Portal de Periódicos da CAPES

Was the ancestral angiosperm flower whorled throughout?

2018; Wiley; Volume: 105; Issue: 1 Linguagem: Inglês

10.1002/ajb2.1003

1537-2197

Dmitry D. Sokoloff, Margarita V. Remizowa, Richard M. Bateman, Paula J. Rudall,

Plant and Fungal Species Descriptions

Inferring directions of character transformation during angiosperm evolution is widely appreciated as one of the most important goals of systematic botany, offering the potential to resolve the enigma of the origin of flowering plants. Many theories on the evolution of floral characters that were developed during the 20th century, such as those of Engler (1904), Takhtajan (1991), and Cronquist (1988), were based entirely on expert opinion. The subsequent cladistic approach to morphological character coding and use of topologies of molecular phylogenies as a framework for mapping phenotypic characters opened a more objective era in generating such reconstructions. Most notably in the context of the present study, Doyle and Endress (2000, 2011) and Endress and Doyle (2007, 2009, 2015) investigated floral character evolution in basal1 angiosperms in detail using maximum parsimony. These influential studies provided robust evidence for character states of a common ancestor of all living flowering plants. For example, in contrast with the earlier ideas of Takhtajan (1991) and Cronquist (1988), pronouncedly ascidiate (lacking a postgenitally sealed ventral slit) carpels were revealed as plesiomorphic in angiosperms (see Endress and Igersheim, 2000a), as opposed to carpels in which closure is achieved either partly or exclusively by postgenital fusion of the margins, including the plicate (conduplicate) condition, where a ventral slit proceeds through the entire length of a carpel. However, reconstructions of ancestral states were equivocal for many key characters in parsimony-based optimizations due largely to (1) the high structural diversity of extant lineages of basal angiosperms and (2) the inapplicability of most floral characters to the panoply of potential gymnospermous outgroups, both extant and extinct (e.g., Bateman et al., 2006; Rudall et al., 2011). For example, it remained unclear whether the ancestral angiosperm flower was unisexual or bisexual and whether its floral phyllotaxis was ancestrally spiral or whorled. Indeed, among extant members of the basal angiosperm grade, Amborella is dioecious with uniformly spirally arranged floral organs whereas, for example, Cabombaceae possess bisexual flowers with uniformly whorled organs (Endress, 2001). Figure 1 shows the complex nature of distribution of spiral and whorled patterns among major clades of angiosperms. The recent study by Sauquet et al. (2017), pursued as a key element of the eFLOWER program, represents a potential keystone in investigating the evolution of floral characters in angiosperms. This work is pioneering in many respects and greatly increases the scope of earlier related studies (e.g. O'Meara et al., 2016); it will likely determine avenues of research into evolutionary floral morphology for at least the immediate future. Sauquet et al. (2017) used model-based approaches to infer transformations between states of numerous characters mapped across time-calibrated molecular phylogenetic trees that were generated under no less than 136 fossil constraints. Most importantly, their study was based on a data set of unprecedented size and quality. The data set not only included the now familiar lineages of basal angiosperms, but also benefited from representative coverage of all other more derived clades, which today encompass the vast majority of species-level diversity among flowering plants. In total, 792 species from 98% orders and 86% families of living angiosperms were included (Sauquet et al., 2017). This extensive coverage allowed the authors to infer the ancestral states of 27 floral characters for all key nodes of living flowering plants, including monocots, commelinids, eudicots, asterids, and rosids. Using an innovative PROTEUS database, each of 13,444 individual records of character states in 792 species was documented by a reference to the original data source. In contrast with many earlier studies, the morphological data set of Sauquet et al. (2017) used as its basis individual plant species—thereby mirroring the equivalent molecular trees—rather than employing the widespread approach of character interpolation for all members of a terminal group such as a genus or family. The data set encompasses only information that is clearly documented for a given species. A huge online appendix contains all methodological foundations of evolutionary floral morphology necessary for their ambitious large-scale analyses. Applying model-based approaches to codified traits allowed Sauquet et al. (2017) to resolve ancestral characters despite strong homoplasy among basal angiosperm clades and the inapplicability of outgroup comparison due to the huge phenotypic disparity separating extant angiosperms from any other group of seed plants (e.g., Bateman et al., 2006). Their study did not present an exact set of ancestral character states but rather an array of reconstructions based on different methods, models, and tree topologies. Most importantly, degrees of uncertainty were estimated quantitatively in all reconstructions. The ancestral flower was recognized as most probably bisexual and polysymmetric, consisting of more than two perianth whorls of three separate tepals each, more than two androecial whorls of three separate stamens each, and more than five spirally arranged separate carpels (Sauquet et al., 2017). The phyllotaxis of these organs—particularly that of the tepals—was emphasized in the many media accounts of this much discussed paper (e.g., Gabbott, 2017; Monahan, 2017; Vallejo-Marin, 2017). Sauquet et al. (2017) were careful to explain their scoring of characters in critical taxa, not least when differentiating between whorled and spiral phyllotaxis of floral organs (e.g., scoring the controversial perianth of Ranunculus as spiral). Nevertheless, they admitted that inferring the phyllotaxis of organs in an ancestral flower is a problematic task (see also Endress and Doyle, 2007), and degrees of uncertainty were relatively high in some analyses. For example, in Amborella, the putative sister of all other extant angiosperms (Fig. 1), all the floral organs are arranged along a spiral (Endress and Igersheim, 2000b), yet the choice of ancestrally whorled rather than spiral phyllotaxis proved impossible using maximum parsimony approaches. Earlier traditional accounts such as those of Takhtajan (1991) viewed angiosperm flowers as ancestrally spiral, based primarily on classical studies of floral evolution such as that of Arber and Parkin (1907). However, the study of Sauquet et al. (2017) added robust quantitative support to the growing evidence for a secondary evolutionary gain of spiral organ arrangement in most (if not all) extant angiosperm flowers, at least with respect to their perianth and androecium. This important conclusion challenges accounts of this topic in many textbooks. Sauquet et al. (2017) used separate characters to code the phyllotaxis of respectively the perianth, androecium, and gynoecium. In their highly novel study of correlations in evolution of different characters, they found that phyllotactic patterns of different flower parts are strongly positively correlated with each other. Nevertheless, when selecting character states to draw what has already become a much-reproduced hypothetical diagram of the ancestral flower, they favored a combination of whorled perianth and whorled androecium but spiral gynoecium (Fig. 2). The authors by no means insisted that this is the only possible solution, as the numerous analyses that they conducted revealed an array of possible combinations of character states, each character in each analysis having its own quantified degree of uncertainty. Still, considering the overall evidence of a huge number of statistical reconstructions, they found the combination of whorled perianth and androecium plus spiral gynoecium as the most plausible aggregate condition for not only ancestral angiosperms but also ancestral (eu)magnoliids and ancestral eudicots. Thus, hemicyclic flowers (with some organs arranged along a spiral but others in whorls) were hypothesized to have played an important role in early stages of angiosperm evolution. This suggestion stimulated us to critically re-examine the intriguing phenomenon of hemicyclic flowers. The published data set of Sauquet et al. (2017)—if assumed to be a representative sample of overall floral diversity—provides a unique opportunity to estimate frequencies within angiosperms of flowers with different patterns of phyllotaxis. Using the data set of Sauquet et al. (2017), we calculated occurrences of different combinations between whorled, spiral, and uncertain conditions in adjacent regions of angiosperm flowers; specifically, between gynoecium and androecium (Table 1) and between androecium and perianth (Table 2). We consider this approach as complementary to calculating actual correlations between the three characters. Sauquet et al. (2017) did not record transitions from a spiral calyx to a whorled corolla but instead scored such taxa as having a cyclic perianth, an approach with which we agree; the spiral calyx could indeed represent a transition from helically arranged vegetative leaves, but alternatively an impression of spirality can merely reflect a non-integer merism of 2½ (see below). An emphasis on the phyllotaxis of the corolla is sensitive to the problem of the likely nonhomology of all angiosperm petals (Zanis et al., 2003; Ronse De Craene, 2007). However, distinguishing between andropetals and bracteopetals is not always simple; controversial views on certain taxa preclude defining the phyllotaxis of andropetals and bracteopetals as two different characters. With all cases of uncertainty (including irregular and polymorphic conditions) set aside, the data set of Sauquet et al. (2017) contained only four species in which phyllotaxis of the gynoecium differs from that of the androecium (Table 1). In all four cases, the androecium was recorded as spiral and the gynoecium as whorled. On closer examination of these four species, we believe that none of them provides a convincing example of the hypothesized transition from spiral to whorled phyllotaxis at the boundary between the androecium and the gynoecium. Flowers of Androstachys johnsonii (Picrodendraceae: Malpighiales) reportedly possess a spiral androecium and a whorled gynoecium (Leandri, 1958), but as they are unisexual (lacking even sterile organs of the opposite sex), these differences can be viewed simply as a manifestation of dimorphism between male and female flowers, comparable with the differences in perianth phyllotaxis between male and female flowers of Buxaceae (von Balthazar and Endress, 2002). Multistaminate male flowers of A. johnsonii have not yet been studied anatomically and developmentally, and the spiral interpretation of their androecium has since been questioned (Webster, 2014). Forstera bidwillii (Stylidiaceae: Asterales) has only two stamens (Mildbraed, 1908). As in most other Stylidiaceae, these are posterior-lateral stamens of the originally pentamerous androecium (Eichler, 1875). A spiral arrangement remains in the eye of the beholder if only two organs are involved. Limeum africanum (Limeaceae: Caryophyllales) has six or seven stamens arranged in a single whorl (Brockington et al., 2013). The sequence of stamen initiation is difficult to observe, as their primordia appear almost simultaneously. Subsequent stamen growth allows their arrangement in a sequence approximating a 3/8 pattern (though the divergence angle between stamens 3 and 4 often exceeds 180°). The direction of the spiral in the androecium is opposite to that of the outer sepaloid members of the perianth (Brockington et al., 2013). It is therefore more logical to interpret the androecium of Limeum as whorled with unequal stamens rather than as spiral. Ronse De Craene (2013) considered the androecium of Caryophyllales as basically whorled and interpreted the fertile stamens of L. africanum as belonging to two whorls. Peumus boldus (Monimiaceae: Laurales) has male and female flowers with all organs arranged according to a continuous Fibonacci spiral (Staedler and Endress, 2009), so the whorled condition assigned to the gynoecium in the data set of Sauquet et al. (2017) is probably an error. Despite a thorough literature search, we were unable to find convincing published records of hemicyclic flowers with a change of phyllotaxis between the androecium and gynoecium in species not already encompassed by the extensive taxon sampling of Sauquet et al. (2017). An earlier record of a spiral gynoecium in Octolobus (Malvaceae: Malvales) is most probably erroneous (Endress, 2014). Sargentodoxa cuneata (Lardizabalaceae: Ranunculales) was scored by Sauquet et al. (2017) as having a whorled perianth, numerous spirally arranged carpels, and an unknown arrangement of stamens. This species merits special attention as a candidate for the possession of a whorled androecium but a spiral gynoecium. However, the detailed developmental data of Zhang and Ren (2008) clearly show that stamen and carpel arrangement is whorled or sometimes irregular in Sargentodoxa. Ren et al. (2010) reported exceptional variation in patterns of organ arrangement in another group of Ranunculales, Ranunculaceae-Anemoneae; spiral, whorled, and somewhat irregular patterns sometimes co-occur in a single species. However, in no case was a change of phyllotactic pattern documented at the transition from androecium to gynoecium. Once again setting aside all cases of uncertainty, the data set of Sauquet et al. (2017) then contains only five species where perianth phyllotaxis was scored as differing from that of the androecium (Table 2). In three of these species, the perianth was recorded as whorled and the androecium as spiral (Table 2). Forstera bidwillii does not represent a convincing example, as its androecium consists of only two stamens (see above). Sampson (1969) described a transition from a whorled perianth to an apparently spiral androecium in Hedycarya arborea (Monimiaceae: Laurales), but provided no illustration of the actual stamen arrangement. Staedler and Endress (2009) studied in great detail male flowers of a congeneric species, H. angustifolia, and provided strong evidence for whorled perianth and androecium, though they documented the occurrence of transitions between cyclic and spiral phyllotaxis in some other Laurales such as Doryphora aromatica (Atherospermataceae) and Glossocalyx longicuspis (Siparunaceae). Liriodendron chinense (Magnoliaceae: Magnoliales) has a whorled perianth but a spiral androecium resembling those of certain Magnoliaceae not sampled by Sauquet et al. (2017) (Xu and Rudall, 2006; Leins and Erbar, 2010). The transition from a spiral to a whorled condition is recorded in two species in the data set (Table 2). The perianth is spiral but the gynoecium is whorled in Nelumbo lutea (Nelumbonaceae: Proteales), as well as in another species of the genus, N. nucifera. However, the stamens appear on an androecial ring primordium, and no orthostichies or parastichies can be determined (Hayes et al., 2000), so stamen arrangement could be irregular. Xanthorhiza simplicissima (Ranunculaceae: Ranunculales) is described as having a spiral perianth and a whorled androecium, though the perianth appears cyclic in some published illustrations (e.g., fig. 4E of Endress, 2010). Some other angiosperms show a transition from spiral to whorled phyllotaxis at the boundary between the perianth and the androecium, an example being Berberidopsis (Berberidopsidaceae: Berberidopsidales). As in Nelumbo, flowers of Berberidopsis possess a clearly spiral perianth and a clearly whorled gynoecium (Ronse De Craene, 2017). Even though stamen number is not precisely fixed in both of the investigated species and they are somewhat unequal during development, it is most plausible to envisage a transition point between spiral and whorled phyllotaxis above the perianth and to interpret the stamens as forming a whorl (see also Ronse De Craene, 2010). As with many other morphological characters, recognizing "natural" character states of floral phyllotaxis is problematic. One major problem is the relative importance of early flower development (including positions and sequence of organ initiation) compared with anthetic flowers. A widely cited example is Illicium (Schisandraceae: Austrobaileyales; Robertson and Tucker, 1979; Endress, 2001; Leins and Erbar, 2010). When early developmental stages are considered, there is no doubt of the spiral arrangement of all the floral organs. In contrast, at anthesis, carpels of Illicium are arranged in a whorl around a large apical residuum of the receptacle. (Interestingly, mature stamens also appear whorled in some published images of Illicium [e.g., fig. 7H of Endress, 2001], so the relative arrangements of both stamens and carpels appear to become adjusted during ontogeny.) The converse condition occurs in genera such as Limeum (see above), in which stamen primordia are initiated almost simultaneously (or the plastochrons are so short that they cannot readily be recognized), but subsequent growth of stamens is unequal, resulting in an apparently spiral androecium (Brockington et al., 2013). Similar observations are available for many other angiosperms. Thus, in our view, distinguishing between "developmental" and "definitive" phyllotaxis is at least partially misleading. Recognizing spiral and whorled phyllotaxis in problematic situations requires data sets that are both comparative and developmental. Distinguishing between states of a morphological character is, by definition, an a priori homology assessment, for which formalized criteria should be (1) explicitly stated and (2) used with caution, as they sometimes contradict each other. The importance of using comparative morphology and developmental data can be illustrated by interpreting floral phyllotaxis in the basal eudicot family Sabiaceae, of which Sauquet et al. (2017) used two species, Sabia swinhoei and Meliosma veitchiorum. In the former, phyllotaxis of perianth, androecium, and gynoecium are all scored as whorled, whereas in the latter all floral parts are interpreted as spiral. Detailed structural and developmental data are available for Meliosma veitchiorum and species of Sabia other than S. swinhoei (Wanntorp and Ronse De Craene, 2007; Ronse De Craene et al., 2015). Interpreting floral phyllotaxis in both genera is problematic due to the lack of alternation between the apparently pentamerous whorls of sepals, petals, and stamens, and the sequential initiation of at least the outermost floral organs. Flowers can be interpreted as having either a spiral of 2/5 or several non-alternating whorls. We suspect that there is no method more conclusive than expert decision-making available to resolve this question, but we can at least suggest that scoring of floral phyllotaxis should be uniform in Sabia and Meliosma. Problems resembling those with Sabia exist with interpretation of flowers in some Caryophyllales. Sattler (1973) documented floral development in Chenopodium album, in which the five perianth members are initiated in a spiral sequence, the five stamen primordia are initiated in the same spiral sequence, and the central gynoecium is initiated as a girdling primordium that becomes two-lobed, with two carpels whose positions we can speculatively consider as continuing the androecial spiral. These data could allow a spiral interpretation of the flower in Chenopodium (with organs arranged along a 2/5 spiral, as in Sabia). However, Olivera et al. (2008) rejected a spiral interpretation for flowers of Chenopodiaceae-Amaranthaceae based on their observations of Beta. Given the presence of only five perianth organs in Chenopodium (compared with 10 in Sabia), interpretation of their flowers as spiral or whorled is highly dependent on outgroup comparison, as is resolving the question whether five organs (e.g., petals) alternating with the stamens were lost during the course of evolution. Alternatively, the perianth of Chenopodium (which resembles the condition found in a broad range of Caryophyllales) could be interpreted as whorled with a merism of 2½; that is, intermediate between 2 and 3 (Choob and Yurtseva, 2007), so that the five tepals (or sepals) belong to two distinct whorls (two organs in the outer whorl, two organs in the inner whorl and a "hybrid" organ partly belonging to either whorl). This interpretation resembles ideas on the derivation of a pentamerous whorl from two trimerous whorls (Ronse De Craene et al., 1998) or from two dimerous whorls (Sauquet et al., 2017). An important feature of theoretically constructed 2½-merous flowers is that their organ arrangement can be the same as in flowers with organs arranged along a 2/5 spiral (Fig. 3C). Should we interpret whorled flowers (or perianths) with 2½ merism and spiral flowers with a divergence angle of 2/5 (and a pentamerous perianth with quincuncial aestivation) as identical or different states with respect to character scoring in data matrices? Resolving this issue is important for interpretation of the origin of pentamery in Pentapetalae (Fig. 3). The terms hemicyclic and spirocyclic are commonly viewed as synonyms (e.g., Leins and Erbar, 2010), encompassing all cases of co-occurrence of whorled and spiral patterns in a given flower. Thus, there exist only two major types of switches in phyllotaxis between organ types within a single flower—from outer spiral to whorled inner organs and from outer whorled to spiral inner organs. Although it is possible that in some cases more than one such switch occurs within a flower, published examples of hemicyclic flowers normally describe only one, as was consistently the case in the taxa discussed in the previous section. We propose hereafter to use "spirocyclic" to describe flowers with outer spiral and whorled inner phyllotaxis and "cyclospiral" for flowers with outer whorled and inner spiral phyllotaxis. As outlined above, the data set of Sauquet et al. (2017) provides two convincing examples of spirocyclic and two examples of cyclospiral flowers; in all four cases, the switch occurs between the perianth and the androecium. When all species with uncertain conditions in the perianth and/or the androecium are set aside, phyllotaxis is categorized identically in both regions in 450 species (Table 2). Therefore, spirocyclic and cyclospiral flowers together appear to characterize fewer than 1% of extant angiosperms. A change from whorled to spiral phyllotaxis at the transition to the androecium (in cyclospiral flowers) could be associated with the occurrence of a long plastochron preceding initiation of the first stamen and much smaller stamen primordia relative to perianth members. As noted by Staedler and Endress (2009), the long plastochron before sporophyll initiation may shape a circular dome-shaped floral apex, which in turn would generate Fibonacci-spiral phyllotaxis. A decrease in primordium size at the transition to the androecium is apparently not restricted to cyclospiral flowers. For example, in the spirocyclic flowers of Nelumbo, the stamen primordia are only about half the diameter of the petal primordia (Hayes et al., 2000). Data on plastochron length are clearly important, but most published data on flower development acquired using scanning electron microscopy provide no direct evidence of time intervals. The absence of convincing examples of a switch in phyllotaxis at the transition from androecium to gynoecium is remarkable and implies the presence of a fundamental developmental constraint. Our focused literature search, combined with arguably the most representative data set of floral characters ever assembled (Sauquet et al., 2017), together provide ample evidence for such a constraint. Unfortunately, to our knowledge, this topic has received comparatively little attention in the literature on genetic regulation of floral development. It is tempting to suggest that the C-class genes that are involved in regulation of stamen and carpel identity as well as flower meristem determinacy (e.g., Mizukami and Ma, 1997; Lenhard et al., 2001; Dreni et al., 2011; Thomson et al., 2017) also influence mechanisms that regulate floral phyllotaxis. Why did the data set of Sauquet et al. (2017), which contains no actual example of an extant taxon showing both a spiral gynoecium and a whorled androecium, lead the authors to infer such a combination as being the most plausible for not only the ancestral flower but also flowers reconstructed for two other key nodes of the extant angiosperm phylogeny (those subtending magnoliids and eudicots)? It is reasonable for Sauquet et al. (2017) to expect that the ancestral flower may not have possessed the precise combination of character states found in any extant species, given that 27 characters were included in their analyses. Indeed, Arber and Parkin (1907) considered that different rates of evolution of different character states made it unlikely that an accurate example of their hypothetical ancestral flower ("euanthostrobilus"), which included all organs spirally arranged, ever actually existed. However, it is difficult to accept that the ancestral flower featured a combination of two characters, tightly linked developmentally, that is apparently unknown in any extant species (at least, a combination that was wholly absent from the analyzed data set). Clearly, the statistically predicted possibility of the existence of this character combination derives from different patterns of distribution of gaps in data sets for different characters. As can be calculated from Tables 1 and 2, the proportion of uncertain situations is unequal among data on phyllotaxis of different floral parts (gynoecium 49.2%; androecium 40.5%; perianth only 11.2%). These accounts include cases of irregular and polymorphic phyllotaxis, but such cases are rare in the data set and do not affect the overall picture. The high proportion of gaps in scoring gynoecium phyllotaxis is in part related to the fact that the character cannot be properly scored in taxa with monomerous gynoecia. Pseudomonomerous gynoecia provide further difficulties, as patterns of arrangement of sterile carpels are often especially difficult to discern. In a situation where up to half of the included taxa were not characterized with respect to gynoecium phyllotaxis, the degree of confidence in the inferred reconstruction is weaker than those for the androecium and especially the perianth. Indeed, Sauquet et al. (2017) readily admit to a considerable level of uncertainty in their reconstruction of the ancestral phyllotaxis of the gynoecium. Because spirocyclic and cyclospiral flowers are rare among extant angiosperms, and apparently none undergoes a switch of phyllotactic patterns between androecium and gynoecium, we believe that the absence of such a switch was also the case in flowers of the earliest angiosperms. We suggest that the ancestral flower was either entirely whorled (up to the gynoecium) or entirely spiral. Given the overall arguments for whorled nature of androecium and perianth in the ancestral flower provided by Sauquet et al. (2017), it is possible that the ancestral flower was whorled. Admittedly, we cannot readily exclude the more traditional hypothesis that the ancestral flower was spiral, as this condition occurs in extant basal angiosperms such as Amborella and Austrobaileya (Fig. 1). An additional argument supporting the idea of an ancestrally whorled gynoecium comes from the Cretaceous fossil Canrightia, which is putatively related to extant Chloranthaceae and possesses a whorl of united carpels (Friis et al., 2011). As all extant Chloranthaceae are unicarpellate (Endress, 1987, 2001), this potentially important lineage of basal angiosperms, which has an extensive and early fossil record, was not characterized with respect to gynoecium phyllotaxis by Sauquet et al. (2017). The data on Canrightia strongly suggest that flowers of the stem group of Chloranthaceae were whorled up to the gynoecium (Doyle et al., 2014; Friis et al., 2015). Cecilanthus polymerus is a Cretaceous polymerous flower of a basal angiosperm with whorled arrangement of all organs up to the gynoecium (Herendeen et al., 2016). In this example, the excellent preservation of the fossil allowed precise determination of phyllotaxis type. However, many other fossils are problematic as sources of information on floral phyllotaxis. For example, the much-debated Archaefructus illustrates the difficulty of confidently characterizing phyllotaxis in even well-preserved two-dimensional fossils; opinions differ on whether the reproductive axis represents (1) an elongate hermaphroditic flower consisting of a spiral androecium and less clearly categorized helical/pseudowhorled/subopposite gynoecium (Sun et al., 2002) or (2) an inflorescence of numerous unisexual flowers that grade toward the apex from opposite to subopposite or spiral phyllotaxis (Friis et al., 2003). Minimizing gaps when scoring the floral phyllotaxis of angiosperms could obviously be achieved by more detailed investigations of morphology in species that have not yet been sufficiently studied. However, this approach will be unable to resolve problems like perianthless flowers, unistaminate androecia, and unicarpellate gynoecia. An alternative approach would be to analyze floral phyllotaxis as a whole. Setting aside the cases of irregularities, four character states can readily be recognized: spiral, whorled, spirocyclic, and cyclospiral. Unfortunately, analysis of a four-state character using model-based methods is technically problematic (Sauquet et al., 2017). As spirocyclic and cyclospiral flowers are apparently so rare in angiosperms, we believe that scoring them as uncertain in order to recognize only two character states, whorled and spiral flowers, would provide an opportunity to produce a data set containing fewer than 10% of cells coded as uncertain. This could potentially lead to more confident reconstruction of phyllotaxis in the hypothesized ancestral angiosperm flower. A further advantage of such simplification of the character would be to increase the probability of scoring non-angiosperm seed-plant outgroups. The absence of a universally accepted theory on homologies of the angiosperm flower and its parts, especially stamens and carpels, dictates severe difficulties in employing any outgroups to help reconstruct the evolution of floral characters. On the other hand, if adopting a broad definition of the character "floral phyllotaxis", use of data on phyllotaxis in gymnosperm reproductive structures becomes feasible. In almost all extant gymnosperms, whorled versus spiral phyllotaxis is conserved between reproductive and vegetative structures (in Metasequoia, phyllotaxis is normally decussate, including cones, but spiral on leading branches: Liguo et al., 1999). Therefore, whatever the fundamental interpretation of the angiosperm flower—as a simple or compound strobilus (reviewed by Rudall and Bateman, 2010)—data on phyllotaxis in extant gymnosperms can be used in outgroup comparisons for analyses directed toward floral phyllotaxis. Other gymnospermous groups identified by morphological cladistic analyses as close relatives of the angiosperms (those not belonging to "acrogymnosperms": Doyle, 2012) are wholly extinct. Use of fossil outgroups is problematic in many cases due to their incomplete preservation and inevitable disarticulation, examples of such a key taxon being the caytonialean Caytonia and the bennettitalean Williamsoniella. The superbly anatomically preserved bisporangiate reproductive structures of the bennettitalean Cycadeoidea (e.g., Crepet, 1974) might have been expected to provide unequivocal evidence of the phyllotaxis of the ovulate and pollen-producing zones, but (1) it is unclear which structures should be analyzed with respect to their arrangement (interseminal scales, ovule stalks, or both) and (2) the arrangement of such densely spaced organs should be studied with the aim of recognizing and counting clockwise and anticlockwise parastichies to establish a pattern of phyllotaxis (Endress, 2006); to our knowledge, no study of this sort has yet been attempted. Arber and Parkin (1907) reported that in Bennettitales, the sterile "perianth" organs and the ovule-bearing organs were spirally arranged, but not the microsporophylls, which were "cyclic". They regarded the cyclic condition as derived within both bennettites and angiosperms. The fact that whorled and spiral phyllotaxis are not interchangeable during the transition from vegetative to reproductive structures in extant gymnosperms is significant when viewed in this context. In contrast, many angiosperms, including members of the basal grade and magnoliids, successfully achieved the transition from spiral (or distichous) vegetative leaves to a whorled perianth or from decussate vegetative leaves to a spiral perianth (Staedler et al., 2007; Endress and Doyle, 2007; Staedler and Endress, 2009). Whorled floral phyllotaxis in many respects resembles whorled (including decussate) phyllotaxis on vegetative shoots. There is, however, evidence of important differences, at least in some angiosperms. Patterning of vegetative leaves is strictly acropetal. Positions of new leaf primordia are determined at the shoot apex using positional information of older leaves (or primordia) situated below them (termed unidirectional prepatterning). At least in some whorled flowers (e.g., Choob and Penin, 2004; Choob and Yurtseva, 2007; Sokoloff et al., 2007; Ronse De Craene, 2010; Penin and Logacheva, 2011; Brockington et al., 2013), the number and position of stamens (or inner-whorl stamens) is regarded as dependent on positional information relating to the carpels (bidirectional prepatterning). A significant number of angiosperms deviate from a strict centripetal developmental progression of visible appearance of organs (Rudall, 2010). In some angiosperms, visible initiation of the gynoecium takes place before stamen primordia are initiated. Even in flowers where the gynoecium is the final structure in the visible sequence of appearance of organ primordia, such as in Arabidopsis and other Brassicaceae (Smyth et al., 1990; Erbar and Leins, 1997), there is evidence of bidirectional prepatterning of positions of organ primordia, carpel prepatterning occurring before stamen prepatterning (Choob and Penin, 2004; Skryabin et al., 2006; Rudall, 2010; Penin and Logacheva, 2011). Thus, although precise identification of all whorled flowers as having unidirectional (Fig. 4A) or bidirectional (Fig. 4B) prepatterning is problematic without intensive experimentation, there is strong evidence that bidirectional prepatterning is common among whorled angiosperm flowers. We hypothesize that spiral gynoecia with a clear Fibonacci pattern of carpel arrangement cannot develop in flowers with bidirectional prepatterning. Indeed, in contrast to acropetal patterning (e.g., Reinhardt et al., 2003; Berleth et al., 2007), it is difficult to imagine a developmental mechanism capable of creating a Fibonacci spiral in a reverse direction (i.e., basipetally). It is even more difficult to imagine the formation of a continuous spiral (and continuous sets of left and right parastichies) from the perianth through the stamens to the carpels in a flower that is subject to bidirectional patterning (Fig. 4D). So spiral flowers most likely have unidirectional prepatterning. Whereas traditional ideas on ancestrally spiral flowers implied that bidirectional prepatterning appeared during the evolution of crown-group angiosperms, the hypothesis of ancestrally whorled flowers opens a door for the possibility of ancestrally bidirectional prepatterning (Fig. 4B), followed by several independent transitions to the unidirectional condition, including spiral flowers (Fig. 4C). Data on variation in floral phyllotaxis in the basal angiosperm family Nymphaeaceae suggest that these flowers have bidirectional prepatterning. Indeed, their perianth is whorled, and the multistaminate androecium commences development as whorled, though it later displays increased irregularity in the positions of the inner stamens (Endress, 2001). The carpels of water lilies form a clear whorl, an outcome that appears unlikely if prepatterning of all floral organs was strictly acropetal. The concept of ancestrally bidirectional prepatterning is potentially testable using methods of developmental genetics; such testing should now be made a priority. The contrast between our conclusions on floral phyllotaxis and those of Sauquet et al. (2017) ably illustrates one of the hazards inherent in dividing the phenotype of an organism into putatively homologous characters and characters into particulate character states ("traits" sensu Sauquet et al.). This reductionist approach is essential in any branch of science, not least for developing matrices to enable any mathematical approach to exploring phylogeny reconstruction and trait evolution. However, when interpreting the results of such reductionist studies, a more holistic viewpoint often proves helpful, given that the developmental processes that together ultimately generate a mature organism are undeniably holistic. Developmental transitions that are predicted through reductionist statistics will not always prove feasible in the more holistic reality dictated by "developmental parsimony". The simplest strategy that can be adopted by the apical meristem that generates leaf phyllotaxis in a shoot is to continue that pattern of organ initiation into the flower that it bears. Although angiosperms have acquired the ability to perform such a transition, to further switch phyllotactic patterns between the closely spaced zones of perianth, androecium, and gynoecium initiation appears to constitute a developmental bridge too far, instead providing an example of a significant developmental constraint. A slightly earlier, but similarly taxonomically well-sampled, analysis of several floral characters of extant angiosperms was conducted by O'Meara et al. (2016), who focused on seeking positive correlations between character states thought to enhance speciation rates. They concluded that corolla presence, bilateral symmetry, and few stamens together constituted an aggregate key innovation, but also found that this character combination required on average tens of millions of years to achieve. We would argue that such periods are far longer than would be required by any known evolutionary mechanism if each supposedly advantageous trait were free to evolve independently—initial constraints on development (a topic not directly addressed by O'Meara et al.) once again provide the most obvious and most likely explanation for this profound evolutionary deficit. We are grateful to the editor and two anonymous reviewers for helpful suggestions. The work of D.D.S. and M.V.R. was supported by the Russian Foundation for Basic Research (project 15-04-05836).