To the Editor:
Over the past few years, large investments have been made in genome-wide association studies (GWAS) with the expectation that some of these studies would lead to the identification of novel therapeutic modalities or allow selection of patients who would respond better to therapeutic interventions. Although the results have provided valuable biological insights for many common diseases, the translation of the genetics findings from GWAS into the clinic remains limited and a topic of intense debate. Among the factors that could explain this situation are that the road from a gene target to an approved marketed drug takes in general more than ten years and most GWAS results have only been obtained over the past four years. Furthermore, because the effect size of the common variants identified by GWAS, alone or in aggregation, is generally modest, the impact in terms of personalized, individually tailored medicine has been negligible. We present here an analysis of another potential application of GWAS data—drug repositioning. In the following study, we assess the utility of GWAS in systematically and rapidly identifying alternative or refined indications for existing drugs.
The complete analysis of our workflow is described in Figure 1. Our approach began with the construction of a list of GWAS genes associated with disease traits. We used the catalog of published GWAS data from the US National Human Genome Research Institute (NHGRI; Bethesda, MD; http://www.genome.gov/gwasstudies). This resource contains an exhaustive description of trait/disease-associated single nucleotide polymorphisms (SNPs). At the time of our analysis (February 14, 2011) the GWAS catalog contained 796 publications with 4,818 rows of data, each row corresponding to an association between a trait and an index SNP. In an attempt to minimize the inclusion of false-positive signals, we eliminated associations annotated as not replicated, and with P > 1e−7 (Supplementary Methods). An additional 400 associations listed in Supplementary Table 1 were excluded because the associated traits were anthropometric and not relevant in the drug discovery context of our analysis. The remaining 1,515 rows from 361 publications referred to 1,099 gene names. Of these, 991 genes with recognizable HUGO gene names from Entrez Gene constituted the starting list for further analysis.