As secondary sensitivity analysis in both GWAS discovery samples, we investigated the called CNV states at a genome-wide level, i.e. the unphased total number of copies at a genomic locus. We determined CNV calls for each individual and each marker using the Affymetrix Genotyping Console 3.0. In particular, we applied a five-state hidden Markov model (HMM) (39) to smooth and segment the data with default values of 0.2 for each prior probability and 1000 Mb as a transition decay. The five possible HMM copy number states (CNS) were: 0, homozygous deletions; 1, heterozygous deletions; 2, (normal) diploid case; 3, single copy gains; 4+, cases of four or more copies. The CNV calling requires comparison of individual signal intensities against a reference sample. Owing to computational constraints, we processed a reference sample of 106 parental pairs from the family-based GWAS sample. To check whether and to what extent the choice of the reference sample had an impact on CNS calling, we decided to choose two different reference samples, each comprising 106 parental pairs. Reference Sample 1 was a random sample of
