SEARCH

SEARCH BY CITATION

Keywords:

  • Comparative analysis;
  • developmental pathways;
  • evolution of development;
  • parallel evolution;
  • pleiotropy

Abstract

  1. Top of page
  2. Abstract
  3. The Need for Metamodels in Evo–Devo
  4. Examples from Specific Metamodels
  5. Perspectives: What Makes a Good Metamodel?
  6. ACKNOWLEDGMENTS
  7. LITERATURE CITED
  8. Appendix

Molecular genetic analysis of phenotypic variation has revealed many examples of evolutionary change in the developmental pathways that control plant and animal morphology. A major challenge is to integrate the information from diverse organisms and traits to understand the general patterns of developmental evolution. This integration can be facilitated by evolutionary metamodels—traits that have undergone multiple independent changes in different species and whose development is controlled by well-studied regulatory pathways. The metamodel approach provides the comparative equivalent of experimental replication, allowing us to test whether the evolution of each developmental pathway follows a consistent pattern, and whether different pathways are predisposed to different modes of evolution by their intrinsic organization. A review of several metamodels suggests that the structure of developmental pathways may bias the genetic basis of phenotypic evolution, and highlights phylogenetic replication as a value-added approach that produces deeper insights into the mechanisms of evolution than single-species analyses.


The Need for Metamodels in Evo–Devo

  1. Top of page
  2. Abstract
  3. The Need for Metamodels in Evo–Devo
  4. Examples from Specific Metamodels
  5. Perspectives: What Makes a Good Metamodel?
  6. ACKNOWLEDGMENTS
  7. LITERATURE CITED
  8. Appendix

The genetic basis of phenotypic evolution has always been one of the central problems in evolutionary biology. A major achievement of molecular genetics and developmental biology was to translate this fundamental question into specific, testable hypotheses: What is the relative importance of changes in gene regulation and protein activity? Do changes occur primarily at the more upstream or downstream levels in developmental hierarchies? How many genes contribute to phenotypic changes, and what is the typical effect of a single locus? And, when a single gene has a large phenotypic effect, is it due to one major mutation, or to the cumulative effect of many subtle changes?

Recent work in comparative developmental biology and quantitative genetics is allowing us not only to answer these questions on a case-by-case basis, but also to assess the “generality” of these answers. Research in a variety of models has shown that changes in gene regulation, rather than protein sequences, play the principal role in the evolution of morphological structures (Davidson 2006; Carroll 2008; Stern and Orgogozo 2008). In evaluating these results, it is necessary to consider potential ascertainment biases. Because the methods of comparative developmental biology (evo–devo) are aimed primarily at the analysis of gene expression, any work based exclusively on evo–devo techniques may lead us to overestimate the frequency of regulatory changes. Much of the population-genetic work supporting the importance of protein changes suffers from an opposite bias, because coding sequences are more often targeted for population surveys. A truly unbiased assessment of the relative contributions of regulatory and coding sequence changes to phenotypic evolution can only be obtained by going beyond candidate genes and embracing genome-wide genetic and genomic approaches.

Genetic analysis shows that phenotypic differences within and between species often have an oligogenic basis, and that some loci have large individual effects (Orr 2001; Doebley 2004; Shapiro et al. 2004; Carbone et al. 2005; Cresko et al. 2007; Steiner et al. 2007). However, many traits deviate from this pattern in either direction. Some show a highly polygenic architecture and are not associated with any large-effect QTLs (Weber et al. 2001; Fishman et al. 2002; Mezey et al. 2005), whereas differences in other traits are controlled by a single locus (Sucena and Stern 2000; Wittkopp et al. 2003a; Wang and Chamberlin 2004). In the few cases where large-effect QTLs were dissected at the molecular level, a general pattern has also failed to emerge. In some instances, phenotypic effects of the gene are due to a single mutation (Hoekstra 2006; Protas et al. 2006), whereas in others the large QTL effect reflects the accumulation of many, presumably subtle, mutations (McGregor et al. 2007; R. Bickel, W. Schackwitz, L. Pennachio, S. Nuzhdin, and A. Kopp, unpubl. ms.).

This diversity of patterns suggests another fundamental question—is the genetic basis of phenotypic evolution constrained by the structure of developmental pathways? For example, are some traits “predisposed” to regulatory changes, and others to changes in protein activity, by the differences in their genetic control? Are monogenic and macromutational changes more likely for some traits than others? Is genetic variation more likely to be fixed in particular components or modules of developmental pathways during species divergence? Do gains and losses of traits follow different genetic trajectories? Finally, genetic basis of phenotypic changes may need to be considered in conjunction with the evolutionary forces responsible for these changes. It is conceivable, for instance, that strong directional selection favors large-effect mutations even if they have pleiotropic deleterious effects, whereas weaker selection favors more subtle mutations that avoid negative trade-offs. If so, different combinations of selective and demographic forces may lead to a different balance between regulatory and coding sequence, or upstream and downstream, changes in developmental pathways.

It is ironic that despite so much work on the genetic basis of phenotypic evolution, we still find it difficult to assess the generality of emerging results. The main reason is that it is not clear how to integrate the lessons from different model traits. As we search for general evolutionary rules, comparing apples and oranges is not only inevitable but essential as well. However, it is hard to compare one apple to one orange and one banana. To overcome this problem, we need to seek out “evolutionary metamodels”—traits that have undergone multiple independent changes in different species, and whose developmental basis is reasonably well understood. The metamodel approach is the comparative equivalent of experimental replication. If we can identify the genetic basis of multiple evolutionary changes in the same trait, we can infer whether this trait has a “predominant” mode of evolution. Only then can we compare different traits in a meaningful fashion.

It is important to note that changes in the same trait are not necessarily caused by changes in the same developmental pathway. For example, convergent evolution of elongated body shape in burrowing salamanders is due to the lengthening of individual vertebrae in one species and to an increased number of vertebrae in another (Parra-Olea and Wake 2001). Variation in wing size in D. melanogaster is a result of changes in cell number in some populations whereas in others differences in cell size make a larger contribution (James et al. 1997; Zwaan et al. 2000). Floral morphology in legumes (Tucker 2003) and sex combs in Drosophila (Tanaka et al. 2009) offers further examples in which similar adult traits are produced by distinct cellular mechanisms. In this review, however, I focus on the evolutionary changes in specific developmental pathways. I describe several metamodels and review the emerging evidence suggesting that the genetic basis of phenotypic evolution may indeed be biased by the structure of developmental pathways. Finally, I outline several requirements that will be essential for a metamodel approach to the evolution of development.

Examples from Specific Metamodels

  1. Top of page
  2. Abstract
  3. The Need for Metamodels in Evo–Devo
  4. Examples from Specific Metamodels
  5. Perspectives: What Makes a Good Metamodel?
  6. ACKNOWLEDGMENTS
  7. LITERATURE CITED
  8. Appendix

DROSOPHILA TRICHOME PATTERNS

Drosophila larvae are covered by a stereotypical, segmentally repeated pattern of single-cell cuticular projections (ventral denticles and dorsal trichomes). This pattern has undergone parallel evolutionary changes in different Drosophila lineages. A large fraction of the dorsal trichome field in every segment has been transformed into naked cuticle once in the melanogaster species group (in D. sechellia) and probably three times in the virilis species group (Dickinson et al. 1993; Sucena et al. 2003). Genetic mapping has shown that the loss of trichomes in D. sechellia is due to changes in a single gene, the transcription factor ovo/shavenbaby (svb) (Sucena and Stern 2000). svb is expressed in the embryonic epidermis and induces trichome formation in a cell-autonomous manner (Payre et al. 1999). In each abdominal segment, svb is expressed in three rows of 1° and 3° cells, which develop stout trichomes, and six to seven rows of 4° cells that form finer trichomes. In D. sechellia, svb expression in the 4° cells has been lost, and the lawn of fine trichomes is replaced by naked cuticle (Sucena and Stern 2000). A similar correlation between narrower svb expression and narrower trichome belts is observed in the virilis group species that show the derived cuticular pattern (Sucena et al. 2003). Genetic evidence, although not conclusive, suggests that cis-regulatory changes at the svb locus may be responsible for the partial loss of svb expression in the virilis group as well as in D. sechellia (Sucena et al. 2003).

Fine-scale genetic mapping and transgenic analysis reveal that the apparently simple monogenic basis of species differences is in fact due to the accumulation of multiple mutations at the svb gene (McGregor et al. 2007). svb expression in the dorsal epidermis is controlled by at least three enhancers spread over 50 kb. In D. sechellia, all three enhancers drive derived expression patterns, and each enhancer has undergone at least one change that contributes to the phenotypic differences between D. sechellia and its closest relatives (McGregor et al. 2007). This finding prompts an obvious question—if species differences are caused by the fixation of multiple mutations, why was their fixation confined to a single locus?

In the genetic pathway that controls trichome development, svb is positioned at the nexus point between upstream regulators and downstream effectors (Fig. 1 and Appendix). Stripes of svb expression are defined by a complex network of signaling pathways and transcription factors that control all aspects of segmental patterning and epithelial differentiation (Sanson 2001). In turn, svb cell-autonomously activates the expression of multiple structural genes that mediate different steps in trichome formation (Chanut-Delalande et al. 2006). A mutation that alters the expression of any upstream regulator of svb in the embryonic epidermis would disrupt not only trichome development, but also the formation of muscle attachments and other cellular processes (Fig. 1). Conversely, a mutation in any one of the downstream targets of svb would alter trichome morphology but not lead to the loss of trichomes. svb may be the only “optimally pleiotropic” locus where mutations have the capacity to abolish trichome development without changing other aspects of the segmental pattern (Delon and Payre 2004; McGregor et al. 2007). Indeed, loss of svb expression in D. sechellia is accompanied by predictable changes in all of its targets (Chanut-Delalande et al. 2006). Interestingly, although ectopic expression of svb in the naked cuticle region is sufficient to induce denticle formation, these ectopic denticles have abnormal shapes (Chanut-Delalande et al. 2006). This suggests that an evolutionary gain of a new trichome or denticle field, were it to occur, would require genetic changes at other loci in addition to svb.

image

Figure 1. A simplified schematic of the regulatory pathway that controls larval denticle and trichome development. Positive and negative regulatory interactions are shown by pointed and blunt arrows, respectively. In this and following figures, a star next to a gene indicates that mutations in this gene are responsible for phenotypic evolution.

Download figure to PowerPoint

All genes involved in epithelial patterning and cell differentiation also play important roles in other, unrelated developmental pathways. For example, in addition to its function in the embryonic epidermis, svb is essential for female germline development (Garfinkel et al. 1994; Mevel-Ninio et al. 1995). However, given the modular organization of tissue-specific enhancers in developmentally controlled genes, pleiotropic gene functions can be easily uncoupled in the course of evolution through fixation of cis-regulatory mutations in different enhancers (Carroll 2008). This suggests that the role of each gene in the evolution of a particular developmental pathway depends on the extent of pleiotropy of this gene within that pathway, whereas its functions in other developmental contexts are likely irrelevant.

ANTHOCYANIN PIGMENTATION IN PLANTS

Flower color often varies dramatically within and among species. This variation has profound adaptive consequences, because color largely determines the spectrum of pollinators attracted to flowers, and changes in color can lead to pollinator shifts (Bradshaw and Schemske 2003; Fenster et al. 2004; Whibley et al. 2006; Hoballah et al. 2007; Rausher 2008). Flower color is controlled in part by flavonoids, a group of plant secondary metabolites (Holton and Cornish 1995). Red, blue, and purple colors are conferred by a particular type of flavonoids called anthocyanins, although other factors such as vacuolar pH, ultrastructure of epidermal cells, and the presence of colorless co-pigments are also important (Mol et al. 1998; Koes et al. 2005). Anthocyanin synthesis is carried out in a stepwise fashion by several enzymes (Fig. 2 and Appendix 2). Importantly, this metabolic pathway includes a number of side branches that produce nonanthocyanin flavonoids that contribute to various physiological functions including pollen fertility, heat stress tolerance, UV resistance, pathogen and herbivore defense, etc. (Koes et al. 1994; Winkel-Shirley 2002; Strauss and Whittall 2006). Multiple steps in anthocyanin synthesis, ranging from several downstream reactions to the entire core pathway, are co-regulated as a single unit by a transcriptional complex including bHLH and MYB-domain transcription factors and a WD40-repeat scaffolding protein (Quattrocchio et al. 1998; Koes et al. 2005; Morita et al. 2006). Both the biosynthetic enzymes and the regulatory genes that control their expression are widely conserved among flowering plants, although many pathway components are encoded by paralogous gene families that expanded independently in different taxa (Holton and Cornish 1995; Winkel-Shirley 2001; Koes et al. 2005).

image

Figure 2. A generalized scheme of the anthocyanin synthesis pathway. Genes are shown in black and metabolites in gray. Transcriptional interactions (direct or indirect) are indicated by black arrows, and chemical reactions by gray arrows. The three upstream enzymes are co-regulated by the MYB/bHLH/WD-40 complex only in some plant lineages (dashed arrows).

Download figure to PowerPoint

Genetic basis of evolutionary changes in flower color has been studied in several plant genera including Petunia, Antirrhinum (snapdragon), Ipomoea (morning glory), Aquilegia (columbine), and Mimulus (monkeyflower) (Quattrocchio et al. 1998; Bradshaw and Schemske 2003; Clegg and Durbin 2003; Schwinn et al. 2006; Whittall et al. 2006; Hoballah et al. 2007; Coberly and Rausher 2008; Streisfeld and Rausher 2009; D. L. Des Marais and M. D. Rausher, unpubl. ms). One of the most common changes is the loss of anthocyanin pigmentation, leading to the evolution of white or yellow flowers (Rausher 2008). For example, Petunia integrifolia has reddish-violet flowers and is pollinated primarily by solitary bees, whereas P. axillaris has white flowers and is pollinated by nocturnal hawkmoths. The loss of anthocyanins in P. axillaris is associated with mutations in the coding sequence of the an2 gene, which encodes a MYB-domain transcription factor (Hoballah et al. 2007). an2 is part of the protein complex that controls the expression of several enzymes in the downstream portion of the anthocyanin synthesis pathway (Fig. 2) (Quattrocchio et al. 1999). Remarkably, different isolates of P. axillaris carry at least six different nonsense or frameshift mutations in an2, indicating that loss-of-function alleles of this gene arose and were fixed repeatedly (Hoballah et al. 2007). In the absence of genome-wide QTL analysis, it is uncertain whether an2 was the only locus responsible for anthocyanin loss during species divergence. However, neither the enzyme genes regulated by an2 nor the other components of the regulatory complex (an1 and an11, which encode the bHLH and WD-40 proteins, respectively [de Vetten et al. 1997; Spelt et al. 2000]]) play major roles in the phenotypic differences between Petunia species. Similarly, in Antirrhinum, complete or partial loss of anthocyanin pigmentation in some species and subspecies is due to changes in the MYB-domain genes rosea and venosa, whereas delila and mut, the bHLH genes, do not make major contributions (Schwinn et al. 2006; Whibley et al. 2006).

Why is repeated evolutionary loss of anthocyanin pigmentation associated with changes in the MYB genes rather than any other components of the regulatory complex or biosynthetic enzymes? The anthocyanin pathway is active in many vegetative tissues in addition to flowers, and the different flavonoids produced by this pathway perform many adaptive functions in addition to pollinator recruitment (Koes et al. 1994; Winkel-Shirley 2002). The expression of anthocyanin synthesis enzymes is controlled by tissue-specific regulatory complexes that share common bHLH and WD-40 components but include different MYB-domain proteins in different tissues (Koes et al. 2005). In Petunia, for example, an2 controls anthocyanin synthesis in the flower corolla, whereas its paralog an4 performs the same function in anthers (Quattrocchio et al. 1993; Spelt et al. 2000). Loss-of-function mutations in the bHLH or WD-40 genes (an1 and an11) result in completely white flowers, whereas an2 and an4 mutations have spatially restricted effects. an1 and an11 also have pleiotropic effects on seed coat development and other cellular processes (de Vetten et al. 1997; Spelt et al. 2002; Koes et al. 2005). At the same time, mutations in the MYB genes are sufficient to eliminate or severely reduce the expression of all enzymes regulated by the MYB/bHLH/WD-40 complexes in the flower corolla in both Petunia and Antirrhinum (Quattrocchio et al. 1999; Schwinn et al. 2006). Thus, tissue-specific MYB genes may be the “optimally pleiotropic” components of the anthocyanin pathway: mutations in these genes have the greatest potential to change pollinator recruitment while having the fewest deleterious side effects. For example, different white-flower alleles segregating in natural populations of Ipomoea purpurea are due to mutations in either an MYB transcription factor (the W locus), or in chalcone synthase (the A locus), the most upstream enzyme in the anthocyanin pathway. Consistent with the differential pleiotropy hypothesis, the white aa plants show reduced survival in the field, whereas the white ww individuals have apparently normal fitness (Chang et al. 2005; Coberly and Rausher 2008).

The optimally pleiotropic genes may be different in different plant taxa due to the changing organization and tissue-specific expression of paralogous gene families (Quattrocchio et al. 1993; Durbin et al. 2003; Des Marais and Rausher 2008), and to changes in the transcriptional co-regulation of the upstream and downstream parts of the anthocyanin pathway (Quattrocchio et al. 1998; Koes et al. 2005; Morita et al. 2006). However, changes in transcription factors that regulate anthocyanin synthesis appear to be a common evolutionary mechanism. In Mimulus aurantiacus, loss of anthocyanin pigmentation is associated with the loss of F3H, DFR, and ANS expression specifically in flowers, but not in vegetative tissues, and genetic analysis suggests that this change is controlled by a single trans-acting locus (Streisfeld and Rausher 2009). In Aquilegia, repeated losses of anthocyanin pigmentation are correlated with reduced expression of multiple enzymes that are known to be controlled by the MYB/bHLH/WD-40 complexes (F3H, DFR, ANS, and 3GT) (Whittall et al. 2006). Again, the genetic basis of color differences is monogenic in at least two different pairs of Aquilegia species (Prazmo 1965; Hodges et al. 2002), suggesting that reduced expression of the biosynthetic enzymes is due to mutations in a single trans-acting factor. Similarly, in Ipomoea, changes in flower color are also correlated with tissue-specific downregulation of multiple anthocyanin synthesis enzymes (Durbin et al. 2003). Interestingly, in Aquilegia, Petunia, and Antirrhinum, expression of enzymes that are regulated by the MYB/bHLH/WD-40 complex and act late in the anthocyanin pathway is more likely to be lost than the expression of early-acting enzymes that are not controlled by these genes (Quattrocchio et al. 1999; Durbin et al. 2003; Schwinn et al. 2006; Whittall et al. 2006). Interruption of this biosynthetic pathway at an upstream step could have negative pleiotropic effects because, in addition to flower color, it would affect the synthesis of nonanthocyanin flavonoids that are important for UV resistance, pathogen defense, and other physiological adaptations (Fig. 2) (Winkel-Shirley 2002; Strauss and Whittall 2006). Thus, the branched organization of the anthocyanin pathway makes later enzymatic steps more evolutionarily labile than earlier ones (Lu and Rausher 2003; Whittall et al. 2006). In Ipomoea, however, the entire pathway starting with CHS is co-regulated by the MYB/bHLH/WD-40 complex, so that even tissue-specific MYB mutations are perforce more widely pleiotropic than in other plants (Chang et al. 2005; Morita et al. 2006). It is intriguing that genetic changes in the downstream enzymes such as DFR, ANS, or 3GT do not appear to contribute to the evolutionary losses of anthocyanin pigmentation in Petunia, Ipomoea, Mimulus, or Antirrhinum. In principle, regulatory mutations in any of these genes could produce white or yellow flowers with relatively few pleiotropic consequences. One possible explanation is that these genes may lack tissue-specific cis-regulatory elements so that their expression cannot evolve independently in different tissues.

The loss of anthocyanins is not the only common mode of flower color evolution. Changes in the relative amounts of red, blue, and yellow pigments are associated with pollinator shifts in many plant genera (Fenster et al. 2004). In Ipomoea, the difference between blue, insect-pollinated and red, hummingbird-pollinated species is due to changes in the flavonoid 3′-hydroxylase (F3′H) gene (Zufall and Rausher 2004; D. L. Des Marais and M. D. Rausher, unpubl. ms). Expression of this enzyme, which converts a precursor of the red-colored pelargonidins into a precursor of the blue-colored cyanidins, is strongly reduced in the red-flowered compared to blue-flowered species. This change is observed in three independent lineages within Ipomoea that evolved red flowers from the blue ancestral state (M. A. Streisfeld and M. D. Rausher, unpubl. ms). In all cases, reduced expression of F3′H is observed in the floral, but not in vegetative tissues, and transgenic assays suggest that these changes are due to cis-regulatory mutations. Thus, evolutionary changes again appear to occur in the optimally pleiotropic component of the developmental pathway, but this component is different for different phenotypes. Continuing work in Ipomoea, Aquilegia, Mimulus, and other genera will tell which features of anthocyanin pathway evolution are truly universal, and what these features reveal about the role of pathway topology in shaping the fixation of natural genetic variation.

DROSOPHILA COLOR PATTERNS

Drosophila color patterns vary extensively both among and within species. Differences in pigmentation can be either global, where one species or morph is uniformly darker than the other, or local, where the taxa differ in the spatial distribution of darkly and lightly pigmented areas. Both types of differences have evolved repeatedly in many different lineages (Wittkopp et al. 2003a), making Drosophila color patterns an excellent metamodel for analyzing the genetic basis of phenotypic evolution.

A key conclusion emerging from recent work is that genetic changes at different loci are responsible for color pattern differences in different Drosophila species (Table 1). Several genes involved in pigment synthesis (including yellow (y), tan (t), and ebony (e)) or in the spatial patterning of pigmentation (optomotor-blind (omb) and bric a brac (bab)) show an association with intra or interspecific pigmentation differences. For example, dark abdominal pigmentation is present in most species of the D. melanogaster species subgroup but has been lost in D. santomea (Llopart et al. 2002; Carbone et al. 2005). Genetic mapping and transgenic analysis have identified cis-regulatory changes at the t locus as one of the key causes of the difference in pigmentation between D. santomea and its closest relative, D. yakuba (Jeong et al. 2008). A similar difference in color pattern is found between D. m. malerkotliana and D. m. pallens in the ananassae subgroup. In this case, however, t makes no significant contribution to the phenotypic change (Ng et al. 2008). Similarly, omb and bab are associated with intraspecific variation in abdominal color patterns in D. polymorpha and D. melanogaster, respectively (Kopp et al. 2003; Brisson et al. 2004), but not in D. malerkotliana. More global differences in the intensity of pigmentation appear to be controlled by changes in ebony and tan in D. americana and D. novamexicana (Wittkopp et al. 2003b; P. Wittkopp, E. Stewart, L. Arnold, A. Neidert, B. Haerum, E. Thompson, S. Arkhas, G. Smith-Winberry, and L. Shefner, unpubl. ms), by ebony and yellow in D. elegans and D. gunungcola (S.-D. Yeh and J. True, pers. comm.), and by ebony in some populations of D. melanogaster (Pool and Aquadro 2007; Takahashi et al. 2007). None of these loci make a detectable contribution to the difference in pigmentation between D. m. malerkotliana and D. m. pallens (Ng et al. 2008). In general, the genetic architecture of color pattern differences ranges from a single Mendelian factor in D. kikkawai and D. jambulina (Ohnishi and Watanabe 1985) to polygenic systems involving complex gene interactions in D. arawakana and D. nigrodunni (Hollocher et al. 2000a,b). Most studied species fall somewhere between these extremes, and a moderately oligogenic basis of variation appears to be typical for this trait (Martinez and Cordeiro 1970; Spicer 1991; Wittkopp et al. 2003b; Carbone et al. 2005; Ng et al. 2008). In all cases that have been dissected at the molecular level, phenotypic changes are associated with regulatory mutations. This is probably not surprising, because many of the pigmentation enzymes and their products play additional roles in neurotransmission, so that coding sequence mutations are likely to have pleiotropic effects on nervous system function and behavior (True 2003).

Table 1.  Genes associated with pigmentation differences in different Drosophila species.
 babombebonyyellowtanMethod1References
  1. 1Methods: 1 =Genetic analysis in experimental crosses; 2 =Genetic association in natural populations.

  2. 2Not tested.

D. melanogaster (USA)YESNONONONO1(Kopp et al. 2003)
D. melanogaster (Africa)nt2ntYESntnt2(Pool and Aquadro 2007)
D. m. malerkotliana/D. m. pallensNONONONONO1(Ng et al. 2008)
D. ananassaeNOntYESYESNO1A. Kopp, unpubl. data
D. kikkawaiNONONONONO1A. Kopp, unpubl. data
D. serrataNONONONONO1A. Kopp, unpubl. data
D. polymorphantYESntntnt2(Brisson et al. 2004)
D. santomea/D. yakubaNOntntNOYES1(Carbone et al. 2005; Jeong et al. 2008)
D. elegans/D. gunungcolantntYESYESNO1S.-D. Yeh and J. True, pers. comm.
D. americana/D. novamexicanaNONOYESNOYES1(Wittkopp et al. 2003b; P. Wittkopp, E. Stewart, L. Arnold, A. Neidert, B. Haerum, E. Thompson, S. Arkhas, G. Smith-Winberry, and L. Shefner, unpubl. ms)

Different genetic basis of similar color patterns in different species may be explained by the branched organization of the Drosophila pigmentation pathway (Fig. 3 and Appendix 3). In this pathway, different metabolic reactions draw on a shared pool of soluble precursors to produce several distinct light and dark pigments (Wright 1987; True 2003; Wittkopp et al. 2003a). At least one of these reactions is reversible, with the opposing reactions catalyzed by the products of the ebony and tan loci (True et al. 2005). This nonlinear pathway structure means that changes in the output of different enzymatic reactions can produce similar phenotypes. For example, darker pigmentation can in principle be due either to increased expression or activity of enzymes required for the synthesis of dark pigments (e.g., Ddc, yellow, or tan), or to decreased expression or activity of enzymes involved in the synthesis of light pigments (such as ebony, black, or Dat). Increased or decreased expression of these enzymes can in turn be caused either by mutations in the regulatory regions of these loci, or to changes in the expression of their upstream regulators such as bab, omb, and Abd-B. Thus, the genetic target for producing any given phenotypic change is quite extensive, so that similar phenotypic adaptations can take distinct genetic paths in different evolutionary lineages.

image

Figure 3. A simplified schematic of the regulatory pathway that controls Drosophila pigmentation. Genes are shown in black and metabolites in gray. Transcriptional interactions (direct or indirect) are indicated by black arrows, and chemical reactions by gray arrows. Positive and negative regulatory interactions are shown by pointed and blunt arrows, respectively. Hypothesized regulatory interactions are indicated by dashed lines.

Download figure to PowerPoint

COAT AND PLUMAGE COLOR IN VERTEBRATES

Pigmentation is as variable in vertebrates as it is in insects, with some intraspecific color variation found in most species. In particular, melanic phenotypes segregate in many mammals, birds, and reptiles (Majerus 1998). Decades of research in mice (Hoekstra 2006), chickens (Smyth 1990), and other domestic animals have identified over a hundred genes affecting coat and plumage color, and the pathway responsible for vertebrate pigmentation is understood in some detail (Fig. 4 and Appendix 4). Unlike insects, vertebrate melanin pigments are produced not by all epithelial cells but by migratory cells (melanocytes) derived from the neural crest, so that genetic changes that affect melanocyte migration lead to altered color patterns (Jackson 1994; Parichy 2006). Subsequent steps in the development of pigmentation are controlled by an endocrine signaling mechanism involving peptide hormones generated as cleavage products of the POMC protein (Fig. 4) (Pritchard and White 2007). Melanocortin hormones are widely pleiotropic, affecting behavior, metabolism, and immunity in addition to pigmentation (Ducrest et al. 2008). Melanin synthesis is regulated by the binding of melanocortins and their paracrine antagonist Agouti to the MC1R receptor expressed in melanocytes (Garcia-Borron et al. 2005). A signal transduction cascade initiated by MC1R activates the expression of MITF, a transcription factor that regulates multiple aspects of melanocyte differentiation (Lin and Fisher 2007). MITF regulates multiple genes involved in pigment synthesis, including the Tyrosinase enzyme that catalyzes the rate-limiting step in this process (Slominski et al. 2004).

image

Figure 4. A simplified schematic of the regulatory pathway that controls vertebrate pigmentation. Genes are shown in black and metabolites in gray. Transcriptional interactions (direct or indirect) are indicated by black arrows, and chemical reactions by gray arrows. Positive and negative regulatory interactions are shown by pointed and blunt arrows, respectively.

Download figure to PowerPoint

Genetic basis of color variation has been studied in dozens of mammal, bird, and reptile species. A key result to emerge from these studies is the frequent involvement of MC1R. Changes in the coding sequence of MC1R, leading to altered receptor activity, are associated with melanism in pocket mice, snow geese, jaguars, lizards, and other vertebrates (Eizirik et al. 2003; Nachman et al. 2003; Mundy et al. 2004; Rosenblum et al. 2004). There are at least 16 independent cases of genetic association between MC1R and melanism in natural populations, and more examples exist in domesticated species (Hoekstra 2006; Ducrest et al. 2008). Although MC1R mutations have mainly been characterized in relation to melanism, this gene is also involved in the evolution of light coat color in beach mice (Steiner et al. 2007). In this case, MC1R is only one of two major QTLs responsible for phenotypic variation; an even greater contribution is made by the Agouti locus. In contrast to MC1R, genetic changes at Agouti are regulatory rather than coding (Steiner et al. 2007). Interestingly, skin color variation in humans is associated with both MC1R and Agouti loci (Valverde et al. 1995; Bonilla et al. 2005), and Agouti may also be responsible for coat color differences in deer mice (Hoekstra 2006).

One factor that probably contributes to the prominent roles of MC1R and Agouti in color variation is the structure of the vertebrate pigmentation pathway (Fig. 4). Although multiple melanocortin receptors are expressed in most tissues and regulate many physiological processes, MC1R is the main if not the only receptor expressed in melanocytes, and its functions in other cell types appear to be relatively minor (Pritchard and White 2007; Ducrest et al. 2008). Agouti is more widely pleiotropic, but its expression is controlled by multiple cis-regulatory elements (Dinulescu and Cone 2000) and evolutionary changes are associated with regulatory rather than coding Agouti mutations (Steiner et al. 2007). In contrast, the upstream parts of the melanocortin pathway are based on endocrine signaling, and any changes in the expression or processing of POMC are bound to have pleiotropic effects on physiology and behavior (Pritchard and White 2007; Ducrest et al. 2008). Signal transduction downstream of MC1R is mediated by a ubiquitous secondary messenger system, and its main downstream target MITF regulates many aspects of melanocyte function including survival (Steingrimsson et al. 2004; Levy et al. 2006). Thus, MC1R and Agouti may well be the “optimally pleiotropic” components of this pathway: genetic changes at these loci alone are sufficient to produce major changes in pigmentation while having the fewest pleiotropic effects (Mundy 2005; Hoekstra 2006).

However, a serious ascertainment bias undoubtedly contributes to the central position of MC1R coding changes in pigmentation literature. MC1R is a small, conserved, single-exon gene, making its coding sequence by far the easiest target for genetic association studies (Mundy 2005; Hoekstra 2006). Other loci, or, indeed, regulatory changes at MC1R are more difficult to study. Thus, the vast majority of papers on the genetic basis of color variation consist of two types: those that report an association between MC1R and pigmentation, and those that report a lack of such association. The frequency of these outcomes is roughly equal; there are at least 24 cases in which color variation is not linked to MC1R (Ducrest et al. 2008). In fact, the two alternative scenarios may be found in the same species: MC1R is associated with melanism in one natural population of rock pocket mice, but not in three others (Hoekstra and Nachman 2003; Nachman et al. 2003); a similar situation is found in different populations of beach mice (Steiner et al. 2007, 2009). Despite clear evidence that MC1R is only part of the story, no concerted effort has been made to identify the other genes, leaving us with a highly biased picture of the genetic basis of evolutionary change (but see [Steiner et al. 2007] for a welcome exception).

What could these other genes be? Pigment synthesis enzymes could well play a role in the evolution of color patterns in vertebrates, as they do in flies. Although these enzymes are essential in many tissues, their expression in different cell types is controlled by independent cis-regulatory elements. In particular, Tyrosinase and Tyrp1 expression in melanocytes is conferred by melanocyte-specific enhancers (Murisier and Beermann 2006). Null Tyrosinase alleles cause oculocutaneous albinism that would be strongly deleterious in nature, but hypomorph mutations lead to subtle changes in coat color (Beermann et al. 2004). Thus, regulatory changes that cause increased or decreased expression of Tyrosinase or other enzymes specifically in melanocytes could become fixed in natural populations. In humans, for example, noncoding variation at the Aim-1 enzyme locus is associated with natural skin color variation (Graf et al. 2007).

More upstream pathway components may also play a role in evolutionary change. Color patterns play an important adaptive role as a defense against visual predators (Caro 2005; Hoekstra 2006). However, due to the pleiotropy of the melanocortin system, pigmentation is often associated with a wide range of physiological and behavioral phenotypes, especially if it involves changes in the upstream (endocrine) part of the pathway (Ducrest et al. 2008). In lions, dark mane color correlates with high testosterone levels, fighting ability, and sexual activity (West and Packer 2002). In alpine swifts, plumage color correlates with stress resistance and body size (Roulin et al. 2008), whereas in great tits it correlates with aggression, body size, and resting metabolic rate (Roskaft et al. 1986; Kölliker et al. 1999). Melanic pigmentation itself may also be antagonistically pleiotropic due, for example, to increasing heat stress (West and Packer 2002; Caro 2005). Thus, evolutionary changes in color patterns may tend to have a different genetic basis depending not only on the selective regime, but also on what trait is under selection—pigmentation itself, or the behavioral syndromes associated with pigmentation. In cases in which physiological or behavioral changes are favorable or at least neutral, the evolution of pigmentation may be due to mutations in the upstream, highly pleiotropic components of the melanocortin pathway. In contrast, selection on color patterns is more likely to fix mutations in the “nexus” components such as MC1R and Agouti, especially when physiological changes that could result from increased melanocortin production are deleterious.

Perspectives: What Makes a Good Metamodel?

  1. Top of page
  2. Abstract
  3. The Need for Metamodels in Evo–Devo
  4. Examples from Specific Metamodels
  5. Perspectives: What Makes a Good Metamodel?
  6. ACKNOWLEDGMENTS
  7. LITERATURE CITED
  8. Appendix

A tentative conclusion emerging from this analysis is that the structure of developmental pathways may bias the genetic basis of evolutionary change. In pathways that have “optimally pleiotropic” genes—that is, genes that have the most widespread effects on the trait under selection and the fewest effects on other traits—genetic changes tend to accumulate at these loci. In contrast, pathways with a more distributed architecture may tend toward a more polygenic and less predictable mode of evolution.

It is important to clarify what “pleiotropy” means in this context. The expression of many genes that regulate animal and plant development is controlled by modular cis-regulatory elements. Thus, for a gene that affects different phenotypes in different tissues or at different times in development, each of its functions can evolve independently of all others (Wray 2007; Carroll 2008). What matters in the evolution of developmental pathways is not the pleiotropy of the gene at the level of the entire organism, but rather its pleiotropy in the particular cell type and at the stage of development where the phenotype under selection is produced. For example, the fact that svb is essential for oogenesis and female fertility does not prevent it from becoming the central player in the evolution of larval cuticle, whereas the involvement of MITF is multiple aspects of melanocyte differentiation and function may well prevent the fixation of evolutionary changes at this locus that would lead to altered pigmentation.

The main purpose of this review, however, is not to draw any specific conclusions about the evolution of developmental pathways, but simply to emphasize the advantages of examining the genetic basis of similar phenotypic changes in multiple evolutionary lineages. By analyzing the evolution of development in a systematic and replicated fashion, we may understand whether it follows any general rules, and what these rules might be.

So, what makes a good metamodel? First or all, detailed knowledge of the developmental pathway is essential for an unbiased study of evolution. In the absence of such information, repeated tests focused on the same small set of candidate genes will inevitably lead to a biased and misleading picture of evolutionary change. However, the knowledge of development does little good if it is not used. A glaring example is vertebrate pigmentation, where most studies have continued to focus on a single gene despite clear evidence that other loci may be equally important. Unfortunately, few pathways are understood so well. It is no accident that three of the four metamodels discussed in this review involve pigmentation—the simplest morphological trait one can imagine. As our knowledge of more complex developmental processes continues to grow, new metamodels will become amenable to increasingly systematic analysis. A sample of such potential metamodels is presented in Table 2. For all these traits, it will be necessary not only to identify the genes responsible for evolutionary changes, but also to characterize the positions of these genes within the developmental pathways that control each trait; without this information, the role of development in channeling the fixation of natural variation cannot be understood. The taxonomic scale of metamodels can be quite broad as long as the regulatory interactions that comprise developmental pathways are substantially conserved—as, for example, in the anthocyanin and vertebrate melanin synthesis pathways. Naturally, all pathways tend to diverge over time due to the gain and loss of regulatory connections and to the turnover of genes in paralogous gene families. Thus, the scale at which the metamodel approach remains informative will need to be decided on a trait-by-trait basis.

Table 2.  A sample of emerging and potential metamodels.
TaxonTaxonomic scaleTraitsEvolutionary patternCrossesToolsReferences
Cave fish (Astyanax)Within species; independent cave populations and surface relativesLoss of pigmentation and eyes; adaptive changes in trophic morphology and behaviorMultiple gains and losses of traitsYesZebrafish genome; linkage maps; ectopic expression and RNAi by injectionProtas et al. 2006, 2007; Jeffery 2008; Protas et al. 2008; Gross et al. 2009; Jeffery 2009
Stickleback fish (Gasterosteidae)Different populations within species; different species and generaLoss of spines and lateral armor plates; changes in body and jaw shape, pigmentation, dentition, and reproductive behaviorMultiple gains and losses of traitsYesGenome of threespine stickleback; linkage maps; transgenic tools; genotyping arrays(Cresko et al. 2004; Boughman et al. 2005; Colosimo et al. 2005; Kimmel et al. 2005; Shapiro et al. 2006; Albert et al. 2008)
Heliconius butterfliesDifferent local populations within species; different speciesWing color patterns (Mullerian mimicry)Divergence between populations within species; convergence between speciesYesH. melpomene genome (soon); linkage maps; genomic libraries(Joron et al. 2006a,b; Baxter et al. 2008; Papa et al. 2008; Baxter et al. 2009)
LepidopteraDifferent familiesWing eyespotsLikely independent gain in different butterfly and moth familiesNoLinkage maps, genome (soon), genomic libraries, mutant collection, and transgenic tools in Bicyclus anynana(Nijhout 1991; Brunetti et al. 2001; Monteiro et al. 2003; Reed and Serfas 2004; Monteiro et al. 2006; Beldade et al. 2009)
Scarabeid beetles (esp. Onthophagus)Different species and genera; polyphenism and quantitative variation within some speciesHead and thoracic hornsPossible multiple gains in Scarabeidae; independent origin in other beetle familiesNoGenomic libraries and EST sequences; microarrays; RNAi(Emlen et al. 2005; Moczek et al. 2006; Emlen et al. 2007; Moczek 2009; Moczek and Rose 2009)
Sepsidae (Diptera)Different species and generaAdult abdominal appendagesPossible multiple gainsNoNone so far(Eberhard 2001; Bowsher and Nijhout 2007; Su et al. 2008)
Rhabditid nematodesDifferent species and generaHermaphroditismIndependent originsNoGenetic resources in C. elegans; multiple genome sequences; systemic RNAi by feeding; transgenic tools(Kiontke et al. 2004; Haag 2005; Kiontke and Fitch 2005; Hill et al. 2006; Kelleher et al. 2008)
Rhabditid nematodesWithin species; between species and generaVulval development: cell lineage, competence, and fates; division patterns; inductive signalsMultiple changesNoSee above(Delattre and Felix 2001; Sommer 2005; Kiontke et al. 2007)
Poeciliid fishDifferent speciesPlacentaLikely multiple gainsRareNone so far(Reznick et al. 2002; O’Neill et al. 2007; Turcotte et al. 2008)
DrosophilaBetween speciesMale sex combsMultiple gains and lossesYesGenetic resources in D. melanogaster; multiple genome sequences; transgenic tools(Barmina and Kopp 2007; Randsholt and Santamaria 2008; Atallah et al. 2009; Tanaka et al. 2009)
Eudicot plantsWithin and between species in different familiesLeaf shapeParallel variation in many species/genera; convergent changes in different genera and familiesYesGenome sequences, genetic resources and transgenic tools in multiple model taxa(Bharathan et al. 2002; Champagne et al. 2007; Blein et al. 2008; Kimura et al. 2008)
Arabidopsis and other BrassicaceaeWithin species, in multiple speciesSelfingMultiple gainsYesGenetic resources in A. thaliana; multiple genome sequences; transgenic tools(Kusaba et al. 2001; Nasrallah et al. 2004; Charlesworth and Vekemans 2005; Tang et al. 2007; Shimizu et al. 2008)

The second rule of metamodel selection is genetic accessibility. Although traditional evo–devo approaches based on comparative analysis of gene expression provide essential insights into the developmental basis of evolutionary changes, such approaches alone cannot provide a complete picture of the genetic basis of evolution. In combination with genetic analysis (Jeong et al. 2008) or transgenic assays (Wang and Chamberlin 2004), developmental-genetic techniques can identify some of the key loci responsible for the evolution of development. However, unbiased genetic approaches such as QTL mapping are invaluable for quantifying the contribution of these loci to phenotypic changes and for identifying other contributing genes (Carbone et al. 2005; Steiner et al. 2007). The best metamodels are those that permit the integration of developmental-genetic and quantitative-genetic approaches (Table 2). This requirement favors rapidly evolving traits—another reason why pigmentation features so prominently in this review. A further advantage of such traits is that once the genes responsible for phenotypic changes are identified, population-genetic analysis of these genes may reveal the selective and demographic forces that shape the evolution of developmental pathways within and between species.

The third rule is phylogenetic independence. This is especially important among recently diverged taxa, where apparently independent phenotypic changes may in fact evolve from standing genetic variation in ancestral populations (Barrett and Schluter 2008). For instance, parallel loss of armor in limnetic sticklebacks is likely caused by repeated independent fixation of related Ectodysplasin alleles that segregate at low frequency in marine populations (Colosimo et al. 2005). Although the independence of evolutionary changes can sometimes be established after the fact, for example if different functional changes are found in the same gene in different populations (Protas et al. 2006), a robust phylogenetic framework is essential for identifying parallel phenotypic transitions.

The nature of evolving traits and the mode of their evolution are also important considerations for providing a balanced picture of developmental changes. The evolution of three-dimensional organs might be expected to have a more complex genetic basis than changes in simple traits such as two-dimensional color patterns. It may, for example, be more likely to involve more regulatory and fewer coding sequence changes (Carroll 2008), or have a more polygenic basis (Mezey et al. 2000; Wagner et al. 2008). At the same time, although most of the best-studied models to date involve rapid loss of characters, the evolution of novel traits is arguably a more interesting process and may require a different type of changes in developmental pathways. The evolution of sex combs in Drosophila (Barmina and Kopp 2007; Tanaka et al. 2009) or the mechanosensory system and feeding apparatus in cave fish (Jeffery 2008; Yamamoto et al. 2009) are promising examples; other metamodels showing gains of complex characters are listed in Table 2. One hopes that such “constructive” changes will play a more prominent role in future evo–devo research.

Finally, the metamodel approach has implications for science funding. The fact that a particular trait has been investigated in one model organism should not necessarily preclude the funding of similar work in other organisms—provided, of course, that this research is unbiased and systematic rather than a “me-too” study of everyone's favorite genes. In many cases, parallel research in multiple model organisms may produce a combined outcome that is greater than the sum of its parts.

Associate Editor: M. Rausher

ACKNOWLEDGMENTS

  1. Top of page
  2. Abstract
  3. The Need for Metamodels in Evo–Devo
  4. Examples from Specific Metamodels
  5. Perspectives: What Makes a Good Metamodel?
  6. ACKNOWLEDGMENTS
  7. LITERATURE CITED
  8. Appendix

I am grateful to J. Bowsher, T. Caro, A. Chandler, V. Irish, A. McGregor, R. Meier, A. Moczek, C-S Ng, M. O’Neill, M. Rausher, D. Reznick, J. True, J. Whittall, and T. Wittkopp for comments on the manuscript, information about their work, and sharing unpublished data. Work in my laboratory is supported by NSF grants DEB-0823673 and IOS-0815141.

LITERATURE CITED

  1. Top of page
  2. Abstract
  3. The Need for Metamodels in Evo–Devo
  4. Examples from Specific Metamodels
  5. Perspectives: What Makes a Good Metamodel?
  6. ACKNOWLEDGMENTS
  7. LITERATURE CITED
  8. Appendix

Appendix

  1. Top of page
  2. Abstract
  3. The Need for Metamodels in Evo–Devo
  4. Examples from Specific Metamodels
  5. Perspectives: What Makes a Good Metamodel?
  6. ACKNOWLEDGMENTS
  7. LITERATURE CITED
  8. Appendix

Appendix. Developmental Pathways Discussed in this Review

DEVELOPMENT OF DROSOPHILA LARVAL TRICHOMES

Segmental patterns in the larval cuticle are initially established by hedgehog (Hh) and wingless (Wg) signaling and further elaborated by epidermal growth factor (EGFR) and notch signaling (Sanson 2001) (Fig. 1). Inputs from these signaling pathways are integrated by the HMG-box transcription factors SoxNeuro (SoxN) and Dichaete (D) and the zinc finger transcription factor ovo/shavenbaby (svb) (Payre et al. 1999; Overton et al. 2007). A negative feedback loop between Wg signaling and SoxN/D, and a positive feedback loop between SoxN/D and svb, translate a quantitative balance between Wg and EGFR activities into sharp boundaries between svb-ON and svb-OFF expression domains (Overton et al. 2007). svb acts as the sole mediator of the Wg and EGFR pathways in establishing trichome pattern; every cell that expresses svb produces a trichome, and every cell that does not express svb produces naked cuticle (Payre et al. 1999; Chanut-Delalande et al. 2006). svb controls multiple cellular processes involved in the development of cuticular projections (Chanut-Delalande et al. 2006). First, it regulates cytoskeletal dynamics in the apical extensions of epithelial cells by inducing the expression of actin-associated factors forked (f), singed (sn), and shavenoid (sha), and the activator of the Arp2/3 actin nucleation complex wasp (wsp). Second, it activates the expression of miniature (m), which encodes a transmembrane protein required for membrane-cuticle attachment. Finally, svb induces localized expression of yellow, a secreted protein involved in trichome pigmentation. Thus, a battery of target genes activated by svb translates a spatial prepattern into a distinct mode of cell differentiation (Chanut-Delalande et al. 2006). In addition to controlling svb expression, the Hh, Wg, and EGFR signaling pathways induce spatially restricted expression of the stripe transcription factor (Hatini and DiNardo 2001), which in turn regulates a set of structural target genes responsible for the developmental of epithelial-muscle attachments (tendons) (Vorbruggen and Jackle 1997). Note that all regulatory and structural genes in this pathway are pleiotropic and perform multiple functions in other tissues and at other times in development.

THE ANTHOCYANIN PATHWAY

The core pathway consists of six enzymes that convert colorless precursors into colored anthocyanin derivatives: chalcone synthase (CHS), chalcone isomerase (CHI), flavanone-3-hydrolase (F3H), dihydroflavonol reductase (DFR), anthocyanidin synthase (ANS), and flavonoid 3-glucosyltransferase (3GT) (Fig. 2). In some plants, this pathway branches after the 3-OH-flavonol stage to produce two or three different pigment types: red pelargonidin, blue cyanidin, and blue-purple delphinidin, but the three parallel subpathways share most of the enzymes. A number of side branches in this pathway produce flavonoids that perform a wide variety of physiological functions including UV and heat stress protection, herbivore and pathogen defense, pollen development, etc. (Koes et al. 1994; Mol et al. 1998; Winkel-Shirley 2002; Strauss and Whittall 2006). In all studied plants, many enzymes in the anthocyanin pathway are encoded by multigene families where different paralogs have distinct but sometimes overlapping tissue-specific expression (Quattrocchio et al. 1993; Durbin et al. 2003). For simplicity, this overview conflates information from different taxa by ignoring most instances of paralogy.

Expression of the anthocyanin synthesis enzymes is controlled by protein complexes consisting of bHLH and MYB-domain transcription factors and a WD40-repeat scaffolding protein (de Vetten et al. 1997; Quattrocchio et al. 1999; Spelt et al. 2000; Koes et al. 2005; Morita et al. 2006; Schwinn et al. 2006). In at least some plants, these regulatory complexes are tissue specific. Different complexes share the same bHLH proteins (encoded by an1 and jaf13 in Petunia, B and R in maize, and delila in snapdragon) and the same WD40-repeat proteins (an11 in Petunia, PAC in maize, and an unidentified gene in snapdragon), but differ by the inclusion of different tissue-specific MYB-domain transcription factors (Spelt et al. 2002; Koes et al. 2005). Transgenic experiments show that the bHLH, MYB-domain, and WD40-repeat proteins that regulate anthocyanin synthesis are functionally conserved between monocots (maize) and eudicots (Petunia), suggesting that both the metabolic pathway and its regulatory architecture are similar in all flowering plants (Goodrich et al. 1992; Quattrocchio et al. 1993; Koes et al. 1994; Holton and Cornish 1995; Mol et al. 1998).

A key feature of this regulatory architecture is that several downstream steps in anthocyanin synthesis are co-regulated (Koes et al. 1994, 2005). In Petunia, DFR, ANS, 3GT, and several more downstream enzymes are all controlled by the an1/an11/an2 complex in the corolla and an1/an11/an4 in anthers (Quattrocchio et al. 1993, 1998, 1999; Mol et al. 1998; Spelt et al. 2000). In snapdragon, this coregulated block extends further upstream to include F3H (Martin et al. 1991; Schwinn et al. 2006). In Ipomoea and maize, however, the entire metabolic pathway is regulated as a single block (Quattrocchio et al. 1993; Mol et al. 1998; Morita et al. 2006). The transcriptional regulation of anthocyanin synthesis enzymes by the WD40/bHLH/MYB complex is likely direct (Spelt et al. 2000; Koes et al. 2005), but the cis-regulatory elements of the enzyme loci have not been characterized in detail.

DEVELOPMENT OF DROSOPHILA PIGMENTATION

Cuticular pigments of Drosophila are polymerized catecholamine derivates (Wright 1987) (Fig. 3). Pigment precursors are secreted by epidermal cells and incorporated into the overlying cuticle, so that pigmentation is nearly cell-autonomous. The synthesis of all pigments begins with the conversion of tyrosine to dihydroxyphenylalanine (dopa) by the tyrosine hydroxylase encoded by the pale gene. Some dopa is then converted to black melanin by extracellular enzymes encoded by the yellow gene family (Han et al. 2002; Wittkopp et al. 2002b). In another branch of the pathway, Dopa Decarboxylase (Ddc) converts dopa to dopamine, which serves as a precursor for brown melanin. Alternatively, dopamine can be shunted toward the production of light pigments. The product of the ebony gene converts dopamine to N-β-alanyldopamine (NBAD), the precursor of yellowish sclerotin. NBAD synthesis is reversible, and some portion of it is converted back into dopamine by an NBAD hydrolase encoded by the tan gene (True et al. 2005). Finally, a family of dopamine-acetyl-transferases (DATs) converts dopamine to N-acetyl dopamine (NADA), which serves as a precursor for colorless sclerotin. The final polymerization of cuticular pigments is controlled by extracellular phenoloxidases (Wright 1987; Kawabata et al. 1995; True et al. 2001). Ddc, ebony, tan, and yellow are transcribed and translated in the epidermis during the pupal stage (Kraminsky et al. 1980; Walter et al. 1996; Wittkopp et al. 2003a). However, the most upstream step in pigment synthesis—the conversion of tyrosine into dopa by the Ple enzyme—does not take place until eclosion, when Ple is activated post-transcriptionally by a hormonal cascade involving ecdysis-triggering hormone and eclosion hormone (Davis et al. 2007).

Ddc, ebony, tan, and yellow are transcribed and translated in the epidermis during the pupal stage (Kraminsky et al. 1980; Walter et al. 1996; Wittkopp et al. 2003a). The spatial color pattern is determined, at least in part, by the differential expression of these genes in different regions (Wittkopp et al. 2002a,b; Futahashi and Fujiwara 2005; Ninomiya et al. 2006), which depends on several regulatory pathways. In D. melanogaster, most abdominal segments bear a posterior stripe of dark pigment. This pattern is regulated by the Hh signaling pathway acting through the transcription factor optomotor-blind (omb) (Kopp and Duncan 1997; Kopp et al. 1997). Most members of the melanogaster species group have an additional sex- and segment-specific pattern: the last two abdominal segments in males have uniform black pigmentation that masks the usual pigment stripes. This pattern is repressed in females by the expression of two related transcription factors encoded by the bric a brac (bab) locus, which is regulated in turn by the HOX gene Abdominal-B (Abd-B) and the sex determination gene doublesex (dsx) (Kopp et al. 2000; Williams et al. 2008). The regulatory connections between the transcription factors and the enzymes that ultimately mediate their functions are not yet clear. Although Abd-B is known to regulate yellow expression directly (Jeong et al. 2006), other direct transcriptional targets of Abd-B, bab, and omb remain unknown, and we cannot rule out the existence of an intermediate regulatory layer.

DEVELOPMENT OF VERTEBRATE PIGMENTATION

The production of melanins in melanocytes is stimulated by melanocortin (MC) hormones. MCs, which include α-, β-, and γ-MSH (Melanin Stimulating Hormones) and the adrenocorticotropic hormone (ACTH), are encoded by the proopiomelanocortin (POMC) gene (Pritchard and White 2007; Ducrest et al. 2008). POMC is expressed in the pituitary gland, CNS, reproductive tract, skin, and other tissues. Several tissue-specific prohormone convertases cleave the POMC protein into multiple peptide hormones that, in addition to MCs, include lipotropins and endorphins. MCs act by binding to several melanocortin receptors (MC1R–MC5R) that are expressed in tissue-specific patterns (Fig. 4). Melanocytes express mainly MC1R, although other MCRs are expressed in the skin at low levels and MC1R itself is also expressed in other tissues such as pituitary and adrenal glands and immune cells. Melanocortins and MCRs have overlapping binding affinities. MC1R binds mainly α-MSH but also β-MSH and ACTH, whereas each of these hormones binds to at least one other MCR (Garcia-Borron et al. 2005; Pritchard and White 2007; Ducrest et al. 2008). The Agouti signaling protein acts as an antagonist of MCs, competing for the binding to MCRs and inducing opposite effects upon binding (Dinulescu and Cone 2000; Garcia-Borron et al. 2005; Ducrest et al. 2008). Agouti is expressed in the dermal papilla cells adjacent to melanocytes as well as in other tissues such as adipocytes, muscle, and gonads. The MC signaling system is conserved across vertebrates (Schioth et al. 2005), although differences in tissue-specific expression of MCRs, Agouti, and prohormone convertases exist between birds and mammals, and even between mice and humans (Dinulescu and Cone 2000; Ducrest et al. 2008).

MC1R encodes a G-protein coupled receptor. The binding to an MC ligand leads to the activation of cAMP secondary messenger system and the CREB transcription factor, which activates the expression of MITF (microophthalmia-associated transcription factor) (Garcia-Borron et al. 2005; Lin and Fisher 2007). In addition to melanocytes, MITF is essential for the development of osteoclasts, mast cells, and retinal pigmented epithelium (RPE). MITF expression is controlled by multiple cis-regulatory elements, including a melanocyte-specific enhancer (Steingrimsson et al. 2004; Levy et al. 2006). In addition to melanin production, MITF is required for cell survival, migration, and cell cycle control (Steingrimsson et al. 2004; Levy et al. 2006). Like the melanocortin signaling system, MITF is conserved in all vertebrate species.

MITF regulates pigmentation by activating the expression of multiple enzymes involved in melanin synthesis (Slominski et al. 2004; Levy et al. 2006). The rate-limiting step in melanin production is the hydroxylation of tyrosine to L-Dopa, which is catalyzed by the Tyrosinase enzyme (del Marmol and Beermann 1996; Hearing 1999; Slominski et al. 2004). The binding of MCs to MC1R leads to increased Tyrosinase expression and DOPA production; conversely, Agouti decreases Tyrosinase expression and results in lower Dopa levels (Le Pape et al. 2008). Tyrosinase has multiple enzymatic activities, which also include the oxidation of Dopa to Dopa-quinone and the conversion of indoles to indolequinones (del Marmol and Beermann 1996; Hearing 1999). Tyrosinase is expressed in multiple cell types, and its expression is controlled by several tissue-specific enhancers. In particular, different cis-regulatory elements activate Tyrosinase expression in melanocytes and RPE (Murisier and Beermann 2006; Murisier et al. 2006, 2007a). The melanocyte enhancer is co-activated by MITF and Sox10 (Murisier et al. 2007b). In addition to Tyrosinase, MITF induces the expression of several downstream components of melanin synthesis, including Tyrp1, Tyrp2, silver, Aim-1, and others (Slominski et al. 2004; Steingrimsson et al. 2004; Murisier et al. 2006; Lin and Fisher 2007).

The melanin pathway branches below Dopa-quinone to produce either dark eumelanins or light pheomelanin (Hearing 1999; Le Pape et al. 2008). The mechanisms that regulate the ratio of these pigments are not fully understood. Because Agouti expression correlates with the spatial distribution of pheomelanin and constitutive MC1R activation results in overabundance of eumelanin, it is hypothesized that Agouti acts as a switch that favors the production of pheomelanin at the expense of eumelanin (Dinulescu and Cone 2000). However, MC1R; Agouti double mutants produce pheomelanin, and Agouti treatment decreases both eu- and pheomelanin synthesis in cell culture assays, suggesting that Agouti may act by downregulating the melanin pathway as a whole rather than any specific branch (Le Pape et al. 2008).