Controls for occult stratification do not appear to provide convincing alternative explanations for the data obtained here. Only 22 of the 937 clustered, reproducibly positive SNPs that we identify here using approach (1) also display sizable allele frequency differences based on ethnicity. Since we would have expected 24 by chance, it appears highly unlikely that stratification based on racial/ethnic differences between each abuser and corresponding control samples provides a major basis for the addiction-associated allelic variants identified herein. Principal components analyses identify robust principal components that cleanly separate African-American from European-American samples and account for about 94% of the variance when both samples are analyzed together (data not shown). However, there is significant distinction between substance dependent vs control pools based on additional principal components that are orthogonal to those that distinguish individuals with African vs European heritages. Analyses of the likelihood that substance dependent samples would be distinguished from control samples by this principal component based on chance yield p = 0.00003 and p = 0.057 probabilities in European- and African-American samples, respectively (data not shown).
