SUMMARY Despite renewed interest in the role of natural selection as a catalyst for the origin of species, the developmental and genetic basis of speciation remains poorly understood. Here we describe the genetics of Müllerian mimicry in Heliconius cydno and H. melpomene (Lepidoptera: Nymphalidae), sister species that recently diverged to mimic other Heliconius. This mimetic shift was a key step in their speciation, leading to pre- and postmating isolation. We identify 10 autosomal loci, half of which have major effects. At least eight appear to be homologous with genes known to control pattern differences within each species. Dominance has evolved under the influence of identifiable “modifier” loci rather than being a fixed characteristic of each locus. Epistasis is found at many levels: phenotypic interaction between specific pairs of genes, developmental canalization due to polygenic modifiers so that patterns are less sharply defined in hybrids, and overall fitness through ecological selection against nonmimetic hybrid genotypes. Most of the loci are clustered into two genomic regions or “supergenes,” suggesting color pattern evolution is constrained by preexisting linked elements that may have arisen via tandem duplication rather than having been assembled by natural selection. Linkage, modifiers, and epistasis affect the strength of mimicry as a barrier to gene flow between these naturally hybridizing species and may permit introgression in genomic regions unlinked to those under disruptive selection. Müllerian mimics in Heliconius use different genetic architectures to achieve the same mimetic patterns, implying few developmental constraints. Therefore, although developmental and genomic constraints undoubtedly influence the evolutionary process, their effects are probably not strong in comparison with natural selection.
The emerging field of evolutionary developmental biology seeks to explain changes in ontogeny that lead from altered genotype to altered phenotype, a process that has been called “developmental reprogramming” (Arthur 2000). It will provide a fundamental contribution to evolutionary theory if it yields answers to two major questions.
First, to what extent does development constrain or drive evolution? In other words, are there emergent properties of developmental reprogramming so that directionality in evolution is not the sole preserve of natural selection acting on random variation? Such “developmental drive” would act in conjunction with selection rather than in opposition, leading to beneficial but suboptimal evolution, so that detectable phylogenetic inertia should result (Arthur 2001). Does adaptive divergence typically proceed by the substitution of many genes of minor effect (Fisher 1930), or can genes of large effect contribute to adaptation and speciation (Orr and Coyne 1992; Coyne and Orr 1998)? Recent theory suggests that when natural selection optimizes quantitative traits, an exponential distribution of gene effects will become fixed, with many factors of small effect and a few of large effect (Orr 1998, 1999).
Genetic dissection of speciation may also provide answers to questions such as what roles do linkage, dominance, and epistasis play in adaptive speciation? Are these genetic constraints incidental properties of newly evolved genes, or do the constraints themselves evolve? For instance, tight linkage between functionally related loci may exist because of recent tandem duplication, or it may be adaptive, because clustering of loci into linked blocks can preserve associations between coadapted alleles (Noor et al. 2001; Rieseberg 2001). Epistasis in particular is fundamental to speciation, because it is the production of ecologically or intrinsically maladapted gene combinations in hybrids that causes reproductive isolation (Whitlock et al. 1995; Turelli and Orr 2000).
The genetics of mimicry is usually studied within an ecological genetics (microevolutionary) framework. However, divergence in mimicry of the butterflies Heliconius cydno and H. melpomene(Fig. 1A) has strongly affected mate preferences and has led to maladaptive nonmimetic hybrids (Mallet et al. 1998; Jiggins et al. 2001). The species also display female hybrid sterility (Naisbit et al. 2002) and reduced hybrid mating success (Naisbit et al. 2001), but these almost certainly evolved after initial divergence, whereas traits such as differences in microhabitat (Mallet and Gilbert 1995) and host plant use (Smiley 1978) provide only weak barriers to gene flow. Divergence in mimicry was therefore a key step in speciation of H. cydno and H. melpomene, and the developmental genes controlling color pattern differences are as much “speciation genes” as those for hybrid sterility in Drosophila (Orr and Presgraves 2000; Ting et al. 2000).
Here we investigate genes that determine differences in mimicry between H. cydno and H. melpomene to answer the following questions:
1What is the genetic architecture of mimicry and does it suggest developmental constraints? Are individual gene effects major or minor? What roles do epistasis, linkage, and dominance play in the evolution of color pattern?
2Are mimicry genes that contribute to speciation homologous to those used in mimetic shifts within each species?
3Do partners in Müllerian mimicry use homologous genetic variation to achieve the same patterns? If so, developmental constraints could be much more important in the evolution of mimicry (Goldschmidt 1945; Nijhout 1991) than classically believed.
MATERIALS AND METHODS
Crosses were performed in Gamboa, Panama between September 1999 and March 2000 using Heliconius cydno chioneus and H. melpomene rosina collected from nearby forest in Soberanía National Park. To obtain crosses, we isolated virgin females with older males. After mating, females were kept individually in 1 × 1 × 2-m outdoor insectaries and supplied with pollen sources (Lantana and Psiguria), artificial nectar (10% sugar solution), and Passiflora vines for oviposition. Eggs were collected daily, and caterpillars fed on new growth of Passiflora biflora.
We were able to obtain offspring only from crosses between male H. melpomene and female H. cydno due to strongly asymmetrical mate preferences. Sterility of F1 females conformed to Haldane's rule and prevented F2 crosses (Naisbit et al. 2002), so color pattern segregation was examined in backcrosses using fertile F1 males. Single gene control was inferred when a 1:1 ratio of distinct phenotypes segregated in the backcross to the parental species bearing the recessive form of the trait, and unless indicated in the results, all tests are for deviation from this 1:1 ratio. Where a continuous distribution of intermediate phenotypes was produced for any single pattern element, control was judged polygenic. Homology was inferred if gene effects and linkage were identical with loci previously described from interracial crosses within either species. Summaries of individual genotypes are given in Tables 1–3, but segregation ratios and recombination frequencies include data from 19 additional individuals (14 from backcrosses to H. cydno and 5 from backcrosses to H. melpomene) that could not be scored at all loci due to wing damage or failure to eclose fully.
Table 1. Genotypes produced without crossing-over in the backcross to H. melpomene (female H. melpomene× male F1)
Genes within square brackets are linked, with maternal H. melpomene alleles given first. Genes in parentheses are not expressed on that genetic background (K on N BN B, and J on Sb1 Sb1). A dash indicates an allele that cannot be determined due to dominance of the alternative allele. Counts are given as females/males.
[BG2 br][-G2 br][ybSb1Vf2 N B][ybSb1Vf2 N B](K yK y)(J1 J1)Ac- [BG2 br][-G2br][ybSb1Vf2 N B][ybSb1Vf2 N B](K yK y)(J1 J2)Ac- [BG2 br][-G2br][ybSb1Vf2N B][ybSb1Vf2 N B](K yK w)(J1 J1)Ac- [BG2 br][-G2 br][ybSb1Vf2 N B][ybSb1Vf2 N B](K yK w)(J1 J2)Ac-
[BG2 br][-G2 br][ybSb1Vf2 N B][Ybc Sb3Vf1N N]K yK wJ1 J1Ac-
[BG2 br][-G2 br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK wJ1 J2Ac-
[BG2 br][-G1Br][ybSb1Vf2 N B][ybSb1Vf2 N B](K yK y)(J1 J1)Ac- [BG2 br][-G1 Br][ybSb1Vf2 N B][ybSb1Vf2 N B](K yK y)(J1 J2)Ac- [BG2 br][-G1 Br][ybSb1Vf2 N B][ybSb1Vf2 N B](K yK w)(J1 J1)Ac- [BG2 br][-G1 Br][ybSb1Vf2 N B][ybSb1Vf2 N B](K yK w)(J1 J2)Ac-
[BG2 br][-G1 Br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK yJ1J1 Ac-
[BG2 br][-G1 Br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK yJ1 J2 Ac-
[BG2 br][-G1 Br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK wJ1 J1 Ac-
[BG2 br][-G1 Br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK wJ1 J2 Ac-
Table 2. Genotypes produced by crossing-over in the backcross to H. melpomene (female H. melpomene mated to male F1)
Conventions as in Table 1. Loci affected by crossing-over are shown in bold.
[BG2 br][-G2 br][ybSb1Vf2 N B][YbcSb1Vf2 N B](K y-)(J1-)Ac-
[BG2 br][-G1 Br][ybSb1Vf2 N B][ybSb1Vf2NN ]K yK w(J1-)Ac-
[BG2 br][-G1 Br][ybSb1Vf2 N B][ybSb1Vf2NN ]K yK y(J1-)Ac-
[BG2 br][-G2 Br][ybSb1Vf2 N B][ybSb1Vf2N N]K yK y(J1-)Ac-
[BG2 br][-G1 br][ybSb1Vf2 N B][Ybc Sb3Vf1N B](K y-)J1 J1 Ac-
[BG2 br][-G1 br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK wJ1 J1 Ac-
[BG2 br][-G1 br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK wJ1 J2 Ac-
[BG2 br][-G1 br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK yJ1 J2 Ac-
[BG2 br][-G1 br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK yJ1 J1 Ac-
[BG2 br][-G1 br][ybSb1Vf2 N B][ybSb1Vf2 N B](K y-)(J1-)Ac-
[BG2 br][-G2 Br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK wJ1 J2 Ac-
[BG2 br][-G2 Br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK yJ1 J1 Ac-
[BG2 br][-G2Br][ybSb1Vf2 N B][Ybc Sb3Vf1 N N]K yK yJ1 J2 Ac-
[BG2 br][-G2 Br][ybSb1Vf2 N B][ybSb1Vf2 N B](K y-)(J1-)Ac-
Table 3. Genotypes in the backcross to H. cydno (female H. cydno mated to male F1)
Genes within square brackets are linked, with the maternal H. cydno alleles given first. Genes in parentheses are not expressed on that genetic background (Vf on bb). Certain genotypes cannot be distinguished due to epistasis involving Sb and J, so that full expression of white hindwing margin could be produced by Sb3Sb3 or Sb1Sb3 J2J2. Counts are given as females/males. Crossing-over was not observed in this backcross and would have been detectable between only two pairs of loci: B and G and Yb and Sb (in the latter case only in certain genotypes due to the interaction between J and Sb).
[bG1 Br][bG1-][Ybc Sb3(Vf1) N N][ybSb3--]K w-J2 J2 acac [bG1 Br][bG1-][Ybc Sb3(Vf1) N N][ybSb3--]K w-J1 J2 acac [bG1Br][bG1-][YbcSb3(Vf1)NN][ybSb1--]Kw-J2J2acac
[bG1 Br][bG1-][Ybc Sb3(Vf1) N N][ybSb3--]K w-J2 J2 Acac [bG1 Br][bG1-][Ybc Sb3(Vf1) N N][ybSb3--]K w-J1 J2 Acac [bG1 Br][bG1-][Ybc Sb3(Vf1) N N][ybSb1--]K w-J2 J2 Acac
[bG1 Br][bG1-][Ybc Sb3(Vf1) N N][ybSb1--]K w-J1 J2 acac
[bG1 Br][bG1-][Ybc Sb3(Vf1) N N][ybSb1--]K w-J1 J2 Acac
[bG1 Br][BG2-][Ybc Sb3Vf1 N N][Ybc Sb3--]K w-J2 J2 acac [bG1 Br][BG2-][Ybc Sb3Vf1 N N][Ybc Sb3--]K w-J1 J2 acac [bG1 Br][BG2-][Ybc Sb3Vf1 N N][Ybc Sb1--]K w-J2 J2 acac
[bG1 Br][BG2-][Ybc Sb3Vf1 N N][Ybc Sb3--]K w-J2 J2 Acac [bG1 Br][BG2-][Ybc Sb3Vf1 N N][Ybc Sb3--]K w-J1 J2 Acac [bG1 Br][BG2-][Ybc Sb3Vf1 N N][Ybc Sb1--]K w-J2 J2 Acac
[bG1 Br][BG2-][Ybc Sb3Vf1 N N][ybSb3--]K w-J2 J2 acac [bG1 Br][BG2-][Ybc Sb3Vf1 N N][ybSb3--]K w-J1 J2 acac [bG1 Br][BG2-][Ybc Sb3Vf1 N N][ybSb1--]K w-J2 J2 acac
[bG1 Br][BG2-][Ybc Sb3Vf1 N N][ybSb3--]K w-J2 J2 Acac [bG1 Br][BG2-][Ybc Sb3Vf1 N N][ybSb3--]K w-J1 J2 Acac [bG1 Br][BG2-][Ybc Sb3Vf1 N N][ybSb1--]K w-J2 J2 Acac
[bG1 Br][BG2-][Ybc Sb3Vf1 N N][ybSb1--]K w-J1 J2 acac
[bG1 Br][BG2-][Ybc Sb3Vf1 N N][ybSb1--]K w-J1 J2 Acac
We investigated the clustering of color pattern loci into tight linkage groups by comparing the extent of linkage among the 10 color pattern loci with a null model assuming random distribution of loci across chromosomes to perform a test similar to that in Turner (1984, , p. 158). The null distribution of loci per chromosome is approximately but not exactly Poisson distributed, where the dispersion = variance/mean = 1. Clustering was tested here numerically by assigning 10 loci randomly onto 21 chromosomes one million times. We used as a test statistic the log-likelihood ratio (G), where 10/21 loci are expected on average per chromosome. A more conservative test was also run using eight loci, because two pairs of putative loci that did not recombine (B and G, Sb and Vf) might each represent pleiotropic effects of a single gene.
Crosses between Heliconius cydno chioneus and H. melpomene rosina revealed a number of loci with major effects on color pattern.
Alleles at the locus control the presence (BB, Bb) or absence (bb) of the red forewing band of melpomene (backcross to cydno Bb:bb 94:80 G1 = 1.13, P> 0.05). There is epistatic interaction with the unlinked N locus, so that in B- N N- individuals the red is moved distally in comparison with the position in melpomene(Fig. 1B). The pattern of gene action, epistasis with N, and linkage (see below) are very similar to those of the B locus in interracial crosses of melpomene (Turner 1972; Sheppard et al. 1985), confirming homology.
This locus controls the presence (N NN N, N NN B) or absence (N BN B) of an area of white or yellow in the forewing band seen in cydno (backcross to melpomene N BN B:N NN B 54:65 G1 = 1.02, P> 0.05). The mode of gene action and its linkage (see below) are identical to that of the N locus segregating in interracial crosses of melpomene (Sheppard et al. 1985). The locus is apparently distinct from the L locus controlling the forewing band in crosses between several Colombian races of cydno (Linares 1996, 1997). L lacks the linkage seen here of N to Yb and Sb, and absence of the band is almost completely dominant, whereas at N absence is recessive.
Alleles at the Yb locus control the presence (ybyb) or absence (YbcYbc, Ybcyb) of the hindwing yellow bar of melpomene(Fig. 1C). In heterozygotes the bar is usually visible as a shadow of melanic scales with altered reflectance, but occasionally very sparse yellow scales are present (backcross to cydno YbcYbc:Ybcyb 80:86 G1 = 0.22, P> 0.05, backcross to melpomene Ybcyb:ybyb 61:57 G1 = 0.14, P> 0.05). On the basis of identical gene action and linkage, these crosses confirm the homology of this locus between species with that previously described from crosses within melpomene (Sheppard et al. 1985), and within cydno (Linares 1997).
This locus controls the presence (Sb3) or absence (Sb1) of the white submarginal band on the hindwing of cydno(Fig. 1D). The strength of expression in heterozygotes depends on at least one unlinked modifier (J; see below). Linkage and gene action are identical to that of Sb in interracial crosses of cydno (Linares 1996, 1997).
Forewing band color (Fig. 1F) is expressed as white (K wK w, K wK y) or yellow (K yK y) (backcross to melpomene in N NN B individuals, K wK y:K yK y 35:30 G1 = 0.38, P> 0.05). This is probably homologous with the K locus segregating in interracial crosses of cydno (Linares 1997). There is no obvious influence on the color of the yellow hindwing bar, but the locus can cause the inclusion of yellow scales in the normally white hindwing submarginal band. Fore- and hindwing band color is jointly controlled in polymorphic cydno populations in Ecuador (Kapan 1998). However, in some Colombian races a yellow forewing band is found with a white hindwing margin (Linares 1997).
In addition, several loci have less dramatic effects on mimicry.
Scale color on the ventral surface of the red forewing band (Fig. 1F) is either dark (Vf1Vf1, Vf1Vf2), or pale (Vf2Vf2) as in melpomene (backcross to melpomene Vf1Vf2:Vf2Vf2 62:56 G1 = 0.31, P> 0.05). This locus has not been described in either species, although its action has been noted in interspecific crosses (Gilbert 2003). The pale ventral surface scales are white in melpomene but can be white or yellowish in backcross Vf2Vf2 individuals.
At the Ac locus, alleles control the presence (accacc) or absence (AcAc, Acacc) of the anterior triangle of a white hourglass shape in the main forewing cell of cydno(Fig. 1F) (backcross to cydno Acacc:accacc 97:79 G1 = 1.84, P> 0.05). Gene action suggests homology with the Ac locus that segregates in crosses between a melpomene race from Trinidad with the red forewing band and Amazonian races in which the hourglass is present (Sheppard et al. 1985). Variation in the posterior half of the hourglass is more difficult to interpret: It is present in all of the backcross to cydno but is very variable in the backcross to melpomene.
Alleles at the Br locus control the presence (BrBr, Brbr) or absence (brbr) of a forceps-shaped brown marking on the hindwing ventral surface seen in cydno(Fig. 1E) (backcross to melpomene Brbr:brbr 56:63 G1 = 0.41, P> 0.05). Expression is variable in heterozygotes, which lack most of the distal part of either or both arms of the forceps, and is complicated by an epistatic interaction with the yellow bar, which occupies a similar position. The color is also variable, brown in cydno, but typically more orange in hybrids (compare Fig. 1, A and E). Linkage with B suggests that this locus is homologous with the D locus in melpomene, which controls orange “Dennis” and “ray” patterns on fore- and hindwing (Sheppard et al. 1985). There appear to be separable loci controlling the anterior and posterior components both of the cydno forceps (Gilbert 2003; M. Linares, personal communication) and of the rayed pattern in melpomene (Mallet 1989).
This locus controls the presence (G2) or absence (G1 ) of a short red line at the base of the costal vein on the forewing ventral surface of melpomene(Fig. 1H). Expression is intermediate and variable in heterozygotes (backcross to cydno G1G1 :G1G2 82:95 G1 = 0.96, P> 0.05, backcross to melpomene G1G2 :G2G261:57 G1 = 0.14, P> 0.05). This locus was first described from interracial crosses of Colombian cydno (Linares 1996).
Several of these loci are involved in epistatic interactions, in addition to that between N and B controlling forewing band shape and color. Some traits are expressed only in certain genotypes, for instance K only in N N- individuals, and Vf only in the B-genotype. Modifier loci that adjust the strength and position of expression of other loci are discussed below.
This is a modifier of incompletely dominant Sb control of hindwing submarginal band (Fig. 1D). In Sb1Sb3 heterozygotes, J1J1 genotypes express the band as melanic scales with altered reflectance on the ventral surface only; J1J2 individuals show a mixture of white and melanic scales producing a gray band expressed most strongly on the dorsal surface; Sb1Sb3 J2J2 individuals are indistinguishable from the Sb3Sb3 phenotype with a white dorsal and ventral submarginal band like that of cydno. In the backcross to melpomene, this gives an expected 2:1:1 ratio of absent (Sb1Sb1 J1J1 and Sb1Sb1 J1J2 ) to altered reflectance (Sb1Sb3 J1J1 ) to scattered white scales (Sb1Sb3 J1J2) (57:28:32, G2 = 0.34, P> 0.05). In the backcross to cydno, a 3:1 ratio of full expression (Sb3Sb3 J1J2, Sb3Sb3 J2J2 and Sb1Sb3 J2J2) to scattered white scales (Sb1Sb3 J1J2 ) is expected (137:40, G1 = 0.56, P> 0.05).
Forewing band width
In both backcrosses, variation in the position of the distal edge of the white or yellow part of the N N- forewing band is continuous, suggesting additive polygenic control (Fig. 1G). However, relatively few modifier loci must control this variation, because extreme phenotypes are common in backcrosses: Around 8% of individuals in the backcross to melpomene and 6% in the backcross to cydno have band widths similar to that in the F1. In the background of these modifiers, there is evidence of a slight effect of N and B loci on the width of the pale part of the forewing band. On average, the band is slightly wider in N NN N than N NNB individuals (Turner 1972) (using linkage with Yb to distinguish heterozygous from homozygous N), and the band is increasingly wide in BB, Bb, and bb individuals (using evidence from linkage to G to distinguish Bb heterozygotes from BB homozygotes). The distal edge of the red part of the forewing band is also variable in position. Its boundary is much less sharply defined and often more distal in hybrids than in melpomene(Fig. 1F). In the backcross to melpomene, the distal boundary varies from a melpomene-like to an F1-like position, independently of the effect of the N locus on the proximal boundary (Fig. 1G). In the backcross to cydno, the distal boundary is generally similar or slightly distal to that in F1 hybrids.
Dorsal forewing band color
There is continuous variation of the red hue in the dorsal forewing. In the backcross to melpomene this ranges from the scarlet of melpomene to an orange-red (Fig. 1F, center left), whereas in the backcross to cydno it varies from orange-red to brownish (Fig. 1F, center right). This continuous variation suggests that control is not homologous with the Or locus controlling red versus orange coloration in melpomene (Sheppard et al. 1985).
Melpomene has a variable number of red spots at the base of the hindwing ventral surface (Fig. 1A). There is often a single spot in the angle between the first anal vein and discal cell, but there may be up to three more, in the angles of the second anal vein and wing margin, the discal cell, and where the subcosta meets the discal cell. Penetrance is variable in hybrids, suggesting epistasis with modifier genes. Spots are absent from many F1 offspring, but present in almost all offspring of backcrosses to melpomene. They are absent from almost all offspring of some backcrosses to cydno but are overrepresented in one brood (brood 342 present:absent 46:26, G1 = 5.63, P <0.05, compared with a 1:1 expectation).
The black areas of the wing are iridescent blue in cydno and matt black in melpomene(Fig. 1A). Iridescence is difficult to score but appears to be under polygenic control: Iridescence is strong in the backcross to cydno, intermediate in the F1 and many from the backcross to melpomene, but absent in others.
Seven of the 10 loci fall into two linkage groups, Br-B-G and N-Sb-Vf-Yb. Three loci, K, Ac, and J, are unlinked to any other, and none are sex linked. The recombination fraction between Br and G is 23/118 (19.5% with support limits 12.9%, 27.4%) in the backcross to melpomene(Table 2). B and G are very tightly linked or may be pleiotropic effects of the same locus, because no recombinants appear among 174 individuals in the backcrosses to cydno. The loci in the other linkage group can be ordered by assuming double recombinants are very rare. In the backcross to melpomene, heterozygotes can be distinguished from homozygotes at all four loci so that crossing-over between any pair of loci can be detected. Gene order is most likely N-Sb-Vf-Yb, with 4.3% recombination between N and Sb-Vf-Yb (5/115, support limits 1.5%, 9.2%) and 0.9% between Yb and Vf-Sb-N (1/115, support limits 0.05%, 3.9%). Further recombinants between Yb and Sb are known from other crosses (Linares 1989), suggesting the two are indeed distinct loci. For recombinant phenotypes see Figure 1, H, F center left (lower Vf1-wings), and D lower right. Sb and Vf are tightly linked or pleiotropic effects of the same locus, with no recombinants among 115 individuals. Because there are 21 chromosomes in cydno and melpomene (Brown et al. 1992), the clustering of 10 loci into linkage groups of three and four loci far exceeds that expected if color pattern genes were distributed randomly across chromosomes (dispersion = 2.44, P < 0.001). The level of clustering remains significant even under the assumption that pairs of genes that show no recombination (Sb with Vf and B with G) are in fact pleiotropic effects of the same gene, giving eight genes with largest linkage groups of three and two loci (dispersion = 1.7, P = 0.015).
Presumed genotypes for the two species are [BrBrbbG1G1][N NN NSb3Sb3Vf1Vf1YbcYbc]K wK waccaccJ2J2 for Heliconius cydno chioneus and [brbrBBG2G2][N BN BSb1Sb1Vf2Vf2 ybyb]K yK yAcAcJ1J1 for H. melpomene rosina. These loci produce almost perfect resemblance to respective co-mimics H. sapho and H. erato. This extends to such minor details as the lightening of the ventral forewing produced by Vf2, the red line produced by G2, and the red spots at the base of the hindwing, all of which are seen in H. erato and replicated in H. melpomene. The only exceptions that do not contribute to mimicry are the K y allele (yellow) revealed on the pale forewing of hybrids, which cannot be expressed on the normal red forewing band of H. melpomene rosina, and the brown forceps-shaped mark on the hindwing of H. cydno produced by Br. Heliconius sapho differs from H. cydno in that it has large red patches near the base of the hindwing and has a red line in the forewing costa as in H. erato and H. melpomene.
Developmental drive and the role of major genes in mimicry
The 8–10 genes found here act together with several polygenic traits to control the mimicry difference between H. cydno and H. melpomene. Half of the loci are of major effect, and epistatic interactions and linkage affect most of them.
The gene effects are major in the sense that individual loci control a large fraction of the differences between the two species, affect large areas of the wing surface, and cause changes far beyond normal within-population variation (True et al. 1997; Orr 2001). Polymorphisms exist in just a few Heliconius populations, including H. cydno in Colombia and Ecuador (Linares 1996; Joron et al. 2001; Kapan 2001; Mallet 2001). Color pattern genes have major effects on pigmentation and scale morphology in specific areas of the wing (Gilbert et al. 1988) and are under very strong selection arising from mate choice (Jiggins et al. 2001) and mimicry (Mallet and Barton 1989; Kapan 2001). A similar distribution of mutational effects separates bee- and hummingbird-pollinated Mimulus flowers, recently interpreted as speciation due to floral mimicry (Bradshaw et al. 1998; Bleiweiss 2001). Mimetic adaptation apparently lacks selective constraints on fixation of developmental genes with major effects on fitness, even though Fisher (1930) argued that adaptation would typically proceed by fixation of many mutations of small effect. However, Fisher ignored the effect of selective advantage on the fixation probability of a new mutation (Kimura 1983), and the fact that adaptation involves sequential substitution of numerous alleles as the optimum is approached. When the entire process is considered, adaptation is expected to fix an exponential distribution of gene effects (Orr 1998, 1999).
Nonetheless, mimicry and perhaps many other traits involved in speciation do not fit the Fisher/Kimura/Orr model of adaptation. Mimicry is likely to evolve in two steps (Turner 1977; Sheppard et al. 1985; Mallet and Joron 1999). Only a major mutation yielding approximate resemblance can cross the adaptive valley between very distinct protected color patterns (Sheppard et al. 1985); this is followed by improvement of resemblance through natural selection on genes of more minor effect. If this model is correct, developmental mutations of major effect can in a sense be said to be “driving” adaptive evolution (Goldschmidt 1945; Turner 1984). The rugged adaptive landscape of mimicry contrasts with the smooth adaptive surface envisioned by Fisher and Orr, where gradual evolution occurs toward a single optimum. Despite these differences, a few large and several small mutations are expected under both theories, as in our empirical results (Turner 1984; Orr 1998). The distribution of gene effects therefore appears to reveal little about the adaptive landscape upon which evolution occurred.
Linkage and the evolution of Müllerian mimicry
Seven of the 10 loci described here fall into two linked groups. Tight linkage of color pattern loci like this is expected in polymorphic Batesian mimics but not in monomorphic Müllerian mimics (Turner 1984). Polymorphisms are expected in edible Batesian mimics of unpalatable models, because predators will learn to attack common mimetic forms more easily. In Batesian mimics such as Papilio memnon, polymorphisms are indeed found, with much of the pattern and wing shape changes inherited at a single “supergene.” The supergene consists of multiple epistatic elements controlling traits whose separateness can be demonstrated via occasional recombinants (Clarke et al. 1968; Clarke and Sheppard 1971). Far from being evidence for “systemic mutations,” as Goldschmidt (1940) proposed, it was now suggested that these supergenes had been constructed gradually from multiple unlinked genes that became more and more tightly linked. Later, Charlesworth and Charlesworth (1975) demonstrated difficulties with the Clarke and Sheppard hypothesis, because unlinked elements in polymorphic populations would normally be selected against after producing abundant nonmimetic recombinants. Turner (1984) therefore proposed an alternative explanation: The supergene evolved because only already linked epistatic elements will survive selection.
In contrast, Müllerian mimics such as H. cydno and H. melpomene are under purifying frequency-dependent selection, and monomorphic populations are indeed generally observed. Although previously known mimicry genes from H. erato and H. melpomene were often linked, they were not significantly clustered (Turner 1984). The significant clustering of color pattern loci we have found in H. cydno and H. melpomene is therefore unexpected. There are two possible explanations. First, linkage in Heliconius might be due to color pattern divergence arising in sympatry. Hybridizing incipient species would in effect form a polymorphic population, into which new epistatic mutations must be linked as in Batesian mimicry (Turner 1984) to become established. However, parapatric divergence, perhaps along a habitat or altitudinal gradient, is more probable in H. cydno and H. melpomene (Mallet 1993). Adaptive evolution of linkage is therefore possible but less likely than for Batesian sympatry. Second, clustering of mimicry genes might result from developmental and genomic constraints on the number of chromosomal regions that affect pattern (Mallet 1989) rather than a selective constraint on the location of substitutions that can become established. The tightly linked genetic architecture we observe could have arisen by duplication of regulatory genes, so that color pattern evolution proceeds within limited linked blocks (Mallet 1989; Force et al. 1999). Some process of gene duplication followed by the acquisition of new functions seems especially likely in Heliconius, where linked genes sometimes promote development of similar pattern elements; for example, red markings are determined by loci in the B linkage group and white/yellow markings by the N group. Similar associations are observed in H. erato and H. himera (Jiggins and McMillan 1997). In Papilio, supergenes include more diverse tightly linked loci controlling wing shape (tails on the hindwing) and body color as well as wing color (Clarke and Sheppard 1971; Turner 1984). However, development of butterfly color pattern, scale morphology, and wing shape (the latter produced by cell death at the wing imaginal disc margin) can occur in response to a single signaling pathway (Carroll et al. 1994). A complete explanation of linkage in both systems must await molecular characterization of the loci involved, but it is tempting to predict that supergene inheritance in Müllerian mimics implies developmental and genomic constraints on color pattern control, instead of their construction by natural selection alone.
Role of epistasis in adaptive speciation
Epistasis plays a profound role in speciation: It is the source of genomic or ecological incompatibilities in hybrids. Here epistasis affects several levels. There is epistasis between specific pairs of loci, for example, the lack of K expression on an N BN B background, the interaction between N and B in positioning the forewing band, and that between Sb and J in the strength of expression of the hindwing submarginal band. More general epistasis is dependent on genetic background, so that in hybrids the color pattern elements are less sharply defined. This suggests that canalization of pattern development breaks down in the absence of coadapted modifier genes (Clarke and Sheppard 1960; Mallet 1989). Finally, epistasis at the fitness level selects against nonmimetic pattern combinations, as in classic hybrid inviability and sterility (Turelli and Orr 2000). Quantitative genetic analyses of morphological differences between species typically find little evidence of epistasis (Orr 2001), but disruptive selection will generate epistasis for fitness even where the underlying genetic basis of the ecologically important trait is additive (Whitlock et al. 1995). In mimicry, as well as in classic postmating isolation, epistatic hybrid dysfunction should be a common incidental by-product of adaptive divergence.
Genetic architecture of intra- and interspecific divergence
Color pattern is strikingly diverse within Heliconius, involving convergence between the major clades of the genus, racial differentiation, and speciation (Turner 1976; Jiggins and McMillan 1997; Mallet et al. 1998; Gilbert 2003). Both H. cydno and H. melpomene have diversified into color pattern races across Central and South America, matching those of respective co-mimics H. sapho+H. eleuchia or H. erato and co-mimics (Brown 1979). Most of the loci encountered here have been described previously from interracial crosses within H. melpomene (N, B, Yb, Ac) (Sheppard et al. 1985; Mallet 1989) or within H. cydno (Sb, Yb, K, G) (Linares 1996, 1997). Linkage relationships are also similar to those previously described: N with Yb in H. melpomene (Sheppard et al. 1985) and Sb with Yb in H. cydno (Linares 1997). The linkage between B and Br suggests homology of Br (found here to control the brown forceps shape on the hindwing of H. cydno) with the D locus linked to B in H. melpomene (D controls the “Dennis” pattern of orange on the proximal part of the fore- and hindwings in Amazonian races; Sheppard et al. 1985; Mallet 1989. Homology also exists between the two major loci responsible for color pattern differences between another pair of sister species, H. erato and H. himera, and loci controlling pattern variation within H. erato (Jiggins and McMillan 1997). There is therefore no obvious distinction between the genetic control of inter- and intraspecific pattern differences: Divergence at both taxonomic levels is effected by many of the same loci. The participation of genes of large effect in both cases also suggests that the “evolution by jerks” involved in mimetic shifts (Turner 1983) can promote rapid speciation.
Although Heliconius sister species typically belong to different mimicry rings (Turner 1976), mimetic shift does not always lead to speciation. Geographic variation in color pattern within species, particularly within H. erato, H. melpomene, and H. cydno, is often as dramatic as that between H. cydno and H. melpomene (Brown 1979). Yet contact zones between geographic races are characterized by rampant hybridization despite strong selection against nonmimetic hybrids (Mallet and Barton 1989; Mallet 1993). Initiation of speciation depends on the mimetic shift leading to a change in the predominant color that would normally act as a courtship releaser (Crane 1955). In most races of H. melpomene, red is the predominant color, in comparison with white or yellow in H. cydno, and this major shift may have driven male preference to coevolve, reducing male courtship toward females with the ancestral color pattern (Jiggins et al. 2001). Any initial reduction in gene flow due to pleiotropy with mate choice and selection against nonmimetic hybrids will facilitate further adaptive divergence and completion of speciation (Rice and Hostert 1993).
Evolution of dominance
In our crosses, dominance and penetrance are variable and are not simply intrinsic properties of alleles. Dominance is influenced by genetic background (Doebley et al. 1995), and a specific dominance modifier, J, can be identified that affects the hindwing margin. In heterozygotes at Sb, the J locus controls penetrance, from complete dominance of the melanistic allele, through to strong expression of white. Such variation casts doubt on the argument that the recessive phenotype comprises the ancestral color pattern in H. melpomene or H. erato (Turner 1984; Sheppard et al. 1985). If dominance normally evolves to this extent, it will be of little use in determining the ancestral phenotype (Mallet 1989).
Genetic architecture and introgression
Clustering of genes for the major pattern differences into just three chromosomal locations, near the B, N, and K loci, may prohibit gene flow between species at adjacent loci but leave other regions to introgress relatively freely (Barton 1979). Also, due to linkage of N, Sb, Vf, and Yb and dominance modifiers such as J, parental genes cosegregate so that almost half the backcross offspring are similar to H. cydno or H. melpomene. These phenotypes form passable Müllerian mimics, so that backcrosses may occur more commonly in nature than inferred from collections of aberrant hybrids (Mallet et al. 2001). Cryptic hybrids would largely escape the strong selection due to predation on nonmimetic patterns. Thus, although mimicry may drive speciation, the clustered architecture of color pattern directly reduces its efficiency as a barrier to gene flow, creating a semipermeable species boundary.
Homology of genes across Heliconius and developmental constraints in macroevolution
The mimetic pair H. melpomene and H. erato are congeneric species that have diversified in parallel across South and Central America. Goldschmidt (1945) suggested that Müllerian co-mimics like these might often exploit the same developmental pathways to achieve identical color patterns but that the specific genes involved would probably differ. More recently, the even more extreme “Goldschmidtian” argument has been made that mimicry genes are homologous between Müllerian mimics H. erato and H. melpomene (Turner 1984; Nijhout 1991). Given that genes acting late in butterfly color pattern determination are the same as those acting early in embryonic development of Drosophila (McMillan et al. 2002), at first sight the suggestion is not implausible. If these ideas are correct, the construction of the mimetic color pattern would depend far more on constraints imposed by the developmental system than envisaged in traditional mimicry theory (e.g., Fisher 1930).
Linkage patterns suggest some homologies between Heliconius mimics; for example, the orange Dennis and ray patterns in melpomene and erato are both inherited as dominant supergenes (Turner 1984; Sheppard et al. 1985; Mallet 1989) and the hindwing yellow bar is tightly linked to white margin in cydno, melpomene, and erato (Turner and Sheppard 1975; Jiggins and McMillan 1997). However, in most cases mimetic patterns show key differences in genetic control. The Dennis allele of melpomene, for example, expresses an orange hindwing bar for which there is no homolog in erato. In H. erato, forewing band color (red or yellow) is controlled together with Dennis and ray pattern elements by a supergene (DRy), unlinked to the Cr locus controlling yellow hindwing bar (Sheppard et al. 1985; Mallet 1989). In heterozygotes for band color, the forewing band is red but sometimes “overprinted” with faint yellow pigment (Sheppard et al. 1985; Mallet 1989; Jiggins and McMillan 1997). This contrasts with gene action and linkage in the melpomene group, where red and yellow/white forewing band elements are controlled by separate loci (B and N); only the former is linked to Dennis and ray, whereas the latter is linked to yellow hindwing bar (Yb). Forewing bands with both red and yellow elements in melpomene and cydno show distal displacement of red rather than overprinting as in erato. Some details of genetic control also differ between H. erato and its sister species H. himera. In himera×erato crosses, red forewing band and red hindwing bar are both controlled by the D R supergene, whereas yellow forewing band is unlinked and controlled by the Cr locus affecting yellow hindwing bar (Jiggins and McMillan 1997); in erato, yellow forewing band is tightly linked to D R. There is even geographic variation in the genetic control of similar patterns within H. erato: The hindwing yellow bar has a disjunct distribution among races and is under the control of two loci in Peru (Mallet 1989) and Brazil (Sheppard et al. 1985) but only a single locus in Central America (Sheppard et al. 1985; Mallet 1986) and Ecuador (Jiggins and McMillan 1997).
Overall, these results suggest some role for developmental constraints in the evolution of Müllerian mimetic patterns. Yet linkage and gene action are evolutionarily labile between the erato and melpomene groups and even within the erato group. A variety of mutable loci therefore appears to be able to affect a single pattern element. Developmental mechanisms may thus only weakly interfere with the evolution of mimicry, whereas the major work of pattern construction is apparently achieved by natural selection. However, the final elucidation of these possibilities awaits mapping and molecular developmental characterization of the genes discovered here and in previous studies.
We are very grateful to the Smithsonian Tropical Research Institute, where this work was carried out, and to the Autoridad Nacional del Ambiente for permission to work in Panama. The project was funded by a Natural Environment Research Council grant and a studentship from the Biotechnology and Biological Sciences Research Council.