Many groups have followed up GWAS by sequencing associated regions in large numbers of samples, with the hope of identifying common causal variants as well as additional less common, more highly penetrant mutations. The WTCCC, for instance, sequenced hundreds of samples in 16 GWAS regions (hundreds of kilobases to a megabase surrounding the most significant SNP), a design that is able to identify nearby variants causing a synthetic association. Indeed, had NOD2 been sequenced, all three low frequency causal mutations would have been discovered. No clear examples of synthetic association were reported [22], which suggests that synthetic association is not commonplace. Nonetheless it is clear that both low- and high-frequency alleles do play a role in complex disease. Nejentsev et al. [23] recently sequenced several genes identified in a GWAS of type 1 diabetes in pools of cases, and discovered four rare coding mutations, each conferring an approximately two-fold increase in risk for diabetes. In contrast to the synthetic association model, however, these rare SNPs are not correlated with the common GWAS SNP, and a conditional analysis including these rare mutations did not affect the strength of association at the common GWAS SNP.
