For each of the three traits studied, the lead SNP is a variant other than rs671, although presumably the results are mediated by rs671, with its known functional charactersitics. Neither of the other lead variatants is itself predicted to be functional. And in fact, the lead SNPs are rather distant from rs671, although both are strongly correlated with that variant: rs143894582, lead SNP for flushing, is 665107 basepairs (bp) from rs671; and rs149212747, lead SNP for AD-CrC and MAXDRINKS, 404995 bp distant on the other side – a total span of over a megabase. This result most likely reflects random variation in the context of a large effect size in a comparatively small sample, and is illustrative of the difficulties commonly faced in identifying the true risk variant from a long associated region. Additionally, rs671 was genotyped directly, but the other lead variants were both imputed. Imperfect imputation could be an additional source of noise obfuscating the signal from rs671. In this case we know the major risk variant; it is a well-validated and well known functionally-null allele. Without that
