Faster-X Sequence Divergence

To determine whether the specific set of genes we examined showed faster-X sequence evolution in the mel-mel, sim-sech, and/or mel-sim comparisons, we contrasted percent sequence divergence for coding (all transcribed sites) and noncoding regions of individual genes between genes on the X chromosome and genes on the autosomes (see Materials and Methods). We found that sequence divergence was significantly greater on the X chromosome than on autosomes for both coding (fig. 1A) and noncoding sequences (fig. 1B) in both interspecific comparisons (Mann–Whitney U test, P_{sim-sech coding} = 9.0 × 10⁻⁸, P_{mel-sim coding} = 2.2 × 10⁻¹¹, P_{sim-sech noncoding} = 7.1 × 10⁻¹², P_{mel-sim noncoding} = 0.0017), consistent with previous studies (Begun et al. 2007; Langley et al. 2012; Mackay et al. 2012; Hu et al. 2013). Greater differentiation of X-linked sequences between two strains of the same species (mel-mel) is not predicted by the faster-X theory, yet we also observed greater differentiation for the X chromosome than autosomes for the mel-mel comparison (Mann–Whitney U test, P_{mel-mel coding} < 1.0 × 10⁻¹⁵, P_{mel-mel noncoding} < 1.0 × 10⁻¹⁵). Prior studies of sequence differentiation between North American and African strains of D. melanogaster have also observed a faster-X pattern of sequence differentiation (Yukilevich et al. 2010; Langley et al. 2012), and North American and African strains of D. melanogaster have been suggested to be in the early stages of speciation (Hollocher, Ting, Pollack, et al. 1997; Hollocher, Ting, Wu, et al. 1997). Our mel-mel comparison therefore most likely captures a level of differentiation somewhere between intraspecific polymorphism and interspecific divergence.

Open in a separate window

Fig. 1.

Faster-X divergence of DNA sequence. Percent sequence divergence is shown for coding (A) and noncoding (B) regions of each of the 4,851 genes analyzed in all three comparisons (mel-mel, sim-sech, mel-sim). Maximum-likelihood estimates of (C) nonsynonymous (d_N) and (D) synonymous changes (d_S) as well as (E) the ratio of nonsynonymous to synonymous changes (d_N/d_S) are shown. Boxplots show the distribution of these values separately for genes on the X chromosome and autosomes with the median (center line) as well as 25th and 75th percentiles indicated (box upper and lower bounds) and whiskers extending 1.5× the interquartile range. Statistical significance of the difference in median values between X-linked and autosomal genes was determined using Mann–Whitney U tests, where NS indicates not significant, *P ≤ 0.05, **P ≤ 0.001, and ***P ≤ 1 × 10⁻⁴.

To investigate the evolutionary forces shaping the patterns of sequence variation we observed, we used PAML (Yang 2007) to generate maximum-likelihood estimates of coding sequence divergence at synonymous sites (d_S), nonsynonymous sites (d_N), and nonsynonymous relative to synonymous sites (d_N/d_S) for each comparison. We found that both d_N (fig. 1C) and d_S (fig. 1D) were significantly greater for genes on the X chromosome than genes located on autosomes in all comparisons (Mann–Whitney U test, P_{mel-mel dN} = 0.005, P_{sim-sech dN} = 0.05, P_{mel-sim dN} = 0.003, P_{mel-mel dS} = 1.0 × 10⁻¹⁵, P_{sim-sech dS} = 1.0 × 10⁻¹⁵, P_{mel-sim dS} = 1.9 × 10⁻⁷). However, we found no significant difference between the X chromosome and autosomes for the d_N/d_S ratio in any comparison (fig. 1E; Mann–Whitney U test, P_{mel-mel dN/dS} = 0.71, P_{sim-sech dN/dS} = 0.98, P_{mel-sim dN/dS} = 0.07). The absence of accelerated d_N/d_S on the X chromosome for this set of genes is surprising given a recent study showing evidence of this pattern genome-wide between D. melanogaster and D. sechellia (Avila et al. 2014), but is consistent with the mixed-results of prior studies that performed similar analyses in the melanogaster species group (Thornton et al. 2006; Begun et al. 2007; Connallon 2007; Hu et al. 2013). A greater mutation rate on the X chromosome than on autosomes might explain the higher d_N and d_S values observed in all comparisons; however, the higher recombination rate on the X chromosome reducing the effects of background selection (Vicoso and Charlesworth 2009b) or less effective purifying selection on the X chromosome resulting from differences in effective population size between the X chromosome and autosomes could also produce this pattern. The extent to which each of these and/or other evolutionary processes (Vicoso and Charlesworth 2006) or demographic factors (Pool and Nielsen 2008) contribute to the observed patterns of sequence variation will require further investigation.

Faster-X Expression Divergence

To determine whether the 4,851 genes analyzed showed greater expression divergence on the X chromosome than on autosomes, we used the nonparametric Spearman’s correlation coefficient (ρ) to compare levels of mRNA expression between strains or species in each comparison and then compared the degree of divergence for genes on the X chromosome and genes on autosomes. Spearman’s ρ has previously been used to contrast expression divergence between the X chromosome and autosomes (Brawand et al. 2011; Kayserili et al. 2012; Meisel et al. 2012) and compares the rank order of gene expression levels for orthologous genes between samples. We found that expression divergence (1 − ρ) was greater for genes on the X chromosome than for genes on the autosomes in all three comparisons (fig. 2A), although this difference was only statistically significant in mel-mel and sim-sech (one-sided permutation test, P_mel-mel = 0.01, P_sim-sech = 0.005, P_mel-sim = 0.1). The “faster-X” pattern of expression divergence in the sim-sech and mel-sim comparisons is consistent with previous comparisons between pairs of Drosophila species with greater evolutionary distances (Kayserili et al. 2012; Meisel et al. 2012), but the faster-X expression change observed in the mel-mel comparison conflicts with previous reports of intraspecific expression polymorphism in D. melanogaster that found no evidence for faster-X expression change (Hutter et al. 2008; Müller et al. 2011; Kayserili et al. 2012; Llopart 2012; Meisel et al. 2012). As described above for sequence variation, this difference likely results from the fact that the North American and African strains used for our mel-mel comparison are more divergent than those used in prior studies.

Open in a separate window

Fig. 2.

Faster-X divergence of gene expression. (A) Overall expression divergence (1 − ρ) is shown for X-linked genes and autosomal genes for each of the three comparisons. Error bars indicate the 2.5 and 97.5 percentiles of 1 − ρ from bootstrapping 10,000 times. Statistical significance for the difference between the X chromosome and autosomes was determined by permuting the chromosome assignment of individual genes relative to their expression levels 100,000 times. (B) The magnitude of expression divergence measured for each gene was determined and boxplots showing the distribution for the X chromosome and autosomes with the median (center line) as well as 25th and 75th percentiles indicated (box upper and lower bounds) and whiskers extending 1.5× the interquartile range. Statistical significance of the difference between the X chromosome and autosomes was determined with Mann–Whitney U tests. (C) The proportion of genes with significant expression differences in each comparison is shown and error bars represent the 2.5 and 97.5 percentiles from bootstrapping 10,000 times. Statistical significance of the difference between the X chromosome and autosomes was determined with Fisher’s exact tests (NS indicates not significant, *P ≤ 0.05, **P ≤ 0.001, and ***P ≤ 1 × 10⁻⁴).

Spearman’s ρ captures differences in the magnitude of expression divergence for individual genes as well as differences in the number of genes that are differentially expressed. To examine these two factors separately, we quantified the magnitude of expression divergence on a gene-by-gene basis and examined the number of genes with evidence of a significant change in expression based on binomial exact tests. We found that the magnitude of expression differences for individual genes was higher for genes on the X chromosome than autosomes in all three cases, but only significantly so in sim-sech and mel-sim (fig. 2B, Mann–Whitney U test, P_mel-mel = 0.059, P_sim-sech = 0.0015, P_mel-sim = 0.04). By contrast, the proportion of genes with a significant difference in expression was significantly higher on the X chromosome in all three comparisons (fig. 2C, Fisher’s exact test, P_mel-mel = 6.1 × 10⁻⁶, P_sim-sech = 8.6 × 10⁻⁸, P_mel-sim = 0.0017), indicating that the faster-X pattern of gene expression we observed is primarily driven by a greater number of genes with expression differences on the X chromosome.

Faster-X Cis-Regulatory Divergence

To test the hypothesis that faster-X divergence in gene expression is driven by faster-X divergence in cis-regulation (Kayserili et al. 2012; Meisel et al. 2012; Meisel and Connallon 2013), we examined relative allele-specific expression in F₁ hybrids, which provides a direct readout of relative cis-regulatory activity (Cowles et al. 2002; Wittkopp et al. 2004). We found that overall divergence (1 − ρ) in cis- regulatory activity was significantly greater for genes on the X chromosome than on autosomes in both sim-sech and mel-sim, but not in mel-mel (one-sided permutation test, P_mel-mel = 0.99, P_sim-sech = 0.006, P_mel-sim = 0.046; fig. 3A). This result is consistent with prior studies showing that cis-regulatory changes contribute more to expression divergence between species than within species (Wittkopp et al. 2008; Coolon et al. 2014). It also underscores that differentiation in the mel-mel comparison is distinct from interspecific divergence despite showing a pattern of faster-X sequence and expression evolution (figs. 1 and ​and2).2). The magnitude of cis-regulatory divergence between genes on the X chromosome and autosomes showed no significant difference in any comparison (Mann–Whitney U test, P_mel-mel = 0.97, P_sim-sec = 0.30, P_mel-sim = 0.48; fig. 3B), even after scaling by total regulatory divergence (% cis; Wittkopp et al. 2008), (Mann–Whitney U test, P_mel-mel = 0.97, P_sim-sec = 0.88, P_mel-sim = 0.96; fig. 3C); however, the number of genes with evidence of a significant difference in cis-regulatory activity was significantly higher on the X chromosome than on autosomes in sim-sech and mel-sim (Fisher’s exact test, P_mel-mel = 0.41, P_sim-sec = 0.002, P_mel-sim = 0.001; fig. 3D). Similar to our findings for total expression differences, these analyses indicate that the number of genes with cis-regulatory differences has a bigger impact on the overall cis-regulatory divergence we observed than the magnitude of cis-regulatory differences.

Open in a separate window

Fig. 3.

Faster-X divergence of cis-regulation. (A) Divergence in cis-regulatory activity (1 − ρ) is shown for genes on the X-chromosome and autosomes for all three comparisons and error bars represent the 2.5 and 97.5 percentiles from 10,000 bootstrap replicates. Statistical significance for the difference between the X chromosome and autosomes was determined with permutation tests (100,000 permutations). (B) Boxplots showing the magnitude of cis-regulatory divergence in F₁ hybrids (abs(log2(allele 1/allele 2))) for the X chromosome and autosomes for each comparison are shown with median (center line) as well as 25th and 75th percentiles indicated (box upper and lower bounds) and whiskers extending 1.5× the interquartile range. (C) The magnitude of cis-regulatory divergence scaled by total regulatory divergence ([|cis|/(|cis|+|trans|)] × 100, %cis; Wittkopp et al. 2008) was determined and boxplots with the median (center line) as well as 25th and 75th percentiles indicated (box upper and lower bounds) and whiskers extending 1.5× the interquartile range. (B, C) Statistical significance of differences between the X chromosome and autosomes was determined using Mann–Whitney U tests. (D) The proportion of genes with significant cis-regulatory differences in F₁ hybrids for each comparison is shown and error bars represent the 2.5 and 97.5 percentiles from bootstrapping 10,000 times. Statistical significance of the difference between the X chromosome and autosomes was determined with Fisher’s exact tests (NS indicates not significant, *P ≤ 0.05, **P ≤ 0.001, and ***P ≤ 1 × 10⁻⁴).

The data from the sim-sech and mel-sim comparisons are consistent with the hypothesis that cis-regulatory divergence contributes to the faster-X pattern of gene expression divergence between species. However, these cis-regulatory differences appear to be insufficient to fully explain the faster-X pattern of expression divergence: The magnitude of overall expression differences (compare fig. 2A with ​with33A) as well as the proportion of genes with significant expression differences (compare fig. 2C with ​with33D) are both smaller for cis-regulation than total expression differences.

Faster-X Trans-Regulatory Divergence

Changes in gene expression that are not explained by cis-regulatory differences can be attributed to trans-regulatory differences (Wittkopp et al. 2004), suggesting that trans-regulatory divergence might also contribute substantially to the faster-X pattern of expression divergence. Consistent with this hypothesis, overall trans-regulatory divergence (1 − ρ) was significantly higher on the X chromosome than on autosomes in mel-mel (permutation test, P_mel-mel = 8 × 10⁻⁵, P_sim-sech = 0.43, P_mel-sim = 0.78; fig. 4A). The magnitude of trans-regulatory divergence was also greater for genes on the X chromosome than on autosomes for all three comparisons (Mann–Whitney U test, P_mel-mel = 0.02, P_sim-sech = 0.0025, P_mel-sim = 0.0017; fig. 4B), but remained significant only for mel-mel and mel-sim after scaling for total regulatory divergence (% trans, Mann–Whitney U test, P_mel-mel = 0.024, P_sim-sech = 0.12, P_mel-sim = 0.034; fig. 4C). The proportion of genes with evidence of significant trans-regulatory divergence was also significantly greater for X-linked than autosomal genes in all three comparisons (Fisher’s exact test, P_mel-mel = 1.2 × 10⁻⁶, P_sim-sech = 1.9 × 10⁻⁵, P_mel-sim = 1.5 × 10⁻⁶; fig. 4D), with more genes showing evidence of trans-regulatory divergence than cis-regulatory divergence in each comparison (compare fig. 4D with ​with33D). These data indicate that faster-X trans-regulatory divergence is responsible for the faster-X pattern of expression divergence in mel-mel and contributes to this pattern in sim-sech and mel-sim. It remains to be seen whether trans-regulatory divergence also contributes significantly to the faster-X patterns of gene expression divergence observed among species with greater divergence times (Brawand et al. 2011; Kayserili et al. 2012; Meisel et al. 2012).

Open in a separate window

Fig. 4.

Faster-X divergence of trans-regulation. (A) Divergence in trans-regulatory activity (1 − ρ) is shown for genes on the X-chromosome and autosomes for all three comparisons and error bars represent the 2.5 and 97.5 percentiles from 10,000 bootstrap replicates. Statistical significance for the difference between the X chromosome and autosomes was determined with permutation tests (100,000 permutations). (B) Boxplots showing the magnitude of trans-regulatory divergence (abs(log2(species 1/species 2)) − abs(log2(allele 1/allele 2))) for the X chromosome and autosomes for each comparison are shown with the median (center line) as well as 25th and 75th percentiles indicated (box upper and lower bounds) and whiskers extending 1.5× the interquartile range. (C) The magnitude of trans-regulatory divergence scaled by total regulatory divergence ([|trans|/(|cis|+|trans|)] × 100, %trans) was determined and boxplots with the median (center line) as well as 25th and 75th percentiles indicated (box upper and lower bounds) and whiskers extending 1.5× the interquartile range. (B, C) Statistical significance of differences between X and autosome was determined using Mann–Whitney U tests. (D) The proportion of genes with significant trans-regulatory differences for each comparison is shown and error bars represent the 2.5 and 97.5 percentiles from bootstrapping 10,000 times. Statistical significance of the difference between the X chromosome and autosomes was determined with Fisher’s exact tests (NS indicates not significant, *P ≤ 0.05, **P ≤ 0.001, and ***P ≤ 1 × 10⁻⁴).

The large contribution of trans-regulatory divergence to differences seen between the North American and African strains of D. melanogaster is consistent with the excess of trans-regulatory variation previously observed segregating within species of Drosophila (Wittkopp et al. 2008; Coolon et al. 2014) and Saccharomyces (Emerson et al. 2010). However, the large contribution of trans-acting variation to the faster-X pattern of expression divergence is surprising because trans-regulatory factors controlling expression of X-linked genes are generally assumed to be distributed randomly across the genome. This is in stark contrast to cis-regulatory sequences, which are, by definition, always located on the X chromosome for X-linked genes. If this assumption about the genomic distribution of trans-acting factors is wrong, and trans-acting factors encoded by genes on the X chromosome preferentially regulate expression of X-linked genes, trans-regulatory divergence could indeed explain the faster-X pattern of expression divergence.

Preferential Regulation of X-Linked Genes by X-Linked Transcription Factors

Although many types of genes can have trans-acting effects on gene expression, genes encoding transcription factors (TFs) are commonly assumed to have the most direct effects (Yvert et al. 2003). Although X-linked TFs have previously been shown to disproportionately contribute to expression divergence genome-wide between species, we wanted to determine whether TFs encoded by genes on the X chromosome are likely to disproportionally affect expression of X-linked genes. To do this, we used a recently constructed gene regulatory network for D. melanogaster that contains 310,634 predicted regulatory interactions among 617 TFs and 12,286 target genes (Marbach et al. 2012). This network was constructed by integrating TF binding motif conservation, ChIP-chip and ChIP-seq for TFs and chromatin marks, and microarray and RNA-seq measures of gene expression (Marbach et al. 2012). We identified 4,379 genes common to our expression analyses and the published regulatory network and using the edges in this network; we compared the proportion of X-linked target genes for X-linked and autosomal TFs (fig. 5A). For the 213 TFs included in this set of 4,379 genes, 46 of which are X-linked (21.6%), we found that TFs encoded by genes on the X chromosome had a larger proportion of their target genes on the X chromosome than TFs encoded by genes on autosomes (Mann–Whitney U test, P₄₃₇₉ < 1.0 × 10⁻¹⁵). This was again true when we considered all 617 TFs in the network, 102 (16.5%) of which were X-linked (Mann–Whitney U test, P_all < 1.0 × 10⁻¹⁵), showing that this pattern is not limited to the set of genes that were analyzed. We also compared the proportion of TFs that are X-linked for target genes on the X chromosome and autosomes (fig. 5B). We found that target genes on the X chromosome were also regulated by a higher proportion of X-linked TFs than target genes on autosomes for both the 4,379 genes examined here (Mann–Whitney U test, P₄₃₇₉ = 1.4 × 10⁻¹¹) and the full set of genes included in the network (Mann–Whitney U test, P_all < 1.0 × 10⁻¹⁵). Neither pattern was observed for any of the individual autosomes (supplementary figs. S1 and S2, Supplementary Material online), indicating that this finding does not result from a genome-wide pattern of preferential interactions between regulators and genes on the same chromosome. This finding that X-linked TFs seem to preferentially regulate X-linked genes suggests that accelerated sequence divergence of X-linked TFs (affecting their cis-regulatory sequences and/or coding sequences) might have a greater impact on expression of X-linked genes than autosomal genes.

Open in a separate window

Fig. 5.

X-linked TFs preferentially regulate X-linked target genes. (A) For TFs on either the X chromosome or autosomes, the proportion of downstream target genes on the X chromosome was determined using a Drosophila regulatory network (Marbach et al. 2012). Boxplots show the distribution of these values for the X-chromosome and the autosomes with the median (center line) as well as 25th and 75th percentiles indicated (box upper and lower bounds) and whiskers extending 1.5× the interquartile range. Mann–Whitney U tests were used to compare the X chromosome and autosomes. (B) For target genes on the X chromosome and autosomes, the proportion of upstream TFs encoded by genes on the X chromosome. Boxplots show the distribution of these values for the X-chromosome and the autosomes with the median (center line) as well as 25th and 75th percentiles indicated (box upper and lower bounds) and whiskers extending 1.5× the interquartile range. Mann–Whitney U tests were performed to compare the X chromosome and autosomes (NS indicates not significant, *P ≤ 0.05, **P ≤ 0.001, and ***P ≤ 1 × 10⁻⁴).

To further explore this possibility, we tested whether the 46 TFs on the X chromosome showed greater sequence divergence than the 167 TFs on the autosomes. We found evidence of a faster-X pattern of sequence evolution for both coding (Mann–Whitney U test, P_mel-mel = 2.3 × 10⁻⁸, P_sim-sech = 0.0007, P_mel-sim = 0.0016; supplementary fig. S3A, Supplementary Material online) and noncoding (Mann–Whitney U test, P_mel-mel = 8.04 × 10⁻⁷, P_sim-sech = 0.00015, P_mel-sim = 1.4 × 10⁻⁵; supplementary fig. S3B, Supplementary Material online) sequences in all three comparisons, but a statistically significant faster-X pattern for d_N and d_S only in mel-mel and sim-sech (Mann–Whitney U test, P_{mel-mel dN} = 0.04, P_{sim-sech dN} = 0.03, P_{mel-sim dN} = 0.12, P_{mel-mel dS} = 2.9 × 10⁻⁷, P_{sim-sech dS} = 0.0003, P_{mel-sim dS} = 0.44; supplementary fig. S3C and D, Supplementary Material online). Although not statistically significant, median d_N and d_S were larger for X-linked than autosomal TFs even in the mel-sim comparison. The absence of a statistically significant faster-X pattern in mel-sim for d_N or d_S despite a significant faster-X pattern for coding sequence divergence presumably results from reduced power of the statistical tests when coding sequences were divided into these two categories. The d_N/d_S ratio was not significantly elevated for X-linked TFs compared with autosomal TFs in any comparison (Mann–Whitney U test, P_mel-mel = 0.99, P_sim-sech = 0.27, P_mel-sim = 0.08; supplementary fig. S3E, Supplementary Material online), consistent with the absence of this pattern in the full set of genes. These data suggest greater amino acid sequence divergence of X-linked TFs compared with autosomal TFs, but provide no evidence that this excess is due to positive selection.

We also tested whether X-linked TFs showed greater divergence in total expression, cis-regulation, or trans-regulation than autosomal TFs. Comparing overall differences using 1 − Spearman’s ρ, magnitude of difference, and proportion of genes with differences, we found that none of these measures showed a significant difference between the X chromosome and autosomes in any comparison (P > 0.05 in all cases). That said, there is a finite number of TFs in the genome and it is possible that these statistical tests lack sufficient power to detect small differences. To test this possibility, we determined the magnitude of effect that could be detected with 80% power for each metric of expression or regulatory divergence. We found that, in all cases, the difference between the X chromosome and autosomes would need to have been 2–4 times greater than what we observed to achieve 80% detection power. This suggests that if there are differences between the X chromosome and autosomes for the regulation and/or expression of TFs, they are subtle. Given the proposed role of X-linked TFs in generating the faster-X expression pattern of expression divergence, elevated sequence divergence for X-linked TFs in the absence of elevated expression divergence could suggest that TFs on the X chromosome may have altered protein function.

Another prediction that follows from our hypothesis that greater divergence of X-linked TFs contributes to the faster-X pattern of expression divergence is that genes with a greater proportion of trans-acting regulators located on the X chromosome might tend to show greater differences in total expression and trans-regulatory divergence. Consistent with this prediction, we observed weak but statistically significant correlations between the proportion of trans-acting regulators of a gene that are X-linked and the magnitude of expression and trans-regulatory divergence in all but one comparison (trans-regulatory divergence in mel-sim), and even this case was marginally significant (supplementary fig. S4A and C, Supplementary Material online; permutation test, total expression: P_mel-mel < 1 × 10⁻⁴, P_sim-sech = 2 × 10⁻⁴, P_mel-sim = 0.049; trans-regulation: P_mel-mel < 1 × 10⁻⁴, P_sim-sech < 1 × 10⁻⁴, P_mel-sim = 0.07). The proportion of X-linked regulators for a gene is not expected to be related to its levels of cis-regulatory divergence, and we found no evidence of a significant correlation in any of the comparisons (supplementary fig. S4B, Supplementary Material online, permutation test, P_mel-mel = 0.73, P_sim-sech = 0.23, P_mel-sim = 0.81).

Quantification of Percent Sequence Divergence

To determine percent sequence divergence in each comparison (mel-mel, sim-sec, and mel-sim), we created reverse chain files to LiftOver coordinates from D. melanogaster dm3 space to each of the other strain or species genomic space (zhr, z30, Tsimbazaza, droSec1; Coolon et al. 2014) using the chainSwap utility from the UCSC Genome Browser (Kent et al. 2002). We downloaded the D. melanogaster genomic annotations for coding (all transcribed sites) and noncoding (all intergenic sequences to the next gene and intronic sequences) regions for each gene from FlyBase (St Pierre et al. 2014). Using the chain files, we converted the dm3 genomic coordinates for each coding and noncoding region for each gene used for quantification in this study into their respective strain- or species-specific genomic coordinates. Using these coordinates, sequences for each region were extracted from each strain- or species-specific genome (provided in the supplementary material for Coolon et al. 2014). These sequences were aligned in pairs using Fast Statistical Alignment (FSA version 1.15.9) (Bradley et al. 2009) and the number of divergent sites per gene (coding and noncoding) was determined using a custom Perl script (pairwise_aln_FSA.pl).

Calculation of d_N, d_S, and d_N/d_S Sequence Metrics

We extracted the shortest transcript for all genes from the D. melanogaster dm3 genome build on FlyBase (St Pierre et al. 2014), and using the chain files described above converted the dm3 genomic coordinates for each transcript into their respective strain- or species-specific genomic coordinates using the liftOver tool from the UCSC Genome Browser (Kent et al. 2002). Genes with a valid start codon and an in-frame stop codon were retained for analyses. Coding sequences were aligned in pairs for each comparison (mel-mel, sim-sech, and mel-sim) with FSA (Bradley et al. 2009). We then used PAML version 4.7a (Yang 2007) to generate estimates of d_N, d_S, and d_N/d_S for each gene using the yn00 package and default settings. Ratios of d_N/d_S estimated using maximum-likelihood methods have great uncertainty when d_S is very low so all genes with no observed synonymous differences were removed. After filtering, 4,351 genes were used in mel-mel (X: 926, A: 3,425), 3,328 in sim-sech (X: 529, A: 2,799), and 3,699 in the mel-sim (X: 622, A: 3,077) comparison.

Mining the D. melanogaster Regulatory Network

As stated in the main text, the supervised gene regulatory network was obtained from the online supplementary material from Marbach et al. (2012). This network consisted of 310,634 predicted regulatory interactions among 617 TFs (regulators) and their 12,286 target genes. The complete set of genes and their chromosomal locations was downloaded from FlyBase (St Pierre et al. 2014). We limited the predicted regulatory interactions to those whose TF and target gene were both present in the gene expression data (4,851 genes from Coolon et al. 2014). We then determined the proportion of X-linked targets for each regulator on each chromosome, as well as the proportion of X-linked regulators for each target on each chromosome, using custom Perl scripts (chromosomal_regulators_subset.pl and chromosomal_targets_subset.pl). This process was repeated with different focal chromosomes. To make sure that the features we observed were not due to the specific gene set used in expression quantification, we also determined these same proportions for the complete supervised gene regulatory network using custom Perl scripts (chromosomal_regulators.pl and chromosomal_targets.pl).

The authors thank Brian Metzger, Richard Lusk, Fabien Duveau, and Alisha John, as well as the editor and three anonymous reviewers, for helpful comments on the manuscript. Funding for this work was provided by the National Institute of Health (5F32GM089009-02; to J.D.C. and 5R01GM095296 to B.R.G.), the National Science Foundation (NSF 0903629; to K.R.S. and MCB-1021398 to P.J.W.), and the Alfred P. Sloan Research Foundation (fellowship to P.J.W.). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Institute of Health, National Science Foundation, or Sloan Foundation.