them collectively was larger than that if we only considered the leading SNP at that locus. For the loci at which the increasing alleles of at least two SNPs were negatively correlated, the SNP effects were underestimated in single-SNP analyses, meaning that some associated variants may be undetected. In other words, the joint analysis is more powerful than the single-SNP analysis in detecting such SNPs7,8. For example, rs1367226 at the EFEMP1 locus on chromosome 2 (Pvalue from the single-SNP meta-analysis (PM) = 0.198) and rs6784185 at the IGF2BP2 locus on chromosome 3 PM = 0.088) did not even show nominally significant association in single-SNP analyses, but they both reached genome-wide significance when fitted jointly with the leading SNPs at these two loci (Table 1). At the same time, the significance and effect sizes of the leading SNPs at the two loci also increased where PM = 1.1 × 10–28 versus the P value from the joint analysis (PJ) = 3.0 × 10–37 for rs3791675 and PM = 1.8 × 10–14 versus PJ = 1.5 × 10–22 for rs720390 (Table 1). There were 11 loci harboring more than 2 associated SNPs, with a maximum number of 4, and the length of
