From complex patterns, like the stripes of a tiger, to the simple changes in the presence/absence of pigment, as in arctic hares, the diversity in mammalian pigmentation is vast [1]. But in addition to diversity among species, there is often appreciable variation in pigmentation within species. Because members of the same species that differ in their pigmentation phenotype can be crossed, this within-species variation is amenable to traditional genetic analyses. In addition, the molecular genetic factors that regulate mammalian pigmentation are relatively well known [reviewed in 2]–[4], thus enabling the genetic bases of these phenotypes to be explored. Furthermore, a nontrivial advantage to studying pigmentation traits is that variation is often easily detectable by eye. Mutant phenotypes that affect the coloration of the entire body are especially conspicuous and are easily recognized by both captive breeders and field biologists. One such phenotype is dark pigmentation or melanism. It is clear that melanism has evolved many times in wide variety of animal taxa [5].

The genes that can cause darkening of coat color have been studied most thoroughly in the laboratory mouse. Although experimentally induced mutations in over 25 genes can produce dark fur in lab mice [6], spontaneous coat-darkening mutations have been reported in only four genes: the Agouti signaling protein (Agouti), attractin (Atrn), melanocortin-1 receptor (Mc1r), and mahogunin (Mgrn) [7]–[10]. The protein products of three of these genes, Mc1r, Agouti, and Atrn, interact at the surface of pigment-producing cells (melanocytes) and constitute the machinery responsible for “pigment type switching,” the ability of melanocytes to switch between the production of dark brown/black (eumelanin) and light yellow/red pigment (pheomelanin). Mc1r is a membrane-bound receptor that, when active, signals the melanocyte to produce eumelanin, whereas Agouti is a paracrine signaling protein that antagonizes Mc1r, causing the melanocyte to produce pheomelanin. Thus, mutations that cause either constitutive- or hyper-activation of Mc1r or loss-of-function of Agouti will result in a melanic phenotype. The functions of Atrn and Mgrn are not as well understood, although Atrn is thought to stabilize interactions between Mc1r and Agouti [11]. Here, we focus on Mc1r and Agouti because their interaction has been well characterized in the lab mouse and thus can be extended to the study of melanism in other taxa.

Melanic phenotypes have evolved both in nature and in captivity in a wide diversity of animals and in some cases their genetic basis has been identified. In captive vertebrates, spontaneous mutants of Agouti, Mc1r, Mgrn, and Atrn have all been found to cause melanism [e.g.], [12], [13], [10,8]. In natural populations, however, mutations in Mc1r are most commonly associated with melanism [e.g. 12], [14]–[16], although both Agouti and Atrn are larger mutational targets. In addition, mutations in Agouti or Atrn that reduce protein expression or activity lead to melanism; these knock-out mutations are certainly more likely to occur than the gain-of-function Mc1r mutations that cause melanism because there are more ways to “break” a gene than to “improve” a gene's activity. Thus, it is unclear why Mc1r has repeatedly been shown to be associated with melanism in nature and a key question is: are melanism-inducing mutations in Agouti not found because they occur less often, or are they simply more difficult to detect?

To address this question, we studied melanism in the deer mouse, Peromyscus maniculatus (Figure 1). Melanism has been reported in several populations of Peromyscus; melanic individuals have been captured in a number of locations in North America, including New Hampshire [17], California [18], Michigan [P. Myers, pers. comm.], and Alaska [C. Conroy, pers. comm.]. Although it is unclear if these melanic phenotypes affect fitness, their repeated occurrence provides us with multiple comparisons of the same phenotype in the same genetic system (i.e. species). Horner et al. [17] showed that, in mice from New Hampshire, melanism is caused by a recessive allele at a single locus. The authors suggested the locus might be Agouti, based on its similarity to the nonagouti phenotype in Mus. Here we uncover the molecular variation that causes melanism in P. maniculatus from New Hampshire and show that the Agouti gene is responsible. We also investigate the molecular basis of melanic phenotypes from geographically distant populations of P. maniculatus and find that melanism has independently arisen at least three times and by different mutations in the same gene, Agouti, in two of those cases.

The results of our laboratory crosses confirmed that melanism in New Hampshire P. maniculatus is caused by a single, recessive allele. In laboratory mice, dominant melanism is usually caused by alleles of Mc1r, while recessive melanism is usually caused by alleles of Agouti. Consistent with this dominance hierarchy, we found that melanism in P. maniculatus is perfectly correlated with the presence of an allele (a^Δ125kb) with a large deletion at the Agouti locus. When mice are homozygous for this allele, the abundance of Agouti transcript in the skin is significantly lower than that in individuals with a single copy of the wild type Agouti allele. This accords with the observation that the deleted region contains the 5′ untranslated regions that are important for temporal and spatial regulation of Agouti and probably any associated cis-regulatory information. The deletion also encompasses exon 2, which contains the start of the Agouti protein (amino acids 1–54). Together, this evidence strongly suggests that the a^Δ125kb allele causes melanism in P. maniculatus from New Hampshire.

Sequencing of Agouti and Mc1r coding regions in melanic individuals from other geographic locations shows that melanism arose independently at least three times in P. maniculatus. Melanic individuals from Shrubby Island, AK are homozygous for an allele (a^Q65term) of Agouti that contains a premature stop codon in exon 3. This mutation is predicted to result in a non-functional protein. Although we cannot rule out contributions of linked variation to the melanic phenotypes possessed by mice from New Hampshire and Alaska, given the likely effects of the Δ125kb and Q65term mutations and the known effects of null Agouti alleles in other taxa, it is very likely that these mutations represent the causative variation underlying these melanic phenotypes. The melanic individual from Michigan possesses neither the a^Δ125kb allele nor the a^Q65term allele; melanism in this population must be caused either by variation at another locus or possibly by unexamined variation at the Agouti or Mc1r loci.

This study presents two cases in which a specific molecular variant at the Agouti locus appears to cause melanism in a natural population. Mc1r mutants represent the vast majority of cases of melanism in natural populations of mammals, despite many occurrences of melanic Agouti mutants in captive and domestic stocks (Table 1). There are a number of possible explanations for this discrepancy.

One possible explanation involves dominance. Haldane [23] suggested that, when natural selection acts on new (i.e., rare) beneficial mutations, adaptation will be biased toward fixing dominant alleles, which are immediately visible to selection (but see [24]). Thus, we expect that when melanism is adaptive, we may see a prevalence of melanic Mc1r mutants. On the other hand, if melanism is deleterious and is being held at mutation-selection equilibrium, we might expect melanism caused by mutations in Agouti if they are recessive. Thus, depending on environmental conditions, expectations regarding the fixation probabilities of Mc1r versus Agouti alleles are different. In Peromyscus, the melanic alleles in both populations described in this study were found at low frequencies – 3–7% assuming Hardy-Weinberg equilibrium ([17]; data not shown) – and there is no obvious association between melanism and environmental conditions as observed in other species (e.g., pocket mice; [25]), suggesting these alleles may not be adaptive. Thus, if melanic phenotypes are often fixed from new dominant mutations rather than standing genetic variation, this may explain the prevalence of melanism caused by Mc1r.

Second, if mutations in Agouti have greater negative pleiotropic effects than mutations in Mc1r, then we would expect to see more evolution in the latter. Having fewer negative pleiotropic consequences of mutations at a locus translates to less evolutionary constraint (or higher net selection coefficients). While deleterious effects may be tolerated when organisms are raised in captivity, they could have important fitness consequences in nature. Whether differing amounts of pleiotropy of mutations at these loci affects the evolution of melanism is difficult to say, because mutations in both Agouti and Mc1r may affect traits other than pigmentation. Mutations in Mc1r, for example, have recently been discovered to have effects in the nervous system [26]. Pleiotropy is especially well documented in Agouti: ectopic expression of Agouti in Mus can result in obesity and lethality [27], [28] and null mutants in Rattus and Peromyscus exhibit behavioral differences [29], [30]. But pleiotropic consequences may be mitigated by the precise type and location of mutations. It has been predicted that for any given gene, mutations in the cis-regulatory elements may minimize antagonist pleiotropic effects relative to those in coding regions because such mutations can alter the time or place of gene expression in some tissues while preserving gene function in others [31]–[33]. Our data provide examples of mutations that are associated with morphological diversity: in one case, a premature stop codon, and in a second, a large deletion of both regulatory and exonic DNA. Thus, our data show, despite potential pleiotropic effects, both cis-regulatory and coding mutations in a highly pleiotropic gene, Agouti, cause a visible melanic phenotype that segregates in natural populations. Alternatively, it is possible that the melanic alleles in this study do generate negative pleiotropic effects that prevent them from increasing in frequency.

The third possibility is that a bias exists toward detecting mutations in the small Mc1r locus versus the larger, more complex Agouti locus. In fact, one would expect that there are more possible mutations that can cause a null Agouti allele than a constitutively active Mc1r allele. Many cases of melanism that have not yet been assigned a precise mutational cause (e.g., some populations of pocket mice [34]; pocket gophers [35]; leaf warblers [36]) may be caused by variation at Agouti, or indeed other loci.

Understanding the genetic basis of phenotypes that have arisen independently underpins studies of convergence by natural selection. While the fitness consequences of the melanic phenotypes in this study are unknown, studies of pigmentation may be uniquely positioned to identify convergence and to uncover its molecular basis because pigmentation traits are easily recognizable and many of the genes involved in producing pigments are well characterized. As the number of cases of convergence on a particular phenotype increases, so does our understanding of the constraints limiting the ways that phenotypes can evolve. In some cases, like stomach lysozyme [37], [38], pelvic reduction in sticklebacks [39], [40], or cyclodiene resistance in a number of insect taxa (reviewed in [41]), evolution appears to be tightly constrained, and the same gene is the repeated target of natural selection. In other cases, such as pigmentation, many different genetic mechansims can produce the same phenotype (beach mice [42], [43]; pocket mice [34]; Drosophila [44]; cavefish [45], [46]; Heliconius [47]). However, in these cases and others, it seems that a handful of proteins at key regulatory points in the pigmentation pathway are major targets of evolution change (e.g., Mc1r/Agouti in vertebrates; ebony/yellow in Drosophila; DFR in flowering plants [48]) Thus, natural selection may repeatedly target either the same key points in a genetic pathway or even the same genes to produce the most beneficial phenotype while minimizing deleterious pleiotropy. Future work on additional phenotypes in additional taxa will shed light on the myriad ways that evolution can generate morphological diversity.