Given the replicated associations between rs603600 and childhood adversity predicting cannabis dependence symptoms and the biological extension of that interaction to predict right basolateral amygdala habituation, a post-hoc structural equation model (SEM) was tested in an opioid-dependent subset of CATS (n = 1,182) that integrated a dichotomous measure of using cannabis to control mood6 (see Supplement for additional methodological details; Figure 4). The rs604300 × CSA interaction significantly predicted using cannabis to control mood (bGxE = -2.04, 95% CI [-3.78, -0.20], p = .025), which, in turn, significantly predicted cannabis dependence symptoms (bM = 0.84, 95% CI [0.71, 0.96], p < .001). The rs604300 × CSA interaction was specific to cannabis: it did not predict the use of other types of substances, or of substances in general, to control mood (all ps> 0.20). The SEM overall demonstrated good fit (relative χ2< 0.001, RMSEA < 0.001, WRMR = 0.40, CFI = 1.00). The indirect pathway from the GxE interaction to cannabis dependence symptoms through using cannabis to control mood was also significant (bIND = -1.71, 95% CI [-3.26, -0.21], p =
