The Evolution of the Drosophila Sex-Determination Pathway
Andrew Pomiankowski, Rolf Nöthiger, Adam Wilkins


The molecular complexity of the Drosophila somatic sex-determination pathway poses formidable intellectual challenges for attempts to explain its evolutionary origins. Here we present a reconstruction of how this regulatory cascade might have evolved in a step-by-step fashion. We illustrate how mutations in genes, which were already part of the pathway or were recruited as new regulators of the pathway, were favored by sexual selection acting on the discriminatory sex-determining signal. This allows us to explain the major features of the pathway, including multiple promoter sites, alternative splicing patterns, autoregulation, and stop codons. Our hypothesis is built on the available data from Drosophila and other insect species, and we point out where it is amenable to further experimental and comparative tests.

CONSIDERING the seemingly simple task that the Drosophila sex-determination pathway performs, namely the production of males and females, it exhibits a remarkable degree of complexity (reviewed in Schutt and Nöthiger 2000). A primary signal, which in itself consists of several interacting gene products, initiates an intricately integrated multi-genic cascade whose components show a variety of regulatory mechanisms. These include multiple promoter sites, alternative splicing patterns, autoregulation, and the presence of stop codons. Comparative analyses of Dipteran relatives of Drosophila and more distantly related species reveal further structural and regulatory complexity, as the sex-determination mechanisms employ some of the same genes but frequently in different ways along with a suite of other control elements.

Although there is a voluminous literature on the diversity of sex-determination systems in general and even within particular phylogenetic groups such as the Diptera (see, for instance, Bull 1983; Marin and Baker 1998), comparatively little effort has been given to considering how specific sex-determination systems may have evolved. The early theoretical literature on the evolution of sex determination is extensive (see Bull 1983 for a review) but has not proved very illuminating, principally because it is too abstract, concentrating on conditions for the spread of hypothetical “modifiers.” The classical literature, in the virtual absence of genetic and molecular information, was unable to address how real genetic networks were constructed. Today, however, our knowledge about the genetic and molecular basis of several sex-determination systems (Cline and Meyer 1996; Schutt and Nöthiger 2000) renders an evolutionary approach both possible and desirable to make sense of the evident variety and complexity of these systems. A few analyses have made a start in this direction, focusing on the known sex-determination pathways in insects and nematodes and their possible evolutionary origins (Nöthiger and Steinmann-Zwicky 1985; Hodgkin 1992; Wilkins 1995; Raymondet al. 1998). In particular, Nöthiger and Steinmann-Zwicky (1985) showed how the multiplicity of insect sex-determination systems might, in principle, reflect diversity only in the most upstream switches of the pathway, while Wilkins (1995) argued that the long Caenorhabditis elegans pathway might have grown by successive addition of upstream control elements to an ancient conserved downstream module.

None of the previous discussions, however, have explained the complexity seen in sex-determination pathways in terms of evolutionary dynamics and selectional forces. Our goal is to redress this gap through a hypothetical reconstruction of the main evolutionary steps that led to the Drosophila sex-determination system. We have chosen to concentrate on Drosophila as this is the best characterized of all the sex-determination pathways and, by virtue of its complexity, provides a challenging test of our general approach. In addition, there is now considerable knowledge about sex determination in other Dipteran insects, which permits informative comparisons and the inference of ancestral states. Our focus is on the underlying genetic events, rather than on morphological or developmental change. By specifying (a) the order in which genes were added to the pathway and (b) the selective reasons for their recruitment, our reconstruction can be broken down into a series of hypotheses, many of which can be tested via comparative studies with other species.

Our wider aim is to develop a framework to study the evolution of the apparently quite different mechanisms of sex determination seen among Diptera (e.g., Drosophila, Ceratitis, Musca, and Sciara), other insects (e.g., Lepidoptera and Hymenoptera), and beyond (e.g., nematodes and mammals). The burgeoning wealth of data on the sex-determination pathways of animal species should serve to facilitate both the formulation of new hypotheses and the testing of ideas, including the ones we propose here.

A major point in the scheme to be described is that it relies on sexual selection as a principal motor for evolutionary change in sex-determining systems. Sexual selection is known to be a strong and temporally variable selective force that has contributed to the exaggeration and diversity of secondary sexual characters involved in courtship display and mating success (Darwin 1871; Andersson 1994). It is not commonly appreciated, however, that sexual selection can also act on the primary mechanism that determines sex. The differences in behavior, physiology, morphology, etc., that affect sex-specific fitness arise from differential gene expression, which is set up and maintained by the sex-determining gene network. Hence, genetic variation in the sex-determining mechanism, even in the primary sexual signal (e.g., in the strength or timing of this signal), will sometimes have consequences for sexual fitness. We show how this can occasionally lead to major transitions in the sex-determining mechanism, such as the recruitment of new elements, changes in heterogamety, new promoter regions, or alternative splicing.

We begin with a hypothetical ancestral state, from which we derive, through a series of mutational changes, the current system of Drosophila melanogaster. The reconstruction that we outline shows how a sequence of individually conventional mutational changes could have generated the pathway that determines somatic sex in Drosophila today. We will begin with a short review of the contemporary system and then proceed to our conceptual reconstruction of its possible evolution from a much simpler ancestral state.


The basic features of the somatic sex-determination cascade in D. melanogaster are outlined in Figure 1 (for a complete description, see Schutt and Nöthiger 2000). Many other genes are essential to the pathway, but these are equally expressed in both sexes and thus have no discriminatory role. The primary genetic signal is provided by the ratio of X-linked numerator genes [three sisterless genes (sisA, sisB, and sisC) and runt (run)] to one major autosomal denominator gene, deadpan (dpn). In females, with two X chromosomes, this X:A ratio is 2:2 while in males, which carry only one X, it is 1:2. The products of these genes are transcription factors that regulate the expression of Sex-lethal (Sxl). This gene is unusual in having a stop codon (UAG) embedded in exon 3. Exon 3, therefore, has to be removed during RNA processing for transcripts to produce functional SXL protein. In females, the products of two doses of the X-linked numerators activate the early promoter of Sxl (Pe) shortly after fertilization at the cellular blastoderm stage (Esteset al. 1995). The Pe promoter produces RNA transcripts from which exons 2 and 3 are constitutively spliced out, resulting in an early burst of active SXL protein. Pe, however, is only transiently active between embryonic cleavage cell cycles 12 and 14 and is quickly replaced by the maintenance promoter Pm, which is active in both sexes and is not regulated by the numerator and denominator transcription factors. Transcripts from this promoter do not undergo constitutive excision of exons 2 and 3. Nevertheless, in females, exon 3 is spliced out of Sxl primary transcripts because the SXL protein that was initially produced from the Pe promoter can bind to its own pre-mRNA and, as a splice enhancer, enforces the elimination of exon 3. This establishes an autoregulatory loop, which maintains itself throughout development. Females, therefore, continue to produce SXL protein. In contrast, there is no alternative splicing in males because they do not produce the initial burst of SXL protein from Pe, which cannot be activated by only a single dose of the numerators. Hence in males, the autoregulatory loop is never established. Male Sxl transcripts produced from the late Pm promoter retain exon 3, and this results in premature termination of translation and absence of functional SXL protein.

Sxl codes for an RNA-binding protein that regulates production of not only its own transcripts but also those of transformer (tra), the next gene in the sex-determination pathway (Figure 1). Like Sxl, tra produces transcripts that contain several stop codons at the beginning of exon 2. In females, SXL protein blocks the canonical splice site and forces use of a cryptic splice site just downstream of the stop codons. This creates an open reading frame, which now allows the production of active TRA protein. In males, however, the absence of SXL results in mRNAs that retain the stop codons in exon 2, which leads to premature termination of translation and absence of any functional TRA protein. tra codes for another RNA-binding protein that causes alternative splicing of doublesex (dsx), the next downstream element in the pathway. In males, the absence of TRA protein results in the default splice of dsx transcripts and the loss of exon 4. Hence male dsx mRNA contains exons 1–3 and 5–6. This produces the male-specific DSXM isoform. In females, in contrast, the presence of TRA protein, together with the cofactor TRA2, initiates an alternative splicing pattern, which includes and terminates with exon 4. Thus female dsx mRNA contains exons 1–4 and produces the female-specific DSXF isoform. Most somatic sexual characters are differentially determined by the two dsx proteins. These act as transcription factors that sex-specifically enhance or repress a number of downstream male- and female-specific genes, which implement the two different routes of sexual differentiation (Christiansenet al. 2002).

Figure 1.

—The genetic cascade regulating somatic sexual development in Drosophila today. The female pathway is on the left; the male pathway is on the right. Open gene symbols indicate functional inactivity. The gene tra2 is active in both sexes, but is required only in the female soma. Vertical arrows show the flow of information. The genes forming the primary signal (N, numerators; D, denominators) and dsx encode transcription factors; Sxl, tra, and tra2 encode proteins involved in RNA processing. Boxes represent exons; for Sxl, only the first 5 exons (of 10) are drawn. Bars in exon 3 of Sxl and in exon 2 of tra symbolize stop codons.

Less is understood about the genetic basis of sex determination in related insects, but it is clear that there is both evolutionary conservation and divergence. A significant conserved element is the last regulatory gene in the cascade, dsx. The transcripts of dsx are alternatively spliced in males and females in several insects, including other Diptera such as Bactrocera tryoni (Shearman and Frommer 1998), Ceratitis capitata (Paneet al. 2002), Megaselia scalaris (Sievertet al. 1997), and Musca domestica (Dubendorferet al. 2002) and, more distantly, the Lepidopteran Bombyx mori (Ohbayashiet al. 2001). Sex-determination genes with strong sequence similarity to dsx in the regions encoding their DNA-binding domains have also been found in a diverse range of still more distantly related animals from nematode worms to mammals (Raymond et al. 1998, 2000). This phylogenetic conservation suggests an ancient origin and role for dsx in sex determination.

Other genes in the Drosophila pathway appear to be more recent additions. tra is thought to be a common factor in Dipteran insects as dsx undergoes sex-specific alternative splicing in several species of this group (Sievertet al. 1997; Shearman and Frommer 1998; Dubendorferet al. 2002; Paneet al. 2002; Hedigeret al. 2004). This was confirmed for the fruit fly C. capitata, in which TRA protein causes sex-specific alternative splicing of dsx (Paneet al. 2002). Recently, a gene with molecular similarity to tra has been identified outside the Diptera in the honeybee Apis mellifera (Beyeet al. 2003), where again it appears to be involved in dsx splicing. However, tra is unlikely to be involved in the Lepidoptera, as dsx in the silkworm B. mori lacks TRA-binding sites and has a reversed pattern of alternative splicing with the default splice in females (Ohbayashiet al. 2001; Suzukiet al. 2001). Outside the insects, there is no evidence for the involvement of tra in sex determination in any species. For instance, the dsx homolog mab-3 of the nematode C. elegans is under completely different genetic control (Shen and Hodgkin 1988). This suggests that the role of tra in sex determination is limited to certain insect groups. The involvement of Sxl in sex determination appears to be even more restricted than that of tra, being limited to the genus Drosophila (Boppet al. 1996; Penalvaet al. 1996). In Chrysomya rufifacies (Muller-Holtkamp 1995), M. scalaris (Sievertet al. 1997), M. domestica (Meiseet al. 1998), and C. capitata (Sacconeet al. 1998), Sxl is equally expressed in both sexes and thus cannot act as a discriminator of sex (the function of Sxl in these Dipteran species is not known). A similar limitation to the genus Drosophila is likely to pertain to the numerator and denominator genes involved in sex determination (Erickson and Cline 1998).

The general picture is diversity among upstream and conservation among downstream control genes, as proposed by Wilkins (1995). The Drosophila cascade has apparently seen the repeated recruitment of upstream elements as regulators of the sex-determination pathway. To envisage how this could have happened, we have reconstructed the evolution of the current genetic system of sex determination in Drosophila from an ancestral state that had dsx as the discriminatory signal of the pathway. This seems a reasonable assumption, given that dsx is involved in sex determination in all insect species examined, whereas tra, Sxl, and the numerator/denominator genes are more phylogenetically restricted. In our reconstruction, we have concentrated on explaining the evolution of the pathway from the X:A ratio to dsx for somatic sex determination. The evolution of germline sex determination and dosage compensation has been left for the future (see discussion). Our efforts have been guided by the extensive experimental knowledge of gene expression in Drosophila and related insect species. Many parts of the evolutionary sequence that we are proposing can, in principle, be tested by future comparative studies.

For the special case of Drosophila, we begin with a putative ancestral sex-determination pathway and postulate a degree of weakness in it that could not be “solved” by simple changes in gene expression of the component elements. We then show how a single genetic change could have rectified the initial weakness while simultaneously laying the ground for further change. The complete sequence of changes that we propose to account for the present-day Drosophila sex-determination system involves both the addition of new control elements to the gene network and major changes to the component genes.


The ancestral state

We assume that control of sex determination in the ancestor of Drosophila was through heterogamety at the dsx locus. In the proposed ancestral state, males were heterogametic dsxM/dsx+, and females were homogametic dsx+/dsx+. A tra+ allele was at fixation and produced TRA protein equally in both sexes. TRA acted as a female splice enhancer of dsx+ transcripts, as occurs today in conjunction with its cofactor TRA2 (Schutt and Nöthiger 2000), but had no effect on dsxM transcripts. Hence, dsx+/dsx+ individuals generated transcripts that followed the female splice to give DSXF and a female phenotype. dsxM/dsx+ individuals produced a greater amount of DSXM and a male phenotype. In this scheme, the dsxM allele behaves as a dominant, generating transcripts that exclusively follow the male splice. dsxM mutants with these characteristics have been reported in D. melanogaster (Baker and Wolfner 1988; Nagoshi and Baker 1990). It has also been shown that Drosophila with a surplus of DSXM over DSXF (XX flies with two copies of dsxM plus one of dsx+) are phenotypically male (Nöthiger and Leutholdet al. 1987).

In addition to DSXM, the dsxM/dsx+ genotype produced DSXF from its dsx+ allele, thereby potentially reducing the strength of the male-determining signal in this genotype. We hypothesize that this ambiguity in dsx expression was the main selective force that led to the first expansion of the pathway with the evolution of tra as a differential upstream regulator of dsx. Mutant forms of dsx+ with less expression and hence less DSXF production would have been favored in males. But such mutants would have reduced DSXF production in females and would have been disadvantageous in this sex. The sharing of gene expression across the sexes limited the possible improvement of sex-specific adaptation by mutational change in the dsx gene (Rice 1984, 1998).

Conversion of tra to an upstream regulator of dsx

Two features of the contemporary tra gene need to be accounted for: first, tra carries a stop codon in exon 2, and second, this part of the exon is skipped in females. In our model, these features evolved in two steps, each resulting in an increase in the distinctiveness of the two sexes. In the first stage, a mutation occurred in the tra+ allele, creating a stop codon (UAG) in exon 2. This mutation, which we call traS, caused premature termination of translation of tra transcripts and production of a truncated and inactive form of the TRA protein, similar to that seen in D. melanogaster today. The traS mutation was beneficial in males (traS/tra+; dsxM/dsx+) because less TRA protein lowered the efficiency of the female splice of dsx+ transcripts and hence increased the production of DSXM relative to DSXF (Table 1). Conversely, the reduction in DSXF in mutant females (traS/tra+; dsx+/dsx+) was disadvantageous as the female splice was less efficient and these females suffered a reduction in DSXF, possibly even accompanied by production of some DSXM. Note that this reconstruction is consistent with the observation that flies heterozygous for a null allele of tra are not sex reversed but female in D. melanogaster (Sturtevant 1945).

The balance of benefit to males and harm to females determined the fate of the traS mutant. This can be followed by assigning fitness values to genotypes (Table 1) and allowing evolution using a standard population genetic simulation (Figure 2; see appendix for details). When rare, the traS mutant invades if the gain in male heterozygote (traS/tra+; dsx+/dsxM) fitness is greater than the loss in female heterozygote (traS/tra+; dsx+/dsx+) fitness (i.e., m > f in Table 1). The mating of male and female mutant heterozygotes generates traS homozygous males, which produce no TRA protein and lack any female splice of dsx+ transcripts and so are male irrespective of their dsx genotype.

Our simulations show that if the fitness of traS homozygous males was greater than or equal to the fitness of traS heterozygous males (i.e., m2 ≥ m), then the traS allele would rise to a frequency of 0.75 (Figure 2). This causes elimination of the dsxM allele and fixation of the dsx+ allele (Figure 2). As a result, the tra gene becomes the upstream regulator of dsx, with females being heterozygous (tra+/traS) and males being homozygous (traS/traS). Thus, the discriminating signal moved one gene upward, from dsx to tra, with a simultaneous reversal in heterogamety from male to female. Our hypothesis shows how a stop mutation in tra ameliorated the problem that arose because the main sex-determining products, DSXM and DSXF, initially were both present in males.

View this table:

Fitness of traS stop mutants

In D. melanogaster, dsx and tra are linked, and recombination is absent from males. Our simulations show that the same evolutionary transition (i.e., tra as the new upstream regulator with the loss of dsxM) occurs if the traS mutation arises in linkage with dsx+, although the conditions for spread are more restrictive than those with free recombination (see appendix for details). In contrast, if the traS mutation occurs in linkage with dsxM, it is limited to males, as dsxM is a dominant masculinizer. The traS dsxM mutant spreads as it reduces the amount of TRA produced by males. But due to the lack of male recombination, the traS mutant cannot cause the elimination of the dsxM allele or the recruitment of tra as an upstream discriminatory regulator of sex determination (at equilibrium, females remain tra+ dsx+ homozygotes and males become traS dsxM/tra+ dsx+ heterozygotes).

Figure 2.

—The spread of the traS mutant is shown (broken line) from introduction at low frequency (0.01) to its equilibrium frequency (0.75). This leads to the loss of the dsxM allele (solid line). The traS allele is set to have a selective advantage in males of m = m2 = 0.03, and a selective disadvantage in females of f = 0.015. The initial frequency and exact values of the selection coefficients do not alter the outcome as long as m = m2 > f. In the simulation shown, the dsx and tra loci were assumed to be unlinked and to undergo free recombination in both sexes (see the appendix for the case of linkage).

Recruitment of Sxl

The evolution of traS, although of overall benefit, itself caused a problem. Females now produced TRA from a single tra+ allele rather than from two copies and were less efficient in splicing dsx+ transcripts in the female mode and may even have produced some DSXM. To some extent, these disadvantages could have been counteracted by selection for higher expression of the single tra+ allele. However, we know that evolution took a different path, which led to the recruitment of Sxl.

Sxl is a general RNA-binding protein that has multiple roles in RNA processing and translation suppression (Kelly et al. 1997; Gebaueret al. 2003). In insects other than Drosophila, however, Sxl is not involved in somatic sex determination, as discussed previously. Hence, we make the parsimonious assumption that the ancestral version Sxl+ was likewise not involved and had no, or at least no significant, binding affinity for tra+ or traS transcripts. In the lineage leading to Drosophila, we assume that the traS allele had two important preexisting features: a poly(U)-binding site in the intron upstream of exon 2 and a cryptic splice acceptor site downstream of the stop codon in exon 2. Both of these are present in the tra allele of D. melanogaster (Sosnowskiet al. 1989) and C. capitata (Paneet al. 2002). Neither of them would have had any consequences in the ancestral sex-determination system.

View this table:

Fitness of SxlF mutants capable of splicing traS

We propose that the first step in the recruitment of Sxl was the occurrence of a mutant allele, which we designate Sxl F, with affinity for the poly(U)-rich binding site in traS. The binding of SXLF protein to traS transcripts blocked this canonical splice acceptor site at the start of exon 2 and forced acceptance of the downstream cryptic splice site by the splicing machinery. The resulting removal of the stop codon from traS transcripts converted this null allele into one that produced active TRA protein. We assume that this TRA protein, although somewhat shorter at its N terminus, retained normal activity, as the product of the contemporary Drosophila tra gene does.

The SxlF mutant in heterozygous form (Sxl F/Sxl+) had the following consequences for sexual phenotype and fitness (Table 2). In tra+/traS female heterozygotes, the splicing-mediated conversion of the traS allele from a null to a functional form allowed both tra alleles to produce TRA protein. In traS/traS homozygotes, both alleles also produced TRA protein, causing feminization, given that the amount of TRA was similar to that found in Sxl F/Sxl+; tra+/traS heterozygotes. This is a reasonable assumption as heterozygotes for null alleles of Sxl in D. melanogaster, where both tra alleles carry stop codons, are female.

The SxlF allele spread if selection favored an increase in TRA production in females (i.e., g, g2 > 0, Table 2). But owing to the pattern of inheritance, this caused a large increase only of traS homozygous females. The reason for this is that Sxl F/Sxl+; traS/tra+ females generate equal numbers of Sxl F/Sxl+; traS/tra+ and SxlF/Sxl+; traS/traS daughters, whereas SxlF/Sxl+; traS/traS females produce only SxlF/Sxl+; traS/traS female offspring. The spread of the SxlF allele thus caused the elimination of the tra+ allele and fixation of traS, resulting in a population of Sxl+/Sxl+; traS/traS males producing no TRA and of Sxl F/Sxl+; traS/traS females producing TRA from both traS alleles. Assuming that Sxl and tra were unlinked, as they are in D. melanogaster, our simulations show that the frequency of SxlF rises to 0.25.

The addition of SxlF as an upstream regulator of tra boosted female fitness at no cost to males. Females remained the heterogametic sex. We note, however, that Sxl is a poor splice enhancer of tra in D. melanogaster, with over half of tra mRNAs in females retaining the stop codon in exon 2 (Beloteet al. 1989). Thus, females would have benefited from two doses of Sxl. But this genotype could not arise in this system, because Sxl F heterozygotes were females, as the equivalent genotype is in D. melanogaster today (Cline and Meyer 1996). Thus homozygous SxlF individuals could not have been generated. This imbalance led to the next evolutionary step.

Addition of a stop codon in Sxl

In D. melanogaster, Sxl, like tra, has an exonic stop codon (in exon 3), which is skipped in females, but prevents production of SXL protein in males (see Figure 1). Sex-specific alternative splicing and stop codons are absent from copies of Sxl in other insects in which Sxl has no role in sex determination (Muller-Holtkamp 1995; Sievertet al. 1997; Meiseet al. 1998; Sacconeet al. 1998). This implies that a stop codon mutation was added to Sxl in the lineage leading to Drosophila. We assume this occurred after Sxl F was recruited to the sex-determination pathway. Our hypothesis is that the stop mutation, which we call Sxl FS, initially promoted female fitness. This effect is paradoxical at first glance, as a stop mutation is expected to reduce protein production. But, as we will see, the ability of Sxl to autoregulate by self-splicing allowed the evolution of an allele with stop codons, which resulted in SXL protein being produced from both copies of Sxl in females and by neither copy in males.

View this table:

Fitness of SxlFS stop mutants

We make the key assumption that Sxl autoregulation arose prior to the origin of the stop mutation. Specifically, the mutation in Sxl F that allowed recognition of tra poly(U)-rich sequences also permitted the recognition of poly(U)-rich sequences within SxlF transcripts. Such poly(U)-rich tracts occur in D. melanogaster in both introns that neighbor the exon with the stop codons. These attract SXL binding and the splicing out of this exon in Sxl transcripts (Horabin and Schedl 1993). We propose that the ancestral Sxl+ allele lacked these poly(U)-rich sequences as autoregulation of Sxl is not known outside the Drosophilids. In the Drosophila lineage, these sequences were introduced and amplified, perhaps by replication slippage or unequal recombination, after the SxlF allele had evolved. This hypothesis can be tested by examination of the genomic sequence of Sxl in non-Drosophilids. While the exact timing of the proposed event is not crucial, we must postulate that some poly(U) sequences existed in Sxl F and allowed self-splicing of exon 3 from Sxl F transcripts prior to the origin of the Sxl stop mutation. The introduction of autoregulation presumably was advantageous to females carrying the SxlF allele, but made no distinct change to the sex-determining system.

Consider the effect of a stop mutation Sxl FS. The mutation arose in the Sxl F allele in a female, which then generated Sxl+/Sxl FS heterozygotes (Table 3). These genotypes produced no SXLF protein as SxlFS transcripts carry the stop codon and so were converted into males. But when mated to Sxl+/Sxl F females, some of the progeny were Sxl F/Sxl FS heterozygotes. These female individuals were favored by selection as they produced SXLF protein from both alleles. SXLF was constitutively produced from the SxlF allele and this protein spliced out the stop-containing exon from Sxl FS transcripts. In turn, these females could have mated with Sxl+/Sxl FS males to generate Sxl FS homozygotes, which produced only transcripts carrying the stop codon. These individuals were male. Selection in favor of Sxl FS in females (Table 3) led to its spread, replacing Sxl+ and resulting in an equilibrium frequency of 0.75 with Sxl F/Sxl FS females and SxlFS/SxlFS males.

It is worth emphasizing the counterintuitive nature of the Sxl FS stop mutation. Its spread boosts SXLF production in females, now from both Sxl alleles, without any cost to males, who continue to produce only non-functional Sxl transcripts. The spread of Sxl FS again causes no change in heterogamety, with females remaining the heterogametic sex. Note that we assume that the original version of Sxl+ did not have the ability to autoregulate through self-splicing. If it did, a stop mutation would not have conferred any benefit, as Sxl+/Sxl FS mutants would have produced SXLF protein and thus have been female. In this situation, homozygous Sxl FS males could not have been generated.

The early promoter of Sxl and the recruitment of sis

In D. melanogaster today, all copies of Sxl carry a stop codon in exon 3 (Figure 1). In females, the stop codon is removed, allowing the production of SXL protein. This process is initiated by the turning on of the early promoter (Pe)of Sxl in females due to a twofold higher dose of the X-linked numerator genes sisA, sisB, sisC, and run. We envisage that the early promoter evolved first, followed by the recruitment of a numerator gene as an upstream control element. Once this system of genetic control was established, further numerators were added.

We can reconstruct these evolutionary steps by considering a mutation arising in the first intron of SxlFS, which generated a new early promoter site, Pe. We designate this allele Sxl FSPe. For simplicity, we consider a single numerator gene, sis; the fact that several sis genes are required today does not demand a qualitatively different argument. Since the numerator genes of D. melanogaster have developmental roles other than sex determination (Cline and Meyer 1996; Walkeret al. 2000), we assume that the sis+ allele was originally at fixation and acted as a general transcription factor with expression early in development. We further assume that the SIS protein also bound to the Pe promoter, leading to the early production of Sxl FSPe transcripts. As today, mRNAs deriving from the Pe promoter lacked exons 2 and 3, which led to an early burst of SXLF protein. This was sufficient to activate the autoregulatory loop and to maintain the production of SXLF protein when the maintenance promoter, Pm, took over.

View this table:

Fitness of the early promoter Pe of Sxl

The Sxl FSPe allele was initially a heterozygote with either Sxl F or Sxl FS. These genotypes received an early burst of SXLF protein from the Sxl FSPe allele and later produced SXLF from both alleles when Pm took over (Table 4). Both genotypes were female. Selection favored them as they produced higher titers of SXLF protein. As males were SxlFS homozygotes, the SxlFSPe allele was thereafter found only in Sxl FSPe/SxlFS heterozygous females. Given that selection favored these females (i.e., j > 0, Table 4), the SxlFSPe allele spread and replaced SxlF, the allele that did not contain a stop codon. After this point, females were SxlFSPe/Sxl FS heterozygotes and males were SxlFS homozygotes. We note that any mutation to create a Pe promoter in the Sxl F allele would have been selectively neutral, as the Sxl F allele already produced SXLF protein from the Pm promoter. For selection to have favored Pe, it must have occurred in an allele that already contained a stop codon.

We can now see how sis alleles were recruited as regulators of Sxl at the top of the sex-determination pathway. We assume that the sis locus was linked to the same chromosome as Sxl and that there was no recombination in males, as seen today in D. melanogaster. The effect of sis alleles is dose dependent in D. melanogaster (Cline 1993), so we further assume that sis- null mutants failed to activate the early promoter Pe and so acted as dominant masculinizers (Table 5). The original sis- null mutant could have arisen in linkage with either the Sxl FSPe or the SxlFS allele. For simplicity, we consider a mutation in linkage with SxlFSPe, which occurred in the germline of a female (see appendix for the alternative pattern of linkage). This mutant chromosome produced a sis- Sxl FSPe/sis+ SxlFS male because a single dose of sis+ does not activate Pe. Matings of this male with standard sis+ Sxl FSPe/sis+ SxlFS females gave rise to sis- Sxl FSPe/sis+ SxlFSPe offspring, which were males homozygous for SxlFSPe. Further “backcrosses” of these males with nonmutant females in the F3 and in later generations produced females homozygous for sis+ SxlFSPe. Assuming that two active early promoters were advantageous to females, because they established the autoregulatory loop more quickly or with greater reliability (k > 0, Table 5), the Sxl FSPe allele could spread. In turn, this selected for the sis- allele because of its male-determining activity when combined with Sxl FSPe homozygosity.

The final system consisted of females that were sis+ Sxl FSPe/sis+ SxlFSPe and males that were sis- SxlFSPe/sis+ Sxl FSPe. The Sxl FS allele was lost, the Sxl FSPe allele went to fixation, and the sis locus took over as the upstream regulator of sex determination. This set the stage for the last step, the degeneration of the chromosome carrying sis- to become the Y (see discussion). The sis- allele constitutes a second paradoxical mutation in our scheme; sis- initially promoted the development of one sex (male), but ultimately contributed to improving the fitness of the other sex (female).

To sum up, our model proposes that the major features of Sxl as seen in Drosophila sex determination evolved in the following order: the ability of SXL protein to bind poly(U) in tra, Sxl autoregulation, the addition of a stop codon, followed by that of the early promoter, and then the recruitment of a sis null as an upstream regulator of Sxl. This final step led to a change from female heterogamety to homogametic sis+/sis+ females and heterogametic sis-/sis+ males. All these changes served to limit and strengthen the autoregulatory loop of Sxl in females. The reliability of this signal was augmented by the recruitment of more numerator genes with mutant null alleles linked to the original sis null and of an autosomal denominator gene that acted as an antagonist to the numerators. The introduction of a denominator converted the analog system (more or less SIS protein) into a digital system (active or inactive SIS protein), thereby improving the reliability of Sxl regulation, i.e., “on” in females and “off” in males (Cline 1993).


Our reconstruction sets out a hypothesis for the evolution of the gene network that determines sex in Drosophila. We assume that the system evolved from an ancestral state in which dsx provided the discriminatory signal and a tra allele without stop codons facilitated female-specific splicing of dsx+ transcripts. There followed a series of gene recruitments and major transitions at the top of the pathway in the following order (Figure 3): (1) a stop codon in tra, which created a null allele; (2) the recruitment of Sxl as an RNA-binding factor, which caused alternative splicing of tra transcripts and the removal of the exon containing the stop codon; (3) Sxl autoregulation through the presence of SXL-protein-binding sites in Sxl transcripts; (4) a stop codon in Sxl, which created a null allele; (5) the presence of an early promoter of Sxl that was activated by binding of SIS transcription factor, leading to the early production of Sxl transcripts that lost the stop-containing exon; and (6) a null allele of sis, which caused dose-dependent activation of the early promoter of Sxl. Each of the transitions was favored because they increased the distinctiveness of gene expression in one sex or the other and hence the strength and reliability of the sex-determining signal. We postulate that sexual selection was the fundamental force driving these evolutionary changes in the sex-determining mechanism, as sexual selection leads to differential effects on the two sexes and is known to be a strong force that can yield rapid evolutionary change (Andersson 1994).

View this table:

Fitness of sis null mutants

Figure 3.

—Sequence of events leading to the Drosophila sex-determining pathway. Alleles in boxes are shown in the order in which they are proposed to have arisen. Below the boxes are the genotypes (in italics) that determine the sex at a given phase of evolution: females on the left side, males on the right side. Vertical solid arrows mark the direct phylogenetic route as described in the text. Broken arrows pointing to the right indicate possible side branches, that is, alternatives that were not taken or were discarded (gene symbols in outlined type) or may have led to systems operating in other Diptera. N, numerators; D, denominators; Df, deficiency; Dp, duplication

In the course of developing our scheme, we considered many other possibilities, which were rejected. The main criteria used for rejection were incompatibility with current knowledge about the sex-determining genes in D. melanogaster and parsimony. For example, we postulate that a null allele of the original numerator gene was unable to activate the early promoter of Sxl. This follows from the observation that the numerator genes in D. melanogaster act in such a dose-dependent manner (Cline 1993). We therefore rejected the alternative idea that there was active recruitment of a sis+ allele to turn on the early promoter of Sxl.

We also rejected some hypotheses because they were implausible and required too many additional steps or failed to account for some aspects of the Drosophila pathway. For example, if the original SxlF mutant had been a recessive, SxlF/Sxl+ heterozygotes would have been male (traS/traS) or female (traS/tra+). Matings between these genotypes would have generated SxlF homozygotes that would have been female. Because SxlF homozygotes produce SXL protein from both alleles, we assume that they were favored by selection, and the spread of this allele would have resulted in the fixation of traS with SxlF/SxlF females and SxlF/Sxl+ males. However, it is difficult to see how the Sxl stop codon could subsequently have been favored. This allele would have been a dominant male determiner, but in heterozygous or homozygous condition it would not have contributed any obvious benefit to male fitness. It is likewise difficult to see how autoregulation could then have been favored. Finally, the hypothesis of SxlF recessivity is contradicted by the fact that an active Sxl allele today is dominant over a null allele; so we would have to find, in addition, an explanation for a change in dominance.

In our presentation, we reconstructed a direct and short route from dsx to Sxl and the numerator/denominator system of Drosophila today (Figure 3). It is conceivable, even likely, that evolution has tried alternative routes. For example, a tra null or a deficiency for tra would have improved the original situation just as well as did a stop codon in tra. But with a tra null, there would have been no chance to recruit a correcting upstream regulator that could have turned tra “on” in females and “off” in males. Similarly, Sxl F without poly(U)-binding sites would have helped females at that stage of evolution (Figure 3), but could not later have acquired autoregulation. A duplication of Sxl F would also have increased the amount of SXL protein; but without the early promoter, it could not have come under transcriptional control by sis.

The linear pathway of Drosophila as it exists today is probably the stem left over from a “shrub” whose many side branches have disappeared or led to the pathways now encountered in other insects. In our view, the pathway has gone through many “trials and errors,” with the outcome not being the optimal solution, but just the one that evolved through short-term advantage. Another area of uncertainty is the temporal sequence of the events underlying the evolution of the Drosophila sex-determining pathway. Were the steps evenly distributed or clumped in time? Did each step reach equilibrium before the next was initiated or did some changes occur simultaneously? For example, one can imagine that Sxl was recruited before its target traS allele had reached an equilibrium frequency of 0.75. It is beyond the scope of this article to deal with these possibilities and complications.

We have not considered the evolution of dosage compensation, which is also under the control of Sxl, through the repression in females of the male-essential gene msl2 (Kelleyet al. 1997). We believe that dosage compensation was added after the recruitment of Sxl and sis to the somatic sex-determination pathway. Dosage compensation is required once X and Y chromosomes have differentiated and the dose of X-linked genes needs to be equalized across the sexes (Charlesworth 1996). We propose that this occurred after the recruitment of Sxl and sis, both of which are X-linked in Drosophila today. The key was the functional loss of one or more of the numerator genes in males. As a dominant masculinizer, the chromosome carrying the sis- null allele was limited to males, in which recombination is reduced or absent; thus, deleterious mutations could have accumulated on this chromosome, leading to its gradual degeneration until it eventually became the Y (Lucchesi 1978). Its homolog carrying the sis+ allele became the X. The numerators themselves are transcription factors with important roles in processes other than sex determination; for example, sisA is involved in midgut formation (Walkeret al. 2000), and sisB has multiple roles in bristle formation and neurogenesis (for a review see Cline and Meyer 1996). Thus, there must have been selective pressure for upregulation of the single copy of these numerators and other vital genes on the X chromosome in males. In D. melanogaster, this is achieved through the binding of the msl complex to the single X in males, which causes enhanced transcription (for review see Lucchesi 1996). In females, this complex does not form because Sxl inhibits msl2, an essential component of the msl complex (Kelleyet al. 1997). We have not discussed this step in detail, but these ideas could be tested by looking at the phylogenetic distribution of sex-specific expression in the numerators, Sxl and msl genes.

A second area that we have neglected is germline sex determination. This is much less well understood than somatic sex determination. Interestingly, Sxl is essential to oogenesis, but not via its downstream somatic targets, tra and msl2. In additon, Sxl in the germline is not regulated by the numerator/denominator ratio (for review see Steinmann-Zwicky 1994). As for dosage compensation, we suspect that these germline-specific features involving Sxl followed the recruitment of Sxl to the somatic sex-determination pathway. We should also note that some of the genes of the somatic pathway of D. melanogaster have other sex-determination functions; for example, tra controls sex-specific expression of fru in the central nervous system (Heinrichset al. 1998). Again, we suspect that these are secondary adaptations and additions made once the basic pathway has been established. Future data from other insects may clarify this point.

In this article, we have proposed a step-by-step hypothesis for the evolution of the Drosophila sex-determination system from a hypothetical ancestral state. Comparative data were used to infer the ancestral state and the general order in which genes were added to the sex-determination pathway. In particular, comparative data support the assumption that dsx was ancestral, tra was added next, followed by Sxl and sis. In principle, we might have used an alternative approach. This would have been to deduce the evolutionary history from comparative analyses of related sex-determination systems, for example, those of other insects like B. mori, C. capitata, or M. domestica. The best-characterized is that of the Mediterranean fruit fly, Ceratitis (Paneet al. 2002), which shares many features with D. melanogaster (tra stop codons, tra control of dsx), but also differs in important aspects (no sex-determining role of Sxl or sis, dominant masculinizing factors, tra autoregulation, and possibly maternal initiation of tra autoregulation). We have not followed this approach because these other systems are far less well characterized, and it is difficult or impossible to infer the sequence of evolutionary events before we have more genetic and molecular information about sex determination in these related insect genera.

We have put forward our hypothesis and set out its assumptions so as to stimulate further research. Our evolutionary hypothesis is as simple as we could make it while being consistent with the known facts. Even though some or many of the details may well need to be refined in the light of subsequent comparative and experimental findings, we believe our general approach is a constructive one: proposing the order in which the major features of sex-determining genes were recruited and how selection favored these changes. This is an advance on an overly abstract evolutionary analysis that does not take into account the known facts about the gene networks involved. It also provides an informative perspective on the known developmental and molecular biology of Drosophila sex determination. This system is as complicated as it is not because of some intrinsic engineering constraints that require it to be so, but because it is a reflection of its evolutionary “bricolage” (Jacob 1977; Duboule and Wilkins 1998).


Simulation technique: Simulations were carried out to follow the fate of mutant alleles, introduced at low frequency in an infinite diploid population with non-overlapping generations. The sex of the possible genotypes was defined (Tables 1, 2, 3, 4, 5). The next step was to apply selection and then to allow random mating between the survivors and the production of offspring for the next generation. So the flow of events iterated was zygotes → selection → reproduction. We initially assumed that different loci were not linked and so underwent free recombination in both sexes. Simulations were run until equilibrium conditions were reached (defined as decreasing changes in allele frequency of <10-6/generation at all loci).

Initial conditions were set up by finding the equilibrium gene frequency when there was only allelic variation at the original locus. A mutant allele was then introduced at the second locus at low frequency (usually 10-3) and followed until equilibrium was reached. A range of values for the selective coefficients in Tables 1, 2, 3, 4, 5 (i.e., m, m2, f, g, g2, h, j, and k) was examined to identify the general patterns reported in the text. Only positive values of the selection coefficients were considered. For simplicity, no conditions of heterozygote advantage were investigated (i.e., m2 ≥ m in Table 1 and g2 ≥ g in Table 2).

Linkage: In D. melanogaster, dsx and tra are located on chromosome 3, and Sxl and sis are located on the X chromosome. Nothing is yet known about the linkage of tra and dsx,orof Sxl and sis, outside of the Drosophilids. In addition, in Drosophila and many, if not all, families of the higher Diptera, male meiosis is achiasmatic and recombination is thought to take place only in females (White 1973). In the following paragraph, we investigate how linkage and the absence of male recombination influenced the evolution of traS and sis- alleles.

For dsx-tra, the evolutionary outcome depends on the initial linkage of the traS mutation. If the traS mutation occurs in linkage with dsx+, then traS heterozygous males (traS dsx+/tra+ dsxM) and females (traS dsx+/tra+ dsx+) will form, and matings between them will generate traS homozygous males. When rare, the conditions for the spread of the traS mutant linked to dsx+ are approximately twice as restrictive as for the unlinked case. This can be explained as follows. When traS is linked to dsx+, it can never occur in physical linkage with the dsxM allele, as dsxM is limited to males and there is no recombination in males. This means that the traS dsx+ linked mutant is more likely to occur in females than in males. Specifically, when the traS mutant is rare, traS dsx+/tra+ dsxM males typically mate with standard females (tra+ dsx+/tra+ dsx+). The offspring of this mating that carry the traS mutation are always female. The net effect is that the traS allele is about twice as likely to be in a female and so suffers about twice as much negative selection (in females) as positive selection (in males). In contrast, when traS is unlinked, it is equally likely to occur in males as in females because traS/tra+; dsx+/dsxM males mated to standard females have offspring with the traS mutation that are equally likely to be males as females. Using the selective coefficients in Table 1, the condition for the spread of traS when linked to dsx+ is approximately m > 2f. The alternative linkage pattern for traS is with dsxM, which is discussed in the text.

For Sxl-sis, the evolutionary outcome is independent of the initial linkage of the sis- mutation. We discuss linkage of sis- to SxlFSPe in the text. Alternatively, this mutant may have occurred in linkage with the SxlFS allele. If we assume that this mutation arose in the germline of a female, then the offspring carrying the chromosome would have been male sis- SxlFS/sis+ Sxl FS. Matings of this male with standard sis+ SxlFSPe/sis+ SxlFS females gave rise to sis- SxlFS/sis+ Sxl FSPe, which were also male. Further “backcrosses” of these males with nonmutant females in the F3 and in later generations produced females homozygous for sis+ SxlFSPe, which are favored by selection (Table 5). This led to the spread of both the SxlFSPe and the sis- alleles. At equilibrium males were sis- Sxl FS/sis+ Sxl FSPe and females were sis+ Sxl FSPe/sis+ SxlFSPe. This differs from the case with original linkage to the SxlFSPe allele only in the retention of SxlFS in coupling with sis- in males. When this chromosome degenerates to become the Y, Sxl FS will have disappeared.


We thank Daniel Bopp for his fruitful discussion of our work. This interdisciplinary research was carried out through a number of meetings sponsored in part by the British Council and the Wissenschaftskolleg zu Berlin.


  • Communicating editor: T. Schüpbach

  • Received September 22, 2003.
  • Accepted December 10, 2003.


View Abstract