Search tips
Search criteria 


Logo of annbotAboutAuthor GuidelinesEditorial BoardAnnals of Botany
Ann Bot. Aug 2008; 102(2): 153–165.
Published online May 28, 2008. doi:  10.1093/aob/mcn082
PMCID: PMC2712368
Evolutionary Trends in the Flowers of Asteridae: Is Polyandry an Alternative to Zygomorphy?
Florian Jabbour,1,2* Catherine Damerval,2 and Sophie Nadot1
1Université Paris-Sud, Laboratoire Ecologie, Systématique, Evolution, CNRS UMR 8079, AgroParisTech, Orsay, F-91405, France
2UMR de Génétique Végétale, INRA – Univ Paris-Sud – CNRS – AgroParisTech, Ferme du Moulon, 91190 Gif-sur-Yvette, France
*For correspondence. E-mail florian.jabbour/at/
Received February 16, 2008; Revised March 17, 2008; Accepted April 21, 2008.
Background and Aims
Floral symmetry presents two main states in angiosperms, actinomorphy (polysymmetry or radial symmetry) and zygomorphy (monosymmetry or bilateral symmetry). Transitions from actinomorphy to zygomorphy have occurred repeatedly among flowering plants, possibly in coadaptation with specialized pollinators. In this paper, the rules controlling the evolution of floral symmetry were investigated to determine in which architectural context zygomorphy can evolve.
Floral traits potentially associated with perianth symmetry shifts in Asteridae, one of the major clades of the core eudicots, were selected: namely the perianth merism, the presence and number of spurs, and the androecium organ number. The evolution of these characters was optimized on a composite tree. Correlations between symmetry and the other morphological traits were then examined using a phylogenetic comparative method.
Key Results
The analyses reveal that the evolution of floral symmetry in Asteridae is conditioned by both androecium organ number and perianth merism and that zygomorphy is a prerequisite to the emergence of spurs.
The statistically significant correlation between perianth zygomorphy and oligandry suggests that the evolution of floral symmetry could be canalized by developmental or spatial constraint. Interestingly, the evolution of polyandry in an actinomorphic context appears as an alternative evolutionary pathway to zygomorphy in Asteridae. These results may be interpreted either in terms of plant–pollinator adaptation or in terms of developmental or physical constraints. The results are discussed in relation to current knowledge about the molecular bases underlying floral symmetry.
Key words: Floral symmetry, architectural constraints, Asteridae, comparative analysis, composite tree, correlated evolution, evolutionary scenario
The flower is a key innovation that arose in plants around 130 mya and investigating floral architecture is a way to understand better the rise and diversification of Angiosperms. The flower is commonly described as the combination of male and female reproductive structures within a small space on the same shoot, surrounded by sterile perianth organs (Irish, 2000; Theissen et al., 2002; Bateman et al., 2006). However, the flower–inflorescence boundary is sometimes blurred (Rudall and Bateman, 2006). A flower is usually characterized by the position, the number and the shape of its organs, constituting floral architectures that are characteristic of more or less high order taxa.
One of the components of floral architecture is symmetry. In its broader acception, floral symmetry points to perianth and androecium structures, very rarely encompassing the gynoecium that often presents specific reductions (Endress, 1999). Among the diversity observed in flowers, two main types of symmetry are recognized. Actinomorphy (polysymmetry or radial symmetry) refers to flowers where all the organs in the same whorl are identical and regularly placed on the receptacle. In contrast, zygomorphy (monosymmetry or bilateral symmetry) involves intrawhorl differentiation that generates a single symmetry plane, most often dorso-ventral. Exceptions are scarce and concern only a few taxa, such as some Fumariaceae that have disymmetric flowers (e.g. Hypecoum and Dicentra sensu lato) or some Leguminosae and Cannaceae that possess asymmetric flowers (Dahlgren et al., 1985; Tucker, 2003). Flowers with a spiral phyllotaxis are considered almost but not fully symmetric (Endress, 2006). It is striking that the most speciose families are those which contain zygomorphic flowers (Sargent, 2004) (e.g. Fabaceae and Asteraceae in the eudicots, Zingiberaceae and Orchidaceae in the monocots). Furthermore, these taxa generally have a closed ground plan, i.e. a fixed organ number and arrangement, as is the rule in monocots and core eudicots. In contrast, zygomorphy is quite rare in basal angiosperms and early diverging eudicots (Endress, 1999; Ronse De Craene et al., 2003) where open ground plans are widespread. A link may thus exist between symmetry and other traits of the flower. Such a link was investigated recently in a group of basal eudicots, the Ranunculales (Damerval and Nadot, 2007). In contrast to previous studies in which the evolution of floral symmetry was examined independently from other morphological characters (Donoghue et al., 1998; Ree and Donoghue, 1999), in the study of Damerval and Nadot (2007) the evolution of other floral traits was also considered in order to identify the morphological context of the evolution of symmetry. These authors showed that zygomorphy evolved three times independently in the Ranunculales, always in the context of a closed ground plan [consistent with the hypothesis of Endress (2001)]. Changes in symmetry occurred independently from changes in perianth merism and stamen number. Moreover, in bisexual species, zygomorphy coincided with the presence of a single nectar spur.
A similar approach is used in this paper, focusing on the Asteridae, one of the two major clades of the core eudicots (APG, 2003), in which flowers are characterized by closed ground plans. Subclass Asteridae includes up to 25 % of all angiosperms (Bremer et al., 2001). About half of the 65 000 species (Ree and Donoghue, 1999) have zygomorphic flowers (Thorne, 1992). The evolution of perianth symmetry was investigated in relation to perianth merism, the number of stamens and the number of floral spurs. The evolutionary history of each character was reconstructed on a composite phylogenetic tree of Asteridae. Correlated evolution between perianth symmetry and the characters listed above was then examined through phylogenetic comparative analyses. This type of analysis looks for correlated transitions between character states, and tests whether the evolution of a character considered as dependent is correlated with the evolution of another character considered as independent, taking into account the phylogenetic relationships between the taxa studied. These analyses have great potential for finding the taxa on which it is relevant to focus and for identifying cohorts of traits to be analysed from a developmental and molecular point of view. In the present study, correlated evolution was tested exclusively between morphological characters. This approach was undertaken to identify the architectural context in which zygomorphy can, or cannot, emerge in Asteridae, and to understand better the morphological context in which genes implicated in the shaping of the flower act.
Building a composite tree of Asteridae and taxon sampling
A composite tree is a combination of phylogenetic trees obtained from different datasets, each focused on a different taxonomic group. It is the only way to obtain a detailed phylogeny of high taxonomic level when no comprehensive phylogenetic study exists for this group based on a homogeneous dataset. Composite trees are useful when characters of genera or even species must be optimized on the phylogeny of a high taxonomic level clade (Sillen-Tullberg, 1993; Weiblen et al., 2000). Using the phylogenetic tree of the families of Asteridae available from the Angiosperm Phylogeny Website (Stevens, 2001 onwards), a composite tree was constructed by grafting onto this backbone detailed phylogenies of families and genera (see Supplementary Information, available online) where needed for character optimization. All families of Asteridae are represented by at least one leaf in the composite tree. If no evolutionary transition for any trait under scrutiny was recorded, the family was represented by one leaf only. In the case of existing evolutionary transitions in a family, as many taxa (genera or species) were incorporated in the database as needed to obtain contrasting pairs of states for the characters considered in the study. This strategy had been used previously in character optimization studies (Ronse De Craene et al., 2003; Sargent, 2004).
The final composite phylogeny contained 163 taxa and was assembled using the Mesquite 2·01 software (Maddison, 2000). Since the individual phylogenies were inferred from various morphological and molecular characters, branch lengths were given a default value of 1. Families for which the phylogenetic position remains unclear, namely Icacinaceae, Oncothecaceae, Boraginaceae and Vahliaceae (belonging to Untitled Group 1, see Data Set in Supplementary Information, available online) were added to the composite tree by creating a polytomy with their sister taxa. Icacinaceae and Oncothecaceae were linked to Garryales, and Boraginaceae and Vahliaceae to Lamiales, Gentianales and Solanales. Bruniaceae, Columelliaceae, Eremosynaceae, Paracryphiaceae, Sphenostemonaceae, Escalloniaceae and Polyosmaceae (Untitled Group 2) were grafted to their sister taxa Apiales, Asterales and Dipsacales. Other polytomies were introduced when phylogenetic links between taxa were unresolved.
For the composite tree to be read by the BayesDiscrete method implemented in BayesTraits (Pagel et al., 2004; Barker and Pagel, 2005; Pagel and Meade, 2006), available from, polytomies had to be resolved and were given the length of 10–5, which was considered as negligible in comparison with the default length of one for all other branches. Algorithms that allow for resolution of polytomies exist (Maddison and Maddison, 1989) that examine all possible resolutions including the least parsimonious ones (Maddison, 1989). In the case of the Asteridae, many polytomies occur at deep nodes in the tree, resulting in a very large potential number of resolved trees. Phylogenetic comparative analyses were performed on a set of resolved trees and it was found that random resolution of polytomies did not influence the results of the statistical tests. Therefore, polytomies were manually resolved in the most-parsimonious way according to character evolution. In all the cases where a polytomy occurred and housed a change in states, only one character of the four considered was concerned by the change. This is why a single resolution was enough to reconcile the most-parsimonious evolution for all of the four characters for each polytomy.
Coding the characters and inferring ancestral states
Table 1 lists the four floral traits considered in this study, together with the states defined for each character. Character states were principally compiled from Cronquist (1981), Bremer et al. (2001) and Kubitzki (2004) and from the electronic databases Angiosperm Phylogeny Website (Stevens, 2001 onwards), Delta (Watson and Dallwitz, 1992) and eFloras ( In this study, flower symmetry was considered at the perianth (calyx + corolla) level only. The primary objective was to test the potential correlations between perianth symmetry and other floral traits supposedly linked to it, like perianth merism, the intra-whorl differentiation of perianth (due to sepals or petals bearing spurs) and androecium organ number. Also, it was assumed that in Asteridae, perianth zygomorphy was more appealing to pollinators than androecium zygomorphy (Stebbins, 1974). A binary coding for androecium organ number was adopted. A flower with ten or less stamens (ten corresponding to twice the ancestral Asteridae perianth merism; see Results), was considered oligandrous. When the number of stamens exceeded twice the ancestral perianth merism, the flower was considered polyandrous. The multistate coding that was used for the perianth merism corresponded exactly with the observed character states. For the comparative analyses, however, multistate characters could not be used and therefore a binary coding system had to be established as input values for BayesDiscrete. For this purpose, perianth merism was coded as variable (state 0) or fixed (state 1).
Table 1.
Table 1.
List of the characters and character states used in this study
The number of spurs was coded as state 0 if absent or equal in number to perianth merism, and as state 1 if inferior in number to perianth merism, in conformity with the absence/presence of an intra-whorl differentiation of petals and/or sepals. When the number of spurs was inferior to perianth merism, it was equal to one, except for the genus Diascia (Scrophulariaceae), which has flowers characterized by two lateral spurs. Binary codings were thus defined for perianth symmetry, androecium organ number, perianth merism and the number of spurs. In addition to requiring binary character states, the BayesDiscrete method does not accept polymorphisms in terminal taxa. To remove polymorphisms, it was decided to split each of the polymorphic branches into two branches with a length of 10–5. Each newly created branch referred to the same species but had a single state of the character (this is indicated by notations 1 and 2 in the Data Set in the Supplementary Information, available online; it was used for instance in the Asteraceae, which is characterized by inflorescences that may bear both zygomorphic and actinomorphic flowers). Phylogenetic trees were represented as cladograms, with terminal taxa coloured according to order. Branches were coloured according to the ancestral state (see Fig. 1) inferred by maximum parsimony using Mesquite 2·01.
Fig. 1.
Fig. 1.
Composite tree of the Asteridae where the evolution of perianth merism is optimized using the maximum parsimony method. Branches are coloured according to the observed (terminal branches) and inferred (internal branches) character state. Yellow, Trimery/hexamery; (more ...)
Detecting a phylogenetic signal in the characters and optimizing the characters
Before inferring the ancestral states in the composite tree, the presence of a phylogenetic signal in the distribution of the character states in the terminal taxa was tested. To test the phylogenetic signal, trees were randomly generated (10 000 simulations) from a set of 163 taxa, allowing an empirical distribution of the number of steps required to explain the distribution of character states (length of the tree) to be obtained. The original tree length was then compared with the empirical distribution: if <5 % of the randomly generated trees are shorter than the original tree, a phylogenetic signal was considered to be present for the character. The simulations were run under the TreeFarm package of Mesquite 2·01. The uniform speciation algorithm (Yule model) was used to simulate the trees, with a tree depth fixed by default at 10. The evolution of each character (with both types of coding for perianth merism and the number of spurs) was optimized on the composite tree using the maximum parsimony method implemented in Mesquite 2·01.
Running the comparative analyses
Comparative analyses were performed using two different methods: maximum likelihood and pairwise comparisons. The first one is implemented in the BayesDiscrete method of BayesTraits and uses a likelihood ratio (LR) test for correlated evolution between pairs of binary traits. It calculates the likelihood of two models applied to the data. The first model allows the traits to evolve independently, whereas the second assumes correlated evolution between the two traits (omnibus test). Evidence for correlated evolution is found when the dependent model fits the data better than the independent one. This is evaluated by the LR statistic that compares the goodness-of-fit of the dependent model (H1) to the data with that of the independent model (H0). The LR statistic is defined as follows: LR = – 2 loge [L(H0)/L(H1)] and is asymptotically distributed as a chi-square variable with degrees of freedom equal to the difference in the number of parameters between the two models. The four parameters of the null model are the independent transition rates between two binary traits (from state ‘0’ to state ‘1’, and conversely). The model of correlated or dependent trait evolution considers the four possible pairs that two binary characters can form: 1 = (0,0); 2 = (0,1); 3 = (1,0) and 4 = (1,1). Therefore eight transition rates qij [i, j Є(1, 2, 3, 4)] have to be estimated assuming that simultaneous changes in the two characters are not likely to occur. In the case of the omnibus test the LR must be compared with χ2 with d.f. = 4 (χ24 = 9·49 with a confidence threshold of 5 %).
BayesDiscrete also tests for temporal ordering and directional trait evolution (Contingent Change Test) of the two characters. The aim of the temporal order test is to identify the first character to change starting from a pair of ancestral characters. The contingency test detects if, when a pair of traits evolve in a correlated fashion, the acquisition of the second state in a character is more probable when the other character has already acquired its second state. To perform these two tests, evolutionary models where one transition rate is constrained to the value of another one are compared with the model with eight parameters (Pagel, 1994, 1997). The LR is then compared with χ2 with d.f. = 1 (χ21 = 3·84 with a confidence threshold of 5 %).
These tests can be summarized in a flow diagram of evolutionary changes. It is a schematic representation of the evolutionary sequence between the ancestral and the derived states of two characters (Rolland et al., 1998). One could then determine which character changes before the other or, in other words, which transition is more likely knowing the ancestral combination of characters.
Symmetry was considered as the ‘dependent’ character in the tests for correlation between ‘perianth merism and symmetry’, ‘androecium organ number and symmetry’ and ‘number of spurs and symmetry’ because it was assumed that, during the early development of the flower, symmetry can be at least partly influenced by perianth merism, the number of stamens and the number of spurs. Alternatively the number of spurs can be considered as a consequence of the evolution of floral symmetry for two reasons: (1) all transitions towards single spurs or pairs of spurs occur in zygomorphic clades; and (2) no transition towards zygomorphy was found in actinomorphic + spurred clades. Flower symmetry was also considered as the independent character for the test of correlated evolution between ‘the number of spurs and symmetry’.
The algorithm used in BayesDiscrete infers the ancestral states at each node of the tree by a Markov process at each step of the simulations. The parameters used were those implemented by default in BayesDiscrete except for the scaling, which was added to take into account branch length.
The second method (Maddison, 2000) is implemented in the software Mesquite 2·01. Ancestral states are inferred at each node of the tree by maximum parsimony and phylogenetic pairwise comparisons are made (100 000 repetitions). Pairs contrasting in two binary characters are selected for the comparison (Read, 1995). This test determines if a contrast in state ‘A’ vs. state ‘a’ predicts a contrast in state ‘B’ vs. state ‘b’. The lineages between selected pairs are not shared; the algorithm cannot therefore consider all the lineages passing through a polytomy. Consequently, pairwise comparisons were run on the completely resolved composite tree. It should be noted that the pairs selected for pairwise comparison are not necessarily statistically independent even though they are phylogenetically independent (Maddison, 2000).
Detecting a phylogenetic signal in the data
According to Laurin (2004), if the observed distribution does not significantly differ from a random one, then little confidence should be placed in the inference of ancestral states using parsimony, since state optimization might just as well be randomly inferred. The presence of a phylogenetic signal for a given character reflects its systematic utility (lack of homoplasy) and therefore gives more value to the inference of ancestral states. A phylogenetic signal was found in the three following characters: perianth merism 2, perianth symmetry and androecium organ number (P < 0·005). No phylogenetic signal was detected in the evolution of the number of spurs (P = 1·0) or in perianth merism 1 (P = 0·33). The absence of a signal for the number of spurs is possibly due to the scarcity of transitions towards the presence of spurs. In contrast, there were many transitions among states for perianth merism 1, and >5 % of the simulated trees were shorter than the composite tree. However, the optimization of perianth merism 1 on the tree was only used descriptively, and this character was not used in the statistical tests as a consequence of its multistate coding.
Inference of ancestral states and independent evolution of characters
The inference of ancestral character states using maximum parsimony indicates that the hypothetical ancestral flower of Asteridae is pentamerous, actinomorphic and oligandrous. Tetramery evolved at least once in each order of the Asteridae and is the second-most frequent type of perianth merism after pentamery. Variable perianth merism evolved seven times independently (and each time from a pentamerous perianth merism; see Fig. 1) which is surprisingly frequent for a clade where closed ground plans are supposed to be the basic floral Bauplan.
Zygomorphy evolved 15 times independently (see Fig. 2A), each time in the context of a fixed perianth merism as shown from the comparison with the tree displayed on Fig. 1. There were ten reversals to actinomorphy (see Fig. 2A). Zygomorphic families were especially abundant in the orders Lamiales and Dipsacales, where all the reversals to actinomorphy were found. The families and genera that showed a reversal to actinomorphy were Patrinia (Valerianaceae), Byblidaceae, Oleaceae, Peltanthera (Gesneriaceae), Plantago (Plantaginaceae), Nuxia (Stilbaceae), Anisacanthus, Sanchezia (Acanthaceae), Nashia and Petrea (Verbenaceae). It is noticeable that almost all zygomorphic flowers in this study are pentamerous. Zygomorphy was found in tetramerous taxa only when these tetramerous taxa were derived within a zygomorphic lineage, as revealed by the comparison between Figs 1 and and22A.
Fig. 2.
Fig. 2.
Mirror trees of the Asteridae showing the evolution of perianth symmetry and androecium organ number optimized using maximum parsimony. (A) Evolution of perianth symmetry. Black, Actinomorphy; red, zygomorphy. (B) Evolution of androecium organ number. (more ...)
Polyandry evolved three times in the Cornales and eight times in the Ericales, the two basalmost clades of Asteridae, yet only two times in the rest of the clade (see Endress, 2002). Overall, oligandry was associated with zygomorphy (see Fig. 2A, B), with two exceptions, namely the genera Couroupita (Lecythidaceae) and Tupidanthus (Araliaceae), where zygomorphy was acquired in a polyandrous context.
Multiple spurs (as many spurs as perianth merism) evolved only twice in the whole Asteridae clade (see Fig. 3B), once within the genus Utleya (Ericaceae, Ericales) and once in Halenia (Gentianaceae, Gentianales). Both genera have actinomorphic flowers. Conversely, flowers with one or two spurs evolved seven times independently. Such flowers were zygomorphic (see Fig. 3A, B). The acquisition of zygomorphy always preceded that of the single spur except in Balsaminaceae (Ericales) where both transitions occurred simultaneously in the tree shown in Fig. 3.
Fig. 3.
Fig. 3.
Mirror trees of the Asteridae showing the evolution of perianth symmetry and the number of spurs optimized using the maximum parsimony method. (A) Evolution of perianth symmetry. Black, Actinomorphy; red, zygomorphy. (B) Evolution of the number of spurs. (more ...)
Comparative analyses
The LR tests indicate that androecium organ number and the number of spurs are both correlated with perianth symmetry (LR = 15·84, P < 0·05; LR = 19·52, P < 0·001). Table 2 summarizes the tests of character correlation conducted in this study as well as the tests performed to assess the transition rate values. The flow diagram of correlated evolution between symmetry and androecium organ number established on the basis of the temporal order test and the contingency test (see Fig. 4) demonstrates that transitions towards a zygomorphic and polyandrous flower are scarce (q24 and q34 are not significantly different from zero, LRs = 0·56, 2·42; P = 0·454, 0·120, respectively). The transition rates indicated on the diagram, namely q12, q21, q13, q31, q43, are significantly different from zero (LRs = 5·72, 5·66, 8·26, 34·42, 4·87, P < 0·025, P < 0·0025, P < 0·005, P < 0·001, P < 0·05, respectively). The acquisition of zygomorphy or polyandry starting from an ancestral actinomorphic and oligandrous flower is equally probable (q12 is not significantly different from q13, LR = 3·34, P = 0·067). Reversals to the ancestral state starting from an actinomorphic and polyandrous flower are not more likely than reversals from a zygomorphic and oligandrous one (q31 is not significantly different from q21, LR = 3·44, P = 0·064). Because q12 was not significantly different from q13, it was not possible to detect any temporal order in the acquisition of zygomorphy vs. polyandry. Furthermore, changes in perianth symmetry from actinomorphy to zygomorphy and the reverse are equally probable (q21 is not significantly different from q12, LR = 1·80, P = 0·180). It is interesting to note that q31 is significantly different from q13 (LR = 25·51, P < 0·001), indicating that in an actinomorphic background, transitions from polyandry to oligandry are more likely than transitions from oligandry to polyandry. q43 is significantly superior to q21 (LR = 4·60; P < 0·05) suggesting that the reversal to actinomorphy is more likely when the flower is polyandrous than oligandrous.
Table 2.
Table 2.
Tests of character correlation and transition rates values under BayesDiscrete
Fig. 4.
Fig. 4.
Flow diagram summarizing the transitions between the different states of androecium organ number and perianth symmetry. Parameter qij is the transition rate between state ‘i’ in androecium organ number and state ‘j’ in (more ...)
Similar tests were performed to examine the relative timing of evolution between perianth symmetry (considered as the independent state) and the number of spurs. The only transition rates that were significantly different from zero were those linking a zygomorphic flower without spurs [state (1,0)] and a zygomorphic flower with a single spur or two spurs [state (1,1)] (LR = 17·30, P < 0·001) and conversely (LR = 6·12, P < 0·025). In a zygomorphic context, the acquisition of one or two spurs occurs less frequently than reversals towards the absence of spurs (LR = 5·46, P < 0·025). Since the transition rates between the states (0, 0) and (1, 0); (0, 0) and (0, 1) were not significantly different from zero, the precedence of zygomorphy and the emergence of spurs could not be traced statistically.
According to the pairwise comparisons (Mesquite 2·01), the evolution of symmetry is correlated with androecium organ number, the number of spurs and perianth merism 2 (P < 0·01). Zygomorphy is strictly associated with oligandry and a fixed perianth merism, and a single nectar spur or a pair of oil spurs are only found in zygomorphic flowers.
Powers and pitfalls of the comparative analysis
Phylogenetic comparative methods (PCMs) are statistical tools that were originally developed by evolutionary ecologists for testing hypotheses about adaptation. This is done by looking at the correlated evolution between morphological and ecological characters in a set of species, taking into account the phylogenetic relationships of these species (Harvey and Pagel, 1991; Rolland et al., 1998; Cézilly et al., 2000; Chapman et al., 2006; Summers et al., 2007). This is a means to distinguish correlation due to adaptation from historical correlation. PCMs are also relevant to the analysis of correlations between morphological characters (Cubo and Arthur, 2001), as a method of detecting developmental constraints, i.e. factors limiting the potential range of forms into which an organism can grow (Rudall and Bateman, 2003), which is the focus of the present study. Phylogenetic frameworks can be used to (a) suggest the direction and likelihood of character changes, (b) understand better ontogenic sequences, and (c) select the appropriate taxa for detailed developmental studies (Kramer et al., 1998; Kramer and Irish, 1999; Donoghue and Ree, 2000).
The phylogenetic tree used for the Asteridae in the present study, while well supported in its overall architecture, still contains poorly resolved regions. Moreover, since it is a composite tree produced by the combination of several phylogenies inferred from various data types, all branches, except those resulting from the resolution of polytomies or polymorphisms, had a default length of 1. New branches formed to resolve polytomies and polymorphisms had a default length of 10–5. Other lengths were tested for resolving polytomies and polymorphisms (10–2 and 1) without significantly changing the results. Thus, even though transition rates are inflated in these shortened branches, the length of newly created branches was kept at 10–5 to reflect as accurately as possible the polytomous state of the branching. Since the distribution of the states for perianth symmetry, androecium organ number and perianth merism was strongly associated with the composite tree, it was possible to infer ancestral states for these characters using parsimony. However, estimating ancestral states using likelihood on the basis of this kind of tree with equal branch lengths remains questionable (for discussion, see Sannier et al., 2007).
The opportunity for scoring evolutionary transitions in a given character can be influenced by taxon sampling and assumptions of branch length (Ree and Donoghue, 1999). In the composite tree of the Asteridae in the present study, zygomorphy and actinomorphy were not equally distributed and this could potentially introduce a bias in estimating transition rates between the two morphs (Thorne, 1992; Ree and Donoghue, 1999). To reflect as accurately as possible the relative abundance of zygomorphy in Asteridae, zygomorphic and actinomorphic species should have been represented in similar proportion (Thorne, 1992). Since the goal of the present study was to reveal a possible impact of several architectural traits on the evolution of symmetry, we deliberately chose to focus on families and genera that presented variation for the characters considered. This allowed us to account for the actual number of transitions as precisely as possible without using a comprehensive tree of the Asteridae clade. Actinomorphic taxa presented more variation in the three other characters studied than zygomorphic taxa, and are therefore over-represented in our tree. Equilibrating the proportion of actinomorphic and zygomorphic taxa would have introduced a bias in the representation of the other traits. Consequently, no attempt was made to correct the proportion of the different states in symmetry. The taxa sampled were selected so that the ancestral states inferred by the maximum parsimony method would not be altered by pruning unvariable taxa off the clade.
The evolution of floral symmetry
The present results show that within Asteridae at least 15 independent transitions from actinomorphy to zygomorphy occurred and ten reversals to actinomorphy. This result is consistent with a previous study showing that the most-parsimonious scenario for the evolution of symmetry in Asteridae assumes at least eight independent origins for zygomorphy and at least nine reversals to actinomorphy (Donoghue et al., 1998). Their study, as well as the present one, was carried out with equally weighted transitions between actinomorphy and zygomorphy. However, it has been demonstrated that inferring the most-parsimonious ancestral states is contingent upon a particular set of transformation costs (Ree and Donoghue, 1998). For example, for zygomorphy to be the most-parsimonious ancestral state for the Asteridae the cost of gains must be weighted more than three times the cost of losses (Ree and Donoghue, 1999). Molecular studies on developmental genetics in model species support the view that zygomorphy is controlled by a complex network of transcription factors and target genes (Corley et al., 2005; Costa et al., 2005), which suggests that the loss of this character state might be easier than its acquisition and supports an unequal weighting of transitions. However, it is also modulated and constrained by interactions with other architectural floral traits (see below) and by the ability of pollinators to exploit the new morphs in order to give rise to an interspecific differentiation. Thus, due to the lack of information to compare the transitions between actinomorphy and zygomorphy, they were assigned the same cost in the present study.
Our approach suggests that perianth symmetry does not evolve independently from the other architectural traits. Our tentative scenario (Fig. 5) summarizes all the possible changes affecting the ancestral flower of Asteridae, which is pentamerous, actinomorphic, oligandrous and spurless. It shows that the reversal to actinomorphy only occurred in oligandrous, spurless and pentamerous clades. Possible evolutionary trends revealed by this scenario are discussed below.
Fig. 5.
Fig. 5.
Proposed scenario for the evolution of perianth symmetry and related characters in Asteridae. The characters are, from top to bottom, perianth symmetry, androecium organ number, the number of spurs, perianth merism. New acquisitions (derived states) are (more ...)
Correlated evolution between perianth symmetry and perianth merism
The Asteridae clade is a good candidate for studying the relationship between perianth symmetry and merism, due in particular to the high number of independent evolutions of variable perianth merism in this clade. Indeed, there are eight independent transitions towards variable perianth merism from an ancestral fixed state, which is rather surprising in derived angiosperms. This phenomenon can be interpreted as a reversal. The present results show that in Asteridae, zygomorphy always appears in the context of fixed perianth merism, suggesting that the evolution towards zygomorphy is only possible in the context of a fixed floral ground plan. Furthermore, the correlated evolution between perianth symmetry and perianth merism indicates that zygomorphy mostly appears in pentamerous taxa. Zygomorphy is present in tetramerous taxa only when derived within a pentamerous + zygomorphic lineage. It can be hypothesized that these tetramerous taxa originate from pentamerous ones through a fusion of two petals (Donoghue et al., 1998). The only case where zygomorphy was associated with hexamery was in the genus Couroupita (Lecythidaceae; Tsou and Mori, 2007).
Correlated evolution between perianth symmetry and the number of spurs
As suggested by the comparison of the mirror trees on Fig. 3, spurs with rewards (nectar or oil) are not restricted to zygomorphic flowers in Asteridae. Single nectar spurs or pair of oil spurs are strictly contingent to zygomorphic flowers, since among the seven taxa presenting spurred zygomorphic flowers, six have a single nectar spur and one has a pair of oil spurs. This is logical since a flower that has one or two spurs is de facto zygomorphic. The genus Diascia is the only case recorded for a flower with a pair of spurs. This genus has pentamerous zygomorphic flowers with two spurs originating from the lateral petals, which is linked to a particular pollination syndrome (Endress, 1994). These are oil spurs and not nectar spurs. Bees collect the oil with their forelegs, and not with the proboscis (Vogel, 1974). Thus the structures that produce floral oil tend to be paired. Even though nectar and oil spurs are not homologous, they were both considered to be morphological differentiations of the corolla. Flowers that have as many spurs as corolla merism evolved only twice in the Asteridae, in the tetramerous genus Halenia (Gentianaceae) and the pentamerous genus Utleya (Ericaceae). In both cases the flowers are actinomorphic. It is clear from Fig. 3 that zygomorphy evolves before the emergence of spurs and that these two acquisitions are not simultaneous. It is therefore tempting to suggest that floral symmetry constrains the emergence of spurs. The genetic determinants of spurs are not hitherto elucidated, but mutations in KNOX homeobox genes in A. majus (Golz et al., 2002) have been shown to affect corolla development and the elaboration of an invaginated spur-like structure.
Correlated evolution between perianth symmetry and androecium organ number
In the Asteridae, the evolution of perianth symmetry is correlated with androecium organ number. This is not the case for Ranunculales, where shifts in perianth symmetry are not necessarily associated with changes in stamen number (Damerval and Nadot, 2007). The statistically significant correlation between perianth zygomorphy and oligandry leads to the suggestion that androecium organ number might be a developmental or spatial constraint exerted on floral symmetry. A high number of stamens may possibly prevent the spatially or temporally heterogeneous development of organs, which is a prerequisite for zygomorphy. This could explain why transitions towards ‘zygomorphy + polyandry’ were recorded so scarcely (see Fig. 4). The two genera recorded in this study as ‘zygomorphic + polyandrous’ were Tupidanthus in Araliaceae and Couroupita in Lecythidaceae. The perianth of Tupidanthus is composed of a calyptra and a calyx reduced to a rim (Sokoloff et al., 2007) and Couroupita is characterized by an androecium held by an elongated structure of the flower (Tsou and Mori, 2007) and therefore placed well above the perianth. These two particular floral morphologies are highly specialized and cannot be interpreted as easily in terms of physical constraint of androecium on perianth symmetry. In addition, it is shown that the ancestral state ‘actinomorphy + oligandry’ is the most stable state (see Fig. 6), evolutionarily easy to reach and hard to lose (see the transition rates in Fig. 4, the differences between q21 and q12 as well as between q31 and q13 being positive). We suggest that the acquisition of polyandry in actinomorphic flowers has a non-negligible morphogenetic cost but may be beneficial in the presence of adapted/suitable pollinators (trade-off), and could therefore be considered like zygomorphy, as a key innovation. For example, polyandry is often observed in beetle-pollinated systems that are derived rather than ancestral in angiosperms (Bernhardt, 2000). Another strategy, namely ‘zygomorphy + oligandry’, is also an unstable combination of symmetry and androecium states. Although zygomorphy can be selected for the advantage it confers to flowers in optimizing the physical interaction between the flower and its pollinators, this trait could potentially be easy to lose through single mutations in the gene network determining floral symmetry (see below). However, this is not the only possible mechanism of reversion to actinomorphy. Citerne et al. (2006) showed that in the legume Cadia the reversal to actinomorphy could be a homeotic transformation due to heterotopic change in gene expression.
Fig. 6.
Fig. 6.
Proposed stability diagram of the ‘perianth symmetry + androecium organ number’ states. The ‘actinomorphic + oligandrous’ (A, <) state is the most stable (square), whereas the ‘zygomorphic + polyandrous’, (more ...)
When comparing the two lower unstable states on the scale of morphogenetic cost, it appears that it is much easier to revert to ‘actinomorphy + oligandry’ from polyandry than from zygomorphy (the value of q31 is more than twice that of q21).
The last combination of states, ‘zygomorphy + polyandry’, is the most unstable. Transitions to this combination of traits are very rare.
As discussed previously, the proportion of zygomorphic taxa is underestimated in the present tree because these taxa appear devoid of evolutionary transitions in stamen number, perianth merism and presence of spurs. This may suggest that zygomorphy promotes high speciation rates, but once it is established, it is less amenable to subsequent evolutionary changes than is actinomorphy, maybe because of closer interdependency with specialized pollinators and/or complexity of genetic networks. At first, this could seem contradictory with the fact that ‘zygomorphy + oligandry’ appears quite unstable compared with ‘actinomorphy + oligandry’. These statements can be reconciled if it is considered that ‘zygomorphy + oligandry’ corresponds to a canalized development, probably maintained by additional forces, like selective pressures operating on the pollination syndrome. The cohort of pollinators could be more affected by a change in perianth shape than by a reduction in the androecium organ number.
Interestingly, in the present study all reversals from zygomorphy to actinomorphy in an oligandrous context occur in taxa where the number of stamens is equal to perianth merism and there are no staminodes, except for the genus Peltanthera (Gesneriaceae, Lamiales). In this particular case, we suggest that the staminode exerts the same spatial constraint on symmetry as a fully developed stamen would do. A study on the monocots giving particular attention to the presence of staminodes in relation with floral symmetry showed that the differential development of one or two stamens emphasizes one plane of symmetry over the two other planes, rendering the stamen whorl zygomorphic (Rudall and Bateman, 2004). It would be interesting to examine androecium structure in the Asteridae further, especially to describe the relationship between perianth and androecium symmetry, and androecium organ number and symmetry, in order to determine any contingency between these traits.
Remarkable progress has been achieved in understanding the molecular bases underlying floral symmetry in the model species Antirrhinum majus (Asteridae) and Lotus japonicus (Rosidae), in which zygomorphy affects both corolla and androecium. Antirrhinum majus has pentamerous tubular bilabial flowers, with one of the stamens reduced to a staminode. Petal identity and abortion of the dorsal stamen are controlled by the interplay of four genes belonging to two families of transcription factors (Corley et al., 2005). CYCLOIDEA is the keystone of this network (Luo et al., 1995, 1999; Reeves and Olmstead, 2003; Corley et al., 2005); it retards the growth of the dorsal domain and inhibits the development of the dorsal stamen. LjCyc2, a CYC paralogue plays a similar role in L. japonicus (Feng et al., 2006). Extension or restriction of the expression domain of CYC paralogues has been shown to play a role in perianth symmetry and androecium development in the two Antirrhineae Linaria vulgaris, Mohavea confertiflora and the Fabaceae Cadia purpurea (Cubas et al., 1999; Hileman et al., 2003; Citerne et al., 2006). The correlated evolution of perianth symmetry and androecium organ number in these species may reflect either ancestral or parallel adaptation in relation to the pollination syndrome, relying on a regulatory network promoting zygomorphy at the perianth and androecium levels.
The present analyses on Asteridae reveal that polyandry and variable perianth merism largely prevent perianth symmetry from evolving towards zygomorphy and point out that the acquisition of zygomorphy precedes the emergence of single or paired spurs. Interestingly, the evolution of polyandry in an actinomorphic context may appear as an alternative pathway to zygomorphy in Asteridae in terms of key innovations. This type of statistical analysis on morphological data in a phylogenetic context will be useful in evo-devo studies trying to unravel the genetic bases of symmetry in eudicots and can potentially be extended to all flowering plants.
Supplementary Information is available online at and includes details of the phylogenetic trees assembled to build the composite tree of Asteridae used in this study, and a data set of taxa represented in the composite phylogenetic tree with associated character states.
We thank Julie Sannier (ESE Lab., UMR 8079, Orsay, France) for helpful discussions on phylogenetic comparative methods, and Julie Dawson (Winter Wheat Breeding and Genetics, Department of Crop and Soil Sciences, Washington State University, Pullman), for comments on the manuscript. We also thank two anonymous reviewers for their constructive comments on the first version of the manuscript. This work was supported by the IFR 87 ‘La Plante et son Environnement’, through a 2007 grant, and the Agence National de la Recherche, through an ANR-07-BLAN-0112 grant. The first author was supported by a fellowship from the Ministère de l'Enseignement Supérieur et de la Recherche.
  • APG. An update of the Angiosperm Phylogeny Group classification for the orders and families of flowering plants: APGII. Botanical Journal of the Linnean Society. 2003;141:399–436.
  • Barker D, Pagel M. Predicting functional gene links using phylogenetic-statistical analysis of whole genomes. PLOS Computational Biology. 2005;1:24–31.
  • Bateman RM, Hilton J, Rudall PJ. Morphological and molecular phylogenetic context of the angiosperms: contrasting the ‘top-down’ and ‘bottom-up’ approaches used to infer the likely characteristics of the first flowers. Journal of Experimental Botany. 2006;57:3471–3503. [PubMed]
  • Bernhardt P. Convergent evolution and adaptive radiation of beetle-pollinated angiosperms. Plant Systematics and Evolution. 2000;222:293–320.
  • Bremer K, Backlund A, Sennblad B, Swenson U, Andreasen K, Hjertson M, et al. A phylogenetic analysis of 100+ genera and 50+ families of euasterids based on morphological and molecular data with notes on possible higher level morphological synapomorphies. Plant Systematics and Evolution. 2001;229:137–169.
  • Cézilly F, Dubois F, Pagel M. Is mate fidelity related to site fidelity? A comparative analysis in Ciconiiforms. Animal Behaviour. 2000;59:1143–1152. [PubMed]
  • Chapman EG, Foote BA, Malukiewicz J, Roeh WR. Parallel evolution of larval morphology and habitat in the snail-killing fly genus Tetanocera. Journal of Evolutionary Biology. 2006;19:1459–1474. [PubMed]
  • Citerne HL, Pennington RT, Cronk QCB. An apparent reversal in floral symmetry in the legume Cadia is a homeotic transformation. Proceedings of the National Academy of Sciences of the USA. 2006;103:12017–12020. [PubMed]
  • Corley SB, Carpenter R, Copsey L, Coen E. Floral asymmetry involves an interplay between TCP and MYB transcription factors in Antirrhinum. Proceedings of the National Academy of Sciences of the USA. 2005;102:5068–5073. [PubMed]
  • Costa MMR, Fox S, Hanna AI, Baxter C, Coen E. Evolution of regulatory interactions controlling floral asymmetry. Development. 2005;132:5093–5101. [PubMed]
  • Cronquist A. An integrated system of classification of flowering plants. New York, NY: Columbia University Press; 1981.
  • Cubas P, Vincent C, Coen E. An epigenetic mutation responsible for natural variation in floral symmetry. Nature. 1999;401:157–161. [PubMed]
  • Cubo J, Arthur W. Patterns of correlated character evolution in flightless birds: a phylogenetical approach. Evolutionary Ecology. 2001;14:693–703.
  • Dahlgren R, Rasmussen FN, Yeo PF. The families of the monocotyledons: structure, function and taxonomy. Berlin: Springer; 1985.
  • Damerval C, Nadot S. Evolution of perianth and stamen characteristics with respect to floral symmetry in Ranunculales. Annals of Botany. 2007;100:631–640. [PMC free article] [PubMed]
  • Donoghue MJ, Ree RH. Homoplasy and developmental constraint: a model and an example from plants. American Zoologist. 2000;40:759–769.
  • Donoghue MJ, Ree RH, Baum DA. Phylogeny and the evolution of flower symmetry in the Asteridae. Trends in Plant Science. 1998;3:311–317.
  • Endress PK. Diversity and evolutionary biology of tropical flowers. Cambridge: Cambridge University Press; 1994.
  • Endress PK. Symmetry in flowers: diversity and evolution. International Journal of Plant Sciences. 1999;160(Suppl. 6):S3–S23. [PubMed]
  • Endress PK. Origins of flower morphology. Journal of Experimental Zoology. 2001;291:105–115. [PubMed]
  • Endress PK. Morphology and angiosperm systematics in the molecular era. Botanical Review. 2002;68:545–570.
  • Endress PK. Angiosperm floral evolution: morphological developmental framework. Advances in Botanical Research. 2006;44:1–61.
  • Feng X, Zhao Z, Tian Z, Xu S, Luo Y, Cai Z, et al. Control of petal shape and floral zygomorphy in Lotus japonicus. Proceedings of the National Academy of Sciences of the USA. 2006;103:4970–4975. [PubMed]
  • Golz JF, Keck EJ, Hudson A. Spontaneous mutations in KNOX genes give rise to a novel floral structure in Antirrhinum. Current Biology. 2002;12:512–522. [PubMed]
  • Harvey PH, Pagel MD. The comparative method in evolutionary biology. Oxford: Oxford University Press; 1991.
  • Hileman LC, Kramer EM, Baum DA. Differential regulation of symmetry genes and the evolution of floral morphologies. Proceedings of the National Academy of Sciences of the USA. 2003;100:12814–12819. [PubMed]
  • Irish VF. Variations on a theme: flower development and evolution. Genome Biology. 2000;1 Reviews 1015·1–1015·4. [PMC free article] [PubMed]
  • Kramer EM, Irish VF. Evolution of genetic mechanisms controlling petal development. Nature. 1999;399:144–148. [PubMed]
  • Kramer EM, Dorit RL, Irish VF. Molecular evolution of genes controlling petal and stamen development: duplication and divergence within the APETALA3 and PISTILLATA MADS-box gene lineages. Genetics. 1998;149:765–783. [PubMed]
  • Kubitzki K. The families and genera of vascular plants. Vol. 6. Berlin: Springer; 2004.
  • Laurin M. The evolution of body size, Cope's rule and the origin of Amniotes. Systematic Biology. 2004;53:594–622. [PubMed]
  • Luo D, Carpenter R, Vincent C, Copsey L, Coen E. Origin of floral asymmetry in Antirrhinum. Nature. 1995;383:794–799. [PubMed]
  • Luo D, Carpenter R, Copsey L, Vincent C, Clark J, Coen E. Control of organ asymmetry in flowers of Antirrhinum. Cell. 1999;99:367–376. [PubMed]
  • Maddison WP. Reconstructing character evolution on polytomous cladograms. Cladistics. 1989;5:365–377.
  • Maddison WP. Testing character correlation using pairwise comparisons on a phylogeny. Journal of Theoretical Biology. 2000;202:195–204. [PubMed]
  • Maddison WP, Maddison DR. Interactive analysis of phylogeny and character evolution using the computer program MacClade. Folia Primatologica. 1989;53:190–202. [PubMed]
  • Pagel M. Detecting correlated evolution on phylogenies: a general method for the comparative analysis of discrete characters. Proceedings of the Royal Society of London Series B – Biological Sciences. 1994;255:37–45.
  • Pagel M. Inferring evolutionary processes from phylogenies. Zoologica Scripta. 1997;26:331–348.
  • Pagel M, Meade A. Bayesian analysis of correlated evolution of discrete characters by reversible-jump Markov chain Monte Carlo. The American Naturalist. 2006;167:808–825. [PubMed]
  • Pagel M, Meade A, Barker D. Bayesian estimation of ancestral character states on phylogenies. Systematic Biology. 2004;53:673–684. [PubMed]
  • Read AF. Inference from binary comparative data. Journal of Theoretical Biology. 1995;173:99–108.
  • Ree RH, Donoghue MJ. Step matrices and the interpretation of homoplasy. Systematic Biology. 1998;47:582–588. [PubMed]
  • Ree RH, Donoghue MJ. Inferring rates of change in flower symmetry in Asterid angiosperms. Systematic Biology. 1999;48:633–641.
  • Reeves PA, Olmstead RG. Evolution of the TCP gene family in Asteridae: cladistic and network approaches to understanding regulatory gene family diversification and its impact on morphological evolution. Molecular Biology and Evolution. 2003;20:1997–2009. [PubMed]
  • Rolland C, Danchin E, de Fraipont M. The evolution of coloniality in birds in relation to food, habitat, predation, and life-history traits: a comparative analysis. The American Naturalist. 1998;151:514–529. [PubMed]
  • Ronse De Craene LP, Soltis PS, Soltis DE. Evolution of floral structures in basal angiosperms. International Journal of Plant Sciences. 2003;164(Suppl. 5):S329–S363.
  • Rudall PJ, Bateman RM. Evolutionary change in flowers and inflorescences: evidence from naturally occurring terata. Trends in Plant Science. 2003;8:76–82. [PubMed]
  • Rudall PJ, Bateman RM. Evolution of zygomorphy in monocot flowers: iterative patterns and developmental constraints. New Phytologist. 2004;162:25–44.
  • Rudall PJ, Bateman RM. Morphological phylogenetic analysis of Pandanales: testing contrasting hypotheses of floral evolution. Systematic Botany. 2006;31:223–238.
  • Sannier J, Asmussen-Lange CB, Harley M, Nadot S. Evolution of microsporogenesis in Palms (Arecaceae) International Journal of Plant Sciences. 2007;168:877–888.
  • Sargent RD. Floral symmetry affects speciation rates in angiosperms. Proceedings of the Royal Society of London Series B – Biological Sciences. 2004;271:603–608. [PMC free article] [PubMed]
  • Sillen-Tullberg B. The effect of biased inclusion of taxa on the correlation between discrete characters in phylogenetic trees. Evolution. 1993;47:1182–1191.
  • Sokoloff DD, Oskolski AA, Remizowa MV, Nuraliev MS. Flower structure and development in Tupidanthus calyptratus (Araliaceae): an extreme case of polymery among asterids. Plant Systematics and Evolution. 2007;268:209–234.
  • Stevens PF. Angiosperm Phylogeny Website. 2001. onwards Version 8.
  • Stebbins GL. Flowering plants – evolution above the species level. London: Edward Arnold; 1974.
  • Summers K, McKeon CS, Heying H, Patrick W. Social and environmental influences on egg size evolution in frogs. Journal of Zoology. 2007;271:225–232.
  • Theissen G, Becker A, Winter K, Munster T, Kirchner C, Saedler H. In: How the land plants learned their floral ABCs: the role of MADS-box genes in the evolutionary origin of flowers. Cronk QCB, Bateman RM, Hawkins JA, editors. London: Taylor and Francis; 2002. pp. 173–205. Developmental genetics and plant evolution.
  • Thorne RF. An updated phylogenetic classification of the flowering plants. Aliso. 1992;13:365–390.
  • Tsou C-H, Mori SA. Floral organogenesis and floral evolution of the Lecythidoideae (Lecythidaceae) American Journal of Botany. 2007;94:716–736. [PubMed]
  • Tucker SC. Floral development in legumes. Plant Physiology. 2003;131:911–926. [PubMed]
  • Vogel S. Ölblumen und ölsamme lnde Bienen. Tropische und Subtropische Pflanzenwelt. 1974;7:1–267.
  • Watson L, Dallwitz MJ. The families of flowering plants: descriptions, illustrations, identification, and information retrieval. 1992. Version 1 June 2007. .
  • Weiblen GD, Oyama RK, Donoghue MJ. Phylogenetic analysis of dioecy in monocotyledons. The American Naturalist. 2000;155:46–58. [PubMed]
Articles from Annals of Botany are provided here courtesy of
Oxford University Press