Phylogenetic analyses suggest that violations of “Dollo’s law”—that is, re-evolution of lost complex structures—do occur, albeit infrequently. However, the genetic basis of such reversals has not been examined. Here, we address this question using the Drosophila sex comb, a recently evolved, male-specific morphological structure composed of modified bristles. In some species, sex comb development involves only the modification of individual bristles, while other species have more complex “rotated” sex combs that are shaped by coordinated migration of epithelial tissues. Rotated sex combs were lost in the ananassae species subgroup and subsequently re-evolved, ∼12 million years later, in Drosophila bipectinata and its sibling species. We examine the genetic basis of the differences in sex comb morphology between D. bipectinata and D. malerkotliana, a closely related species with a much simpler sex comb representing the ancestral condition. QTL mapping reveals that >50% of this difference is controlled by one chromosomal inversion that covers ∼5% of the genome. Several other, larger inversions do not contribute appreciably to the phenotype. This genetic architecture suggests that rotating sex combs may have re-evolved through changes in relatively few genes. We discuss potential developmental mechanisms that may allow lost complex structures to be regained.
DOLLO’S “law of irreversibility” posits that complex morphological structures, once lost during evolution, cannot be regained in the same form. This principle makes intuitive sense: resurrecting an extinct developmental pathway in close to its ancestral condition seems biologically as well as statistically implausible. And yet, phylogenetic analyses have revealed several cases where Dollo’s law is apparently violated. Examples include re-evolution of lost digits in lizards (Kohlsdorf and Wagner 2006; Brandley et al. 2008; Kohlsdorf et al. 2010), eggshells and oviparity in boas (Lynch and Wagner 2010), wings in stick insects (Whiting et al. 2003), shell coiling in limpets (Collin and Cipriani 2003), mandibular teeth in frogs (Wiens 2011), molars in lynx (Kurten 1963), and others. There is a growing consensus that lost structures can sometimes be regained, especially if that happens soon after the initial loss (Wiens 2011) (but see Goldberg and Igic 2008 and Galis et al. 2010 for counterarguments). This shifts the question from the realm of phylogenies to developmental genetics: how can complex structures re-evolve? What is the genetic basis of such reversals?
In this report, we examine the genetic basis of a likely violation of Dollo’s law that occurred during the evolution of Drosophila sex combs. Sex combs are male-specific arrays of modified mechanosensory bristles (“teeth”) that evolved within the genus Drosophila and are used by males during courtship and mating (Kopp 2011). These structures develop from either transverse or longitudinal bristle rows that are present on the front legs of both sexes and show extensive morphological diversity but essentially fall into three distinct types. “Rotating” sex combs develop from one or several transverse bristle rows (TBRs) that undergo a 90° rotation. This rotation is driven by a precisely coordinated rearrangement of several hundred epithelial cells that, assisted by strong homophilic adhesion between adjacent bristle cells, moves the embedded bristle rows from a transverse to a more longitudinal orientation (Atallah et al. 2009a,b; Tanaka et al. 2009). In contrast, “transverse” sex combs are simply TBRs composed of modified bristles. In this case, sex comb development is limited to the modification of individual bristle shafts and does not involve any morphogenetic movements (Kopp 2011). Finally, “longitudinal” sex combs resemble rotating sex combs in adult flies but actually develop from longitudinal bristle rows and are not homologous to the rotating sex combs on a cell-by-cell basis (Atallah et al. 2009b; Tanaka et al. 2009). Sex comb evolution presents many examples of divergence and convergence, and each developmental mechanism has evolved more than once (Atallah et al. 2009b, 2012; Tanaka et al. 2009) (Figure 1A).
An apparent violation of Dollo’s law is observed in the ananassae subgroup of the melanogaster species group. This lineage consists of over 20 species, most of which have simple transverse sex combs (Matsuda et al. 2009) (Figure 1, A and B). The only exception is found in one pair of sibling species, D. bipectinata and D. parabipectinata, which have much more dramatic sex combs that develop by active rotation (Figure 1, A and C). Detailed analysis of cell behavior during the pupal stage shows that this rotation occurs by the same mechanism as in distantly related groups such as the obscura species group and D. melanogaster and its relatives (Atallah et al. 2009b; Tanaka et al. 2009). D. bipectinata and D. parabipectinata branch deeply within the ananassae subgroup, while all basal lineages have transverse sex combs. Phylogenetic analysis strongly suggests that the common ancestor of these species has re-evolved a rotating sex comb following a previous loss at the base of the ananassae subgroup (Kopp and Barmina 2005; Barmina and Kopp 2007; Matsuda et al. 2009) (Figure 1A).
D. bipectinata can be hybridized with its close relative D. malerkotliana (Bock 1978; Kopp and Barmina 2005), which has simple transverse sex combs, opening the way for a direct genetic analysis of sex comb re-evolution. We used a QTL mapping approach to identify the genomic regions responsible for the differences in sex comb morphology between D. bipectinata and D. malerkotliana. Surprisingly, we find that much of the species difference maps to a single chromosomal inversion that covers ∼5% of the genome.
Materials and Methods
Drosophila strains, crosses, and phenotypic analysis
Drosophila strains were obtained from the Drosophila Species Stock Center or provided by Y. Fuyama and M. Matsuda and maintained on standard cornmeal media. Polytene chromosome spreads were prepared from salivary glands of female larvae using acetic orcein staining. Strains D. malerkotliana 14024-0391.00 and D. bipectinata 14024-0381.03 were inbred by single-pair, full-sib crosses for 12 and 18 generations, respectively, to generate the derivative strains mal0-sc2 and bip3-isoA. Interspecific hybridizations and subsequent crosses were performed in mass cultures with at least 20 males and 20 females per generation. For phenotypic analysis, male legs were removed just above the tibia–tarsus joint, mounted in Hoyers media between two 60- × 22-mm coverslips, and examined under a high-power compound microscope with brightfield illumination. Right and left legs from each male were kept together on an individual slide, and the average number of sex comb teeth per leg was recorded.
Transcriptome sequencing and identification of fixed differences between parental strains
To identify single nucleotide polymorphisms that distinguish mal0-sc2 and bip3-isoA, normalized cDNA libraries were synthesized from whole-body adult RNA samples extracted from each strain. Paired-end libraries were prepared from the sheared, normalized cDNA using the standard Illumina protocol and sequenced with 85-base paired-end reads on an Illumina Genome Analyzer II at the University of California Davis Genome Center (Supporting Information, Table S1). The partial transcriptomes of mal0-sc2 and bip3-isoA were assembled de novo using ABySS (Birol et al. 2009; Simpson et al. 2009; Miller et al. 2010) followed by Trans-ABySS (Robertson et al. 2010) and CAP3 (Huang and Madan 1999) (Table S2; File S1a and File S1b; and File S2a and File S2b). For genetic analysis, we required fixed differences (FD) between parental strains, i.e., nucleotide positions where bip3-isoA is fixed for one allele and mal0-sc2 for a different allele. To identify FDs for the first round of genotyping, mal0-sc2 and bip3-isoA read libraries were each mapped separately to both bip3-isoA and mal0-sc2 de novo transcriptome assemblies using SOAP2 (Li et al. 2009). Same-species mapping (mal0-sc2 reads to mal0-sc2 transcriptome and bip3-isoA reads to bip3-isoA transcriptome) serves to correct assembly errors, gauge coverage at each position, and identify nucleotide positions where multiple alleles segregate within each parental strain despite inbreeding. Cross-species mapping allows us to identify polymorphic positions and select FDs that meet coverage cutoffs. Transcriptome alignments yielded >40,000 FDs between bip3-isoA and mal0-sc2 that are located in pairs of orthologous transcripts (File S3). For the second round of genotyping, bip3-isoA and mal0-sc2 cDNA reads were mapped to the D. bipectinata reference genome (GenBank AFFE00000000.1) (File S4). To identify the genomic locations of FDs, we BLASTed the sequence flanking each FD against the D. ananassae FlyBase 1.3 (July 2011) reference genome and transcriptome. The transcriptome assemblies and SNP data sets for D. malerkotliana, D. bipectinata, and several related species are described in detail elsewhere (Signor et al. 2013).
Genotyping and genetic mapping
To construct linkage maps and identify QTL intervals, we genotyped the progeny of two separate F2 backcrosses and an F38 introgression line (see Results). For the first round of genotyping, we selected 32 FDs that were evenly distributed among the major chromosome arms (Muller elements A–E) and were located at least 2.4 Mb from each other in the D. ananassae genome (Table S3). Because the initial analysis suggested the presence of one or more strong QTL on Muller E or proximal Muller D, we performed a second round of genotyping with 32 additional FDs concentrated on these chromosome arms. Genotyping primers were designed using Typer (Sequenom) based on the sequences of at least 70 bp upstream and 70 bp downstream from each candidate FD, after accounting for within-strain polymorphisms (Table S4). Individual flies were genotyped using MassARRAY (Sequenom) single-base extension in a 32-plex format using standard protocols (Table S5).
Linkage maps were constructed using R-QTL (Broman et al. 2003; Broman and Sen 2009). To map QTL responsible for the differences in sex comb morphology, we applied the Haley–Knott, multiple imputation, and expectation-maximization models (Dempster et al. 1977; Haley and Knott 1992; Sen and Churchill 2001) to our data using the R-QTL package. All three methods gave nearly identical peak locations, LOD scores, and significance levels, indicating that the data are robust to overparameterization. We first performed single-QTL scans to identify likely regions of genotype–phenotype association. For each detected QTL, we performed composite interval mapping and determined that genotypes at neighboring markers did not significantly affect the peak LOD score or width. We calculated the additive and epistatic interactions between all markers with scans utilizing two-QTL models. To determine the statistical significance of QTL peaks, we used a genome scan-adjusted P-value corresponding to the observed LOD score. The null distribution was derived through a standard permutation test. To test for the presence of a QTL on the nonrecombining fourth chromosome (Muller F), we genotyped two FDs in eight of the lightest and eight of the darkest individuals in each F2 backcross using cleaved amplified polymorphism sequences (Darvasi and Soller 1992; Konieczny and Ausubel 1993). QTL association power analysis was performed using R-QTLDesign (Sen et al. 2006). A more detailed description of sequencing and genotyping methods is provided in the Expanded Methods Section online (File S5).
Mapping candidate genes to the D. bipectinata genome
To determine the locations of Scr and dsx on our linkage maps, we BLASTed the full-length sequences of the D. ananassae genomic regions encompassing each gene, as well as the mature transcript of each gene, against the D. bipectinata genome assembly. Both the genomic and the transcript sequences mapped unambiguously to genome scaffolds that contained several of our genotyping markers. Scr mapped to scaffold scf7180000396708, which also contained markers E-In(2L)D-u7 through E-In(2L)D-u13, while dsx mapped to scaffold scf7180000395971, which also contained marker E-In(2L)D-u16. On the linkage map, both of these scaffolds are located in the distal-most, nonrecombining segment of Muller E (2L) corresponding to the inversion in In(2L)D. A similar BLAST analysis shows that genotyping markers linked to the major Muller E QTL are located in a different, more proximal nonrecombining region corresponding to the inversion In(2L)M (Table S6).
Allele-specific expression analysis
Allele-specific pyrosequencing was performed in male F1 hybrids between D. bipectinata bip3-isoA and D. malerkotliana mal0-sc2. First and second pupal legs between 16 and 20 hr after pupariation were dissected and stored in TRIzol (Invitrogen) at −70°. Three replicates of 24–43 individuals each were collected. RNA samples were extracted and reverse-transcribed using oligo(dT) primer and Superscript II (Invitrogen) following the manufacturer’s protocols. Second-strand synthesis was performed using DNA polymerase I and RNase-H. Nucleotide substitutions in the Scr sequence were identified by amplifying and sequencing an ∼500-bp fragment of the second coding Scr exon from each parental strain. A 163-bp region flanking the chosen SNP was amplified using primers Fwd CATGTGGTACGGCACGATGTTCA and Rev biotin-GAGTTCCACTTCAACCGCTACCTG. Extension primer CTTGTGCTCCTTCTTCCACTTCA, which anneals upstream from the targeted SNP site, was used to measure allele-specific gene expression as described (Wittkopp et al. 2004). The sequence immediately downstream of this primer is TAC in D. bipectinata and TGC in D. malerkotliana. The polymorphic A/T site (underlined) corresponds to position 1871 in the D. melanogaster Scr-A transcript.
Phenotypic and chromosomal variation in D. bipectinata and D. malerkotliana
To examine the genetics of re-evolution of large rotated sex combs, we carried out a series of crosses between D. bipectinata and D. malerkotliana. Prior to that analysis, we analyzed several strains of each species to estimate the degree of intraspecific variation. The number of sex comb teeth per leg varied from 5.2 ± 0.72 to 6.9 ± 1.03 in D. malerkotliana and from 13.6 ± 1.2 to 18.7 ± 1.37 in D. bipectinata (Table 1). There was no detectable variation in other aspects of sex comb morphology such as their position and orientation or the shape and color of teeth. Two strains of D. bipectinata and two of D. malerkotliana were crossed in all possible combinations and in both directions. F1 hybrid males showed sex comb morphology that was intermediate in all respects (number of teeth, orientation, and tooth morphology and color). Males from reciprocal crosses showed only slight differences in sex comb size (Table S7), indicating that this phenotype is controlled predominantly by autosomal genes.
D. bipectinata and D. malerkotliana differ by several fixed chromosomal inversions, and each species is also polymorphic for many inversions (Tomimura et al. 2005). By examining the polytene chromosomes of multiple strains of each species and their F1 hybrids, we determined that the strains D. malerkotliana 14024-0391.00 and D. bipectinata 14024-0381.03 differed by the smallest number of inversions, all but one of which are completely fixed between the two species. The remaining inversion In(2L)D, which occupies the distal part of the chromosome arm 2L from 18A to 28D (Table S8), is polymorphic within D. bipectinata. However, repeated attempts to cross the only available strain that lacked this inversion to different strains of D. malerkotliana did not succeed.
We inbred D. malerkotliana 14024-0391.00 and D. bipectinata 14024-0381.03 for 12 and 18 generations by single-pair full-sib crosses, respectively. The resulting strains, called mal0-sc2 and bip3-isoA, were used for all subsequent experiments. These strains differ by one inversion each on XL and XR (Muller element A), two adjacent inversions on 2L (Muller E), none on 2R (Muller D), one on 3L (Muller C), and several overlapping inversions that cover almost the entire 3R (Muller B). The chromosome order of each strain is given in Table S8, and the autosomal inversions are shown in Figure S1. In crosses between these two strains, ∼50% of the euchromatic genome is locked inside chromosomal inversions.
QTL mapping in F2 hybrids identifies a major QTL on Muller-E
To estimate the number of genomic regions contributing to the dramatic difference in sex comb morphology between D. bipectinata and D. malerkotliana, we first performed two F2 backcrosses between mal0-sc2 and bip3-isoA. F1 females from the cross between D. malerkotliana females and D. bipectinata males were crossed separately to mal0-sc2 or bip3-isoA males. In the former cross, F2 recombinant males are either heterozygous for the D. malerkotliana and D. bipectinata alleles or homozygous for the D. malerkotliana allele at each locus; in the latter, F2 males are either heterozygous or homozygous for the D. bipectinata alleles. We examined the sex combs of 427 and 528 F2 males from these two crosses, respectively. In both panels, all aspects of sex comb morphology were correlated: males with the largest number of sex comb teeth had fully rotated sex combs with curved dark teeth, males with the smallest number of teeth had unrotated sex combs with straight light teeth, and those with an intermediate number of teeth were also intermediate in sex comb orientation and tooth morphology. The number of teeth could be quantified unambiguously, while the angle of rotation proved difficult to quantify due to the variation in the orientation of legs mounted on slides. We therefore used the number of teeth (average between the left and right forelegs) as proxy for species-specific sex comb morphology in subsequent analyses. The distribution of sex comb size in each F2 backcross is shown in Figure 2, A and B.
A total of 188 F2 males from the mal0-sc2 backcross and 163 males from the bip3-isoA backcross were genotyped for 28 SNP markers distributed among all major chromosome arms. Since recombination was low and uneven due to inversion heterozygosity in F1 females, linkage maps were inferred using a combination of empirical genetic distances and physical locations of the markers in the genomes of D. bipectinata and D. ananassae (Figure S2). QTL analysis showed that, in both crosses, a large fraction of the difference between species was explained by a single genomic region that spanned most of Muller E (2L) (Figure 3). In the mal0-sc2 backcross, mean sex comb size was 9.1 ± 1.12 for malerkotliana homozygotes and 12.3 ± 1.17 for bipectinata/malerkotliana heterozygotes for all markers on distal Muller E. In the bip3-isoA backcross, mean sex comb size was 14.2 ± 1.31 for bipectinata/malerkotliana heterozygotes and 17.2 ± 1.79 for bipectinata homozygotes. Thus, this QTL interval accounts for ∼6–6.4 teeth, or slightly >50% of the total difference between parental strains.
Other QTL intervals had much weaker effects. A QTL located on Muller C + B (3L + 3R) was significant at the 95% level and had a total effect of ∼1.43 teeth in the D. malerkotliana backcross, but was not significant with an effect of ∼0.93 teeth in the D. bipectinata backcross (Figure 3). A possible QTL on Muller A (XL + XR) had an effect of ∼0.98 teeth in the D. bipectinata backcross and ∼0.65 teeth in the D. malerkotliana backcross and did not reach significance in either panel. No QTL were detected on Muller D (2R) or Muller F (the dot chromosome). Two-QTL scans did not reveal any epistatic interactions between the Muller E, Muller C + B, and Muller A QTL, suggesting that the loci controlling variation in sex comb size act in an additive manner. Under the additive model, and assuming that both Muller C + B and Muller A contain true QTL, all detected QTL together explain only 7.9–8.4 teeth, or ∼67–72% of the difference in sex comb size between bip3-isoA and mal0-sc2. The rest of this difference is likely to be controlled by even weaker QTL that are below our power of detection. Assuming that all QTL are fully additive, the number of genotyped males provides 95% power to detect a QTL with an effect size of at least 1.01 teeth in the D. malerkotliana backcross, or at least 1.41 teeth in the D. bipectinata backcross, for QTL located on Muller A–E (Lynch and Walsh 1998; Sen et al. 2007). These estimates are in agreement with the effect size of the weak QTL detected on Muller C+B.
Refined mapping localizes the major QTL to a single chromosomal inversion
The 2L/Muller E chromosome arm carries two adjacent interspecific inversions, In(2L)D (18A; 28D) and In(2L)M (28D; 34A) (Table S8 and Figure S1). The proximal 2L (34A–45D) and all of 2R/Muller D are free of inversions. To examine the genetic basis of interspecific differences more closely, we sought to increase the amount of recombination between the D. bipectinata and D. malerkotliana genomes using an introgression approach. F2 males were sterile in both backcrosses. We crossed F2 females from the mal0-sc2 backcross to mal0-sc2 males. Some fraction of F3 males was fertile when crossed to mal0-sc2 females. In the F4, we selected males with the largest sex combs, which were used to found an introgression strain. There is some recombination in D. bipectinata males, but it is low compared to females (Singh and Banerjee 1996). Since the sex comb phenotype can be scored only in males, we used the following crossing scheme: in even-numbered generations, hybrid males with the largest sex combs were selected and crossed to mal0-sc2 females, while in odd-numbered generations randomly chosen hybrid females were crossed to mal0-sc2 males (Figure 4A). These crosses should eventually make the introgression strain homozygous for D. malerkotliana alleles at all loci, with the exception of genomic regions that are strongly linked to genes responsible for the interspecific differences in sex comb morphology.
We examined polytene chromosomes in the introgression strain after 20 generations. We found that it was polymorphic for In(2L)D and In(2L)M, indicating that it was heterozygous for the D. bipectinata and D. malerkotliana alleles on Muller E. On all other chromosome arms, the introgression was homozygous for the D. malerkotliana arrangement. This result suggested that only the 2L (and potentially 2R) made a major contribution to species differences, confirming the results of F2 QTL mapping. Throughout the introgression process, different aspects of sex comb morphology (orientation and the number, shape, and color of teeth) continued to be correlated (Figure 4B).
After 36 total generations of introgression (corresponding to 19 recombining female generations), males with the largest sex combs were crossed to mal0-sc2 females, and the resulting F37 males and females were crossed to each other en masse. This resulted in F38 males that could in principle be homozygous for the D. bipectinata allele, heterozygous, or homozygous for the D. malerkotliana allele at any locus. We examined the sex combs of 590 F38 males. The distribution of sex comb sizes was more clearly bimodal than in the F2 (Figure 2C), suggesting that this phenotype was largely controlled by a single genomic region and that some of the weaker QTL have been removed by repeated backcrossing. A total of 185 F38 males were first genotyped for the same 28 SNP markers as the F2 panels; preliminary QTL mapping was consistent with the F2 results. We therefore genotyped the F38 panel for 25 additional markers on Muller E and 6 additional markers on Muller D.
In the F38 recombinant panel, all marker loci located on the proximal Muller E, most of Muller D, and all other chromosome arms were homozygous for D. malerkotliana alleles, while the distal portion of Muller E that carries the inversions In(2L)D and In(2L)M was polymorphic for the D. bipectinata and D. malerkotliana alleles and showed a strong association with the sex comb phenotype (Figure 5). We did not observe any males homozygous for the D. bipectinata alleles on Muller E, suggesting that some interaction between one or more D. bipectinata genes in this region with homozygous D. malerkotliana alleles elsewhere in the genome results in hybrid lethality. With a single exception (see below), the entire distal region of Muller E segregated as a single block, as expected from the inversion heterozygosity. Males homozygous for the D. malerkotliana alleles throughout this region had small sex combs (8.3 ± 1.13 teeth), whereas the malerkotliana/bipectinata heterozygotes had sex combs that were intermediate in size between the parental species (11.3 ± 1.04) (Figure 5C). Thus, in the F38 as well as in the F2, this single QTL region accounts for ∼6 teeth, slightly >50% of the total difference between the D. bipectinata and D. malerkotliana parents. Sex comb size in the F38 males homozygous for the D. malerkotliana Muller E (8.3 ± 1.13 teeth) is larger than in the mal0-sc2 parent (7.13 ± 1.13 teeth) but smaller than in the F2 males homozygous for the D. malerkotliana Muller E (9.1 ± 1.12 teeth). This suggests that some of the minor QTL persist in the introgression strain but are below our power of detection.
We observed a single recombination event between In(2L)D and In(2L)M. This rarity is not surprising, given the close proximity between the inversion breakpoints. This fortuitous event allowed us to localize the region responsible for the differences in sex comb morphology more precisely. The recombinant male was homozygous for D. malerkotliana alleles in the more proximal block of markers, presumably corresponding to In(2L)M, but heterozygous for the D. bipectinata and D. malerkotliana alleles in the more distal block that presumably corresponds to In(2L)D (Figure 5C). The number of sex comb teeth in this male (8) almost exactly matches the mean sex comb size of the 93 males that are homozygous for D. malerkotliana alleles over the entire Muller E (8.30 teeth), but is clearly different from the mean phenotype of the 91 males that are heterozygous for the D. bipectinata and D. malerkotliana alleles in both In(2L)D and In(2L)M (11.32 teeth). This indicates that the major Muller E QTL interval corresponds to the In(2L)M inversion.
The markers that cosegregate with In(2L)M map to four scaffolds in the genome of D. bipectinata. Together, these scaffolds cover 6049 kb and contain several hundred genes including transcription factors, Polycomb and Trithorax group genes, and other regulatory genes that could potentially affect sex comb development (Table S9). Since recombination mapping within the In(2L)M inversion is not feasible, we cannot determine whether this QTL corresponds to a single locus or reflects the cumulative effect of several weaker QTL.
Scr and dsx are not directly responsible for the interspecific differences
The HOX gene Sex combs reduced (Scr) and the sex determination gene doublesex (dsx) play central roles in sex comb development. The evolutionary origin of sex combs coincides with the origin of a new dsx expression domain and novel regulatory interactions between Scr and dsx (Barmina and Kopp 2007; Tanaka et al. 2011). A combination of experimental and comparative evidence suggests that changes in dsx and Scr expression were responsible for sex comb evolution (Kopp 2011; Tanaka et al. 2011). Moreover, dsx and Scr expression differs between D. bipectinata and D. malerkotliana in a way consistent with their morphological differences (Barmina and Kopp 2007; Tanaka et al. 2011). Since both genes are located on Muller E (2L of D. bipectinata), we tested whether they could be responsible for the differences in sex comb morphology between these species.
We used allele-specific pyrosequencing (Cowles et al. 2002; Wittkopp et al. 2004) in F1 hybrids between bip3-isoA and mal0-sc2 to test whether the interspecific differences in Scr expression had a cis-regulatory component. We found that the D. bipectinata allele of Scr was expressed at a significantly higher level than the D. malerkotliana allele in the prothoracic, but not in the mesothoracic, pupal legs of F1 hybrid males (t-test P = 0.0003; Figure 6). Thus, Scr expression has diverged between D. malerkotliana and D. bipectinata due at least in part to changes at the Scr locus.
To localize Scr and dsx relative to the linkage map and inversion boundaries, we BLASTed the coding sequences of these genes and the transcriptome contigs that contained our genotyping markers against the D. bipectinata genome scaffolds. We found that both genes were located on genomic scaffolds that also included SNP markers that were part of the In(2L)D linkage block (Figure 5, Table S6). In contrast, phenotypic differences between D. malerkotliana and D. bipectinata are associated with the In(2L)M block (Figure 5C). Thus, despite the evidence for cis-regulatory divergence at the Scr locus, neither that gene nor dsx are directly responsible for species divergence.
A single inversion, In(2L)M, is responsible for slightly more than half of the difference in sex comb morphology between D. bipectinata and D. malerkotliana. Minor QTL located elsewhere in the genome have much weaker effects. In(2L)M spans chromosomal bands 28D–34A, out of the total of 100 chromosome divisions. In other words, much of the interspecific difference maps to ∼5% of the genome. Other chromosomal inversions that are also completely fixed between D. malerkotliana and D. bipectinata, including several inversions that are much larger than In(2L)M, make little or no contribution to the differences in sex comb morphology. Thus, although we cannot determine the number of genes in the In(2L)M inversion that contribute to its total effect, it is possible that this number is relatively small.
Re-evolution of large rotating sex combs in the last common ancestor of D. bipectinata and D. parabipectinata represents a major developmental change. The number of bristles recruited into the sex comb, rotation of the surrounding epithelium, the shape of bristle shafts, and their pigmentation have very different molecular underpinnings (Kopp 2011). Epithelial rotation in particular is a complex morphogenetic process involving localized convergent extension and must require numerous genes involved in cell polarity, cell adhesion, and cytoskeletal dynamics (Atallah et al. 2009a; Tanaka et al. 2009). What types of genetic changes could allow this entire suite of processes to re-evolve following an earlier loss (Barmina and Kopp 2007; Matsuda et al. 2009)?
We suggest that violations of Dollo’s law can be made more likely by the modular organization of developmental pathways. Many pathways have “nexus” regulatory genes that activate multiple downstream targets responsible for different cellular processes. The loss of a complex trait can happen easily if the expression of the nexus gene in the progenitor tissue is disrupted, since the entire downstream developmental program will be lost automatically. In fact, rapid loss of morphological structures by loss-of-function mutations in a single regulatory gene has been documented in several cases (McGregor et al. 2007; Chan et al. 2010). The same property of developmental pathways could also explain the re-evolution of lost traits: as long as the downstream pathway remains intact, regaining the expression of the nexus gene or genes in the progenitor tissue will be sufficient to restore much of the original structure. For example, several lineages of swordtail fish (Xiphophorus) have secondarily lost the male swords (extended tail fins) (Meyer et al. 1994, 2006). In at least some of these species, small vestigial “swordlets” can be restored by exposure to abnormally high levels of testosterone (Gordon et al. 1943). In X. maculatus, a swordless species, a single mutation is sufficient to form a similar (although not identical) fin extension, presumably by enhancing cell response to endogenous testosterone (Offen et al. 2008). These observations suggest that the pathway responsible for sword development has decayed only partially following the loss of the sword and can be brought back by changes in a relatively small number of genes. On a much deeper timescale, teeth are absent in all modern birds; however, a single mutation in the chicken talpid2 gene can partially restore tooth development, inducing integumentary outgrowths that resemble crocodilian teeth (Harris et al. 2006). Of course, this is only possible because some of the regulatory landscape that controls odontogenesis has been retained in avian oral tissues (Chen et al. 2000; Mitsiadis et al. 2006).
The question, then, is, why would the downstream pathway (i.e., a large set of regulatory interactions among genes) stay intact and not succumb to mutation accumulation in the absence of selection on the defunct structure? One possible explanation is that the vast majority of genes have pleiotropic functions. In the simplest case, the coding sequence of a gene whose expression has been lost in one tissue will remain under purifying selection as long as it continues to be expressed in other tissues (Marshall et al. 1994; Collin and Miglietta 2008). More generally, the modular organization of development suggests that pleiotropy can protect entire pathways: selection pressure on most regulatory interactions may still be present if the pathway acts in other tissues. This mechanism may be particularly important for structures that have serial homologs, which share largely the same developmental programs. For example, re-evolution of lost digits in lizards and mandibular teeth in frogs is likely to be enabled by the fact that other digits, and other teeth, have always been retained (Brandley et al. 2008; Kohlsdorf et al. 2010; Wiens 2011).
Even if selection does not preserve the entire pathway, preservation of its component modules may be enough to retain capacity for re-evolution. In Drosophila, the cellular module responsible for making thickened, rounded, darkly pigmented sex comb teeth may also be deployed to make similar bristles in other body parts such as male genitalia; the module responsible for sex comb rotation may act in other epithelial sheets that undergo convergent extension, and so on. If selection on pleiotropically linked traits preserves most regulatory interactions from mutational decay, the pathway as a whole could remain largely intact and require changes in only a few genes to restore it to a modestly functional state that can then be refined by additional mutations. The genetic architecture of sex comb morphology in D. bipectinata vs. D. malerkotliana could conceivably be the result of such a process.
Understanding the molecular mechanisms that enable re-evolution of complex traits will require a developmental genetics perspective; QTL mapping is only the first step in this direction. Although the limitations imposed by fixed chromosomal inversions preclude us from identifying the major gene or genes responsible for the re-evolution of large rotating sex combs in D. bipectinata by linkage mapping, our growing understanding of sex comb development may ultimately allow us to overcome these limitations.
We are grateful to Muneo Matsuda for his help in analyzing chromosome order in the bipectinata species complex; to Y. Fuyama and the San Diego Drosophila Species Stock Center for providing fly strains; and to Rachael Curtis, Andrew Hamilton, and Mary Magsombol for help with phenotypic analysis and DNA extractions. Illumina sequencing was performed at the University of California Davis Genome Center. Genotyping was performed at the University of California Davis Veterinary Genetics Laboratory. This work was supported by National Institutes of Health grant 5-R01GM082843-02.
Communicating editor: C. D. Jones
- Received August 30, 2012.
- Accepted September 28, 2012.
- Copyright © 2012 by the Genetics Society of America