Monetianthus

Monetianthus mirus is based on a single coalified flower from Vale de Agua, Portugal, of probable early Albian age (see Material and Methods). It was briefly reported as nymphaealean by Friis et al. (2001) and formally described by Friis et al. (2009) using SEM and X-ray microtomography (synchrotron radiation X-ray tomographic microscopy). It has an inferior ovary, a star-shaped ring of 12 fused carpels, and broken bases of tepals and stamens. Pollen adhering to the carpels and areas between the stamen bases is all of the same reticulate monosulcate type and was therefore assumed to be from this species (and probably this flower). The relationship of this fossil to Nymphaeales was questioned by Crepet et al. (2004) and Gandolfo et al. (2004), based on a phylogenetic analysis that showed its characters were equally compatible with Illiciaceae and other angiosperm families. However, the new observations of Friis et al. (2009) clarified several characters, including presence of several ovules per carpel and laminar placentation, a distinctive feature of most Nymphaeales. Friis et al. (2009) argued that the flower was at the anthetic stage because of the irregular height of the broken-off organ bases and the fact that the ovules had not matured into seeds. They distinguished 9–10 perianth parts and ∼20 stamens by the elliptical versus rhombic shape of their bases and the different form of their vascular bundles. The only characters of ovule morphology that could be determined were anatropous curvature and presence of two integuments.

Friis et al. (2009) analyzed the phylogenetic position of Monetianthus using the data set of Saarela et al. (2007), with the addition of three characters observed in the fossil, namely, “star-shaped” carpel arrangement, septal slits, and ovules not filling the carpel locule. In Doyle and Endress (2010), we adopted star-shaped carpel arrangement as a state of a character that includes both carpel number and arrangement, defined as more than five carpels in one whorl or series. This does not specify whether the carpel phyllotaxis is whorled or spiral, which may be difficult to determine in fossils; we use the term “ring” to describe this state. Among extant taxa, Friis et al. (2009) scored Lactoris as having the star-shaped state, like the related Aristolochiaceae, but because Lactoris has three carpels, rather than six in most Aristolochiaceae, it falls in our state for one whorl of 2–5 carpels, like most monocots. However, we have not added their new characters for septal slits, because these are restricted to Nymphaea subgenus Anecphya (Conard 1905, as Apocarpiae), which is nested within Nymphaea in the phylogeny of Borsch et al. (2011), and are therefore presumably derived within the genus, or for ovule size, because this character may vary with developmental stage, which is difficult to assess in fossils, and appears to be correlated with ovule number (one vs. more).

New characters of Doyle and Endress (2010) that were not used by Friis et al. (2009) are (44) flowers not in heads (inferred from presence of a pedicel), (45) pedicel present, (49) short receptacle, (50) carpels not sunken in pits in the receptacle, (51) no cortical vascular system, (52) protruding floral apex, (65) more than two stamen whorls or series, (86) sulcus not branched, and (89) uniform reticulum. Other new characters were the result of splitting of older characters, such as (53) presence of a perianth, which was previously implicit in the character for number of whorls; (61) absence of a bract-derived calyptra, whose presence was a state of the character for outer perianth whorl; and (62) more than one stamen.

A potentially important character for placement of Monetianthus is floral phyllotaxis, which is whorled in Nymphaeales but spiral in Amborella and Austrobaileyales. Friis et al. (2001) interpreted all floral parts as whorled, but on more critical examination Friis et al. (2009) were unable to detect either orthostichies or clockwise and counterclockwise parastichies that are inclined at different angles, which would indicate whorled or spiral phyllotaxis, respectively (Endress 2006; Endress and Doyle 2007). Based on these observations and the absence of Fibonacci numbers, they concluded that phyllotaxis was probably not spiral, but whorled phyllotaxis could not be established either. They considered a third possibility, that phyllotaxis was irregular, to be unlikely because of the relatively small number of parts. They suggested that the flower had a whorled arrangement that was distorted during fossilization. Based on their figures, we suspect that the flower is not distorted and its phyllotaxis was truly irregular, a condition that we score as unknown (except when phyllotaxis varies between irregular and well defined, as may occur within species; Ren et al. 2010). We therefore follow Friis et al. (2009) in scoring phyllotaxis of the perianth (54) and androecium (63) as unknown. Given the uncertainty on phyllotaxis, we follow Friis et al. (2009) in scoring merism of the perianth (55) and androecium (64) as unknown. Friis et al. treated the number of perianth whorls (or series if spiral) as unknown, but based on the presence of tepal scars at two levels (Friis et al. 2009, fig. 1C), we score this number (56) as either two or more. The number of stamen whorls or series (65) was clearly more than two.

Considering more substantive differences in character analysis, Friis et al. (2009) scored pollen size (82), which is 18–20 μm, as small (<20 μm), but we score it as either medium (20–50 μm) or small (1/2) because pollen in fossil flowers studied with SEM is often smaller than dispersed pollen of the same type studied with light microscopy, suggesting the possibility that in situ pollen may undergo shrinkage (Doyle et al. 2008). Friis et al. (2009) scored carpel form (97) as ascidiate, but on page 1097 they compared carpels of the fossil with those of Barclaya, which has our intermediate state (plicate above and ascidiate below, with ovules in the ascidiate zone), versus ascidiate in other Nymphaeales. Schneider (1978) showed that carpels of Barclaya have a plicate zone that extends below the stigmatic area to the top of the area bearing ovules, so they are distinctly more plicate than carpels of Nymphaea and Nuphar, in which there is a shorter external slit surrounded by stigmatic tissue. In Monetianthus the carpel margin begins well above the level with ovules, so it may have had the condition seen in Nymphaea and Nuphar. Monetianthus has a ventral slit in the free part of the carpel (fig. 2B of Friis et al. 2009), but what happened below this is unclear. Because it is often impossible to distinguish plicate and ascidiate structure without developmental data (Endress 2005), we have scored this character as unknown. Friis et al. (2009) scored the character for stigmatic papillae, which Doyle and Endress (2010) split into two characters (103, 104), as smooth or with unicellular papillae, but they (Friis et al. 2009, p. 1092) stated that the stigma “appears papillate with short, perhaps unicellular, papillae.” This would correspond to the state of character 104 for papillae unicellular or with one emergent cell, but because the nature of the papillae is too uncertain, we score this character as unknown. Friis et al. (2009) scored septal nectaries (111) as absent, but because of the possibility that the septal slits between the carpels were nectaries, we score this character as unknown.

Using the data set of Saarela et al. (2007), Friis et al. (2009) found that the most parsimonious position of Monetianthus was nested within the family Nymphaeaceae, as the sister group of Barclaya and Nymphaeoideae (Nymphaea, Euryale, and Victoria). Its next-best positions, which were one step less parsimonious, were sister to Nuphar, Barclaya, or Nymphaeoideae alone.

Our analysis using the D&E backbone tree (fig. 2) gave very similar results, despite the differences in character scoring. Results using the J/M backbone are virtually identical. Monetianthus has three most parsimonious positions, all nested within Nymphaeaceae: as the sister group of Barclaya plus Nymphaeoideae, of Barclaya, and of Nymphaeoideae. Considering unequivocal synapomorphies that support these results (derived states unambiguously placed on the branch indicated), Monetianthus is linked with Cabombaceae plus Nymphaeaceae by more than two ovules per carpel (112), with Nymphaeaceae by a ring of more than five carpels (96) and eusyncarpy (106), and with Barclaya and Nymphaeoideae by inferior ovary (48) and globose pollen (83) versus superior ovary and boat-shaped pollen in Nuphar (for sources of data on Recent taxa, see Material and Methods). Another synapomorphy of Monetianthus and Nymphaeales is laminar placentation (113), in which we include both the typical laminar placentation of Nymphaeaceae and related conditions (such as “dorsal”) in Cabombaceae, but where this feature arose is equivocal, because the position of the single apical ovule in Trithuria (=Hydatellaceae) is uncertain (Rudall et al. 2007) and was therefore scored as unknown. The positions with Barclaya and Nymphaeoideae alone conflict with the monosulcate pollen of Monetianthus, since the two living groups are united by zonasulculate pollen (84), but the extra step in this character is balanced by the fact that Monetianthus shares medium-sized pollen (82) with Barclaya and a protruding floral apex (52) with Nymphaeoideae. Other positions in Nymphaeales are at least two steps less parsimonious.

As noted by Friis et al. (2009), two features of Monetianthus that are anomalous for Nymphaeaceae are reticulate pollen tectum (88) and ascendent rather than pendent ovule orientation (114). If Monetianthus is nested in Nymphaeaceae, it is most parsimonious to interpret these features as autapomorphies at the scale of Nymphaeales and as convergences with other groups. Friis et al. (2009) cited Moseley (1971) and Igersheim and Endress (1998) as showing that ovules may be horizontal or ascendent in Nuphar and Barclaya, but Moseley’s figures of Nuphar show that ovules are initially pendent, with some becoming horizontal or ascendent at anthesis. Based on Igersheim and Endress (1998, fig. 27), ovule direction in Barclaya appears to be irregular, possibly related to the fact that the ovules are orthotropous versus anatropous in other Nymphaeales. Ascidiate carpel form tends to be correlated with pendent ovules, so the ascendent orientation in Monetianthus might be explained if its carpels were more plicate than those of modern Nymphaeales, although ovules in multiovulate plicate carpels are more commonly horizontal.

A position of Monetianthus with Illicium in the Austrobaileyales is only one step less parsimonious, in part because the tectum (88) is reticulate, as in Austrobaileyales, rather than continuous, as in Nymphaeales. In addition, Illicium is as much like Monetianthus as Nymphaeaceae in having a ring of carpels (96). All other positions outside Nymphaeales are at least two steps less parsimonious. Determination of the floral phyllotaxis in Monetianthus could affect this picture. If the relevant characters (54, 63) of Monetianthus are scored as spiral, a relationship with Illicium (which is spiral) becomes one step more parsimonious than the position in Nymphaeaceae (which are whorled), but if these characters are scored as whorled, the position with Illicium becomes three steps worse.

Other characters not included in our data set could favor a relationship with Nymphaeales or clarify its position within the order if shown to be valid. Two are characters of Friis et al. (2009) that we rejected: ovules not filling the locule, as in Monetianthus, Cabombaceae, and Nymphaeaceae (vs. Illicium), which appears to be correlated with ovule number, and presence of septal slits between the carpels, as in Monetianthus and Nymphaea subgenus Anecphya (Conard 1905), which is nested in Nymphaea (Borsch et al. 2011). Unless Monetianthus is also nested in Nymphaea, which is unlikely in view of its small size and pollen morphology, septal slits are best interpreted as a convergence in the two taxa. Friis et al. (2009) noted that Monetianthus is like those Nymphaeales that have a relatively low number of ovules per carpel (e.g., Nuphar), since the ovules appear to be linear on the septa, rather than scattered over the surface, as in taxa with higher ovule numbers (e.g., Nymphaea, Barclaya); this character could favor a position below the Barclaya-Nymphaeoideae clade. Among characters that Friis et al. (2009) did not include in their data set, they suggested that the lowest appendage on the flower may be a closely associated floral subtending bract, a feature of many Nymphaeaceae (Chassat 1962; Endress and Doyle 2009). Friis et al. (2009) also noted that Monetianthus was like Nymphaeales and unlike Illicium in lacking resin bodies, assumed to be remnants of oil cells, in the carpels. Oil cells are absent in Amborella and Nymphaeales, and their origin is a presumed synapomorphy of Austrobaileyales and mesangiosperms (Doyle and Endress 2000).

From a broader evolutionary perspective, one of the most interesting features of Monetianthus is its small size. Living Nymphaeaceae have large flowers, but like most Early Cretaceous flowers (Friis et al. 2011) Monetianthus was much smaller (∼3 mm long and 2 mm across without stamens or perianth parts, estimated to be ∼1 cm across with these parts; Friis et al. 2011), more like flowers of Cabombaceae, and Carpestella (treated next) was even smaller (∼0.65 mm). In terms of morphological parsimony, this could mean either that small flowers were ancestral in Nymphaeaceae and there was a parallel size increase trend in Nuphar and other living Nymphaeaceae or that flower size increased on the line to Nymphaeaceae and secondarily decreased in Monetianthus. However, it could be argued that the age of Monetianthus and the fact that Carpestella was also small might shift the balance toward the first scenario, which would bring Nymphaeales in line with the broader picture of early floral evolution. Discovery of fossils situated elsewhere in Nymphaeaceae could resolve this problem.

Reticulate monosulcate pollen like that associated with Monetianthus was the predominant dispersed angiosperm type prior to the rise of tricolpates, but it contrasts with the pollen of living Nymphaeales, which has a continuous tectum and is usually larger (except in Barclaya and Trithuria). As already noted, pollen characters are partly responsible for the fact that it is only one step less parsimonious to place Monetianthus in Austrobaileyales. Parsimony optimization unambiguously indicates that the reticulate tectum of Monetianthus was derived within Nymphaeales and not homologous with the same state in other groups, which originated at the node connecting Austrobaileyales and mesangiosperms (Doyle 2005). This scenario could change if reticulate pollen is found to be more widespread in early Nymphaeales.

This discussion depends on the assumption that the pollen adhering to the flower was from Monetianthus, but this is uncertain. Although Friis et al. (2009) included pollen characters in their analysis, E. M. Friis (personal communication, 2012) cautioned that the association of the pollen was not demonstrated, as only stamen bases are preserved and therefore no pollen was actually found in situ in anthers. If the pollen was not from Monetianthus, the anomaly would disappear. To address this issue, we analyzed the data set with pollen characters (81–95) scored as unknown. In this analysis the most parsimonious position of Monetianthus is sister to Nymphaeoideae, since this is no longer contradicted by the aperture character and is favored by the protruding floral apex, and a position with Illicium is three steps less parsimonious. Clearly, discovery of Monetianthus specimens with pollen in the anthers could resolve these problems.

Carpestella

Carpestella lacunata is based on a single charcoalified flower from the middle Albian Puddledock locality in Virginia, described by von Balthazar et al. (2008) using SEM and X-ray microtomography. Carpestella is like Monetianthus in having an inferior ovary, a star-shaped ring of fused carpels with septal slits, and numerous tepals and stamens (represented by ∼15 larger oval and ∼60 smaller quadrangular scars, respectively). However, it is smaller (0.65 mm long, 0.45 mm wide) and less well preserved, with the top of the 13-carpellate gynoecium missing and no adhering pollen. One carpel contains a poorly preserved structure that von Balthazar et al. (2008) interpreted as a possible seed, but this provides no reliable evidence on ovule characters.

Von Balthazar et al. (2008) examined the phylogenetic position of the fossil using the data set of Saarela et al. (2007), with the addition of characters for star-shaped carpel arrangement and septal slits. As with Monetianthus, there are several new characters in our data set that can be scored: (44) flowers not in heads, (45) pedicel present, (49) short receptacle, (50) carpels not sunken in pits, (53) presence of a perianth, (61) absence of a bract-derived calyptra, (62) more than one stamen, and (65) more than two stamen whorls or series. The gynoecium has a “central bump” suggestive of the protruding floral apex of Nymphaeoideae and Illicium (52), but because the top of the carpels is missing, it cannot be established that the apex protruded. An existing character not scored by von Balthazar et al. (2008) is (47) bisexual flowers.

Von Balthazar et al. (2008) and Friis et al. (2009) interpreted phyllotaxis of the perianth and androecium as spiral, but we score both characters as unknown. In their figure 1C, von Balthazar et al. marked one set of parastichies in the androecial zone (slanting to the left as seen in surface view), but in order to distinguish spiral from whorled, two sets must be considered. A second set (slanting to the right) is visible in their figure; assuming there was no deformation during fossilization, it is at a slightly higher angle than the first set, indicating that the phyllotaxis was not perfectly whorled, but this angle is not as much higher than the first angle as would be expected with a Fibonacci spiral (the type almost always found in flowers with spiral phyllotaxis; Endress and Doyle 2007). Parastichies with similar angles were observed by Wolf (1991) in the androecium (though not the perianth) of Nymphaea alba, along with rarer whorled and Fibonacci conditions, so whether this sort of phyllotaxis is interpreted as spiral or whorled, it is consistent in Carpestella and Nymphaea. Von Balthazar et al. (2008) scored perianth and androecium merism as irregular, but Endress and Doyle (2009) eliminated this state because it is redundant with spiral phyllotaxis; we score both characters (55, 64) as unknown. As in Monetianthus, von Balthazar et al. (2008) scored septal nectaries (111) as absent, but we treat this character as unknown.

When von Balthazar et al. (2008) analyzed the phylogenetic position of Carpestella, they obtained highly ambiguous results, but in discussion they emphasized similarities to Nymphaeaceae and Illicium. With the greater number of characters in our current data set and both backbone trees, we found the same three most parsimonious positions for Carpestella as for Monetianthus (fig. 3)—nested within Nymphaeaceae, as the sister group of Barclaya, Nymphaeoideae, or both. There are no unequivocal synapomorphies that associate Carpestella with Nymphaeales as a whole or with Cabombaceae plus Nymphaeaceae, but it is linked with Nymphaeaceae by the ring of carpels (96) and eusyncarpy (106) and with Barclaya and/or Nymphaeoideae by the inferior ovary (48). There are no characters that support or contradict a relationship with Barclaya or Nymphaeoideae alone. Remarkably, a position linked with the eudicot Trochodendron is only one step less parsimonious, supported by more than two whorls (series) of stamens (65) and the ring of carpels (96), plus the inferior ovary (48), as in both Trochodendron and its sister group Tetracentron. This illustrates how lack of pollen in fossil flowers can be a serious handicap, since Trochodendron and related eudicots differ sharply from Nymphaeales in having reticulate tricolpate pollen. Other positions outside Nymphaeales are at least two steps worse, including one with Illicium, which has a similar gynoecium. If floral phyllotaxis is assumed to be spiral, a relationship with Illicium becomes as parsimonious as one with Nymphaeaceae but not more so. Von Balthazar et al. (2008) stressed the septal slits as a feature shared with Nymphaea but not Illicium, but as discussed in connection with Monetianthus, this is unlikely to be homologous in the fossil.

Despite their identical most parsimonious positions on the tree, Carpestella and Monetianthus are clearly not the same taxon. Carpestella has a few more perianth scars and carpels and many more stamen scars than Monetianthus, but the two taxa do not differ in the scoring of any of our characters that can be determined in both, except more than two perianth whorls or series (56), rather than either two or more than two in Monetianthus. As a result, our data provide no evidence on whether they form a clade or two lines located at different points in Nymphaeaceae. If it could be shown that they formed two lines, this would add support for the hypothesis of a parallel size increase trend within Nymphaeaceae.

Pluricarpellatia

Pluricarpellatia peltata is based on impression fossils of rhizomes with attached roots, leaves, and flowers in the fruit stage, locally permineralized with iron oxide, from the Crato Formation of NE Brazil (late Aptian). These remains were figured by Mohr and Friis (2000) and formally described by Mohr et al. (2008). Pluricarpellatia resembles Cabombaceae in having slender stems and more or less peltate leaves but differs in having more carpels (6–12), while it differs from Nymphaeaceae in being apocarpous. Whereas there is no direct evidence on the ecology of Monetianthus and Carpestella, the vegetative morphology of Pluricarpellatia, with rhizomes bearing roots, leaves with long petioles, and flowers with long pedicels, clearly indicates an aquatic habit, consistent with the lacustrine origin of the sediments.

Using a data set of vegetative characters only (Taylor 2008), Mohr et al. (2008) and Taylor et al. (2008) found two most parsimonious positions for Pluricarpellatia, depending on how they scored the leaves, which they described as varying between eccentrically and centrally peltate: sister to Cabombaceae when the leaves were scored as centrally peltate but sister to Nymphaeales as a whole (not including Trithuria) when the leaves were scored as eccentrically peltate. However, Taylor (2008) used the term “peltate” to describe attachment of the petiole to the plane of the blade at a high angle (D. W. Taylor, personal communication, 2012), rather than in the well-established sense of extension of the blade around the adaxial side of the petiole. Taylor (2008) scored all living Nymphaeales as having one of four peltate states, but in most Nymphaeaceae (except Euryale and Victoria) the petiole is located at the top of a narrow notch between the basal lobes and there is no adaxial extension of the blade; we score this condition as nonpeltate. Because we define the peltate state as including cases in which some but not all leaves of the plant are peltate (32), we score Pluricarpellatia as peltate.

Mohr et al. (2008) described the vegetative phyllotaxis (21) as “unclear,” but the leaves were certainly not regularly opposite, and most were clearly alternate. However, whether they were spiral or distichous (22) is unknown. The angle and shape of the petiole attachment and the lack of thickening of the node indicate a nonsheathing base (25) with no stipules (26).

Single flowers with long pedicels were borne along the stem, with no visible subtending leaves or bracts or bracts on the pedicel, but the preservation is not good enough to establish whether such appendages were absent or had been shed. Because we interpret lateral flowers as solitary if they have more than two bracts on the pedicel and borne in racemes if they have two or fewer bracts (Endress and Doyle 2009; Endress 2010), we score the inflorescence character (42) as either solitary or raceme (0/2) and floral subtending leaves or bracts (46) as unknown. By contrast, we score extant Cabombaceae and Nymphaeaceae, which have single flowers borne along a rhizome that also bears vegetative leaves, as having racemes (Endress and Doyle 2009). Rudall and Bateman (2010, p. 405) implied that this interpretation was due partly to our inclusion of the Early Cretaceous fossil Archaefructus, which has more obvious racemes, in Nymphaeales, but this is incorrect. Rather it followed from the morphological analysis of branching in Nymphaeales by Chassat (1962) and the general principles of inflorescence classification summarized in Endress (2010).

In the flowers, only characters of the gynoecial zone are preserved. Mohr et al. (2008) thought that the carpels were most likely attached in a spiral, but we consider this uncertain. The number of carpels (6–12) suggests that they may have been borne in more than one whorl or series, as in Brasenia, but because it is possible that they were in one whorl, as in Nymphaeaceae, we score carpel number/arrangement (96) as uncertain (2/3). Mohr et al. (2008) described the carpels as slightly ascidiate when young, whereas Friis et al. (2009, p. 1098) said that they “appear to be plicate.” Many carpels show a dark line running down the middle, but it is uncertain whether this was the ventral suture of a plicate carpel, a stigmatic crest like that of the ascidiate carpel of Brasenia (Endress 2005), or a vascular bundle, so we score carpel form (97) as unknown. Mohr et al. (2008) suggested that the fruits dehisced along this line, but in the absence of actual dehisced carpels we treat fruit dehiscence (125) as unknown. Mohr et al. described the stigmatic area as “not modified”; it is possible that a style was present but fell off in the fruit stage, but because the style does persist and is conspicuous in fruits of most ANITA lines and magnoliids that have a style (including Brasenia and Cabomba: Takhtajan 1988), we score style (101) as absent. Mohr et al. (2008) described the ovules as “most likely with laminar attachment,” but because this cannot be determined from their figures, we score placentation (113) as unknown. The seeds were clearly anatropous (115), but the only other ovule and seed characters that can be scored are presence of an operculum (134) and a palisade exotesta (128), which is somewhat degraded but appears to be comparable to that in better-preserved dispersed seeds from Portugal described by Friis et al. (2010a).

In our analysis (fig. 4), Pluricarpellatia has three most parsimonious positions, with both backbone trees: attached to the stem lineage of Cabombaceae, before origin of an elongate style; to Brasenia; or to the stem lineage of Nymphaeaceae, before the origin of syncarpy. It is associated with Nymphaeales as a whole by palisade exotesta (128) and operculum (134). Two additional synapomorphies that cannot be localized precisely, because Trithuria is too modified to score, are palmate venation (30) and entire leaf margin (35). It is linked with Cabombaceae plus Nymphaeaceae by more than two ovules per carpel (112). The position with Cabombaceae is supported by peltate leaves (32), while the two other positions correspond to different assumptions on the carpel number character (96): with Nymphaeaceae if the carpels are in a whorl (series) of more than five (state 2), with Brasenia if they are in more than one whorl (state 3). The presumed absence of a style results in an extra step in this character (101) when the fossil is sister to Brasenia. Positions sister to Cabombaceae plus Nymphaeaceae, to Cabomba, to Nuphar, and to Barclaya plus Nymphaeoideae are one step less parsimonious.

The tentative nature of these results is underlined by the fact that it is only one step worse to link Pluricarpellatia with Nelumbo, in the eudicot order Proteales, which also has entire-margined (31) and peltate (32) leaves, carpels in more than one series (96), and a sessile stigma (101). Discovery of pollen of Pluricarpellatia could strengthen or refute this alternative, since Nelumbo has reticulate tricolpate pollen. This problem is all the more relevant because peltate leaves thought to be related to Nelumbo, called Nelumbites, are a conspicuous element in Albian floras of Virginia (Berry 1911; Hickey and Doyle 1977; Upchurch et al. 1994), Kazakhstan (Vakhrameev 1952), and Siberia (Samylina 1968). In Virginia, Upchurch et al. (1994) associated Nelumbites with floral receptacles that resemble those of Nelumbo in having carpels borne in pits but differ in being round rather than flat topped, and our phylogenetic analysis (Doyle and Endress 2010) supported a relationship to Nelumbo. Mohr et al. (2008) discussed similarities and differences between Pluricarpellatia and Nelumbites, but they confused the picture by referring to Nelumbites as nymphaealean and not mentioning the evidence that it was related to Nelumbo. However, there are other peltate leaves in the Albian of Jordan (Scutifolium; Taylor et al. 2008) and Kansas (Brasenites; Wang and Dilcher 2006) that are more like Brasenia than Nelumbo and Nelumbites in tending to be longer than wide, rather than round or wider than long, and in other characters used by Taylor et al. (2008).

Archaefructus

Another fossil that has been associated with Nymphaeales is Archaefructus, from the Barremian-Aptian Yixian Formation of China (Sun et al. 1998, 2001, 2002), an aquatic plant with finely dissected leaves and reproductive axes bearing pairs of stamens, single or paired carpels, and no accessory parts. A phylogenetic analysis by Sun et al. (2002) concluded that Archaefructus was a stem relative of all living angiosperms, but Friis et al. (2003) argued that its proposed primitive features could be the result of reduction in an aquatic habitat, while the ternate leaf organization could support a relationship to basal eudicots and/or the Early Cretaceous fossil Vitiphyllum. The Friis et al. interpretation was in turn criticized by Crepet et al. (2004) and Crepet (2008). However, analyses by Endress and Doyle (2009) with the D&E backbone nested Archaefructus in Nymphaeales with the highly reduced Hydatellaceae (now treated as one genus, Trithuria; Sokoloff et al. 2008), while analyses with the J/M backbone linked it with Ceratophyllum. An analysis in the context of seed plants as a whole (Doyle 2008) found that it was five steps more parsimonious to link Archaefructus with Trithuria than to place it below the angiosperm crown group.

There have been several more recent observations on Archaefructus; none of these are well enough corroborated to justify a new analysis, but we review them briefly because of the proposed relationship of Archaefructus to Nymphaeales. Critical support for this relationship came from presumed in situ pollen grains studied with SEM by Sun et al. (2001, 2002), namely, their monosulcate aperture, boatlike shape, and continuous tectum. However, Friis et al. (2003, 2011) questioned whether these structures were indeed pollen, because of their irregular size and shape. Without pollen characters, Endress and Doyle (2009) found that the most parsimonious position of Archaefructus with the D&E backbone was nested among basal eudicots, with Euptelea; with the J/M backbone, the connection with Ceratophyllum was strengthened. Eudicot affinities might be supported by the fact that the stamens are borne in pairs, suggesting the dimerous condition of Papaveraceae and other basal eudicots (Drinnan et al. 1994; Endress and Doyle 2009), and by the ternate leaf dissection. However, ternate dissection is inferred to have evolved within Ranunculales (Doyle 2007), which have tricolpate pollen, but there are only doubtful reports of tricolpate pollen before the Albian in this paleolatitudinal belt (see Doyle 2012). Endress and Doyle (2009) scored ovule curvature in Archaefructus as unknown, but Ji et al. (2004) interpreted the ovules of Archaefructus eoflora as orthotropous; if this is confirmed and holds for Archaefructus as a whole, it would strengthen a relationship to Ceratophyllum. Wang and Zheng (2012) interpreted the ovules as “dorsal” (attached to the midrib), combined with laminar in our placentation character (113), rather than ventral as assumed by Sun et al. (2002), which would strengthen a relationship to Nymphaeales. More work is needed to determine whether the seeds have a palisade exotesta (as assumed by Endress and Doyle 2009) or an operculum, a nymphaealean character previously scored as unknown.

Despite the remarkable preservation of Archaefructus, there are so many questions concerning critical aspects of its morphology that we consider its systematic position highly ambiguous. It should certainly not be used to set a minimum age for crown group Nymphaeales in molecular dating analyses.

Implications for the history of Nymphaeales

Although the position of Nymphaeales in most molecular trees implies that the nymphaealean line existed in the Early Cretaceous, other evidence is needed to determine whether the crown group existed then. This question was addressed by a molecular dating analysis of Yoo et al. (2005). Under the assumption that the angiosperm crown group is not much older than the first definite angiosperm fossils (131.8 Mya, Hauterivian), this study concluded that crown group Nymphaeales (not including Trithuria) did not originate until the Eocene (44.6 ± 7.9 Mya). Yoo et al. recognized that this result conflicted with the reports of Monetianthus (Friis et al. 2001) and Microvictoria, from the late Turonian (∼90 Mya) of New Jersey, which an analysis by Gandolfo et al. (2004) linked with Victoria or Victoria plus Euryale. Yoo et al. therefore suggested that these fossils were not crown group Nymphaeales but rather stem relatives.

These conclusions were criticized by Nixon (2008), who reaffirmed the crown group position of Microvictoria and argued that the conflict was more likely due to failure of the dating method due to rate heterogeneity—e.g., if a rapid radiation of Nymphaeales during invasion of aquatic habitats was followed by stasis. However, there is reason to doubt that Microvictoria belongs in Nymphaeales (Endress 2008, p. 855): it has spiral rather than whorled floral phyllotaxis and outermost tepals that did not enclose the flower, and the shape of the floral base is inconsistent with the supposed presence of an inferior ovary. Furthermore, as noted by Yoo et al. (2005), the analysis of Gandolfo et al. (2004) included only Nymphaeales and could not test the hypothesis that Microvictoria belonged elsewhere.

Whatever the status of Microvictoria, our results concerning Monetianthus, Carpestella, and Pluricarpellatia strengthen the view that crown group Nymphaeales, and in fact Nymphaeaceae, are much older than inferred by Yoo et al. (2005)—at least 110 Mya. This was also noted by Taylor et al. (2008), who concluded that Pluricarpellatia and Scutifolium were related to Cabombaceae, and by Friis et al. (2011). Characters of the best-understood fossil, Monetianthus, do not make sense for a stem relative of Nymphaeales, since the most recent common ancestor of the order can be reconstructed as having hypogynous flowers and a smaller number of free carpels containing one ovule each (Endress and Doyle 2009). Crown group angiosperms are almost surely older than Hauterivian (Doyle 2012), but as Yoo et al. (2005) noted, if the relative ages found in their analysis are correct, the existence of crown group Nymphaeales in the Aptian-Albian would imply that angiosperms are far older (∼330 Mya, mid-Carboniferous). Given the broad consistency of the angiosperm fossil record with molecular phylogenies (Doyle 2012), it seems more likely that the conflicts between molecular and fossil dates for Nymphaeales are due to problems of molecular dating methods in dealing with rate heterogeneity.

Dispersed seeds with a palisade exotesta and an operculum from the Aptian-Albian of Portugal and the Potomac Group have been considered evidence for Nymphaeales (Friis et al. 2000b, 2006, 2010a, 2011). However, both features appear to be ancestral in Nymphaeales, including Trithuria (Hamann 1975), so they do not indicate whether these seeds are from stem relatives or crown group members. Furthermore, the report by Yamada et al. (2003) of a small operculum in Trimenia (see appendix) raises the possibility that this feature was once more widespread in the ANITA grade.

Because all extant Nymphaeales are aquatic, except for an obvious reversal to wet terrestrial habitats in the well-nested species Barclaya rotundifolia (Schneider and Carlquist 1995; Feild et al. 2004), it is most parsimonious to assume that the aquatic habit had originated by the time the crown group evolved. Thus, recognition of crown group fossils in the Aptian-Albian indirectly implies that these were aquatic plants, and this is confirmed by the vegetative morphology of Pluricarpellatia, Scutifolium, and Brasenites (Wang and Dilcher 2006; Taylor et al. 2008; Friis et al. 2011).

Anacostia

Anacostia includes four species described by Friis et al. (1997a) based on isolated lignitized and charcoalified carpels at the fruit stage, each with a single enclosed seed, and adhering pollen: Anacostia portugallica and Anacostia teixeirae from Buarcos, Famalicão, and Vale de Agua, Portugal (early Albian); Anacostia virginiensis from Puddledock, Virginia (middle Albian); and Anacostia marylandensis from Kenilworth, Maryland (middle Albian). We also include an axis bearing numerous spirally arranged immature carpels from Puddledock (Friis et al. 1997a, fig. 6A), which Friis et al. associated with Anacostia based on epidermal structure and adhering pollen.

The pollen, which varies from monosulcate to trichotomosulcate, is of a type once compared with monocots because its sculpture grades from coarsely reticulate to much finer on different parts of the grain, a pattern not known in other monosulcate angiosperms (Doyle 1973; Walker and Walker 1984). Such pollen was first identified as Retimonocolpites (Doyle 1973) or Liliacidites (Doyle and Hickey 1976; Doyle and Robbins 1977; Walker and Walker 1984), but because it differs from species originally assigned to Liliacidites in having finer sculpture at the proximal pole and around the middle of the sulcus, rather than at the ends of the grain, it was transferred to the new genus Similipollis by Góczán and Juhász (1984). Because Doyle and Hotton (1991) could find no reports of this pattern of grading in monocots, they questioned whether Similipollis was monocotyledonous. However, Harley (1997) and Dransfield et al. (2008) described similar grading in two derived genera of palms (Chamaedorea, Pseudophoenix).

Friis et al. (1997a) recognized that Anacostia shows no sign of monocot affinities. They noted that it has similarities to several “magnoliid” groups, particularly Canellaceae, Winteraceae, Illiciaceae (=Illicium), and Schisandraceae (=Schisandra), now assigned to Canellales and Austrobaileyales, but they hesitated to relate it to any particular group (cf. Friis et al. 2011). Friis et al. (1997a) emphasized that the seed coat of Anacostia is like that of the taxa listed in having a palisade exotesta. However, this is underlain by a layer of cells with digitate anticlinal walls, which resembles the sclerotic mesotesta of Austrobaileyales (Corner 1976; Takhtajan 1988; Oh et al. 2003) but has no counterpart in Canellales.

Because Similipollis had been compared with monocots, Doyle et al. (2008) analyzed Anacostia as part of a review of Early Cretaceous monocots, using the data set of Endress and Doyle (2009). They used the Puddledock floral axis to score Anacostia as having a pedicel (45), superior ovary and no hypanthium (48), elongate receptacle (49), and apocarpous gynoecium (106). Among characters introduced or redefined since Doyle et al. (2008), this specimen also allows us to score flowers as not in heads (44), carpels not sunken in pits in the receptacle (50), and carpels in more than one whorl or series (96). No characters of the perianth and androecium can be scored. The Puddledock axis shows that the carpels had a distinctly spiral phyllotaxis, but there are few traces of other floral parts below these, so there is no convincing evidence on their identity or arrangement. Probable stamen bases apparently in the same spiral phyllotaxis as the carpels are preserved in another specimen from Puddledock (Crane et al. 1994, fig. 1a), but its relation to Anacostia is uncertain because it has no associated pollen.

As in Doyle et al. (2008), we score pollen size (82), measured by Friis et al. (1997a) as 12–18 μm, as either small or medium, because similar dispersed pollen is larger than 20 μm, suggesting the possibility of shrinkage in the in situ pollen, and nexine thickness (95), which appears thick in TEM sections (Friis et al. 1997a, fig. 3), as either thick or thin (1/2) because the sections are oblique and the nexine is thinner in dispersed pollen. One of the pollen grains figured by Friis et al. (1997a, fig. 8C) has a uniform reticulum and an extended sulcus, so it was probably from a different taxon.

Doyle et al. (2008) treated placentation (113) as unknown, but the basal position of the single seed in the laterally compressed fruits suggests that it had our “ventral” state (0), like the basal ovule of modern taxa such as Illicium and Myristicaceae. The cells with digitate anticlinal walls that make up the inner layer of the seed coat are very similar to those of the sclerotic mesotesta in Illicium (Oh et al. 2003), which supports scoring of the mesotesta character (129) as the same state (1; see appendix for different interpretations of the mesotesta in Austrobaileyales by Yamada et al. 2003). Because there is too little space between the two layers for a fleshy sarcotesta, we score this character (130) as absent.

As in Doyle et al. (2008), our analyses using both backbone trees nest Anacostia within Austrobaileyales (fig. 5), as the sister group of either Illicium plus Schisandra or Schisandra alone. Synapomorphies associating it with Austrobaileyales as a whole are more than one whorl or series of carpels (96) and sclerotic mesotesta (129); it is linked with Austrobaileyales other than Austrobaileya by palisade exotesta (128) and with Illicium plus Schisandra by ascendent ovule orientation (114; modified to horizontal in Schisandra, an autapomorphy in this context). When Anacostia is sister to Illicium plus Schisandra, the two living taxa are united by tri/hexacolpate pollen (84); when Anacostia is linked with Schisandra alone, there is an extra step in this character, but this is balanced by the fact that Anacostia and Schisandra share an elongate receptacle (49). Other positions above Austrobaileya are one step worse, while those with Austrobaileya or on the stem lineage of the order are two steps worse. All positions outside Austrobaileyales are at least three steps less parsimonious.

These results indicate that the crown group of the third ANITA line had originated by the early Albian. Demonstration that Anacostia had spiral phyllotaxis in the perianth and androecium, which is likely in view of the spiral carpel arrangement, would strengthen this conclusion. Fossil evidence for the Amborella line is discussed in connection with Appomattoxia.

Other mid-Cretaceous fossils have been compared with Austrobaileyales, but these comparisons are based on fewer characters. Yamada et al. (2008) assigned a seed from the late Albian of Japan to Trimeniaceae, based on presence of a thick lignified layer that they interpreted as a multilayered exotesta (see appendix) and an operculum, a feature typical of Nymphaeales but not known in Austrobaileyales before it was reported in Trimenia by Yamada et al. (2003; see appendix). Seeds resembling Illicium were described by Frumin and Friis (1999) from the Cenomanian-Turonian of Kazakhstan. Other potential Austrobaileyales are leaves from the lower Potomac (Aptian–early Albian) that Upchurch (1984) compared with Amborella and Austrobaileya based on cuticle structure and leaves (Longstrethia) near the Albian-Cenomanian boundary in Nebraska that Upchurch and Dilcher (1990) assigned to “Illiciales” based on cuticles and venation. Periporate pollen from the late Albian to Turonian of Africa and Brazil, Cretacaeiporites scabratus (Herngreen 1973), was compared with Trimenia by Muller (1981) and Friis et al. (2011), but its exine structure is very different (Sampson and Endress 1984; Ward and Doyle 1994). Pollen called Trisectoris from the Santonian to Paleocene of North America (Tschudy 1970) and the Aptian and early Albian of Brazil (Heimhofer and Hochuli 2010) is like Illicium in having three colpi that join at the poles, such that the grains usually split into three sectors. However, Illicium has uniformly reticulate sculpture, whereas the Early Cretaceous pollen has conspicuous longitudinal “costae” with reticulate sculpture. Pollen more like that of Illicium and Schisandra is known from the Maastrichtian of California (Chmura 1973; Muller 1981).

The variation between monosulcate and trichotomosulcate apertures in pollen of Anacostia (Similipollis) may provide evidence on the origin of the unusual pollen of Illicium and Schisandra, which has three colpi joined at one pole (plus three alternating colpi in most Schisandra species) but differs from the tricolpate pollen of eudicots in orientation of the colpi in the tetrad (Huynh 1976). The three fused colpi have been interpreted as the extended arms of a trichotomosulcus (Doyle et al. 1990; Friis et al. 1997a; Doyle 2005). Similipollis could support this view by representing an earlier state where the trichotomosulcate condition was not yet fixed. As noted by Friis et al. (1997a), the graded reticulum of Similipollis differs from the uniform reticulum of Illicium and Schisandra. However, this does not rule out a relationship, since the only extant taxa with the Similipollis sculpture pattern are palms (Harley 1997; Dransfield et al. 2008) and tricolpate eudicots with finer sculpture at the poles, both clearly unrelated based on other characters. In Anacostia this sculpture appears to be an autapomorphy of an extinct side line that has no direct bearing on relationships.

Couperites

Couperites mauldinensis is based on isolated carpels at the fruit stage with pollen of the Clavatipollenites type on the sessile stigma, described by Pedersen et al. (1991) from Mauldin Mountain, Maryland (early Cenomanian). Friis et al. (1997b) reported an isolated stamen with similar pollen from Puddledock, Virginia, but because it is possible that such pollen is systematically heterogeneous and Puddledock is appreciably older (middle Albian), we base our scoring on the Mauldin Mountain material only.

Discovery of Couperites was a breakthrough in botanical understanding of pollen of the Clavatipollenites type. However, it is important to recognize that the associated pollen differs from the best-known dispersed material from the Aptian (Doyle et al. 1975; Walker and Walker 1984): it has a narrower sulcus with sharply defined, thickened, infolded margins and a tendency for the columellae to detach from the nexine. Pierce (1961) described apparently similar pollen from the Cenomanian of Minnesota as Retimonocolpites dividuus, but the identity of this species is uncertain because of the low magnification of the figures; it may differ in having a sulcus that extends more than halfway around the grain. Brenner (1963) transferred Pierce’s species to Liliacidites, as Liliacidites dividuus, which he reported throughout Zone II of the Potomac Group, but later workers (Doyle and Hickey 1976; Doyle and Robbins 1977; Hickey and Doyle 1977) reserved Liliacidites for pollen with graded sculpture and identified the Potomac pollen as Retimonocolpites cf. or aff. dividuus, which they extended downward into upper Zone I. Some Potomac pollen of this type is probably identical to Clavatipollenites rotundus of Kemp (1968), which enters in the early Albian of England (see Material and Methods). However, the few available EM data show variation in Potomac pollen of this type; a grain from upper Subzone II-B (late Albian) that Walker and Walker (1984) identified as R. dividuus has a thinner nexine than Couperites pollen and transverse ridges on the muri rather than microverrucae. Resolving whether these pollen types form a natural group may require new EM observations and/or discoveries of pollen in situ.

The carpels of Couperites are like those of Chloranthaceae in containing one pendent ovule, but they differ in other characters. The ovule is anatropous rather than orthotropous, which Pedersen et al. (1991) recognized might mean that Couperites was outside the crown group of Chloranthaceae, assuming that the orthotropous ovules of the family are derived (as inferred from molecular trees; Endress and Doyle 2009). The seed coat has two structural layers; the outer is probably a palisade exotesta, which occurs in Nymphaeales and most Austrobaileyales but not in Chloranthaceae.

The phylogenetic position of Couperites was analyzed by Eklund et al. (2004) in a morphological analysis that included 38 species of Recent Chloranthaceae and 10 outgroups, including the ANITA lines and three exemplars of the mesangiosperms. This treatment needs modification in light of the larger number of outgroups and different characters and character state definitions in our present data set, as well as more recent observations.

According to Pedersen et al. (1991), pollen size (82) is 22–25 μm, which is unambiguously medium. Eklund et al. (2004) did not include a character for sculpture on the sulcus membrane (93). However, although the sulcus was narrow, SEM and TEM (Pedersen et al. 1991, figs. 5D, 6D) show that it had at least some verrucae. Another character not used by Eklund et al. (2004) is extra-apertural nexine structure (94). Pedersen et al. (1991) reported thick endexine under the sulcus and thin endexine in other areas, but the latter is comparable to a very thin and discontinuous layer of questionable identity that occurs in Ascarina and other Chloranthaceae, which we include in the state (0) for foot layer only (Doyle 2005).

No characters of floral organization can be scored. If it could be assumed that the flower had other parts, the character for floral base/ovary position (48) could be scored as superior, either with or without a hypanthium (0/1), but because the flower may have consisted of a single carpel, with no other floral parts to define ovary position, we treat this character as unknown, as in living taxa with flowers that consist of one carpel. The carpels have a short stalk that could be the pedicel of a unicarpellate flower, but because it could equally well be the stipe of the carpel, we score floral pedicel (45) as unknown. As in unicarpellate extant taxa and other fossils with isolated carpels, we score carpel fusion (106) as unknown, in order to avoid artifactual steps in cases where a syncarpous gynoecium was reduced to one carpel (Endress and Doyle 2009). The shape of the carpel and the apical ovule attachment suggest that the carpel was ascidiate, but in the absence of developmental data or anatomical markers, we score carpel form (97) and placentation (113) as unknown. The stigma was sessile (101) but is too degraded for scoring of related characters (102–104). Pedersen et al. (1991) interpreted resin bodies in the carpel wall as probable altered oil cells (they are consistent in size, ∼50 μm, to oil cells in the seed coat of Magnolia figured by Corner 1976), but these are not of the intrusive type visible at the surface (107).

Pedersen et al. (1991, p. 588) stated that the seed of Couperites is like that of living Chloranthaceae (except Hedyosmum, which has no lignified layer in the seed coat) in having “cuboid and lignified palisade cells in the testa.” However, as they recognized, potentially related taxa have two sorts of testal palisade layers: Chloranthaceae have an endotesta (derived from the inner epidermis of the outer integument), whereas Nymphaeales and Austrobaileyales (except Austrobaileya) have an exotesta (from the outer epidermis). They left open the question of whether the palisade layer in Couperites was exo- or endotestal, but Friis et al. (2011) and Friis and Pedersen (2011) interpreted it as an exotesta (128), which we accept. In addition, the seed coat had an inner structural layer of longitudinally elongate cells. Although the nature of this layer is not fully established, we follow Pedersen et al. (1991) and Eklund et al. (2004) in interpreting it as a fibrous exotegmen (132), a feature known in Ascarina and Chloranthus.

The analysis of Eklund et al. (2004) found two most parsimonious positions for Couperites: sister to the crown clade of Chloranthaceae and nested within it as the sister group of Ascarina, Sarcandra, and Chloranthus (here designated the “ASC clade”). With the present data set, the inferred relationships of Couperites differ with the two backbone trees.

With the J/M tree (fig. 9B), where Ceratophyllum is well separated from Chloranthaceae, Couperites has one most parsimonious position, as the sister group of Chloranthaceae as a whole. Synapomorphies that link it with the family are two of the pollen characters stressed by Walker and Walker (1984), supratectal spinules (91) and thick nexine (95), while its position below the crown group is a consequence of its anatropous ovule (115), which becomes orthotropous in living Chloranthaceae. Similarities such as sculptured sulcus membrane (93) and one pendent ovule (112, 114) are not evidence for a special relationship, since they are inferred to be ancestral for angiosperms (Endress and Doyle 2009). The palisade exotesta (128) and fibrous exotegmen (132) are convergences with other taxa where they occur. However, three positions within Chloranthaceae are only one step less parsimonious: sister to Hedyosmum, Ascarina, and the ASC clade. The third position is less parsimonious than it was in Eklund et al. (2004) because we rescored the tegmen character (132) in Chloranthus as uncertain (Doyle and Endress 2010): a fibrous exotegmen like that of Ascarina occurs in Chloranthus erectus, one of the two species studied, but not in Chloranthus spicatus (Corner 1976; Endress 1987). Positions outside the chloranthaceous line are at least two steps worse.

With the D&E tree, where Ceratophyllum is linked with Chloranthaceae, Couperites has four most parsimonious positions (fig. 9A): sister to the Chloranthaceae-Ceratophyllum clade, based on the combination of thick nexine and anatropous ovule; sister to the ASC clade, based on the sessile stigma (101), with the three modern genera united by one-layered endoreticulate endotesta (131); linked with Ascarina by fibrous exotegmen; and sister to mesangiosperms as a whole, supported by nexine consisting of foot layer only (94). The positions of Couperites within Chloranthaceae require a reversal from orthotropous to anatropous ovules (115) and convergent origin of the palisade exotesta (128), but because both Ceratophyllum and Hedyosmum share an elongate style, this is counteracted by one less step in the style character (101). If Couperites is located below Ceratophyllum and Chloranthaceae, the ancestral state of the style character in this line is equivocal and it undergoes two steps, but if Couperites is nested in Chloranthaceae, a style is ancestral and there is only one step, namely, loss of the style in the common ancestor of Couperites and the ASC clade (with the J/M tree, there is only one change in the chloranthaceous line, wherever Couperites is located, since the style of Hedyosmum is an autapomorphy). Nine positions outside the Chloranthaceae-Ceratophyllum clade are only one step less parsimonious, including six in the ANITA grade, such as on the stem lineage of Nymphaeales or Austrobaileyales, sister to Trimenia (which has a similar carpel with one pendent anatropous ovule), or sister to the eudicots. Most of these positions are more consistent with the palisade exotesta, which occurs in Nymphaeales, Austrobaileyales, and many basal eudicots.

Still more possibilities emerge when Couperites is added to the analysis together with the four fossils most securely associated with Chloranthaceae (Canrightia, Zlatkocarpus, Pennipollis plant, Asteropollis plant). With the D&E backbone tree (fig. 7A–7D), Couperites is located above Canrightia and Zlatkocarpus in 16 of the 18 most parsimonious trees (fig. 7A, 7B): either below the Chloranthaceae-Ceratophyllum crown clade, with or without the Pennipollis plant, in any of their three possible arrangements; with Ceratophyllum, with or without the Pennipollis plant; or sister to or nested within Chloranthaceae, as the sister group of Hedyosmum plus the Asteropollis plant, the ASC clade, or Ascarina. In most of these trees its anatropous ovule is best interpreted as a reversal (this is equivocal in the two trees where Couperites is just above Zlatkocarpus). Finally, in two trees (fig. 7C, 7D) Couperites is united with the eudicots by palisade exotesta. With the J/M backbone (fig. 7E–7N), Couperites is associated with Chloranthaceae in all 45 trees, but in eight trees it is a stem relative of the family, above Canrightia and/or Zlatkocarpus, with or without the Pennipollis plant, whereas in the other 37 it is nested at various points within the crown group.

In view of these results, a relationship of Couperites to Chloranthaceae is possible but uncertain. It cannot be used as secure evidence for the chloranthaceous line and certainly not for the palynologically most similar genus, Ascarina, which it resembles largely in shared ancestral states. Information that might resolve this problem could include any indication on organization of the flowers that bore the carpels—whether they had a perianth or not, had one carpel or several, or were unisexual or bisexual. The carpels differ from those of Hedyosmum, Zlatkocarpus, and Canrightia in lacking an adnate perianth, but this could mean that they came from either a flower like that of Ascarina or a flower with several free carpels, which might be unrelated to Chloranthaceae. It should be borne in mind that these results do not necessarily apply to other dispersed pollen identified as Clavatipollenites, which may well be systematically heterogeneous, especially considering that most of its features are plesiomorphic.

Asteropollis plant

This taxon, not yet formally described, is based on isolated female flowers with adhering pollen of the Asteropollis type from Torres Vedras, Catefica, Vale de Agua, and Buarcos, Portugal, figured by Friis et al. (1994, fig. 3c, 3d; 1997b, fig. 6.3; 1999, pollen type J.4, figs. 105–107; 2000b, fig. 3E; 2006, fig. 7F–7L; 2011, fig. 8.15), and axes bearing numerous stamens that contain similar pollen from the same localities (Friis et al. 1994, fig. 1; 2006, fig. 7A–7E; 2011, fig. 8.13). Torres Vedras may be either Aptian or earliest Albian, but the other localities are more securely dated as early Albian.

Asteropollis, first described by Hedlund and Norris (1968) from the middle Albian of Oklahoma, differs from Clavatipollenites in having a 4–5-branched sulcus. Similar pollen is known only in the chloranthaceous genus Hedyosmum (Doyle 1969; Muller 1970, 1981; Walker and Walker 1984). The flowers were first figured by Friis et al. (1994) and associated with the pollen by Friis et al. (1997b, 1999), who pointed out their similarity to Hedyosmum. As in Hedyosmum, the female flower consists of one carpel with three tepals on top and three window-like depressions on the sides, while the male axes are also like those of Hedyosmum, in which the individual stamens are interpreted as unistaminate flowers with no perianth or subtending bracts (Endress 1987; Eklund et al. 2004).

Eklund et al. (2004) analyzed the phylogenetic position of this plant based on the female flowers and pollen alone, because of uncertainty that the male structures were associated, but this association is now well established (Friis et al. 2006, 2011). The female flower illustrated by Eklund et al. (2004, fig. 2A) is from Torres Vedras (Friis et al. 1994, fig. 3c; 2006, fig. 7F), whereas the pollen figured by Eklund et al. (2004, fig. 2B), which had a verrucate sulcus with a distinct margin, is from a stamen at Buarcos (E. M. Friis, personal communication, 2004). According to E. M. Friis (personal communication), the pollen adhering to female flowers at Torres Vedras differs in having a “fragmented” reticulum over the sulcus, but the two types are similar in characters in our data set.

The scoring of this plant by Eklund et al. (2004) requires revision along lines discussed for Couperites. In general, in taxa with unisexual flowers, we score inflorescence and flower characters based on the sex with more complex structures (Endress and Doyle 2009). Because female inflorescences are unknown, we score inflorescence type (42) based on the male structures, which we interpret as spikes of unistaminate flowers (as did Friis et al. 2006, 2011), but inflorescence partial units (43) as unknown, because the female flowers may or may not have been borne in cymose units, as in Hedyosmum (which has male spikes and female thyrses, both included in the same state of character 42). Contrary to earlier authors (Endress 1987; Todzia 1988; Eklund et al. 2004; Endress and Doyle 2009), Doria et al. (2012) interpreted the partial units in female inflorescences of Hedyosmum as spikes, rather than monochasial cymes. Under this hypothesis, the total number of bracts within and subtending each unit should be one more than the number of flowers. Doria et al. (2012) claimed this was true for the three species that they studied, but this is not clear from their figures, and in other species the number of bracts is the same as the number of flowers, as expected if each unit is a cyme (Hedyosmum mexicanum: Endress 1987, fig. 23; Hedyosmum brenesii: Todzia 1988, fig. 15A). We therefore continue to score the partial units in Hedyosmum as cymes. Fossil flowers of both sexes lack a pedicel (45). Friis et al. (1994) did not observe floral subtending bracts below the male flowers, but because there is no evidence on whether such bracts occurred in the female inflorescences (as they do in Hedyosmum), we score this character (46) as either present in female and absent in male flowers or absent in all flowers (1/2).

Doria et al. (2012) interpreted the ovary of Hedyosmum as superior rather than inferior, with three connate tepals free from the base of the ovary. However, as shown in their figure 5E and in figures 5, 8, and 16 of Endress (1971), the lobes that develop into the tepals are initiated above the level of the ovary. As support for their interpretation, Doria et al. (2012) cited the superior ovary position in “abnormal bisexual flowers” of Hedyosmum orientale figured by Yamazaki (1992), but these specimens appear to be misidentified flowers of a member of the Ulmaceae; they differ from Hedyosmum in having a papillate stigma, wide ovary locule, and short dorsifixed anthers. Doria et al. (2012) interpreted the distinctive “windows” in the ovary wall below the tepals as schizogenous, as suggested by D’Arcy and Liesner (1981). However, Endress (1971) showed that the windows form well after tepal initiation by morphogenetic differentiation of circular rims on the floral surface. A section in Doria et al. (2012, fig. 4J) that shows a discontinuity between the ovary and an outer layer appears to be a nonmedian section through the edge of a window “frame,” which is continuous with the rest of the flower just above the level of the ovule.

Scoring of perianth organization (53–56) is based on the female flowers. As in other taxa with one whorl of sepaloid tepals, we score tepal differentiation (57) as either all sepaloid or outer sepaloid and inner petaloid (Endress and Doyle 2009), on the assumption that the one whorl of sepaloid tepals could be derived from the outer whorl of either polycyclic type. Eklund et al. (2004) scored the tepals as free rather than connate, following terminology used by Todzia (1988) to describe variation within Hedyosmum, but Doria et al. (2012) argued that all Hedyosmum species have fused tepals. Closer examination shows that the variation noted by Todzia (1988) and Eklund et al. (2004) is between tepals that are fused only at the very base and much more fused. As illustrated by Friis et al. (1994, fig. 3d), the tepals in the fossil are basally fused into a shallow ring around the ovary, as in H. orientale (Yamazaki 1992, fig. 3), one of the species that Todzia (1988) and Eklund et al. (2004) described as free. Our present character (60), which is defined in a broader angiosperm context, treats this condition as fused, so we score both the Asteropollis plant and Hedyosmum as fused, thus bringing our data in line with Doria et al. (2012). As with other unistaminate flowers, we score all characters of androecial organization except single stamen (62) as unknown. Friis et al. (2011, p. 182) stated that the connective “extends apically into a short sterile extension,” but their figures show that the apex (72) was like that of taxa that Eklund et al. (2004) and Endress and Doyle (2009) scored as truncate or rounded. The stigma is usually broken off, but Friis et al. (2000b, fig. 3E; 2006, fig. 7I) figured one flower with a large, basally constricted stigma, which is more like the stigma of Hedyosmum, which we score as having a style (101), than the sessile stigma of other Chloranthaceae. Seeds have not been figured, but according to Friis et al. (2011, p. 183) the carpel contained a single orthotropous ovule (112, 115).

In Eklund et al. (2004), the Asteropollis plant was either sister to Hedyosmum or nested within the basal grade of the genus. The present analysis, which treats living Hedyosmum as a single taxon, confirms that the fossil is related to Hedyosmum (fig. 10). With the D&E backbone tree, synapomorphies that link it with Chloranthaceae and Ceratophyllum are sessile flower (45), one stamen (62), embedded pollen sacs (73), one carpel (96), and orthotropous ovule (115). Despite the remarkable similarity of the fossil and Hedyosmum, their only unequivocal synapomorphy is the branched sulcus (86). They share several other derived features, namely, absence of bracts subtending the stamens (46), inferior ovary (48), one perianth whorl (56), and basally fused tepals (60), but where these characters arose is equivocal, since they are uncertain or inapplicable in Ceratophyllum, which has female flowers that consist of a single carpel. The last three characters were also scored as unknown in Ascarina, which also has naked female flowers, and in Sarcandra and Chloranthus, where the single stamen or three-lobed androecium is attached to the back of the carpel, which might or might not mean that the ovary is inferior. As a result, positions sister to Chloranthaceae and/or Ceratophyllum and to the ASC clade are only one step less parsimonious, but the best position outside the whole line (with Myristicaceae) is eight steps worse. With the J/M backbone, a sister group relationship of the Asteropollis plant to Hedyosmum is two steps more parsimonious than the best alternatives (sister to Chloranthaceae and to the ASC clade). The fossil is associated with Chloranthaceae by the five synapomorphies listed above and with Hedyosmum by both the branched sulcus and the absence of bracts subtending the stamens. Its best position elsewhere, with Ceratophyllum, is six steps worse. It is also sister to Hedyosmum in all analyses that include several fossils (figs. 6–8).

The unique “windows” in the fruit wall (not included in our data set) would strengthen a link between the Asteropollis plant and Hedyosmum rather than Canrightia or Zlatkocarpus, which both have an inferior ovary but no windows. However, it would not affect the relative parsimony of other positions among living Chloranthaceae, as it would not be justifiable to score such a character in the ASC clade, where ovary position cannot be defined.

The Asteropollis plant presents one of several conspicuous cases in which fossils that closely resemble an extant clade are much older than ages of the crown clade inferred from molecular dating analyses. Another concerns Early Cretaceous fossils that resemble Ephedra (Rydin et al. 2004, 2010). This has led to views that the fossil and molecular data are in conflict (Rydin et al. 2004; Friis et al. 2005; Nixon 2008). However, this is not necessarily true if the fossils have no derived features that arose within the crown group, in which case they could be stem relatives (Doyle and Donoghue 1993; Pirie and Doyle 2012); the problem becomes explaining the long period of morphological stasis between the time of the fossil and the radiation of the crown clade (cf. Friis et al. 2005). In the case of Ephedra, closer examination showed that the fossils are more plesiomorphic than living Ephedra in having more valves surrounding the ovule, supporting the stem relative hypothesis (Rydin et al. 2010). In the case of Hedyosmum, Eklund et al. (2004) found seven most parsimonious positions for the Asteropollis plant, one sister to Hedyosmum and six nested within the crown group. Of these, the latter would conflict with molecular dating analyses (Zhang and Renner 2003; Antonelli and Sanmartín 2011; Zhang et al. 2011), which have given Eocene to Oligocene ages for the crown group. Eklund et al. (2004) took these results as support for the view that the Asteropollis plant was a stem relative of Hedyosmum, rather than a crown group member. They noted one character (not used in their analysis) that might favor a stem position, namely, the fact that the sulcus in the Asteropollis grain in their figure 2B (from Vale de Agua) had a distinct margin and verrucate sculpture, as opposed to an indistinct margin and a fragmented reticulum in extant species. However, according to E. M. Friis (personal communication, 2004), the sulcus in grains from Torres Vedras has a fragmented reticulum.

Despite these problems, the Asteropollis plant does provide a secure minimum age of early Albian for the crown node of Chloranthaceae. Use of dispersed pollen for this purpose is problematic, because many authors have used the name Asteropollis for trichotomosulcate pollen as well as pollen of the original type with more than three sulcus branches. Heimhofer et al. (2007) recorded Asteropollis throughout the Aptian and Albian in well-dated marine sections in Portugal, but the only figured specimen with a four-branched sulcus was early Albian, and U. Heimhofer (personal communication, 2009) confirmed that he observed no four-branched grains in the Aptian. Asteropollis has also been reported from the Aptian of Argentina (Archangelsky et al. 2009), but the aperture condition is trichotomosulcate (V. Barreda, personal communication, 2009), poorly defined and irregular (Llorens 2003; M. Llorens, personal communication, 2009), or polycolpoidate (Prámparo et al. 2007; M. Prámparo, personal communication, 2009). The trichotomosulcate pollen may well be related to Asteropollis and Hedyosmum, but because the several-branched type is unique to Hedyosmum today, whereas trichotomosulcate pollen is more widespread, we consider this uncertain (see also Friis et al. 2011). It therefore seems most prudent to base any calibrations on the flowers, but there are stratigraphic problems associated with these too. As discussed in Material and Methods, the Vale de Agua and Buarcos localities are probably early Albian, but Torres Vedras is older, either earliest Albian or Aptian. The conservative approach would therefore be to use the Asteropollis plant to set a minimum age of early Albian (∼110 Mya) for crown group Chloranthaceae.

Zlatkocarpus

Zlatkocarpus is based on compressions of female inflorescences at the fruit stage, Zlatkocarpus brnikensis from Brník and Zlatkocarpus pragensis from Hloubětín-Hutě, Czech Republic (middle Cenomanian), with adhering pollen of the Retimonocolpites type. The latter species was first described by Kvaček and Eklund (2003) as Myricantheum pragense and segregated as the new genus Zlatkocarpus by Kvaček and Friis (2010).

Zlatkocarpus shows a combination of characters not seen in any living Chloranthaceae. It has spikes of female flowers that recall Ascarina, but it is like Hedyosmum in having a presumed reduced perianth adnate to the ovary. However, its pollen differs from that of Hedyosmum in having a normal sulcus, and the stigma is sessile, as in the other living genera. In its exine sculpture, it is unlike both Hedyosmum and Ascarina and like Sarcandra and Chloranthus in having smooth rather than spinulose muri.

The spikes (known to be compound in Z. pragensis, a character not included in our data set) bear sessile flowers in the axils of bracts. Because male inflorescences are unknown, we score floral subtending bracts (46) as either present in all flowers or present in female flowers but absent in male (0/1). The lower part of the carpel is surrounded by a cup, which has one abaxial and one or two adaxial tips (with reference to the inflorescence axis) in Z. brnikensis. Kvaček and Friis (2010) interpreted the cup as most likely a perianth of fused tepals, but Friis and Pedersen (2011, p. 24) described it as a hypanthium, two alternatives that are difficult to distinguish. Under either interpretation the fossil has the inferior ovary state in character 48, which covers both fully and partially inferior. Kvaček and Friis (2010) considered the possibility that the cup consisted of fused bracts of a reduced cyme, but in that case we would expect not one abaxial bract aligned with the subtending bract but two bracts, one on either side. Hence we score the flower as having a perianth made up of one whorl of sepaloid tepals. Because the number of tips on the cup is variable, we score perianth merism (55) as unknown. Because the rim of the cup could be formed by laterally connate tepals or the edge of a hypanthium bearing separate tepals, we score tepal fusion (60) as unknown.

With pollen size 12–18 μm in Z. brnikensis and 8–10 μm in Z. pragensis, we score pollen size (82) as either medium or small (1/2) to allow for possible shrinkage. Kvaček and Friis (2010) did not describe the membrane of the sulcus (93), which is usually strongly infolded, but verrucae are visible at the end of the sulcus in the obliquely flattened grain in their figure 3B.

The gynoecium can be scored as consisting of one carpel (96) with no style (101) and an extended stigma (102). Kvaček and Eklund (2003, fig. 5B) illustrated a macerated fruit of Z. pragensis containing one ovule, which they described as orthotropous, with a notch at the lower end that they identified as the micropyle. However, it is possible that this notch is a random fracture, so we follow J. Kvaček (personal communication, 2010) in treating ovule curvature (115) as unknown. No other ovule or seed characters can be determined, but the fruit is a berry (123–125).

With the D&E backbone tree (fig. 11A), Zlatkocarpus is associated with Chloranthaceae and Ceratophyllum by sessile flowers (45) and one carpel (96); two advances that it shares with Hedyosmum but are inapplicable in other Chloranthaceae and Ceratophyllum are inferior ovary (48) and one perianth whorl (56). Its most parsimonious position is sister to the ASC clade, based on a shift to sessile stigma (101) and retention of a perianth (53). However, it only “costs” one more step to attach Zlatkocarpus to all other branches within the Chloranthaceae-Ceratophyllum clade, except Sarcandra and Chloranthus, and to its stem lineage. With the J/M backbone too (fig. 11B), Zlatkocarpus may be sister to the ASC clade, but it has two other most parsimonious positions, sister to Chloranthaceae and to Hedyosmum, because polarity of the sessile stigma character is equivocal.

With both backbone trees, the most parsimonious position of Zlatkocarpus outside the chloranthaceous line, with Myristicaceae (which also have unisexual flowers with one perianth whorl and one carpel), is three steps less parsimonious. With the J/M backbone, a sister group relationship to Ceratophyllum is four steps worse. Kvaček and Friis (2010) noted that similar pollen occurs in Portuguese fossils assigned to Araceae (Friis et al. 2010b) and suggested that a monocot affinity for Zlatkocarpus cannot be excluded. However, its best positions in monocots, with Acorus and Araceae, are seven steps less parsimonious with the D&E tree and eight steps less parsimonious with the J/M tree.

The fact that Zlatkocarpus has a perianth is significant for floral evolution in Chloranthaceae. With the J/M tree, with or without Zlatkocarpus, the perianth of Hedyosmum is an ancestral feature retained from the base of the angiosperms (as in Doyle et al. 2003). By contrast, with the D&E tree and no fossils, because Ceratophyllum has no perianth, it is equally parsimonious to assume that a perianth was ancestral in the Chloranthaceae-Ceratophyllum line and lost twice, in Ceratophyllum and the ASC clade, or that it was lost on the stem lineage of the two groups and regained in Hedyosmum. However, the former scenario is favored if Zlatkocarpus is nested within Chloranthaceae as the sister group of the ASC clade.

Although Zlatkocarpus is nested within Chloranthaceae when added to the D&E tree by itself, it is more basal in analyses that include other chloranthoid fossils. In both trees found when Canrightia, Zlatkocarpus, the Pennipollis plant, and the Asteropollis plant are added to the D&E backbone (fig. 6A, 6B), Zlatkocarpus is attached to the stem lineage of Chloranthaceae and Ceratophyllum, immediately above Canrightia. It has the same position when either Couperites or Appomattoxia is added as well (figs. 7A–7D, 8A–8D). In all these trees it is linked with the living taxa by unisexual flowers (47) and one carpel (96) but more basal because it lacks supratectal spinules (91), which unite the Pennipollis plant and living Chloranthaceae. However, in analyses with the J/M backbone and several chloranthoid fossils (figs. 6C–6K, 7E–7N, 8E–8O), Zlatkocarpus may be either attached to the stem lineage of Chloranthaceae or nested within the crown group, at the base of the line to either Hedyosmum or the ASC clade.

Pennipollis plant

This plant was reconstructed by Friis et al. (2000a) based on association of Pennipollis pollen with isolated carpels (Pennicarpus tenuis) and stamens (Pennistemon portugallicus) and a fragment of a multistaminate axis, from Vale de Agua and Buarcos, Portugal (early Albian). They interpreted the staminate axis as a spike of male flowers that consist of a single stamen. The carpels contain a single orthotropous ovule.

Monosulcate pollen of the Pennipollis type, characterized by an unusually coarse reticulum that tends to detach from the nexine, is a conspicuous element in Aptian-Albian palynofloras worldwide (Penny 1988a; Friis et al. 2000a, 2011). Such pollen was first described by Brenner (1963), who assigned it to Peromonolites, a genus for monolete spores with a perispore, as Peromonolites peroreticulatus and Peromonolites reticulatus. However, Norris (1967) and Doyle (1969) suggested that it was monosulcate and angiospermous, and it was transferred to Liliacidites by Singh (1971) and Retimonocolpites by Doyle et al. (1975). Using SEM and TEM, Doyle et al. (1975) and Walker and Walker (1984) showed that this pollen resembles “Clavatipollenites” in having supratectal spinules and a thick nexine made up of foot layer, except for some endexine under the sulcus, but differs in having no columellae. It was transferred by Juhász and Góczán (1985) to the new genus Brenneripollis, in which they included species both with and without columellae. Because Juhász and Góczán (1985) described the type species of Brenneripollis, Brenneripollis pellitus, as having columellae, Friis et al. (2000a) transferred pollen of Brenner’s original type to the new genus Pennipollis, expressly defined as lacking columellae, as P. peroreticulatus. They noted that their in situ pollen had a thin layer of fine granules below the tectum.

Based on its exine characters, Doyle and Hotton (1991) suggested that Pennipollis was produced by extinct relatives of Chloranthaceae that had lost their columellae. By contrast, Friis et al. (2000a) argued that the combination of a reticulate tectum and granular infratectum is known only in some Alismatales, including some Araceae, and that the Pennipollis plant was therefore a monocot. The most similar modern pollen that we know is that of Aponogeton (=Aponogetonaceae, Alismatales), not cited by Friis et al. (2000a), which is monosulcate and has a reticulum with supratectal spinules and infratectal granules (Thanikaimoni 1985). However, Pennipollis differs from Aponogeton and other Alismatales and resembles Chloranthaceae in having a thick nexine consisting of foot layer. Affinities of Pennipollis with Araceae were rejected by Wilde et al. (2005) and Hesse and Zetter (2007), who favored a relationship to Chloranthaceae, particularly Ascarina. However, the characters in which Pennipollis is most like Ascarina (unbranched sulcus, supratectal spinules) are probable symplesiomorphies that do not support a special relationship to the living genus.

Because Friis et al. (2000a) interpreted the Pennipollis plant as a monocot, Doyle et al. (2008) analyzed its position in their review of Early Cretaceous monocots, using the data set of Endress and Doyle (2009). However, new data and arguments require a few minor changes in character scoring.

As in Doyle et al. (2008), we assume that the stamen-bearing axis was a spike of naked, unistaminate flowers with no subtending bracts, as favored by Friis et al. (2000a, 2011), rather than a multistaminate flower. Because female inflorescences are not known, we score inflorescence units (43) as unknown and floral subtending bracts (46) as either present in female and absent in male flowers or absent in all flowers (1/2), as for the Asteropollis plant. Doyle et al. (2008) scored floral base/ovary position (48) as superior, either with or without a hypanthium (0/1), but as discussed for Couperites, because of lack of information on organization of the female flowers, we have rescored this character as unknown.

Friis et al. (2000a, 2011) described stamen dehiscence as extrorse, but this referred to orientation relative to the inflorescence axis, whereas for homology assessment it should be defined relative to the axis of the individual flower. If the stamen is abaxial relative to the inflorescence axis (anterior), it is extrorse, whereas if it is adaxial (posterior), it is introrse. Because stamen position in these highly reduced flowers is unknown and there are no clues such as orientation of the vascular bundle in the stamen, we score orientation of dehiscence (75) as either introrse or extrorse (0/2), as in Doyle et al. (2008). As discussed in Doyle et al. (2008), analogies with Chloranthaceae and other unistaminate taxa, where the stamen is abaxial, would favor the extrorse interpretation, but use of this argument would depend too much on the assumption that the fossil and Recent taxa were related. Doyle et al. (2008) did not include the character for food bodies on stamens or staminodes (69), but the stamens seem well enough preserved to score food bodies as absent.

Both the associated pollen and similar dispersed grains (Doyle et al. 1975; Walker and Walker 1984) measure less than 20 μm, so we score pollen size (82) as small. Following Friis et al. (2000a), Doyle et al. (2008) scored infratectal structure (87) as granular; Hesse and Zetter (2007) questioned this interpretation, but convincing granules are visible on the inner surface of the muri illustrated by Friis et al. (2010b, fig. 2d). Friis et al. (2000a) described the nexine (94) as consisting of a thick foot layer and a very thin endexine, the latter thickening under the aperture, but like the similar nexine with discontinuous endexine in Couperites and Chloranthaceae, we include this in the state for foot layer only (0).

Friis et al. (2000a) tentatively interpreted the carpel as having a sessile stigma, but to be cautious Doyle et al. (2008) scored the style character (101) as unknown. However, because Friis et al. (2000a) reported not observing a style in over 100 fruits examined, we have rescored style as absent. Friis et al. (2000a, 2011) considered the single ovule to be most likely orthotropous (115) but cautioned that an anatropous curvature cannot be ruled out. However, as argued in Doyle et al. (2008), the shape of the seed suggests an orthotropous ovule with the chalaza displaced toward one side of the base, as in Amborella (Endress and Igersheim 2000; Tobe et al. 2000) and Chloranthaceae (Endress 1987; Yamada et al. 2001). Friis et al. (2000a) did not characterize ovule direction (114), but Friis et al. (2011) interpreted the seed as “apparently basally attached” (ascendent), because the micropylar end appeared to be directed toward the presumed stigmatic end of the fruit (E. M. Friis, personal communication, 2006). However, as discussed in Doyle et al. (2008) and Endress (2011), asymmetry of the base of the ovule of the sort seen here is generally correlated with apical and pendent ovule attachment (as in Amborella and Chloranthaceae), whereas basal and ascendent ovules (as in Piperaceae and some Araceae) have a symmetrical base. Furthermore, the outline of the enclosed seed in the carpel shown in figure 6A of Friis et al. (2000a) appears to be more asymmetrical toward the presumed stigmatic end of the carpel, like the chalazal end of the isolated seed in their figure 6G, which would support interpretation of the ovule as pendent. We therefore score ovule direction (114) as unknown.

Only the outer cuticle of the seed coat is preserved, so there is no evidence on the number of integuments (116) or cell layers. This precludes scoring of most seed coat characters, which are defined in terms of the integument (outer, inner) and original cell layer (outer or inner epidermis, mesophyll) from which each seed coat layer is derived. In many extant unitegmic angiosperms, it has been inferred that the single integument is derived from both original integuments (Kelley and Gasser 2009; Endress 2011). Under this hypothesis, the outermost cell layer in the seed coat can be interpreted as an exotesta and the innermost layer as an endotegmen, but intermediate layers cannot be identified (they may not even exist: in Ceratophyllum the integument is only two cells thick). Application of our tegmen character is also problematic, since it covers modifications of both the exo- and endotegmen. In living unitegmic taxa in our data set (Siparunaceae, Ceratophyllum, Circaeaster) we therefore score only the exotesta character (128) and treat others (129–132) as unknown.

Consistent with this reasoning, Doyle et al. (2008) scored all seed coat characters of the Pennipollis plant except exotesta as unknown. They scored exotesta as either unspecialized or with tabular cells (0/2), but because the elongation of cells is not as marked as in those monocots scored as having tabular cells (Aponogeton, Scheuchzeria), we have rescored this character as unspecialized (0).

Both Doyle et al. (2008) and the present analysis strongly contradict the hypothesis that the Pennipollis plant was a monocot. When the fossil is added to the J/M tree (fig. 12B), its most parsimonious position is sister to Chloranthaceae; when it is added to the D&E tree (fig. 12A), it is sister to the Chloranthaceae-Ceratophyllum clade. With the J/M tree, synapomorphies uniting it with Chloranthaceae are unisexual flowers (47), one stamen (62), supratectal spinules (91), thick nexine (95), and orthotropous ovule (115); it is located below the crown group because it has protruding rather than embedded pollen sacs (73) and either introrse or extrorse rather than latrorse anthers (75). Its next-best position, one step less parsimonious, is with Hedyosmum, supported by absence of bracts in the male inflorescences (46). Its best position outside the chloranthaceous line, with Ceratophyllum, is four steps worse, while its best positions in monocots, with Acorus and Aponogeton, are eight steps worse, despite the similar exine in Aponogeton, because these plants have such different multiparted flowers. With the D&E tree, the Pennipollis plant is linked with Chloranthaceae and Ceratophyllum by absence of bracts in the male inflorescence (46), a feature of Ceratophyllum and Hedyosmum, one stamen, thick nexine, and orthotropous ovule, while its protruding pollen sacs place it below the crown group. However, positions sister to Ceratophyllum, Chloranthaceae, and the ASC clade are only one step worse. Its best positions in monocots, again with Acorus and Aponogeton, are seven and eight steps worse, respectively.

The basal (ascendent) seed attachment favored by Friis et al. (2011) would contrast with the apical (pendent) condition in Chloranthaceae. However, if this character (114) is scored as ascendent rather than unknown, the relative parsimony of the positions in monocots improves by only one step or not at all for a relationship with Acorus in the J/M tree.

With the J/M backbone and the four most securely associated chloranthoid fossils, the position of the Pennipollis plant is more ambiguous: it is sister to Chloranthaceae, above Canrightia and Zlatkocarpus (fig. 6D); sister to Hedyosmum and the Asteropollis plant (fig. 6C, 6F, 6G), supported by loss of bracts in the male spikes (46) and/or supratectal spinules (91), depending on the position of Zlatkocarpus; sister to the ASC clade (fig. 6E), based on extended connective apex; sister to Ascarina (fig. 6H–6J), supported by supratectal spinules (equivocally in fig. 6J); or sister to Sarcandra and Chloranthus (fig. 6K). In the 45 trees found when Couperites is also added (fig. 7E–7N), the Pennipollis plant is either a stem relative of Chloranthaceae, above Zlatkocarpus, with or without Couperites, or nested within the crown group, on the line to Hedyosmum and the Asteropollis plant, the ASC clade, or Ascarina, with or without Couperites, or sister to Sarcandra plus Chloranthus. In the 48 trees found with the four main fossils and Appomattoxia (fig. 8E–8O), the Pennipollis plant is in one of the five positions found with four chloranthoid fossils (fig. 6C–6K); linked with Appomattoxia, as the sister group of either crown group Chloranthaceae or the Asteropollis-Hedyosmum clade; or sister to Appomattoxia plus the Asteropollis-Hedyosmum clade.

Analyses with the D&E backbone tree and other chloranthoid fossils raise the intriguing possibility that the Pennipollis plant is a stem relative of Ceratophyllum. In one of the two trees found with the four most securely associated fossils (fig. 6A), the Pennipollis plant is sister to the Chloranthaceae-Ceratophyllum clade, above Canrightia and Zlatkocarpus, based on its combination of supratectal spinules and protruding pollen sacs, but in the other tree (fig. 6B) it is linked with Ceratophyllum by a shift from latrorse to introrse or extrorse anthers (75). In the 18 trees found when Couperites is also added (fig. 7A–7D), where Canrightia and Zlatkocarpus are basal, the Pennipollis plant is either basal to the crown clade, alone or with Couperites, in all three arrangements, or linked with Ceratophyllum, alone or with Couperites, in all three arrangements. When Appomattoxia is added, the Pennipollis plant may be in the same two positions found with the four fossils (fig. 8C, 8D); sister to the ASC clade (fig. 8A), based on extended connective apex (72) and a reversal to fleshy fruit wall (123), or sister to Appomattoxia and Ceratophyllum (fig. 8B), based on introrse or extrorse anthers and intermediate infratectum (87), as in Appomattoxia (counted as a step toward the granular structure of Pennipollis because the character is ordered). Implications of a relationship of the Pennipollis plant and Ceratophyllum are discussed in the section on Appomattoxia.

These analyses do not address the possibility that the Pennipollis plant was nested within Araceae, a family that is reported from the Aptian-Albian (Friis et al. 2004, 2010b), but as discussed in Doyle et al. (2008) this is unlikely. Although various Araceae have highly reduced unisexual flowers and exine similarities, these features do not occur in the same living clades, and in the context of current phylogenies (Cabrera et al. 2008; Cusimano et al. 2011) it is unparsimonious to assume that they occurred together in the past. The stamens of Araceae (illustrated in Mayo et al. 1997) are also very different from those of the Pennipollis plant.

Friis et al. (2010b, 2011) rejected a relationship of the Pennipollis plant to Chloranthaceae, but their main argument was the fact that the combination of reticulate tectum and granular infratectum is not known outside monocots. However, the spinulose muri are not fundamentally different from those of Chloranthaceae, and the granular infratectum is only one character out of many. As with any character, there is no reason to assume that it could not have originated by homoplasy in extinct relatives of a group where it does not occur today, especially considering the many shifts from columellar to granular structure elsewhere in angiosperms (Doyle 2009). Except for its coarser reticulum and absence of columellae, Pennipollis resembles Barremian-Aptian monosulcates that do have columellae (e.g., Retisulc-dentat; Hughes et al. 1979; Hughes 1994), and some Pennipollis-like grains in the earliest Aptian have rare columellae (Retisulc-dubdent; Hughes et al. 1979; Penny 1988a; Hughes 1994). These problems might be resolved easily if the floral remains were associated with vegetative parts.

Friis et al. (2000a, 2011) stated that Pennipollis ranges from the Barremian to the Cenomanian or Turonian, but the oldest well-dated records are early Aptian, confirming the use of Pennipollis as a guide fossil for Aptian and younger sediments (Penny 1988a; Doyle 1992; Heimhofer et al. 2007). Hughes et al. (1979) labeled the oldest grains of the Pennipollis type (Retisulc-dubdent) as Barremian-Aptian, but these were from the upper Vectis Formation, which has been redated by magnetostratigraphy as earliest Aptian (Kerth and Hailwood 1988; Allen and Wimbledon 1991; Hughes 1994). Pennipollis also appears above the base of the Aptian in Portuguese marine sections (Heimhofer et al. 2005, 2007).

Canrightia

Canrightia resinifera is based on lignitized and charcoalified flowers and fruits, one with an attached stamen, and associated pollen of the Retimonocolpites type, from Catefica and other early Albian localities in Portugal, described by Friis and Pedersen (2011) using X-ray microtomography. Some of these fossils were illustrated by Friis et al. (1999). A conspicuous feature is the presence of resin bodies in the fruit wall, assumed to be altered oil cells, which appear to be comparable to the intrusive oil cells of Sarcandra and Chloranthus (Endress and Igersheim 1997).

These flowers are of special significance for the origin of Chloranthaceae. They resemble Hedyosmum, the Asteropollis plant, and Zlatkocarpus in having a perianth adnate to the ovary, but they differ in being bisexual, with scars of approximately four stamens just above the ring of highly reduced tepals. As in Chloranthaceae the ovules are orthotropous and pendent, but the gynoecium consists of two to five fused carpels. As Friis and Pedersen (2011) noted, this means that in several respects Canrightia is more like flowers of Piperales, which are usually bisexual, with either one whorl of three tepals (Lactoris, Aristolochiaceae, Hydnoraceae) or no perianth (Saururaceae, Piperaceae), and with several carpels, which contain orthotropous ovules in Saururaceae and Piperaceae.

Friis and Pedersen (2011) analyzed the phylogenetic position of Canrightia using the data set of Doyle and Endress (2010). Here we make several minor changes in scoring, many following policies adopted in Endress and Doyle (2009) when parts are absent or hard to interpret.

Inflorescences are not known, but some flowers have a subtending bract (46) at the base. This suggests that the inflorescence units (43) were single flowers, but without direct evidence we follow Friis and Pedersen (2011) in scoring this character as unknown. We have scored some floral characters that Friis and Pedersen (2011) did not score: carpels not sunken in pits in the receptacle (50), floral apex not protruding (52), and tepals all sepaloid or outer sepaloid and inner petaloid (57). Conversely, we have not scored other characters that Friis and Pedersen (2011) did score: presence of petals (58) and petal nectaries (59), treated as unknown in flowers with one perianth whorl; fusion of outer tepals (60), for reasons like those cited in Zlatkocarpus; and stamens not in double positions (66), which requires better evidence on the relation of the stamens to perianth parts. The perianth appears to be represented by four tips on the cup surrounding the ovary, but because this is usually unclear and the merism of the gynoecium is variable, we follow Friis and Pedersen (2011) in treating perianth merism (55) as unknown. The attached stamen seems too poorly preserved to score food bodies (69) and connective apex (72), which Friis and Pedersen (2011) described as prominently extended, but may be like the rounded apex in the Asteropollis plant, but stamen dehiscence (75) is latrorse rather than unknown.

Friis and Pedersen (2011, p. 24) did not score pollen aperture (84) and elongate versus round distal aperture (85), on the grounds that it is uncertain whether the elongated sulcus was polar. However, as discussed below, the fact that the furrow did not extend all the way around the grain strongly implies that it was polar. Friis and Pedersen (2011, p. 8) did not score nexine thickness (95), but because they stated that SEM of broken grains shows a “relatively thin” foot layer, we score it as thin.

We concur with Friis and Pedersen’s (2011) interpretation of the gynoecium as parasyncarpous (106), usually with four main vascular bundles alternating with four ovules, which were vascularized from the upper part of the ovary. This implies that the bundles were most likely synlateral, but because the positional relationships of the carpels and ovules are uncertain, we follow Friis and Pedersen (2011) in scoring placentation (113) as unknown. Friis and Pedersen (2011) scored long hairs on or between the carpels (108), curved hairs on the carpels (109), and dorsal (110) nectaries as absent, but we question whether these characters would be preserved and therefore score them as unknown. They scored the chalaza (113) as unextended, but although it is theoretically possible for an orthotropous ovule to have an extended chalaza, the distinction between the two states can be made easily only in anatropous ovules (Periasamy 1962), so we follow our previous procedure (Endress and Doyle 2009) of scoring orthotropous ovules as unknown. We follow Friis and Pedersen (2011) in considering the endotesta (131) the only lignified layer of the seed coat. Friis and Pedersen (2011) scored mesotesta (129) as unlignified and sarcotesta (130) as absent, but these characters are inapplicable because the outer integument was only two cells thick (119). They identified the innermost seed coat layer of nonlignified, apparently tannin-filled cells as an endothelium, a character not included in either their or our data set; the most similar layer in our extant taxa is the endothelium of Lactoris, but this is collapsed at maturity (Takhtajan 1988; Tobe et al. 1993).

When Friis and Pedersen (2011) added Canrightia to the Doyle and Endress (2010) data set, its best position was sister to Chloranthaceae with the J/M backbone tree and sister to Chloranthaceae and Ceratophyllum with the D&E tree. Friis and Pedersen (2011) stated that a position nested within Piperales, sister to Saururaceae and Piperaceae, was two or three steps less parsimonious, depending on the backbone tree, but it appears to be three or four steps less parsimonious in their figure 19. Despite our different scoring decisions, we obtained essentially the same results.

With the D&E backbone (fig. 13A), Canrightia is linked to Chloranthaceae and Ceratophyllum by sessile flowers (45), inferior ovary (48), one perianth whorl (56), and orthotropous ovule (115), while the crown clade shows a shift to unisexual flowers (47) and reduction to one stamen (62) and one carpel (96). Its next-best positions, which are two steps less parsimonious, are nested within Chloranthaceae, as the sister group of the ASC clade, supported by sessile stigma (101, a reversal) and one-layered endoreticulate endotesta (131), or linked with Sarcandra and Chloranthus by bisexual flowers (47), a reversal with this tree, and intrusive oil cells in the carpel wall (107). The best positions outside the Chloranthaceae-Ceratophyllum line are three steps worse: (a) sister to mesangiosperms as a whole and (b) linked with Saururaceae and Piperaceae by sessile flowers and orthotropous ovule, intrusive oil cells in the carpel wall, and latrorse stamen dehiscence (75).

With the J/M backbone (fig. 13B), Canrightia is sister to Chloranthaceae, based on the same four synapomorphies that link it with Chloranthaceae and Ceratophyllum with the D&E backbone, plus one stamen whorl (65), and the lack of synapomorphies of crown Chloranthaceae, which include not only one stamen and one carpel but also thick nexine (95). Again, a position sister to the ASC clade is two steps worse, but the position sister to Sarcandra and Chloranthus is three steps worse, rather than two; this is because bisexual flowers are not necessarily a shared derived state of the three taxa but could equally well be ancestral in Chloranthaceae. A position sister to Saururaceae and Piperaceae is only two steps worse, supported by the same four synapomorphies listed above.

Canrightia is basal on the Chloranthaceae-Ceratophyllum or Chloranthaceae line in nearly all trees found in analyses that include other chloranthoid fossils, most often followed by Zlatkocarpus, which is associated with the remaining taxa by a shift to unisexual flowers and reduction to one carpel (figs. 6, 7A–7M, 8). The only exception is one tree found with the J/M backbone, Canrightia, Zlatkocarpus, the Pennipollis plant, the Asteropollis plant, and Couperites, in which Canrightia is nested within Piperales, as the sister group of Saururaceae and Piperaceae (fig. 7N).

These results imply that Canrightia diverged from the chloranthaceous line at an intermediate stage in reduction of the original multiparted angiosperm flower. Canrightia was clearly not directly ancestral to Chloranthaceae, since it coexisted with more derived forms, so some of its features may be autapomorphic rather than primitive. Parsimony optimization indicates that this is the case for outer integument with two cell layers (119) and intrusive oil cells in the carpel wall (107), plus one-layered endoreticulate endotesta (131) in trees that include several fossils. It may also be true for the syncarpous gynoecium; optimization of this character (106) is equivocal because other members of the clade have one carpel and such taxa were scored as unknown, on the grounds that it cannot be determined whether a single carpel originated by reduction from several free carpels or from a syncarpous gynoecium. The number of floral parts per whorl, which is most often four but was not scored because it is variable, may also be derived: the perianth and androecium at the node where Canrightia diverged are reconstructed as trimerous.

Friis and Pedersen (2011) compared the pollen of Canrightia to Retimonocolpites and Dichastopollenites; it resembles Retimonocolpites dividuus of Pierce (1961) in that the sulcus extends more than halfway around the grain, thus approaching Dichastopollenites, where it forms a ring. It differs from pollen of the Clavatipollenites rotundus type associated with Couperites in having smooth rather than sculptured muri and a thin rather than thick nexine. Friis and Pedersen (2011) suggested that the smooth muri of both Canrightia and Zlatkocarpus might be evidence for relationship, but on trees where these taxa are successive lines, it is a symplesiomorphy, like retention of a perianth. Friis and Pedersen (2011, p. 24) also suggested that the sulcus might not be distal but in “a derived, non-polar position as for instance in some Nymphaeales,” but this is unlikely in terms of the principle that the polar axis in pollen corresponds to its main axis of symmetry. The aperture of Barclaya and Nymphaeoideae forms a complete ring that is perpendicular to the axis of symmetry of the grain, which is known to correspond to the polar axis (Gabarayeva and Rowley 1994). By contrast, the sulcus of Canrightia does not form a complete ring, so the axis of symmetry runs through the midpoint of the sulcus, implying that it is polar.

Friis and Pedersen (2011) suggested that the similarities between Canrightia and both Chloranthaceae and Piperales support the older view that the two living taxa are related, which conflicts with molecular evidence that Piperales are nested within the magnoliid clade. We tested this hypothesis by comparing the morphological parsimony of the D&E and J/M trees with each of two trees in which the chloranthoid line and Piperales are sister groups, with and without Canrightia: (a) with the Chloranthaceae-Ceratophyllum clade (D&E) or Chloranthaceae (J/M) moved into magnoliids and (b) with Piperales moved out of magnoliids as the sister group of Chloranthaceae-Ceratophyllum (D&E) or Chloranthaceae (J/M).

These experiments indicate that Canrightia provides only weak support for a closer relationship between Chloranthaceae and Piperales. With the D&E tree, addition of Canrightia has no effect on the morphological parsimony of this relationship. It is two steps less parsimonious to move the Chloranthaceae-Ceratophyllum clade into the magnoliids with Piperales both with Recent taxa only (1018 steps vs. 1016) and with Canrightia added (at any of its three possible positions relative to Piperales and the Chloranthaceae-Ceratophyllum clade or with Saururaceae and Piperaceae; 1023 steps vs. 1021). It is two steps more parsimonious to move Piperales out of magnoliids to the Chloranthaceae-Ceratophyllum line both with Recent taxa only (1014 steps vs. 1016) and with Canrightia added (on the Chloranthaceae-Ceratophyllum line; 1019 steps vs. 1021). With the J/M tree, addition of Canrightia does improve the parsimony of a relationship of Chloranthaceae and Piperales but only by one or two steps. It is one step more parsimonious to move Chloranthaceae into the magnoliids with Piperales with Recent taxa only (1026 steps vs. 1027) but three steps more parsimonious with Canrightia added (with Piperales; 1028 steps vs. 1031), an improvement of two steps. It is three steps more parsimonious to move Piperales out of magnoliids to Chloranthaceae with Recent taxa only (1024 steps vs. 1027) but four steps more parsimonious with Canrightia added (with Piperales or with Saururaceae and Piperaceae; 1027 steps vs. 1031), an improvement of one step. This modest increase in support for a closer relationship between Chloranthaceae and Piperales must be weighed against increasingly strong molecular evidence for the monophyly of magnoliids and the nested position of Piperales within them (cf. Soltis et al. 2005).

Friis and Pedersen (2011) suggested that the presence of an endothelium in seeds of Canrightia and Lactoris (Piperales) was evidence for relationship. However, since these are the only taxa with an endothelium, this feature would support a relationship only if Canrightia and Lactoris were sister groups, which is eight steps less parsimonious than a relationship of Canrightia with Chloranthaceae (J/M) or Chloranthaceae-Ceratophyllum (D&E). Friis and Pedersen (2011) suggested that an endothelium might be a basic feature of magnoliids that was lost in all lines except Lactoris, but this would require at least four losses, not a parsimonious scenario. Furthermore, an endothelium tends to occur in ovules with a thin nucellus (Endress 2011; Lactoris is weakly tenuinucellar: Tobe et al. 1993), which is unlikely to be ancestral in magnoliids.

Appomattoxia

Our concept of this taxon is based primarily on Appomattoxia ancistrophora, known from isolated carpels at the fruit stage with adhering pollen of the Tucanopollis-Transitoripollis type, which were described by Friis et al. (1995) from Puddledock, Virginia (middle Albian). Additional characters are based on isolated stamens containing similar pollen from Torres Vedras, Portugal (Aptian or early Albian), figured by Friis et al. (2006, fig. 10C; 2010a, pl. II, fig. 2).

The most conspicuous feature of Appomattoxia is the presence of hooked hairs on the fruit wall. Friis et al. (1995) compared these with hairs of the eudicot Circaeaster (Ranunculales), but they rejected a close relationship because the pollen of Appomattoxia is monosulcate rather than tricolpate. Friis et al. (2010a) also reported fruits with similar pollen but without hairs at Torres Vedras; if these are related to Appomattoxia, they would tend to neutralize the hairs as evidence of relationship. The fruits contain a single pendent orthotropous ovule, as in Chloranthaceae, but the innermost layer of the seed coat around the micropyle consists of cells with thickened undulate walls. Friis et al. (1995) compared this with a similar layer in Piperaceae and Saururaceae (Piperales), which is known to be a sclerotic ecto- and endotegmen (Corner 1976; Takhtajan 1988). The pollen resembles Clavatipollenites and most Chloranthaceae in having a sculptured sulcus membrane, supratectal spinules, and an unusually thick nexine consisting of foot layer plus endexine under the aperture, but it differs in having a continuous tectum, again like Piperaceae and Saururaceae. Based on these characters, Friis et al. (1995, 2010a, 2011) favored a relationship with Piperaceae and Saururaceae. However, these taxa have a syncarpous gynoecium of 2–4 carpels (except in the derived genus Peperomia), with several ventral (horizontal) ovules per carpel in Saururaceae and one basal (ascendent) ovule per unilocular gynoecium in Piperaceae, rather than one pendent ovule per carpel, as in Appomattoxia and Chloranthaceae.

Dispersed pollen similar to that of Appomattoxia was described from the Barremian-Aptian of Brazil by Regali et al. (1974) as Inaperturopollenites crisopolensis (although it has a conspicuous sulcus, often widened into a large circular ulcus) and transferred to the new genus Tucanopollis by Regali (1989). Góczán and Juhász (1984) described similar but generally less sculptured pollen from the Albian of Hungary as Transitoripollis. Whether the two genera should be distinguished is unclear. Such pollen is a minor element in Southern Laurasia, extending back to the Barremian of England as the Barremian-ring group of Hughes (1994), but one of the most abundant angiosperm pollen types in the Barremian and Aptian of Northern Gondwana (Doyle et al. 1977; Regali 1989). Based on the similarities noted above, Doyle and Hotton (1991) argued that Tucanopollis was produced by relatives of Chloranthaceae that were either more primitive or more derived in having a continuous tectum. However, they also noted similarities to Saururaceae, which lack supratectal spinules and have a thin endexine around the grain but which also have a more or less continuous tectum and a thick foot layer (Smith and Stockey 2007).

As noted by Friis et al. (1995), the oblique position of the carpel stipe suggests that the carpels were from a multicarpellate apocarpous flower, rather than a unicarpellate flower, which would mean that the ovary was superior, but because this is speculative, we have scored floral base/ovary position (48) and other characters of inflorescence and flower organization as unknown. The apical view of the stamen in Friis et al. (2006, 2010a) shows a truncated connective apex (72), protruding pollen sacs (73), four microsporangia (74), and longitudinal dehiscence (76), but no other characters can be determined because of the unsuitable orientation.

Pollen measures 16–19 μm (82), which we score as either small (<20 μm) or medium to allow for possible shrinkage. As noted by Friis et al. (1995), the infratectum (87) appears to consist of granules that are often fused into irregular columellae, falling in our intermediate state. The nexine (94) consists of a thick foot layer and much thinner endexine, but the endexine is darker and more consistently present than the discontinuous and indistinctly staining inner layer of most Chloranthaceae (Doyle 2005), which we treat as having foot layer only, so we score endexine as present (1).

As in most fossils, carpel form (97) cannot be established. We score the stigma (102) as extended because of the large size of the area (above the hooked hairs) that Friis et al. (1995) interpreted as stigmatic; however, we score style (101) as absent because there is no visible constriction at the base of the stigmatic extension. The stigma bears papillae (104) assignable to state 1 (unicellular or with one emergent cell).

The ovule is pendent (114) and orthotropous (115). As in the Pennipollis plant, there is no evidence on the number of integuments (116), so we have scored characters of intermediate cell layers in the seed coat (129–131) as unknown. However, because the innermost layer is so distinctive (relative to the Pennipollis plant and living unitegmic taxa), we score it as a sclerotic ecto- and endotegmen (132, state 1), to express its similarity and potential homology with that layer in Piperaceae and Saururaceae. We assume that the fruit wall (123) was dry because of its thinness and the fact that fleshiness and hooks appear to be mutually exclusive in living plants.

Our analyses give two very different sorts of results for the relationships of Appomattoxia, depending on whether it is added alone to the Recent trees or with other fossils. Both sets of results suggest that Appomattoxia may be important for understanding early angiosperm evolution but in different ways.

When Appomattoxia is added alone to both backbone trees (fig. 14), its four most parsimonious positions are at or near the very base of the tree: sister to all angiosperms, Amborella, all other angiosperms, or Nymphaeales. These positions are favored by the continuous tectum (88), which becomes perforate below Austrobaileyales and mesangiosperms, and they are consistent with the single pendent ovule (as in Amborella and Trithuria) and sessile stigma. The first three positions are equally parsimonious because the backbone trees include no outgroups, so there are no unambiguously derived states that favor one arrangement over another. The orthotropous ovule (115) of Appomattoxia and Amborella favors these three positions over a relationship with Nymphaeales, but this is balanced by truncated connective apex (72), a derived state shared with Nymphaeales. The next-best positions (one step worse) are sister to Austrobaileyales plus mesangiosperms, Chloranthaceae (J/M) or Ceratophyllum and Chloranthaceae (D&E), and Hedyosmum (J/M only).

Of the three positions of Appomattoxia around the basal node, the one sister to Amborella would be favored if the orthotropous ovule of the two taxa was shown to be derived. This would be consistent with the hypothesis that Caytonia was the sister group of angiosperms and its anatropous cupule was homologous with the anatropous bitegmic ovule of angiosperms (Gaussen 1946; Doyle 1978, 2008). Another possible synapomorphy, not included in our data set, is a tendency for low verrucate pollen sculpture (Friis et al. 1995, figs. 20–24), which could be a step toward the more prominent verrucae of Amborella (Sampson 1993; Hesse 2001). Similar pollen with larger verrucae is known from the Hauterivian of England (Hauterivian-cactisulc; Hughes and McDougall 1987; Hughes 1994) and has been compared with Amborella (Doyle 2001; Hesse 2001). If these pollen types were related to Amborella, they would imply that Amborella is a relict of a group that was far more widespread and abundant in the Early Cretaceous.

There are characters that could support the other two near-basal positions for Appomattoxia, but they are also difficult to polarize (and often highly homoplastic). Whereas Amborella has pluricellular stigmatic papillae, Appomattoxia has papillae with one emergent cell (104), the inferred basic state for all other angiosperms; if this state is derived, it would favor a position sister to other angiosperms. With the D&E tree, dry fruit wall (123) is another such character. Appomattoxia differs from the reconstructed common ancestor of all extant angiosperms in its intermediate infratectal structure, thick nexine, hooked hairs, and sclerotic tegmen; if any of these features are ancestral, they would support a position on the angiosperm stem lineage.

If Appomattoxia is in any of these positions, the fact that Tucanopollis pollen is so abundant in Northern Gondwana would be significant for the ecological evolution of angiosperms. The Early Cretaceous climate in this province has long been interpreted as hot and dry, based on sedimentary indicators (salt deposits, lack of coal) and characteristics of the flora (few ferns; abundance of Classopollis, representing the xeromorphic conifer family Cheirolepidiaceae, and Gnetales), except for presumed wetter areas in the Middle East and northern South America (Brenner 1976, 1996; Doyle et al. 1982; Doyle 1999; Mejia-Velasquez et al. 2012). By contrast, Amborella and Austrobaileyales are restricted to wet, shaded forest understory sites. Feild et al. (2004, 2009) reconstructed such habitats as ancestral for angiosperms and argued that the rarity of such environments in the Triassic and Jurassic could explain why angiosperms escaped detection if they existed at that time, as indicated by many molecular dating analyses (Magallón 2010; Smith et al. 2010; Clarke et al. 2011; see Doyle 2012). If Appomattoxia is near the base of the tree, it could challenge this view of ecology of the first angiosperms or else indicate that members of the basal grade were able to break out of the ancestral niche and adapt to dry climates more easily than might be expected.

By contrast, when Appomattoxia is added to the Recent data set along with Canrightia, Zlatkocarpus, and the Pennipollis and Asteropollis plants, it is near basal in some most parsimonious trees but associated with Chloranthaceae in others. In one of the four trees found with the D&E backbone (fig. 8A), it is sister to Chloranthaceae plus Ceratophyllum, based on supratectal spinules (91) and dry fruit wall (123; reversed within Chloranthaceae). In another (fig. 8B) it is linked with Ceratophyllum, above the Pennipollis plant; the three taxa are united by a shift from latrorse to extrorse anthers (75) and intermediate infratectal structure (87; see discussion of the Pennipollis plant), while Appomattoxia is linked with Ceratophyllum by dry fruit wall. It should be noted that the exine features listed are not present in Ceratophyllum; its exine is reduced to a thin structureless layer (Takahashi 1995), so we scored most of its pollen characters as unknown. Finally, in two trees (fig. 8C, 8D) it is sister to Nymphaeales. In 36 of the 48 trees found with the J/M backbone, Appomattoxia is located in one of the three positions around the basal node or with Nymphaeales (fig. 8E, 8F), but it is related to Chloranthaceae in the remaining 12. In one of the latter trees (fig. 8H), it is linked with the Pennipollis plant by intermediate infratectum, and the two are the sister group of crown Chloranthaceae, based on supratectal spinules. In the other 11 trees (fig. 8G, 8I–8O), Appomattoxia is nested in Chloranthaceae, on the line to Hedyosmum and the Asteropollis plant: either by itself, with the three taxa united by dry fruit wall; linked with or above the Pennipollis plant, with the four taxa united by loss of bracts subtending the male flowers (46) and/or supratectal spinules (depending on the position of Zlatkocarpus); or above Zlatkocarpus.

Association of Appomattoxia and/or the Pennipollis plant with vegetative remains could confirm or refute the hypothesis that these fossils are stem relatives of Ceratophyllum, and if they are related, they could clarify the origin of this enigmatic floating aquatic, which has whorled dichotomous leaves and no roots. There are several megafossils that merit investigation as possible relatives of Ceratophyllum. Dilcher and Wang (2009) explicitly related fruits (Donlesia) and associated leafy stems from the latest Albian of Kansas to Ceratophyllum. As in Ceratophyllum, the fruits are one-seeded achenes with prominent spines, and the leaves are whorled and dichotomous, but the fruits differ in having a long “peduncle” and possibly basal (vs. apical) seed attachment and the leaves lack marginal denticles. Montsechia, from Barremian lake beds in Spain, has been interpreted as an aquatic with whorled undivided leaves (Martín-Closas 2003; Gomez et al. 2006). Krassilov (2011) reinterpreted it as a xerophytic marsh plant with long and short shoots and dimorphic opposite leaves, but because whorled phyllotaxis and opposite phyllotaxis are closely related, this would not contradict a relationship to Ceratophyllum. He interpreted Montsechia as a “proangiosperm” with “cupules” of scale leaves containing one ovule, based on the presence of pollen in the nucellar area. However, the pollen is poorly characterized, and the ovule appears to be orthotropous, with the chalaza displaced to one side, as in the Pennipollis plant and Chloranthaceae. Most intriguing is Pseudoasterophyllites, from the late Albian of France and the middle Cenomanian of Bohemia, which has pseudowhorls of opposite, linear, and apparently succulent leaves and was interpreted by Kvaček et al. (2012) as a halophyte. Kvaček et al. (2012) associated stamens containing Tucanopollis-like pollen with the plant, based on close co-occurrence and similar stomata. If Pseudoasterophyllites is related to Appomattoxia, it could support a relationship between Appomattoxia and Ceratophyllum. An aquatic habit for the Pennipollis plant would be consistent with the abundance of Pennipollis pollen in the organic-rich Arundel Clay (e.g., at United Clay Mine; Brenner 1963; Doyle et al. 1975), a classic swamp deposit.

If Tucanopollis pollen came from relatives of Ceratophyllum, it would have little bearing on the original ecology of angiosperms, and its abundance in Northern Gondwana might reflect the presence of its parent plants in local wet and/or saline habitats. It is also possible that pollen referred to Tucanopollis is heterogeneous and was not all produced by plants related to Appomattoxia. However, the exine structure of Tucanopollis from the Barremian of Gabon (Doyle and Hotton 1991) is very similar to that of Appomattoxia pollen (Friis et al. 1995), even at the TEM level.

Appomattoxia again illustrates how having only isolated reproductive organs allows a wide range of hypotheses and how association with vegetative parts or evidence on floral organization could decide among them. For example, an ANITA-grade plant near Amborella would be expected to have a gynoecium of several carpels and a perianth of several whorls or series of tepals (Endress and Doyle 2009), whereas a plant near Ceratophyllum and/or Chloranthaceae should have one carpel and either no perianth or one whorl of tepals.

Summary of evolution in the chloranthaceous line

In a synthesis based on a morphological cladistic analysis of living Chloranthaceae, which included the ANITA lines and three magnoliid taxa as outgroups but not Ceratophyllum, Doyle et al. (2003) were unable to decide between two equally parsimonious scenarios for floral evolution in the family. In one scenario, flowers were still bisexual in the most recent common ancestor of Chloranthaceae and became unisexual independently in Hedyosmum and Ascarina. In the other, flowers were already unisexual in the most recent common ancestor and reversed to secondarily bisexual in the clade consisting of Sarcandra and Chloranthus. In both scenarios, a perianth was still present at the crown group node, as in female flowers of Hedyosmum, but was lost on the line to the Ascarina-Sarcandra-Chloranthus (ASC) clade.

Addition of fossils provides a better-resolved picture. Our discussion centers on a scenario (fig. 15) based on one of the two most parsimonious trees found when the four most securely associated fossils were added to the D&E backbone (fig. 6B), with remarks on variations seen in other trees. Because floral organization of Appomattoxia is largely unknown, its addition would have little impact on this scenario.

Since outgroup relationships imply that Chloranthaceae (and Ceratophyllum) were derived from ancestors that had bisexual flowers with many parts, the first inferred changes, seen in Canrightia, were reduction of the pedicel, resulting in sessile flowers; reduction of the perianth to one whorl of tepals; adnation of the tepals and stamens to a gynoecium of several carpels, resulting in an inferior ovary; and a shift from anatropous to orthotropous ovules. The carpels had a single pendent ovule, which is inferred to be a feature inherited from the first angiosperms (Endress and Doyle 2009). The androecium was also reduced to one whorl of stamens, but in the D&E tree, because of the arrangement of outgroups and treatment of this character as unordered, it is equally parsimonious to assume that this reduction occurred earlier, in the common ancestor of mesangiosperms.

The next major events were a shift to unisexual flowers and reduction to one carpel, still surrounded by adnate tepals, as in Zlatkocarpus, the Asteropollis plant, and Hedyosmum. This holds whether Zlatkocarpus is sister to the crown clade (figs. 6A–6E, 6J, 6K, 15) or nested within it (fig. 6F–6I), except that the number of shifts to unisexuality is ambiguous if Zlatkocarpus is attached to the Hedyosmum line (fig. 6F, 6H). Reduction to one stamen may have occurred on the same branch, but this is ambiguous if Zlatkocarpus is basal (as in fig. 15), since its male structures are unknown. In any case, stamen number had been reduced to one at the crown node.

These changes were followed by loss of the perianth, as in Ascarina, Sarcandra, Chloranthus, and Ceratophyllum. With the J/M backbone (fig. 6C–6K), this loss occurred once, on line to the ASC clade. However, with the D&E backbone (fig. 15), where Ceratophyllum is related to Chloranthaceae, it is equally parsimonious to assume that the perianth was lost twice, on the lines to Ceratophyllum and the ASC clade, or lost once in the common ancestor of Ceratophyllum and Chloranthaceae and regained in the Asteropollis-Hedyosmum line (perhaps a less plausible scenario). Information on whether female flowers of the Pennipollis plant had a perianth could affect these inferences (it clearly did not have an adnate perianth of the type seen in Canrightia, Zlatkocarpus, and Hedyosmum).

In most trees including the four fossils, it is most parsimonious to assume that the bisexual flowers of Sarcandra and Chloranthus were secondarily derived from unisexual; exceptions are two trees with Zlatkocarpus on the line to Hedyosmum (fig. 6F, 6H), where the course of evolution is ambiguous. This hypothesis might be consistent with the bizarre morphology of these flowers—a single carpel with one stamen attached to the back in Sarcandra, one carpel with a dorsal three-lobed androecium in Chloranthus and related Late Cretaceous fossils (Herendeen et al. 1993; Eklund et al. 1997), which is variously interpreted as three fused stamens or one subdivided stamen (Swamy 1953; Endress 1987; Eklund et al. 1997; Kong et al. 2002; Doyle et al. 2003). This raises the possibility that these bisexual structures are actually pseudanthia composed of extremely reduced unisexual flowers. Other reversals were increases in stamen number in male flowers of some Ascarina species (to 2–5; Jérémie 1980) and in Chloranthus, if it has three fused stamens.

Another reduction occurred in inflorescence morphology, namely, loss of the bracts subtending male flowers in Hedyosmum, the Asteropollis plant, the Pennipollis plant, and Ceratophyllum; the situation in Zlatkocarpus is unknown. In three trees found with the J/M backbone (fig. 6C, 6F, 6G), in which the Pennipollis plant, the Asteropollis plant, and Hedyosmum form a clade, these bracts were lost once in Chloranthaceae. However, in trees with both backbones in which the Pennipollis plant is elsewhere, this character is homoplastic. In three trees (figs. 6B, 6D, 6E, 15), it is equivocal whether bracts were lost twice or lost once and regained in the ASC clade; the former scenario is favored in four trees (fig. 6H–6K), the latter in one (fig. 6A).

Pollen in the common ancestor of Canrightia and Chloranthaceae can be reconstructed as globose and monosulcate, with a reticulate-columellar exine, smooth muri (no supratectal spinules), and a sculptured sulcus membrane, as in Canrightia and Zlatkocarpus. All these features were inherited from lower in the tree. This was modified by origin of spinules on the muri, as in Pennipollis, Asteropollis, Hedyosmum, and Ascarina (as well as Appomattoxia-Tucanopollis and Clavatipollenites). In both trees with the D&E backbone (fig. 6A, 6B) and three with the J/M backbone (fig. 6D, 6E, 6K), spinules arose once in the common ancestor of the Pennipollis plant and living groups and were later lost in the Sarcandra-Chloranthus clade. However, in other J/M trees, spinules either originated twice (in the Hedyosmum and Ascarina lines, with the Pennipollis plant linked with one or the other; fig. 6F–6I) or the course of their evolution is ambiguous (fig. 6C, 6J). The distinctive thick nexine of Pennipollis and living Chloranthaceae (except some Chloranthus species), also seen in Tucanopollis and Clavatipollenites, originated after the divergence of Canrightia, but where exactly is uncertain if Zlatkocarpus is basal (fig. 6A–6E, 6J, 6K), since its nexine thickness is unknown. The ancestral sulcus was modified to a several-armed furrow in the Asteropollis-Hedyosmum clade and to scattered pores and several colpoid areas in Sarcandra and Chloranthus, respectively.

If both the Pennipollis plant and Appomattoxia are related to Ceratophyllum (fig. 8B), they show a picture of both progressive and markedly divergent pollen trends. From an ancestor with a reticulate-columellar exine and supratectal spinules, infratectal structure was modified to intermediate (as in Appomattoxia) and then granular (in Pennipollis). The original finely reticulate tectum was modified in opposite ways, becoming extremely coarse in Pennipollis but closed in Appomattoxia-Tucanopollis; however, both retained spinules and a thick nexine. Ceratophyllum neither supports nor contradicts this scenario; any signs of its earlier history were erased during reduction of its entire exine to a thin structureless layer (Takahashi 1995).