The current study utilized subjects ascertained for studying alcohol (COGA), nicotine (COGEND), and cocaine (FSCD) dependence, and none of the studies excluded individuals because of their involvement with other substances. Our analyses did account for mean differences/batch effects between the different samples within SAGE by controlling for study origin in all analyses. However, our estimates are uncorrected for ascertainment because there is currently no way in GCTA to correct for ascertainment on continuous phenotypes, such as liability for substance dependence, which we focused on here. Our estimates of SNP-heritability are therefore relevant to a population with the same distribution of liability as that observed in our sample; we expect that the SNP-heritability is lower in the general (non-ascertained) population (42). Future research, using larger epidemiologic samples recruited for population representativeness rather than substance abuse is needed to replicate the present findings. However, given the low rates of comorbid drug involvement in community- and general-population samples, samples in the tens of thousands will be necessary to obtain sufficient variability across alcohol, tobacco, cannabis, and in particular illicit substances. It should also
