The power of a multivariate approach to genome-wide association studies: an example withDrosophila melanogasterwing shape

Abstract

Due to the complexity of genotype-phenotype relationships, simultaneous analyses of genomic associations with multiple traits will be more powerful and more informative than a series of univariate analyses. In most cases, however, studies of genotype-phenotype relationships have analyzed only one trait at a time, even as the rapid advances in molecular tools have expanded our view of the genotype to include whole genomes. Here, we report the results of a fully integrated multivariate genome-wide association analysis of the shape of the Drosophila melanogaster wing in the Drosophila Genetic Reference Panel. Genotypic effects on wing shape were highly correlated between two different labs. We found 2,396 significant SNPs using a 5%FDR cutoff in the multivariate analyses, but just 4 significant SNPs in univariate analyses of scores on the first 20 principal component axes. A key advantage of multivariate analysis is that the direction of the estimated phenotypic effect is much more informative than a univariate one.Exploiting this feature, we show that the directions of effects were on average replicable in an unrelated panel of inbred lines. Effects of knockdowns of genes implicated in the initial screen were on average more similar than expected under a null model. Association studies that take a phenomic approach in considering many traits simultaneously are an important complement to the power of genomics. Multivariate analyses of such data are more powerful, more informative, and allow the unbiased study of pleiotropy.Forward genetic analyses are generally built on a single measurable quantity, such as size, color, or the presence/absence of a distinctive organismal feature. The rise of phenomics, with its . CC-BY-NC-ND 4.0 International license peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not . http://dx.doi.org/10.1101/108308 doi: bioRxiv preprint first posted online Feb. 14, 2017; P i t c h e r s e t a l .-3 emphasis on high-throughput measurement of high-dimensional traits, is beginning to allow us to address the genetics of more complex traits that no single measurement can capture (Houle 2010; Houle et al. 2010). For instance, any one measurement of the wing of a fly, such as the length, incompletely captures wing size and shape Houle and Fierst 2013;Pitchers et al. 2013).Despite the growing enthusiasm for a more comprehensive approach to the phenotype, the vast majority of genome-wide association studies (GWAS) that include more than one trait have undertaken multiple univariate analyses for each site, rather than a single multivariate analysis (e.g., Teslovich et al. 2010;Battle et al. 2014). Statisticians have long appreciated the value of genuinely multivariate approaches to association studies (Lange et al. 2003;Shriner 2012), and this has led to a recent flowering of multivariate methods and software (O'Reilly et al. 2012;Stephens 2013; van der Sluis et al. 2013;Scutari et al. 2014;Zhou and Stephens 2014;Schaid et al. 2016). ...