Reconstructing the ancestral angiosperm flower and its initial specializations^†

Taxa

Besides adding Ceratophyllum and Hydatellaceae, we have increased our taxon sampling in other groups. In Chloranthaceae, we split Chloranthus and Sarcandra, treated as one taxon in 57, into the two genera and rescored several characters based on 62. In Ranunculales, we have added Circaeaster and split Hydrastis and Glaucidium from “core” Ranunculaceae (their sister group according to 121) because they differ substantially in floral features and may thus affect reconstruction of floral evolution. For the same reason, we split Magnoliaceae into Liriodendron and Magnolioideae (Magnolia s. l. of recent authors) and Trochodendraceae into Trochodendron and Tetracentron. We have modified the scoring of Platanaceae (now Platanus) and Buxaceae, in which we previously included presumed fossil relatives, to apply strictly to the extant crown groups, in anticipation of testing the position of the fossils. We have increased our sampling of Alismatales for future tests of comparisons with fossils. We have added Nartheciaceae because they appear to be related to but less modified than Dioscoreaceae (12), and Melanthiaceae as a relatively plesiomorphic exemplar of Liliales (16).

In Ceratophyllum, the fertile structures have been variously interpreted (70; 127): at one extreme, as female flowers with a single carpel surrounded by tepals and as male flowers with tepals and numerous stamens; at the other, as female flowers with no perianth but bracts lower on the axis and as spikes with basal bracts and numerous male flowers consisting of one stamen, with no perianth or individual subtending bract (73). For the purposes of this analysis, we have provisionally accepted the second interpretation. First, in the female structures, single carpels occasionally occur in the axils of the sterile appendages (1; 127), suggesting that the latter are bracts rather than tepals and the system is a reduced inflorescence. Second, the stamens have an extremely labile phyllotaxis and marked acropetal delay in maturation (70), which is a common pattern in flowers of spicate inflorescences but anomalous within the androecium of a multistaminate flower.

Our scoring of Archaefructus is based on whole plants of A. liaoningensis and A. sinensis (185, 49, 134; 93); features are generally consistent in A. eoflora (129), which may represent either a smaller species or a younger stage of A. sinensis. We analyzed the position of Archaefructus using two alternative scorings, one (Archaefructus inf) assuming that fertile axis was a raceme of male and female flowers consisting of usually two stamens and one or two carpels, the other (Archaefructus flo), following 186, that it was a bisexual flower or preflower with paired stamens below and carpels above. 93 questioned whether the bodies that 187, 134) described as pollen were in fact pollen grains, because of their irregular size and shape, but we provisionally assume that at least some of them are pollen and have scored them based on the most convincing specimen, illustrated in Fig. 2F of 186. To evaluate the alternative interpretation, we have used a third scoring (Archaefructus NP) that corresponds to Archaefructus inf with pollen characters treated as unknown. 185, 134) described the carpels as conduplicate (= plicate), but in extant carpels of similar appearance this cannot be determined without developmental or anatomical evidence (93; 74). They described the fruits as follicles, but they did not actually report dehiscence. The seeds appear to have a palisade exotesta as defined here (character 101, including not only radially elongated but also shorter sclerotic cells): 185, 134) described the surface as consisting of epidermal cells with cutinized anticlinal and periclinal walls.

129 interpreted seeds of A. eoflora as orthotropous, but their published illustrations are not clear enough to determine whether the seeds were anatropous or orthotropous; the figure of the end of a seed in Fig. 2C of 185 is actually more suggestive of an anatropous ovule. Hence we have scored ovule curvature (93) as unknown. 129 described one lateral unit in A. eoflora as bisexual, with one stamen and two carpels, and they also interpreted one unit in the type specimen of A. sinensis (186) as bisexual. Under our character definition, A. eoflora might be scored as uncertain (0/1) for flower sexuality (26). However, we believe it would be premature to rescore Archaefructus as a whole in this way.

Characters

In this study we have not included all the characters in our most recent version of the 57 data set, many of which are not relevant for our present purposes, where we have used fixed backbone constraint trees as a framework for placement of Ceratophyllum and Archaefructus and reconstruction of floral evolution, and would require excessively lengthy documentation and argumentation. We have included all floral characters, including those of stamen, carpel, and ovule morphology. In addition, we include all those nonfloral characters needed for analysis of the position of Ceratophyllum and Archaefructus. Characters omitted because they do not exist or are inapplicable in Ceratophyllum include aspects of secondary xylem and phloem, leaf anatomy, and the inner and outer integuments, which are reduced or fused into a single integument.

Some of the most important and complex arguments for decisions in definition of characters and scoring of particular taxa are discussed in this section, others that are less problematic or significant in Appendix 1. Because of space limitations, we cite only general sources of information for particular character sets and especially important references on particular taxa, and reserve more detailed documentation and resolution of differences between our interpretations and those of other workers for elsewhere. We concentrate on references for new taxa and characters; for those used in 57, readers are referred to that article.

Our general philosophy on definition of characters is explained in more detail in 57. Few of our characters are quantitative in the sense of continuous (e.g., pollen size, nexine thickness), but there are often series of conditions that could be grouped into many states or a few. In general, we have tried to break the variation into a smaller number of states in ways that make morphological (especially developmental) sense and reduce the number of uncertain (“polymorphic”) scorings of taxa (assuming that this reduction is evidence that the variations included in each of the states are related). Several important changes concern replacement of multistate with binary characters, which can sometimes improve resolution of relationships in cases where the optimization of a multistate character would be ambiguous. In several cases we previously used unordered multistate characters to combat the Maddison “long distance” effect (147): where the ancestral state in one clade in which a structure (more generally a character) occurs in two (or more) versions influences the polarity of the character in another clade that has the structure, even though the structure does not exist in the intervening lines and presumably arose independently. This artifact can be avoided by treating lack of the structure as one state of a multistate character and different versions of the structure as other states. However, this procedure weakens the contrast between presence and absence of the structure as an independent source of information on relationships.

An example that underlines the importance of the Maddison effect concerns presence or absence of a perianth. In 57, we treated the number of perianth whorls as an unordered multistate character, with no perianth one of four states. A group where this may cause problems is Chloranthaceae, where Hedyosmum has one perianth whorl and the other genera have no perianth. Since most outgroups have a perianth, its presence might appear to be evidence for the basal position of Hedyosmum, or in other words its loss could be evidence for the monophyly of the remaining genera (which is supported by molecular evidence). However, when presence and number of whorls are treated as a single unordered multistate character, scoring Hedyosmum as having one whorl does not favor a basal position in Chloranthaceae because the outgroups have two or more whorls, states that are not recognized as any more similar to one whorl than to none. For this reason, 62 split the 57 character into two—one for presence or absence of a perianth, the other for number of whorls, with taxa lacking a perianth scored as unknown—and we have followed this solution here (characters 31, 34). 147 pointed out cases where this procedure is unlikely to cause problems, notably where loss of a structure occurs in a terminal clade. With relationships largely inferred from molecular data, cases where this may cause artifacts can usually be recognized and treated in discussion.

Additional important changes involve other characters of floral organization. In 57, we recognized phyllotaxis and merism (merosity) as separate characters in both the perianth and the androecium, but this poses problems for scoring of merism in spiral taxa. Our solution was to treat merism as a multistate character, with spiral taxa scored as (0) irregular and whorled taxa as (1) trimerous or (2) dimerous, tetramerous, or pentamerous. However, this may introduce bias due to redundancy of spiral phyllotaxis and irregular merism. One solution would be to combine phyllotaxis and merism into a single character, but as discussed in 76, the distinction between spiral and whorled appears to be consistent and independent enough to be treated separately. Our solution is to retain both characters (32, 33 for the perianth; 41, 42 for the androecium) but score spiral taxa as unknown for merism. Optimization of this character across the tree produces artifactual reconstructions of merism in spiral taxa, but this can be considered in discussion.

Scoring the number of perianth and stamen whorls (34, 43) is straightforward in whorled taxa (except for seemingly tetramerous flowers that actually have dimerous whorls, as in Proteaceae, Tetracentron, and Buxaceae, as inferred from the fact that the stamens appear to be opposite the tepals: 204; 20), but again the treatment of spiral taxa poses problems. Many spiral taxa (and Nelumbo, with chaotic stamen insertion on an androecial ring meristem: 109) have numbers of tepals and/or stamens that are comparable to those of taxa with more than two whorls, so we have scored them accordingly. We also used the number of series (in which a certain number of parts fills the circumference of the flower; 76) as a rough substitute for number of whorls.

In most basal angiosperms, all perianth parts are best described as tepals (118; 209; 71; 164) because they are less strongly differentiated than the typical sepals and petals of core eudicots (Pentapetalae of 14). These tepals may be uniform (either sepaloid or petaloid) or differentiated into outer sepaloid and inner petaloid parts, distinctions recognized in character 35. We include both tepals and more differentiated petals in the count of whorls, and staminodes as well as fertile stamens. However, we also introduced a separate character (36) for presence or absence of typical petals (mostly in Ranunculales), defined on more pronounced differences in anatomy and delay in development. Taxa with petals may show differentiation within the outer perianth whorls, such as Nuphar, which has outer sepaloid and inner petaloid “tepals” or “sepals” and much smaller petals.

We have made fewer changes from 57 in characters of individual floral parts. Following 62, to reduce uncertain scorings, we modified the stamen base character (48) to combine short and wide and short and narrow in the same state, and the orientation character (53) to combine slightly introrse with latrorse. We previously treated modes of carpel sealing as a multistate character (corresponding to the four types of 76). However, carpel sealing has two potentially independent aspects, degree of postgenital fusion and secretion, which we have split into two characters (76, 77). We separated types of papillae (82) from larger protuberances (81), because pluricellular papillae and protuberances co-occur in Amborella and Trimenia but not in other taxa and therefore appear to represent independent characters.

Many aspects of floral evolution were also treated by 165. They made less effort to ensure independence of characters: for example, lack of perianth was a state in three of their characters. This was not necessarily a problem in their study, in which they plotted characters on a molecular tree, and it may be useful in assessing the implications of different character definitions. However, such redundancy poses problems if characters are used for tree reconstruction, since it may overweight what was presumably a single change—for example, loss of perianth. Because we intend to use our data set in a future combined analysis and have used it to investigate the relationships of Ceratophyllum and Archaefructus in the current study, we have tried to minimize redundancy among characters.

Inflorescence characters deserve special attention as an area where we have made major modifications. In our previous analysis (57), we recognized a relatively crude inflorescence character emphasizing degree of branching, with three states: solitary flowers; racemes, spikes, and botryoids; and more richly branched inflorescences such as panicles and compound inflorescences of racemes, spikes, and botryoids. Thus in inflorescences of the second state, we did not recognize the standard contrast between indeterminate and determinate inflorescences, or the related distinction of 199 and 213 between polytelic systems with no terminal flower (racemes, spikes, thyrses) and monotelic systems in which all axes terminate with a flower (botryoids, thyrsoids, panicles—all sometimes imprecisely described as “cymes”). This was because of a perception that the two types intergrade within taxa, such that many taxa would have to be scored as uncertain. However, closer examination has led us to conclude that a different grouping of traditional types into basically monotelic and polytelic states (in character 22, Fig. 1) leads to fewer problems than we had thought and is more informative: one state includes units lacking a terminal flower (racemes, spikes, thyrses), the other those with a terminal flower (botryoids, thyrsoids, panicles). Although taxa often vary between types within each of these states, there is less variation between types belonging to the different states.

Racemes and thyrses differ on whether the lateral units on the indeterminate axis are single flowers or cymes. Cymes are branching systems that can have one to several branching orders (i.e., a main axis and lateral branches of one to several orders formed by repeated branching of these laterals), but each axis has not more than two lateral branches of the next higher order, and all axes are usually determinate. Botryoids are the other way around: they have only one branching order (i.e., a main branch and lateral branches of only the first order), but the number of first-order lateral branches is not limited until the terminal flower is formed; as in cymes the axes are determinate. When there is only one branching order and axes have only one or two lateral flowers, cymes and botryoids cannot be distinguished unless more highly branched units are found. Racemes and thyrses both occur within taxa such as Chloranthaceae, where Hedyosmum has thyrses of female flowers and spikes of male flowers, and some species of Ascarina have spikes, others thyrses. We have recognized this distinction with a separate character (23, lateral units single flowers or cymes). We have also distinguished racemes from spikes by introducing a character contrasting pedicellate and sessile flowers (24).

Another important distinction concerns the presence or absence of bracts or leaves (pherophylls) subtending the flowers (25). In Archaefructus, 186 cited the absence of bracts below the paired stamens and carpels as evidence that the fertile axis was a flower (or preflower) rather than an inflorescence. However, subtending bracts are absent in several groups in the present data set, such as Hydatellaceae, Acorus, and Araceae.

Other problems concern the distinction between solitary flowers and racemes, specifically when solitary flowers are borne in the axils of more or less unmodified vegetative leaves. Solitary axillary flowers are sometimes distinguished from lateral flowers in a raceme based on whether they are subtended by normal leaves or modified bracts, but this is more a matter of degree than a fundamental difference in organization. This problem is illustrated by cases in which flowers are borne in the axils of bracts on an axis that then reverts to producing vegetative leaves (e.g., Schisandra, Euptelea; 63; 212). In our previous analysis we scored these as solitary. Alternatively, even systems with flowers in the axils of normal leaves are sometimes described as racemes (213). Because mode of branching seems more fundamental than variation between bracts and leaves, we have adopted this approach, grouping systems where flowers are borne in the axils of bracts and regular leaves as racemes. We group flowers that terminate either a long shoot (the classic terminal condition) or an axillary short shoot in the solitary state (in character 22, Fig. 1): there is variation between these two extremes in many taxa, such as Austrobaileya, Eupomatia, Magnoliaceae, and Calycanthaceae. If the axillary branch (pedicel) bearing the flower has no appendages or at most one or two prophylls, we call the system a raceme; if it has more sterile appendages, we call the flower solitary. This definition allows most taxa to be scored unambiguously. Schisandraceae are still mixed (0/1), since the number of bracteoles varies between zero and three or more among species (212; 170). Special problems in interpretation of Nymphaeales are treated in the Discussion, since they make more sense in a phylogenetic context.

When flowers are unisexual and the inflorescences of male and female flowers differ in type, we have scored the taxon based on the more complex type. Thus, we have scored Hedyosmum, with male spikes and female thyrses, as having thyrses; and Ceratophyllum, with solitary female flowers and male spikes, as having spikes.

Except for three pollen characters (see Appendix 1), all multistate characters were treated as unordered.

Analyses

Our analyses (all based on parsimony) used “backbone constraint” trees, with Recent taxa fixed into one of two topologies. Analyses were performed with the program PAUP* version 3.1.1 (188) and involved 10 or 100 heuristic replicates, stepwise random addition of taxa, and tree-bisection-reconnection (TBR) branch swapping. The relative parsimony of alternative relationships was determined by searching for trees less than or equal to a given number of steps and observing the trees obtained or by moving taxa manually with MacClade (146).

The first backbone tree (henceforth labeled D&E) is a modification of the tree found in our morphological and three-gene analysis (57), with changes where accumulating molecular data have most strongly and consistently contradicted relationships found in our previous study. Essentially this is a handmade supertree. Besides linking Piperales with Canellales, as already discussed, we have moved Euptelea from within Ranunculales to the base of the order, following 133; this position is actually more parsimonious in terms of morphology. Taxa added or split for the reasons discussed earlier have been placed following 141, 121, 178, 19, and 16. In a preliminary analysis, we added Ceratophyllum to the data set and constrained all other relationships as described. The tree found in this constrained analysis is the modified D&E backbone tree used in subsequent analyses.

The second backbone tree (labeled J/M) incorporates relationships of major clades found in analyses of whole plastid genomes by 128 and 153, notably with Chloranthaceae linked with magnoliids, Ceratophyllum with eudicots, and the latter two with monocots. The same relationships were found by 169 in analyses of a smaller plastid data set. Relationships within clades (which were sparsely sampled in the plastid studies) are the same as in the D&E backbone tree.

To investigate the position of Archaefructus, we analyzed the data set with Archaefructus added, using both backbone trees. To assess implications of the hypothesis that Amborella and Nymphaeales form a clade, we rerooted trees manually with the program MacClade version 4.03 (146).

We used MacClade to optimize character evolution on trees, reconstruct ancestral states, and identify characters supporting relationships. When we refer to features as unequivocal synapomorphies of particular clades, this does not mean they are uniquely derived, but rather that the change in state unequivocally occurs at this point on the tree, as opposed to cases where the position of change is equivocal (e.g., where an earlier origin followed by a reversal and two later origins are equally parsimonious, or where the character state in neighboring taxa is unknown).

Phylogenetic results

Our inference that Ceratophyllum is related to Chloranthaceae is supported by five unequivocal synapomorphies: sessile flower (character 24), one stamen (40), embedded pollen sacs (51), one carpel (74), and orthotropous ovule (93). Synapomorphies of Chloranthaceae that are not found in Ceratophyllum and thereby place Ceratophyllum outside the family are sheathing leaf bases (12), interpetiolar stipules (13), and stigmatic protuberances (81). Remarkably, it is only one step less parsimonious to nest Ceratophyllum within Chloranthaceae, where its best position is sister to Hedyosmum, supported by loss of bracts subtending the male flowers (25) and dry fruit wall (97).

A sister group relationship of Ceratophyllum and eudicots, as found in the plastid genome analyses of 128 and 153 and many other molecular analyses (e.g., 169), is nine steps less parsimonious and would be supported by only one unequivocal morphological synapomorphy, dry fruit wall (97), a highly homoplastic character. It is eight steps less parsimonious to link Ceratophyllum with monocots, which would be supported by loss of cambium (4). Whether this parsimony differential is sufficient to overrule the molecular support for a relationship with eudicots needs to be tested by future combined analyses. However, it should be noted that bootstrap support for the link between Ceratophyllum and eudicots is only modest (71% in 153; 74–89% in 169); that analyses by 153 using various methods and subsets of data gave different topologies, some with Chloranthaceae in a more basal position; and that other molecular analyses have linked Ceratophyllum with Chloranthaceae (61; 149; 160). The strength of the morphological synapomorphies might be questioned on the grounds that they largely represent reductions and simplifications from ancestral states in angiosperms, which might be expected to give similar results regardless of their starting point. However, this is not in itself evidence that they are systematically worthless: without Ceratophyllum all these features are valid synapomorphies of Chloranthaceae, which are independently supported as a clade by molecular data.

These results suggest the intriguing possibility that Ceratophyllum is an aquatic derivative of a terrestrial stem relative of Chloranthaceae that already had many features of the crown group. Many additional changes would have to occur on the line leading to Ceratophyllum: origin of a protoxylem lacuna (2), loss of cambium (4), loss of pericyclic fibers (6), dissection of the leaves (20) and shift to dichotomous venation (18), loss of pollen aperture (62) (and almost total reduction of the exine: 189), loss of stigmatic papillae (82), reduction or fusion of the integuments to one (94), and large embryo (109). Chloranthaceae and their extinct relatives are emerging as one of the first successful angiosperm lines (62; 85), which included greater diversity than would be inferred from the four living genera alone. Our results concerning Ceratophyllum therefore raise the possibility that some Early Cretaceous carpels or pollen that resemble Chloranthaceae might actually be closer to Ceratophyllum and might therefore provide evidence on steps in its origin.

Our analysis provides provisional support for the speculative suggestion of 169 that Archaefructus is related to Hydatellaceae. This is the most parsimonious position of Archaefructus when its fertile axis is interpreted as a raceme of unisexual flowers and Ceratophyllum is associated with Chloranthaceae, as with the D&E backbone. Unequivocal synapomorphies of the two groups are loss of floral subtending bracts (25) and loss of perianth (31). The fact that the flowers are unisexual is consistent but not indicative because the polarity of this character is equivocal. Other features of Archaefructus that support a relationship to Nymphaeales as a whole are palmate venation (17) (reduced to one vein in Hydatellaceae), boat-shaped pollen (61), and palisade exotesta (101). This result would suggest that Hydatellaceae may be what became of one member of the Archaefructus group after 125 Myr of further reduction in an aquatic habitat.

In contrast, with the J/M backbone tree, in which Ceratophyllum is divorced from Chloranthaceae and associated with eudicots, it is more parsimonious to associate Archaefructus with Ceratophyllum, based on dissected leaves (20), dichotomous venation (18), loss of floral bracts (25), unisexual flowers (26), and loss of perianth (31). This position is four steps less parsimonious with the D&E backbone, where Ceratophyllum is associated with Chloranthaceae, which have more features that conflict with those of Archaefructus, such as opposite leaves (9), pinnate venation (17), round pollen (61), and reticulate tectum (66). Better evidence on the position of Ceratophyllum could therefore have an impact on the best interpretation of Archaefructus. Our results also depend on uncertain assumptions concerning the morphology of the fertile structures of Archaefructus. When the fertile shoot is interpreted as a flower or preflower (186), which we regard as unlikely, one of the most parsimonious positions of Archaefructus is still with Hydatellaceae, but it is equally parsimonious to place it elsewhere in Nymphaeales. Confirmation of the view of 129 that the seeds of Archaefructus were orthotropous would increase the relative parsimony of a link with Ceratophyllum.

Some of the evidence for a relationship of Archaefructus with Hydatellaceae comes from the report by 187, 134) of boat-shaped, tectate monosulcate pollen grains in Archaefructus, which was questioned by 93. With the D&E backbone, removal of pollen characters weakens the connection of Archaefructus with Hydatellaceae and favors a link with the eudicot genus Euptelea, supported in part by absence of a perianth (31) and one stamen whorl (43), as well as palmate venation (17), shared with eudicots as a whole, and several ovules (90), a synapomorphy of mesangiosperms other than Chloranthaceae and Ceratophyllum. The possibility that Archaefructus was related to eudicots was raised by 93, based especially on the ternate, dissected leaf architecture. Such a relationship would imply that Archaefructus had tricolpate rather than monosulcate pollen, which would be surprising in light of its Barremian-Aptian age, when tricolpate pollen was exceedingly rare outside northern Gondwana (46; 122; 120). However, even in the absence of pollen characters, relationships with Hydatellaceae and Ceratophyllum remain almost as parsimonious with the D&E backbone, and the link with Ceratophyllum is strengthened with the J/M backbone. These results underline the need for more convincing evidence on pollen of Archaefructus.

Our analysis does not address the hypothesis that Archaefructus is a stem relative of all living angiosperms rather than a member of the crown group (186): it only specifies the most parsimonious position(s) of Archaefructus if it belongs in the crown group. However, the crown group hypothesis was supported by an analysis of living and fossil seed plants (53), including all the ANITA lines, Chloranthaceae, and three magnoliids. When Archaefructus was interpreted as having an inflorescence of unisexual flowers, its most parsimonious position was with Hydatellaceae, and a position sister to all living angiosperms was five steps worse. When the fertile axis was interpreted as a bisexual flower, it was again more parsimonious to place Archaefructus in Nymphaeales than below living angiosperms, but by three steps rather than five.

168 cited the order of fertile parts in Archaefructus (stamens basal, carpels apical) as an argument against a relationship with Hydatellaceae, where the female flowers in species with bisexual inflorescences are to the outside (assumed to be basal) and male flowers are central (apical). However, if Archaefructus has racemes and Hydatellaceae have modified thyrses, as 168 argued, the order of flowers in the two groups cannot be so easily compared. If the main axis of the inflorescence in Hydatellaceae (as reconstructed by 168, in fig. 5D) is compared with the main axis in Archaefructus, there is no difference in the relative position of male and female flowers in the two groups. In both, the male flowers (plus female flowers in Hydatellaceae) are borne on more basal lateral units (cymes in Hydatellaceae), while the more distal lateral units are entirely female. Furthermore, the argument that an opposite order of male and female flowers precludes a relationship is not compelling because analogies with other groups suggest that the order of flowers in bisexual inflorescences can reverse. For example, in Buxaceae, male flowers are basal and female flowers terminal in Buxus and Styloceras kunthianum, female basal and male apical in Sarcococca and Pachysandra, and inflorescences are unisexual in other Styloceras species. Based on inferred phylogenetic relationships (205), either one or the other bisexual condition could be ancestral, but the other bisexual type would be derived from it.

Ancestral floral states and initial specializations

In the following sections, we consider the ancestral state reconstructions in Table 1 and their general implications. Contrary to some expectations (e.g., 160), trees in which Amborella is sister to all other angiosperms and those in which it is linked with Nymphaeales have only modestly different implications for ancestral states: all seven differences involve cases in which the ancestral state is equivocal with one rooting and one of the same two states with the other. Addition of Archaefructus has even less impact, with a few important exceptions to be discussed. Finally, except for the positions of Chloranthaceae and Ceratophyllum, the differences between arrangements of mesangiosperm lines in the D&E combined and J/M plastid trees (128; 153) have generally minor effects. This result seems due to two factors. First, inferences on ancestral states are most dependent on relationships in the ANITA grade, which are the same with both arrangements. Second, very few morphological changes occurred on the internodes between the three main lineages of mesangiosperms (magnoliids, eudicots, and monocots), however they are arranged, presumably because these lineages radiated in a very short time (153), the same reason their relationships have been so difficult to resolve.

Inflorescence organization

Because of varying views on interpretation of flowers and inflorescences in taxa such as Archaefructus, Hydatellaceae, and Chloranthaceae and recent suggestions that the distinction between inflorescences and flowers may be labile or problematic in basal angiosperms (90; 168), we have considered characters of inflorescences as well as flowers.

Based on our results, with Amborella basal (Fig. 3), the ancestral inflorescence type (character 22) in angiosperms is equivocal: either botryoids, as in Amborella; or racemes (which some authors might describe as stems with solitary axillary flowers), as in Nymphaeales, Chloranthaceae (modified to spikes and thyrses), and basal eudicots and monocots. However, if Amborella is linked with Nymphaeales, the ancestral type can be reconstructed as a raceme. Both hypotheses imply that solitary flowers, often considered ancestral in angiosperms, are instead derived: from racemes in Austrobaileyales (with a shift to botryoids in Trimenia), magnoliids, and Nelumbo, and from botryoids in the Hydrastis-Glaucidium clade in Ranunculaceae. In magnoliids, solitary flowers may have evolved either once from racemes at the base of the Magnoliales-Laurales clade, with a reversal in Myristicaceae and a shift to botryoids in Laurales, or separately from racemes in Magnoliales and from either racemes or botryoids in Laurales (Calycanthaceae).

Thyrses, distinguished from racemes and spikes by the lateral unit character (23; cymes rather than single flowers), appear to be derived from racemes in Hydatellaceae, Chloranthaceae, Aristolochioideae, and Butomus, and from either racemes or botryoids in Siparunaceae and Hernandiaceae. Sessile flowers (24) were derived from pedicellate ones, resulting in spikes in Chloranthaceae and Ceratophyllum (a synapomorphy with the D&E backbone, a convergence with the J/M backbone), the Piperaceae-Saururaceae clade, monocots (separately in Acorus, Araceae, and Aponogeton), and Platanus (modified into heads), and botryoids with sessile flowers (i.e., stachyoids) in Tetracentron. In all these cases, reduction of the pedicel is correlated with general floral reduction. Loss of bracts (25; arrows in Fig. 3) occurred in several lines in which racemes were modified to spikes (Hedyosmum and Ceratophyllum, either once or twice, depending on backbone tree and optimization; Acorus, Araceae, Aponogeton, Platanus) or thyrses of reduced pedicellate flowers (Hydatellaceae) and might also seem correlated with reduced flowers. However, this is not a universal rule because bracts were also lost within Nymphaeaceae, in which flowers are unusually large.

Nymphaeales deserve special attention because interpretation of their inflorescence morphology is both particularly controversial and potentially relevant to ancestral conditions and early trends in angiosperms. 35, 36, 37, 38) described Nymphaeaceae (Nuphar, Nymphaea) as having a unique system of solitary flowers borne in the same phyllotactic spiral as leaves, with no subtending bracts (accepted by 173), which she compared with conditions in ferns. This view was critiqued by 18, who interpreted Nymphaeaceae as having modified racemes, with the apparent position of flowers in the same spiral as leaves due to reduction of the leaf (pherophyll) component of a leaf-bud primordium.

In a phylogenetic context, with Cabombaceae sister to Nymphaeaceae, the interpretation of 18 makes more sense because Cabomba has racemes, with flowers borne in the axils of peltate floating leaves (Brasenia has not been studied in sufficient detail for comparisons). It would also bring Nymphaeaceae in line with the normal shoot organization in angiosperms and other seed plants. Closer examination of inflorescence and floral morphology in Nymphaeaceae supports this view. Nuphar, which is basal in Nymphaeaceae, has a bract near the base of the pedicel, on its abaxial side with respect to the main axis and thus near the position of a subtending bract, and three outer tepals. Nymphaea, however, has no bract on the pedicel and four outer tepals, with the first-formed tepal abaxial relative to the main axis, like the bract in Nuphar. As discussed by 18, this structure might be derived from that in Nuphar either by complete reduction of the subtending bract or by its incorporation into the perianth as the abaxial tepal. The latter hypothesis would explain the change from trimerous to tetramerous organization of the perianth. On the other hand, the earlier development of the abaxial tepal could be a function of the fact that the flower is more developed on the abaxial side at the time the tepals are initiated and somewhat incurved. 36 also homologized the bract in Nuphar with the abaxial tepal in Nymphaea, noting cases in Nymphaea in which this tepal is displaced toward the base of the pedicel, but she argued that the Nuphar condition is derived, as a result of intercalary growth between the abaxial tepal and the rest of the flower. This interpretation is less plausible in terms of outgroup comparison. We have therefore scored both Nuphar and Nymphaeoideae (Nymphaea, Euryale, and Victoria) as having racemes, with the floral subtending bract present in Nuphar but absent in Nymphaeoideae (which could be due either to reduction or to incorporation into the perianth). The condition in Barclaya is unknown, although it appears consistent with that in Nuphar and Nymphaea.

From this perspective, the whole shoot system of Nymphaeaceae can be considered a giant raceme. If the pherophyll-bud primordium is viewed as a complex of two parts (cf. 18), in some cases, the pherophyll part develops into a foliage leaf, and the floral bud is suppressed; in others, the floral bud grows rapidly after initiation, and the pherophyll is reduced to a thin bract or nothing at all. One could also suggest there is competition for space: either the flower or the leaf is reduced, and the other, more precocious part “wins.” This divergence in development may be a function of the gigantism of the shoots, leaves, and flowers of these plants, compared to their outgroups.

Victoria and Euryale may provide indirect support for this interpretation. 7 identified these taxa as the sister group of Nymphaea, but subsequent analyses of more markers (144) indicate they are nested within Nymphaea and are therefore unlikely to represent the ancestral condition in Nymphaeaceae. However, they too can be interpreted in terms of an underlying racemose pattern. In Victoria and Euryale the flowers arise in a Fibonacci spiral. Each flower is associated with a leaf, but it is located not in the middle of the leaf axil but rather toward the inner side, in terms of the direction of the spiral (anodic side; 38; 173). 38 described the leaves and flowers as forming two separate spirals, but an interpretation more consistent with normal angiosperm morphology may be that each flower is in the axil of a foliage leaf but slightly displaced (18). This displacement might be due to the fact that both leaf (petiole) and flower (pedicel) are bulky, so an exact superposition would not allow enough space in the mature state. As in Nymphaea, the abaxial tepal in Victoria develops first; but the fact that Victoria has a subtending leaf as well could be evidence against identification of the abaxial tepal in Nymphaea with the floral subtending bract. Because each pherophyll develops into a leaf and each bud develops into a flower, the number of flowers and leaves in a shoot is the same. In contrast, in Nuphar and Nymphaea only the leaf or only the flower of the pherophyll/flower “complex” develops to maturity, and the numbers of mature flowers and leaves in a shoot are not necessarily equal. If Victoria and Euryale are nested in Nymphaea, in which the subtending leaf is absent, their condition may represent a “reactivation” of the pherophyll portion of the leaf-bud primordium, perhaps related to even more extreme gigantism.

In Hydatellaceae, interpretation of the crowded inflorescences of extremely simple flowers is made difficult by the lack of subtending bracts for the lateral branches. However, 168 tentatively but plausibly interpreted the flowers as forming reduced thyrses.

Based on the inferred phylogenetic relationships, racemes are ancestral in Nymphaeales, either as a synapomorphy or a retention from the first angiosperms. With bracts present in Cabomba and Nuphar, it is most parsimonious to assume that bracts were lost independently on the line to Hydatellaceae and Archaefructus (if these two taxa form a clade) and within Nymphaeaceae (Nymphaea). Archaefructus still had racemes, but these were modified into thyrses in Hydatellaceae, by replacement of single lateral flowers by cymes. One possible adaptive explanation is that the resulting increase in number of flowers compensated for the reduction in number of carpels and ovules, but the small number of cymes in the living group may reflect a later round of reduction.

Floral organization

In recent years most authors have assumed that the first angiosperms had bisexual flowers (26), but because the flowers of Amborella are functionally unisexual the ancestral state is equivocal. With our previous data set, the lineage leading to all other angiosperms could be reconstructed as basically bisexual, but the situation has changed with the addition of Hydatellaceae (and Archaefructus, if it is linked with Hydatellaceae). With the D&E backbone, the state is equivocal up to the basal node of the mesangiosperms, above which the Chloranthaceae-Ceratophyllum line is unisexual and the line leading to the remaining mesangiosperms is bisexual. With the J/M backbone and Amborella basal, the state is equivocal up to the node connecting Nymphaeales and the remaining groups. Rescoring Archaefructus as uncertain (0/1) for this character based on the report of bisexual units by 129 would not modify these inferences.

The view that bisexual flowers were ancestral is supported by the regular presence of one or two sterile stamens in female flowers of Amborella (76; 11). In other words, the flowers are organizationally bisexual. Michael Frohlich, Royal Botanic Gardens, Kew (personal communication) also gave us a likelihood argument in support of the view that the unisexual state in Amborella is derived, namely that Amborella terminates a long branch with no surviving side-branches, whereas the sister branch (including all the remaining angiosperms) is “broken up” by several lineages near its base. This difference in branch length could mean that inference of the initial state is more secure on the latter line than on the longer line leading to Amborella. However, this argument is weakened by the addition of Hydatellaceae (either with or without Archaefructus) to Nymphaeales.

An intriguing case concerns Chloranthaceae, in which Hedyosmum and Ascarina have unisexual flowers, but Sarcandra and Chloranthus have bizarre bisexual flowers consisting of one carpel and one stamen or tripartite androecium (68; 62). Molecular studies and the morphological analysis of 62 agree that Hedyosmum and Ascarina diverged successively below Sarcandra and Chloranthus. In 62, in which Chloranthaceae were nested among bisexual taxa, there were two equally parsimonious scenarios: either bisexual flowers were plesiomorphic for the family and became unisexual independently in Hedyosmum and Ascarina, or flowers became unisexual in the ancestor of the family and reverted to bisexual in Sarcandra and Chloranthus (56). This is still true in trees with the J/M backbone (Fig. 2B). However, in trees with the D&E backbone (Fig. 2A), where Chloranthaceae are linked with Ceratophyllum, which has unisexual flowers, it is most parsimonious to assume that the bisexual flowers of Sarcandra and Chloranthus were derived from unisexual flowers. In “higher” angiosperm groups, we know of no cases where phylogenetic analyses imply that bisexual flowers are derived, but it would be dangerous to assume this was true during the early angiosperm radiation. In any case, it appears that floral sexuality was highly labile in early angiosperms. At least eight reversals from bisexual to unisexual also occurred within magnoliids and basal eudicots: in Lactoris (some flowers), Myristicaceae, Siparunaceae, Mollinedioideae and Monimioideae (once or twice), Lardizabalaceae, Menispermaceae, Platanus, and Buxaceae.

Congenital fusion of all outer floral parts into a hypanthium (27) occurred independently in Amborella (where it could be either ancestral or derived if Amborella alone is basal, but derived if Amborella and Nymphaeales form a clade), Eupomatia, and Laurales, where it is an important synapomorphy of the order. We treated inferior ovary as a state of the same character, on the grounds that it might originate by fusion of either separate outer parts or an existing hypanthium to the ovary. The latter process does appear to have occurred in Laurales, where Gomortega and the clade consisting of Lauraceae (where the ancestral state appears to be inferior, as in Hypodaphnis and other basal genera: 163) and Hernandiaceae are nested within the order. However, there is no phylogenetic evidence for this in the other lines with an inferior ovary (Barclaya plus Nymphaeoideae, Hedyosmum, Saururaceae, Aristolochiaceae, Dioscoreaceae, and Trochodendraceae), whose closest outgroups have no hypanthium.

An elongate receptacle (28), often presented as a primitive feature, appears instead to be an independent advance of Schisandraceae, Magnoliaceae, and Galbulimima. In Magnoliaceae this coincided with origin of cortical vasculature (29) extending from the perianth into the gynoecium, a feature that arose independently in Glaucidium. Cortical vasculature extending only into the androecium arose before an elongate receptacle in the Degeneria-Galbulimima clade and independently in Trochodendron. Protrusion of the floral apex (30), a distinctive feature of Nymphaeoideae and Illicium, is an independent advance in these two groups.

Perianth

Our results indicate that presence of a perianth (31) is ancestral, even with the addition of Archaefructus, which has no perianth (both of its potential extant relatives also lack a perianth). Independent losses occurred in Hydatellaceae (with or without Archaefructus), the Piperaceae-Saururaceae clade, the strange case of Eupomatia (and Galbulimima if its outer petaloid organs are staminodes; we scored perianth as unknown, but it is probably absent), and Euptelea (arrows in Fig. 4A). Loss of the perianth in Ceratophyllum and Chloranthaceae other than Hedyosmum poses more problems. The analyses of 57 and 62 indicated that the presence of three tepals in Hedyosmum was plesiomorphic and their loss a synapomorphy of other Chloranthaceae, and this is still so for the J/M tree. However, if Ceratophyllum is related to Chloranthaceae, as in the D&E tree (Fig. 4A), and if we are correct in interpreting Ceratophyllum as lacking a perianth, it is equivocal whether the perianth of Hedyosmum is a primitive retention or a secondary invention.

As discussed in 76, the ancestral perianth phyllotaxis (32; Fig. 4A) is equivocal: either the spiral state of Amborella and Austrobaileyales is ancestral and the whorled state of Nymphaeales is derived, or vice versa. However, the reconstructed ancestral state in mesangiosperms is unambiguously whorled (see also 218). Cases of spiral perianth in magnoliids, once widely assumed to be primitive, are therefore derived from whorled: in Degeneria in the Magnoliales, and once, twice, or three times in Laurales. In the eudicots, shifts to spiral occurred in Circaeaster, core Ranunculaceae, and Nelumbo. If a spiral perianth originated once at the base of Laurales and was retained into Calycanthaceae, Atherospermataceae, Gomortega, and the monimiaceous genus Hortonia, there was yet another round of reversal, from spiral to whorled, in other Monimiaceae and the Lauraceae-Hernandiaceae clade. As argued by 67, perianth phyllotaxis is therefore a highly labile character, although it is stable over large parts of the tree.

Tracing the evolution of perianth merism (33; Fig. 4B) is potentially confused by the occurrence of many taxa with spiral phyllotaxis, in which merism was scored as unknown but parsimony optimization implicitly treats taxa as having the state of the surrounding groups (cf. 147). Based on those taxa that are whorled, if the ancestral angiosperms had a whorled perianth, it was trimerous. It became polymerous within Nymphaeaceae (specifically tetramerous) and in Hernandiaceae (Gyrocarpoideae, some Hernandioideae), dimerous in Winteraceae. Whether the shift from trimerous to tetramerous perianth in Nymphaeaceae was a result of incorporation of the bract into the perianth, as discussed, would be an intriguing topic for evo-devo studies. Most interesting is the case of eudicots, in which the reconstructed ancestral state is either trimerous, as in most Ranunculales, or dimerous (a possibility first emphasized by 60), as in Papaveraceae, near the base of Ranunculales, and in Proteaceae, Tetracentron (20), and Buxaceae (204), on the line leading to “core” eudicots (Gunneridae, including Pentapetalae, of 14).

The fact that trimery is reconstructed as homologous in Hedyosmum (with three tepals) and other groups (Fig. 4B) might be questioned as an artifact of the Maddison long distance effect (147). With the J/M backbone, where the presence of a perianth in Hedyosmum is reconstructed as ancestral, this poses no problem. However, with the D&E backbone (Fig. 4B), where Chloranthaceae are linked with Ceratophyllum, the perianth of Hedyosmum may be a secondary invention, and if so its trimery would not be strictly homologous with trimery in other taxa. On the other hand, it might be that the reappearance of a perianth in trimerous form was a consequence of reactivation of an existing but suppressed developmental program and therefore homologous at a more fundamental genetic level (cf. 143). This reappearance of trimery could be an intriguing topic for developmental genetic research (cf. “biological homology” of 207, 7). But even if the inferred homology of the trimerous perianth in Hedyosmum is an artifact, it would still be valid to conclude that trimery is ancestral in angiosperms (if they were originally whorled) because this is also inferred if Hedyosmum is deleted.

The ancestral number of perianth whorls (series when spiral) (34) is reconstructed as more than two, and this was retained from the first angiosperms into magnoliids (Fig. 5A). The number of whorls was reduced to two in Cabombaceae, Lauraceae, and (independently or as a synapomorphy with Lauraceae) some Hernandiaceae. With the D&E backbone (Fig. 5A), a shift to two whorls is a conspicuous synapomorphy of monocots and occurred two or three times in eudicots—once in Ranunculaceae, and once or twice in the other branch, depending on whether the numerous series in Nelumbo are ancestral or derived. However, with the J/M backbone, where monocots, Ceratophyllum, and eudicots form a clade, reduction to two whorls may be either an event that occurred three or four times, or a synapomorphy of these groups, with two reversals in eudicots. In Ranunculaceae, reduction to two whorls appears to have been a step toward reduction to one in Hydrastis, and the same may have occurred in Gyrocarpoideae. But one whorl was apparently derived directly from more than two in Hedyosmum (if its perianth is ancestral in Chloranthaceae), the Lactoris-Aristolochiaceae clade, Myristicaceae, and Circaeaster.

The inferred ancestral state of perianth differentiation (35; Fig. 5B) is either all sepaloid tepals, as in Amborella, or outer sepaloid and inner petaloid tepals, the basic state in the line leading to all other angiosperms. If Amborella is linked with Nymphaeales, the differentiated state is unequivocally ancestral. Tepals became all sepaloid in Trimenia and one to three times in Laurales. As with the origin of two perianth whorls, with the D&E trees tepals became all sepaloid in monocots and once or twice in eudicots (Platanus, Proteaceae, Tetracentron, and Buxaceae), depending on whether the differentiated tepals of Nelumbo are ancestral or derived, but with the J/M trees this is a possible synapomorphy of the two clades. Differentiated tepals became all petaloid in Cabombaceae and some Ranunculales. In monocots, the all petaloid state is a synapomorphy of the “core” monocot clade represented by Dioscoreaceae, Nartheciaceae, and Melanthiaceae (Petrosaviidae of 14), apparently derived from all sepaloid. True petals (36) are a synapomorphy of Ranunculales other than Euptelea, with convergent origins in Nuphar and within Asaroideae (Saruma). The adaxial nectar glands (37) on the inner petals of many Ranunculales apparently originated once after origin of petals, after divergence of Papaveraceae, with a convergence in Cabomba.

If core eudicots (Gunneridae of 14) are linked with Buxaceae and/or Trochodendraceae (176), our results support the hypothesis that the typical dicyclic, pentamerous perianth of the gunnerid groups other than Gunnerales (Pentapetalae of 14) was derived from two dimerous whorls of reduced sepal-like organs (contrary to 211). Whether this occurred by increase in the number of parts per whorl or by addition and reorganization of new whorls is beyond the scope of this paper.

At least basal fusion of the outermost perianth parts (38) occurred independently in several lines: Amborella (there is a short zone of fusion among tepals before fusion with the stamens begins), Cabomba (76), Canellales and Aristolochiaceae (either once or twice), Myristicaceae, and Degeneria (Magnoliales). On parsimony grounds, the tepal fusion in Amborella could be ancestral if Amborella is sister to all other angiosperms, but it is derived if Amborella and Nymphaeales form a clade. This fusion is not related to formation of a calyptra (39), apparently derived from one or two floral bracts (64, 56; 131), which intriguingly arose either once or three times in other Magnoliales (Magnoliaceae, Galbulimima, Eupomatia). At the level of angiosperms as a whole, fusion tends to be much more labile in sepals (or outer tepals) than in petals. This phenomenon is reflected by the fact that the contrast of choripetaly vs. sympetaly has been commonly regarded as significant at the macrosystematic level, whereas chorisepaly vs. synsepaly has been relatively neglected. Among basal angiosperms, tepal fusion may be interesting within genera or families, for example in Hedyosmum, where it unites a large derived clade (62). Another interesting feature concerns the fate of tepals after anthesis in the ANITA grade: caducous during or at the end of anthesis (combined with narrow attachment areas) in Austrobaileyales, but persistent (combined with broad attachment areas) in Amborella and Nymphaeales (76). This feature has not been explored throughout basal angiosperms and was therefore not used in our analysis. However, a caducous perianth may be a synapomorphy of Austrobaileyales. The cases of tepal fusion appear to be restricted to clades with persistent tepals.

Androecium

Our results indicate that the single stamen (40) of Hydatellaceae, Ceratophyllum, and most Chloranthaceae is derived and is a synapomorphy of Ceratophyllum and Chloranthaceae with the D&E trees. Our data imply that the presence of a few stamens in some species of Ascarina (scored as 0/1) and the tripartite androecium of Chloranthus (scored as unknown) were derived within Chloranthaceae. This inference is sensitive to the interpretation of the androecium of Chloranthus, which has been variously considered a result of lobation of one stamen or fusion of three (68; 56). If Chloranthus is rescored as having more than one stamen, and if Ceratophyllum is not related to Chloranthaceae (as in the J/M topology), the inferred ancestral state for the family is equivocal. However, if Ceratophyllum is associated with Chloranthaceae, one stamen is reconstructed as ancestral.

Stamen phyllotaxis (41; Fig. 6A) is generally correlated with perianth phyllotaxis, but this correlation breaks down in Magnoliales (76). As with the perianth, the ancestral stamen phyllotaxis is equivocal, and with the D&E topology the basic state for the magnoliid-monocot-eudicot clade (mesangiosperms other than Chloranthaceae and Ceratophyllum, which cannot be scored) is whorled. However, Magnoliaceae have a whorled perianth (except for some probably derived species of Magnolia) but spiral (or somewhat irregularly arranged) stamens. Stamens are also spiral in Eupomatia and Galbulimima (which have no perianth), as are both tepals and stamens in Degeneria. Myristicaceae have three whorled tepals, but their fused stamens vary between spiral and whorled (scored as 0/1). As a result, stamen phyllotaxis in magnoliids appears to have shifted to spiral earlier than perianth phyllotaxis, in the common ancestor of Magnoliales and Laurales, with reversals to whorled in Annonaceae (often becoming chaotic within the androecium: 67) and either once or twice in Mollinedioideae and the Lauraceae-Hernandiaceae clade. With the J/M backbone, because whorled eudicots and monocots are consolidated in a clade and Chloranthaceae are linked with magnoliids, the ancestral state in mesangiosperms is equivocal, and it is possible that the spiral androecium in Magnoliales and Laurales was retained from the first angiosperms, rather than being a reversal or a convergence with Amborella and Austrobaileyales, as we inferred for the perianth. A less consequential discrepancy between perianth and androecium occurs in Nelumbo, where tepals are spiral but stamens are produced chaotically on a ring primordium (109; here scored as unknown).

As with the perianth, a trimerous androecium (42) is predominant and reconstructed as ancestral (if one assumes the androecium was originally whorled). Changes in stamen merism are only sometimes correlated with those in the perianth. The androecium became polymerous before the perianth in Nymphaeaceae (before rather than after divergence of Nuphar) and in Canellales, where the perianth remained trimerous in Canellaceae (the polymerous androecium may be related to connation of the stamens) and became dimerous in Winteraceae. Like the perianth, the ancestral androecium in eudicots may have been either trimerous or dimerous (the latter would be favored if Euptelea is dimerous, as some have suggested, but this is unclear: 121; 161).

The inferred ancestral number of stamen whorls (or series) (43; Fig. 6B) is more than two, as in the perianth. However, in contrast to the situation in the perianth, where more than two whorls were apparently retained from the first angiosperms into magnoliids, with the D&E backbone a shift to two stamen whorls occurred near the base of mesangiosperms (either above or below Chloranthaceae) and reversed to more than two in Magnoliales and Laurales (Fig. 5B). With the J/M backbone both this scenario and persistence of more than two stamen whorls (or series) into magnoliids are equally parsimonious (as was also true for spiral stamen phyllotaxis). With this backbone, origin of two stamen whorls may be a synapomorphy of monocots and eudicots, but it may equally well be homologous with the same state in Piperales and Canellaceae. Other noteworthy changes include increases from two to more than two stamen whorls (or series) in Ranunculaceae, Nelumbo, and Trochodendron, and reductions from more than two to one in Cabomba and Myristicaceae, from more than two to two in Hernandiaceae, and from two to one in Euptelea and Circaeaster.

Production of stamens in double positions (44) evolved independently in Nymphaeales (the condition in Hydatellaceae and Archaefructus is undefined), Aristolochiaceae, Annonaceae, Butomus, and Papaveraceae, and within Winteraceae, Mollinedioideae, and Tofieldiaceae, where both states occur. At the taxonomic level of this analysis, stamen fusion (45) is a separate advance wherever it occurs (Schisandraceae, Canellaceae, Myristicaceae, Eupomatia, and within several other taxa).

The intriguing possibility that inner staminodes (46) might be a primitive feature in angiosperms (65; 41) is not borne out: inferred relationships imply that inner staminodes originated independently in Austrobaileya and in Magnoliales and Laurales. In the latter groups, because we scored Myristicaceae as unknown, since they have highly modified male flowers with a central columnar androecium, it is equally parsimonious to assume that inner staminoides arose once, followed by loss in Magnoliaceae, or twice, in Laurales and the clade consisting of Galbulimima, Degeneria, Eupomatia, and Annonaceae (where inner staminodes are retained in the basal genus Anaxagorea; 145; 172). Glandular food bodies (47) on the inner staminodes are a more secure synapomorphy of the Degeneria-Annonaceae clade.

“Laminar” or “leaflike” stamens have often been considered ancestral in angiosperms (e.g., 13). However, the fact that the sporangia are adaxial in some laminar stamens and abaxial in others suggests that the laminar condition may not be homologous across basal angiosperms. This led 191 to suggest that the ancestral stamen had marginal sporangia, a condition usually associated with a narrow connective. Rather than contrasting laminar and filamentous, we have split stamen morphology into several characters.

One character concerns the stamen base (48; Fig. 7A), with three states: short (either wide or constricted, which often intergrade: 62), long and wide, and long and narrow (= typical filament; see Appendix 1 for limits between states). Stamens with both of the first two states have been described as laminar. With our previous data set (57), the ancestral state was either long and wide (as in Amborella and most Austrobaileyales) or narrow (as in Cabombaceae and Trimenia). This is still true with Recent taxa only and both backbone topologies, because Hydatellaceae also have a long and narrow base, but with the D&E backbone and the addition of Archaefructus, which has a short base, any of the three states may be ancestral. Within Nymphaeales, there is a series from long and narrow to short to long and wide if only living taxa are considered, but if Archaefructus is linked with Hydatellaceae this becomes equivocal. With the D&E backbone, a long and narrow filament is basic for mesangiosperms other than the Chloranthaceae-Ceratophyllum line (most of which have a short base), and a shift to laminar stamens with a short base unites Magnoliales and Laurales. This reversed to long and narrow in the clade consisting of Monimiaceae, Lauraceae, and Hernandiaceae. However, with the J/M backbone, the state at the base of mesangiosperms is entirely unresolved, and the short base of Magnoliales and basal Laurales may be either derived or inherited from lower in the tree.

Paired basal glands (49), a peculiar feature of the stamens of many Laurales, originated either once after divergence of Calycanthaceae, with reversals in Siparunaceae and Mollinedioideae, or twice, in the Atherospermataceae-Gomortega and Monimiaceae-Lauraceae-Hernandiaceae clades. Given the distinctive nature of this advance and the fact that its absence in Siparunaceae and Mollinedioideae is correlated with packing of the stamens in a deep hypanthium, the former scenario may be more likely.

An extended connective apex (50) is also common in laminar stamens. As in 57, this feature is ancestral on most trees, except the J/M trees with Recent taxa only, where the ancestral state is equivocal. If the extended type is ancestral, truncation of the apex occurred an uncertain number of times in four near-basal lines (Hydatellaceae, Cabombaceae, Schisandraceae plus Illicium, and Sarcandra) and in mesangiosperms. With the D&E backbone a truncated apex is reconstructed as basic for the monocot-magnoliid clade, and in all trees it is ancestral in magnoliids. As a result, the extended apex of the classic laminar stamens of Magnoliales is a secondarily derived feature that unites Galbulimima, Degeneria, Eupomatia, and Annonaceae, and in Laurales the same is true for Calycanthaceae. The situation is confused in eudicots: it is equivocal whether the extended apex of Euptelea, Nelumbo, and Buxaceae is ancestral or derived. A peltate apex originated independently in Nuphar and Platanus (and within Annonaceae: 54; 172; see also 76).

Embedded pollen sacs (51) are characteristic of some laminar stamens, as in Degeneria, but they also occur in taxa with filamentous stamens, such as the Piperaceae-Saururaceae clade, and some laminar stamens have protruding pollen sacs, as illustrated most graphically by Austrobaileya. Protruding pollen sacs are inferred to be ancestral, as in Amborella and Austrobaileyales, and embedded pollen sacs were derived several times, often uniting important groups: Nymphaeaceae, Chloranthaceae (reversed in Chloranthus) and Ceratophyllum (if these are related), Piperaceae-Saururaceae, Magnoliales other than Myristicaceae, once or twice in Laurales (in the “atherosperm” clade consisting of Atherospermataceae, Gomortega, and Siparunaceae and the Lauraceae-Hernandiaceae clade), and Trochodendraceae. Reduction to two microsporangia (52) occurred once (with reversals) or more times in Laurales—in the atherosperm clade, Lauraceae (where the ancestral state is unclear: 163), and Hernandiaceae, nearly coinciding with the shift to embedded pollen sacs—and in Circaeaster.

Orientation of dehiscence (microsporangium position) (53) is one of the most homoplastic floral characters (consistency index, C. I. = 0.09), but it shows some noteworthy patterns (Fig. 7B). Problems in scoring of unistaminate flowers are discussed in Appendix 1. In 57, introrse dehiscence (adaxial microsporangia), as in Amborella, Nymphaeaceae, and most Austrobaileyales, was ancestral, but with the addition of Hydatellaceae, which are latrorse, the ancestral orientation is equivocal (introrse or latrorse) in the D&E trees (Fig. 7B). However, with the J/M backbone, introrse is still reconstructed as ancestral, because the latrorse Chloranthaceae are further from the base of the tree. The ancestral state for mesangiosperms is equivocal with all trees, but by the base of the magnoliids dehiscence appears to have become extrorse. These results, together with those on stamen base (48), are therefore consistent with a scenario in which the introrse laminar stamens of the first angiosperms first became more filamentous, and then these stamens were secondarily expanded with a new, abaxial microsporangium position in magnoliids. Reversals to introrse occurred in Magnolioideae, Eupomatia, and Laurales, where introrse is a synapomorphy of groups other than Calycanthaceae (with reversals in Hortonia and Gyrocarpoideae and much plasticity in Lauraceae, often within the same flower). Latrorse is widespread in eudicots, perhaps functionally related to the narrow filament of most groups, but it is equivocal as a synapomorphy of the clade.

Anther dehiscence (54) by longitudinal slits is clearly ancestral. Branching of the ends of the slit, resulting in “H-valvate” dehiscence, occurred in Nuphar, Monimioideae (Laurales), Euptelea, Platanus, within Calycanthoideae (Sinocalycanthus; 182), and as a synapomorphy of Sarcandra and Chloranthus, the Galbulimima-Annonaceae clade (Magnoliales), and Trochodendraceae. In Nymphaeales, Chloranthaceae, and Magnoliales, it appears to have evolved after embedded pollen sacs. This feature is also known from a number of unplaced Cretaceous fossils (91, 5) and the Late Cretaceous calycanthoid flower Jerseyanthus (32). Dehiscence by apically hinged flaps is a distinctive feature of many Laurales, but like basal glands, which have a partially overlapping distribution, its history is equivocal: it may have arisen after divergence of Calycanthaceae, with a reversal within Monimiaceae, or separately in the atherosperm and Lauraceae-Hernandiaceae clades.

In Magnoliales embedded pollen sacs, inner staminodes, food bodies, extended connective apex, and H-dehiscence appear to have evolved in the context of beetle pollination, so it is interesting that some of these advances also occurred in Calycanthaceae, another beetle-pollinated group, as early emphasized by 104.

Gynoecium

We have not recognized separate characters for carpel phyllotaxis and merism because these features are usually correlated with those of the androecium. The most conspicuous deviation from this correlation is presence of one carpel (74), which we contrast with more than one. More than one carpel is reconstructed as ancestral; reduction to one occurred independently in Hydatellaceae, Trimenia, Myristicaceae, Degeneria, the Lauraceae-Hernandiaceae clade, Berberidaceae, Proteaceae, and within several groups. In the D&E trees, reduction to one carpel is an important synapomorphy of Ceratophyllum and Chloranthaceae.

One of the most significant results of the molecular rooting of angiosperms was its implication that the ancestral carpel was not the conduplicate or plicate type of Magnoliales and Winteraceae, similar to a leaf folded down the middle (3), but rather the ascidiate type, which grows up as a cup or tube as the result of a meristematic cross-zone between the primordium margins, and was sealed not by postgenital fusion but by secretion in the resulting canal (76). Earlier, margins of some plicate carpels had been described as unsealed, so that pollen tubes grew to the ovules among stigmatic hairs, but in fact they are sealed by postgenital fusion of the immature epidermises (124). Because these topics were discussed in detail in 57 and 76, we consider them more briefly here, except where the two backbone topologies have different implications. Our conclusions are not affected by addition of Archaefructus, in which we conservatively scored these characters as unknown, because they are often impossible to evaluate in the absence of developmental or anatomical data (74).

In all trees, the ancestral carpel form (75) is ascidiate. With the D&E backbone (Fig. 8A), origin of the plicate carpel is an important synapomorphy of mesangiosperms other than Chloranthaceae and Ceratophyllum, with reversals to ascidiate in Mollinedioideae, Circaeaster, Berberidaceae, and Nelumbo. The intermediate type, with both ascidiate and plicate zones below the stigma and the ovule(s) attached to the ascidiate zone, evolved from ascidiate in Barclaya and Illicium, but also from plicate in Myristicaceae, Laurales other than Calycanthaceae, Acorus, and Euptelea. In contrast, with the J/M backbone (Fig. 8B), with Chloranthaceae and Ceratophyllum nested at different positions in mesangiosperms, scenarios in mesangiosperms are equivocal. Either the ascidiate carpels of Chloranthaceae and Ceratophyllum are primitive and plicate carpels originated separately in eudicots, monocots, and magnoliids, or plicate carpels originated at the base of mesangiosperms and reversed twice to ascidiate in Chloranthaceae and Ceratophyllum. In monocots, the intermediate carpels of Acorus may or may not be evolutionarily intermediate between ascidiate and plicate, and the ascidiate carpels of some Araceae may be either primitive or derived.

As already noted, we split modes of carpel sealing into two characters. Scenarios for origin of postgenital fusion of the carpel margins (76) in mesangiosperms are similar to those for origin of the plicate carpel, differing with the D&E and J/M trees. However, the two characters are not redundant, because complete fusion also arose at the base of Nymphaeaceae, when the carpels were still fully ascidiate, and its history within magnoliids, monocots, and eudicots was different. Laurales other than Calycanthaceae shifted to intermediate carpels but have complete or partial postgenital fusion, and complete fusion may have persisted into Lauraceae and Hernandiaceae. Euptelea and Acorus also have intermediate carpels but complete postgenital fusion. Many monocots (aquatic Alismatales, the Melanthiaceae-Dioscoreales clade) are plicate but have only partial postgenital fusion, and some Ranunculales are plicate but have no or partial postgenital fusion. As a result, with the J/M backbone (Fig. 9A), it is equally parsimonious to assume that lack of postgenital fusion persisted well into both monocots and eudicots, even though they were plicate, or that there were reversals of postgenital fusion within these groups. Secretion in the carpels (77) persisted well into groups with plicate carpels and postgenital fusion, being lost once or twice in Canellales and Piperales; in Degeneria, Eupomatia, Calycanthaceae, Gomortega, and the Lauraceae-Hernandiaceae clade; and an uncertain number of times in eudicots, always either coincident with or subsequent to postgenital fusion. However, secretion was retained in monocots.

A single cell layer of pollen tube transmission tissue (78) originated repeatedly: in Austrobaileyales (once or twice: it is absent in Trimenia), Asaroideae (Piperales), Canellales, the Galbulimima-Annonaceae clade (Magnoliales), Hortonia (Monimiaceae), monocots, and three times in eudicots (Lardizabalaceae, Berberidaceae-Ranunculaceae, and the clade of Proteales, Trochodendraceae, and Buxaceae). However, it seems stable within these groups (except possibly Austrobaileyales). This character needs more study, since the “differentiated” state includes more than one type of cell differentiation. Most significant is a third state, multilayered transmission tissue, which is one of several morphological synapomorphies of Lauraceae and Hernandiaceae.

The most homoplastic character is formation of a style (79; C. I. = 0.05–0.06). The reconstructed ancestral state is either lack of a style (i.e., sessile stigma) or equivocal, depending on the backbone tree, the arrangement of Amborella and Nymphaeales, and addition of Archaefructus, which has a style (see Table 1). The basic state in mesangiosperms is presence of a style with the D&M backbone, followed by many losses (e.g., one or two in Magnoliales), but equivocal with the J/M backbone—a style may have originated once or many times. However, in some clades a style seems to have persisted once formed, as in most of the Laurales, Alismatales other than Araceae, the Melanthiaceae-Dioscoreales clade (petrosaviids), and the Buxaceae-Trochodendraceae clade (probably including gunnerids). Stigma extension (80) is less homoplastic (Fig. 9B): a stigma extending more than halfway down the style-stigma zone is reconstructed as ancestral, and it did not become restricted to the apex until within mesangiosperms. In the J/M trees, where Chloranthaceae (with an extended stigma) are linked with magnoliids, this occurred four times, in the Magnoliales-Laurales clade, monocots, Ranunculales other than Euptelea (Papaveraceae scored as unknown because the stigma is modified by syncarpy; reversed in Berberidaceae), and Proteaceae; but in the D&E trees (Fig. 9B), where Chloranthaceae are basal in mesangiosperms, restriction of the stigma may or may not be homologous in monocots and magnoliids, with a reversal in the Piperales-Canellales clade. In any case, the long stigmatic crest of Degeneria, often interpreted as a primitive feature, appears to be a reversal.

Stigmatic protuberances (81), found in Amborella, may be ancestral if Amborella is basal in angiosperms, but not if Amborella and Nymphaeales form a clade. Other occurrences originated independently, in Trimenia, Chloranthaceae (with a loss in Sarcandra), Idiospermum, and Hydrastis plus Glaucidium. Stigmatic papillae (82) with either a pluricellular (Amborella, Hydatellaceae, Barclaya, Nymphaeoideae) or unicellular emergent portion may be ancestral, but the unicellular state was established in the common ancestor of Austrobaileyales and mesangiosperms, followed by scattered origins of pluricellular papillae in Trimenia, Asaroideae, Degeneria, Eupomatia, and Butomus, and losses of papillae in Ceratophyllum, Sarcandra plus Chloranthus, Berberidaceae, and Hydrastis.

Formation of an extragynoecial compitum (83), where contact between stigmas allows pollen tubes to grow to more than one carpel, appears to be ancestral in angiosperms and was retained through Austrobaileyales, with a loss in Cabombaceae. With the D&E backbone (Fig. 10A), this feature is lost near the base of the mesangiosperms (the state in Chloranthaceae is undefined because they have only one carpel), followed by reappearances in Magnoliales and Laurales (once with a loss in Magnoliaceae, or independently in the Galbulimima-Annonaceae clade and in Laurales) and (once or twice) in Lardizabalaceae and Menispermaceae. With the J/M backbone, both this scenario and one in which an extragynoecial compitum persisted into Magnoliales and Laurales, with parallel losses in the monocot-eudicot and Piperales-Canellales clades, are equally parsimonious. Actual fusion of carpels (84; Fig. 10B) occurred several times by two different routes. Eusyncarpy, with carpels fused at the center of the gynoecium, often resulting in axile placentation, evolved independently in Nymphaeaceae, Aristolochiaceae, monocots, and the Trochodendraceae-Buxaceae (and gunnerid) clade. The topology in monocots implies that the free carpels of Alismatales other than Araceae are not primitive but rather secondarily derived from united carpels, as concluded by 19. However, the situation in the Piperales-Canellales clade is confused: Aristolochiaceae are eusyncarpous (or paracarpous in some presumably derived Aristolochia species), but Piperaceae-Saururaceae, Canellaceae, and Takhtajania in the Winteraceae (76) are paracarpous, with carpels fused into a unilocular ovary with parietal placentation, and other Winteraceae and Lactoris are apocarpous. One scenario is that apocarpy was ancestral; paracarpy originated independently in the Piperaceae-Saururaceae clade, Takhtajania, and Canellaceae; and eusyncarpy evolved from apocarpy in Aristolochiaceae. In other scenarios, paracarpy was ancestral and reversed to apocarpy in Lactoris and Winteraceae. In any case, paracarpy evolved independently in Papaveraceae.

Several minor modifications involve the carpel surface. Intrusive oil cells (85) visible at the carpel surface (scored only in taxa with mesophyll oil cells) originated independently in Austrobaileyales (Schisandraceae, Illicium, and possibly Trimenia, which is mixed), Sarcandra plus Chloranthus, the Piperaceae-Saururaceae clade, and once or twice in Monimiaceae (Hortonia and Mollinedioideae). Long unicellular hairs (86) on and/or between the carpels may be a synapomorphy of Laurales, but this is equivocal because they are absent in Gomortega and Siparunaceae; similar hairs also evolved in Hydrastis and within several taxa. Short curved hairs (87) with a long apical cell (71) are found in Amborella, Nymphaeales (except Nuphar), and sometimes Trimenia and may be ancestral in angiosperms. Abaxial nectaries (88) on the backs of the carpels are a synapomorphy of Buxaceae and Trochodendraceae. A widespread feature in monocots is septal nectaries (89) between the fused carpels, seen in our data set in Dioscoreaceae and some Nartheciaceae. It has been suggested that lateral nectaries on the free carpels of some Alismatales (represented by Tofieldiaceae and Butomus) may be homologous (39; 123), and we scored them as the same state, but they originate independently on the trees.

Consideration of the last two characters and others treated earlier indicates that nectaries originated twice on the adaxial side of the inner perianth parts (37), in Cabomba and Ranunculales; once or twice on the stamen bases (49) in Laurales; once on the backs of the carpels (88) within eudicots; and two or more times on the sides of the carpels (89) in monocots. Other types that we did not include because they differ morphologically and appear to be autapomorphic are abaxial nectaries on the petals of Nuphar (118; 76) and disc-like nectaries of uncertain morphological nature in Proteaceae (44) and Sabiaceae (not included in this data set; 166). These results imply that nectar secretion itself arose independently at these different sites: even if all cases of nectar secretion were treated as a state of one character, none of the different types of nectaries would be inferred to be homologous.

In our previous analysis (57), the ancestral ovule number (90) was equivocal: either one, as in Amborella, Trimenia, Illicium, and Chloranthaceae, or more than two, as in most core Nymphaeales and Austrobaileya. Now, because of the addition of Hydatellaceae, which are uniovulate, our data imply that one ovule is ancestral when only Recent taxa are considered (Fig. 11A) and that this number was retained up to Trimenia and Illicium in the Austrobaileyales and to Chloranthaceae (plus Ceratophyllum in D&E trees). However, if Archaefructus, which has several ovules per carpel, is linked with Hydatellaceae, the ancestral state is still equivocal (Fig. 11B). Scenarios in mesangiosperms vary with the backbone tree. With the D&E trees (Fig. 11), where Ceratophyllum and Chloranthaceae form a clade at the base of the mesangiosperms, the basic ovule number in the remaining mesangiosperms is entirely equivocal, and the uniovulate condition in magnoliid groups such as Myristicaceae, Galbulimima, and Laurales (other than Calycanthaceae, which have two ovules) may be either a retention from the first angiosperms or the result of reduction. However, the basic number in the Piperales-Canellales clade and in monocots is more than two, and the basic number in eudicots is either two or more. In contrast, with the J/M backbone, where Ceratophyllum and Chloranthaceae are nested at different points within the mesangiosperms, the uniovulate condition is retained from the base of the angiosperms to these groups (unless Archaefructus is included, in which case the uniovulate condition in Ceratophyllum may be either primitive or secondary) and into Magnoliales and Laurales. This scenario would imply there were increases in ovule number in the Piperales-Canellales clade, Magnoliaceae, the Galbulimima-Annonaceae clade, Calycanthaceae, monocots, and eudicots. Secondary reduction to one ovule occurred in Piperaceae and Nelumbo.

Laminar placentation (91; including “dorsal” in Brasenia, actually on the carpel midrib), often noted as a similarity of Nymphaeales and Alismatales, was independently derived from marginal placentation in Nymphaeales (the state in Hydatellaceae is unknown, because the flowers lack orientation marks) and Butomus (and presumably related Alismatales). The ancestral ovule direction (92; Fig. 12) is reconstructed as pendent, both in basal groups with one apical ovule, such as Amborella and Hydatellaceae, and in multiovulate Nymphaeales. Austrobaileyales show one or two shifts to horizontal, in Austrobaileya and Schisandraceae, and one to ascendent, in Illicium. With the D&E tree, pendent persists to Chloranthaceae and into basal eudicots, but in the monocot-magnoliid clade ovule direction may either remain pendent or shift to ascendent. With the J/M tree, because Chloranthaceae and Ceratophyllum are nested in mesangiosperms, the pendent state unambiguously persists up to Acorus in the monocots. With both trees, a shift to the ascendent state occurs within Ranunculales (Menispermaceae, Berberidaceae, and Ranunculaceae). The basic state in monocots other than Acorus is ascendent; this may be either ancestral for monocots (D&E) or derived from pendent (J/M). The ancestral state in magnoliids is entirely unresolved, but horizontal is basic in Piperales-Canellales and Magnoliales other than Myristicaceae (which have one basal, ascendent ovule). Ascendent is ancestral in Laurales, reverting to pendent in Gomortega and in the clade consisting of Monimiaceae, Lauraceae, and Hernandiaceae.

Inferences on the ancestral ovule curvature (93) depend on interpretation of Amborella, described as anatropous by 2, orthotropous by 76 and 215, and hemianatropous (= hemitropous) by 196. Based on the illustrations of both 76 and 196, the funicle is attached to one side of the base of the ovule, not halfway along the side of the ovule, as in the typical hemitropous condition. The asymmetry of the ovule base may be a consequence of the “apical” position of the ovule, which is actually attached to the adaxial cross zone; in our experience, a truly symmetrical base is restricted to taxa with either basal ovules or apical ovules in a spacious locule (e.g., Acorus; 10). Because the Amborella condition seems closer to typical orthotropous than to anatropous, we have included it in the orthotropous state. With this scoring, the ancestral ovule curvature is equivocal if Amborella is sister to all other angiosperms. However, the fact that the outer integument is asymmetric during development in both Amborella and Chloranthus (215) suggests that orthotropous was derived from anatropous (76). This is the most parsimonious hypothesis if Amborella and Nymphaeales form a clade. It is also favored if the bitegmic ovule is homologous with the cupule of Caytonia (100; 183; 45), as indicated by some cladistic analyses of seed plants (30; 55; 51, 53; 119). Orthotropous ovules evolved several other times, sometimes uniting clades: in Barclaya (Nymphaeaceae), Chloranthaceae and Ceratophyllum (a clade in D&E trees), Piperaceae and Saururaceae, Gomortega, Acorus, Circaeaster, and the Platanus-Proteaceae clade.

Summary of character discussion

Our results (Table 1) imply that the ancestral angiosperm flower had more than two whorls (or series) of tepals, more than two whorls (series) of stamens, probably with adaxial microsporangia (introrse), and several ascidiate carpels that were sealed by secretion rather than postgenital fusion, most likely with one pendent bitegmic ovule, which was probably anatropous. These flowers were borne either in racemes (which some authors might call shoots with axillary solitary flowers) or in botryoids. Perianth and androecium phyllotaxis is uncertain, but if parts were whorled they were trimerous. The most striking uncertainty is whether the ancestral flower was bisexual or unisexual. This is an area where comparative studies on the genetic control of development and better understanding of fossil diversity could be most interesting. The clearest effect of considering Archaefructus concerns ovule number—whereas analysis of Recent taxa alone implies that one ovule was ancestral, the ancestral state becomes ambiguous (either one or more than two) if Archaefructus is linked with Hydatellaceae.

The “explosive” radiation of angiosperms appears to have begun with the origin of the mesangiosperm clade, after the origin of crown group angiosperms and divergence of the more basal ANITA lines (cf. 153). It should be noted that the beginning of this radiation, corresponding to the initial splitting of the main mesangiosperm lines, must have predated the radiation of angiosperms observed in the latter half of the Early Cretaceous fossil record (Barremian through Albian), which involved diversification within the magnoliid, eudicot, and monocot clades (59; 122; 49; 95; 58). There is no unequivocal floral synapomorphy at this point in the tree that might be interpreted as a key innovation responsible for this radiation (cf. 14). Most mesangiosperms differ from the more basal lines in having plicate rather than ascidiate carpels, sealed by postgenital fusion rather than secretion. However, Chloranthaceae have ascidiate carpels that are notably similar to those of more basal groups (68, 49; 76). If the 57 arrangement based on combined molecular and morphological data is correct and Chloranthaceae (with or without Ceratophyllum) are basal in mesangiosperms, plicate carpels originated at the next node; if Chloranthaceae are nested within mesangiosperms (128; 153), plicate carpels may have originated either at the base of mesangiosperms (but soon reversed) or several times within the clade. Furthermore, the fact that Chloranthaceae are one of the most prominent recognizable groups in the Early Cretaceous fossil record (94, 5; 62; 85) suggests they were part of any accelerated radiation. An apomorphy more closely tied to the base of the mesangiosperms is origin of the typical eight-nucleate female gametophyte and resultant formation of triploid rather than diploid endosperm (88, 8; 87; 167), but the history of this character is ambiguous because Amborella has a nine-nucleate female gametophyte (86).

The fact that molecular data firmly nest Magnoliidae (in the restricted monophyletic sense of 14) well within the angiosperms calls into question the traditional use of magnoliid groups such as Magnoliales and Winteraceae as models for the original angiosperm flower (e.g., 34; 191). However, flowers of these groups are more like our present reconstruction of the ancestral flower than some that were being discussed before the molecular rooting—e.g., the simple flowers of Chloranthaceae, suggested as one of several alternative prototypes by 66 and found to be basal in some morphological cladistic analyses (154; 117). According to our analysis, many putatively “primitive” magnoliid features were indeed retained from the first angiosperms, such as more than two whorls (or series) of perianth parts, several free carpels, and probably bisexuality. However, possession of more than two whorls (or series) of stamens may be a secondary increase from an intermediate stage with two whorls near the base of the mesangiosperms. The laminar form of many magnoliid stamens may also be a reversal from more filamentous stamens in earlier mesangiosperms, and the abaxial position of the microsporangia in many magnoliids, which may reflect this secondary expansion, is more definitely derived. Whether the first angiosperms had spiral or whorled floral phyllotaxis, the spiral perianth of some magnoliids appears to be derived from a whorled perianth lower in the mesangiosperms, and the same may be true for the androecium. Most notably, the plicate carpels often illustrated in textbooks as primitive are instead derived, as is the elongate “strobilar” receptacle of Magnoliaceae. Evidence is also strong that the solitary flowers of many magnoliids are derived. Many of these derived features may be related to a general increase in flower size and specialization for beetle pollination, like the inner staminodes of many Magnoliales and Laurales.

Floral groundplan of simple floral structure in near-basal angiosperms

The drastically simple reproductive structures of several extant basal angiosperms and Early Cretaceous fossils have provoked much recent discusion. In Ceratophyllum, flowers are unisexual; female flowers are unicarpellate, with an ascidiate carpel. The structures commonly interpreted as multistaminate male flowers (70; 127) are more likely inflorescences of unistaminate flowers without subtending bracts (73), as we argued earlier. This interpretation becomes even more plausible if a relationship of Ceratophyllum and Chloranthaceae is envisaged (61; 149; 160), as supported by our analysis. The male structures of Hedyosmum are also inflorescences of unistaminate, perianthless flowers without subtending bracts (67). Although 140 interpreted these male shoots as flowers, the inflorescence interpretation is more plausible based on comparison with male inflorescences in Ascarina, in which each stamen (or two to three stamens) is (are) subtended by a small subtending bract. Male shoots similar to those in Hedyosmum are also known from the Early Cretaceous (94, 5).

Hydatellaceae and Archaefructus (186) exhibit a similar pattern of very simple flowers. In Hydatellaceae, several stamens are commonly surrounded by a number of ascidiate carpels, and these by two or more bracts (168). The simplest interpretation is that the flowers are unisexual, unicarpellate or unistaminate, and perianthless (106, 59, 108; 168). In Archaefructus, we interpret the flowers as unisexual, 1–2-carpellate or usually 2-staminate, and perianthless (93). Some authors have suggested that such structures may represent a prefloral state (186; 90) or may be a result of secondary dissolution of the flower–inflorescence boundary due to loss of floral identity (168). 186 speculated that the fertile structures of Archaefructus might represent an evolutionary stage at which the genetic programs for flower and inflorescence formation were not yet strictly separated.

Our inferences on inflorescence, perianth, and androecium evolution do not support the suggestion that these taxa represent a prefloral state. Although developmental studies show that a perianth is completely missing in Hydatellaceae, Ceratophyllum, and most Chloranthaceae (except female flowers of Hedyosmum), and not even present as rudiments in early developmental stages (68, 70; 134; 127; 168), our results imply that the simple floral structure of these groups and Archaefructus is a result of reduction (i.e., decrease in organ number, loss of a perianth, and probably loss of bisexuality) of more “complete” ancestral flowers with a perianth and several stamens and carpels. On parsimony grounds, it is equivocal whether the ancestral flower was bisexual, but we suspect it was, for the reasons discussed.

On phylogenetic grounds, it is less easy to eliminate the hypothesis that floral reduction in these groups occurred not by gradual loss of parts but by loss of floral identity, one of three possibilities discussed by 168. However, if Archaefructus is related to either Hydatellaceae or Ceratophyllum, the fact that its flowers had usually two stamens and one or two carpels may support a reduction scenario. Loss of floral organs may have been easier in more basal angiosperms because they had less floral synorganization than more derived clades. However, we see no reason to think that this involved loss of the basic floral program. A more conservative hypothesis is that the basic floral program was still present but the flowers became simpler by loss and reduction of organs, and selection for maintenance of sufficient reproductive output favored production of more flowers per individual or per inflorescence.

This view is supported by recent molecular developmental studies on Amborella and Nymphaeales, which suggest that the floral genetic program originally found in Arabidopsis and Antirrhinum (21) is present with similar function in these basalmost extant angiosperm lines. Furthermore, the program does not differ profoundly between Amborella and Nymphaeales, in which Nuphar has been studied (181). The same is true for Illicium, another member of the ANITA grade, and some magnoliids (Asimina, Eupomatia, Liriodendron, Magnolia, Persea: 131, 132; 15; 181, 7; 9). In Chloranthus class A genes (responsible for perianth identity in Arabidopsis and Antirrhinum) are present, although its flowers do not have a perianth (143). It appears that one of the mechanisms for the early evolutionary modification and elaboration of the floral developmental program was repeated events of gene duplication and sub- or neofunctionalization (126; 216; 125; 135). This is certainly a promising track to follow in evo-devo studies.

Extant angiosperms that flower under water and peculiarities of their reproductive structures

Because Hydatellaceae, Archaefructus, and Ceratophyllum are or were water plants, it is useful to consider if and how the scenarios developed here may be functionally related to a shift from a terrestrial to an aquatic habitat. For this purpose, we cursorily surveyed reproductive morphology in all angiosperm families comprising taxa with underwater flowers. Comparisons with these taxa suggest that floral simplicity comparable to that in more basal aquatic groups has frequently arisen by reduction. We emphasize that we are not using these correlations as evidence for one or another concept of relationships or direction of evolution; rather, we present them as possible biological explanations for the changes inferred from phylogenetic trees. However, it would be fair to take functional correlations among reduced floral features as evidence that these characters have less individual weight than many others and a reason to exercise caution in accepting relationships based on them. Physical conditions are very different in the water and in the air, and the evolutionary transition from the air to the water has a profound impact on plant structure and flower biology (174; 203; 202). A morphological interpretation of submerged flowers should take into account these differences.

Among angiosperms systematic surveys show many instances of evolution from the land to the water (28), and shifts from pollination in the air to under the water (or on its surface) also occurred a number of times, whereas there are no clear reversals. Based on our survey, there is clearly a general tendency for underwater-flowering plants to evolve simple, unisexual, perianthless, and unistaminate or unicarpellate flowers. This reduction trend is especially conspicuous in near-basal monocots (Alismatales), but it is also recognizable (usually in less extreme form) in other angiosperms. In Alismatales there is a broad diversity of plants with underwater flowering, diverse modes of water-surface flowering, and flowering in the air. Phylogenetic analyses show that the taxa with submerged flowers are derived from taxa with aerial flowers. That there is really an evolutionary trend is shown by the fact that underwater pollination did not evolve just once in Alismatales but several times (141; 19).

Another possibility to consider is that the underwater-flowering habit may have often evolved from wind pollination. A number of water plants have emergent inflorescences and wind-pollinated flowers (24). Wind pollination and underwater pollination can lead to similar reductions in floral structure (see discussion in 69). Thus features adapted to wind pollination may have been preadaptations for underwater pollination. This evolutionary pathway is consistent with the fact that there are genera in which wind pollination and underwater pollination co-occur, such as Groenlandia (Potamogetonaceae) (105) and Callitriche (Plantaginaceae) (156). However, in general, floral reduction is even stronger in water-pollinated than in wind-pollinated flowers. This scenario is a real possibility for Ceratophyllum if it is related to Chloranthaceae, which appear to be ancestrally wind-pollinated, as in Hedyosmum and Ascarina (68).

In Alismatales, there are eight families in which some or all members have submerged flowers (Cymodoceaceae, Hydrocharitaceae, Juncaginaceae, Posidoniaceae, Potamogetonaceae, Ruppiaceae, Zannichelliaceae, and Zosteraceae). These families are representatives of both major aquatic clades in the order (16; 142). Of these eight families, five have unisexual flowers. Six have perianthless flowers, some with a spathe-like envelope (vs. two with a single whorl of perianth organs). It is uncertain whether this envelope is a modified perianth or a bract. Five have unistaminate or bistaminate male flowers (vs. three with three or more stamens). Seven have at least partly unicarpellate or bicarpellate female flowers (vs. one with three or more carpels) (43; 197, 198; 27; 110, 111, 114, 116; 136; 123).

In contrast, there are seven families in Alismatales without submerged flowers (Alismataceae, Aponogetonaceae, Araceae, Butomaceae, Limnocharitaceae, Scheuchzeriaceae, Tofieldiaceae). In all of these, the flowers are either always or sometimes bisexual. All have a perianth of two whorls, six or more stamens, and at least three carpels, except for derived groups of Araceae (130; 26; 112, 113, 115; 151; 201; 123).

A family of Alismatales that particularly well illustrates the evolutionary flexibility of flowers is Juncaginaceae. It shows not only an evolutionary transition from aerial to submerged flowers and correlated morphological changes but also conspicuous lability of flower forms in the group with submerged flowers. Some Juncaginaceae have spikes of aerial flowers (e.g., Triglochin), which are trimerous and bisexual, with two whorls of tepals. Lilaea, however, is a submerged water plant, also with spikes. Phylogenetic analyses nest Lilaea within the family (141; 19; 206), implying that its submerged habit is derived. There are five different floral morphs within one species, all very simple, including from the top downward in a spike (157): (1) male flowers, unistaminate, associated with a bract-like organ, which could be a floral subtending bract or a tepal (interpreted as a tepal by 157); (2) bisexual flowers, unistaminate and with a single carpel, associated with a bract-like organ; (3) female flowers of a single short-styled carpel, associated with a bract-like organ; (4) female flowers of one short-styled carpel with no bract-like organ; and (5) female flowers of one very long-styled carpel with no bract-like organ. Whether the bract-like organ is a floral subtending bract or a perianth part, the flowers are extremely simple in terms of organ numbers.

Thus, although more surveys from a phylogenetic perspective would be desirable, Alismatales can serve as a model group to show functional changes in reproductive structures correlated with the transition from aerial to submerged flowers. Similar patterns can be seen in other groups with completely or partly submerged flowers. In Lamiales, Callitriche (Plantaginaceae) has unisexual, perianthless flowers, male flowers unistaminate, female flowers with two carpels; Hydrostachys (Hydrostachyaceae) has unisexual, perianthless flowers, male flowers uni- or bistaminate, female flowers with two carpels (82, 8). The two genera have the most reduced flowers in Lamiales, which usually have bisexual flowers with a perianth of two whorls and at least two stamens. In Malpighiales, some Podostemaceae have submerged (cleistogamous) flowers, many perianthless and uni- or bistaminate (29); some species of Bergia and Elatine (Elatinaceae) have submerged, very small, di- or trimerous flowers (25). In Myrtales, some species of Rotala (Lythraceae) have submerged flowers, some apetalous, some unistaminate (23). In all these groups, reduction (or loss) of the perianth and reduction of stamens are obvious from outgroup comparison. However, the tendency for reduction in carpel number inferred in Alismatales is not evident in the eudicot examples, perhaps because of more intimate synorganization of the carpels.

In addition to reduction of the perianth, another apparent trend in water plants is reduction or loss of the floral subtending bract. In underwater-flowering Alismatales, for example, the bract is often absent in female flowers (and rarely in male flowers) of Najas (Hydrocharitaceae; 110) and in some flowers of Lilaea (Juncaginaceae) (157) (discussed earlier). However, this trend is also seen in taxa with pollination above the water, including not only Acorus and various Alismatales (Aponogeton, Araceae, Juncaginaceae, Potamogetonaceae), but also some Nymphaeaceae (Nuphar, Nymphaea). This observation is of interest in view of the fact that the putative flowers of Archaefructus do not have a subtending bract (186; 93). An ecological explanation for this trend is that in water plants, whether they flower in the air or in the water, the flowers begin their development in the water and therefore do not need protection against desiccation. In nonwater plants the floral subtending bract provides such protection for the delicate young floral organs before the outer perianth organs are differentiated enough to take over this function (69).

Groups with submerged flowers often have bracts below the whole inflorescence, even when they lack floral subtending bracts, as in Hydatellaceae and some Alismatales. The apparent absence of inflorescence bracts in Archaefructus (186; 93) might be considered evidence against the hypothesis that it was adapted to an underwater flowering habit. However, when there are no floral subtending bracts that individually protect youngest floral stages, inflorescences may have some protection by more basal leaves. In Archaefructus, inflorescences were enclosed by leaves in bud, as shown in the type specimen of A. eoflora (Figs. 2, 26 in 129), which has younger stages of reproductive parts than A. sinensis in 186. In Cabombaceae, young inflorescences are enclosed by regular leaves below the inflorescence and floral subtending leaves, and the same is true of some Alismatales. In the nonaquatic family Chloranthaceae, the inflorescences are enclosed in early development by the fused stipules of adjacent leaf pairs, which form a sheathing structure. In many species of Hedyosmum, the unistaminate male flowers, which lack a subtending bract, are protected by a massive sterile apical part (68, 77).

Concerning vegetative parts, submerged leaves of water plants tend to be either entire and linear or dissected with linear lobes; thus the presence of linear parts is characteristic. Both modes occur, e.g., in aquatic species of Ranunculus (R. reptans entire and linear, R. fluitans dissected with linear lobes), and also in Nymphaeales (Hydatellaceae entire and linear, Cabomba dissected with linear lobes, in addition to peltate, floating leaves). Archaefructus is also dissected with linear lobes.

These observations, together with the evidence from our phylogenetic analysis that Archaefructus may be related to Hydatellaceae or Ceratophyllum, strengthen the view that its simple flowers are the result of reduction in an aquatic habitat. However, the correlations among floral features and the fact that our results were sensitive to assumptions concerning the position of Ceratophyllum underline the dangers of associating it with a particular extant group. More secure conclusions on its affinities may require recognition of other fossils that link it with one rather than another living taxon. Whether Archaefructus affects ideas about the first flower, it reveals important early trends in floral evolution and the early ecological radiation of angiosperms (cf. 85). Especially if it is related to the Albian genera Vitiphyllum and Caspiocarpus (93), which had similar but less finely dissected leaves, it represents an important trend for invasion of Early Cretaceous aquatic ecosystems by angiosperms (cf. 148), represented today only by Hydatellaceae, core Nymphaeales, and probably Ceratophyllum.