GRS comprised of >110 000 SNPs only captured very modest proportions of the variance in any alcohol-related measure (<1%). This observation is consistent with other studies.30, 31, 32 For instance, Vink et al.32 used GRS constructed from a large meta-analysis of GWAS of tobacco smoking measures to predict variance in alcohol, tobacco and cannabis-related outcomes. In that study, polygenic scores that were associated with tobacco-related measures at P<10−70 explained, at most, 1.5% of the variance in any substance-related outcome. Similarly, Power et al.30 examined the relationship between cannabis involvement and GRS generated from a meta-analysis of schizophrenia (N=13 833 cases, 18 310 controls) which included 13 genome-wide significant loci. Even though schizophrenia GRS were significantly associated with cannabis use, the scores, even at P<0.05 explained <1% of the variance in cannabis-related phenotypes. For alcohol-related measures, Salvatore et al.31 found that GRS generated for alcohol problems (N=4304) only predicted 0.6% of the variance in a similar measure in an independent sample. Despite relying on a smaller discovery sample (COGA, N=1788), our findings are consistent with these estimates. Nonetheless, the small sample
