Inference of Natural Selection from Interspersed Genomically coHerent elemenTs

Drosophila melanogaster polymorphism sequence data was obtained from the Drosophila Genetic Refrence Panel (DGRP). This resource contains the complete genome sequences of 205 unrelated fly lines from a homogeneous Raleigh, North Carolina fruit-fly population. Genotype calls for these lines were extracted from the Variant Call Format (VCF) files downloaded from the DGRP freeze 2 website on May 2013. We considered only single nucleotide polymorphisms (SNPs) and discard sites deemed as structural variants within these VCF files. All other positions not reported in the VCF files were assumed to be monomorphic for the reference allele (according to UCSC dm3 / BDGP Release 5).

The polymorphism data were summarized by recording for each position in dm3 the allele count for each of the four basses (A,C,G, and T) across the 205x2=410 chromosomes. Sites with more than two observed alleles (i.e. tri-allelic sites, etc) were masked. Additionally, sites with missing data for 5 or more of the 205 fly lines were masked (see filters).

Divergence was inferred using four melanogaster-subgroup outgroup genomes: D. simulans (droSim2), D. sechellia (droSec2), D. yakuba (droYak3) and D. erecta (droEre2). We created a custom 5-way alignment of these outgroup species and D. melanogaster (dm3) using the LASTZ and chain/net procedure prescribed by the UCSC genome browser. All genome assemblies, excluding D. simulans, were downloaded from the UCSC genome browser. We utilized a higher-quality, recently-released D. simulans genome assembly, named droSim2, in our custom alignment.

For each position in dm3, we recorded the aligned base from each of the 4 non-melanogaster fruit-fly species, or an indication that no syntenic alignment was available at that position (see filters).

A prior distribution for the ancestral state (Z) was computed for all non-filtered sites in dm3, by assuming a phylogeny estimated from four-fold degenerate sites (see Figure), and applying the postprob.msa function in RPHAST. The D. melanogaster (dm3) sequence was masked in this computation, so that the computed distribution corresponds to the distribution over the bases in the ancestral sequence (Z) given the other four genomes, which is used as a prior distribution (P(Z|O)) by the INSIGHT model.

DGRP freeze 2 datasets were only provide for the homozygous blocks of the autosomes and the X chromosomes (i.e. 2L, 2R, 3L, 3R, 4, X). Polymorphic sites within heterozygous blocks (i.e. 2LHet, 2RHet, 3LHet, 3RHet, XHet, and YHet) or for the mitochondrial DNA were not provided. Thus, our analysis was restricted to the homogous chromsomal regions. We applied various filters to reduce the impact of technical errors from alignment, sequencing, genotype inference, and genome assembly. Our filters included repetitive sequences (simple repeats), recent transposable elements, recent segmental duplications, and CpG site pairs. CpG site pairs (prone to hypermutability) were identified as position pairs having a “CG” dinucleotide in the D. melanogaster reference genome. Non-syntenic regions, as identified from synthenic blocks residing on low-quality Nets, and gaps in the outgroup alignment were hard-masked (by “N”s) individually in each outgroup genome. This uncertainty was incorporated when estimating the prior distribution over the ancestral sequence (Z, see above).

Sites with missing data in greater than or equal to 5 of the 205 fly lines were masked out completely (see above). Additionally, sites with more than two observed alleles in the fly population data (i.e. tri- or quad-allelic sites) were masked. We treated lines without a reported genotype and lines with "N" base assignments as lines with "missing" data. At polymorphic sites where less than 5 individuals contained missing data, we subsampled 200 fly lines without replacement in order to arrive an a uniform number of 400 alleles per site. This subsampling approach is a common method utilized in processing genotype data with missing information, and offers consistency and accuracy in the treatment of polymorphisms within INSIGHT.

For estimation of genome-wide neutral polymorphism and divergence rates, we used a fixed collection of 1kb non-overlapping windows. We used the putative neutral sites in each 1 kb window to estimate a neutral polymorphism rate (θ_b) and a neutral divergence rate (λ_b), and those estimates were then associated with the appropriate genomic block.