To demonstrate the conservative nature of our model selection strategy, we performed selection of SNPs with a less stringent P value threshold. We constrained the analysis to region 1 Mb up- or downstream of the 180 known height-associated SNPs and chose a P value threshold of 5 × 10–7, as only ~13% of the genome is covered by these 2-Mb regions. We identified 85 additional associated SNPs at 60 loci, which explained 2.4% of variance in the discovery set. We validated the joint effects of these 85 SNPs by our prediction analyses. The prediction R2 were 2.4% (P = 8.6 × 10–37) and 1.9% (P = 5.3 × 10–18) in the ARIC and QIMR cohorts, respectively, suggesting that we could detect more associated variants with a less stringent threshold but, of course, might increase the risk of including false positives. Nevertheless, this analysis confirms more heritability can be explained at a substantial proportion of loci that affect the trait. It also suggests a model of genetic architecture of a large number of loci and multiple causal variants at many of these regions.
