consequences in different ways—there is a high degree of heterogeneity, both in terms of individual susceptibility, and in terms of environmental influences. AUD is treated with a range of methodologies, from participation in organizations like Alcoholics Anonymous to various types of psychotherapy to pharmacological treatments. The success of treatment is variable, and it is hard to predict who might benefit from any particular treatment. It would be very helpful to resolve some of the biological heterogeneity that underlies this set of traits. If we could understand interindividual differences in alcohol use biology, this might lead to improvements in prevention and treatment. Accordingly, there have been many efforts to identify subtypes of alcohol use traits, with phenotype-based typologies going back decades, such as “type 1 versus type 2” (
1) and “type A versus type B” (
2). While these were of interest, they have not found additional support in the genomics era. We would really like to know, What are the basic traits that, taken individually or in combination, make up the set of observed alcohol use traits? The data available in the pre-genomics era didn’t provide enough information to answer this very complex question.
Those of us who work in genetics have long held out the hope that with better understanding of genetics, we could develop improved typologies of alcohol use behaviors (AUBs), with better biological relevance, and a study from Savage et al. in this issue of the
Journal (
3) moves us closer to this goal. Indeed, studies in alcohol use genetics—principally, genome-wide association studies (GWASs)—have already resulted in major advances in our understanding of alcohol use phenotypes, such as separating quantity and frequency of use traits genetically from AUD (
4,
5). Before large genomics studies were possible, quantity/frequency phenotypes (like Alcohol Use Disorders Identification Test consumption [AUDIT-C] scores and drinks per week [DPW]) and AUD were not considered genetically dissimilar. But now multiple studies support that traits based on how much alcohol is consumed differ genetically—and therefore biologically—from traits defined based on physiological dependence or medical consequences of alcohol use (
6).
Alcohol use and AUD-related traits have genetic commonalities, including strong association with alcohol-metabolizing enzyme variants (
7,
8).
ADH1B, for example, is associated with both problematic alcohol use (PAU) and DPW at very high levels of significance (PAU, p=2.22×10
−283 [
9]; DPW, p=5.89×10
−284 [
10]), and this has been seen for most other alcohol-related traits. Pretty much every alcohol use–related trait studied to date has been associated with alcohol-metabolizing enzyme variants, with numerous replications.
There has been huge progress in understanding the specifics of the genetics of alcohol-related traits, with hundreds of risk loci mapped at high levels of significance. This has been made possible by the availability of large datasets—including large numbers of genotyped individuals—either by means of meta-analyses or by access to large biobanks (
6). The key insight in the Savage et al. study is to extend analysis to all available alcohol-related traits on a biobank level, in this case the UK Biobank, rather than accessing only those traits already thought to be related to relevant human pathologies. In all, 18 alcohol use traits were extracted, capturing a broad array of alcohol consumption traits, related problems, and preferences, to be used for genomic investigation.
The authors performed a sophisticated series of genetic analyses to study the interrelationship of these 18 alcohol traits by means of genomic structural equation modeling (gSEM) and additional downstream analyses to identify latent factors that may better characterize the genetic architecture underlying alcohol use traits. They found evidence via gSEM to suggest that a four-factor model best captured the genetic effects across this spectrum of alcohol use behaviors. This four-factor model fit the data well and was supported by traditional fit indices, aligning with factors that the authors identified as
Consumption,
Problems,
Beer Preference (
BeerPref), and
Atypical Preference (
AtypicalPref). Of these four factors, two, as the authors note, resemble previously identified AUD and quantity/frequency traits that have been studied extensively in the field—
Problems and
Consumption. They report a genetic correlation (r
g) of
Problems with PAU (
11) of 0.76, and of
Consumption with DPW (
10) of 0.80 (see Table S7 in the article’s supplement). However, the other two factors identified are less expected and more novel. The highest r
g for
BeerPref with a previously described alcohol trait is 0.56 (with alcohol dependence), and the highest r
g for
AtypicalPref, 0.34 (with DPW). The latter two,
BeerPref and
AtypicalPref, are designated according to the preferred type of alcohol consumed.
Is it possible that this really reflects an underlying genetic truth about alcohol use? It’s conceivable; food (
12) and beverage (
13) preferences are genetically influenced, and come down to factors such as how bitterness is experienced. Bitter beverage preference and alcohol intake share genetic risk loci (
5,
13). The underlying biology here may not reveal something fundamental about alcohol use directly; alcohol itself is the same no matter what drink it’s found in—it’s always ethanol. It seems more likely that this finding reveals something different about factors that influence beer preference and “atypical” preference, such as taste perception and sociodemographic factors. For example,
BeerPref and
AtypicalPref showed negative correlations with educational attainment—compared to
Consumption, which was positively correlated with that trait. The authors argue reasonably that they “uncover previously unobserved characteristics of the genetic architecture of AUBs.” However, the two drink selection–related clusters are not associated with
ADH1B, which could be an indication that they mostly index something less closely related to alcohol itself, or to physiological response to alcohol.
Another issue to consider is the trait-versus-state nature of the AUBs considered. Some are more related to current use (e.g., AUDIT-C score), and some more to habitual use (e.g., personal history of alcoholism)—more “state” for the former and more “trait” for the latter. For genetic studies, we are concerned mostly with traits, not states. We are born with our genomes and they don’t change. We do have experience from large genetic studies that state measures are surprisingly informative and often perform very well, but we must keep the distinction in mind. For example, as AUDIT-C is a state measure, this basic fact about the measure comes into play when we use it as a genetic marker when we consider people who have an AUDIT-C score of 0, that is, they are current nonusers, but who previously had alcohol use disorder (
14). In this case the state measure says no current alcohol use, but the trait measure indicates alcohol use disorder and any associated genetic predisposition for that trait, whether or not it is presently expressed. Type of alcoholic beverage preferred could be a trait measure, but it could also be influenced by age or, more broadly, time of life.
Notwithstanding what these factors capture specifically, the authors point out that they still “may be useful indices to identify individuals with different types of risk.” And thus, with a more comprehensive view of AUBs, we may be able to gain more insight into how people end up having different experiences with alcohol—experiences that may lead to differing outcomes. As the authors note, there are potential biases present in behavioral investigations and cross-cohort inconsistencies in terms of the presence of alcohol use problems, so replication in other large cohorts will be important. Especially if these findings reflect to some extent differences in environmental exposures, we should expect that the gSEM solutions may differ where the exposures differ.
In summary, considering all 18 possible alcohol-related measures from the UK Biobank that the authors could identify, a factor analysis landed on four factors. Two in effect recapitulated previously studied alcohol traits, and the other two centered on factors related to type of alcohol consumed. This work thus confirms and extends previous work on alcohol traits, demonstrates again the power of a deeply phenotyped biobank, and increases our understanding of what lies behind interindividual differences in alcohol consumption. In so doing, it advances the field in terms of our understanding of the major genetic factors that underlie the biology of a range of alcohol use behaviors.