From 2019... Educational attainment impacts drinking behaviors and risk for alcohol dependence: results from a two-sample Mendelian randomization study with ~780,000 participants

From 2019... Educational attainment impacts drinking behaviors and risk for alcohol dependence: results from a two-sample Mendelian randomization study with ~780,000 participants. Daniel B. Rosoff, Toni-Kim Clarke, Mark J. Adams, Andrew M. McIntosh, George Davey Smith, Jeesun Jung & Falk W. Lohoff. Molecular Psychiatry volume 26, pages1119–1132. Oct 2019. https://www.nature.com/articles/s41380-019-0535-9

Rolf Degen's take: The genetic instruments associated with higher educational attainment are linked to reduced binge drinking, reduced amount of alcohol consumed per occasion, but to increased frequency of alcohol intake, especially of wine

Abstract: Observational studies suggest that lower educational attainment (EA) may be associated with risky alcohol use behaviors; however, these findings may be biased by confounding and reverse causality. We performed two-sample Mendelian randomization (MR) using summary statistics from recent genome-wide association studies (GWAS) with >780,000 participants to assess the causal effects of EA on alcohol use behaviors and alcohol dependence (AD). Fifty-three independent genome-wide significant SNPs previously associated with EA were tested for association with alcohol use behaviors. We show that while genetic instruments associated with increased EA are not associated with total amount of weekly drinks, they are associated with reduced frequency of binge drinking ≥6 drinks (ßIVW = −0.198, 95% CI, −0.297 to –0.099, PIVW = 9.14 × 10−5), reduced total drinks consumed per drinking day (ßIVW = −0.207, 95% CI, −0.293 to –0.120, PIVW = 2.87 × 10−6), as well as lower weekly distilled spirits intake (ßIVW = −0.148, 95% CI, −0.188 to –0.107, PIVW = 6.24 × 10−13). Conversely, genetic instruments for increased EA were associated with increased alcohol intake frequency (ßIVW = 0.331, 95% CI, 0.267–0.396, PIVW = 4.62 × 10−24), and increased weekly white wine (ßIVW = 0.199, 95% CI, 0.159–0.238, PIVW = 7.96 × 10−23) and red wine intake (ßIVW = 0.204, 95% CI, 0.161–0.248, PIVW = 6.67 × 10−20). Genetic instruments associated with increased EA reduced AD risk: an additional 3.61 years schooling reduced the risk by ~50% (ORIVW = 0.508, 95% CI, 0.315–0.819, PIVW = 5.52 × 10−3). Consistency of results across complementary MR methods accommodating different assumptions about genetic pleiotropy strengthened causal inference. Our findings suggest EA may have important effects on alcohol consumption patterns and may provide potential mechanisms explaining reported associations between EA and adverse health outcomes.

Discussion

Using large summary-level GWAS data and complementary two-sample MR methods, we show that EA has a likely causal relationship with alcohol consumption behaviors and alcohol dependence risk in individuals of European Ancestry. More specifically, higher EA reduced binge drinking (six or more units of alcohol), the amount of alcohol consumed per occasion, frequency of memory loss due to drinking, distilled spirits intake, and AD risk. EA increased the frequency of alcohol intake, whether alcohol is consumed with meals, and wine consumption. We found evidence that our results may be driven by genetic pleiotropy in only two of the eight alcohol consumption behaviors (average weekly beer plus cider intake and alcohol usually taken with meals) and significance remained after additional analysis using EA instruments with SNPs nominally associated with either cognition or income suggest that EA may be an important factor responsible for variation in alcohol use behaviors. Consistency of our results across MR methods also strengthens our inference of causality.

Educated persons generally have healthier lifestyle habits, fewer comorbidities, and live longer than their less educated counterparts [52], and our results suggest EA is causally associated with different likelihoods of belonging to variegated alcohol consumer typologies. We found that an additional 3.61 years of education reduced the risk of alcohol dependence by ~50%, which is consistent with results from small community samples [53], and the two most recent alcohol dependence GWASs findings strong inverse genetic correlations with educational attainment [27, 54]. Notably, binge drinking significantly increases the alcohol dependence risk [55], and distilled spirits and beer consumption account for the majority of hazardous alcohol use [56]. Furthermore, compared to wine drinkers, beer and spirits drinkers are at increased risk of becoming heavy or excessive drinkers [57], for alcohol-related problems and illicit drug use [58, 59], and AD [57]. Our findings related to alcoholic drink preferences, when combined with our results showing increased binge drinking, memory loss due to alcohol, and a suggestive relationship with remorse after drinking, imply a pattern of alcohol consumption motivated to reduce negative emotions or becoming intoxicated [14].

In contrast to the often-reported positive association between EA and total amount of alcohol consumption reported from observational studies [18, 60], we found little evidence of a causal relationship. This null finding may be reconciled by the opposing influences on alcohol intake frequency and total alcohol consumed per occasion, which, while not leading to an overall change in total consumption, nonetheless significantly affect the pattern. Our null finding regarding total consumption does support similar results from Davies et al. [52], who used the 1972 mandated increase in school-leaving age in the UK as a natural experiment instrumental variable design to investigate the causal effects of staying in school on total alcohol consumption (from individuals in the UKB sample who turned 15 in the first year before and after the schooling age increased). Davies et al. may have found a significant effect of staying in school had they included the disaggregated behavioral dimensions of alcohol consumption behaviors. Nevertheless, even if no EA-total alcohol consumption relationship exists, studies have reported that both the specific alcoholic beverage and the pattern with which it is consumed, controlling for total consumption, independently contribute to risky health behaviors [61, 62].

Natural experiments [52, 63], and twin studies have found that differences in EA, even after controlling for shared environmental factors, still significantly impact mortality risk [64,65,66], and recent large Mendelian randomization studies have demonstrated inverse relationships between EA on smoking behaviors [35] and coronary heart disease (CHD) risk [34] add to the growing body of literature, suggesting a causal effect of increased EA on health and mortality. Other observational studies have linked alcohol consumption patterns to health, disease, and mortality risk [67,68,69]. In particular, binge drinking may have dramatic short-term consequences, including motor vehicle accidents, alcoholic coma, cerebral dysfunction, and violent behavior [70], as well as long-term effects such as hypertension, stroke, and other cardiovascular outcomes [71]. A recent MR study showed that smoking mediates, in part, the effect of education on cardiovascular disease [72], and our results suggest that differences in alcohol consumption patterns may also be another mediator. Health consequences incur significant costs with binge drinking accounting for ~77% of the $249 billion alcohol-related costs (lost workplace productivity, health care expenses, law enforcement, and criminal justice expenses, etc.) in the United States in 2010 [55].

While we do not fully understand the underlying biological mechanisms through which the instrument SNPs influence EA, they are primarily found in genomic regions regulating brain development and expressed in neural tissue. These SNPs demonstrate significant expression throughout the life course, but exhibit the highest expression during development [36]. For example, rs4500960, which was associated with reduced EA, is an intronic variant in the transcription factor protein, T-box, Brain 1 (TBR1), that is important for differentiation and migration of neurons during development [36], while rs10061788 is associated with cerebral cortex and hippocampal mossy fiber morphology [36]. It is, however, important to note that interpreting these SNPs as representing “genes for education” may be “overly simplistic” since EA is strongly affected by environmental factors [36]. Our results remained when using an EA instrument with SNPs nominally associated with income removed, suggesting that an individual’s genetics may impact behavior development, which then increases EA [73]. Conversely, genetic estimates of EA and its correlations with other complex social phenotypes using population-based samples may be susceptible to biases, such as assortative mating and dynastic effects that provide pathways alternate to direct biological effects [40]. For example, EA-associated genetic influence on parental behavior could causally affect the child’s environment [73]. Using polygenic scores for EA, Belsky et al. [73] recently found the mothers’ EA-linked genetics actually predicted their children’s social attainment better than the child’s own EA-linked genetics, suggesting an effect mediated by environmental effects. While policies are not able to change children’s genes, or their inherited social status, they can provide resources [73], and our results suggest that interventions to increase education may help improve health outcomes through changing alcohol consumption patterns.

Notably, there was evidence for some causal effects of alcohol consumption patterns on EA, and the divergent effects again demonstrate the importance of separating drinking variables. However, we failed to find evidence that total alcohol consumed, binge drinking, or AD impacts EA, which is in line with observational studies finding no, or small effects [21], and suggests that other studies findings a negative effect [21] may be due to confounding. Alternatively, EA may not be sensitive enough to detect changes in schooling, e.g., grade point average [21], falling behind in homework and other academic difficulties that also reported association  with heavy drinking [74]. Further, there are currently no adolescent drinking behavior GWAS, so the temporal sequence of these analyses should be considered during their interpretation. Our findings, therefore, need replication when GWAS on adolescent alcohol consumption patterns becomes available.

Exploratory sex-specific analyses revealed differences in certain aspects of the relationship between EA and alcohol consumption. For men, the relationship between their consumption of red wine, beer, and whether they drink with meals was more sensitive to changes in EA than for women. Conversely, the reduction in binge drinking with increased EA may be driven by its effect for women since its effect on men was not significant. In addition, in women the negative effect of EA on spirit consumption was more than double its effect on men. We found no differences among the AUDIT question.

There are noted gender gaps in alcohol use and associated outcomes due to a combination of physiological and social factors [39]. Notably, Huerta et al. [75] found sex-specific effects of EA and academic performance on the odds of belonging to different alcohol consumption typologies (ranging from “Abstainer” to “Regular Heavy Drinker with Problems”). The absence of any association in males may be due to their inability to model binge drinking [75]; however, our results suggest otherwise. Additionally, the recent Clarke et al. [28] total weekly alcohol GWAS found sex-specific genetic correlation differences with an rg = 0.1 in men and 0.33 in women. Taken together, our findings suggest EA may partially account for some of these observed gender gaps in alcohol consumption, but not others. We should note that the only available sex-specific EA GWAS had significant overlap ( ≥18.9%) with the outcome datasets, so our exploratory sex-specific analysis used the same EA GWAS combining men and women. The lack of available sex-specific AD GWAS also meant we were unable to examine differences in AD risk. Notably, the sex-specific EA GWAS demonstrated nearly identical effect sizes between men and women, which support the validity of the estimates derived from the combined-sex EA GWAS, but future studies using sex-specific instruments are required.

Strengths and limitations

We note several strengths. We have analyzed multiple alcohol-related behavioral phenotypes, which support the consistency of our results. We have implemented multiple complementary MR methods (IVW, Egger, weighted median, and weighted mode MR) and diagnostics. Consistency of results across MR methods (accommodating different assumptions about genetic pleiotropy) strengthens our causal interpretation of the estimates [76]. We also used the largest publicly available GWASs for both exposure and outcome samples; large summary datasets are important for MR and other genetic analysis investigating small effect sizes [77]. We also note limitations and future directions. There is minimal sample overlap between the exposure SSGAC GWAS and the outcome PGC GWAS (AD), but there may still be individuals participating in multiple surveys, which event we cannot ascertain with available summary-level GWAS statistics. Further, the GWASs cohorts are from Anglophone countries, where beer is the preferred drink [78]; therefore, applicability to other countries with different alcohol preferences may be limited. Further still, it has been reported the UKB sample is more educated, with healthier lifestyles, and fewer health problems than the UK population [79], which may limit the generalizability to other populations. Replication of these findings using alcohol use information from different ethnicities is necessary. EA only measured years of completed schooling; determining how various aspects of education differentially impact alcohol consumption was not possible but should be a topic of future work. Finally, alcohol consumption is not stable over time [15]; however, the alcohol consumption outcomes correspond to current drinking behavior, which may have led to the misclassification of some individuals. The current drinking also impacts the temporal relationship of our bidirectional analyses, since the current alcohol intake likely occurred after maximum educational attainment for most of the participants. Future GWAS that evaluate drinking behavior during adolescence, or other longitudinal studies are necessary to confirm these findings and better elucidate the impact of alcohol intake on EA.