To infer the haplotypes occurring between the two ZNF804A SNPs, we phased the study genotypes in each ancestry group using the ShapeIT (version 2) program (Howie et al., 2012) with all 1000 Genomes reference haplotype panels, 500 conditioning states, recommended effective population sizes of 15,000 for African Americans and 11,418 for European Americans, and default settings for all other program options specified. Logistic regression models were used to test the two-SNP haplotypes (coded as dummy variables) for association with heroin addiction, adjusted for sex and eigenvectors. Ancestry-specific SNP and haplotype association results were combined using fixed-effects, inverse variance-weighted meta-analysis, the most commonly used approach for large-scale genetic association meta-analysis (Panagiotou et al., 2013). Fixed-effects methods, which optimize power for discovery analyses (Pereira et al., 2009), assume that heterogeneity across study-specific results are due to random error rather than true population differences; the inverse variance weighting involves taking the inverse of each study-specific standard error estimate and accounting for the direction of association (i.e., sign of the β) to generate combined regression coefficients.
