To reduce the scope of hypothesis testing, multi-SNP haplotypes were constructed only when two or more phenotypes were significant (p≤0.05) within any one gene. To avoid constructing haplotypes based on SNPs providing redundant information, only SNPs with low pairwise LD (r2<0.50) were used in haplotype analysis. Each haplotype was then examined to determine whether significant association results were due to the overtransmission of a particular haplotype to affected individuals or to the differential transmission of particular haplotypes to siblings discordant for the phenotype. Except for the severe withdrawal phenotype, haplotypes were estimated using phase-certain genotyped individuals in the program UNPHASED (Dudbridge, 2003). Due to the small number of both affected and unaffected subjects for the severe withdrawal phenotype, missing haplotypes were estimated using the EM algorithm. All haplotypes with a frequency less than 0.05 were omitted from association analyses.
