Meta-analysis of genome-wide association studies

Individual genome-wide association studies have only limited power to find novel loci underlying complex traits and common diseases. With relatively modest sample and effect sizes, a true association between genotype and phenotype may never meet genome-wide statistical significance (P < 5 x 10(-8)) in a single study. Through meta-analysis, novel susceptibility loci can be discovered by effectively summing the statistical evidence of individually underpowered studies. Most genetic discoveries for complex traits are now made through meta-analysis collaborations, which so far have been restricted to single-locus analyses, testing for main effects at a single polymorphism at a time. A key benefit of this approach is that individual-level genotype (and phenotype) data do not need to be exchanged between research groups. In this article, we focus on meta-analysis at individual single-nucleotide polymorphisms (SNPs), paying particular attention to how imputation uncertainty can be incorporated into the association analysis and subsequent meta-analysis. Probably the most important aspect of genome-wide association meta-analysis is harmonization of the study results. As studies differ in design, sample collection, genotyping platforms, and association analysis methods, it is important that the association results (per SNP) of each study can be formatted, exchanged, and analyzed in such a way that the statistical evidence can be combined appropriately and that no valuable information is lost. Without minimizing the importance of having a clear phenotype definition (and corresponding measurements), we will assume that investigators representing the various studies have made sensible agreements about phenotype definitions, necessary sample exclusions, and appropriate covariate modeling.