Individuals of the same genotype do not have the same phenotype for quantitative traits when reared under common macro-environmental conditions, a phenomenon called micro-environmental plasticity. Genetic variation in micro-environmental plasticity is assumed in models of the evolution of phenotypic variance, and is important in applied breeding and personalized medicine. Here, we quantified genetic variation for micro-environmental plasticity for three quantitative traits in the inbred, sequenced lines of the Drosophila melanogaster Genetic Reference Panel. We found substantial genetic variation for micro-environmental plasticity for all traits, with broad sense heritabilities of the same magnitude or greater than those of trait means. Micro-environmental plasticity is not correlated with residual segregating variation, is trait-specific, and has genetic correlations with trait means ranging from zero to near unity. We identified several candidate genes associated with micro-environmental plasticity of startle response, including Drosophila Hsp90, setting the stage for future genetic dissection of this phenomenon.

A major challenge in modern biology is to understand how naturally occurring variation in DNA sequences affects complex organismal traits through networks of intermediate molecular phenotypes. This question is best addressed in a genetic mapping population in which all molecular polymorphisms are known and for which molecular endophenotypes and complex traits are assessed on the same genotypes. Here, we performed deep RNA sequencing of 200 Drosophila Genetic Reference Panel inbred lines with complete genome sequences and for which phenotypes of many quantitative traits have been evaluated. We mapped expression quantitative trait loci for annotated genes, novel transcribed regions, transposable elements, and microbial species. We identified host variants that affect expression of transposable elements, independent of their copy number, as well as microbiome composition. We constructed sex-specific expression quantitative trait locus regulatory networks. These networks are enriched for novel transcribed regions and target genes in heterochromatin and euchromatic regions of reduced recombination, as well as genes regulating transposable element expression. This study provides new insights regarding the role of natural genetic variation in regulating gene expression and generates testable hypotheses for future functional analyses.

Predicting complex phenotypes from genomic data is a fundamental aim of animal and plant breeding, where we wish to predict genetic merits of selection candidates; and of human genetics, where we wish to predict disease risk. While genomic prediction models work well with populations of related individuals and high linkage disequilibrium (LD) (e.g., livestock), comparable models perform poorly for populations of unrelated individuals and low LD (e.g., humans). We hypothesized that low prediction accuracies in the latter situation may occur when the genetics architecture of the trait departs from the infinitesimal and additive architecture assumed by most prediction models. We used simulated data for 10,000 lines based on sequence data from a population of unrelated, inbred Drosophila melanogaster lines to evaluate this hypothesis. We show that, even in very simplified scenarios meant as a stress test of the commonly used Genomic Best Linear Unbiased Predictor (G-BLUP) method, using all common variants yields low prediction accuracy regardless of the trait genetic architecture. However, prediction accuracy increases when predictions are informed by the genetic architecture inferred from mapping the top variants affecting main effects and interactions in the training data, provided there is sufficient power for mapping. When the true genetic architecture is largely or partially due to epistatic interactions, the additive model may not perform well, while models that account explicitly for interactions generally increase prediction accuracy. Our results indicate that accounting for genetic architecture can improve prediction accuracy for quantitative traits.

The ability to accurately predict complex trait phenotypes from genetic and genomic data is critical for the implementation of personalized medicine and precision agriculture; however, prediction accuracy for most complex traits is currently low. Here, we used data on whole genome sequences, deep RNA sequencing, and high quality phenotypes for three quantitative traits in the ~200 inbred lines of the Drosophilamelanogaster Genetic Reference Panel (DGRP) to compare the prediction accuracies of gene expression and genotypes for three complex traits. We found that expression levels (r = 0.28 and 0.38, for females and males, respectively) provided higher prediction accuracy than genotypes (r = 0.07 and 0.15, for females and males, respectively) for starvation resistance, similar prediction accuracy for chill coma recovery (null for both models and sexes), and lower prediction accuracy for startle response (r = 0.15 and 0.14 for female and male genotypes, respectively; and r = 0.12 and 0.11, for females and male transcripts, respectively). Models including both genotype and expression levels did not outperform the best single component model. However, accuracy increased considerably for all the three traits when we included gene ontology (GO) category as an additional layer of information for both genomic variants and transcripts. We found strongly predictive GO terms for each of the three traits, some of which had a clear plausible biological interpretation. For example, for starvation resistance in females, GO:0033500 (r = 0.39 for transcripts) and GO:0032870 (r = 0.40 for transcripts), have been implicated in carbohydrate homeostasis and cellular response to hormone stimulus (including the insulin receptor signaling pathway), respectively. In summary, this study shows that integrating different sources of information improved prediction accuracy and helped elucidate the genetic architecture of three Drosophila complex phenotypes.