Building bridges from genome to physiology using machine learning and Drosophila experimental evolution

Abstract

Drosophila experimental evolution, with its well-defined selection protocols, has long supplied useful genetic material for the analysis of functional physiology. While there is a long tradition of interpreting the effects of large-effect mutants physiologically, in the genomic era identifying and interpreting gene-to-phenotype relationships has been challenging, with many labs not resolving how physiological traits are affected by multiple genes throughout the genome. Drosophila experimental evolution has demonstrated that multiple phenotypes change due to the evolution of many loci across the genome, creating the scientific challenge of sifting out differentiated but noncausal loci for individual characters. The fused lasso additive model method (FLAM) allows us to infer some of the differentiated loci that have relatively greater causal effects on the differentiation of specific phenotypes. The experimental material used in the present study comes from 50 populations that have been selected for different life-histories and levels of stress resistance. Differentiation of cardiac robustness, starvation resistance, desiccation resistance, lipid content, glycogen content, water content, and body masses was assayed among 40 to 50 of these experimentally-evolved populations. Through FLAM, we combined physiological analysis from eight parameters with whole-body pooled-seq genomic data to identify potentially causally linked genomic regions. We have identified approximately 1,900 significantly differentiated 50 kb genomic windows among our 50 populations, with 161 of those identified genomic regions highly likely to have a causal effect connecting specific genome sites to specific physiological characters.