when considering illegal drugs39,41. Although information from Yale-Penn and Army STARRS cohorts were previously combined to replicate genetic associations related to SD and suicidality8,11, the limited availability of information regarding polysubstance dependence in Army STARRS participants likely prevented us from replicating the findings observed in the Yale-Penn participants. Additionally, the differences in the demographic characteristics of the two cohorts may also have contributed to reducing the power of our replication analysis. Accordingly, the findings presented will need to be replicated in independent samples that are adequately powered. Finally, accounting for heritable covariates in association tests can lead to spurious associations due to collider bias52. Although the StructLMM interaction test is robust to this confounding effect27, gene-exposure associations may alter the interpretation of interactions, reflecting epistatic relationships between genetic factors. The loci identified in the present study were not previously identified as associated with substance use, abuse, and dependence. This supports that the interactions identified are not due to epistatic effects.
