either heroin or amphetamine dependence. Separate analyses of case and control data found significant effects in both groups, demonstrating that our findings were not limited to heroin dependent cases. BIS data were not available on all participants and this could have introduced some bias into analyses examining mediation. An examination of risk for missing BIS data using logistic regression found a significant effect for case status (66.1% of those missing BIS data versus 80.3% with data; OR 0.48; 95%CI 0.38–0.61), but not for PTSD. Including case status as a covariate in an examination of missing BIS data, no significant effects were found for the various +/− amphetamine dependence, +/− rs12364283 G allele status combinations. Thus, these missing data likely resulted only in reduced power which likely contributed to the comparison of risk associated with the amphetamine dependence +, any rs12364283 G allele + versus amphetamine dependence −, any rs12364283 G allele + falling just below statistical significance. Another potential limitation, population admixture, was addressed by observing that the inclusion of two principal components as covariates in analyses did not change our findings. Regardless, generalizability of our findings to samples from other regions and of differing ethnicity will need to be
