Long-read sequencing technology promises to greatly enhance de novo assembly of genomes for nonmodel species. Although the error rates of long reads have been a stumbling block, sequencing at high coverage permits the self-correction of many errors. Here, we sequence and de novo assemble the genome of Drosophila serrata, a species from the montium subgroup that has been well-studied for latitudinal clines, sexual selection, and gene expression, but which lacks a reference genome. Using 11 PacBio single-molecule real-time (SMRT cells), we generated 12 Gbp of raw sequence data comprising ∼65 × whole-genome coverage. Read lengths averaged 8940 bp (NRead50 12,200) with the longest read at 53 kbp. We self-corrected reads using the PBDagCon algorithm and assembled the genome using the MHAP algorithm within the PBcR assembler. Total genome length was 198 Mbp with an N50 just under 1 Mbp. Contigs displayed a high degree of chromosome arm-level conservation with the D. melanogaster genome and many could be sensibly placed on the D. serrata physical map. We also provide an initial annotation for this genome using in silico gene predictions that were supported by RNA-seq data.

The interaction of evolutionary processes to determine quantitative genetic variation has implications for contemporary and future phenotypic evolution, as well as for our ability to detect causal genetic variants. While theoretical studies have provided robust predictions to discriminate among competing models, empirical assessment of these has been limited. In particular, theory highlights the importance of pleiotropy in resolving observations of selection and mutation, but empirical investigations have typically been limited to few traits. Here, we applied high-dimensional Bayesian Sparse Factor Genetic modeling to gene expression datasets in 2 species, Drosophila melanogaster and Drosophila serrata, to explore the distributions of genetic variance across high-dimensional phenotypic space. Surprisingly, most of the heritable trait covariation was due to few lines (genotypes) with extreme [>3 interquartile ranges (IQR) from the median] values. Intriguingly, while genotypes extreme for a multivariate factor also tended to have a higher proportion of individual traits that were extreme, we also observed genotypes that were extreme for multivariate factors but not for any individual trait. We observed other consistent differences between heritable multivariate factors with outlier lines vs those factors without extreme values, including differences in gene functions. We use these observations to identify further data required to advance our understanding of the evolutionary dynamics and nature of standing genetic variation for quantitative traits.