whole-genome estimation approach we recently developed, which is also lower than the corresponding explained heritability for height (~56%)14,18. In a previous analysis partitioning genetic variance onto individual chromosomes, the variance explained by each chromosome showed a strong linear relationship with chromosome length for height, but such a relationship was rather weak for BMI18. To investigate whether additional variants for BMI could be detected, we performed a conditional and joint analysis with a less stringent P value threshold of 5 × 10–6, with the LD structure estimated from the ARIC cohort. We identified 19 multiple associated SNPs (9 leading and 10 additional SNPs) at 9 loci (Supplementary Tables 1 and 5), which is still much lower than the number of additional variants detected for height. The ten additional SNPs explained 0.21% of the variance in the discovery set. When using these SNPs to predict the BMI phenotypes in the QIMR cohort, the prediction R2 was 0.13%, which is nominally significant (P = 0.037). Taken together, the previous and current results are consistent and suggest that the genetic architectures for height and BMI might be different in terms of the allelic spectrum of causal variants within and between loci, the distribution of
