Evolution of petal identity

Abstract

Petals appear in many angiosperm taxa, yet when and how these attractive organs originated remains unclear. Phylogenetic reconstructions based on morphological data suggest that petals have evolved multiple times during the radiation of the angiosperms. Based on the diversity of petal morphologies, it is likely that the developmental programmes specifying petal identity are distinct in different lineages. On the other hand, molecular genetic analyses have suggested that the specification of petal identity in different lineages utilizes similar genetic pathways. Together, these observations indicate that the evolution of petals has relied on the repeated recruitment of a suite of interacting developmental control genes, albeit in different ways in different lineages. These observations suggest that this gene regulatory network represents a ‘deep homology’ in plant evolution. A major challenge is to understand how this ancestral developmental pathway has been redeployed in different angiosperm lineages, and how changes in the workings of this pathway have led to the myriad shapes, colours, and sizes of petals.

Angiosperm, Arabidopsis, development, evolution, flower, MADS, petal

Introduction

Angiosperm flowers generally contain petals that display considerable diversity in their size, morphology, and colour. The brilliant hues and complex forms of petals are thought to have evolved to facilitate pollinator attraction (Galliot et al., 2006; Cronk and Ojeda, 2008). Other floral organ types can occasionally display these showy attributes, but petals can be uniquely defined based on their position surrounding the stamens, as well as by their morphology (Hiepko, 1965; Weberling, 1989). Flowers generally consist of a number of perianth organs surrounding the male and female reproductive organs. The sterile laminar organs that make up the perianth can be ‘undifferentiated’, consisting of similar leaf-like organs, often referred to as tepals (Doyle and Endress, 2000; Endress, 2006; Ronse De Craene, 2007; Endress and Doyle, 2009). Alternatively, perianths can be ‘differentiated’ into distinct outer and inner organ types. Most often, differentiated perianths are composed of outer sepals and inner petals, but some species possess novel perianth organ types, such as the grasses that possess a perianth composed of paleas, lemmas and lodicules.

Petals generally have a narrow base, one vascular trace, and usually lag behind the stamens in their growth (Endress and Doyle, 2009). The petal blade is often large and showy, and often white or pigmented, due to the presence of leucoplasts or chromoplasts instead of chloroplasts. Furthermore, the adaxial epidermal cells are generally conical or elongate, giving a characteristic velvety sheen to petals (Weberling, 1989). Volatile oils, including terpenoids, benzenoids, and other aromatic compounds, are frequently produced in petal epidermal cells, giving flowers their characteristic scents (Dudareva and Pichersky, 2000).

In contrast to true petals, some species have petaloid organs that possess morphological attributes of petals in other locations. For instance, Cornus florida, the flowering dogwood, possesses large, showy, petal-like bracts. It seems likely that such examples represent cases where aspects of petal developmental pathways have been ectopically expressed in novel locations.

Independent origins of petals

Across the angiosperms, there is considerable variation in whether flowers possess tepals or differentiated perianths with sepals and petals (Fig. 1). Tepals are common in the monocots; for instance, lilies possess showy perianths consisting of tepals. By contrast, most core eudicot species possess distinct sepals and petals. Morphological and phylogenetic evidence has been used to suggest that petals (e.g. distinct inner perianth organs) have evolved multiple times. In some angiosperm lineages, petals are thought to have arisen as modifications of stamen-like structures (andropetaloidy), while in other lineages, petals appear to be derived as modifications of bract- or leaf-like organs (bracteopetaloidy) (Eames, 1961; Hiepko, 1965; Weberling, 1989; Takhtajan, 1991). These distinctions are based on a number of morphological characteristics, including the number of vascular traces, and patterns of primordium initiation and phyllotaxy. By mapping differences in perianth morphologies (Fig. 1) onto a robust phylogeny, it has been estimated that there have been as many as six independent origins of differentiated perianths within the angiosperms (Zanis et al., 2003). The idea that petals arose as a secondary modification of pre-existing organ types is also implicit in more recent models, which invoke alterations in organ identity through changes in the expression of homeotic genes as underlying the evolution of petals (Albert et al., 1998; Baum, 1998; Ronse De Craene, 2007). Variations on this idea have also been applied to explain the graded transitions from outer to inner tepals in some clades of basal angiosperms (Soltis et al., 2007a).

Fig. 1.

Phylogeny of the angiosperms, showing distribution of perianth types among major angiosperm clades. It is clear that a differentiated perianth arose multiple times during the course of angiosperm evolution. Core eudicots are thought to have had a single independent origin of a bipartite perianth containing petals. The ancestral character state for the angiosperms was likely to have been an undifferentiated perianth. (Adapted from Hileman and Irish, 2009, and reproduced by kind permission of the Botanical Society of America.)

Given this diversity in differentiated perianth types, there has been some debate as to the probable morphology of the ancestral angiosperm flower. The recent advances in resolving angiosperm phylogeny have considerably refined our views as to how perianths may have evolved (Soltis et al., 2005). It seems likely that the ancestral angiosperm possessed relatively small, few-parted flowers with an undifferentiated perianth. The identification of Amborella trichopoda, with its diminutive spirally-arranged tepals, as the probable basal-most extant angiosperm species has provided support for this conclusion (Mathews and Donoghue, 1999; Parkinson et al., 1999; Qiu et al., 1999; Soltis et al., 1999; Zanis et al., 2002). These phylogenetic analyses also strongly support the placement of the Nymphaeales, Illiciaceae, Schisandraceae, Trimenaceae, and Austrobaileyaceae, all with undifferentiated perianths, as the successive basal branches of the angiosperm tree. In fact, most basal angiosperm species possess tepals, although such tepals can show gradations from outer, more sepaloid, to inner, more petaloid characteristics (Hiepko, 1965; Endress, 2001; Endress and Doyle, 2009).

Fossil evidence, although subject to considerable debate, has also been used to suggest that the ancestral angiosperm flower lacked petals. In particular, the aquatic fossil plant Archaefructus, which lacked petals, has been suggested to represent the sister group to all extant angiosperms (Sun et al., 2002). However, this interpretation has been questioned, with more recent analyses suggesting that this fossil represents a more recently evolved, derived aquatic form, in which petals may have been secondarily lost (Friis et al., 2003). Nonetheless, given the preponderance of extant basal angiosperm species lacking petals it seems likely that the ancestral angiosperm flower possessed an undifferentiated perianth, although it is still formally possible that all these instances represent secondary losses of petals.

If one accepts the hypothesis that the ancestral angiosperm flower possessed an undifferentiated perianth composed of a series of tepals, how did differentiated perianth organs evolve in different lineages? Understanding the course of perianth evolution can have ramifications for developing models for how gene regulatory networks have diversified to give rise to petals.

Molecular genetics of petal specification

With the advent of molecular and developmental genetic approaches to dissecting the flowering process, there has been great interest in examining whether such data can be profitably used to suggest mechanisms by which differentiated perianths evolved in different lineages. Most work has focused on the potentially changing roles of a set of MADS box genes that have been shown in Arabidopsis, as well as other core eudicot species, to play critical roles in the specification of different floral organ identities, including petal identity. Far less is known of the mechanisms involved in regulating petal growth, size, and shape (Anastasiou and Lenhard, 2007; Irish, 2008).

The genetic control of petal identity, that is, the specification of unique cell types, colours and morphologies associated with petals in given taxa has been investigated extensively in the core eudicots Arabidopsis thaliana and Antirrhinum majus. A number of pioneering studies in these two species were instrumental in the development of the ABC model (Schwarz-Sommer et al., 1990; Bowman et al., 1991; Coen et al., 1991; Coen and Meyerowitz, 1991; Trobner et al., 1992). This model posits that combinations of homeotic gene expression can uniquely specify different organ identities. The subsequent molecular characterization of these homeotic genes has shown that most encode MADS-domain containing transcription factors (Weigel and Meyerowitz, 1994; Jack, 2001). In Arabidopsis petals, the combined expression of the APETALA1 (AP1), PISTILLATA (PI), APETALA3 (AP3), and SEPALLATA (SEP) MADS box genes are both necessary and sufficient for petal identity specification (Jack et al., 1994; Krizek and Meyerowitz, 1996; Honma and Goto, 2001; Pelaz et al., 2001). Furthermore, the products of these genes appear to act as components of a transcriptional complex and presumably act to up-regulate the transcription of petal-specific factors (Honma and Goto, 2001). What has given credence to the idea that this ABC programme is widely conserved across angiosperms were the early findings that the homologous genes in Antirrhinum play essentially the same role (Schwarz-Sommer et al., 1990; Coen, 1992; Egea-Cortines et al., 1999).

Nonetheless, there are notable difficulties with this simple assumption. First of all, orthologues of some of these genes simply do not exist in some angiosperm clades. In other cases, orthologues exist, but do not possess the same biological function as their Arabidopsis or Antirrhinum counterparts. This is a critical issue, as in many cases the expression of a given orthologue has been interpreted as defining the function of that gene. However, expression clearly does not always equate to function. The number of species for which functional approaches can be taken are limited, yet these studies have been essential in developing more sophisticated models for the evolving roles of the MADS box genes. Below, the evidence (or lack thereof) for a functional role of AP1, AP3, PI, and SEP genes in petal identity specification across the angiosperms is summarized.

APETALA1

In Arabidopsis, AP1 is required for both petal and sepal development, since ap1 mutants show homeotic transformations of sepals to bract-like structures while petals are lacking (Irish and Sussex, 1990; Bowman et al., 1993). These defects in sepal and petal development in ap1 mutants, in combination with the localized expression of AP1 in sepal and petal primordia, have led to the idea that AP1 plays a critical role in organ identity specification (Mandel et al., 1992). AP1 also has a role in specifying floral meristem identity, in that ap1 mutants show a partial transformation of flowers to inflorescence structures (Irish and Sussex, 1990; Bowman et al., 1993). This effect appears to be due to the up-regulation of at least three genes, AGL24, SOC1, and SVP (Yu et al., 2004; Liu et al., 2007); double ap1 agl24 mutants show a restoration of nearly normal flowers, indicating that it is the ectopic expression of AGL24 that is largely responsible for the partial inflorescence transformation seen in ap1 mutants (Yu et al., 2004). Thus Arabidopsis AP1 clearly has a distinct role in meristem specification, and it is this role in meristem specification that appears to be conserved in other core eudicot species. The Antirrhinum orthologue of AP1, SQUAMOSA (SQUA), lacks any overt role in floral organ development, since squa mutants only show defects in meristem identity, but not in sepal or petal formation (Huijser et al., 1992). Similarly, mutations in the tomato and Pisum sativum AP1 orthologues only affect meristem identity, and do not show defects in sepal or petal development (Berbel et al., 2001; Taylor et al., 2002; Vrebalov et al., 2002).

Phylogenetic analyses have shown that a major gene duplication event occurred in the AP1 lineage at the base of the core eudicots, leading to the euAP1 (containing Arabidopsis AP1), euFUL, and FUL-like clades of genes (Litt and Irish, 2003). As such, defining orthologous relationships among AP1-related genes in angiosperms is complicated by gene duplication and consequent sequence diversification. Functional studies have been carried out for one non-core eudicot AP1 paralogue, the wheat WAP1/VRN1 gene which is required for the vernalization-dependent transition to flowering (Danyluk et al., 2003; Murai et al., 2003; Trevaskis et al., 2003; Yan et al., 2003; Shitsukawa et al., 2007). Thus the function of the monocot WAP1/VRN1 gene may have some similarities to Arabidopsis AP1 in terms of regulating aspects of meristem identity, but clearly this paralogue does not possess equivalent functions in terms of organ identity. Together, these observations throw into considerable doubt the idea that AP1 lineage genes play a role in petal identity specification across the angiosperms (Litt, 2007).

APETALA3

A duplication in the AP3 lineage at the base of the core eudicots resulted in the paralogous euAP3 and TM6 lineages (Kramer et al., 1998; Kramer and Irish, 1999), with Arabidopsis AP3 being a member of the euAP3 lineage. Arabidopsis AP3 is required for specifying both petal and stamen identities, as ap3 mutations cause homeotic transformations of both petals and stamens (Bowman et al., 1989; Jack et al., 1992). Similar homeotic transformations are observed for the loss of the euAP3 orthologue function from a number of core eudicot species, including A. majus, Petunia hybrida, Nicotiana benthamiana, Solanum lycopersicum, and Gerbera hybrida (Sommer et al., 1990; van der Krol et al., 1993; Yu et al., 1999; Liu et al., 2004; de Martino et al., 2006). Together with ectopic expression studies (Krizek and Meyerowitz, 1996), these analyses suggest that the role of euAP3 genes in specifying petal identity has been conserved across core eudicots. By contrast, loss of function analyses of the paralogous TM6 lineage genes in P. hybrida and S. lycopersicum suggest that TM6 genes function in the determination of stamen identity, and have little or no role in petal development (de Martino et al., 2006; Rijpkema et al., 2006).

The functions of paleoAP3 lineage genes, present in non-core eudicots, have now been analysed in several species. Loss-of-function analyses of the maize and rice paleoAP3 genes has demonstrated that these gene products are required for both lodicule and stamen development in these grass species (Ambrose et al., 2000; Nagasawa et al., 2003). This observation has been interpreted in two ways: on the one hand, the corresponding roles of paleoAP3 and euAP3 genes may reflect an underlying homology of lodicules with petals (Whipple et al., 2004). Although the term homology has been used in a variety of ways, each with its own implicit biases (Hall, 1994; Fitch, 2000), morphological homology is generally taken to mean the presence of a structure in the most recent common ancestor; as such, this interpretation would suggest a morphological continuity between grass lodicules and dicot petals. However, it is not clear that lodicules are morphologically homologous to petals, since the transitional morphological forms between non-grass monocot petals and grass lodicules are tepals, bracts, or lacking (Dahlgren et al., 1985; Whipple et al., 2007).

Alternatively, the role of paleoAP3 genes in maize and rice may reflect the recruitment of paleoAP3 gene function to the specification of a grass-specific novel structure, the lodicule (Irish, 2000). This interpretation would imply that there is ‘deep homology’ between petals and lodicules, despite the lack of morphological homology. The term deep homology (or process homology) was coined to describe situations in which genetic processes are conserved, even if the resulting morphological features are not (Shubin et al., 1997; Gilbert and Bolker, 2001; Shubin et al., 2009). Analyses of paleoAP3 gene expression across the grasses has provided evidence for the consistent demarcation of an expression domain in the stamens and in the adjacent tissue, suggesting that the expression of the paleoAP3 genes is defining a specific region within the flower (Whipple et al., 2007). Furthermore, the paleoAP3 gene of the monocot Asparagus officinalis is expressed in inner tepals and stamens, indicating that various organ types can arise in the domain specified by paleoAP3 gene expression (Park et al., 2003). As such, the appearance of lodicules in the grasses and of petals in the core eudicots could reflect the parallel evolution of different organ types based on a common developmental genetic framework specifying a domain within the floral meristem.

Further evidence for the deep homology of morphologically non-homologous second whorl organs has come from recent analyses of paleoAP3 function in Papaver somniferum, a basal eudicot (Drea et al., 2007). Papaver possesses two paleoAP3 genes that appear to have arisen by a recent duplication within the Ranunculales. Loss-of-function analyses of these gene duplicates indicate that one is required predominantly for petal identity, while the other is required for stamen identity specification. Phylogenetic analyses have been used to suggest that a differentiated perianth has evolved multiple times, with a probable independent origin of petals in the Ranunculales (Zanis et al., 2003; Soltis et al., 2005). Thus Papaver petals appear to be non-homologous with core eudicot petals, again suggesting a case of parallel evolution of a novel organ type via reutilization of an AP3 dependent pathway.

How, then, has the diversity of differentiated perianths arisen from a common conserved developmental programme? Presumably, the action of AP3 lineage gene products, in concert with their interacting MADS box protein partners, has diversified extensively in the regulation of downstream genes. This could occur by diversification in the binding capabilities of AP3 proteins, and/or through diversification in cis-regulatory elements of downstream genes. Microarray analyses in Arabidopsis and Antirrhinum have identified a number of downstream target genes regulated by AP3 lineage proteins (Zik and Irish, 2003; Bey et al., 2004; Wellmer et al., 2004; Mara and Irish, 2008). Comparative analyses of these target genes, and the modes by which they are regulated in different species, should be valuable in determining how this diversification occurred.

PISTILLATA

The evolutionary history of the PI gene lineage is peppered with both relatively ancient and many more recent gene duplications (Chung et al., 1995; Kramer et al., 1998, 2003; Aoki et al., 2004; Kim et al., 2004; Stellari et al., 2004; Vandenbussche et al., 2004; Jaramillo and Kramer, 2007). In Arabidopsis, PI is required to specify both petal and stamen identities (Goto and Meyerowitz, 1994), as is its Antirrhinum counterpart (Trobner et al., 1992). In other core eudicots, there is some variation in the roles of PI orthologues that appears to reflect subfunctionalization of PI gene action as a consequence of gene duplication. For instance, in Petunia there are two PI orthologues; mutations in each show only mild defects in petal and stamen development, but the double mutant displays a strong homeotic transformation of both petals and stamens, indicating that these gene products have partially redundant functions (Angenent et al., 1993; Vandenbussche et al., 2004). In G. hybrida, the function of only one PI orthologue has been described and has a predominant role in petal specification (Yu et al., 1999). It seems likely though, that an as yet undescribed PI duplicate exists in Gerbera which functions to specify stamen identity.

Functional studies of PI orthologues have also been carried out in the basal eudicot species Papaver somniferum and Aquilegia vulgaris. P. somniferum possesses two PI lineage genes that arose from a recent gene duplication event within the Papaveraceae (Drea et al., 2007). Loss-of-function analyses have shown that one of these PI genes is required for both petal and stamen specification, while the other gene has no overt functional role and may be on the way to becoming a pseudogene (Drea et al., 2007). By contrast, A. vulgaris contains only one PI gene, which appears to be responsible for specifying petals and stamens (Kramer et al., 2007). However, the observed loss-of-function phenotype may also be due to co-ordinate down-regulation of the A. vulgaris AP3 homologues (Kramer et al., 2007).

Two PI lineage genes, OsMADS2 and OsMADS4, have been identified in rice, a monocot grass. As demonstrated by RNAi loss of function studies, OsMADS2 functions in lodicule development (Prasad and Vijayraghavan, 2003; Yadav et al., 2007). Functional analyses of OsMADS4 have produced inconsistent results, with reports of a role for OsMADS4 in both lodicule and stamen development (Kang et al., 1998), while other studies have indicated that loss of OsMADS4 function has no phenotypic effects (Yoshida et al., 2007; Yao et al., 2008). However, RNAi-induced loss of function of OsMADS2 and OSMADS4 together affects the development of both lodicules and stamens, indicating that these genes operate redundantly in stamen development (Yao et al., 2008).

Together, these analyses suggest that the ancestral role of the PI lineage was in specifying a floral domain consisting of stamens and the adjacent region. In cases where there are duplicate PI genes, the combined effects of PI gene function appear generally to encompass the specification of stamens and adjacent organ types. This is analogous to the situation for AP3, again suggesting a deep homology for PI gene function in terms of a role in regional specification in the flower.

SEPALLATA

The SEP gene family has also undergone considerable expansion through multiple gene duplications. One duplication occurred prior to the radiation of the angiosperms, with many subsequent duplications (and losses) at all phylogenetic levels (Malcomber and Kellogg, 2005; Zahn et al., 2005). Arabidopsis possesses four SEP genes, which individually show only subtle mutant phenotypes (Pelaz et al., 2000; Ditta et al., 2004). The quadruple sep1 sep2 sep3 sep4 mutant, though, displays a complete loss of floral organ identities, indicating that these genes are redundantly required for specification of all floral organ types (Ditta et al., 2004).

The redundancy of the SEP genes in Arabidopsis, coupled with the duplication history of this gene lineage, has made it difficult to tease out a common functional role for the SEP genes. In many cases, only one or a few of the SEP genes have been analysed in a given species, making it difficult to draw conclusions as to the common functions of these genes. For instance, the Petunia fbp5 (SEP1/2 orthologue) mutant shows no obvious phenotype while the fbp2 mutant (SEP3 orthologue) displays mild transformation of petals to more sepaloid structures (Vandenbussche et al., 2003). The double fbp2 fbp5 mutant, though, shows a more pronounced transformation of floral organs to a more leaf-like state, demonstrating the redundancy of these gene products in Petunia (Vandenbussche et al., 2003). At least two other Petunia SEP genes exist (Immink et al., 2003; Vandenbussche et al., 2003), but triple or quadruple mutants have not yet been analysed. By contrast, antisense-mediated loss of function of GRCD1, the G. hybrida SEP3 orthologue, results in a transformation of sterile stamens to more petaloid structures (Kotilainen et al., 2000). Furthermore, cosuppression of the G. hybrida SEP1/2 orthologue, GRCD2 results in a novel phenotype, in which the carpels are petaloid, the flowers are indeterminate, and the inflorescences themselves are larger (Uimari et al., 2004). These Gerbera results have suggested that SEP gene functions may have diversified beyond simply controlling floral organ identities (Teeri et al., 2006). A diversification in SEP functions is also hinted at by the phenotype of the tomato rin mutant. This mutant corresponds to a complex rearrangement at the tomato SEP4 gene and fails to produce ripe fruit (Vrebalov et al., 2002). However, the aberrant fusion protein produced by the rin locus may, in fact, have novel functions and so not truly reflect the role of the endogenous locus (Vrebalov et al., 2002; Ito et al., 2008).

A number of monocot-specific SEP-gene duplications have also been documented (Malcomber and Kellogg, 2005; Zahn et al., 2005). In rice, a semi-dominant negative mutation in one of these SEP genes displays an increase in leafiness of floral organs as well as defects in organ number (Jeon et al., 2000). However, true loss-of-function analyses have not yet been carried out for any of the monocot SEP genes, leaving open the question as to the extent to which these genes have similar functional roles to their eudicot counterparts.

From these observations, it is difficult to define a common role for SEP genes across angiosperms, While their role in Arabidopsis petal specification seems clear, it is unclear the extent to which such a role can be extrapolated even to other core eudicot species. In fact, these observations indeed may reflect a considerable diversification of SEP gene function in different clades. Analyses of SEP gene products from tomato and Arabidopsis have indicated that they have considerable promiscuity in their ability to heterodimerize (Leseberg et al., 2008; Immink et al., 2009), also supporting the idea that SEP gene products may have a relatively non-specific role in organ identity specification.

MADS box protein complexes

Implicit in the ABC model is the idea that the MADS box genes function together to specify distinct organ identities. Protein interaction studies have provided a mechanistic basis for understanding how this combinatorial function may play out in vivo (Honma and Goto, 2001; Jack, 2001; Theissen and Saedler, 2001). Most transcription factors bind to DNA as dimers, and the MADS box gene products are no exception. In Arabidopsis, various interactions have been observed between the gene products necessary for petal specification. AP3 and PI proteins do not form DNA-binding homodimers, but instead bind to DNA as obligate heterodimers (Riechmann et al., 1996a, b). Furthermore, the AP3/PI heterodimer forms multiprotein complexes with both SEP3 and AP1 in vitro, suggesting that interactions between multiple MADS box proteins are necessary in vivo for specifying organ identities (Egea-Cortines et al., 1999; Honma and Goto, 2001; Jack, 2001; de Folter et al., 2005; Immink et al., 2009). This interaction serves to bring the necessary transactivation capabilities of AP1 and SEP3 to the complex since both AP3 and PI lack a transcriptional activation domain (Honma and Goto, 2001). These complexes may potentially be larger and incorporate other, non-MADS proteins, as evidenced by the interactions of AP1 and SEP3 with transcriptional co-repressor proteins required for petal development (Franks et al., 2006; Sridhar et al., 2006).

To what extent are these protein complex interactions conserved across angiosperms? This has been examined most extensively by assessing the dimerization capabilities of AP3 and PI orthologues. These interactions have been examined in a number of core eudicot species, and, in general, AP3 and PI homologous proteins display obligate heterodimerization (Davies et al., 1996; Riechmann et al., 1996a). In some monocot and gymnosperm species, AP3 and PI lineage gene products have been shown to homodimerize (Winter et al., 2002; Tzeng et al., 2004; Tsai et al., 2008), suggesting that heterodimerization evolved from homodimerization (Winter et al., 2002). There is evidence for positive selection acting on residues involved in AP3–PI heterodimerization, suggesting that the evolution of this protein complex had adaptive significance (Hernandez-Hernandez et al., 2007). In fact, across the eudicots, there is no evidence for homodimerization of AP3 or PI gene products, suggesting that obligate heterodimerization evolved prior to the radiation of this clade (Winter et al., 2002).

It is also clear that there is considerable plasticity to AP3–PI heterodimerization, in that the products of AP3 and PI gene duplicates can have qualitatively distinct interaction specificities. For instance, both tomato and Petunia possess duplicate AP3 and PI genes (Vandenbussche et al., 2003, 2004; de Martino et al., 2006; Hileman et al., 2006). The subsequent diversification of these duplicate copies has resulted in preferential interactions with particular partners, suggesting that these proteins may have non-equivalent roles in specifying petal identity (Immink et al., 2003; Vandenbussche et al., 2004; de Martino et al., 2006; Rijpkema et al., 2006; Leseberg et al., 2008). Similarly, in the basal eudicot species Papaver somniferum and Aquilegia vulgaris, duplicated AP3 and PI gene products have been shown to have differential interaction capabilities (Drea et al., 2007; Kramer et al., 2007).

Given the flexibility in interactions, it seems likely that evolutionary shifts in MADS complex function could easily have occurred through a shift in protein partner specificities, as opposed to, or in addition to, shifts in expression domains of MADS box proteins. Changes in protein complex interactions could have two potential outcomes. One the one hand, the co-evolution of protein partners could occur to maintain a particular regulatory interaction. Alternatively, a shift in complex composition could drive changes in protein–DNA interactions, resulting in changes in the pathways regulated by orthologueous MADS box genes. Global gene expression analyses are beginning to define the transcriptional targets of MADS box gene products in diverse species (Zik and Irish, 2003; Bey et al., 2004; Wellmer et al., 2004; Sundstrom et al., 2006; Mara and Irish, 2008), but it is still unclear the extent to which trans-acting versus cis-acting shifts in regulatory networks are responsible for petal specification in different taxa.

A model for the evolution of petals

The accumulating functional data on the action of MADS box genes in specifying inner perianth identity in different clades indicates that, where tested, AP3 and PI orthologues are involved in specifying a regional domain consisting of inner perianth and stamens (Fig. 2). It is important to point out that this inner perianth domain does not necessarily correspond to petals. Organs that arise in this domain specified by AP3/PI can be inner tepals (e.g. asparagus: Park et al., 2003), or grass-specific lodicules (e.g. maize, rice: Ambrose et al., 2000; Nagasawa et al., 2003; Prasad and Vijayraghavan, 2003; Whipple et al., 2007; Yadav et al., 2007). Furthermore, petals have arisen independently in different clades (Fig. 1), and so petals in core eudicots, for instance, are not homologous to petals of the Ranunculales (Drea et al., 2007). Together, these observations underscore the idea that the ancestral function of AP3/PI was to specify a regional domain, not a particular organ type. Thus the network controlled by AP3/PI represents an instance of deep homology, where a cascade of regulatory interactions has been redeployed repeatedly in different contexts (Shubin et al., 1997, 2009).

Fig. 2.

A model for the evolution of different inner perianth types. A simplified phylogeny of the angiosperms, showing major clades, and cartoon examples of flower forms in each of these clades. For each example, the type of inner perianth organ is indicated. The presumed dimerization capabilities of the MADS box gene products for each clade are summarized, with the respective gene products represented by a circle. SEP lineage gene products, green; paleoAP3/TM6, dark blue; euAP3, light blue; PI, purple; AG, red. AG lineage MADS box genes have been implicated in reproductive organ identity in a number of angiosperm species (see Kramer et al., 2004; Causier et al., 2005; Zahn et al., 2006). The presumed domains of function (not expression) of these dimers are indicated by boxes.

This idea of deep homology can accommodate the many independent origins of differentiated perianth types, in that it does not invoke a specific organogenic role for AP3/PI genes. Rather, this model implies that differences in perianth types arose from differences in the ways in which the AP3/PI regulatory network was deployed in different lineages. For instance, this model can explain both instances of andropetaloidy and bracteopetaloidy, in that it is plausible to suppose that AP3/PI homologues may have evolved the capacity to regulate genes involved in aspects of stamen differentiation in some lineages, while in other clades AP3/PI acquired the capacity to regulate genes previously involved in aspects of leaf-like differentiation. In other words, differences in perianth types would be a consequence of the evolution of distinct differences in the cascade of genes regulated by AP3/PI. These kinds of shifts in the regulatory roles of AP3/PI would presumably be due to changes in their protein partners and/or changes in the cis-regulatory elements of downstream genes (Fig. 2).

The model presented here differs conceptually from earlier models such as the sliding borders or fading borders models (Bowman, 1997; Kramer and Jaramillo, 2005; Soltis et al., 2007b) that explicitly propose a role for AP3/PI genes in specifying ‘petaloid’ identity. The sliding borders and fading borders models presuppose that simple shifts in the domains or levels of expression of MADS box genes are the drivers of organ identity in different angiosperm clades. While shifts in MADS box gene expression are likely to play a role in the specification of some novel morphologies (e.g. the petaloid sepals of tulips; van Tunen et al., 1993; Kanno et al., 2003), expression changes alone cannot explain the diversity of perianth organ types.

If indeed the ancestral role of the homeotic MADS box genes was to specify regional domains, then the question remains as to how morphologically similar petals evolved multiple times independently. Discerning the similarities and differences in the genetic circuitry controlled by the MADS box genes will be crucial in determining what is conserved and what is novel in the specification of petal identities across angiosperms.

Work in my laboratory on the evolution of petal identity has been supported by grants from the National Science Foundation. I thank the members of my laboratory, past and present, who have been such valuable colleagues in developing some of the ideas presented here.