Many earlier discussions of the origin of angiosperms concluded that angiosperms were “derived from” seed ferns (e.g., Long 1966, Cronquist 1968, 1988, Takhtajan 1969). From a cladistic viewpoint, such statements are not very informative, since all phylogenetic analyses of living and fossil seed plants have indicated that “seed ferns” are a paraphyletic grade made up of lines that are basal to more derived groups (Crane 1985, Doyle and Donoghue 1986, Doyle et al. 1994, Nixon et al. 1994, Rothwell and Serbet 1994, Doyle 1996). By definition, the ancestor of a clade is not a paraphyletic group but a single species, which might be impossible to recognize as ancestral in the fossil record. However, taxa traditionally called seed ferns could still be the closest relatives of angiosperms, and these could say almost as much as a direct ancestor, by revealing the order of evolution of the various new features of the extant clade (the crown group) and more plesiomorphic homologs of its characteristic structures. This is especially true if we can recognize a series of successive outgroups, which may allow us to distinguish character states that existed on the stem lineage leading to the crown group from autapomorphies that arose in extinct side lines.

Whether or not seed ferns include the closest relatives of angiosperms, they may be important for understanding the origin of angiosperms and their distinctive features. Many authors have argued that the ovulate structures of other groups were too derived to be prototypes for the carpel, usually interpreted as a folded leaf bearing ovules on its adaxial side. For example, Bennettitales have been associated with angiosperms because they had flower-like structures (Arber and Parkin 1907) but rejected as ancestors because their ovules were borne directly on an ovuliferous receptacle, intermixed with interseminal scales, rather than on a leaflike structure. Similarly, other groups have been excluded as angiosperm ancestors because they are too advanced in their wood anatomy or other features (Bailey 1944, 1949, Cronquist 1968, 1988, Takhtajan 1969). Taxa with more derived sporophylls could still be the closest relatives of angiosperms, and they might have much to say about the origin of other features. However, it would be necessary to look lower in the phylogenetic tree for plants with leaflike sporophylls that might be transformed into a carpel, and these might be called seed ferns. This point was recognized by Arber and Parkin (1907), who argued that angiosperms were related to Bennettitales and Gnetales but all three groups came from a common ancestor with pinnate megasporophylls, which they hypothesized was derived from some group of seed ferns (cf. also Takhtajan 1969). The question becomes which group or groups of plants traditionally called seed ferns are related to angiosperms and how.

Understanding the homologies of the carpel also requires consideration of the angiosperm ovule, which usually differs from the ovules of other seed plants in having two integuments (bitegmic) and being bent back on itself (anatropous). The inner integument is presumably homologous with the single integument of other seed plants, but what is the outer integument? This problem is illustrated by the “Mesozoic seed fern” Caytonia, first described from the Yorkshire Jurassic by Thomas (1925). The ovulate structures of Caytonia consisted of an axis bearing two rows of fleshy cupules, each of which contained several ovules. Thomas (1925) compared these cupules with angiosperm carpels. However, this idea soon fell out of favor. First, Harris (1940) found that pollen got inside the cupule, to the micropyles of the ovules—the plant was functionally gymnospermous. This would not rule out the hypothesis that Caytonia was related to angiosperms but more primitive. However, other aspects of the morphology of Caytonia are inconsistent with a cupule-carpel homology. First, angiosperm carpels are thought to be modified leaves borne on a stem, whereas Caytonia cupules were borne in two rows on a dorsiventral axis, like leaflets on the rachis of a compound leaf. Second, as emphasized by Bailey and Swamy (1951), the cupules of Caytonia were enrolled circinately, from tip to base, whereas supposedly primitive carpels are folded lengthwise (conduplicate, or plicate). Finally, the ovules of Caytonia had only one integument, not two.

Similar problems affect Long's (1966) derivation of the carpel from the lobate, dichotomously organized cupule of Carboniferous seed ferns. On recognizing the problem of the bitegmic ovule, Long was forced to postulate that the second integument arose de novo as an outgrowth of the first. Meeuse and Bouman (1974) tried to circumvent the problem by homologizing the inner integument with the wall of the lagenostome (distal part of the nucellus) in early seed ferns. However, as recognized by Meeuse and Bouman, the inner integument of angiosperms develops from a ring of meristematic tissue (cf. Robinson-Beers et al. 1992, Umeda et al. 1994), whereas the lagenostome wall represents the epidermis of the nucellar apex, which separated from the central tissue of the apex to form a pollen chamber (Sporne 1965, Stewart and Rothwell 1993).

Such arguments led some to conclude that Caytonia was not related to angiosperms (e.g., Bailey 1949, Harris 1951, Cronquist 1968). However, there is another way to formulate homologies that might salvage the Caytonia-angiosperm relationship and solve the problem of the angiosperm ovule at the same time, first proposed by Gaussen (1946) and later supported by Stebbins (1974) and Doyle (1978; Fig. 1). Under this hypothesis the cupules of Caytonia correspond not to carpels, but rather to bitegmic ovules. The only change needed in the cupule would be reduction of the ovule number to one. The cupule wall would thus become the outer integument, and because of the circinate character of the cupule the bitegmic ovule would already be anatropous. However, this leaves a problem in explaining the carpel. In terms of positional relationships, the carpel should correspond to the Caytonia rachis, but this was narrow and not very leaflike. In Doyle (1978) I argued that the rachis was probably larger relative to the cupules early in ontogeny, so that it could be transformed into a carpel by modification of development at an early stage.

Stebbins (1974) preferred to compare angiosperms with another group, the Permian glossopterids of Gondwana, and this idea was adopted by Retallack and Dilcher (1981; Fig. 2). Glossopterids also had ovule-bearing structures that have been called cupules, but these were more leaflike—they were described as megasporophylls by Gould and Delevoryas (1977) and Taylor and Taylor (1992)—and did not enclose the ovules so completely. Either one or several of these cupules were borne on the adaxial side of a normal leaf. The resulting structure has been variously called a fertiliger (Schopf 1976), a bract-sporophyll complex (Doyle 1996), or a bract-cupule complex (Doyle 1998a); in this article I will call it a leaf-cupule complex, because the subtending leaf was essentially unmodified. The cupule has been interpreted in many ways (Retallack and Dilcher 1981, Pigg and Trivett 1994): as a sporophyll fused to a leaf, a sporophyll on an axillary shoot fused to a leaf, or an adaxial fertile segment of a leaf (analogous to the fertile segment of Ophioglossales: Kato 1990). Whatever the cupule was, reduction to one ovule per cupule would yield an organ like a bitegmic ovule. Furthermore, the subtending leaf could be folded around the cupule to form the carpel wall—an advantage over the Caytonia hypothesis. Actually, the two hypotheses may not be mutually exclusive, since it is possible that Caytonia and glossopterids are related. This would be consistent with their simple reticulate leaf venation, with a midrib and one order of laminar venation. The main difference is that the Glossopteris leaf was simple but the Caytonia leaf was palmately compound, with four leaflets each resembling a Glossopteris leaf.

There are good reasons to believe that the many structures called cupules were not all homologous. Cupules of the first Late Devonian-Carboniferous seed ferns (Archaeosperma, Elkinsia, other hydraspermans, Lyginopteris) were dichotomously organized and borne apically on special fronds or segments of fronds (Kidston 1924, Long 1961, 1979, Galtier 1988, Retallack and Dilcher 1988, Serbet and Rothwell 1992). Kenrick and Crane (1997) argued that these cupules were homologous with the dichotomous fertile appendages of “progymnosperms,” while the ovules themselves were derived from groups of sporangia (or sporangium-bearing telomes), with the integument derived from the outer sporangia (telomes) by sterilization and fusion. In contrast, the cupules of Caytonia and glossopterids were dorsiventral and therefore more like modified leaves or leaflets (cf. Reymanówna 1974). This view is consistent with the stratigraphic distribution of the two types of cupules: dichotomously organized cupules appeared near the origin of typical compound fronds, whereas cupules of the dorsiventral type appeared much later.

The first cladistic analysis to address these questions was by Crane (1985). The best-known result of this study is the inference that angiosperms were related to Bennettitales, the Cretaceous genus Pentoxylon, and Gnetales, forming the “anthophyte” clade. However, the closest outgroups of anthophytes were Mesozoic seed ferns: corystosperms, Caytonia, and glossopterids. As Crane argued, these results were consistent with the cupule-ovule homology. He proposed a scheme starting with a megasporophyll bearing multiovulate cupules, as in Caytonia. Then the number of ovules per cupule was reduced to one, as in corystosperms. The carpel was derived from the whole sporophyll by expansion and folding of the rachis. Crane postulated that Pentoxylon and Bennettitales underwent further reduction to one cupule per sporophyll and a shift of the cupule to an orthotropous orientation. This relied in part on the view of Harris (1954) that some Bennettitales had a cupule, an interpretation recently rejected by Rothwell and Stockey (2002) and Stockey and Rothwell (2003) based on observations on Cycadeoidea and Williamsonia. Doyle and Donoghue (1986) obtained results similar to those of Crane, but with angiosperms at the base of the anthophytes and Caytonia as their sister group. They interpreted the flowers of Gnetales as still more reduced, with reduction to one ovule per flower, loss of the cupule, and formation of a new outer integument from two perianth parts (cf. Doyle 1994).

These results also supported the view that not all cupules were homologous. The basal Paleozoic seed ferns with dichotomous cupules were separated from Mesozoic groups by lines that lacked cupules and bore seeds directly on more or less leaflike sporophylls, such as medullosans, Callistophyton, and cycads. This implies that the original cupule was lost (or perhaps less plausibly that the original integument fused with the nucellus and the cupule was transformed into a new integument: Walton 1953, Meyen 1984), and that the cupules of more advanced seed ferns were modified ovule-bearing leaves or leaflets.

Other morphological analyses kept the anthophytes together but separated them from Mesozoic seed ferns (Nixon et al. 1994, Rothwell and Serbet 1994). Gnetales were monophyletic in Rothwell and Serbet (1994), but paraphyletic in Nixon et al. (1994), with angiosperms nested within them. In both analyses anthophytes were linked with conifers, while Caytonia and glossopterids formed a clade situated lower in the tree.

These results have been called into question by molecular phylogenetic analyses of living seed plants. Obviously such studies say nothing directly about relationships of angiosperms to fossil taxa, but they do address the view that angiosperms are related to Gnetales. Only a few molecular analyses have linked angiosperms and Gnetales, and this with low statistical support (Hamby and Zimmer 1992, Stefanovic et al. 1998, Rydin et al. 2002). Some analyses have placed Gnetales at the base of seed plants (Hamby and Zimmer 1992, Albert et al. 1994, Sanderson et al. 2000, Rydin et al. 2002), but tests using likelihood and other methods suggest that this arrangement is a result of long-branch attraction, particularly affecting third codon positions (Sanderson et al. 2000, Magallón and Sanderson 2002, Rydin et al. 2002, Soltis et al. 2002, Burleigh and Mathews 2004). Most analyses, especially those based on combining several genes, have associated Gnetales with conifers, either as their sister group or nested within them, as the sister group of Pinaceae (Goremykin et al. 1996, Chaw et al. 1997, 2000, Hansen et al. 1999, Qiu et al. 1999, Samigullin et al. 1999, Shindo et al. 1999, Winter et al. 1999, Bowe et al. 2000, Frohlich and Parker 2000, Sanderson et al. 2000, Rydin et al. 2002, Soltis et al. 2002, Burleigh and Mathews 2004, Nickerson and Drouin 2004, Kim et al. 2004). Trees of this sort offer a more plausible alternative to the anthophyte hypothesis, because many earlier authors pointed out similarities between Gnetales and conifers, such as linear leaves, elimination of scalariform pitting even in the primary xylem, and compound strobili constructed on a cordaite-like plan (Bailey 1944, 1949, Eames 1952, Bierhorst 1971, Doyle 1978, Carlquist 1996a).

Most such “gnetifer” and “gnepine” trees indicate that no other living gymnosperm group is any more closely related to the angiosperms: angiosperms and living gymnosperms are sister groups. This does not mean that angiosperms and gymnosperms were derived independently from non-seed plants, or that the molecular results conflict with the fossil record and should therefore be rejected (Axsmith et al. 1998). Any number of Paleozoic seed fern lines might branch off below the common ancestor of living angiosperms and gymnosperms, and other Paleozoic and Mesozoic taxa might be attached to the stem lineage leading to angiosperms. Consistent with this view, an analysis of the morphological data set of Doyle (1996) with living taxa constrained into the molecular arrangement indicated that both angiosperms and living gymnosperms are nested among Paleozoic seed ferns (Doyle 2001). However, the molecular results do mean that the search for relatives of angiosperms and steps in their origin must concentrate on fossil seed plants.

To address this question requires use of a morphological data set to determine how fossils fit into the tree of living taxa. This is a daunting task now that previous morphological analyses appear to have been so wrong about the relationship of angiosperms and Gnetales. However, the fact that morphology was wrong about Gnetales does not mean it is misleading everywhere—molecular phylogenetic analyses have confirmed many groups that were first recognized based on morphology—and in any case it is the only tool we have for the job. A general reassessment of previously used morphological characters is desirable, but a critical reevaluation of those characters that supported the anthophyte hypothesis is especially necessary, as emphasized by Donoghue and Doyle (2000). This article presents such an analysis, which incorporates a critique of supposed anthophyte synapomorphies, previously overlooked similarities between Gnetales and conifers, and new developments on the morphology of fossil seed plants and attempts to synthesize morphological data from living and fossil taxa with results of molecular analyses.

Materials and Methods

The starting point for this study was the data set of Doyle (1996), but many characters have been redefined, added, or eliminated after critical evaluation. All changes in characters and scoring of taxa are listed in Appendix 1. Some that are most relevant to angiosperm relationships or pose problems that require special argumentation are discussed here.

Results

Analyses of the two data sets (A and B) differing in interpretation of seed characters in Bennettitales gave identical trees of the same lengths, both with and without constraints. Subsequent remarks will refer to data set B, for reasons discussed further below.

The unconstrained analysis yielded eight most parsimonious trees of 321 steps (strict consensus in Fig. 6). These show the same arrangement of Devonian-Carboniferous seed ferns found in all previous analyses (allowing for variations in taxon sampling), with Elkinsia, Lyginopteris, and medullosans branching successively below a “platysperm” clade that contains cycads, the Late Carboniferous seed fern Callistophyton, and all remaining taxa, including coniferophytes (Crane 1985, Doyle and Donoghue 1986, Nixon et al. 1994, Rothwell and Serbet 1994, Doyle 1996). As in previous studies, Gnetales are the closest living relatives of angiosperms, but the two taxa belong to an anthophyte clade that also includes Bennettitales and Pentoxylon. These are linked with glossopterids and Caytonia, together forming a clade called glossophytes by Doyle (1996), but with a different internal arrangement of taxa. Glossophytes are nested within coniferophytes, as the sister group of conifers (including the Paleozoic genus Emporia).

Although this result reaffirms the anthophyte hypothesis, trees in which Gnetales are related to conifers rather than anthophytes are only one step less parsimonious. When I forced Gnetales and living conifers into a clade, using a backbone constraint tree of living taxa with all other relationships unresolved, I obtained 16 trees of 322 steps (representative tree in Fig. 7, with nodes not found in all trees indicated by arrows). Gnetales are nested within conifers, but not linked with Pinaceae; they may be sister to either Araucariaceae or a clade consisting of Araucariaceae, Taxodiaceae (including Cupressaceae), Cephalotaxus, and Taxaceae. Two pollen characters that support both positions are loss of air sacs and granular exine structure. The remaining glossophytes are much lower, linked with cycads by simple pinnate leaves and seed shed with mature embryo (scored as unknown in glossopterids, Pentoxylon, and Caytonia). Their internal topology is more like that of Doyle (1996), with glossopterids and Pentoxylon forming a basal clade and Caytonia the sister group of angiosperms. Although trees in which cycads are related to glossophytes are most parsimonious, trees in which they are on the line leading to other living gymnosperms or basal to both lines are only two steps longer.

In the “one-off” trees in which Gnetales are nested in conifers (Fig. 7), the topology of living conifer families (setting aside Gnetales) is the same as that found by Doyle (1996), which differs from molecular trees (e.g., Magallón and Sanderson 2002, Quinn et al. 2002) only in that Podocarpaceae are linked with Pinaceae (based on two microsporangia per sporophyll) rather than Araucariaceae. However, in the unconstrained trees (Fig. 6), Podocarpaceae may be linked with either Pinaceae or the remaining conifers (based on tangential bands of phloem fibers).

Nymphaeales are sister to all other angiosperms in both sorts of trees (Figs. 6, 7), as in the analysis of Doyle (1996), which included only one other member of the basal ANITA grade (Austrobaileya). In contrast to molecular trees, the other ANITA taxa (Amborella, Austrobaileya, Trimenia, Illicium, Schisandraceae) form a clade that also includes Chloranthaceae, separated from Nymphaeales by Piperales (Saururaceae, Asaroideae) and Winteraceae. The only variation is that Winteraceae are linked with Asaroideae and Saururaceae in the unconstrained analysis (Fig. 6) but form an adjacent line in some trees with Gnetales in conifers (Fig. 7). These relationships are similar to those found in the morphological analysis of Doyle and Endress (2000), which included many more taxa, and identical to those in the unconstrained analysis of Eklund et al. (2004), which used the same sampling of angiosperms, allowing for the fact that those studies rooted the angiosperms on Amborella.

The analysis with living taxa forced into the molecular topology yielded 18 most parsimonious trees of 339 steps (Fig. 8), which form two islands of 6 and 12 trees. The same relationships outside angiosperms were found when Amborella and Nymphaeales were constrained to form a basal clade, as in Barkman et al. (2000). This represents an increase of 18 steps over the unconstrained analysis. Most of the extra steps are due to the different arrangement of taxa in angiosperms (10) and conifers (3) and the association of cycads with other living gymnosperms (2). Paleozoic seed ferns branch in the previously observed order at the base. Medullosans are united with the platysperms by bilateral pollen symmetry, loss of the original lobed cupule, loss of the central column in the lagenostome, sarcotesta, and vascularized nucellus. Synapomorphies of the platysperms include a sulcus, honeycomb-alveolar exine structure, sealed micropyle, platyspermic seeds (defined by presence of two vascular bundles or other anatomical signs of bilateral or bisymmetric organization, not necessarily flattened shape), and, in some trees, endarch primary xylem, simple pinnate leaves, abaxial microsporangia, and linear megaspore tetrad.

The poor resolution at the base of the platysperms (Fig. 8) reflects the existence of two islands of trees and the ambiguous position of Callistophyton. In island 1 (representative tree in Fig. 9, with nodes not found in all trees indicated by arrows), glossophytes diverge at the base of the platysperms, and their internal topology (not counting the molecular arrangement in angiosperms) is the same as in one-off trees with Gnetales in conifers (Fig. 7). Relationships are more poorly resolved in the consensus of island 2 (representative tree in Fig. 10), with a basal polytomy in the platysperms involving all groups except coniferophytes and the clade made up of Bennettitales, Caytonia, and angiosperms. However, inspection of individual trees shows that this lack of resolution is due to “jumping” of Callistophyton between two widely separated parts of the tree: nested within the clade including living gymnosperms in seven trees, between glossopterids and coniferophytes, along with corystosperms and peltasperms; and just below the common ancestor of living gymnosperms and angiosperms in five trees. In all 12 trees, cycads, Pentoxylon, and glossopterids are attached in that order at the base of the gymnosperm line. Thus the two islands represent different rootings of the platysperms and different unrooted relationships in the vicinity of cycads, glossopterids, and Pentoxylon.

Character state changes on the tree shown in Fig. 9, with nodes numbered in Fig. 11, are listed in Table 1. Unequivocal synapomorphies of each clade or terminal taxon are listed first, then equivocally optimized changes. Positions of the latter changes were sometimes chosen assuming accelerated transformation (acctran), with an early origin of the derived state followed by reversals, sometimes delayed transformation (deltran), with later multiple origins. Deltran was chosen when the derived state represented loss of a structure or some other such change that seemed more likely to have occurred twice than to have reversed. In angiosperms, results of the more extensive analysis of Doyle and Endress (2000) were sometimes used to choose between equally parsimonious optimizations. Synapomorphies of extant groups that involve characters not preserved in fossil outgroups were placed at the crown-group node; such features may have arisen lower on the stem lineage.

Consistent with the ambiguous rooting of the platysperms, there is no unequivocal synapomorphy for the clade that includes extant gymnosperms; a possible candidate (equivocal because the basic state in glossophytes is unknown) is abaxial microsporangia. The only unequivocal synapomorphy of the glossophyte clade is loss of the lagenostome. Simple, pinnately veined leaves (i.e., with a midrib, unlike simple, dichotomously veined leaves in coniferophytes) are homologous throughout glossophytes (modified to palmately compound in Caytonia) but equivocal as a synapomorphy of the clade because they may or may not be homologous with the similar leaves of cycads (Fig. 9). Another derived feature that is restricted to glossophytes but equivocal as a synapomorphy is adaxial ovules, discussed further below. Glossophytes also share a thick nucellar cuticle, versus thin in all members of the living gymnosperm line where this character is known, but its status is equivocal because appropriate data are lacking for basal seed ferns. Loss of the megaspore membrane is a derived feature of all glossophytes except glossopterids, but because Pentoxylon is linked with glossopterids, it is equivocal whether this loss occurred independently in Pentoxylon and other glossophytes or (perhaps less plausibly, considering that glossopterids are the oldest members of the clade) occurred once and was reversed in glossopterids.

Within glossophytes, glossopterids and Pentoxylon are united by uniseriate rays, secondarily free microsporangia, and paddle-like megasporophylls. Bennettitales are linked with Caytonia and angiosperms by presence of scalariform pitting in the secondary xylem, endarch leaf traces, siphonogamy, and reduced pollen chamber (all but the last being characters that are unknown in Caytonia). Caytonia is linked with angiosperms by reticulate venation (initially of one vein order), unraised guard cell poles, anatropous cupules (including bitegmic ovules), and loss of nucellar vasculature. The present topology implies that reticulate venation is not homologous in glossopterids and Caytonia, because it would have to be lost independently in Pentoxylon and Bennettitales. Saccate pollen may be homologous throughout the living gymnosperm clade (Callistophyton, Autunia, corystosperms, cordaites, conifers), but not in glossopterids and Caytonia. Problems concerning these characters and other potential synapomorphies are discussed below.

Decay and bootstrap values from the unconstrained analyses (Fig. 6) reveal fairly strong support for relationships at the base of the seed plants, including the association of medullosans with platysperms (decay index 5 steps, bootstrap frequency 93%) and the monophyly of platysperms (4 steps, 90%), as well as for Gnetales (8 steps, 98%) and angiosperms (5 steps, 98%). However, support values for near-basal nodes in the platysperms are low, including those for nodes in the glossophyte clade (1 step, <50%). When living taxa are constrained into the molecular topology (Fig. 8), support values for relationships that involve fossil taxa are generally similar to those found without constraints. Decay support for the relationship of angiosperms with Caytonia and Bennettitales is slightly higher (2 steps) than other relationships in glossophytes, but bootstrap support remains low (61% for association of Caytonia with angiosperms, <50% for association of Bennettitales).

Discussion

The fact that both anthophyte and conifer relationships of Gnetales became almost equally parsimonious after the character revisions made here suggests that morphology does not conflict as strongly with molecular data on the position of Gnetales as it seemed. Apparently much of the conflict between previous morphological and molecular results was due to difficulties in assessing homology in certain morphological characters. A skeptic might argue that this change in parsimony only indicates that morphological characters can be reinterpreted at will to support any desired relationship. However, this would be unduly pessimistic. In the process of reciprocal illumination outlined above, closer examination showed that some characters thought to support the anthophyte hypothesis are instead equally consistent with either relationship, whereas others had been interpreted incorrectly. At the time of the first molecular analyses it seemed that morphology strongly favored the anthophyte hypothesis, whereas molecular data were equivocal, suggesting the morphological result should be accepted (Doyle 1998b). But now the situation is reversed: more voluminous and better-analyzed molecular data strongly contradict the anthophyte hypothesis, whereas morphological data are ambiguous. On a more positive note, the inference that morphology is less positively misleading than it seemed may be grounds for optimism about the prospects of using morphological data to fit fossil taxa into a molecular framework of living taxa.

As in Doyle (1996), the bootstrap and decay analyses (Figs. 6, 8) indicate that the strongest results concern relationships near the base of the seed plants, including the monophyly of platysperms, which correspond roughly to crown-group seed plants (depending on the position of Callistophyton), and the monophyly of Gnetales and angiosperms. In the constrained analysis (Fig. 8), bootstrap values below 100% in extant conifers and angiosperms, within which relationships were fixed, must be due to “infiltration” of fossil outgroups into the crown groups in some bootstrap replicates. A similar effect can be seen by examining trees found during the decay analyses. For example, in trees found when angiosperms were specified as not forming a clade (five steps longer than the shortest trees), Caytonia was nested within angiosperms. Thus the present numbers may underestimate the true support for angiosperm monophyly, because trees with Caytonia nested in angiosperms assume that Caytonia had angiosperm synapomorphies in embryological and other characters that are not preserved, which may be true in some cases, but probably not in all.

Unfortunately, the low support for relationships in basal platysperms and glossophytes means that the question of angiosperm relationships is still far from resolved. In the unconstrained analyses, the low values may reflect almost equal support for placement of Gnetales in glossophytes and in conifers, plus uncertain relationships among other fossil platysperms. Support in the vicinity of conifers is slightly higher in the constrained analysis (Fig. 8), where Gnetales were not allowed to “jump” out of conifers. The slightly higher decay support for relationships of angiosperms with Caytonia and Bennettitales in the constrained analyses may reflect the fact that Gnetales are no longer in the picture, but why bootstrap support for these relationships remains low is unclear. These results mean that the position of angiosperms among glossophytes is only a best guess, which may however serve as a focus for future investigations in paleobotany and in evolution and development. The strongest inference is that both angiosperms and other living seed plants are nested among Paleozoic seed ferns, in the platysperm clade, and that homologies for angiosperm organs are to be sought among fossil members of this clade, rather than in more basal seed ferns.

The unconstrained trees (Fig. 6, with leaf character in Fig. 12) are reminiscent of some trees that were one step longer than the shortest trees in the analysis of Doyle (1996, Fig. 9), which differed in relationships among glossophytes and the fact that their sister group was cordaites rather than conifers. This result recalls the view of Schopf (1976) that glossopterids were related to coniferophytes, with their simple, pinnately veined leaves derived from cordaite- or Ginkgo-like simple leaves with dichotomous venation by aggregation of veins into a midrib, and with their “fertiliger” (leaf-cupule complex) derived from a bract-axillary fertile short shoot unit of the type found across coniferophytes. It also recalls the suggestion of Crane et al. (2004) that among modern plants angiosperms are related to conifers and Gnetales, but cycads and Ginkgo are more basal, based on the restriction of siphonogamy to conifers, Gnetales, and angiosperms. Ironically, though, siphonogamy would not be a valid synapomorphy if fossil taxa are interpolated as found here, given the discovery of zooidogamy in glossopterids (Nishida et al. 2003, 2004) and its presumed occurrence in cordaites (e.g., Poort et al. 1996), in addition to more basal seed ferns (Benson 1908, Stewart 1951). However, trees found when Gnetales were forced together with conifers (Fig. 7) or living taxa were constrained into the molecular topology (Figs. 8, 9, 10) imply rather that the glossophyte line diverged earlier from platyspermic seed ferns. This emphasizes again that relationships near the base of the platysperms are poorly resolved, whether because of the smaller number of known characters in Permian and Mesozoic fossils than in Carboniferous forms, very rapid radiation after origin of the clade, or both. Better evidence on anatomical and life cycle features in Permian and Mesozoic taxa could have a significant impact and should be a high priority for students of seed plant phylogeny.

Given that the analysis with Gnetales forced together with conifers placed cycads on the line leading to angiosperms (Fig. 7), but trees with cycads on the line leading to other living gymnosperms (as in most molecular analyses) or below both lines are only two steps longer, the conflict between morphological and molecular data on cycads is not severe. The position of cycads is one of the more weakly supported aspects of molecular phylogenies (cf. Magallón and Sanderson 2002, Soltis et al. 2002). Considering extant taxa alone, it might seem that different positions of cycads would have very different implications for character evolution in seed plants. However, trees with all three positions of cycads actually imply rather similar scenarios for the evolution of most characters, such as leaf morphology; differing positions of Permian and Mesozoic groups have a greater effect (compare Figs. 7, 9, 10, and 12). This is because Paleozoic seed ferns remain at the base of seed plants and relationships within the two main platysperm lines are often similar. This reaffirms the view that incorporation of fossil taxa into molecular trees can be necessary in order to gain a proper understanding of character evolution in ancient groups such as seed plants, even when relationships among living taxa are not affected (Doyle and Donoghue 1987, Donoghue et al. 1989).

All these trees are troubling in indicating that pollen germination through a distal sulcus originated at the base of platysperms, implying that the tetrad scar and proximal pollen germination of cordaites and Paleozoic conifers (such as Emporia) are not primitive features, as generally assumed (e.g., Poort et al. 1996), but rather reversals. This, together with the fact that cordaites are the oldest known platysperms, could be evidence for a more basal position of coniferophytes.

In discussing the implications of these results, there would be little justification for using a tree from the unconstrained analysis, since some of the relationships within angiosperms are strongly contradicted by molecular data. This is evident from the combined analysis of Doyle and Endress (2000), where molecular data overruled morphological data in most cases, for example in placing both Nymphaeales and other ANITA taxa together in a basal grade. Exceptions, where morphology overcame weakly supported molecular relationships, concerned taxa not included in the present data set (e.g., association of Lauraceae with Hernandiaceae, Piperales with monocots). The ideal procedure would be to combine the present data with DNA sequences and analyze them together, but in the absence of this, and because the results of Doyle and Endress (2000) indicate that DNA would dominate in the taxa sampled here, I will instead concentrate on the analysis with living taxa constrained into the molecular arrangement (Fig. 8). I will consider data set B, with seed characters of Bennettitales scored as uncertain where they differ between Vardekloeftia and other taxa. Although I accept the interpretation of Cycadeoidea and Williamsonia by Rothwell and Stockey (2002) and Stockey and Rothwell (2003), I find the interpretation of Vardekloeftia by Pedersen et al. (1989b) more convincing than Rothwell and Stockey's (2002) critique of it, and it seems premature to assume which set of characters is ancestral. In any case, this is not a critical issue, because the two scorings of Bennettitales had no effect on inferred relationships.

Of the two types of trees found in the constrained analysis (Figs. 9, 10), trees from island 1 are more plausible in terms of the stratigraphic distribution of taxa and morphotypes. Island 2 (Fig. 10) implies that a line consisting of some of the youngest taxa of seed plants—Bennettitales and Caytonia, both unknown before the Late Triassic, and angiosperms—diverged at the same time as a line with members extending back to the Late Carboniferous (cordaites, conifers, probably cycads). This implies a long ghost lineage for the former line (where its existence is predicted by the tree but not attested in the fossil record: cf. Doyle 1998b). Furthermore, it implies that the first members of both lines had pinnately veined simple leaves, while the fernlike leaves of Autunia, Peltaspermum, corystosperms, and (in some trees) Callistophyton were secondarily compound (Fig. 10). In fact, compound leaves predominated in the Carboniferous, while simple pinnate leaves did not appear until the latest Carboniferous (Taeniopteris) and (with the notable exception of Permian glossopterids) remained subordinate until the Mesozoic. In contrast, trees in island 1 (Fig. 9) place older groups (including glossopterids) near the base of both lines and Mesozoic groups in more nested positions, and they allow pinnately compound leaves to be interpreted as primitive in all taxa where they occur (except Caytonia). Therefore I will use a tree from island 1 (Fig. 9) as a basis for discussion of evolutionary implications.

Implications for the evolution of ovulate structures can be introduced in terms of the ovule position character (Fig. 13), with apical, abaxial, adaxial, and marginal states. At the base are Paleozoic seed ferns—first Elkinia and Lyginopteris with cupules of the dichotomous type, then medullosans, with no cupule, in all of which the ovules appear to be apical. Seed ferns with abaxial ovules—Callistophyton (the most plesiomorphic example of this type), peltasperms, and corystosperms—form a grade on the line between cycads (in which ovules are basically marginal: Norstog and Nicholls 1997) and coniferophytes, except in one tree in which these taxa are linked with cordaites. Coniferophytes (including Gnetales), whose more plesiomorphic members (cordaites, Ginkgoales, and Paleozoic conifers such as Emporia) had ovules that were apparently apical on simple sporophylls, are nested in this clade. Their simple sporophyll morphology is presumably a consequence of a general shift from fernlike fronds to simple leaves, ascribed by Rothwell (1982) to heterochronic substitution of cataphylls for fronds. The glossophyte line includes all taxa with adaxial ovules, with glossopterids and Pentoxylon at the base, and with Caytonia linked with angiosperms. Because lines with adaxial, abaxial, and marginal ovules diverge from adjacent nodes, the most parsimonious ancestral state in platysperms is equivocal, and it is not clear which of these states are synapomorphies. One alternative is that all three states were separately derived from apical and are therefore synapomorphies of their respective clades.

Whereas most versions of the anthophyte hypothesis implied that the “flowers” of Gnetales were reduced from more complex structures (Crane 1985, Doyle and Donoghue 1986, Doyle 1994), the molecular results support the hypothesis that they are homologous with the axillary fertile short shoots of coniferophytes, best seen in cordaites and Paleozoic conifers. This view was proposed by Eames (1951) for Ephedra but later extended to the other genera (Bierhorst 1971, Doyle 1978), and it was discussed by Doyle (1994) in relation to trees in which anthophytes were nested in coniferophytes (e.g., Nixon et al. 1994). Shindo et al. (1999) argued that it was supported by developmental genetic data. The position of Gnetales within conifers, linked with Pinaceae, implies that the female fertile short shoot, which corresponds to the cone scale of living conifers (Florin 1951), was either transformed into a woody cone scale twice, in Pinaceae and in other conifers, or that the cone scale reverted to a shoot with scale-like appendages in Gnetales, a less plausible scenario. In either case, the constrained trees imply that the fertile short shoot, which had become dorsiventral in the common ancestor of Emporia and crown-group conifers, reverted to bisymmetric in Gnetales. In addition, there was a shift from simple to compound male strobili, perhaps by remodeling of the male structures on the female plan. Future studies may clarify whether the compound male strobili of the Paleozoic conifer Thucydia (Hernandez-Castillo et al. 2001) are relevant to this question. There is no character in the present data set that unequivocally links Pinaceae and Gnetales, although double fertilization of the Ephedra type, which has been reported in some Pinaceae but is not confirmed for the whole family (which was therefore scored as 0/1), could be such a synapomorphy (Friedman and Floyd 2001).

It may be significant that anthophyte trees implied that the inferred ancestral megasporophyll and cupule were completely lost by reduction in Gnetales (Doyle and Donoghue 1986, Doyle 1994, 1996). In hindsight the absence of these structures was a danger signal suggesting that Gnetales belonged elsewhere.

If a leaf-cupule complex of the glossopterid type existed on the line leading to angiosperms, their bitegmic ovule could be derived by enrolling of the cupule and reduction of the number of ovules on its adaxial surface to one, and the carpel wall could be derived from the subtending leaf (Stebbins 1974, Retallack and Dilcher 1981, Doyle 1996). These homologies are diagrammed in Fig. 14, with the abaxial surface of foliar structures indicated in black. In glossopterids, I have illustrated a unicupulate leaf-cupule complex (Fig. 14a) and two interpretations of the multicupulate type (e.g., Lidgettonia: Surange and Chandra 1975, Schopf 1976, Retallack & Dilcher 1981; Fig. 2J–L). In Fig. 14b the cupules are interpreted as leaflets of a single compound sporophyll, in Fig. 14c as several simple sporophylls. For angiosperms, I have shown an ascidiate carpel with one ovule (Fig. 14d), a common type in basal angiosperms (Endress and Igersheim 2000), and a classic plicate carpel with several ovules (Fig. 14e). The parallels between the unicupulate glossopterid type (Fig. 14a) and the ascidiate carpel (Fig. 14d) are especially close: the bitegmic ovule develops from the cross-zone on the ventral side of the carpel primordium, where one might expect an axillary branch, and the ovule has the proper orientation, allowing for tilting toward the carpel midrib. The positional relationships are less clearly comparable in the plicate carpel. However, the precise geometry may not be critical, considering the great flexibility in placentation within angiosperms.

The hypothesis that the carpel was derived from a glossopterid leaf-cupule complex is consistent with a widespread view among developmental geneticists (e.g., Skinner et al. 2004), based on mutants and gene expression patterns, that the placenta and carpel wall are distinct structures: the carpel wall corresponds to the leaf, the placenta to the axillary fertile branch. This also recalls the “gonophyll theory” of Melville (1963), but without his reliance on now-refuted reconstructions of glossopterids and his concept that angiosperm gynoecia were derived polyphyletically from glossopterid structures.

A weakness of this scheme is uncertainty in reconstructing characters of the first glossophytes. The present trees imply that several features of glossopterids are derived: uniseriate rays, unfused microsporangia, and paddle-like megasporophylls in both glossopterids and Pentoxylon, plus reticulate venation, air sacs and striations on the pollen, and loss of the sarcotesta in glossopterids alone. However, scenarios in which many of these features are ancestral in glossophytes are only one step less parsimonious (discussed below for reticulate venation). Uniseriate rays, an important aspect of the pycnoxylic wood syndrome, are suspiciously correlated with the cool temperate distribution of glossopterids and Pentoxylon (as well as conifers, ginkgos, and corystosperms). The inference that paddle-like megasporophylls were derived may be an artifact of the present character definition if they were homologous with the cupules of Caytonia and the bitegmic ovules of angiosperms. Alternatively, if multicupulate leaf-cupule complexes were ancestral in glossopterids and the ovule-bearing portion was a compound megasporophyll (Fig. 14b), glossopterids would have the same state as Caytonia.

The concept that the leaf-cupule complex was ancestral in glossophytes is consistent with the age of glossopterids, and it would be enhanced if they were shown to be paraphyletic. Many authors have expressed suspicions that glossopterids were heterogeneous, noting for example the contrast between unicupulate types, in which the cupule was quite leaflike, and multicupulate types, in which the cupules were more modified and contained fewer ovules (cf. Pigg and Trivett 1994, Taylor 1996). There is little reason to believe that glossopterids were polyphyletic, since the different conditions could be a result of variation in the number of sporophylls per branch and the number of ovules per sporophyll, rather than fundamentally different starting points. It may be easier to imagine that some traditional glossopterids were more closely related to Mesozoic taxa than others, making the group paraphyletic. However, to demonstrate this would require showing that the apparent synapomorphies of glossopterids were outweighed by synapomorphies of some glossopterids and Mesozoic taxa.

The present scheme also requires that something like the leaf-cupule complex persisted on the line between glossopterids and angiosperms, but other taxa attached to this line show no sign of such a structure, with the possible exception of the cupule in the bennettitalian genus Vardekloeftia (Harris 1932, 1954, Pedersen et al. 1989b). Bennettitales had stalked ovules borne on a radial receptacle, while Pentoxylon had ovules borne on all sides of a structure that was originally assumed to be radial but was shown by Rothwell and Serbet (1994) to have a bilateral, leaflike anatomy. This may not be a problem if these conditions were autapomorphic specializations of structures of a glossopterid type, which I allowed by scoring Bennettitales as unknown for megasporophyll morphology and ovule position. The most common interpretation is that each stalked ovule of Bennettitales was a highly reduced sporophyll (Crane 1985, Rothwell and Stockey 2002). A more exotic alternative is that the receptacle was a sporophyll shifted to a terminal position and radialized (Doyle and Donoghue 1986, Doyle 1996), on analogy with Pentoxylon. However, both hypotheses imply that the ancestral leaf-axillary branch organization was lost, whether by reduction, fusion, or heterotopic transfer of the sporophyll to an axis of a lower order.

This problem would be less severe if Bennettitales and Pentoxylon formed a clade, as in Crane (1985), putting both deviant taxa on a sideline. This relationship is only one step less parsimonious with no other constraints, but three steps worse with molecular backbone constraints. The problem would disappear if it was shown that Bennettitales were not related to glossophytes but to some other group, such as cycads, a relationship that is two steps less parsimonious with no other constraints, three steps worse with the molecular backbone. Perhaps the situation is analogous to that of Gnetales, where lack of any vestige of the ancestral megasporophyll or cupule now appears to be evidence that Gnetales did not belong in the anthophytes. Discovery of more plesiomorphic relatives of Bennettitales or Pentoxylon could show either conditions more compatible with the glossopterid type or something different, strengthening or refuting the present scheme.

Caytonia fits better between glossopterids and angiosperms: it had cupules that correspond to the predicted intermediate, in containing several ovules but being anatropous, plus angiosperm-like advances in its seeds (no pollen chamber, thick nucellar cuticle, loss of the megaspore membrane). However, other aspects of its morphology are hard to explain in glossopterid terms. Several interpretations are possible, each of which raises new questions (Fig. 14f–h). Was the ovulate structure a compound leaf, with a rachis and leaflets converted into cupules, as believed by Harris (1940, 1951) and Reymanówna (1974) and assumed here in scoring Caytonia as having pinnate megasporophylls? If so, was it borne directly on a main stem (Fig. 14f), with the adaxially enrolled cupules facing upward? This would be difficult to reconcile with a glossopterid prototype, except by postulating that the sporophyll was transferred from the axillary branch to a stem of a lower order. Such a sporophyll might be transformed into an angiosperm carpel by expansion of the rachis (Gaussen 1946, Stebbins 1974, Doyle 1978; Fig. 1), an alternative to the proposed homologies with glossopterids. Or was the ovulate structure a compound leaf borne on an axillary shoot (Fig. 14g)? Such a system could be compared with a Lidgettonia leaf-cupule complex as interpreted in Fig. 14b, where the cupules correspond to leaflets. This would predict that Caytonia cupules faced downward, toward a subtending leaf or bract. Or, despite its dorsiventral appearance, was the Caytonia ovulate structure actually an axillary branch bearing several simple sporophylls (Fig. 14h)? This would correspond to Lidgettonia as reconstructed in Fig. 14c, where each cupule is a simple sporophyll. In either of the latter schemes (Fig. 14g, h), was the subtending leaf distinct, or was it fused to the cupule-bearing axis, so that what appears to be a rachis was actually a composite structure?

Specimens that show the ovulate structures of Caytonia attached to a stem are needed to decide among these alternatives, although they might not be easy to interpret. Retallack and Dilcher (1988) reconstructed the cupules as facing downward, based on a sporophyll apparently attached to a stem in a specimen at Cambridge University, with no sign of a subtending leaf or bract. They interpreted this orientation as evidence that the ovules were on the abaxial surface of the cupules, although it might be consistent with Fig. 14g, in which the ovules are adaxial, if the bract was highly reduced or fused to the rachis. However, after examining this specimen I am not convinced that the relative orientation of the parts can be determined.

If nothing comparable to a glossopterid leaf-cupule complex can be found in the Mesozoic taxa associated here with glossopterids and angiosperms, it could mean that this structure was an autapomorphy of glossopterids that never existed on the line leading to angiosperms, thus refuting the proposed homologies. Or it could suggest an exclusive link between angiosperms and glossopterids. Answers to these questions, which might come from better information on the anatomy of glossopterids and Mesozoic fossils, could have a major impact on the angiosperm question.

Origin of the angiosperm stamen in terms of potential outgroups is less widely discussed, but it poses as many problems as origin of the carpel. Male structures of glossopterids consisted of a branched sporangium-bearing unit adnate to the adaxial side of a leaf (Surange and Maheshwari 1970, Schopf 1976, Gould and Delevoryas 1977, Retallack & Dilcher 1981), reminiscent of the leaf-cupule complex. Bennettitales had what appear to be sporophylls bearing microsynangia on their adaxial side, but it is worth considering that these were derived from compound structures of the glossopterid type. Whatever the origin of these structures, extreme reduction in the number of sporangia in either group might result in something like the stamens of basal angiosperms, which have two pairs of microsporangia borne on their adaxial side (Doyle and Endress 2000). Each pair of microsporangia would represent a separate synangium. As with the female structures, it is unclear how glossopterid and bennettitalian organs relate to the branched microsporophylls of Pentoxylon and Caytonia.

Another aspect of the angiosperm problem is origin of the angiosperm leaf. There is a wide morphological gap between the simple leaves of extant basal angiosperms, with pinnate secondary veins and reticulate fine venation, and Paleozoic seed ferns, which were pinnately compound and had dissected pinnules with open dichotomous venation. Leaves of the glossopterid type could fill part of this gap: not only were they already simple, but they also had reticulate laminar venation. The main difference is that the reticulum was simple, consisting of veins of only one order, as opposed to complex in angiosperms, consisting of several orders, the finest of which are freely ending veinlets. Elaborating on ideas of Stebbins (1974), Doyle and Hickey (1976) proposed that transformation of a seed fern frond into an angiosperm leaf involved radical reduction in a semiarid environment, but this is unnecessary under the present scheme, since glossopterid leaves were simple but not highly reduced. Origin of the angiosperm leaf could instead involve origin of a hierarchy of coarse to fine veins without any major change in size, which might reflect a shift in the type of meristematic activity responsible for production of the blade, from marginal to diffuse (Boyce 2005). This would be consistent with arguments of Feild et al. (2004) that the first angiosperms were adapted to disturbed habitats in the wet forest understory.

A problem for this scenario is the fact that taxa without reticulate venation, namely Pentoxylon and Bennettiales, are interpolated between glossopterids and angiosperms, although glossopterid-like venation does occur in the palmately compound leaves of Caytonia. As a result, although simple leaves with pinnate venation (like the Taeniopteris-type leaves of Pentoxylon and some Bennettitales) are reconstructed at the base of glossophytes, it is most parsimonious to assume that reticulate venation originated independently in glossopterids and the Caytonia-angiosperm clade. The alternative, that it originated once but was lost in Pentoxylon and Bennettiales, is one step less parsimonious. This picture would change if Pentoxylon and Bennettiales formed a clade, which would require only one reversal from reticulate to open venation. Also, some Bennettitales had reticulate venation (Dictyozamites); if these were shown to be plesiomorphic in the group, the hypothesis that reticulate venation was homologous in glossopterids, Caytonia, and angiosperms would be strengthened.

These observations bring out a general problem for the present scheme, the stratigraphic gap in the record of plants with glossopterid-like features after the mass extinction at the end of the Permian (Retallack 1995). The scheme predicts that some derivatives or relatives of glossopterids, with or without additional advances, survived into the Mesozoic. This picture could change with the discovery of new Mesozoic fossils or association of known but isolated organs. A possible example is Mexiglossa, a Glossopteris-like leaf from the Jurassic of Mexico (Delevoryas and Person 1975), which co-occurs with branched microsporophylls (Perezlaria) suggestive of the male structures of Caytonia (Delevoryas and Gould 1971). In general, if the plants associated as glossophytes do form a clade, they are probably not its only members. This is suggested by Petriellaea (Taylor et al. 1994), which also had cupules with adaxial ovules. Anderson and Anderson (2003) have described a remarkable array of plants from the Triassic Molteno flora of South Africa, some with anatropous cupules, that might also belong here.

The present results also provide a new perspective on the origin of siphonogamy, a feature of angiosperms and Gnetales that seemed to be an anthophyte synapomorphy and was more recently proposed as a synapomorphy of angiosperms, conifers, and Gnetales (Crane et al. 2004). Stockey and Rothwell (2003) showed that it probably existed in Bennettitales. A major new element is the report by Nishida et al. (2003, 2004) that glossopterids had motile sperm. This cannot be taken as evidence that glossopterids are not related to angiosperms. The molecular arrangement of living taxa (Fig. 5), where cycads and Ginkgo, with motile sperm, are attached between conifers and angiosperms, implies that siphonogamy arose independently in conifers (including Gnetales) and on the line leading to angiosperms. Considering living taxa alone, its origin on the angiosperm line could have occurred at any time between the Carboniferous and the Cretaceous. In terms of the tree of fossil and living taxa (Fig. 15), the discovery of motile sperm in glossopterids implies that siphonogamy arose between glossopterids and Bennettitales. This tree also predicts that Pentoxylon had motile sperm; discovery that Pentoxylon too was siphonogamous might be evidence that it was closer to Bennettitales and angiosperms. Such examples illustrate not only the uncertainties caused by incomplete preservation, but also the potential impact of new data on previously unknown characters.

Of course, there are other fossils that might invalidate this scheme, such as Permian gigantopterids (cf. Taylor and Li 1997), which had even more angiosperm-like leaf venation but are too incompletely known to be included in an analysis. What is needed to test these hypotheses is better understanding of Permian and Mesozoic seed plant diversity and the morphology of fossil taxa that are already known. Examples include information on the nodal anatomy of glossopterids and Caytonia (two-trace unilacunar in Pentoxylon and basal angiosperms); wood anatomy, sporophyll attachment and associated structures in Caytonia; seed cuticle characters based on coordinated observations on both petrified and compressed material; details of the life cycle in Mesozoic fossil taxa; and recognition of more plesiomorphic relatives of Bennettitales, Pentoxylon, and Caytonia.