Colour polymorphisms (CP’s) continue to be of interest to evolutionary biologists because of their general tractability, importance in studies of selection and potential role in speciation. Since some of the earliest studies of CP, it has been evident that alternative colour morphs often differ in features other than colour. Here we review the rapidly accumulating evidence concerning the genetic mechanisms underlying correlations between CP and other traits in animals. We find that evidence for genetic correlations is now available for taxonomically diverse systems and that physical linkage and regulatory mechanisms including transcription factors, cis-regulatory elements, and hormone systems provide pathways for the ready accumulation or modification of these correlations. Moreover, physical linkage and regulatory mechanisms may both contribute to genetic correlation in some of the best-studied systems. These results raise the possibility that negative frequency-dependent selection and disruptive selection might often be acting on suites of traits and that the cumulative effects of such selection, as well as correlational selection, may be important to CP persistence and evolution. We consider additional evolutionary implications. We recommend continued efforts to elucidate the mechanisms underlying CP-correlated characters and the more frequent application of comparative approaches, looking at related species that vary in character correlations and patterns of selection. We also recommend efforts to elucidate how frequency-dependent selection may act on suites of characters.
Colour polymorphisms (CP’s) are usually defined as occurring when two or more distinct, genetically determined colour morphs are found within a single interbreeding population, the rarest morph being too frequent to be solely the result of recurrent mutation (Huxley 1955). They have long been of interest to evolutionary biologists, for at least three reasons. First, colour patterns are easily scored compared to many other traits and thus are tractable for study, this being even more important in the pre-molecular genetics era. Second, CP’s have provided key models for studies in some of the most active and controversial areas in evolutionary biology including sexual selection, speciation, aposematism and mimicry. Finally, CP’s provide the sort of raw material widely thought to facilitate sympatric speciation (Sinervo & Svensson 2002; Bolnick & Fitzpatrick 2007; Gray & McKinnon 2007; Fitzpatrick et al. 2009; Mallet et al. 2009). It has also been argued that CP is of considerable ecological significance (Forsman et al. 2008).
Since some of the earliest studies of CP, it has been evident that alternative colour morphs often differ in features other than colour (e.g. Cain & Sheppard 1954; Jones et al. 1977; Brodie 1992; recent reviews by Sinervo & Svensson 2002; Roulin 2004; Gray & McKinnon 2007; Fig. 1). Indeed, some investigators have used colour patterns essentially as markers for different traits that were the true focus of the study, whereas in other instances colour patterns were quickly recognized as associated with a set of distinct alternative mating or life history strategies (Oliveira et al. 2008). Even in systems where alternative colour morphs do not obviously represent distinct multi-trait strategies, colour patterns and other traits are often found to covary. These correlations are in some respects surprising since it has frequently been argued that selection on colour patterns should minimize pleiotropic effects of mutations (e.g. Prud’homme et al. 2007). Here we review our current state of knowledge concerning the evidence for genetic correlations between colour patterns and other traits, for which genetic mechanisms underlie such correlations, and whether these correlations are likely to be evolving under the influence of selection. We also explore possible evolutionary implications of the results of our survey, particularly for the processes that might maintain CP and for how readily new species may evolve from CP systems. With increasing interest in how sets of traits are acted on by selection and evolve in response (e.g. Lande 1979; Blows 2007; Walsh & Blows 2009; Sih et al. 2004 and Bell 2007 with regard to behavioural syndromes), and much improved methods for elucidating the molecular basis of quantitative traits, new results are rapidly emerging on the genetic mechanisms responsible for correlations between CP’s and other characters. In light of our own expertise and the very extensive literature, we have focused our review on animals (but see Box 1).
Box 1 Plant colour polymorphisms
Although we have focused our review on animals, we briefly survey results from plants to ask if the patterns and phenomena observed in animals obtain more broadly. Here we rely heavily upon recent reviews on plant CP and colour evolution (particularly Strauss & Whittall 2006; Rausher 2008; Bomblies 2010).
It is clear from even the most cursory survey that plant CP’s are frequently associated with correlated traits that themselves may affect fitness and experience selection. For example, in a study of 10 populations of four species of South African Protea, white morphs produced seeds 10% heavier and 3.5 times more likely to germinate than those of red morphs, whereas red morphs were less susceptible, at least at times, to seed predation by endophagous larvae (Carlson & Holsinger 2010). In the common morning glory, Ipomoeia purpurea, an allele for white flowers has a transmission advantage resulting from increased selfing rates in white-flowered morphs, but when homozygous is associated with reduced survival from germination to flowering (Coberly & Rausher 2008). The latter study is an example of a case in which the trait correlation has been shown to have a genetic basis and additional such cases have been documented (reviewed by Strauss & Whittall 2006; Rausher 2008).
Most molecular genetic/developmental studies of plant CP’s have focused on anthocyanin pigments and the pathways associated with their production and regulation, which are relatively well characterized. As for animals, both structural and regulatory loci have been found to contribute to the relevant genetic variation. For example, in the morning glory study described above, the a allele responsible for white flower coloration is the result of an insertion in the gene that codes for chalcone synthase, the first enzyme in the anthocyanin pathway. The pleiotropic effects of a may occur because chalcone synthase is also the first enzyme of the more general flavonoid pathway and its inactivation should affect not just floral but also vegetative tissues, where flavonoids perform several physiologically important functions; anythocyanins have been shown to be especially important in stress tolerance in several taxa (Strauss & Whittall 2006). A different white-causing allele in common morning glory, w, illustrates how pleiotropic effects may be relatively lower when the underlying mutations are regulatory and tissue-specific, as discussed earlier for animals. The w allele causes white flowers by inactivating a transcription factor coded by the gene Ipmyb1. ww individuals still have anthocyanins and other flavenoids in their vegetative tissues and do not exhibit any deleterious pleiotropic effects (reviewed in Strauss & Whittall 2006; Coberly & Rausher 2008). Evolution of regulatory mechanisms has now been implicated in several floral CP’s but the level of resolution of the molecular genetics is less than in some of the recent fruit fly studies described above (Bomblies 2010).
In contrast to animal studies, hormonal mechanisms seem not to have been implicated extensively in CP pleiotropy in the plant literature. ‘Supergenes,’ another important concept in the animal CP literature, have been discussed for plants mainly in the context of alternative morphs in incompatibility systems (e.g. Gilmartin & Li 2010) rather than colour variants.
Evidence of genetic correlations with CP
Table 1 presents a survey of colour patterns for which evidence exists of a genetic correlation between CP and one or more additional traits. Genetic correlation is defined as the correlation of breeding values (Falconer & Mackay 1996), or the proportion of variance that two traits share due to genetic causes. Such correlations may arise from linkage disequilibrium (LD), i.e. the non-random association of alleles at two or more loci, or from pleiotropy, which occurs when a single gene influences two or more characters (Falconer & Mackay 1996). Where information is available, we have indicated which mechanism appears to be responsible, as well as summarizing other information about genetic mechanisms and pertinent microevolutionary processes. While this summary is certainly not exhaustive, it does represent an extensive survey of examples for which either substantial genetic data are available or substantial information on relevant evolutionary processes has been gathered—the situations of greatest interest here. We have been somewhat liberal in applying the definition of CP, including some cases where variation is not clearly discrete (indeed the discreteness of some putative CP’s is contentious, e.g. Côtéet al. 2008; Vercken et al. 2007, 2008). We have focused on cases where genetic correlations (or LD or pleiotropy) are known from naturally occurring variants, as opposed to laboratory mutational studies that may not reflect the situation in nature (Wittkopp & Beldade 2009).
Table 1. A survey of colour polymorphisms for which there is evidence of genetic correlations with other traits
Evidence for genetic correlation‡
Inferred genetic mechanism
Forms of selection§
*Lower case indicates expression present but weaker than in other gender.
†Where evidence of genetic correlation varies, best documented traits are listed first.
‡‘Mendelian crosses’ indicates traits approximately mendelian and in crosses appear to map to the same or linked loci; ‘animal mode’ indicates a mixed effects quantitative genetic analysis, usually restricted maximum likelihood of pedigreed natural populations; ‘quanititative genetic’ indicates analysis of correlations among relatives or a closely related standard method; others self-explanatory.
§Frequency-dependence is negative unless indicated otherwise.
Blue-tailed damselfly Ischnura elegans
Body colour pattern
Morphology, dimorphism, others
Phenotypic correlations with heritable colour; mendelian crosses, others
As is evident from Table 1, colour has been found to correlate with a wide range of traits. These include body size, life history characters such as clutch and egg size, development time, morphological traits, immune system traits, characters associated with stress response, and various aspects of behaviour including aggression, mating strategy and mate choice. It is noteworthy that examples have emerged of female colour patterns that are correlated with reproductive and immune system traits, the suite of characters together constituting alternative reproductive strategies. Alternative reproductive strategies have been extensively studied in males (e.g. Oliveira et al. 2008) but have generally received less attention in females (Alonzo 2008). Many of the colour patterns studied involve form and extent of melanism but colour patterns based on carotenoids and other pigments have also been examined (Table 1).
The nature and quality of the evidence for genetic correlation emerging from our survey is highly variable (Table 1). Perhaps the most easily gathered and common, though not definitive, form of evidence involves documenting the heritability of a colour pattern then showing covariation with other traits (e.g. Calsbeek et al. 2008). Building on this approach, a trait putatively correlated to colour can be studied in sibs displaying different heritable colour patterns in order to infer pleiotropy (or tight linkage) of the trait with colour loci (e.g. Dijkstra et al. 2009). Formal demonstrations of genetic correlations between colour patterns and other traits, generally involving either the more traditional correlations among close kin or the more current ‘animal models’ and pedigrees (Åkesson et al. 2008), are now available for an increasing, taxonomically diverse set of systems. Mendelian analyses, usually accompanied by calculations of LD or other analyses, also continue.
Genetic correlations have been elucidated at the molecular level in a small but growing number of studies (Table 1). In a diverse array of species for which the molecular genetic basis of melanic polymorphisms has been assayed, the melanocortin-1 receptor (Mc1r) has proven critical (reviewed by Kingsley et al. 2009; Hubbard et al. 2010), but these examples are not included in Table 1—possible correlations with other traits have generally either not been assayed or not been reported. Indeed, it has been suggested that Mc1r is so ubiquitous precisely because of its minimal pleiotropic effects (Mundy 2005; Kingsley et al. 2009).
Mechanisms of Genetic Correlation and their Evolution
It has been widely argued that adaptive evolution will frequently occur through those genetic mechanisms, such as changes in cis-regulatory elements (CRE’s) or hormone receptors, that have the most limited pleiotropy and thus are least likely to disrupt adaptation in other traits (e.g. Prud’homme et al. 2007; Kingsley et al. 2009). At the same time, it is increasingly clear that pleiotropy can potentially evolve as an adaptation, particularly through changes in linkage relationships, CRE’s or pathways of hormonal action. Indeed, theory predicts that disruptive selection, especially across multiple traits, may favour the genetic control of polymorphic characters by a limited number of loci. Such a pattern should result in robustness to recombination and a better match to the adaptive landscape (van Doorn & Dieckmann 2006; ten Tusscher & Hogeweg 2009). We next examine studies of genetic mechanisms with particular regard to these issues and the evolution of CP-correlated traits. In general, the data available remain limited but new molecular tools and a growing awareness of the role of nonlinear selection are stimulating rapid advances on the molecular underpinnings of polymorphisms, with tantalizing results.
Restricted recombination is the hallmark of loci in tight physical linkage along a chromosome or occurring within regions where recombination is suppressed, for example within inversions, in the proximity of sex determining regions, or through the direct involvement of recombination modifiers (Butlin 2005). It has been noted (Pepper 2003) that physical linkage, particularly when originating from epistatic clustering, might generate modularity with respect to recombination, rather than the more commonly considered mutational modularity (Wagner et al. 2007).
Chromosomal rearrangements such as inversions, fusions and translocations are frequently implicated in the generation of correlations through suppression of recombination and/or by bringing unlinked loci together (Ford 1964; Nei 1967; Dobzhansky 1970; Lenormand & Otto 2000; Rieseberg 2001; Butlin 2005; Kirkpatrick & Barton 2006). In particular, inversions have been long the subject of intense theoretical and empirical work, given their frequent underdominance and hence their possible role in reproductive isolation (e.g. Wright & Dobzhansky 1946; Dobzhansky 1970; White 1978; King 1993; Noor et al. 2001; Rieseberg 2001). By generating LD between previously unlinked genes on the same chromosome, inversions could give rise to ‘coadapted gene complexes’ (Mayr 1963) behaving as ‘supergenes’ (Ford 1964; Turner 1967; Dobzhansky 1970). Of particular relevance is the analysis by Kirkpatrick & Barton (2006) suggesting that inversions can spread in a population when they incorporate locally adapted alleles even in the absence of epistatic interactions for fitness between them: the inversion spreads at migration-selection equilibrium because each of these favourable alleles captured by the inversion will have higher fitness than its copies in non-rearranged chromosomes since it will always co-occur with the other favourable alleles at other loci in the inversion. This theoretical result expands the range of potentially correlated traits in inversions by relaxing the requirement for ‘coadapted alleles’ characteristic of previous models. Recently, Feder & Nosil (2009) examined the conditions under which loci responsible for adaptive divergence and speciation are expected to accumulate within inversions. Under the adaptive scenario envisaged by Rieseberg et al. (2001) and with low levels of recombination within inverted regions, Feder & Nosil (2009) found that the level of differentiation within inversions was higher than in collinear regions when more than two loci were considered and migration was low. These models suggest that inversions might allow the maintenance of polymorphisms as long as these involve suites of traits clustered within the inversion, even when some recombination is present.
Evidence for correlations between colour and other traits mediated by chromosomal rearrangements is at this time limited. Hatadani et al. (2004) suggested that two inversions on the second chromosome of Drosophila mediopunctata might be responsible for a polymorphism in temperature adaptation and abdominal dark spot patterns but no gene associated with either trait has been unequivocally identified so far. The crown plumage colouration of the white-throated sparrow (Zonotrichia albicollis; Fig. 1) is associated with a pair of included pericentric inversions (Lowther 1961; Thorneycroft 1966; Thomas et al. 2008) and a remarkable suite of behavioural and life history traits correlates with the colour (and inversion) polymorphism. Strong disassortative mating (and likely other processes) between the two colour morphs is thought to maintain a balanced polymorphism in natural populations (Lowther 1961; Falls & Kopachena 1994). Thomas et al. (2008) found that the included inversions extend for more than 98 Mb and might possibly contain around 1000 protein-coding genes: the authors suggested that genes for crown plumage might be linked in the inversion to genes determining the behavioural and life history polymorphism. Preliminary comparative mapping of the rearranged segment identified a number of candidate loci for colouration and aggressive behaviour but no gene, as yet, has been characterized in this system. Alternatively, a single switch gene within the inversion affecting neuroendocrine pathways might be responsible for both colour and behavioural polymorphism (discussed below).
A butterfly CP revives the idea that inversions might be associated with reproductive isolation by ‘protecting’ genetic correlations between mating preference and mate recognition traits from the effects of gene flow, in the early stages of divergence. In the species Heliconius cydno and Heliconius pachinus, males use the colour of patches on female wings as a species recognition cue: H.cydno exhibits a large white patch on the forewings while H.pachinus has yellow stripes. Quantitative trait loci (QTL) mapping revealed that male preference for forewing colouration perfectly co-segregates with the main autosomal locus responsible for forewing colour variation (Kronforst et al. 2006; Chamberlain et al. 2009). In turn, this colour locus (K) is very close to another locus (wingless) previously implicated in butterfly pigmentation (Carroll et al. 1994). Since a white/yellow polymorphic population of H.cydno showed no erosion of LD between preference and wing colouration, Kronforst et al. (2006; also see Chamberlain et al. 2009) suggested that the correlation may be maintained by suppressed recombination through an inversion. Alternatively, a single gene with pleiotropic effects on pigmentation and colour preference might explain these results. In conclusion, while there is only limited evidence for correlations between colour and other traits mediated by chromosomal rearrangements, theoretical work (e.g. Rieseberg 2001; Kirkpatrick & Barton 2006) and the abundance of morphological and life history traits associated with inversions in model systems such as Drosophila species (Hoffmann et al. 2004) suggest that these might not be rare. The emergence of novel and cost-effective genomic techniques and the future whole genome sequencing of more non-model systems (e.g. Heliconius Genome Consortium) allowing rapid fine-scale comparative mapping, will provide powerful tools to assess the nature and function of those sets of genes, ‘recombination escapees,’ that have found a safe haven within chromosomal rearrangements.
As noted above, selection need not be strictly correlational to lead to tight linkage of sets of genes. In other words, correlations generated by physical linkage do not necessarily involve coadapted gene complexes (Kirkpatrick & Barton 2006). An example of this might be sexually antagonistic selection. Rice (1987) observed that sexually antagonistic traits (whether sexually selected or not) should accumulate on the sex chromosomes and predicted their location on the Y (W) or X (Z) chromosome according to their dominance and their effect on the sexes. If there is a tendency for sexually antagonistic colouration to accumulate in proximity to the sex-determining region, correlation with other antagonistic genes will evolve in a genomic area of reduced recombination. By this process, even sets of genes with no fitness epistasis (with each other—though certainly with sex determination) but sharing the common element of being sexually antagonistic might end up clustered together in proximity to the sex determining locus and contribute to the integration of sexually dimorphic phenotypes. We can therefore imagine the generation of Y(W)-linkage between colour and other genes as a process in which co-localization and recombination suppression are simultaneously achieved via independent accumulation of each gene in the proximity of the sex-determining locus, i.e. as a by-product of sexually antagonistic selection. A slightly different scenario is expected when sexually antagonistic genes accumulate on the recombining homolog X(Z), as predicted by Rice (1987). In this case, following independent localization on the X(Z) chromosome, a second step is required to tighten linkage between the now X(Z)-linked genes. When this second stage is driven by selection, the process might be penalized by the lower recombination rates characterizing such chromosomes.
In addition, genes linked in the proximity of the sex determining region might frequently give rise to negative genetic correlations due to accumulation of deleterious mutations, as has been suggested for sex-linked colouration in the guppy (Brooks 2000).
Guppies, platyfish and various other members of the family Poeciliidae display an astonishing number of sex-linked colour traits (Volff & Schartl 2001; Lindholm & Breden 2002). In the guppy, sex chromosomes are largely homologous (Winge 1927) however cytological studies, karyological work (Traut & Winking 2001) and QTL mapping (Tripathi et al. 2009) suggest that a large portion of the sex chromosome might be subject to suppressed recombination. Colour traits are mostly sex-linked but few are limited to the Y chromosome and rare crossover events have allowed the identification of a set of distinct colour genes in tight linkage in proximity to the sex-determining region. Further, the recent construction of a genetic linkage map for the guppy revealed the presence of several QTL responsible for body size, body shape and colour patterns on linkage group LG12, known to harbour the sex-determining factor. In the platyfish Xiphophorus maculatus, a set of genes tightly linked to the sex determining locus was identified: the Mdl locus determines onset and location of macro-melanophore patterns (Gordon 1931; Kallman 1975), the RY locus (possibly a group of tightly linked genes) is responsible for red-yellow pigmentation patterns (Kallman 1975) and the pituitary locus (P) determines age and size at sexual maturation via its effects on the hypothalamic-pituitary-gonadal axis (Kallman 1989). Natural populations are highly polymorphic at both colour loci and the pituitary locus and alternative reproductive and life history tactics are associated with such variation (Kallman 1989; Ryan et al. 1990, 1992). In addition, duplications, deletions and transpositions have frequently affected the region in proximity to the SD locus (Froschauer et al. 2001), increasing the levels of genetic variation characteristic of poeciliid sex linked loci. Therefore, the generation of tight linkage between colour loci on the sex chromosomes in guppies and platyfish might have originated both as resolution of sexual conflict through their accumulation in linkage with the sex determining locus and by rearrangements such as tandem duplications followed by neo- or sub-functionalization (Froschauer et al. 2001).
In Xiphophorus, both mechanisms seem to be involved in the evolution of the melanoma-inducing Xmrk oncogene (Schartl et al. 1995; Weis & Schartl 1998). Xmrk originated by tandem duplication of the epidermal growth factor receptor (Egfr). Divergence in the flanking regions of the duplicated copy, possibly affecting transcription factors, is likely responsible for the observed changes in expression patterns leading to melanoma formation. Interestingly, the oncogene is tightly linked to a copy of the melanocortin type 1 receptor Mc1r (Weis & Schartl 1998) and is located in the same region where copies of Mc1r, the pituitary locus P and the RY colour locus are clustered, as described above. It has been suggested that the persistence of the oncogene can be explained by genetic hitchhiking with a positively selected Mc1r copy (Schartl 2008).
In Xiphophorus cortezi, Xmrk is required for the expression of the spotted caudal melanin pattern, which is determined by autosomal genes (Kallman 1971). Fernandez & Morris (2008) showed that in polymorphic populations females prefer males with the caudal spot that are also carriers of Xmrk. In addition, females choose males with enhanced over males with reduced melanin patterns. However, such preference appears to be frequency-dependent: in a population with a high frequency of spotted females, non-spotted males were preferred, a shift probably driven by selection against the generation of offspring carrying two copies of the oncogene, which have reduced viability and shorter life span.
Using mirror image experiments, Fernandez (2010) demonstrated that the caudal spot pattern of X. cortezi is also involved in aggressive signalling. Moreover, individuals carrying the oncogene, regardless of spot pattern expression, are more aggressive, suggesting that the variation in agonistic behaviour is associated with the oncogene (or a factor linked to it) rather than to the autosomal spot pattern. Aggressive behaviour might be mediated by Mc1r for which multiple copies are known to be located in proximity to Xmrk (Froschauer et al. 2002; Selz et al. 2007). Fernandez & Bowser (2010) have confirmed the presence of Xmrk and associated malignant melanomas in both sexes in nature and documented increased body size in individuals with Xmrk; larger size can affect male attractiveness, female fecundity, dominance and survival in the presence of predators. Thus the fitness advantages to Xmrk-carrying males deriving from higher aggressiveness (Fernandez 2010) and intersexual selection (Fernandez & Morris 2008), and to both sexes of increased body size (Fernandez & Bowser 2010), might compensate for the reduction in lifespan due to melanoma formation (reviewed by Summers & Crespi 2010).
In a number of fish species, male and female heterogamety coexist, with X, Y and W sex chromosomes segregating in a single population. In the platyfish X. maculatus, some populations are known to harbour XX, WX and WY females, and XY or YY males (Bull 1983; Kallman 1984; Basolo 1994). While both X and Y chromosomes typically carry many colour genes in this species, none is located on the W sex chromosome: Kallman (1970) suggested that the invasion of a dominant female determiner (W) lacking linked colour traits might have been favoured because of the higher cost of colour traits in females. Similarly, in some populations of Lake Victoria and Lake Malawi cichlid fish species (Seehausen et al. 1999; Roberts et al. 2009), both male and female sex determiners are present and a characteristic blotched colour pattern is linked to the dominant female determining W (Seehausen et al. 1999; Lande et al. 2001; Roberts et al. 2009), giving rise to blotched morphs (i.e. carriers of W-) and a non-blotched morph (i.e. XX, XY). The correlation between colour traits and sex determiners can allow direct selection on the population sex ratio driven by inter- and intra-sexual selection on colour patterns that act as signals of sex chromosome composition (Lande et al. 2001; Pierotti et al. 2008, 2009; Dijkstra et al. 2009). Conversely, the correlation can help maintain multiple morphs characterized by alternative suites of traits located in genomic regions of reduced recombination (Dijkstra et al. 2009; Roberts et al. 2009).
Sex chromosome linkage of traits and preferences involved in species recognition and in lower hybrid fitness has received considerable theoretical attention (Servedio & Sætre 2003; Lemmon & Kirkpatrick 2006). Saether et al. (2007) showed that species recognition in two hybridizing Ficedula flycatchers is Z-linked and so is the male plumage colouration on which females base species recognition, as well as genes causing low hybrid fitness. The authors argue that the location of all three traits on the Z chromosome leads to a substantial reduction in recombination, possibly allowing the maintenance of their association during hybridization. It has been proposed that within-species CP’s might be maintained by a similar process, as in the red and black morphs of the Gouldian finch, in which head colour (Southern 1945) and postzygotic incompatibilities between morphs (Pryke & Griffith 2007) are Z-linked. Recently, Pryke (2009) found patterns of mating preferences between morphs and their hybrids consistent with Z-linkage and suggested that the morph-specific association between trait and preference might be protected from the effects of gene flow by Z-linkage.
An example of the interplay between positive and negative correlations mediated by physical linkage is provided by Gratten et al’s. (2008) study of adaptation in the Soay sheep, which provides a contrast to the mainly favourable genetic correlations so far discussed. The St Kilda’s population is characterised by coat CP determined by a single structural mutation at the autosomal TYRP1 gene: dark brown colouration is dominant over light tawny coat which results from recessive homozygosity at the TYRP1 locus. The dark morph is associated with larger body size because of linkage of the colour gene with a QTL associated with birth weight and adult weight; in addition, size is known to be positively correlated with survival and mating success in this species (Coltman et al. 1999; Wilson et al. 2005). However, rather than observing an increase of the dark morph in the population over generations, Gratten et al. (2008) recorded a decrease in the frequency of dark sheep. The apparent fitness disadvantage of the dark morph appears to be driven by a negative genetic correlation between the dominant TYRP1 allele and a tightly linked QTL affecting lifetime fitness. The evolutionary trajectory of Soay sheep colouration is therefore shaped by correlations with antagonistic effects on fitness and arising from tight linkage of loci affecting colouration, body size and lifetime fitness. Why the allele(s) responsible for the fitness decreasing QTL and the observed patterns of linkage persist remain open questions.
The most extreme form of tight linkage preserving coadapted allelic combinations is represented by the so-called ‘supergenes’, clusters of coadapted genes in tight linkage. Recently, some authors, spearheaded by the influential work of West-Eberhard (2003) and Nijhout (1991, 2003), questioned the role of supergene-type genetic architectures in mediating polymorphisms. In particular, an increased understanding of the mechanisms by which suites of unlinked genes might be regulated via master switches led West-Eberhard to suggest that sets of genes underlying correlated traits might be more commonly connected by ‘regulatory linkage via coexpression’ rather than by physical linkage (West-Eberhard 2003).
The fluctuating fortunes of the ‘supergene’ idea have been historically intertwined with progress in the elucidation of the genetic basis of butterfly mimetic forms (Dobzhansky 1951; Mallet 1989; Charlesworth 1994; Mallet & Joron 1999; Naisbit et al. 2003; Baxter et al. 2010). The pioneering work of Clarke & Sheppard (1959, 1960, 1962) established that colour patterns in the Batesian mimics Papilio dardanus and Papilio memnon were determined by a single autosomal locus acting as a supergene composed of many elements, each responsible for a portion of the wing pattern. Tight linkage may have evolved between the different genes in order to preserve the allelic combinations associated with mimetic patterns. However, the conditions under which linkage would evolve gradually might be very restrictive given that the initially unlinked genes would produce a large number of non-mimetic combinations that should be strongly selected against (Charlesworth & Charlesworth 1975). Alternatively, Turner (1984) proposed that the loci contributing to supergenes are initially linked. Nijhout (1991, 2003) critically reviewed the genetic evidence for supergenes in Papilio and reconsidered the role of individual factors generating the complex mimetic patterns. He identified two major axes of variation that could explain wing pattern polymorphism: the first is associated with differences in background colouration of the wing mainly due to quantitative changes in the proportion of different scale types. The second axis accounts for quantitative changes in width or position of the black bands that define the pattern. If most of the variation is due to changes in the background colouration and in the expansion (contraction) of black areas, simple variation in the activity of a few unlinked structural genes regulated by a regulatory switch (the mimicry gene) might underlie the observed phenotypic variation. Recent evidence consistent with this hypothesis is discussed in the next section.
In the Mullerian mimics Heliconius melpomene and Heliconius erato, a few clusters of tightly linked factors are responsible for a large fraction of the phenotypic variation (Sheppard et al. 1985). Strikingly, clusters with similar phenotypic effects in the two species map to the same genomic regions, suggesting homologous gene clusters regulate pattern formation in different species (Joron et al. 2006). Even more interesting is the finding that the single locus (a ‘supergene’) responsible for most colour pattern variation in Heliconius numata maps to one of the clusters shared by H. melpomene and H.erato (Joron et al. 2006). The tightly linked factors in H. melpomene and H. erato might in fact be clusters of CRE’s of a single switch gene controlling pigmentation genes downstream. This is also supported by the observation that genes forming each tightly linked group control similar pattern elements (e.g. yellow markings on hindwing in one homologous cluster, red patterns of both wings in a second cluster) and are not implicated in pigment production but rather in the spatial distribution of pigment deposition on the wing. In H. numata (and perhaps in the Papilio supergenes), one of these switch genes might then have taken over control of the entire pattern formation (Joron et al. 2006; Baxter et al. 2008, 2010; Counterman et al. 2010).
In many cases, supergenes involve sets of related functions rather than tight linkage of coadapted genes each implicated in a separate pathway. In addition, their origin from unlinked loci is problematic (Charlesworth & Charlesworth 1975). Therefore, tight linkage of a large number of unrelated genes (‘supergenes’) might not be the most common genetic architecture underlying complex alternative phenotypes such as in CP’s involving coordinated expression of colouration, morphology, behaviour and life history. Supergenes might be frequently generated by tandem duplication of regulatory and/or structural genes, followed by sub-functionalization, instantly creating the initial linkage required by models of supergene evolution (Charlesworth & Charlesworth 1975). Such clusters of paralogous regulatory elements might allow concerted temporal and spatial expression of unlinked downstream structural genes. Supergenes originally attracted a lot of attention particularly for their feature of protecting strong genetic correlations between favourable alleles at distinct loci in the face of recombination. However, if molecular work confirms that most of these clusters are effectively composed of paralogs, supergenes might be viewed as the result of selection acting to break, rather than create, correlations. In particular, tandem duplication of regulatory elements (i.e. regulatory element expansion) or structural genes, followed by sub-functionalization, can minimise genetic correlations arising from pleiotropy and allow increased modularity and hence, evolvability.
Transcription factors, hormones and pleiotropy
It is well known that alleles of structural genes may cause differences in multiple phenotypic traits and such classical pleiotropy is well documented (e.g. Majerus 1998). More recently it has become apparent that physically unlinked genes may exhibit correlated patterns of expression if they evolve CRE’s that result in their regulation by the same transcription factors or hormones (Fig. 2). In this instance, particular transcription factors or hormones (and, in turn, their regulatory mechanisms) may have extensive pleiotropic effects.
Some of the most thoroughly elucidated examples of the genetics of colour pattern evolution involve CRE’s, although their role in the evolution of genetic correlations via pleiotropy is less well resolved. Rebeiz et al. (2009) documented the cis-regulatory changes in the ebony gene (which codes for an enzyme required for yellow colouration, its absence causing melanism) that account for apparently adaptive melanism at higher altitudes in Drosophila melanogaster from Uganda. Using reporter transgene constructs bearing the upstream regulatory portions of ebony, it was possible to examine effects on activity in various tissues. Remarkably, light and dark regulatory regions displayed similar activity in a variety of developing tissues including legs and brain. However, the dark allele regulatory region displayed dramatically reduced activity in the abdomen where the adaptive colour pattern is expressed. This result exemplifies the modularity and spatial specificity that have been extensively claimed for CRE’s.
Although these findings illustrate how pleiotropic effects of melanic adaptations can be minimized, it is well established that melanin synthesis in both arthropods and vertebrates involves multi-step pathways with great potential for pleiotropic effects at various points and affecting diverse aspects of the phenotype, including behaviour and immune responses (Ducrest et al. 2008; Wittkopp & Beldade 2009). The evolution of pleiotropic effects has been examined by Gompel & Carroll (2003) for a morphological correlate of melanic pigmentation in the Drosophilinae (the D. melanogaster subfamily). Both abdominal melanic pigmentation pattern (repressed) and trichome distribution (enhanced) are broadly influenced by the different distributions of the transcription factor Bab2 among taxa. However, Bab2 repression of melanic pigmentation appears to have been partially or entirely lost in several species, with other regulatory pathways presumably taking on a greater role—even while Bab2 regulation of trichome patterning is largely retained. The authors suggest that pleiotropic effects of Bab2 have evolved within the Drosophilinae; but the fitness consequences of this variation in pleiotropy are not known (Gompel & Carroll 2003).
More recently, Werner et al. (2010) have documented in Drosophila guttifera how several colour pattern elements that together make up a complex pattern of wing spotting are under the control of a single transcription factor, wingless, that regulates expression of the yellow gene and thus an essential protein needed for formation of black melanin. A comparative analysis of colour pattern and wingless expression suggests that cross vein pigmentation and a wingless -responsive vein spot CRE evolved in the common ancestor of Drosophila virilis and the Drosophila quinaria species group (including D. guttifera), effectively taking advantage of a pre-existing wingless distribution pattern to produce some aspects of the D. guttifera pigmentation pattern. Subsequently, regulation of wingless appears to have evolved in the D. guttifera lineage so as to closely associate wingless expression (and corresponding melanic spots) with the positions of developing campaniform sensilla, a type of wing sense organ. Werner et al. (2010) suggest that the new spots arose by co-option of wingless expression at new sites specified by regulators involved in campaniform sensilla development. This conclusion is of particular interest in the present context because it provides a possible example of how CRE evolution generates an association between a morphological feature and a colour pattern.
In a study of the Mocker swallowtail, P. dardanus, Clark et al. (2008) raise the possibility of extensive, presumably adaptive pleiotropy of a single regulatory factor controlling multiple distinct CP pattern components. Using whole genome mapping combined with a candidate gene approach, Clark et al. (2008) map the previously hypothetical H‘supergene’ locus, which determines colour pattern morph, to a 13.9 cM region that contains the gene for the transcription factor invected. Work continues with this system and it will hopefully clarify whether and how multiple downstream colour pattern genes have come to be under the influence of invected. Certainly CRE evolution for downstream genes will need to be investigated. Comparative studies of CRE’s will be critical in this and other systems in order to resolve whether multi-trait correlations have evolved adaptively to come under control of a single ‘switch’, rather than evolving through physical linkage of multiple distinct loci as long hypothesized. New genomic methods, perhaps including chromatin immunoprecipitation (ChIP: Rokas & Abbot 2009) will be important in this enterprise.
Hormones may act in a manner analogous to transcription factors to regulate the activity of other genes, and in some cases they join with receptors to form complexes that function as transcription factors (Zera et al. 2007; Adkins-Regan 2008; McGlothlin & Ketterson 2008). However, hormones have the potential to exert more systemic effects than are usually associated with the direct action of individual transcription factors.
Alternative colour morphs in some systems have been shown to differ in hormone titers (Miles et al. 2007). In side-blotched lizards, orange morph males have higher plasma testosterone than yellow or blue males; some yellow males transform into blue males late in the breeding season and the transforming males then exhibit elevated plasma testosterone (Sinervo et al. 2000; Sinervo & Calsbeek 2003). Orange males, with highest testosterone, also exhibit higher endurance relative to other morphs. Consistent with a causal role for testosterone in this polymorphic system, testosterone implants improved endurance in blue and yellow males, as well as enhancing access to female territories (Sinervo et al. 2000). Testosterone in turn is regulated in large part by the upstream gonadotropins luteinizing hormone (LH) and follicle-stimulating hormone (FSH), released by the anterior pituitary. Manipulations of LH and FSH in side-blotched lizards had a mixture of indirect (through testosterone levels) and direct effects on innate and adaptive immune function, hematocrit level, endurance, sprint speed, colour and mate attractiveness. Interactions between such effects and morph genotype suggest that the evolution of responses to hormones in different morphs is adaptive and that the yy genotype (yellow) typically operates below maximal capacities for these traits, which can be up-regulated opportunistically (Mills et al. 2008). The observation that immune system performance is influenced by gonadotropins independently of the effects of testosterone is of particular interest given the emphasis that has been placed on testosterone’s role in mediating costs of suppression (the ‘immunocompetence handicap hypothesis’, e.g. Roberts et al. 2004).
Contrasting corticosterone titers of different morphs have also been documented in side-blotched lizards in response to environmental variation, in this case for the female CP. Plasma corticosterone of yellow-throated females increased with higher densities of both yellow and orange morphs whereas orange-throated females showed reduced levels of corticosterone with increased density of orange females (Comendant et al. 2003). In addition, corticosterone manipulations produced different effects on reproductive effort and survival in orange and yellow females in directions consistent with their contrasting r (orange) and k (yellow) life history strategies, results again suggestive of adaptive morph-specific responses to hormone levels (Lancaster et al. 2007).
Variation in hormone titers along the hypotha-lamic-pituitary-gonadal axis has been implicated in the white-throated sparrow CP (reviewed by Maney 2008), a system in which physical linkage, through inversion, is also thought to play an important role in CP evolution (see discussion above; Thorneycroft 1966; Thomas et al. 2008). Males of the white striped morph, which show higher levels of aggressive and sexual behaviours but spend less time in parental activities (Tuttle 2003), were found to have higher levels of testosterone in several studies (Spinney et al. 2006; Swett & Breuner 2008, 2009). Female morphs also differ behaviourally (Tuttle 2003) and in their levels of LH and estradiol (Spinney et al. 2006; Lake et al. 2008). In a recent study (Maney et al. 2009), morph-characteristic behavioural differences persisted after steroid levels were manipulated to be consistently high. However, experimental subjects were captive and non-reproductive, two factors that can affect behaviour and steroid levels in this species, and might explain these unexpected results. Alternatively, morph-specific behaviours may have persisted because organizational roles of steroids are more important than activational roles or because other mechanisms are more important (Maney et al. 2009). Possible interactions between testosterone and corticosterone, mediated by corticosteroid-binding globulin, could also influence levels of bio-actively available testosterone but a recent study does not support this possibility (Swett & Breuner 2009). In contrast, a comparison of baseline corticosterone levels between morphs, together with the results of manipulations, suggest that corticosterone might play a causal role in the parental behaviour differences between male morphs (Horton & Holberton 2009).
Hormonal influences on CP and correlated traits are not confined to steroids or the hypothalamic-pituitary-gonadal axis. Ducrest et al. (2008) compared patterns of pleiotropy suggested by genetic (e.g. knockout) and physiological (e.g. melanocortin injections) manipulations of the melanocortin system with phenotypic correlations between melanism and other traits in natural systems. The melanocortins are a family of peptide hormones that interface with a family of receptors, including the well studied Mc1r, to regulate not only form and extent of melanism but also various other traits. The manipulation results provide consistent support for positive relationships between melanism and sexual behaviour, aggressiveness, and resistance to stressors. The same associations, in the same directions, are observed in a survey of unmanipulated vertebrates (Ducrest et al. 2008). The generality of these relationships is striking and suggests that they are either adaptive or highly constrained in their evolution.
In summary, hormonal profiles often differ between colour morphs, providing a coordinating signal that can potentially entrain additional traits and lead to correlated expression with colour pattern; a hormone that determines, or is itself regulated by, the same factor(s) as the CP can lead to pleiotropic cascades of additional traits. Responses to the same hormone are sometimes observed to differ between colour morphs, suggesting that evolution of downstream pathways has also occurred and in some instances in an arguably adaptive manner. Moreover, patterns of correlation in natural systems may be similar to those observed in mutational studies in at least some instances. Nevertheless, our understanding of the role of hormonal systems in the evolution of colour-correlated suites of characters is at an early stage. We have minimal formal evidence of genetic correlations between colour patterns and hormone titers. We also have little knowledge of the mechanisms underlying differing hormone responses between colour morphs, or of the evolutionary history of response or hormone evolution. Possible causes of differences in hormone activity between morphs include variation in levels of binding globulins, competitive interactions with other hormones, metabolizing enzyme activity, receptor expression or receptor cofactor expression, and these have only begun to be investigated (Ball & Balthazart 2008; Swett & Breuner 2009). Ultimately, for all these various mechanisms, the study of tissue-specific patterns will also be required (Zera et al. 2007).
Selection on certain allelic combinations can have a large influence on the genomic architecture by favouring the mechanisms reviewed above, such as pleiotropic mutations (Lande 1980) and chromosomal rearrangements (e.g. inversions) generating tighter linkage between epistatically interacting alleles. In the absence of such integration, the genetic correlation between genes influencing different traits maintained solely by correlational selection is expected to be small in a randomly mating population (Lande 1984) and in the absence of other evolutionary forces, to rapidly decay (Pomiankowski & Sheridan 1994; Falconer & Mackay 1996). Therefore, the long-term maintenance of polymorphisms purely by correlational selection, in the absence of pleiotropy and linkage, is still an open issue (Brodie 1992) and some contend that the main role of correlational selection might be the forging of elements of genetic architecture that allow more stable genetic correlations (e.g. pleiotropy, physical linkage, dominance). While there is growing awareness that correlational selection might be ubiquitous (Blows & Brooks 2003; Table 1), the genetic architecture responsible for the associated correlations has rarely been elucidated and examples of CP-trait correlations maintained solely by selection are rare. One exception is the side-blotched lizard, in which some genetically correlated traits have been mapped to the vicinity of microsatellites thought to reside on different chromosomes (Sinervo & Clobert 2003; Sinervo et al. 2006).
Non-linear selection on a suite of traits in combination with assortative mating might also lead to long-term stability of their correlation. The genetic coupling generated by correlational selection in the context of habitat or mate selection might be particularly favourable to the emergence of assortative mating, a very powerful force for the maintenance of coadapted complexes and, under restrictive conditions, for divergence and reproductive isolation (Kirkpatrick & Ravigne 2002; Gavrilets 2004).
The central role played by colour in visual communication suggests that colour traits might be frequently recruited in combination with particular alleles at behavioural loci (e.g. mate preference loci) leading to non-random mating between individuals with alternative suites of coadapted alleles. Therefore, assortative mating might represent a major player in the evolution and maintenance of genetic correlations involving CP’s. A famous example of such CP-preference correlation is found in guppies (Houde 1994), although mate choice in this species also involves other factors (e.g. Eakley & Houde 2004; Zajitschek & Brooks 2008).
A different scenario known to lead to genetic correlations between traits is migration between populations (Kimura 1956; Hartl 1980). When selection in different habitats is divergent for colour pattern and other selected traits (i.e. favouring different forms of a trait in each habitat) but gene flow is present, genetic correlations may be maintained as a result; immigrants from one habitat will tend to have alleles for colour patterns and other traits favoured where they originated, whereas residents will more often have suites of traits and alleles well suited to the habitat in which selection has acted on them and their ancestors. A recent example of such a pattern is observed in colour polymorphic Timema stick insects, in which the genetic correlation between host plant preference and colour pattern is greatest at sites with both host plant types and thus ongoing divergent selection and gene flow (Nosil et al. 2006).
Some evolutionary implications
Our review confirms that CP’s are commonly correlated with other traits and reveals that genetic mechanisms are often responsible. Such correlations may be mediated through unlinked suites of traits maintained in LD by selection or other processes; by tight physical linkage of the genes in question; or by pleiotropy, particularly through regulation of sets of genes by the same hormonal systems or transcription factors. Moreover, multiple mechanisms may be present in the same system. Although the details are in most cases poorly understood, it appears that responses to transcription factors or hormones may readily evolve, potentially linking suites of traits together through the same regulatory factor. Here we consider some evolutionary implications of these observations, summarizing, extending and synthesizing relevant points from the preceding sections.
The finding that CP’s frequently incorporate correlated traits has implications for CP persistence. If multiple traits experience negative frequency-dependent and disruptive selection through the same underlying process—for example because they contribute to alternative foraging or reproductive strategies—then effects of selection may build upon one another when traits are expressed in a suitably correlated manner. Stronger negative frequency-dependence of fitness should result, making stochastic effects less likely to eliminate one form from a population. Thus CP’s should be more stable and likely to be observed. An interesting prediction of this hypothesis is that CP’s that involve correlations with other traits should be older on average than those with fewer or weaker such correlations (although the time needed to accumulate multiple traits in a CP correlated suite may also contribute to this pattern). In addition, with stronger disruptive selection, CP’s involving a larger number of correlated traits should be more discrete, all else being equal. As a result of strong genetic correlation, selection on one trait will also be translated into evolution of the other, and vice versa.
Correlational selection may exaggerate both negative frequency-dependent and disruptive effects still further. Correlational selection in CP systems has now been detected in contexts ranging from male mating tactics to female life histories to avoiding predation (Table 1). It is potentially very important here for several reasons. First, it should select directly for the build-up of genetic correlations. Second, it appears that non-linear selection, including disruptive selection, often acts on suites of traits (Blows & Brooks 2003) and so CP’s may be more often favoured when they involve other characters. In addition, Doebeli & Ispolatov (2010) have recently developed a model in which, if the ecological properties of an organism are determined by multiple traits with complex interactions, much as in correlational selection, negative frequency-dependent selection more readily maintains polymorphism. Unfortunately, to the best of our knowledge there is minimal direct evidence (but see Sinervo & Clobert 2003; Sinervo et al. 2006) concerning negative frequency-dependent correlational selection—i.e. whether certain sets of trait values are more strongly favoured at low frequency than others, distinct from frequency-dependence for individual traits (for which evidence is also limited). Nevertheless, evidence for disruptive correlational selection promoting colour-trait correlations in several systems (Table 1) raises the possibility that correlational frequency-dependent selection may also be relatively widespread, since disruptive and frequency-dependent selection commonly emerge from the same aspects of natural history.
The evolution of suites of colour-correlated characters may also be driven by processes sometimes considered formally distinct from negative frequency-dependence but similar in their effects. In particular, if sexual selection is microhabitat-contingent and mate choice occurs entirely within microhabitats (much as for soft selection) rare morphs may be favoured and CP’s maintained (Chunco et al. 2007). Given that different microhabitats may select for divergence in multiple traits, disruptive and correlational selection are at the least very plausible in this context. And if one of the traits involved is habitat choice, CP persistence may be substantially facilitated (Taylor 1975). Similarly, models in which competitive interactions are not transitive (akin to rock-paper-scissors), are sometimes treated as frequency-dependent (e.g. Sinervo & Svensson 2002; Sinervo & Calsbeek 2006) and in other cases considered to exhibit frequency-dependent dynamics while being formally distinct (Harris et al. 2008). In any case, the build-up of trait correlations is likely to be favoured in such scenarios (Sinervo & Svensson 2002; Sinervo & Calsbeek 2006). Moreover, it is noteworthy that examples of three morph systems are rapidly accumulating (Jukema & Piersma 2006; Sinervo & Calsbeek 2006; Rowland & Emlen 2009), though they do not always share the same dynamics or causation.
Colour-correlated traits or character suites could also facilitate the evolution of reproductive isolation and speciation, even in sympatry. By enhancing CP persistence through one or more of the mechanisms described above they may provide the raw material for speciation, through any of several processes that potentially start with a CP (reviewed in Gray & McKinnon 2007).
It is also possible that CP suites could contribute more directly to the evolution of reproductive isolation. It is now established (e.g. Gavrilets 2004; Bolnick & Fitzpatrick 2007) that sympatric speciation is theoretically possible but only under rather restrictive conditions, one of the main difficulties being transmitting the force of disruptive selection to the genes responsible for assortative mating. Therefore the most favourable condition for sympatric speciation is when disruptive selection acts on a trait that also leads to assortative mating, a ‘magic trait’ (Gavrilets 2004). However, conditions for a magic trait might be achieved even when the trait influencing mating is not the same as the trait under disruptive selection as long as the two are strongly genetically correlated, expanding the range of genetic scenarios potentially leading to magic trait sympatric speciation dynamics. Thus the evolution of assortative mating may be more likely if, as a result of sampling more traits, one of the traits incorporated into the CP suite happens to lead to assortative mating—which may ensue solely as a by-product or also because it is favoured. This scenario is related to the ‘sampling process’ effect discussed in the context of ‘multifarious selection’ by Nosil et al. (2009; also see Gray & McKinnon 2007) and may be more likely if the CP is not strictly sex-limited. In addition, if CP-correlated suites of traits experience stronger disruptive selection through the mechanisms described above, any preference that leads to assortative mating based on the CP may be more strongly favoured than would be the case if suites of traits were not involved.
The speciation processes discussed so far would require no explicit role for geography. But if a CP-possessing species is widely distributed, one or more morphs may be lost from some populations either through stochastic processes or as a result of differences in selection. Corresponding mating preferences might also be lost, or mating preferences and correlated characters might evolve differently as a consequence of the changed social environment. Either way, there is the possibility for reproductive isolation to ensue between populations with or without CP or with different mono- or polymorphisms (West-Eberhard 1986, 2003; Sinervo & Svensson 2002). In a recent study of such processes, sometimes called ‘morphic speciation’ (Corl et al. 2010), the ancestral trimorphic condition in the side-blotched lizard (with morphs estimated to be over 10 Myr of age) was found to have given rise on multiple occasions to populations, subspecies or species with just two morphs or even one. The yellow morph is lost disproportionately often, suggesting selection rather than drift is responsible, and male size, female size and size dimorphism evolve at much accelerated rates in the one and two morph populations. Possible effects on reproductive isolation are suggested but remain to be confirmed (Corl et al. 2010). It is also noteworthy that, when multiple traits are correlated with CP as in this and some other systems, changes in environments and associated changes in selection may lead some traits to be pulled across adaptive valleys by selection on others, toward new univariate or multivariate optima. Along similar lines, gene flow across species (or even population) boundaries may have complex effects when transcription factors or hormones have coevolved with CRE’s to yield CP trait suites, again with the possibility of generating novel phenotypes and shifts to new adaptive optima (see Seehausen 2004; Mallet 2007 for reviews of hybridization and speciation).
Different mechanisms of genetic correlation will have different consequences for the persistence of correlations and for their subsequent evolution. Correlations arising from LD between unlinked loci, maintained solely by correlational selection or other microevolutionary processes, should be the most labile. As noted above, when favoured by selection over extended periods they might be expected to evolve more lasting underlying mechanisms (Sinervo & Svensson 2002); but this will depend on the availability of suitable genetic variation. The dynamics associated with correlations arising from physical linkage, for example due to proximity on autosomes or collocation in regions of reduced recombination (including inversions), are more difficult to evaluate and likely variable. Correlations arising from alleles of different genes that reside together in highly derived sex chromosomes might be expected to be especially durable, although chromosomal sex determiners evolve relatively quickly in some taxa (e.g. Ross et al. 2009). However, LD, and associated genetic correlation, between moderately or weakly linked autosomal loci should gradually dissipate in the absence of microevolutionary processes that maintain LD. Further, the recruitment of functionally distinct loci from throughout the genome is arguably most problematic for suites of traits maintained in correlation by physical linkage, as the variation in genomic organization required may often not be available. Genetic correlations resulting from the regulation of multiple loci by a single transcription factor or hormone, while stable over the evolutionary short term, should have the greatest potential for adaptive evolution and the recruitment of functionally distinct traits, given that gene regulation appears to evolve relatively rapidly in many instances (e.g. Werner et al. 2010).
Although it is perhaps early for such speculation in light of our limited knowledge of the mechanisms responsible for most CP-associated trait correlations, our survey also leads to some preliminary predictions about the mechanisms through which different sorts of correlations will evolve. A simple and possibly widespread mechanism for the evolution of closely related characters, in this case usually colour pattern elements, is tandem gene duplication and neo- or sub-functionalization. Such processes can potentially result in physically linked sets of genes and even supergenes and such physical linkage is predicted to be over-represented among related traits relative to sets of more distinct characters. Correlations with contrasting fitness effects in each gender can become correlated through co-localization on sex chromosomes, particularly the Y or W—although such linkage does not necessarily imply coadaptation of traits to each other. Suites of functionally distinct characters may most often come to covary through transcription factor, hormone or (especially) CRE evolution. Given the consistent patterns of correlation documented between melanism and particular classes of character, such as aggression, and the potential role of the melanocortins in such correlations (Ducrest et al. 2008), the continuing discovery of widespread patterns of correlation is also to be expected. An important question is whether these are most usefully thought of as constraints or as ubiquitous axes of variation favoured by selection.
Our understanding of the genetic basis of CP’s and their correlated traits remains relatively rudimentary in even the best-studied systems. What is mainly needed is thorough elucidation of the genetic architecture underlying trait correlations and the form and intensity of selection pressures operating on them. Rapid improvements in the power, speed, and cost of genomic investigations will greatly improve this situation in coming years.
In addition, it will be important to conduct comparative studies to allow better assessment of whether observed correlations and genetic architectures should be thought of as adaptations to particular patterns of selection, ancestral features that sometimes prove beneficial, or constraints.
It would be intriguing to know whether multi-trait CP suites are older and/or more discrete than CP’s involving fewer characters, as predicted if they are adaptive and facilitate the persistence of CP. The number of genes involved in CP’s with sets of correlated traits is another area requiring investigation. The intensity of selection required to generate a genetic correlation will be crucially dependent on the number of loci involved. In turn, the maintenance of a stable multi-trait polymorphism and the probability of divergence and speciation will be limited by this number.
Finally, there is still little integration of the limited number of theoretical treatments of frequency-dependent selection on multi-trait character suites with the emerging genetic data. Even more urgent are empirical studies, both observational and manipulative, examining how traits interact in their influence on frequency-dependent fitness.
It is an exciting time to work on the genetic basis of complex colour-associated polymorphisms but only a deeper integration of theoretical evolutionary genetics, evo-devo, evolutionary ecology and genomic approaches will allow us to shed light on the mechanisms maintaining polymorphism and ultimately, biodiversity.
We thank Louis Bernatchez, Sue McRae, Katie Peichel, Maria Servedio, John Stiller, Kyle Summers, Lengxob Yong and two anonymous reviewers for helpful comments on early drafts of some or all of the MS. Carol Goodwillie helped guide our foray into the botanical literature and Elaina Tuttle provided sparrow photographs.
J. S. M. is a Professor and Chair of the Department of Biology at East Carolina University. He is interested in how sexual selection, natural selection and other processes maintain colour polymorphisms and other forms of within-species variation. He is also interested in speciation and the evolution of sexual dimorphism. M.E.R.P. is a post-doctoral scientist working in McKinnon’s laboratory. His research interests are similar to McKinnon’s, with an additional focus on the evolution of sex determination. Both scientists use a variety of field, experimental, genetic and comparative approaches and work mainly on fishes.