Pigment pattern evolution by differential deployment of neural crest and post-embryonic melanophore lineages in Danio fishes

Neural crest cells give rise to many of the shared, derived traits of vertebrates (Gans and Northcutt,1983; Hall, 1999). These cells arise along the dorsal neural tube during neurulation and then disperse widely throughout the embryo(Knecht and Bronner-Fraser,2002; Halloran and Berndt,2003). Among the cells and tissues derived from this transient,migratory population are pigment cells, glia and neurons of the peripheral nervous system, endocardial cushion cells, chromaffin cells of the adrenal gland, smooth muscle, and bone and cartilage of the craniofacial skeleton(Hörstadius, 1950; Le Douarin, 1999). Not surprisingly, in light of their many derivatives, neural crest cells are associated with a wide range of inherited and acquired disorders ranging from melanoma to neuroblastoma, Hirschsprung disease to Waardenburg syndrome, and Treacher Collins syndrome to craniofacial dysmorphogenesis following fetal ethanol exposure (Matthay,1997; Amiel and Lyonnet,2001; Ahlgren et al.,2002; Chin, 2003; Widlund and Fisher, 2003; Farlie et al., 2004). Changes in the patterning of neural crest cells and their derivatives are similarly thought to underlie much of vertebrate diversity, from variation in pigment pattern to variation in jaw morphology(Kelsh, 2004; Kulesa et al., 2004).

Given the biomedical and evolutionary significance of neural crest cells and their derivatives, it is of paramount importance to identify the mechanisms by which these cells are patterned to generate the particular forms expressed by juveniles and adults. Most studies have focused on the early patterning of neural crest cells during embryogenesis. Yet, recent studies have demonstrated post-embryonic neural crest-derived stem cells in peripheral nerves, gut and skin (Morrison et al.,1999; Bixby et al.,2002; Kruger et al.,2002; Nishimura et al.,2002; Iwashita et al.,2003; Sieber-Blum and Grim,2004; Sieber-Blum et al.,2004; Joseph et al.,2004). These findings suggest that the development and maintenance of adult traits, as well as the evolution of these traits, may depend on contributions from latent stem cells in addition to direct contributions from neural crest cells at embryonic stages.

A useful system for studying the development and evolution of neural crest-derived traits is the pigment pattern of teleost fishes(Quigley and Parichy, 2002; Parichy, 2003; Kelsh, 2004). In the zebrafish Danio rerio, an early larval pigment pattern develops during embryogenesis as neural crest cells differentiate into early larval melanophores and other pigment cell classes. This pattern is largely completed by 3 days post-fertilization (dpf), and includes melanophore stripes along the dorsal and ventral edges of the myotomes, and along the horizontal myoseptum(Milos and Dingle, 1978a; Kelsh et al., 2000). The early larval pigment pattern remains essentially unchanged for about two weeks,until the onset of pigment pattern metamorphosis. At this time, new melanophores appear over the flank in regions not previously occupied by these cells, and during the following two weeks, the pigment pattern is transformed into that of the adult (Fig. 1)(Kirschbaum, 1975; Johnson et al., 1995; Parichy et al., 2000b). Genetic and cellular analyses demonstrate that new melanophores arising at metamorphosis differentiate from latent precursors or stem cells of presumptive neural crest origin (Johnson et al., 1995; Parichy and Turner, 2003b); such melanophores also play a crucial role in pigment pattern regeneration (Goodrich and Nichols, 1931; Rawls and Johnson, 2000; Rawls and Johnson, 2001). Although previous studies provide compelling evidence that metamorphic and regenerative melanophores are derived from post-embryonic latent precursors, specific markers for these cells have not been demonstrated, and their locations, potencies and developmental requirements remain largely unknown. Given these caveats, we refer to these post-embryonic melanophores simply as `metamorphic' melanophores.

The diversity of pigment patterns in species closely related to D. rerio allows the use of comparative and experimental approaches to understand pigment pattern development and evolution(Parichy and Johnson, 2001; Quigley and Parichy, 2002; Parichy, 2003). Besides horizontal stripes of varying width and number, these species exhibit vertical bars and uniform pigment patterns in which melanophores are dispersed evenly over the flank (Fig. 1). This interspecific variation provides an opportunity to dissect the genetic and cellular mechanisms underlying naturally occurring phenotypic variation in an adult trait of neural crest origin.

In this study, we ask whether adult melanophore stripes develop similarly across species, and in particular, whether the relative roles of neural crest-derived early larval melanophores and metamorphic melanophores have been maintained during evolution. To address this question, we first examine D. nigrofasciatus (Fig. 1), a species having fewer melanophores and stripes than D. rerio, but in which stripes that do develop are similar to those of D. rerio. Whereas stripes in D. rerio arise almost entirely from metamorphic melanophores, we show that stripes in D. nigrofasciatus arise from fewer metamorphic melanophores and an increased number of neural crest-derived early larval melanophores that persist into the adult. This interspecific variation led us to test the relative roles of these melanophore lineages during pigment pattern development in several additional species. These analyses demonstrate that a primary role for metamorphic melanophores in adult pigment pattern formation is likely to be ancestral for Danio, and that D. nigrofasciatus exhibits a unique, derived reduction in these cells, with a corresponding increased contribution of early larval melanophores to the adult pigment pattern. We further demonstrate that evolutionary changes within D. nigrofasciatus are extrinsic(non-autonomous) to the melanophore lineages, and we identify a candidate genetic pathway for mediating this change. These analyses highlight the potential for studies of D. rerio and its relatives to reveal basic mechanisms of post-embryonic neural crest development.

Fish were reared at 28.5°C (14 hours light: 10 hours dark). Wild-type D. rerio were the inbred mapping strain AB^ut, or an outbred wild-type stock representing mixed AB^ut, wik^ut,commercially derived `ekkwill', and other backgrounds. No differences in development were observed between wild-type D. rerio strains. D. nigrofasciatus, D. choprae, Devario shanensis and Tanichthys albonubes were derived from stocks purchased originally from a commercial pet supplier (Transship Discounts, Jamaica, NY). D. albolineatus and D. kerri were derived from stocks originally provided by M. McClure(Cornell University). D. `hikari' (used in phylogeny reconstruction)was obtained commercially and resembles D. kerri but has not been described formally. D. kyathit also was obtained commercially, but has more complete stripes than the type described(Fang, 1998), and may represent a variant or subspecies; for ease of presentation we refer to these fish simply as `D. kyathit' in the text, but to acknowledge the uncertainty of their precise taxonomic affinity, we refer to the fish as`D. aff. kyathit' in the phylogram. D. reriomutants have already been described: albino^b4(Kelsh et al., 2000), sox10 (colourless)(Dutton et al., 2001), endothelin receptor b1 (ednrb1, rose^b140)(Parichy et al., 2000a), tfap2a (lockjaw^ts213)(Knight et al., 2003; Knight et al., 2004), mitfa (nacre^w2)(Lister et al., 1999), and puma^j115e1 (Parichy and Turner, 2003b; Parichy et al., 2003). Additional D. rerio mutants were derived from on-going mutagenesis screens (D.M.P., E.H.B. and E.L.M., unpublished). Interspecific complementation tests were performed as previously described(Parichy and Johnson, 2001) by in vitro fertilization. Because of difficulties obtaining fertilizable eggs from heterospecific danios, most complementation tests were performed using D. rerio females and heterospecific males. When the identities or map positions of D. rerio mutant loci were known, heterozygotes were used for generating interspecific hybrids, to randomize effects across unlinked loci, and progeny were genotyped for the presence or absence of the mutant allele by PCR (primers and diagnostic single nucleotide polymorphisms available on request). Finally, puma mutant D. rerio are temperature-sensitive, with growth rate-dependent pigment pattern defects at 25°C, moderate pigment pattern defects at 28.5°C, and more severe defects at 33°C (Parichy et al.,2003); tester puma hybrids were reared at the intermediate temperature of 28.5°C to avoid mortality owing to stresses at the higher temperatures.

Previous studies defined pigment pattern elements in D. rerio(Parichy and Johnson, 2001; Parichy and Turner, 2003b),including: early larval dorsal, lateral and ventral melanophore stripes (ELD, ELL, ELV); adult first-developing (primary) dorsal and ventral melanophore stripes (1D, 1V); and later-developing (secondary) dorsal and ventral melanophore stripes (2D, 2V). Additionally, between adult melanophore stripes are xanthophore-rich `interstripe' regions. For simplicity, we use the term `stripes' to refer exclusively to the adult primary melanophore stripes (1D, 1V), unless indicated otherwise.

To examine melanophore behavior, we repeatedly imaged individual larvae during pigment pattern metamorphosis, allowing us to follow the appearance,disappearance and migration of individual melanophores(Parichy et al., 2000b; Parichy and Turner, 2003b). Individually reared fish were anesthetized with MS222 (Sigma) and imaged every 24 hours using an Olympus SZX-12 stereozoom microscope. To ensure that we could follow cells at the edges of the flank, all fish were imaged lying parallel to the camera, and also on a specially constructed stand providing an angle 30° from normal. Images were transferred to Adobe Photoshop CS for analysis, in some cases in conjunction with the FoveaPro 3.0 image processing and analysis package (Reindeer Graphics).

Individual melanophores were tracked as previously described(Parichy et al., 2000b; Parichy and Turner, 2003b),with newly differentiated melanophores clearly distinguishable from pre-existing melanophores by their initially lighter melanization (and in some instances different color, see below). We identified individual melanophores present in the early larval pigment patterns, then examined the fates of these cells by examining their positions in sequential images. In following early larval melanophores through metamorphosis, we could not formally observe cell division in static image series, so we tracked only one presumptive daughter following likely mitoses. Thus, our counts and estimated proportions of early larval melanophore contributions to later stages in D. nigrofasciatusare conservative and may underestimate true values to some degree. For comparisons of early larval and metamorphic melanophore numbers between species, we defined an area of interest bounded anteriorly by the anteriormost anal fin ray insertion and posteriorly at two myotomes anterior to the caudal peduncle. We counted individual early larval melanophores unilaterally within this region. We determined total adult melanophore numbers within this region,either within the adult ventral primary melanophore stripe, if present, or at an equivalent dorsoventral position as observed in D. rerio, with a height defined arbitrarily as one-quarter the flank height, as measured at the anterior boundary. In final images from each individual, all melanophores in the region of interest were marked and counted, either by eye or by the Count plug-in of FoveaPro 3.0. Numbers of metamorphic melanophores were thus calculated as the difference between the total numbers of melanophores identified in the final images, and the numbers of melanophores that had been followed into the region of interest from early larval stages. Statistical analyses were performed with JMP 5.0.1a Statistical Software (SAS Institute,Cary, NC). Additional information on quantitative image analyses is available on request.

We transplanted cells between mid-blastula stage [3.3-3.8 hours post-fertilization (hpf)] D. rerio and D. nigrofasciatusembryos, using a Narishige IM-9B micrometer-driven microinjection apparatus mounted on a Narishige micromanipulator. We placed embryos in agar-lined dishes containing 10% Hanks solution plus 1% penicillin/streptomycin, and dechorionated embryos with fine forceps. We transplanted 20-100 cells into each recipient and reared chimeric individuals through adult stages. To identify donor D. rerio cells in D. nigrofasciatus hosts, we used donors that were transgenic for EGFP driven by a ubiquitously expressed D. rerio β-actin promoter, kindly provided by Ken Poss(Parichy and Turner, 2003a; Parichy et al., 2003). To identify donor D. nigrofasciatus melanophores in D. reriohosts, we used hosts mutant for albino or nacre(mitfa), which fail to develop melanin and melanophores,respectively. Both of these mutant loci normally act autonomously to the melanophore lineage, as revealed previously(Lin et al., 1992; Lister et al., 1999; Parichy and Turner, 2003a) and confirmed in control experiments performed for the present analyses (data not shown). Previous studies reveal minimal local correlation between the distributions of pigment cells and other tissues in genetic mosaics examined at metamorphic and adult stages(Maderspacher and Nusslein-Volhard,2003; Parichy and Turner,2003a; Parichy et al.,2003). We confirmed that donor melanophores typically develop independently of other local donor tissues in a subset of chimeras in which donor embryos were injected with rhodamine dextran prior to the four-cell stage, then were examined for the distribution of melanophores and other tissues at 4 dpf (data not shown). We sorted chimeras at 3 dpf for the presence or absence of donor melanophores, and as larvae approached metamorphosis, we repeatedly imaged individual larvae to follow the behavior of early larval melanophores and to assess the distribution of metamorphic melanophores. Survival rates for interspecific chimeras were typically 5-10%of that observed for comparable experiments involving only D. rerio(Parichy and Turner, 2003a; Parichy et al., 2003),suggesting some species incompatibilities; ∼1% of chimeras were informative for analyses of pigment pattern formation (see Results).

We used in situ hybridization to detect transcripts for melanophore lineage markers, as described previously (Parichy et al., 2000a; Parichy et al.,2000b; Parichy et al.,2003). Larvae were fixed briefly in 4% paraformaldehyde, 1% DMSO in PBS, decapitated, and then fixed overnight at 4°C. Larvae were transferred to methanol, rehydrated to PBST (PBS with 0.2% Tween-20), then treated for 20 minutes at room temperature with 20 μg/ml proteinase-K in PBST containing 1% DMSO. Larvae were postfixed for 20 minutes at room temperature in 4% paraformaldehyde, 0.005% glutaraldehyde, washed in PBST,then washed three times in hybridization solution lacking tRNA and heparin. Prehybridizations were performed overnight at 60°C in hybridization solution (50% formamide, 5×SSC, 500 μg/ml yeast tRNA, 50 μg/ml heparin, 0.2% Tween-20, 9.2 mM citric acid). Hybridizations were performed at 60°C over two nights, in fresh hybridization solution containing digoxigenin-labeled riboprobes fractionated to ∼300 nucleotides. Larvae were then washed twice, for 15 minutes each, in 2×SSCT, and three times,for 2 hours each, in 0.2×SSCT at 60°C. After graded changes to PBST,larvae were blocked overnight at 4°C in 2 mg/ml BSA, 5% heat-inactivated calf serum in PBST, then incubated at 4°C over two nights in fresh blocking reagent containing 1:5000 anti-digoxigenin alkaline phosphatase-conjugated Fab fragments (Roche). Larvae were washed over two nights in PBST, transferred to alkaline phosphatase buffer [100 mM Tris (pH 9.5), 50 mM MgCl₂, 100 mM NaCl, 0.1% Tween-20], and the color developed with NBT/BCIP.

To assay for tyrosinase activity, larvae were fixed for 2 hours in 4%paraformaldehyde in PBS, rinsed three times in PBS, incubated in 0.1%l-dopa (Sigma) for 1 hour to overnight, rinsed in PBS, then stored in glycerol (Camp and Lardelli,2001; McCauley et al.,2004). We verified the specificity of the assay for melanoblasts by the reduced staining on the flanks of metamorphosing nacre mutant D. rerio, which have defects in the melanophore lineage(Lister et al., 1999; Parichy et al., 2000b), and we verified that newly melanized (tyrosinase⁺) cells are not macrophages by Neutral Red staining(Herbomel et al., 1999) (data not shown).

We reconstructed phylogenetic relationships based on mitochondrial 12S and 16S rDNA sequences, obtained using standard methods and universal primers(Kocher et al., 1989; Palumbi et al., 1991).

12S: H1478, 5′-TGA CTG CAG AGG GTG ACG GGC GGT GTG T-3′; L1091,5′-AAA AAG CTT CAA ACT GGG ATT AGA TAC CCC ACT AT-3′.

16S: 16Sar-L, 5′-CGC CTG TTT ATC AAA AAC AT-3′; 16Sbr-H,5′-CCG GTC TGA ACT CAG ATC ACG T-3′.

Sequences were aligned using CLUSTAL-W, inspected by eye and edited as necessary. We then analyzed combined 12S and 16S sequences (784 nucleotides)using maximum likelihood estimation in PAUP^* 4.0b10 for Macintosh(Swofford, 2002). Maximum likelihood analyses used a general time-reversible plus gamma model. Substitution rate matrix, nucleotide frequencies, and among site rate variation were estimated from the data by preliminary parsimony analyses using a heuristic search strategy. Maximum likelihood, parsimony and distance methods produced trees with the same topology. To estimate confidence values for reconstructed nodes, we performed two independent analyses. First, we performed 100 nonparametric bootstrap replicates using PAUP^*. Second, we performed a Bayesian analysis of the data using MrBayes(Larget and Simon, 1999; Huelsenbeck and Ronquist,2001; Wilcox et al.,2002), with 3000 replicate trees from 300,000 generations following the approach to asymptotic likelihood values. Both approaches gave nearly identical confidence values, which we report as percentages of recovered trees in the phylogram (see Results).

To assess the generality of adult pigment pattern-forming mechanisms, we investigated whether stripes of different Danio species arise through similar underlying cellular behaviors. We chose to compare the closely related species D. rerio and D. nigrofasciatus(Parichy and Johnson, 2001),for which stripes on the flank are superficially similar(Fig. 1).

Closer inspection reveals about twice as many melanophores in D. rerio than in D. nigrofasciatus(Fig. 2A,C; see below). Melanophore colors differ as well. In D. rerio, the dorsal and ventral stripes consist almost entirely of grey-black melanophores. Yet,occasional brownish melanophores occur at the ventral edge of the dorsal stripe (Fig. 2A,B), where a few melanophores derive not from latent precursors at metamorphosis, but from the rearrangement of embryo-derived melanophores originally present in the early larval lateral stripe along the horizontal myoseptum(Parichy and Turner, 2003b). In D. nigrofasciatus, however, both dorsal and ventral stripes contain numerous brown melanophores (Fig. 2C,D), and melanophores are not present along the ventral myotome edge (where the early larval ventral stripe had been). Melanophore color variation is apparent transiently after metamorphosis, and is not equally pronounced in all families; whether this variation reflects the age of the melanin contained within the cells or some other biochemical difference is not clear. Nevertheless, the differences in melanophore colors and their relative frequencies in the adult pigment patterns of D. rerio and D. nigrofasciatus led us to hypothesize that cryptic patterning variation might underlie the superficially similar stripes between these species.

To determine whether stripes develop differently in D. rerio and D. nigrofasciatus, we imaged larvae repeatedly through pigment pattern metamorphosis, allowing us to follow the appearance, disappearance and movements of individual melanophores(Parichy et al., 2000b; Parichy and Turner, 2003b). These analyses demonstrate that D. rerio stripes depend almost entirely on melanophores that differentiate from latent precursors at metamorphosis. By stark contrast, the superficially similar stripes of D. nigrofasciatus arise by the reorganization of early larval melanophores,with far fewer metamorphic melanophores.

In D. rerio, the onset of pigment pattern metamorphosis is marked by the differentiation of single `pioneer' metamorphic melanophores over the middle of most ventral myotomes (Fig. 3A). Subsequently, metamorphic melanophores differentiate widely over the myotomes, between the early larval stripes(Fig. 3B,C). The adult primary stripes become increasingly apparent (Fig. 3D), as initially dispersed metamorphic melanophores migrate short distances to the sites of stripe formation, and as additional metamorphic melanophores differentiate within the stripes themselves(Fig. 3C,D). A few early larval melanophores migrate from the horizontal myoseptum to join the dorsal adult primary melanophore stripe (Fig. 3D,E), but most remain in place and eventually are lost(Parichy and Turner, 2003b). As fish approach the end of metamorphosis, a juvenile pattern emerges, with adult dorsal and ventral primary melanophore stripes consisting almost entirely of melanophores that have differentiated from latent precursors during metamorphosis (Fig. 3E,E′).

In D. nigrofasciatus, pioneer metamorphic melanophores are absent and the onset of pigment pattern metamorphosis is marked by a loss of cohesiveness within the early larval stripes(Fig. 3F). Gradually, early larval melanophores are visible more laterally over the myotomes;rearrangements of melanophores relative to one another indicate that movements result in part from migration (as opposed to passive movement due to growth; Fig. 4). A few metamorphic melanophores differentiate but these are drastically reduced in number compared with in D. rerio (Fig. 3G). As metamorphosis proceeds, early larval melanophores are increasingly observed at sites of adult stripe formation, and additional metamorphic melanophores differentiate at these sites(Fig. 3H). As compared with D. rerio, the D. nigrofasciatus dorsal stripe occurs at approximately the same dorsal-ventral location, whereas the ventral stripe lies closer to the ventral edge of the myotomes(Fig. 3E′,J′). By the end of pigment pattern metamorphosis, the pattern consists of dorsal and ventral stripes comprising a relatively high proportion of early larval melanophores (Fig. 3I,J). Differences in pigment pattern metamorphosis between species are particularly evident in animations compiled from multiple images of individual larvae (see Movies 1, 2 in supplementary material).

Our analyses demonstrate that superficially similar stripes in two closely related species develop in markedly different ways: in D. rerio,stripes arise almost entirely by the differentiation of melanophores from latent precursors during metamorphosis; in D. nigrofasciatus, stripes arise largely through the reorganization of early larval melanophores, with metamorphic melanophores playing a smaller role.

The relative contributions to adult pigment patterns of early larval melanophores and metamorphic melanophores could vary continuously across species. Alternatively, either the D. rerio or the D. nigrofasciatus mode could be typical. To distinguish between these possibilities, and to determine which, if either, mode is ancestral and which is derived, we sought to examine pigment pattern metamorphosis in additional species.

Because danio relationships remain poorly understood, we first sequenced 12S and 16S rDNA from additional taxa to infer phylogenetic relationships(Fig. 5). These analyses confirm the close relationship between D. rerio and D. nigrofasciatus, as well as D. kyathit(Fig. 1). The phylogeny also supports a split between Danio and Devario [formerly within Danio (Fang, 2003)]. Moreover, these data reveal additional pigment pattern diversity within Danio (as defined in Fig. 5): these fish have been known to have horizontal stripes, spots,uniform patterns, and more complex pigment patterns; Danio chopraeadds vertical barring to the repertoire(Fig. 1).

Our analyses of additional species demonstrate that adult pigment patterns form primarily by the recruitment of metamorphic melanophores, rather than by the reorganization of early larval melanophores. Using inferred phylogenetic relationships as a guide (Fig. 5), we chose additional species representing Danio and more distant groups, as well as a range of adult pigment patterns, and we repeatedly imaged larvae through metamorphosis. Fig. 6 shows the ventral flank for representative larvae of D. nigrofasciatus, D. rerio, D. kyathit, D. kerri, D. albolineatus, D. choprae, and Tanichthys albonubes. In each species, the early larval pigment patterns are nearly identical(Fig. 6, row 1; see also Fig. S1 in supplementary material). Subsequently, melanophores differentiate from latent precursors during metamorphosis to generate the adult pigment pattern(Fig. 6, rows 2-8). Only in D. nigrofasciatus do large numbers of early larval melanophores become incorporated into adult stripes; a few early larval melanophores move small distances from their initial positions in D. kerri, D. albolineatus,D. choprae, D. shanensis (data not shown) and T. albonubes, but these do not contribute substantially to the adult pattern.

The greater persistence of early larval neural crest-derived melanophores into the adult pigment pattern of D. nigrofasciatus compared with other species is shown quantitatively in Fig. 7A. Total numbers of early larval melanophores do not differ dramatically across the species, yet many more of these cells contribute to the adult pigment pattern in D. nigrofasciatus. Variation in the total numbers of adult melanophores across species is shown in Fig. 7B. D. nigrofasciatus and D. albolineatus (and to a lesser extent D. kerri) each have reduced numbers of metamorphic melanophores compared with other species. However, only in D. nigrofasciatus is this deficit in metamorphic melanophores partly compensated by a significantly greater contribution from early larval neural crest-derived melanophores.

Adult pigment patterns of these species thus arise principally from metamorphic melanophores with little contribution from early larval melanophores. This finding supports a model in which Danio adult pigment patterns depend ancestrally on metamorphic melanophores, with D. nigrofasciatus exhibiting an evolutionarily derived reduction in metamorphic melanophores and a correspondingly greater role for neural crest-derived early larval melanophores in generating the adult pigment pattern.

D. rerio mutants can identify genes and pathways that contribute to interspecific pigment pattern differences(Parichy and Johnson, 2001). Given the reduced number of metamorphic melanophores in D. nigrofasciatus compared to D. rerio, we investigated whether genes isolated as D. rerio mutants with defects in melanophore development also contribute to the difference between species. We used interspecific hybrids to test for complementation of D. rerio mutant alleles by crossing mutant D. rerio to D. nigrofasciatus and comparing these tester (mutant) hybrids to control (wild-type) hybrids. Tester hybrids exhibiting fewer melanophores than controls identify genes that may contribute to the interspecific difference, whereas tester hybrids that have similar melanophore numbers to controls identify genes less likely to have major effect roles.

Control hybrids between wild-type D. rerio and D. nigrofasciatus have phenotypes intermediate between species. Whereas melanophore numbers in primary adult stripes are increased over D. nigrofasciatus and are closer to D. rerio, melanophore numbers in secondary adult stripes, and the total numbers of stripes, are closer to D. nigrofasciatus than D. rerio(Fig. 8A)(Parichy and Johnson, 2001). Comparing adult hybrid phenotypes does not reveal gross non-complementation of the recessive melanophore mutants sox10^ut.r13e1,tfap2a^ts213, bonaparte^ut.r16e1,cezanne^ut.r17e1, degas^ut.r18e1, oberon^j198e1, pissarro^ut.r8e1,picasso^ut.r2e1, primrose^j199, puma^j115e1or seurat^ut.r15e1 (e.g. Fig. 8), adding to the previously excluded loci ednrb1, fms, kit, mitfa, leopard, fritz and jaguar (Parichy and Johnson,2001). Thus, genes contributing to the differences in the final numbers of adult melanophores between species either are not likely to be represented in this collection of 18 D. rerio pigment pattern mutants, or differences in allelic strengths are not sufficient to reveal non-complementation.

Although the preceding analyses tend to exclude loci from roles in reducing the total numbers of adult melanophores in D. nigrofasciatus, they did not explicitly address differences in early larval melanophore morphogenesis or contributions to the adult pigment pattern. To identify genetic pathways that might be responsible for interspecific differences in early larval melanophore fate and adult stripe composition, we further surveyed D. rerio mutants for morphogenetic similarities to D. nigrofasciatus.

We first investigated whether any of several D. rerio mutants exhibiting stripes dorsally and spots ventrally, as in D. nigrofasciatus, might have similar modes of pigment pattern metamorphosis to D. nigrofasciatus. Examination of one of these mutants, ednrb1 (Parichy et al.,2000a), revealed little contribution of early larval melanophores to the adult pigment pattern, unlike in D. nigrofasciatus(Fig. 9A-D, and data not shown). Thus, a similarity of pigment pattern elements does not predict the underlying mode of pigment pattern metamorphosis.

We next examined additional D. rerio mutants for similarities to D. nigrofasciatus. We identified two recessive D. reriomutants, puma and picasso, with severe deficits in metamorphic melanophores, as well as increased persistence of early larval melanophores into the adult (Fig. 9E-L) (Parichy and Turner,2003b). To see whether puma or picassocontribute to the difference between species, we tested whether early larval melanophore fates and adult stripe compositions in tester hybrids differ from those of control hybrids. Repeated imaging of hybrid larvae reveals that early larval melanophores make little contribution to the adult stripes in control(wild-type) hybrids (Fig. 9M-P; n=9), tester picasso hybrids, or tester hybrids for two additional mutants, pissarro and seurat (n=4,11 and 6, respectively; data not shown). In tester puma hybrids(n=9), however, a greater number of early larval melanophores leave their initial positions along the myotome edges and relocalize over the flank(Fig. 9Q-T), similar to puma mutants and D. nigrofasciatus(Fig. 9E-L, Fig. 3F-J). These hybrids have somewhat reduced numbers of metamorphic melanophores initially, but this deficit is regulated by adult stages so that adults have total melanophore complements that are indistinguishable from control hybrids. The developmental non-complementation phenotype of tester puma hybrids identifies puma or genes in its pathway as candidates for contributing to the evolutionary reduction of metamorphic melanophores, and increased contribution of early larval melanophores to the D. nigrofasciatus adult pigment pattern.

These results indicate that differences in total numbers of adult melanophores between D. rerio and D. nigrofasciatus are not likely to be explained by differences at loci already isolated as D. rerio melanophore mutants. Moreover, similarity of adult pigment pattern alone is not a good predictor for the underlying mode of pigment pattern metamorphosis. By contrast, interspecific complementation tests for melanophore morphogenesis suggest a role for puma or its pathway in determining the relative contributions of metamorphic melanophores and neural crest-derived early larval melanophores to the adult pigment patterns of D. rerio and D. nigrofasciatus.

The reduction in metamorphic melanophores in D. nigrofasciatuscould reflect a failure to recruit committed melanophore precursors(melanoblasts) from uncommitted latent precursors or stem cells during metamorphosis. For example, puma mutant D. rerio exhibit severe reductions in metamorphic melanoblasts compared with wild-type D. rerio (Parichy et al.,2003). If the same pathway affected in puma mutant D. rerio has evolved between D. rerio and D. nigrofasciatus, then fewer melanoblasts should be observed in D. nigrofasciatus compared with wild-type D. rerio. Alternatively,fewer metamorphic melanophores in D. nigrofasciatus could reflect a later block in this lineage, with similar numbers of melanoblasts being recruited from latent precursors then failing to terminally differentiate as melanophores. To distinguish between these possibilities, we used molecular markers and histological assays to compare D. rerio and D. nigrofasciatus during metamorphosis.

Our examination of the melanophore lineage during metamorphosis reveals a severe reduction in the number of melanoblasts in D. nigrofasciatus,suggesting an early block in metamorphic melanophore development. We examined the distribution of cells expressing transcripts for two molecular markers, dopachrome tautomerase (dct) and tyrosinase(tyr), which encode enzymes required for melanin synthesis and thus identify committed melanophore precursors (as distinct from latent stem cells)(Kelsh et al., 2000; Camp and Lardelli, 2001). We observed fewer dct⁺ and tyr⁺ cells throughout metamorphosis in D. nigrofasciatus compared to D. rerio (Fig. 10). Importantly, however, we observed strong staining for each marker in fully differentiated melanophores, and in the more rare, unmelanized cells, in D. nigrofasciatus, demonstrating the efficacy of these probes in this cross-species comparison.

From molecular marker analyses, we predicted that metamorphosing D. nigrofasciatus would exhibit fewer cells that were competent to produce melanin when provided with exogenously supplied L-dopa, which is converted to melanin in cells expressing functional tyrosinase(McCauley et al., 2004). Comparing the same regions of the flank before and after L-dopa incubation, we observe newly melanized cells in D. rerio larvae, but only ∼10%as many of these cells in D. nigrofasciatus larvae(Fig. 11). These findings support the conclusion that D. nigrofasciatus exhibit fewer melanoblasts during metamorphosis than D. rerio do.

Together, these analyses demonstrate that reduced numbers of metamorphic melanophores in D. nigrofasciatus result from an early block in the specification of these cells or their recruitment from latent precursors,rather than from a late block in their terminal differentiation as melanophores.

The different modes of pigment pattern metamorphosis in D. rerioand D. nigrofasciatus could reflect evolutionary changes that are intrinsic (autonomous) or extrinsic (non-autonomous) to melanophore lineages. Although species differences have been attributed to intrinsic factors(Twitty and Bodenstein, 1939; Rawles, 1948; Schneider and Helms, 2003),the extensive migrations and cellular interactions during neural crest and melanophore development imply many opportunities for extrinsic factors to generate differences in form as well(Erickson and Perris, 1993; Parichy, 1996; Halloran and Berndt, 2003). To distinguish between these possibilities, we examined melanophore behaviors and patterns in genetic mosaics. These analyses demonstrate a primary role for extrinsic factors in determining early larval melanophore contributions to adult stripes, as well as the positions of adult stripes on the flank.

We transplanted cells from D. nigrofasciatus to D. rerio,and then reared chimeras through metamorphosis(Parichy and Turner, 2003a; Parichy et al., 2003). To identify donor D. nigrofasciatus melanophores, we used D. rerio hosts mutant for the albino locus, which acts autonomously to the melanophore lineage to promote melanization, but does not otherwise affect melanophore development or pigment pattern formation(Lin et al., 1992); D. nigrofasciatus melanophores thus appear as the only melanized cells in a field of unmelanized but otherwise normal melanophores(Lin et al., 1992; Parichy et al., 1999; Kelsh et al., 2000). To assess the mode of pigment pattern metamorphosis, we identified chimeras that developed D. nigrofasciatus early larval melanophores, then we repeatedly imaged these individuals through metamorphosis.

We predicted that if species differences are autonomous to the melanophore lineages, then donor D. nigrofasciatus early larval melanophores should contribute to the adult ventral melanophore stripe (as in D. nigrofasciatus); if differences between species are non-autonomous to the melanophore lineages, then donor D. nigrofasciatus early larval melanophores should fail to contribute to this stripe (as in D. rerio). Fig. 12A-D shows a representative D. nigrofasciatus→D. rerio chimera. Donor D. nigrofasciatus early larval melanophores are present within the early larval stripe along the ventral myotomes but do not contribute to the adult ventral stripe. Thus, early larval melanophore morphogenesis resembles that of D. rerio rather than that of D. nigrofasciatus (compare with Fig. 3). Moreover, D. nigrofasciatus melanophores that differentiated during metamorphosis did so at the normal location of D. rerio stripes, rather than further ventrally as in D. nigrofasciatus (compare with Fig. 3). These findings indicate that factors non-autonomous to the melanophore lineages determine species differences in early larval melanophore contributions to adult stripes, as well as the positions of adult stripes. These results also obviate the identification of other donor D. nigrofasciatus cells in D. rerio hosts, as the final distributions of donor melanophores cannot easily be explained by a simple coincidence of D. nigrofasciatus melanophores and other D. nigrofasciatus donor tissues (which might have explained the alternative result, had donor melanophores behaved like their own, donor species, rather than the host species).

In reciprocal D. rerio→D. nigrofasciatus chimeras(n>40), we never observed donor D. rerio melanophores in D.nigrofasciatus hosts, although we frequently observed other donor tissues (e.g. myotomes, nerves, skin; data not shown). Similar studies in D. rerio yielded comparable frequencies of these tissues,but also melanophores (Parichy and Turner,2003a; Parichy et al.,2003). These data suggest that D. nigrofasciatus hosts may be compromised in their ability to support D. rerio donor melanophores, supporting a model in which species differences in metamorphic melanophore differentiation are non-autonomous to this lineage as well.

Finally, we investigated whether evolutionary changes in interactions between melanophores themselves might contribute to the different metamorphic modes between species. We reasoned that a reduction in the numbers of metamorphic melanophores, and thus reduced contact inhibition of movement(Tucker and Erickson, 1986),might allow early larval melanophores to leave their initial positions during metamorphosis in D. nigrofasciatus. To test this possibility, we transplanted D. nigrofasciatus cells to nacre mutant D. rerio hosts. nacre mutants lack melanophores owing to an inactivating mutation in mitfa, which normally acts autonomously to the melanophore lineage (Lister et al.,1999). We predicted that if changes in melanophore-melanophore interactions alone are responsible for species differences, then D. nigrofasciatus early larval melanophores in nacre mutant hosts should contribute to the adult ventral stripe (as in D. nigrofasciatus). If other factors contribute to the species differences,the D. nigrofasciatus early larval melanophores should fail to contribute to this stripe (as in D. rerio). Fig. 12F-I shows a D. nigrofasciatus→nacre mutant D. rerio chimera. Repeated imaging demonstrates that donor D. nigrofasciatus early larval melanophores do not contribute to the adult ventral stripe, which forms at a position similar to that seen in D. rerio. These data demonstrate that factors extrinsic to melanophore lineages contribute to differences in pigment pattern metamorphosis between D. rerio and D. nigrofasciatus.

Our analyses provide new insights into the generalized features of adult pigment pattern metamorphosis in danios and their relatives, how these patterns evolve, and the derived mode of pigment pattern metamorphosis in D. nigrofasciatus. These results suggest a model relating early larval and adult pigment pattern formation in D. rerio and other species, and how these processes have been modified in D. nigrofasciatus (Fig. 13).

The cellular and genetic bases for adult form are largely unknown, but elucidating these mechanisms is crucial for a deeper understanding of human development and disease, as well as the evolution of morphology. Of particular interest are traits derived from the neural crest, given the many genetic disorders and cancers associated with these cells(Chin, 2003; Farlie et al., 2004), and their major role in vertebrate diversification(Gans and Northcutt, 1983; Hall, 1999). At least two extreme models can explain the forms taken by adult traits derived from the neural crest. First, the particular features of such traits could be determined entirely by patterning mechanisms during embryogenesis, while neural crest cells are migrating, or shortly after their arrival at target tissues. Second, forms taken by adult neural crest derivatives could depend not on neural crest cells themselves, but on the patterning of post-embryonic neural crest-derived latent precursors or stem cells resident in many tissues(Bixby et al., 2002; Kruger et al., 2002; Nishimura et al., 2002; Sieber-Blum and Grim, 2004; Sieber-Blum et al., 2004; Joseph et al., 2004).

Our results suggest a major role for latent precursors, presumptively of neural crest origin, during the development of adult pigment patterns in danios and their relatives (Fig. 13). Lineage analyses revealed that in each species (except D. nigrofasciatus), adult pigment patterns were formed principally by metamorphic melanophores derived from latent precursors, rather than by early larval melanophores derived from neural crest cells during embryogenesis. The prevalence of this mode of pigment pattern metamorphosis strongly suggests that this is a shared, ancestral trait for Danio. To our knowledge,this represents the first systematic survey across species to define the cellular origins for an adult neural crest-derived trait. Previous studies have demonstrated roles for melanocytes derived from stem cells in the development of mammalian pigmentation(Nishimura et al., 2002), and for melanophores derived from latent precursors in pigment pattern formation of some teleosts, including D. rerio(Johnson et al., 1995; Sugimoto, 2002; Parichy and Turner, 2003b). Latent precursors probably also generate adult pigment patterns of many amphibians (Parichy, 1998). Likewise, the adult epibranchial cartilage of the salamander Eurycea bislineata arises from a discrete population of cells in the perichondrium of the larval neural crest-derived epibranchial cartilage(Alberch and Gale, 1986). Given the presence of post-embryonic neural crest stem cells and specified latent precursors in a variety of tissues (see Introduction), it will be interesting to determine the extent to which other adult traits depend on these cells (as distinct from embryonic neural crest cells) for their initial patterning,maintenance, and repair after injury.

The comparative approach we have taken also implicates post-embryonic,latent precursors of presumptive neural crest origin in the generation of organismal diversity. We examined species exhibiting a variety of adult pigment patterns, including horizontal stripes that are compact (D. rerio,D. kyathit) or diffuse (D. kerri, T. albonubes), as well as vertical bars (D. choprae) and uniform patterns (D. albolineatus). Despite this variation in adults, the larvae of these species exhibit melanophore patterns that are indistinguishable from one another, except for small differences in melanophore numbers (see Fig. S1 in supplementary material; Fig. 7A). Our results demonstrate that much of the pigment pattern diversity of adults reflects interspecific variation in the differentiation and morphogenesis of metamorphic melanophores that are derived from latent precursors or stem cells, rather than the reorganization of embryonic neural crest-derived melanophores. That embryonic/early larval pigment patterns and adult pigment patterns depend on different melanophore lineages suggests a mechanism by which these pigment patterns may be relatively uncoupled across life-cycle stages. Thus, evolutionary responses to selection on the adult pigment pattern may be relatively unconstrained by features of the earlier developing embryonic/early larval pigment pattern, if genetic controls differ to some extent between neural crest-derived and metamorphic melanophore lineages (Haldane, 1932; Ebenman, 1992; Parichy, 1998). Indeed,several D. rerio pigment pattern mutants have defects limited to particular embryonic or metamorphic melanophore lineages(Johnson et al., 1995; Parichy et al., 1999; Parichy et al., 2000a; Parichy et al., 2000b; Parichy and Turner, 2003). Nevertheless, the extent of genetic independence across life-cycle stages – and its evolutionary consequences – remains an empirical question that deserves further analysis.

A central problem in evolutionary developmental biology is the extent to which similar phenotypes depend on the same or different underlying mechanisms. Several recent analyses have demonstrated the repeated,independent evolution of traits via common underlying genetic changes(Sucena et al., 2003; Mundy et al., 2004; Shapiro et al., 2004). Such cases of evolutionary parallelism suggest that pathways of evolutionary change in development may be more limited than classical evolutionary theory might suggest (Barton and Turelli,1989). By contrast, other analyses reveal divergent mechanisms underlying repeated trait evolution(Hoekstra and Nachman, 2003; Wittkopp et al., 2003). Despite this recent focus on traits that have evolved independently, we still know little about developmental variation underlying traits having a common evolutionary origin.

Our analyses reveal substantial differences in stripe development between D. rerio and D. nigrofasciatus, despite the superficial similarity of the final stripes that form, and the close phylogenetic relationship of these species. Cryptic patterning variation has been observed for other traits (Hall, 1984; Minsuk and Keller, 1996; Jungblut and Sommer, 2000),and argues for the importance of a comparative approach in validating conclusions gleaned from studies of model organisms(Parichy, 2005; Bolker, 1995; Metscher and Ahlberg, 1999). Such variation may reflect selection to maintain a particular adult phenotype,in the absence of selection for precisely how this phenotype is achieved. The roles of teleost pigment patterns in predation avoidance, mate recognition,mate choice and shoaling behavior suggests strong selection on adult phenotypes (Endler, 1983; Houde, 1997; Couldridge and Alexander,2002; Engeszer et al.,2004; Allender et al.,2003); the behavioral roles and selective consequences of early larval and metamorphic pigment patterns remain wholly unexplored.

Adult pigment pattern formation in D. nigrofasciatus differs from that of D. rerio in having a lesser contribution from metamorphic melanophores, and a correspondingly greater contribution from persisting early larval melanophores (Fig. 13). Thus, D. nigrofasciatus may be viewed as exhibiting a heterochronic change in pigment pattern development, with a relatively paedomorphic (or juvenilized) mode when compared with the inferred ancestral condition. This uncoupling of pigment pattern and somatic metamorphosis is somewhat similar to several species and subspecies of salamanders, in which adult spots and stripes appear during the larval stage, prior to somatic metamorphosis(Anderson, 1961; Anderson and Worthington, 1971; Parichy, 1998). Dissociability of pigment pattern and somatic metamorphosis may be a generalized feature of post-embryonic development in these ectothermic vertebrates.

The reduction of metamorphic melanophores and the persistence of early larval melanophores in D. nigrofasciatus also is reminiscent of several D. rerio mutants. This concordance highlights the utility of D. rerio mutants both for understanding developmental mechanisms within zebrafish, and for framing hypotheses that can be tested across species to dissect mechanisms of evolutionary change. In this study, we examined roles for several of these mutant loci in pigment pattern diversification using interspecific complementation tests(Parichy and Johnson, 2001). Only hybrids between D. nigrofasciatus and puma mutant D. rerio exhibited non-complementation phenotypes, with fewer metamorphic melanophores and increased early larval melanophores persisting into the adult pattern, when compared with control hybrids. This observation raises the possibility that puma activity differs between species,and might therefore explain the derived mode of pigment pattern metamorphosis in D. nigrofasciatus. Nevertheless, puma acts autonomously to the metamorphic melanophore lineage(Parichy et al., 2003),whereas interspecific genetic mosaics constructed in this study reveal differences that are non-autonomous to melanophore lineages (see below). Thus,it seems unlikely that variation at the puma locus itself contributes to these species differences. Rather, the non-complementation phenotype may reflect interspecific variation in a sensitized puma-dependent pathway. Consistent with this idea, our analyses demonstrate that both puma mutant D. rerio(Parichy et al., 2003) and D. nigrofasciatus exhibit far fewer metamorphic melanophore precursors than wild-type D. rerio. These findings imply a change in the early development or specification of the metamorphic melanophore lineage in D. nigrofasciatus. This early block differs from the situation found in several species of Astyanax cave fish(McCauley et al., 2004) and D. albolineatus (Quigley et al.,2005), in which melanophore numbers are reduced owing to a later block in this lineage, such that melanoblasts develop but then fail to differentiate or survive. Examination of additional species should allow a more complete reconstruction of the evolutionary history of melanophore patterns and melanophore lineage modifications across taxa.

Besides the reduction in metamorphic melanophores, D. nigrofasciatus exhibit a dramatic increase in the contribution of early larval melanophores to the adult pigment pattern. By examining melanophore behaviors and patterns in interspecific genetic mosaics, we demonstrate that factors non-autonomous to melanophore lineages determine the different behaviors of these cells between species. This contrasts with several studies that have identified changes autonomous to neural crest or melanophore lineages in determining species differences(Twitty and Bodenstein, 1939; Twitty, 1945; Rawles, 1948; Epperlein and Löfberg,1990; Schneider and Helms,2003), although some of these results are open to alternative interpretations (Parichy,1996; Parichy,2001). Our findings suggest that the relative roles of intrinsic and extrinsic factors differ both across species and across traits, and it is too simplistic to ascribe evolutionary changes to intrinsic factors alone.

At least two models can be suggested for the non-autonomous factors contributing to the differences in early larval melanophore morphogenesis between D. rerio and D. nigrofasciatus. First, early larval melanophores could reorganize in D. nigrofasciatus owing to the reduced numbers of metamorphic melanophores, which might otherwise prevent the cells from leaving their original positions within the early larval stripes by contact inhibition of movement (Tucker and Erickson, 1986). This explanation would suggest a relatively simple pattern regulatory process, such that early larval melanophores fill gaps resulting from reduced numbers of metamorphic melanophores(Milos and Dingle, 1978b). Our examination of genetic mosaics between D. nigrofasciatus and nacre mutant D. rerio, which lack melanophores, excludes the loss of interactions between metamorphic melanophores and early larval melanophores as being the sole factor underlying the difference between species. Second, early larval melanophores could reorganize owing to changes in other factors in the extracellular environment. Consistent with this idea,the adult ventral melanophore stripe develops closer to the myotome edge in D. nigrofasciatus than in D. rerio. Conceivably, D. nigrofasciatus early larval melanophores may be close enough to respond to stripe-forming cues that D. rerio early larval melanophores do not encounter because these cues are situated further dorsally on the flank. The difference between species also may reflect multiple partially redundant changes in factors extrinsic to the melanophore lineage. These possibilities are now being addressed by additional cell transplantation studies and by seeking the nature of the stripe-forming cues themselves.

Thanks to C. Lee for help rearing fish and E. Herrington for assistance with in situ hybridization. T. Wilcox provided valuable advice on phylogenetic methods and J. Wallingford provided comments on the manuscript. Mutant and transgenic stocks were generously provided by R. Knight, J. Lister, D. Raible,T. Schilling, S. Johnson and K. Poss, as well as by the Zebrafish International Resources Center at the University of Oregon. Supported by NIH R01 GM62182 to D.M.P.