Don’t throw the baby out with the bathwater: Enabling a bottom-up approach in genome-wide association studies: Figure 1.

The current model for conducting genome-wide association studies (GWAS) is primarily phenotype-driven.In this "top-down" approach, the model is the case-control study, where participants are enrolled based on the presence or absence of a clinical phenotype, for example, cardiovascular disease or breast cancer (Pennisi 2007; Wellcome Trust Case Control Consortium 2007).The International HapMap Project has identified a large number of single nucleotide polymorphisms (SNPs) in the human population that enable investigators to genotype subjects for these various polymorphisms and determine associations with a phenotype of interest (Fig. 1A) (International HapMap Consortium et al. 2007).Commercially available SNP arrays allow researchers to easily genotype from 100,000 to 1,000,000 SNPs per individual, providing "whole-genome" coverage (Affymetrix GeneChip System, http://www.affymetrix.com/products/system.affx; Illumina, Inc., http://www.illumina.com/pages.ilmn?ID=39).Typically, only a handful of these SNPs are associated with the particular phenotype under study and are present in only a small fraction of the study population because the minor alleles (variants) tend to be present at low frequencies.The vast majority of SNPs are not associated with the phenotype under study and are ignored because they are not relevant to the phenotype under investigation in the GWAS.For example, the Wellcome Trust Case Control Consortium genotyped 500,000 SNPs in 14,000 cases representing seven common diseases and an additional 3000 controls.From this extensive genotyping (17,000 ‫ן‬ 500,000), they identified 24 independent associations (Wellcome Trust Case Control Consortium 2007).The vast majority of the genotyping was not relevant to the seven phenotypes under investigation.The failure to use this "excess" genotypic information, by allowing future investigators to recontact participants and collect hypothesisdriven phenotypic information not collected in the original study, represents a tremendous missed opportunity.This type of "targeted phenotyping" would maximize the utility of data generated in GWAS and stored in existing databases, such as dbSNP (http://www.ncbi.nlm.nih.gov/projects/SNP/) and dbGaP (http://www.ncbi.nlm.nih.gov/sites/entrez?db=gap).Recently, scientists have begun to examine the functional consequences of human SNPs in the laboratory, based on data available from the HapMap Project and other SNP discovery studies, without any a priori knowledge of a human phenotype associated with a particular SNP.Having identified an interesting phenotype in vitro, the next step is to determine the conse-4