The Genetic Basis of Rapidly Evolving Male Genital Morphology in Drosophila

Abstract

The external genitalia are some of the most rapidly evolving morphological structures in insects. The posterior lobe of the male genital arch shows striking differences in both size and shape among closely related species of the Drosophila melanogaster species subgroup. Here, we dissect the genetic basis of posterior lobe morphology between D. mauritiana and D. sechellia, two island endemic species that last shared a common ancestor ∼300,000 years ago. We test a large collection of genome-wide homozygous D. mauritiana genetic introgressions, which collectively cover ∼50% of the genome, for their morphological effects when placed in a D. sechellia genetic background. We find several introgressions that have large effects on posterior lobe morphology and that posterior lobe size and posterior lobe shape can be separated genetically for some of the loci that specify morphology. Using next generation sequencing technology, we perform whole transcriptome gene expression analyses of the larval genital imaginal disc of D. mauritiana, D. sechellia, and two D. mauritiana–D. sechellia hybrid introgression genotypes that each have large effects on either posterior lobe size or posterior lobe shape. Many of the genes we identify as differentially expressed are expressed at levels similar to D. mauritiana in one introgression hybrid, but are expressed at levels similar to D. sechellia in the other introgression hybrid. However, we also find that both introgression hybrids express some of the same genes at levels similar to D. mauritiana, and notably, that both introgression hybrids possess genes in the insulin receptor signaling pathway, which are expressed at D. mauritiana expression levels. These results suggest the possibility that the insulin signaling pathway might integrate size and shape genetic inputs to establish differences in overall posterior lobe morphology between D. mauritiana and D. sechellia.

MORPHOLOGICAL evolution is an important mechanism for generating biodiversity. One of the goals of evolutionary developmental biology is to identify genes that specify morphology and understand how genetic variation at those loci directs development to give rise to intra- and interspecific differences in morphology. Two different approaches have been primarily used to identify genes important for specifying morphological differences between species. One approach has been to identify genes that are important for development of morphological structures in one species and study their developmental roles in closely related species. The second approach has been to genetically map regions of the genome that have large morphological effects between species that produce viable, fertile hybrid offspring to identify the causative loci. Both approaches have been successful in identifying genes, genetic pathways, and regulatory differences underlying species-specific morphologies in a variety of taxa including plants (Luo et al. 1996; Doebley et al. 1997; Da et al. 1999; Galego and Almeida 2002; Hubbard et al. 2002; Corley et al. 2005; Clark et al. 2006; Hay and Tsiantis 2006), cnidarians (Khalturin et al. 2008), arthropods (Stern 1998; Sucena and Stern 2000; Beldade et al. 2002; Ronshaugen et al. 2002; Wittkopp et al. 2003, 2009; Reed and Serfas 2004; Gompel et al. 2005; Prud’homme et al. 2006; Barmina and Kopp 2007; McGregor et al. 2007; Moczek and Rose 2009; Hrycaj et al. 2010; Loehlin et al. 2010; Shirataki et al. 2010; Wasik et al. 2010; Werner et al. 2010; Wasik and Moczek 2011), echinoderms (Hinman and Davidson 2007), fish (Fraser et al. 2009), birds (Abzhanov et al. 2004, 2006; Mallarino et al. 2011), mammals (Cretekos et al. 2008), as well as morphological differences between more distantly related groups of organisms such as agnathans and gnathostomes (Meulemans and Bronner-Fraser 2002; McCauley and Bronner-Fraser 2006).

One of the most abundant and striking examples of morphological diversity in insects is variation in the size and shape of their external genital structures (Eberhard 1985). Compared to other morphological structures that often appear similar among closely related species, the external genitalia are one of the most rapidly evolving morphological characters. In particular, various modifications of the male genitalia have evolved among phylogenetically widespread species of Drosophila (Eberhard and Ramirez 2004; Jagadeeshan and Singh 2006). Some species that belong to the D. melanogaster species subgroup have evolved a small cuticular projection from the lateral portion of the epandrium, a horseshoe-shaped cuticular structure that surrounds the external genitalia and analia. This cuticular projection, commonly known as the posterior lobe of the genital arch, inserts between the eighth and ninth abdominal tergites of the female during copulation (Robertson 1988), appears important for mounting, genital coupling, and copulation duration (Coyne 1993; Price et al. 2001; Jagadeeshan and Singh 2006), and might also effect sperm transfer (Coyne 1993; Price et al. 2001). The size and shape of the posterior lobe have changed dramatically among D. melanogaster and its sibling species and serves as the most reliable morphological character that distinguishes males of these species from one another (Hsu 1949; Coyne 1983; Ashburner et al. 2005). Because this structure is involved in mating, it has been hypothesized that the differences in morphology among these species have been driven by sexual selection (Eberhard 1985), although the exact mechanism of sexual selection (e.g., cryptic female choice vs. sexually antagonistic coevolution) that might drive morphological divergence in this case remains unclear.

Several studies have mapped the location of genes involved in specifying morphological differences in the posterior lobe among D. mauritiana, D. sechellia, and D. simulans, three species that last shared a common ancestor 300,000–900,000 years ago (Kliman et al. 2000; Tamura et al. 2004). These three species will mate with each other to produce sterile F₁ hybrid males and fertile F₁ hybrid females, making it possible to generate backcross and F₂-like genotypes between any pair of species. In the D. mauritiana–D. simulans species pair, genes specifying posterior lobe morphology map to all major chromosomes (Coyne 1983; Liu et al. 1996), and quantitative trait locus (QTL) mapping experiments identified a minimum of 20 loci underlying the morphological difference between these two species (Laurie et al. 1997; Zeng et al. 2000). In the D. sechellia–D. simulans species pair, mapping experiments also revealed loci on all major chromosomes (Coyne and Kreitman 1986; MacDonald and Goldstein 1999), and QTL mapping revealed a minimum of 13 loci that have effects on posterior lobe morphology (MacDonald and Goldstein 1999). Many of the QTL regions identified in both D. mauritiana–D. simulans and D. sechellia–D. simulans species pairs reside in similar genomic regions, which suggests that some of the same genes might specify differences in morphology among these species. These QTL mapping experiments also revealed that the genetic basis of the posterior lobe morphological differences among these species appears mostly additive, and most of the large effect QTL identified act in the direction of parental trait values in their effects on posterior lobe morphology. Lastly, mapping experiments in the D. mauritiana–D. sechellia species pair showed that both the second and third chromosomes have large effects on morphology, and a small effect appeared to map to the X chromosome (Coyne et al. 1991).

In each of these studies, the loci involved in specifying posterior lobe morphological differences were mapped to large genomic regions, which on average contain hundreds to thousands of genes. In an effort to reduce the size of these genomic regions, we fine map the genome to identify regions with large morphological effects between D. mauritiana and D. sechellia, with the ultimate goal of identifying genes or genetic pathways responsible for differences in posterior lobe morphology between these two species. Our study takes advantage of a large collection of genetic introgression lines that were generated by backcrossing individual D. mauritiana stocks marked with single P-element insertions that bear the visible eye color marker white⁺ (w⁺) into a D. sechellia white⁻ (w) genetic background for 15 generations (Masly and Presgraves 2007). Each of these introgression lines carries a small homozygous region of the D. mauritiana genome linked to the P-element marker in a mostly D. sechellia genetic background. We measure posterior lobe morphology of 45 viable and fertile D. mauritiana introgressions and identify several that have large effects on posterior lobe size and shape. Interestingly, we find some introgressions that have a large effect on either posterior lobe size or posterior lobe shape, which suggests that these two morphological phenotypes might be partly under separate genetic control. Using next generation sequencing technology, we perform whole transcriptome expression analyses of the male genital imaginal disc of the D. mauritiana and D. sechellia parental strains, and two D. mauritiana hybrid introgression lines that have large phenotypic effects on either posterior lobe size or posterior lobe shape. The results of these experiments show that genes in the insulin signaling pathway are differentially expressed between the two pure species, and that each of the introgression hybrids express these genes at expression levels similar to those of D. mauritiana.

Materials and Methods

Drosophila stocks

All stocks were maintained on standard cornmeal, yeast, dextrose medium at 25° with a 12 hr light:dark cycle. The D. mauritiana–D. sechellia hybrid introgression lines were constructed previously using D. sechellia w and a collection of D. mauritiana P-element insertion stocks that each bear a w⁺ visible marker (Masly and Presgraves 2007). Each of these introgression lines carries a small homozygous region of the D. mauritiana genome linked to a P-element in an otherwise D. sechellia w genetic background. The effect of each individual introgression on morphology was compared to the parental pure species, D. sechellia w, and two randomly chosen D. mauritiana P-element insertion lines (Q1 and Wanda1; see True et al. 1996 for a description of the D. mauritiana P-element insertion lines). All morphological measurements from D. mauritiana Q1 and Wanda1 were not significantly different; we thus pooled measurements from both lines in our morphometric analyses and refer to them as “D. mauritiana” in the text. F₁ hybrid males were obtained from crosses between D. sechellia w females and D. mauritiana P-insert Q1 males. Repeated attempts to generate the reciprocal F₁ hybrid genotype using these two lines failed. Microarray experiments were performed using two D. melanogaster strains, Berlin and Canton-S.

Morphological data acquisition

Adult males from the D. mauritiana–D. sechellia introgression lines, D. mauritiana, and D. sechellia w were sampled from four replicate vials per genotype in approximately equal numbers, anesthetized, and dissected. Left and right posterior lobes of the genital arch, and left and right forelegs from each individual were mounted in polyvinyl alcohol mounting medium (Bioquip, Rancho Dominguez, CA) under coverslips on glass slides and allowed to set. Images were captured using a compound microscope equipped with a Zeiss AxioCam digital imaging system; tibias were imaged at a resolution of ×100 magnification and posterior lobes were imaged at a resolution of ×200 magnification.

Morphometric analyses

Morphometric analyses performed here follow those performed in previous studies of genital morphology among species of the D. simulans clade (Coyne 1983; Coyne and Kreitman 1986; Coyne et al. 1991; Laurie et al. 1997; Liu et al. 1996; MacDonald and Goldstein 1999; Zeng et al. 2000). Measurements on all digitized posterior lobes and tibias were performed using ImageJ software (Abramoff et al. 2004; Rasband 1997–2009) calibrated with a stage micrometer. The outline of each posterior lobe was traced and lobe area was delimited by an artificial baseline that lies roughly in line with the edge of the lateral plate. Posterior lobe areas were corrected for differences in overall body size by dividing posterior lobe area by tibia length. Tibia length was measured as the distance between the foreleg tibiofemoral joint and the tibiotarsal joint. Posterior lobe area comparisons were performed on both corrected and uncorrected lobe areas.

Because the posterior lobe possesses no reliable morphological landmarks, posterior lobe shape was represented using an elliptical Fourier analysis. The posterior lobe outline was translated into series of 300–600 (x,y) coordinates that represent lobe shape. Elliptical Fourier coefficients were calculated using the algorithm developed by Kuhl and Giardina (1982). Prior to calculating the coefficients, each outline was configured to standardize for location, orientation, and handedness. Outline coordinates were also divided by the square root of the posterior lobe area to normalize each outline to have an area of one, and the origin of the coordinates system was placed at the centroid of the outline. For each posterior lobe, we obtained 80 Fourier coefficients (20 harmonics of the Fourier series), which when used to recreate the lobe outline, did so with high precision (not shown).

To reduce the number of variables that explain variation in shape, Fourier coefficients for each posterior lobe were analyzed by principal components analysis (PCA). Among the genotypes studied here, principal component 1 (PC1) and principal component 2 (PC2) combined describe >60% of the variation in Fourier coefficients. The first 14 principal components (PC1–PC14) describe ∼97% of the variation in Fourier coefficients. Because the first three coefficients of the first harmonic represent the normalizing coefficients, and are identical among all outlines, they were excluded from PCA as they provide no information in describing differences in posterior lobe shape. The remaining 77 coefficients were therefore used in PCA. We performed PCA using both singular value decomposition and covariance matrix decomposition methods. We present the results of our statistical tests using the principal components scores obtained by singular value decomposition for simplicity, although the results obtained using scores calculated with covariance matrix decomposition remain similar.

Our approach to analyzing posterior lobe shape is briefly summarized as follows: Posterior lobe outline coordinates were adjusted to normalize posterior lobe area. Those coordinates were used to calculate elliptical Fourier coefficients. Fourier coefficients from all lobes of the subset of genotypes being compared were used in PCA to obtain principal components scores for each posterior lobe in that group, and the principal components scores among the genotypes in that group were compared to test for differences in posterior lobe shape.

D. melanogaster microarray assays

We used two-color glass slide microarrays spotted with 16,420 70-mer oligonucleotide probes that include all known and predicted genes in the D. melanogaster genome release 4.1 to assay gene expression in D. melanogaster male and female genital imaginal discs. Details of oligonucleotide probes and microarray construction can be found in Goldman and Arbeitman (2007) and Lebo et al. (2009). Animals were reared on medium supplemented with 0.05% bromophenol blue to allow visualization of gut clearing after the high-titer ecdysone pulse during late third larval instar stage of development (Andres and Thummel 1994). This allowed us to harvest genital discs and assay gene expression during a narrow window of development (∼2–5 hr). Larvae were dissected post gut clearing in ice-cold RNAse-free 1× PBS and genital imaginal discs transferred immediately to TRIzol reagent (Invitrogen) and homogenized. For each sample, total RNA was extracted from ∼50–100 genital discs using a standard protocol, and poly(A)⁺ transcripts were amplified to obtain the requisite amount of starting material for the microarray hybridizations using the Amino-Allyl Message-Amp II aRNA kit (Ambion).

Microarray experiments were performed with five independent biological replicates each for D. melanogaster Berlin and Canton-S using a dye-swap design (i.e., male samples labeled with Cy5 in three replicates were labeled with Cy3 in the other two replicates in that experimental set). Ten micrograms of amplified mRNA from one male and 10 μg from one female were labeled with either Cy5 or Cy3 fluorescent dye (Amersham Bioscience), respectively, and hybridized to the arrays overnight at 42°. Microarrays were scanned using a GenePix 4100A scanner and processed using GenePix Pro 5.0 software from Axon Instruments (Molecular Diagnostics, Sunnyvale, CA). Poor quality spots were removed by visual inspection of the microarray images prior to analysis. Spots were also only considered for further analysis if at least one channel (Cy3 or Cy5) had >75% of the pixels with intensity values at least one standard deviation above background intensity levels. All normalization and count ratio analyses were performed using the LIMMA package of BioConductor in R (Smyth and Speed 2003; Gentleman et al. 2004; Smyth 2005). Global loess normalization was performed for all microarrays, and significant genes were identified using both a t-test and a FDR-corrected q-value (Storey 2003).

Transcriptome preparation and sequencing

We prepared three independent biological replicates of sequencing libraries for each of the following strains: D. mauritiana P-insertion line Q1, D. sechellia w, hybrid introgression line 3Q1(A), and hybrid introgression line Q1(A). Male genital discs were obtained as described above, and total RNA was extracted using RNAqueous-Micro kit (Ambion). Poly(A)⁺ transcripts were isolated subsequently using a MicroPoly(A)Purist kit (Ambion). To facilitate additional quality control of reads across our samples, at this stage of library construction we spiked-in small amounts of exogenous RNA from an ArrayControl kit (Ambion) into each sample of poly(A)⁺ RNA. Spike-in control sequences selected had similar lengths (∼1 kb), had no significant alignment to the D. melanogaster transcriptome using 25-bp alignments, and had no significant alignment among the other spike-in sequences chosen. Five spike-in controls (Ambion ArrayControl RNA spikes 3–7) were added to each of our 12 samples in decreasing amounts following a log₂ scale. The combination of spike-in controls represented, on average, 0.08% of the total RNA pool for each sample.

After adding the spike-in control RNAs, samples were fragmented to ∼250 bp by chemical fragmentation. First-strand cDNA was synthesized using SuperScript II reverse transcriptase (Invitrogen) and a combination of random hexamer and oligo (dT) primers; second-strand cDNA was synthesized using DNA polymerase I in combination with ribonuclease H. Double-stranded cDNA templates were blunt ended using End-It Repair kit (Epicentre), and A-overhangs were added at both ends with Klenow fragment (3′→5′ exo-minus). Illumina GAII sequencing adaptors were then ligated to both ends of the cDNA templates using Fast-Link DNA ligation kit (Epicentre). We then enriched for cDNA templates by performing multiplex incorporating PCR reactions (18 cycles) and isolated 250- to 550-bp fragments by gel purification. Cycle number was kept low during enrichment to avoid introducing biases that might skew the relative abundance of transcripts in each sample (and thus effect the ability to accurately estimate differential gene expression). Analysis of our spike-in control RNAs show that the reactions remained within the exponential range during amplification, and that the relative amounts of each of the five spike-in controls remained proportional to their starting amounts (see Supporting Information, File S1). During PCR, unique index sequences (Illumina) were also incorporated into each biological sample to allow identification of reads from each sample when multiple samples were sequenced on a single lane of the flow cell. Paired-end sequencing was carried out by loading the samples onto four lanes (three samples per lane) of a flow cell and run on an Illumina Genome Analyzer IIx sequencer using 72 cycles per end of each paired-end read. Biological replicates of each genotype were loaded onto separate lanes, which allowed us to detect potential lane effects on sequencing coverage. Images were processed using Illumina’s GenomeStudio software.

Transcriptome analysis

Sequence reads from all samples were mapped using PerM (Chen et al. 2009) to the D. melanogaster genome sequence (FlyBase version FB2010_07) allowing 12 mismatches per 72-bp read. Because we were unable to use species-specific reference genomes to map transcriptome reads, each end of the paired-end read was mapped separately. To maximize the number of mapped reads for detecting differential expression and differences in transcript splice junctions, we also used MapSplice (Wang et al. 2010), which allows reads to map across exon junctions. For reads that span exon junctions, we used more stringent criteria for the number of mismatches (≤4) due to the higher degree of sequence ambiguity compared to read mapping within exons. Uniquely mappable exon and junction reads were included in the downstream analysis.

Gene annotations in D. melanogaster (FlyBase version FB2010_07) were used to assign gene identity for transcriptome quantification. Gene expression levels were quantified using a two-parameter generalized Poisson (GP) model (Srivastava and Chen 2010). Compared to the traditional Poisson model, the GP model introduces an additional parameter, λ, to correct for potential biases in library preparation and the sequencing process. The GP model is more versatile than alternative models (e.g., Poisson and negative binomial) because of its ability to take into account both overdispersion and underdispersion of gene expression data. This allows for more accurate expression estimates, facilitates normalization across samples, and improves identification of differentially expressed transcripts in mRNA-seq experiments. Differentially expressed genes were identified using the log likelihood approach with a FDR cutoff of 0.05. FDR was controlled using the Benjamini–Hochberg method. A more stringent FDR cutoff of 0.01 did not substantially improve the number of genes identified as differentially expressed (File S1). Uncorrected P-values for comparisons between pairs of genotypes that correspond to a FDR cutoff of 0.05 were P < 0.003–0.02. We also performed our analyses using the traditional Poisson model to detect and measure differential gene expression, and all genes identified as differentially expressed using the GP model were also identified as differentially expressed using the Poisson model.

Statistics

The distributions of principal components scores for posterior lobe shape are nonnormal. We therefore tested for differences in posterior lobe shape using a nonparametric test for small sample, multivariate data (Bathke et al. 2008). Briefly, each principal component score (PC1–PC14) was considered as a variable. Ranks were computed among all observations (individual posterior lobe scores) for each variable. The ranks were then used to perform tests using asymptotic methods to calculate nonparametric test statistics. We tested the difference in the distributions of the introgression lines using a finite approximation of the Bartlett–Nanda–Pillai-type test statistic, which is approximately F distributed. This test statistic was shown to be a robust and conservative test of significance when used to test differences between a small number of samples with each having a moderate number of observations (Bathke et al. 2008).

Comparisons of tibia lengths and posterior lobe areas were tested using Welch’s t-test, which makes no assumptions about equality of variance between two samples (Welch 1947). Means are reported ±1 standard error. Programs to perform basic statistics, PCA, nonparametric rank tests, and calculate elliptical Fourier coefficients were implemented using the statistical program R (R Development Core Team 2010).

Results and Discussion

Pure species and D. mauritiana–D. sechellia introgression hybrid morphological data

D. mauritiana and D. sechellia show marked differences in posterior lobe morphology: D. mauritiana has a “finger-shaped” lobe, D. sechellia has a “goose-head–shaped” lobe, and D. sechellia posterior lobe area is roughly three times that of D. mauritiana (Figure 1). We dissected posterior lobes from at least 30 individuals from D. mauritiana, D. sechellia w, and each of 45 D. mauritiana–D. sechellia hybrid introgression lines, which carry small (∼1.5 Mb on average) regions of the D. mauritiana genome in a D. sechellia w genetic background (Masly and Presgraves 2007). The collection of D. mauritiana introgressions are distributed throughout the genome: 12 introgressions reside on the X chromosome, 8 on chromosome arm 2L, 9 on chromosome arm 2R, 11 on chromosome arm 3L, 5 on chromosome arm 3R, and 0 on chromosome 4 (Table S1).

Figure 1 .

Posterior lobe morphology. Posterior lobes from D. sechellia, D. mauritiana–D. sechellia F₁ hybrid, and D. mauritiana are shown. F₁ hybrid shown was obtained from crosses between D. sechellia w females and D. mauritiana males. Posterior lobe outlines were delimited by an artificial baseline to produce closed contours that resemble posterior lobe images shown here. Bar, 25 µm.

The posterior lobe is a thin and slightly curved cuticular structure that can be flattened and visualized in two dimensions. Although flattening the posterior lobe affects its three-dimensional structure slightly, it has little effect on the outline of the cuticle in two dimensions. We obtained digital images of dissected posterior lobes and the outline of each posterior lobe was traced and enclosed by an artificial baseline that lies roughly in line with the edge of the lateral plate (shown in Figure 1). We performed all posterior lobe measurements using this closed contour. Posterior lobe size was measured as the area included within the outline of the contour. Posterior lobe shape was represented with a vector that contained elliptical Fourier coefficients (see Materials and Methods). Elliptical Fourier analysis is a mathematical procedure capable of reconstructing the closed contour of a complex shape using the Cartesian coordinates of the contour's outline as input and has been used to represent posterior lobe morphology with high precision in previous studies of the D. melanogaster subgroup species (Liu et al. 1996; Laurie et al. 1997; MacDonald and Goldstein 1999; Zeng et al. 2000). Prior to calculating Fourier coefficients, posterior lobe outline coordinates were also adjusted to normalize each outline to have an area of one. This allowed the set of Fourier coefficients calculated for each posterior lobe to describe mostly posterior lobe shape and not posterior lobe area. To reduce the number of variables that explain variation in posterior lobe shape, Fourier coefficients were then analyzed by PCA. Among the genotypes studied here, PC1 describes 40–45% of the variation in Fourier coefficients on average and represents the principal component that explains the largest fraction of variation in shape. PC1 can therefore be considered a rough approximation of posterior lobe shape.

When possible, we measured both tibias and both posterior lobes from each individual dissected. Pearson correlations between the left and right measurements of bilateral traits were high both in the pure species and in the entire collection of D. mauritiana–D. sechellia introgression lines (Table 1). Left–right correlations were also relatively high among individuals within each D. mauritiana–D. sechellia introgression genotype (median introgression line values: tibia length, r = 0.71; posterior lobe area, r = 0.73; and PC1, r = 0.56). Because left and right tibia length and left and right posterior lobe area were correlated among individuals within each genotype, we used the mean of the left and right measurements of each individual to calculate the correlation between tibia length and posterior lobe area. Tibia length and posterior lobe area showed low correlation, consistent with previous observations of genital allometry in Drosophila (Table 1; Coyne et al. 1991; Liu et al. 1996; MacDonald and Goldstein 1999; Eberhard 2008; Shingleton et al. 2009). This hypoallometric relationship suggests that differences in overall body size have little effect on posterior lobe size.

Table 1 . Phenotypic correlations.

Genotype	Tibia length (L–R)		Posterior lobe area (L–R)		PC1b (L–R)		Tibia, posterior lobe area		Posterior lobe area, PC1
	r	P	r	P	r	P	r	P	r	P
D. sechellia w	0.91	<0.001	0.57	<0.001	0.83	<0.001	0.11	0.42	−0.14	0.16
D. mauritiana	0.72	<0.001	0.67	<0.001	0.49	<0.01	0.06	0.75	−0.50	<0.001
Introgression hybrids	0.79 (0.71)a	<0.001	0.87 (0.73)	<0.001	0.77 (0.56)	<0.001	0.14 (0.12)	<0.001c	−0.44 (0.02)	<0.001

L–R denotes correlation between left and right traits.

^aParentheses show the median value when correlations were calculated for each D. mauritiana–D. sechellia introgression line individually.

^bPC1 obtained from principal components analysis performed using D. sechellia w, D. mauritiana, and all 45 introgression hybrids.

^cSignificant correlation likely reflects large sample size.

The relationship between change in posterior lobe size and change in posterior lobe shape has been difficult to resolve in previous studies of D. mauritiana, D. sechellia, and D. simulans. In D. sechellia, D. simulans, and their reciprocal backcross hybrids, the correlation between posterior lobe area and PC1 was low (MacDonald and Goldstein 1999), which suggests that change in posterior lobe size and change in posterior lobe shape might be largely independent. However, in backcross hybrid genotypes between D. mauritiana and D. simulans, the correlation between posterior lobe area and PC1 was high (Liu et al. 1996), which suggests that change in posterior lobe size might affect change in posterior lobe shape and vice versa. To test the correlation between posterior lobe area and PC1 in our data, we performed PCA by using an approach similar to that of the previous studies and calculated principal components scores on Fourier coefficients from a single group that consisted of all 47 genotypes we studied. Principal components scores calculated from this group of genotypes allowed us to measure the effects of any given genotype on posterior lobe shape relative to the effects of all of the genotypes we studied. We found that the correlation between posterior lobe area and PC1 was low in D. sechellia w, although significant in both D. mauritiana and the collection of introgression hybrids (Table 1). These results suggest that changes in posterior lobe size and changes in posterior lobe shape might not be independent of one another.

We also calculated principal components scores in another way by performing PCA on 45 different groups each composed of three genotypes, a single D. mauritiana–D. sechellia introgression genotype, and the two pure species parental genotypes, D. mauritiana and D. sechellia w. Principal components scores calculated in this way allowed us to test the effects of a single introgression on posterior lobe shape relative to the posterior lobe shapes of only the two pure species parental lines. When we calculated the correlation between posterior lobe area and PC1 obtained in this way, we found that the correlation between posterior lobe area and PC1 for D. mauritiana, D. sechellia w, and each of the 45 D. mauritiana–D. sechellia introgression lines was generally low and not significant (−0.31< r < 0.30, 43 of 45 groups). These results suggest that change in posterior lobe size and change in posterior lobe shape might be independent of one another, and thus each of these phenotypes might be under separate genetic control.

Because it appears unclear whether change in posterior lobe size and change in posterior lobe shape are in fact correlated, we analyzed the phenotypic effects of each D. mauritiana introgression on posterior lobe area and posterior lobe shape separately to attempt to resolve these conflicting results and test whether these two phenotypes are at all specified by separate loci.

D. mauritiana introgression effects on posterior lobe area

Posterior lobe area appears to be little affected by differences in overall body size (Table 1; Coyne et al. 1991; Liu et al. 1996; MacDonald and Goldstein 1999; Eberhard 2008; Shingleton et al. 2009). Nonetheless, to ensure that differences in body size did not affect our posterior lobe area measurements, we corrected our posterior lobe area measurements to account for differences in body size using tibia length, which has been shown to be a good predictor of body size (Catchpole 1994), and performed our analysis of posterior lobe area on both the body size corrected and uncorrected areas. [Although tibia length serves as a good proxy for overall body size in relation to the size of some body parts, it is worth noting that tibia length may not be the best predictor for all body parts (Shingleton et al. 2009)]. In most cases, correcting for body size using tibia length when applied to posterior lobe area measurements had no effect on the overall results of our comparisons. Table 2 shows the mean posterior lobe area and mean tibia length for D. mauritiana, D. sechellia w, and the 45 D. mauritiana–D. sechellia introgression lines. Because the D. mauritiana–D. sechellia introgression lines are mostly D. sechellia w in their genetic background, we expected most introgressions to show posterior lobe areas similar to that of D. sechellia w, and only those introgressions that contain loci of major effect to possess posterior lobes with smaller areas. As expected, most introgressions show posterior lobe areas similar to that of D. sechellia w (Table 2). Two introgressions, however, showed significantly reduced posterior lobe areas compared to D. sechellia w in both the body size corrected and uncorrected comparisons (Table 2, Figure 2). Introgression lines 4C2(A) and Q1(A) possess D. mauritiana introgressions that reside in different regions of chromosome arm 3L (Table S1). Each of these introgressions reduces posterior lobe area by ∼15% compared to D. sechellia w.

Table 2 . Posterior lobe area and tibia measurements.

Genotype	Posterior lobe area	Tibia length	Area (mm2)/tibia (mm)
	(× 103 mm2)	(mm)	(× 103)
D. sechelia w	4.749 ± 0.041	0.4508 ± 0.0025	10.55 ± 0.1005
D. mauritiana	1.672 ± 0.022***	0.4254 ± 0.0013***	3.93 ± 0.0503***
Q1(A)	3.995 ± 0.043***	0.4681 ± 0.0020***	8.54 ± 0.0850***
4C2(A)	4.130 ± 0.049***	0.4423 ± 0.0021*	9.34 ± 0.1174***
3G1(A)	4.587 ± 0.075	0.4545 ± 0.0030	10.10 ± 0.1656
3Q1(A)	4.670 ± 0.066	0.4607 ± 0.0024*	10.14 ± 0.1353
4N2(B)	4.695 ± 0.086	0.4652 ± 0.0032**	10.11 ± 0.2252
3J1(B)	4.697 ± 0.060	0.4606 ± 0.0019*	10.20 ± 0.1264
2H1(A)	4.704 ± 0.099	0.4616 ± 0.0021*	10.19 ± 0.2139
L1C(B)	4.711 ± 0.104	0.4498 ± 0.0035***	10.48 ± 0.2310
EMMA1(A)	4.729 ± 0.088	0.4642 ± 0.0019***	10.19 ± 0.2026
BECKY1(B)	4.767 ± 0.107	0.4548 ± 0.0019	10.49 ± 0.2381
3L1(A)	4.782 ± 0.090	0.4496 ± 0.0026	10.64 ± 0.1819
4Z3(A)	4.788 ± 0.096	0.4591 ± 0.0030	10.44 ± 0.2166
I1(A)	4.798 ± 0.087	0.4557 ± 0.0047	10.57 ± 0.2610
2X2(A)	4.827 ± 0.077	0.4518 ± 0.0029	10.69 ± 0.1853
2Z1(A)	4.840 ± 0.095	0.4709 ± 0.0029***	10.27 ± 0.1830
YAR1(B)	4.859 ± 0.063	0.4601 ± 0.0016*	10.56 ± 0.1398
YAR1(A)	4.890 ± 0.061	0.4571 ± 0.0025	10.70 ± 0.1290
3Q1(B)	4.902 ± 0.089	0.4632 ± 0.0026**	10.58 ± 0.1688
4R1(B)	4.939 ± 0.084	0.4617 ± 0.0039	10.72 ± 0.2120
D1(B)	4.965 ± 0.104	0.4494 ± 0.0034	11.06 ± 0.2468
4G5(A)	4.971 ± 0.076	0.4513 ± 0.0024	11.02 ± 0.1746
E4(A)	4.972 ± 0.102	0.4634 ± 0.0033*	10.73 ± 0.2187
4G5(B)	4.989 ± 0.058	0.4543 ± 0.0018	10.98 ± 0.1221
RYLAH1(C)	4.994 ± 0.124	0.4558 ± 0.0040	10.95 ± 0.2399
V1(B)	5.044 ± 0.072**	0.4589 ± 0.0025	11.00 ± 0.1634
2J1(A)	5.048 ± 0.132	0.4542 ± 0.0027	11.12 ± 0.2851
I1(B)	5.079 ± 0.101**	0.4650 ± 0.0039*	10.91 ± 0.1780
NENEH2(B)	5.097 ± 0.114**	0.4629 ± 0.0024**	11.02 ± 0.2686
BART1(B)	5.130 ± 0.084**	0.4625 ± 0.0020**	11.10 ± 0.1961
ERNIE1(A)	5.136 ± 0.089**	0.4627 ± 0.0033*	11.09 ± 0.1547*
G2(B)	5.168 ± 0.086***	0.4591 ± 0.0015*	11.26 ± 0.1905*
DEE1(B)	5.215 ± 0.061***	0.4523 ± 0.0019	11.54 ± 0.1395***
4X1(B)	5.226 ± 0.108***	0.4497 ± 0.0026	11.62 ± 0.2344**
2Q2(B)	5.302 ± 0.102***	0.4617 ± 0.0031*	11.50 ± 0.2496*
4I3(B)	5.305 ± 0.086***	0.4551 ± 0.0032	11.66 ± 0.1713***
PANDORA1(A)	5.365 ± 0.071***	0.4665 ± 0.0027***	11.51 ± 0.1598***
G2(A)	5.380 ± 0.121***	0.4559 ± 0.0030	11.80 ± 0.2451***
L1C(A)	5.411 ± 0.113***	0.4655 ± 0.0029**	11.63 ± 0.2491**
2H3(B)	5.546 ± 0.113***	0.4613 ± 0.0024*	12.01 ± 0.2228***
2K3(A)	5.631 ± 0.099***	0.4675 ± 0.0018***	12.05 ± 0.2145***
RYLAH1(B)	5.634 ± 0.097***	0.4680 ± 0.0027***	12.04 ± 0.2004***
4J1(A)	5.682 ± 0.088***	0.4499 ± 0.0033	12.64 ± 0.2125***
NENEH2(A)	5.783 ± 0.092***	0.4636 ± 0.0023**	12.48 ± 0.2026***
2u1(C)	6.078 ± 0.100***	0.4494 ± 0.0033	13.54 ± 0.2312***
4G4C(A)	6.178 ± 0.090***	0.4720 ± 0.0017***	13.10 ± 0.2034***

Comparison to D. sechellia w: *P < 0.01, **P < 0.001, ***P < 0.0001.

Figure 2 .

Reduced posterior lobe area in 4C2(A) and Q1(A) introgression hybrid genotypes. Boxplots show the distribution of posterior lobe areas using the average of the measurements from left and right posterior lobes of each individual measured. Boxes show the interquartile range; vertical lines are drawn out to the extreme values; horizontal lines within boxes show the median. Number of individuals measured: D. sechellia w = 60, 4C2(A) = 31, Q1(A) = 59, and D. mauritiana = 55.

Interestingly, some introgression lines show significantly larger posterior lobe areas than D. sechellia w in both the body size corrected and uncorrected comparisons (Table 2). It seems unlikely that hybrid vigor that might result from removing recessive deleterious mutations segregating in our parental stocks is the cause of these larger posterior lobes. Although both of our parental lines are inbred genotypes, the introgression hybrid genotypes we study suffer most of the D. sechellia w deleterious mutational load (as they are mostly D. sechellia w in genetic background) and they suffer all of the D. mauritiana recessive deleterious mutations within their introgression regions (because all introgressions are homozygous). However, we cannot exclude the possibility that certain D. mauritiana introgressions purge regions of the D. sechellia w genome that experience particularly high mutation loads, which might result in larger posterior lobe areas in those introgression hybrid genotypes.

Two other explanations for larger introgression hybrid posterior lobe areas are also possible. First, larger posterior lobes might reflect transgressive variation—hybrid trait values that lie outside the range of the parental species trait values. Transgressive variation can be caused genetically by segregation of additive genetic factors that have phenotypic effects opposite of that expected on the basis of the trait values observed in pure species (Rieseberg et al. 1999). In the case of posterior lobe area, D. sechellia might possess alleles that specify smaller posterior lobe area relative to the effects of the D. mauritiana alleles at some loci, even though the net additive effect of all species-specific alleles results in larger posterior lobes in D. sechellia and smaller posterior lobes in D. mauritiana. Any introgression that contains such D. mauritiana alleles would produce larger posterior lobe areas than that of D. sechellia w because they possess all or most of the D. sechellia alleles, and some D. mauritiana alleles that act in the direction of larger lobe area. Transgressive segregation has been observed in several plant and animal hybrids (Rieseberg et al. 1999), and the results from the previous study of D. mauritiana–D. sechellia posterior lobe area also support this idea. Coyne et al. (1991) measured posterior lobe area from reciprocal D. mauritiana–D. sechellia F₁ hybrids; these hybrids share their autosomal genotype, but differ in the species origin of their X chromosome. They detected a significant effect of the X chromosome on posterior lobe area, but this effect acts in the direction opposite to that predicted from the parental species trait values: F₁ hybrids that possess a D. sechellia X chromosome have smaller posterior lobe areas than F₁ hybrids that possess the D. mauritiana X.

A second possible explanation is that larger introgression hybrid posterior lobe areas might result as a byproduct of incompatible epistatic interactions, or so-called Dobzhansky–Muller incompatibilities. Functional divergence at interacting loci is a common cause of postzygotic fitness problems in species hybrids (Coyne and Orr 2004), but also has the potential to cause abnormalities in other phenotypes in hybrids between recently diverged species. If loci important for posterior lobe morphology interact epistatically and have functionally diverged between D. mauritiana and D. sechellia, introgression of D. mauritiana alleles into a D. sechellia genetic background might result in abnormal morphology of the posterior lobe in these introgression hybrids. Although the genetic basis of posterior lobe morphology appears mostly additive (Coyne et al. 1991; Liu et al. 1996; MacDonald and Goldstein 1999), loci that affect morphology are known to interact epistatically in some species (Costa et al. 2005; Baxter et al. 2007). Resolution of these three possibilities must wait for identification and characterization of the genes involved in species-specific posterior lobe morphology.

D. mauritiana introgression effects on posterior lobe shape

To measure the phenotypic effects of each individual D. mauritiana introgression on posterior lobe shape relative to the shapes of both parental pure species, we performed PCA on area-normalized Fourier coefficients from 45 separate groups that each consisted of an individual D. mauritiana–D. sechellia introgression line, D. mauritiana, and D. sechellia w. Because all individuals within an introgression line share the same genotype, we used both left and right posterior lobe measurements for all genotypes to capture most within-genotype variation in shape. As with posterior lobe area, we expected that D. mauritiana introgressions that contain loci of large effect will produce hybrids with posterior lobe shapes that are significantly different from that of D. sechellia w and appear more D. mauritiana-like.

We tested for differences in posterior lobe shape by comparing the distribution of PC1–PC14 between each D. mauritiana–D. sechellia introgression genotype and D. sechellia w. On average, PC1–PC14 explain ∼97% of the variation in shape among D. mauritiana, D. sechellia w, and the D. mauritiana–D. sechellia introgression lines. Five of the 45 D. mauritiana–D. sechellia introgression lines show a significant difference in the distribution of PC1–PC14 (Figure 3, Figure S1). Although most (4 of 5) of these introgressions appear to have distributions that move toward that of D. mauritiana, one in particular showed a phenotype that appeared clearly more D. mauritiana-like. Introgression 3Q1(A), which bears a D. mauritiana introgression on chromosome arm 2L, has a significant difference in the distribution of PC1–PC14 (F_14,150 = 2.11, P = 0.014; Figure 3) and a striking visual phenotype: 3Q1(A) introgression hybrids possess a less “beaky” posterior lobe shape compared to the wild-type goose-headed D. sechellia posterior lobe shape (Figure 4). This difference in posterior lobe shape did not, however, affect posterior lobe area, as 3Q1(A) shows no significant reduction in area compared to D. sechellia w (Table 2). Three of the remaining four introgressions that had significant effects on posterior lobe shape did affect posterior lobe area in both the body size corrected and uncorrected comparisons (NENEH2(A), DEE1(B), and 2H3(B); Table 2, Figure S1). However, none of these introgressions produced both smaller posterior lobe area and more D. mauritiana-like posterior lobe shape (File S2). Thus, among the introgressions we identified that had significant phenotypic effects, we found none that had effects on posterior lobe area and posterior lobe shape that act in the direction of the D. mauritiana pure species trait values for both phenotypes.

Figure 3 .

Distribution of the first two principal components calculated using area-normalized elliptical Fourier coefficients for introgression hybrid 3Q1(A). D. sechellia w is shown with red triangles, D. mauritiana is shown with blue circles, and introgression hybrid 3Q1(A) is shown with black diamonds. Ellipses show 70% normal-probability contours for each sample. The percentage of variation in shape that principal component 1 (PC1) and principal component 2 (PC2) explain is shown in parentheses.

Figure 4 .

Introgression hybrid Q1(A) and 3Q1(A) posterior lobe morphology. D. sechellia w, introgression hybrid Q1(A), and introgression hybrid 3Q1(A) posterior lobes are shown. Q1(A) shows reduced posterior lobe area compared to D. sechellia w, but no significant difference in posterior lobe shape. 3Q1(A) shows reduced beak morphology compared to the wild-type goose-headed D. sechellia w posterior lobe morphology, but no significant difference in posterior lobe area. Bar, 25 µm.

The two introgressions that show a significant reduction in posterior lobe area, 4C2(A) and Q1(A), have no significant effect on posterior lobe shape when we calculated area-normalized elliptical Fourier coefficients. As an additional test of their difference in posterior lobe morphology, we calculated elliptical Fourier coefficients for these introgressions without first normalizing posterior lobe area, which allowed the Fourier coefficients to describe both size and shape. Although the distributions of PC1–PC14 appear to move away from that of D. sechellia and toward that of D. mauritiana, both 4C2(A) and Q1(A) show no significant differences in morphology (F_14,96 = 1.05, P = 0.41 and F_14,105 = 1.29, P = 0.23, respectively) when posterior lobe area was included in calculating elliptical Fourier coefficients. These results were unexpected, but not entirely surprising, as the differences in posterior lobe area for these two introgressions are significant, but small (∼15% reduction in area compared to D. sechellia w), and neither introgression has a significant effect on posterior lobe shape. Additional analyses using F₁ hybrid males (which show substantial differences in posterior lobe size and shape compared to D. mauritiana and D. sechellia; Figure 1) suggest that the nonsignificant results we obtained using the combined size and shape morphologies for 4C2(A) and Q1(A) might reflect the relative sensitivity of our morphometric assays (File S2).

The results of our morphometric analyses show that some D. mauritiana genomic regions affect posterior lobe size, but not shape (e.g., Q1(A); Figure 4) and others affect posterior lobe shape but not size (e.g., 3Q1(A); Figure 4). Thus, we find that posterior lobe size and posterior lobe shape are indeed under separate genetic control for at least some of the loci that specify morphology.

Species-specific gene expression differences in the genital imaginal disc

There are several possible genetic causes for the rapidly evolving posterior lobe morphology among the D. melanogaster subgroup species. These include species-specific differences in protein coding sequence, differences in gene expression levels, differences in spatial or temporal gene expression, gene copy number differences, differences in transcript splice isoforms, the evolution of species-specific neomorphic loci, or a combination of some of these factors. In an effort to identify potential candidate genes or regulatory pathways that affect species-specific morphology, we measured gene expression level differences among D. mauritiana, D. sechellia w, and two of the D. mauritiana–D. sechellia introgression lines with large phenotypic effects: 3Q1(A), which has a significant effect on posterior lobe shape, and Q1(A), which has a significant reduction in posterior lobe area.

The external genitalia in Drosophila develop from the larval genital imaginal disc (reviewed in Estrada et al. 2003). To identify the genes expressed in this tissue and to quantify differences in gene expression levels, we constructed cDNA libraries for Illumina sequencing from late third larval instar male genital discs of D. mauritiana P-insertion line Q1 (the parental P-insertion stock used to generate introgression hybrid Q1(A)), D. sechellia w, introgression hybrid 3Q1(A), and introgression hybrid Q1(A); we sequenced these libraries using paired-end sequencing with read lengths of 72 bp on each end. Sequence reads were mapped to the D. melanogaster genome using the programs PerM (Chen et al. 2009) and MapSplice (Wang et al. 2010), which allowed us to map transcriptome sequence reads within exons and also allowed us to map reads that span exon splice junctions. We mapped our transcriptome reads from D. mauritiana, D. sechellia, and both introgression hybrids to D. melanogaster for two reasons. First, although the D. sechellia genome is sequenced, the D. mauritiana genome sequence is not currently available, thus no genomic reference exists for this species. Second, both D. mauritiana and D. sechellia are equally evolutionarily distant from D. melanogaster, thus any potential mapping bias due to sequence divergence should affect mapping sequence reads of both species equally. Because a large fraction of our sequence reads were unlikely to map to the D. melanogaster genome with zero mismatches, we analyzed mapping performance by allowing for 0–20 mismatches for each end of a paired-end read. We found that allowing ≤12 mismatches per end read covered a substantial proportion of the uniquely mapped reads, and that allowing for greater numbers of mismatches did not substantially increase the number of mapped reads (Table S2, Figure S2). We therefore used ≤12 mismatches per end read to map our transcriptome sequence reads.

We sequenced three independent biological replicates for each of the four genotypes to measure transcript abundance and quantified gene expression levels using a two-parameter generalized Poisson model (Srivastava and Chen 2010). Correlations of normalized gene expression levels among biological replicates of each genotype were high (0.92 < r < 0.99), which shows that the gene expression level measurements we observe for each genotype were robust. To determine the efficiency and reproducibility of our sequencing, we also added several spike-in control mRNAs at various concentrations to each of our samples during sequencing library construction (see Materials and Methods). Correlations were high when we compared the expected log of normalized expression levels of our spike-in controls and the log of the actual amount of spike-in control RNAs added to each library after we normalized for expression level using their sequence lengths (r > 0.99 for all comparisons). This shows that sequencing performance was uniform across our samples. Summing over all three replicates for each genotype, we obtained 6.85, 6.99, 8.46, and 6.63 million uniquely mappable mate pairs (13.70, 13.97, 16.92, and 13.25 million uniquely mappable sequence reads total) for D. mauritiana, D. sechellia w, introgression hybrid 3Q1(A), and introgression hybrid Q1(A), respectively.

We detected expression of 7800–8000 genes in the genital imaginal disc among the four genotypes we studied. Among these genes, we identified 2261 genes that were differentially expressed between D. mauritiana and D. sechellia w using a FDR correction of 0.05 and a significance threshold of P < 0.01 (Figure 5, Table S3). To identify potential candidates for gene expression differences that might contribute to the posterior lobe phenotypes we observed in 3Q1(A) and Q1(A) hybrids, we compared expression levels in each introgression hybrid genotype to expression levels in D. sechellia w. Because both 3Q1(A) and Q1(A) are mostly D. sechellia w in their genetic background, we expected to observe expression levels similar to those of D. sechellia w for most genes. We identified differential expression in 410 genes between 3Q1(A) and D. sechellia w (Figure 5, blue circle) and 236 genes between Q1(A) and D. sechellia w (Figure 5, red circle). Of the combined 526 genes that were differentially expressed in both of these comparisons, 289 genes (55%) show significant differential expression when the introgression hybrids are compared to D. sechellia, but they do not show significant expression level differences between pure species D. mauritiana and D. sechellia (Figure 5, gray regions). Gene misexpression is often observed in interspecific hybrid genotypes (Michalak and Noor 2004; Wittkopp et al. 2004, 2008; Landry et al. 2005; Malone et al. 2007; Moehring et al. 2007; Catron and Noor 2008; Malone and Michalak 2008; Graze et al. 2009; Renaut et al. 2009; Tirosh et al. 2009; Hill-Burns and Clark 2010; Lu et al. 2010; McManus et al. 2010; Rosas et al. 2010) and in some cases appears to have little effect on phenotypic or fitness differences outside of germline tissues (Hill-Burns and Clark 2010; Rosas et al. 2010). Because each of the two introgression hybrids we sequenced show posterior lobe phenotypes that move in the direction of the pure species D. mauritiana phenotype, it seems unlikely that misexpression of these 289 genes gives rise to the intermediate posterior lobe morphologies we observed in 3Q1(A) and Q1(A). We therefore excluded these genes from our list of potential expression candidates. Misexpression of these genes might, however, cause abnormal phenotypes in genital traits that we did not measure, or they might cause no obvious phenotypic effects.

Figure 5 .

Differential gene expression in the genital imaginal disc. Number of differentially expressed genes are shown for comparisons between D. mauritiana and D. sechellia w (black circle), introgression hybrid 3Q1(A) and D. sechellia w (blue circle), and introgression hybrid Q1(A) and D. sechellia w (red circle). Overlapping regions show those genes that are shared among comparisons. Yellow regions show genes that are both differentially expressed between D. mauritiana and D. sechellia w and are also differentially expressed between the introgression hybrids and D. sechellia w. Gray regions show genes that are differentially expressed between the introgression hybrids and D. sechellia w, but are not differentially expressed between D. mauritiana and D. sechellia w.

If differences in gene expression levels give rise to species-specific posterior lobe morphology, we hypothesized that introgression hybrids 3Q1(A) and Q1(A) might possess gene expression levels similar to those of D. mauritiana for genes that are important for specifying posterior lobe shape and posterior lobe size. To identify the strongest candidate genes for differences in posterior lobe morphology, we first narrowed our candidate gene set to those genes that were both differentially expressed between D. mauritiana and D. sechellia w, and were also differentially expressed between each introgression hybrid and D. sechellia w (Figure 5, yellow regions). Next, we limited our candidates to those genes within this set that do not show significant expression differences when 3Q1(A) or Q1(A) gene expression levels are compared to those of D. mauritiana. Using these criteria, we identified 58 genes that show D. mauritiana expression levels in 3Q1(A), but not in Q1(A), 24 genes that show D. mauritiana expression levels in Q1(A), but not in 3Q1(A), and 18 genes that show D. mauritiana expression levels and are expressed in both introgression hybrid genotypes (100 genes total; Table S4). Among this set of 100 genes, two interesting patterns emerged. First, most of these genes reside in regions of the genome that are outside of the 3Q1(A) and Q1(A) D. mauritiana introgression regions determined by previous D. mauritiana–D. sechellia introgression breakpoint mapping experiments (Masly and Presgraves 2007). These genes are D. sechellia alleles that are expressed at D. mauritiana expression levels in the introgression hybrids, which suggests the possibility that these genes might be regulated as a consequence of some trans-acting factor(s) that lies within the D. mauritiana introgression(s). Furthermore, some of these differentially expressed genes reside in regions of the genome we identified as having large phenotypic effects from tests of other D. mauritiana introgressions, which suggests the possibility that loci specifying posterior lobe morphology might interact. Second, each introgression hybrid showed D. mauritiana expression levels in genes that were expressed at D. sechellia levels in the other introgression hybrid, but both introgression hybrids also showed D. mauritiana expression levels in identical genes, which suggests the possibility that a common biological pathway might integrate size and shape inputs to establish differences in overall posterior lobe morphology.

Introgression hybrids 3Q1(A) and Q1(A) express insulin signaling pathway genes at D. mauritiana expression levels

We examined the 100 differentially expressed D. mauritiana–D. sechellia genes that also show D. mauritiana expression levels in either 3Q1(A), Q1(A), or both to identify potential candidate genes and/or candidate regulatory pathways that could be involved in establishing posterior lobe morphology on the basis of their known molecular and biological functions. Among the 18 genes shared between 3Q1(A) and Q1(A), there were few genes that had obvious potential to affect morphology, but one that appeared to be a particularly strong candidate. Insulin-like receptor (InR) is part of the insulin/insulin-like growth factor signaling (IIS) pathway, an evolutionarily conserved signaling cascade that regulates cell size, cell number, and tissue growth in animals (see Grewal 2009; Hietakangas and Cohen 2009 for recent reviews). The IIS pathway acts through several downstream targets of InR; these target proteins deploy additional downstream effector genes that regulate cellular processes such as cell cycle control, cell motility, and cell survival/apoptosis.

In D. melanogaster, reduced InR expression in the developing male genital disc causes a reduction in adult posterior lobe area (Shingleton et al. 2005). In the male genital disc of our two parental species, InR is expressed at significantly higher levels in D. sechellia compared to expression levels in D. mauritiana (Table 3), consistent with the observed differences in posterior lobe size between these two species (Table 2, Figure 1). The species difference in InR expression level also appears consistent with the observed differences in overall body size between D. mauritiana and D. sechellia w (Table 2). In D. melanogaster, however, the requirements and action of InR are tissue specific; in particular, InR expression level and activity in the genitalia of D. melanogaster males appear independent of expression and activity in other tissues (Shingleton et al. 2005). Given that mRNA derived from the genital discs of introgression hybrids 3Q1(A) and Q1(A) show InR expression levels similar to that of D. mauritiana, we might expect that both introgression hybrid genotypes might show a consequent reduction in overall body size compared to D. sechellia w. Both 3Q1(A) and Q1(A), however, show no reduction in body size compared to D. sechellia w (Table 2), which suggests that InR abundance and activity in the developing genitalia may be independent of other tissues in D. mauritiana and D. sechellia as well.

Table 3 . Posterior lobe morphology candidate loci.

Locus	Molecular functiona	Normalized expression level (GP)				Expression comparisons (P values)
		D. mauritiana	D. sechellia w	3Q1(A)	Q1(A)	sech-Q1(A)	maur-Q1(A)	sech-3Q1(A)	maur-3Q1(A)
Insulin-like receptor	Insulin receptor	35.603	46.062	34.109	34.672	<1.0 E-15	0.538	<1.0 E-15	0.265
S6kII	Protein kinase	19.338	23.520	21.384	18.381	5.22E-03	0.575	0.337	0.185
widerborst	Protein phosphatase	103.137	112.746	101.080	106.179	0.236	0.462	4.10E-03	0.608
twins	Protein phosphatase	70.683	77.442	67.211	73.287	0.261	0.323	2.95E-05	0.141
Rho1	Protein binding; kinase binding	125.886	143.859	124.376	139.013	0.560	3.44E-03	3.04E-05	0.771
Dredd	Apoptotic protease activator	29.573	24.887	28.724	31.504	1.42E-03	0.352	0.070	0.684
doublesex	Transcription factor	93.822	78.619	74.399	76.308	0.627	4.81E-12	0.217	<1.0 E-15
Pox neuro	Transcription factor	17.377	5.232	6.045	7.268	0.198	7.01E-13	0.621	<1.0 E-15
Drop	Transcription factor	92.561	85.337	88.028	84.674	0.951	0.082	0.722	0.327

^aAbridged description from FlyBase. GP, generalized Poisson.

Genes known to either act downstream of InR, or provide input into the IIS pathway through other parallel growth-related pathways were also included among the 100 genes that were expressed at D. mauritiana levels in either 3Q1(A) or Q1(A) but not in both introgression hybrids (Table 3, Table S4). Similar to species-specific expression level differences in InR, most of these downstream target genes show reduced expression in D. mauritiana and the introgression hybrids relative to expression levels in D. sechellia w. Some of the stronger potential candidates for reducing posterior lobe size that are expressed at levels similar to D. mauritiana in introgression hybrid Q1(A) include the ribosomal protein gene S6kII, which encodes a serine/threonine kinase that regulates cell size through the MAP kinase pathway and acts downstream of many growth factor signaling cascades including IIS (Roux and Blenis 2004). S6kII is an important regulator of the Ras/ERK signaling cascade (Kim et al. 2006), and one of its cellular roles in mammals is in inhibition of apoptotic processes (Anjum and Blenis 2008). Consistent with lower S6kII expression, we observed increased expression of some apoptotic genes in our D. mauritiana and Q1(A) datasets. One of these genes, Dredd, is a member of the caspase gene family and encodes a Ced-3-like/Nedd2-like apoptotic protease activator (Chen et al. 1998). The possibility of apoptosis regulating posterior lobe size is particularly interesting, as differences in apoptotic activity are known to affect species-specific horn morphology in Onthophagus beetles (Kijimoto et al. 2010).

Introgression hybrid 3Q1(A) also showed insulin signaling genes expressed at D. mauritiana levels, but in different downstream targets of the IIS pathway than those found in Q1(A). In particular, 3Q1(A) possesses D. mauritiana expression levels of genes that act through the Akt1/PKB branch of the IIS pathway. Two of these genes, widerborst and twins, both encode β-regulatory subunits of protein phosphatase 2A, an important regulatory component of Akt1 activity (Mayer-Jaekel et al. 1992; Hannus et al. 2002), and appears to be involved in several aspects of cellular morphometric processes, including establishment of normal cell morphology (Pinal et al. 2006; Martin-Belmonte et al. 2007), transcription initiation of more downstream insulin signaling genes (Vereshchagina and Wilson 2006), and regulation of apoptosis (Li et al. 2002). Another interesting potential candidate for differences in posterior lobe shape is Rho1, which encodes a GTP-binding protein that regulates cell shape and cell migration by mediating actin rearrangements in the cytoskeleton (Magie et al. 1999; Jacinto et al. 2002; Yan et al. 2009). Underexpression of Rho1 in Drosophila imaginal disc-derived appendages such as the eye results in abnormal ommatidia morphology and defects in cell polarity (Strutt et al. 1997).

Of the few genes expressed at D. mauritiana levels that reside within either the Q1(A) or the 3Q1(A) introgression regions (seven genes total, Table S4), three are not functionally annotated, and among the four genes that are, none appears intuitively obvious as a strong candidate for specifying morphological differences. We categorized the genes in our 3Q1(A) and Q1(A) introgression hybrid candidate expression datasets as differentially expressed when compared to D. sechellia w expression levels, but not differentially expressed compared to D. mauritiana expression levels using the statistical criteria described above. Although these genes represent potential candidates for specifying differences in posterior lobe morphology between D. mauritiana and D. sechellia, we must emphasize that careful molecular work is needed to test their functional importance.

Species-specific transcript splice isoform differences

By mapping reads that span exon junctions, we were able to determine differences in putative transcript splice isoform abundance among the four genotypes we sequenced (see File S1). We analyzed the 100 genes in our introgression hybrid dataset that possess D. mauritiana-like expression levels to determine whether differential expression of these genes might in part reflect differences in transcript isoform abundance.

We identified at least one putative splice junction difference in 22 differentially expressed genes in 3Q1(A), 4 differentially expressed genes in Q1(A), and 10 genes that were differentially expressed in both 3Q1(A) and Q1(A) among the 100 genes in our candidate gene expression dataset (Table S5). Among the 62 splice junctions that showed greater than twofold splicing ratio differences between D. mauritiana and D. sechellia, 53 (85%) showed D. mauritiana-like splicing ratios in the introgression hybrids. As was the case for differences in gene expression levels, most of the genes that possess splice junctions with D. mauritiana splicing ratios reside outside of the D. mauritiana introgression regions in 3Q1(A) and Q1(A) and are thus D. sechellia alleles spliced at the ratios observed among pure species D. mauritiana transcript isoforms. It therefore appears that our set of candidate genes is not only expressed at D. mauritiana expression levels in introgression hybrids, but that the expression level differences might also reflect the relative abundance of D. mauritiana splice isoforms in 3Q1(A) and Q1(A) introgression hybrids.

Sex-biased gene expression in the genital disc

Evolution at loci with sex-biased expression affects species-specific morphology in some insect traits (Barmina and Kopp 2007; Loehlin et al. 2010; Wasik et al. 2010). The posterior lobe is a male-specific morphological trait, thus the genes important for specifying differences in posterior lobe morphology may show male-biased expression. To identify male-biased genes in the genital disc, we performed a microarray experiment to compare gene expression between D. melanogaster third larval instar male and female genital discs. We compared relative expression levels between males and females of two different laboratory strains, Canton-S and Berlin, which allowed us to account for any strain-specific expression bias, and allowed us to identify genes that were consistently expressed in third larval instar genital discs of both sexes. Because a single Drosophila imaginal disc contains ∼1–4 ng of poly(A)⁺ RNA (Klebes et al. 2002), we amplified poly(A)⁺ RNA pools from the total RNA extracted from male and female genital discs using a linear amplification protocol to obtain the requisite material for visualization on the microarrays. We hybridized five independent biological male vs. female replicates for each strain to our microarrays and considered a gene as expressed if it was detected in at least 7 of 10 arrays in the combined Canton-S and the Berlin array sets. Using this criterion, we identified 4336 genes expressed in the combined male and female expression datasets. This number of genes is far fewer than the number we detected as being expressed in the genital disc in our Illumina transcriptome data. There are two possible explanations for this apparent deficit of genes expressed in the D. melanogaster genital disc: (1) the potential exclusion of low copy number transcripts that may have resulted from the necessary mRNA amplification step and (2) the limit of detection of microarray technology vs. next generation sequencing technology, as gene expression needs to exceed a certain threshold to be detected by our microarrays.

Among the genes expressed in male and female discs, we identified 330 male-biased genes and 6 female-biased genes using P < 0.01 after a FDR correction of 0.05 (q < 0.05) as statistical criteria (Table S6). Several genes known to have sex-biased expression were present in our datasets including the sex-specific doublesex (dsx) isoforms (Burtis and Baker 1989), male-specific dosage compensation genes roX1 and roX2 (Franke and Baker 1999), and bab1, a gene with female-biased expression important for sex-specific abdominal pigmentation differences (Kopp et al. 2000). Many of the genes we identified as sex-biased also agree with those identified by another recent study comparing sex-biased expression in the D. melanogaster genital disc across three developmental time points (Chatterjee et al. 2011).

Among the 2261 genes that are differentially expressed in the genital disc between D. mauritiana and D. sechellia males, we found 27 genes that also show male-biased expression levels in D. melanogaster (Table S7). Two of these genes are particularly noteworthy. The first is the sex determination gene dsx, which directs sex-specific development of the genitalia in Drosophila by regulating the action of patterning genes and Hox genes in the genital imaginal disc (Sánchez and Guerrero 2001; Christiansen et al. 2002; Estrada et al. 2003). Differences in dsx expression level have been implicated in species-specific wing size differences in male Nasonia wasps (Loehlin et al. 2010). The second noteworthy gene is Pox neuro (Poxn), which encodes a transcription factor necessary for proper development of the male genitalia in D. melanogaster. Null alleles of Poxn completely abrogate posterior lobe development (Boll and Noll 2002). Although differences in expression levels of dsx and/or Poxn between D. mauritiana and D. sechellia could potentially give rise to differences in their posterior lobe morphologies, both introgression hybrids 3Q1(A) and Q1(A) express dsx and Poxn at D. sechellia expression levels (Table 3).

A third male-biased gene also represents a strong candidate for species-specific morphological differences. The transcription factor Drop (Dr) is functionally important for the development of several male genital structures in D. melanogaster, including the posterior lobe (Chatterjee et al. 2011). Although Dr is not significantly differentially expressed in the genital disc between D. mauritiana and D. sechellia (P = 0.11; Table 3), one D. mauritiana introgression we identified with significant morphological effects might include this locus. Introgression I1(B) resides on chromosome arm 3R in the region near Dr and possesses a posterior lobe shape that moves in the direction of the D. mauritiana phenotype (Figure S1). The D. mauritiana genetic material included in I1(B) extends beyond the proximal limit of our current introgression breakpoint mapping; Dr lies just outside of this region and thus may reside within introgression I1(B).

The genetic architecture of species-specific genital morphology

We measured the effects of 45 homozygous D. mauritiana genetic regions that had been introgressed into a D. sechellia genetic background. Our analysis of posterior lobe morphology among this collection of hybrid introgression lines revealed two D. mauritiana introgressions that significantly reduced posterior lobe size and four D. mauritiana introgressions that had significant directional effects on posterior lobe shape (Figure 6). Although we were unable to test all of the genome, the introgressions that we identified with effects on morphology reside roughly in some of the same regions identified by the previous QTL mapping studies in both the D. mauritiana–D. simulans (Liu et al. 1996; Laurie et al. 1997; Zeng et al. 2000) and the D. sechellia–D. simulans (MacDonald and Goldstein 1999) species pairs. This similarity in location of loci that specify posterior lobe morphology in mapping studies among these three species suggests that evolutionary change at some of the same loci might give rise to differences in posterior lobe morphology. The D. mauritiana introgressions we measured in this study cover approximately half of the genome and revealed a minimum of six loci that have effects on posterior lobe morphology that act in the direction of the D. mauritiana posterior lobe phenotype. Although the distribution of introgressions is not uniform across the genome, and results from previous mapping experiments in this species group show an excess of loci that reside on the third chromosome, if we extrapolate from our results we estimate that genome-wide there are a minimum of 12 loci that specify posterior lobe morphology between D. mauritiana and D. sechellia, which is similar to the minimum number of loci identified in previous mapping studies among the other species in the D. simulans clade.

Figure 6 .

Genomic distribution of D. mauritiana introgressions with large morphological effects. The five major chromosome arms of D. mauritiana and D. sechellia are shown; the tiny “dot” fourth chromosome is omitted. Black circles show the location of the centromeres. Rectangles on the chromosome arms mark the locations of the six D. mauritiana genetic introgressions that have significant effects on posterior lobe size or posterior lobe shape that act in the direction of the D. mauritiana phenotype. Dots shown above the chromosome arms mark the approximate locations of differentially expressed genes in the genital imaginal disc. Parentheses on chromosome arm 3R mark the location of the breakpoints of a fixed inversion difference between 3R in D. melanogaster and 3R in D. mauritiana, D. sechellia, and D. simulans. Numbers below each chromosome arm show approximate cytological regions. DE, differentially expressed.

Our results are also consistent with the results of the previous mapping study between D. mauritiana and D. sechellia. Coyne et al. (1991) mapped genomic regions with large effects on posterior lobe area using backcross hybrid genotypes between these two species. Among the eight backcross genotypes they analyzed, they found that both the second and third chromosomes had large effects on posterior lobe area, but the X chromosome appeared to have no significant effect. However, when comparing the reciprocal D. mauritiana–D. sechellia F₁ hybrid males, they found that the X chromosome did have a significant effect on posterior lobe area. This effect was in the direction opposite of what might be expected from the posterior lobe areas of both pure species, as F₁ hybrid males that carried the D. sechellia X chromosome had smaller posterior lobes than F₁ hybrid males that carried the D. mauritiana X chromosome. Among our introgression hybrid genotypes, we found D. mauritiana introgressions on the second, third, and X chromosomes that had large effects on posterior lobe area. On the X chromosome in particular, five introgressions had large effects on posterior lobe area (PANDORA1(A), L1C(A), 4J1(B), NENEH2(A), and 2u1(C); Table 2). Each of these introgressions caused a significant increase in posterior lobe area, consistent with the effect of the D. mauritiana X chromosome identified by Coyne et al. (1991) from their analysis of reciprocal F₁ hybrids. Although it appears that the X chromosome might have an effect on posterior lobe area in this species pair, as mentioned earlier, we cannot exclude hybrid vigor as a possible cause of larger posterior lobes in the introgression hybrids. Thus, at the current level of resolution, our data do not allow us to conclude definitively the existence of loci specifying posterior lobe area on the X chromosome. However, it does appear that at least one locus with directional effects on posterior lobe shape resides on the X (NENEH2(A); Figure S1).

The D. mauritiana introgression regions we identified with large phenotypic effects showed that posterior lobe size and posterior lobe shape are, in part, under separate genetic control for at least some of the loci important for specifying posterior lobe morphology. Although we did find some introgressions with effects on both size and shape, none of these D. mauritiana introgressions alone produced both reduced posterior lobe area and more D. mauritiana-like posterior lobe shape. Other regions of the genome that we were unable to test here could, of course, contain loci that cause directional phenotypic effects for both posterior lobe size and posterior lobe shape. The D. mauritiana introgressions of large effect that we identified are also relatively small compared to those identified by previous mapping studies (e.g., Q1(A) includes <80 genes at the current introgression breakpoint mapping resolution), which will facilitate the identification of the causative loci in future work. Some of these D. mauritiana loci might be transcription factors with several downstream targets, as our transcriptome analyses of 3Q1(A) and Q1(A) male genital imaginal discs identified several genes from outside of the D. mauritiana introgressions that were expressed at D. mauritiana expression levels.

The transcriptome analyses also revealed several genes that lie downstream of the IIS and other growth factor-related signaling pathways that showed both differential expression in the genital disc between pure species D. mauritiana and D. sechellia and showed expression levels similar to D. mauritiana in the two introgression hybrids we studied. Although these genes appear intuitive on the basis of their known biological functions and thus make attractive candidates, caution must be taken in interpreting their potential importance in establishing posterior lobe morphology. First, the third larval instar genital imaginal disc gives rise to several structures of the internal and external genitalia and analia. Although most of these appear normal in 3Q1(A) and Q1(A), it is possible that any observed gene expression differences could affect structures that we did not measure carefully. Second, the posterior lobe grows out of the developing epandrium sometime between 45 and 65 hr after puparium formation (V. Orgogozo, personal communication). Because this period of development occurs long after the third larval instar stage, the genes included in our transcriptome dataset might not be expressed in the genitalia at this stage of development or they show no differential expression between genotypes, and therefore might have little or no effect in specifying posterior lobe morphological differences between D. mauritiana and D. sechellia.

Our microarray experiments identifying sex-biased gene expression in D. melanogaster detected fewer total genes expressed in the genital disc than the number we identified in our D. mauritiana–D. sechellia transcriptome sequencing experiment. Consequently, we may have missed some male-biased genes that are important for establishing morphological differences that are differentially expressed between D. sechellia and the 3Q1(A) and Q1(A) introgression hybrids. However, among those genes that we did identify as sex-biased, we still found few genes with male-biased expression that were either expressed at levels similar to D. mauritiana in our introgression hybrid transcriptome datasets (Table S7) or included within the D. mauritiana introgression regions with large morphological effects (nine genes total, Table S6). This result might be best interpreted from the perspective of evolution in a developmentally important gene regulatory network (Erwin and Davidson 2009). Development of the Drosophila genitalia occurs under the control of the sex determination regulatory hierarchy. Genes that act early in the sex hierarchy occur at or near the top of the network topology (e.g., dsx) and specify sex-specific cell fates and sexual identity in developing tissues. In this region of the regulatory network, genes are likely to show substantial sex bias, as they function to direct large-scale, sex-specific developmental changes (see Ehrensperger 1983; Epper 1983a,b for a description of the extensive tissue rearrangements that occur early in the developing genitalia of male and female Drosophila). In contrast, genes that occur at the periphery of the network topology act near the end of the developmental program and may thus have more limited roles in establishing gross sex-specific morphologies. Because these genes act in tissues in which sexual identity and sex-specific morphogenesis have already been established, they may not necessarily exhibit any sex bias, but might instead perform sex-specific functions within their given sexual and developmental context. Introgression hybrids 3Q1(A) and Q1(A) possess posterior lobe phenotypes that can be described as slight modifications of the overall wild-type D. sechellia posterior lobe phenotype; the genes we identified that are expressed at D. mauritiana levels in these two introgression hybrids likely include those that function nearer the periphery of the posterior lobe developmental regulatory subcircuit and thus might exhibit little sex-biased expression. Future molecular analyses of both the sex- and species-biased candidate genes we have identified here will be necessary to test the functional importance and relationships among these genes in directing development of species-specific posterior lobe morphology.