In order to gain insight into this, we analyzed the effect size of these SNP variants on both the expression levels and the phenotypes. While the effect sizes of these trait-associated SNPs on eventual phenotypes were usually small, their intermediate (molecular) effects was often substantially larger. This supports the notion that the effect on e.g. MPV and MCV is through these trans-genes, and suggests the presence of ‘phenotypic buffering’, shown previously in plants [44], in humans (Table 2, Figure 4b): the effects of the 18 converging pairs of SNPs on gene expression levels were often substantially higher than the originally reported effect sizes on the trait-phenotypes. For example, several MPV- and MCV-associated SNPs explain between 1.41% and 10.99% of trans-expression variation within the 1,469 unrelated samples, whereas these SNPs only explain between 0.24% and 1.12% of the MPV and MCV phenotype variation (and as such required over 13,000 samples [41], [42] for identification, Figure 4b).
