Phylogeny of the core Malvales: evidence from ndhF sequence data ^†

Abstract

The monophyly of the group comprising the core malvalean families, Bombacaceae, Malvaceae, Sterculiaceae, and Tiliaceae, was recently confirmed by molecular studies, but the internal structure of this clade is poorly understood. In this study, we examined sequences of the chloroplast ndhF gene (aligned length 2226 bp) from 70 exemplars representing 35 of the 39 putative tribes of core Malvales. The monophyly of one traditional family, the Malvaceae, was supported in the trees resulting from these data, but the other three families, as traditionally circumscribed, are nonmonophyletic. In addition, the following relationships were well supported: (1) a clade, /Malvatheca, consisting of traditional Malvaceae and Bombacaceae (except some members of tribe Durioneae), plus Fremontodendron and Chiranthodendron, which are usually treated as Sterculiaceae; (2) a clade, /Malvadendrina, supported by a unique 21-bp (base pair) deletion and consisting of /Malvatheca, plus five additional subclades, including representatives of Sterculiaceae and Tiliaceae, and Durionieae; (3) a clade, /Byttneriina, with genera traditionally assigned to several tribes of Tiliaceae, plus exemplars of tribes Byttnerieae, Hermannieae, and Lasiopetaleae of Sterculiaceae. The most striking departures from traditional classifications are the following: Durio and relatives appear to be more closely related to Helicteres and Reevesia (Sterculiaceae) than to Bombacaceae; several genera traditionally considered as Bombacaceae (Camptostemon, Matisia, Phragmotheca, and Quararibea) or Sterculiaceae (Chiranthodendron and Fremontodendron) appear as sister lineages to the traditional Malvaceae; the traditional tribe Helictereae (Sterculiaceae) is polyphyletic; and Sterculiaceae and Tiliaceae, as traditionally circumscribed, represent polyphyletic groups that cannot sensibly be maintained with their traditional limits for purposes of classification. We discuss morphological characters and conclude that there has been extensive homoplasy in characters previously used to delineate major taxonomic groups in core Malvales. The topologies here also suggest that /Malvatheca do not have as a synapormophy monothecate anthers, as has been previously supposed but, instead, may be united by dithecate, transversely septate (polysporangiate) anthers, as found in basal members of both /Bombacoideae and /Malvoideae. Thus, “monothecate” anthers may have been derived at least twice, independently, within the /Bombacoideae (core Bombacaceae) and /Malvoideae (traditional Malvaceae).

At the core of the order Malvales are four traditional families; Bombacaceae (∼250 spp.), Malvaceae (∼1500 spp.), Sterculiaceae (∼1000 spp.), and Tiliaceae (∼400 spp.). The close relationship among these core families has been recognized since the time of Linnaeus and was recently confirmed by molecular data (Chase et al., 1993; Soltis et al., 1997; Alverson et al., 1998; Fay et al., 1998; Bayer et al., 1999). Several morphological and anatomical synapomorphies also appear to support their monophyly (Judd and Manchester, 1997; Alverson et al., 1998; Bayer, 1999; Bayer et al., 1999). However, knowledge of the phylogenetic relationships within the four families remains poor. As aptly stated by Judd and Manchester (1997): “It has long been apparent to systematists that the delimitation of families within the ‘core’ Malvales …is problematic.”

The core Malvales constitute one of four major clades within an expanded Malvales clade, which also includes Bixaceae, Cistaceae, Cochlospermaceae, Dipterocarpaceae, Muntingiaceae, Neuradaceae, Sarcolaenaceae, Sphaerosepalaceae, and Thymelaeaceae (Alverson et al., 1998; Fay et al., 1998; Bayer et al., 1999). The expanded Malvales clade (or, simply, “Malvales”; APG, 1998) is associated with the Capparales and the Sapindales within the Rosid II clade of Chase et al. (1993). Now that the broader phylogenetic context has been established, the opportunity exists to clarify relationships within the core Malvales (termed “Malvaceae” by Judd and Manchester, 1997; APG, 1998; and Bayer et al., 1999).

The boundaries between the four core malvalean families have been notoriously difficult to pin down and there are several genera (e.g., Camptostemon, Chiranthodendron, Corchoropsis, Craigia, Fremontodendron, Hampea, Maxwellia, Nesogordonia, and Uladendron) that have been shuffled among them in traditional classification schemes (e.g., Edlin, 1935; Hutchinson, 1967; Robyns, Nilsson, and Dechamps, 1977; Takhtajan, 1987, 1997; Tang, 1990). Within the four families there are similar problems in the delimitation of tribes and subfamilies, which are reflected in some marked differences among traditional taxonomic treatments (Table 1). Furthermore, even those suprageneric taxa that were agreed upon by these systematists are not necessarily reliable. For example, Durio and its relatives (Durioneae) have been universally placed within the Bombacaceae, but this position was recently questioned based on morphology, biogeography, and chromosome number (Baum and Oginuma, 1994; Judd and Manchester, 1997).

The guiding model used when delimiting these traditional families has been one of a progression from the primitive Tiliaceae, through the intermediate Sterculiaceae and Bombacaceae, to the advanced Malvaceae (Cronquist, 1981, 1988; Takhtajan, 1987, 1997; Dahlgren, 1989; Thorne, 1992; Judd, Sanders, and Donoghue, 1994; La Duke and Doebley, 1995; Judd and Manchester, 1997). Although Warming (1895) and Venkata Rao (1952) deviated from this model by considering the Sterculiaceae to be the basal group, the prevailing assumption of an unbroken linear trend, founded mostly on increased staminal fusion and loss of woodiness, has inhibited progress towards an understanding of the phylogeny and evolutionary diversification of these groups.

A recent morphological cladistic analysis attempted to improve knowledge of the relationships within the core Malvales (Judd and Manchester, 1997), but obtained little robust resolution except for confirmation of the monophyly of the traditional Malvaceae. Thus, knowledge of the phylogeny of the core Malvales remains limited, obscuring knowledge of the group's morphological evolution and biogeography. In particular, the androecium is quite varied within the core Malvales (van Heel, 1966), but this variation is refractory to investigation in the absence of a well-resolved phylogeny.

In this study we attempted to advance understanding of the phylogeny of core Malvales using sequences of the chloroplast gene ndhF, which has been used successfully to elucidate phylogenetic relationships at the infraordinal level (e.g., Clark, Zhang, and Wendel, 1995; Smith et al., 1997). This gene encodes a subunit of the nicotinamide dehydrogenase complex and shows approximately twice the average mutation rate of rbcL (Suguira, 1989; Olmstead and Sweere, 1994). Our goal was to identify large-scale phylogenetic structure within core Malvales and to provide a better understanding of the relationships among the major subfamilial groups. This study parallels independent investigations of Malvales phylogeny using rbcL and atpB sequences (Bayer et al., 1999) and the nuclear gene PLD (F. Blattner, IPK, Gatersleben, personal communication).

MATERIALS AND METHODS

Taxon sampling

We generated 66 new ndhF sequences for this phylogenetic analysis of Malvales, to which we added four previously published sequences of Malvaceae (see Table 1 for a complete list of exemplars and voucher information). These exemplars represented 35 of the 39 tribes of the core malvalean families recognized by Takhtajan (1997). Not sampled were Triplochitoneae (Sterculiaceae, one genus); Desplatsieae, Diplodisceae, and Duboscieae (Tiliaceae; four, four, and one genera, respectively), due to lack of material. We also did not include exemplars of Takhtajan's subfamily Neotessmannioideae (Tiliaceae, two genera). Although Burret (1926), Edlin (1935), Hutchinson (1967), Cronquist (1981), Takhtajan (1987, 1997), and Benn and Lemke (1991) place Neotessmannia and its close relatives Dicraspidia and Muntingia within Tiliaceae, sequence analyses of the latter two genera indicate that they are not part of the core Malvales but, rather, more distantly related members of the expanded malvalean clade (Alverson et al., 1998; Bayer, Chase, and Fay, 1998; Bayer et al., 1999). Muntingia and two other genera previously determined as belonging to the expanded Malvales, Cochlospermum, and Rhopalocarpus, were used as outgroups in the current study (Table 1).

Laboratory methods

Our DNA extractions were modified Zimmer or 2X-CTAB procedures (protocols A and C in Smith et al., 1990), or procedures associated with DNeasy columns (Qiagen Corp. number 69104, Valencia, California). We generally used four polymerase chain reaction (PCR) primers that provided two products of 1370 and 1139 bp, with an overlap of 347 bp. For difficult taxa, additional, internal PCR primers were used. PCR products were diluted and used in the cycle sequencing reactions or were first cleaned with QlAquick columns (Qiagen Corp. numbers 28104 and 28704). Overlapping sequence fragments were obtained from both strands of the gene using a total of 12–16 primers. Primers were obtained from Olmstead, Sweere, and Wolfe (1993) and were modified in four instances for Malvales (536F, TTGTAACTAATCGGATAGGCGA; 972F, GTCCCAACTGGGTTATATGATG; and their respective complementary primers, 536R and 972R). Dideoxy-terminated cycle sequencing products were sequenced on ABI 370A and 377 automated DNA sequencers using Long Ranger acrylamide gels (FMC Bioproducts, Rockland, Maine). The contigs were assembled and checked for stop codons using Sequencher 3.0 (Gene Codes Corp., Ann Arbor, Michigan). Bases from positions −52 to −1 and from 2227 to 2252 were removed to avoid primer artifacts. The sequence from each taxon was aligned and gapped by visual inspection, a generally easy process. All newly generated sequences were submitted to GenBank (Table 1) and the aligned data were submitted to TreeBASE (www.herbaria.harvard.edu/treebase). Descriptive statistical data were generated with the BSS program (R. Nyffeler, unpublished data) from the entire ndhF gene, as well as separately for the 5′ and 3′ ends, split between base positions 1350 and 1351, comparable to the boundary used by Kim and Jansen (1995) and corresponding to positions 1329 and 1330 of the Nicotiana sequence (cf. Olmstead, Sweere, and Wolfe, 1993). All ambiguous sites were scored as “?” for these descriptive statistics. Variable sites were scored as 2-fold transitions (two states related by a transition), 2-fold transversions (two states related by a transversion), 3-fold sites (with three states observed), and 4-fold sites. Finally, indel events were referenced to the ndhF sequence of Nicotiana (Olmstead, Sweere, and Wolfe, 1993; GenBank L14953).

Phylogenetic analyses

For a preliminary analysis using all 70 taxa, we ran 100 random addition sequence (RAS) equally weighted parsimony searches with tree bisection reconnection branch swapping (TBR), hold = 1, MULPARS and steepest descent on, gaps scored as missing, and zero-length branches collapsed using PAUP* versions 4.0d64 and 4.01b (Swofford, 1998) on a PowerMac 8500/150. The shortest trees were condensed and then summarized by a strict consensus tree. The phylogenetic signal present in the full ndhF data set was estimated using PAUP* based on the skewness of the length distribution of 100 000 random trees (Hillis and Huelsenbeck, 1992; but see Ka¨llersjo¨ et al., 1992).

An excessive number of trees were attributable to one poorly resolved clade of genera traditionally placed in or near Bombacaceae, tribe Bombaceae. We assessed the robustness of this clade by decay analysis (Bremer, 1988; Donoghue et al., 1992) using inverse constraints and 100 RAS searches with MULPARS off. The core Bombacaceae clade was then reduced from 11 to five taxa for all further analysis. Most parsimonious trees for the 64-taxon data set were estimated using 100 RAS parsimony searches, as above. In order to explore other weighting schemes, we ran five sets of ten RAS parsimony searches under each of the following conditions: (1) transition/transversion stepmatrix weights of 2:1 and 5:1, respectively; (2) same two stepmatrices plus codon weighting of 1, 1, 0.5; (3) same two stepmatrices plus codon weights of 1, 1, 0; (4) same two stepmatrices plus equally weighted codon positions, and gapped characters deleted; (5) no stepmatrices, equally weighted codons, and gapped characters deleted.

The strength of support for each clade found during the 64-taxon equally weighted parsimony searches was assessed using PAUP* 4.0b1 (Swofford, 1998) by obtaining its decay index (di), using 100 RAS inverse constraint searches (otherwise as above), and by bootstrap values (bs), using ten RAS searches per pseudoreplicate (as above but with MAXTREES = 100). We also used the shortest trees in which the monophyly of each clade of interest was not supported (from the decay analyses) and then evaluated the significance of the extra steps required by these trees using a Wilcoxon sign-rank test (Templeton, 1983). Results of these Templeton tests are reported below only when they are significant. The cost of some specific topological arrangements not found on the shortest trees was assessed by conducting inverse monophyly constraint searches. The aligned data set and most parsimonious trees from this study are available from TreeBASE (www.herbaria.harvard.edu/treebase; S388 and M542).

Nomenclature

We refer to clades discovered during the course of this study by a set of unranked, phylogenetic taxon names as published by Baum, Alverson, and Nyffeler (1998) and by traditional, ranked names proposed by Bayer et al. (1999). Although the phylogenetic names have no official status until a code governing such names is published (DeQueiroz and Gauthier, 1992, 1994), these rankless names offer several potential advantages over traditional names (de Queiroz, 1997; Baum, Alverson, and Nyffeler, 1998). Two conspicuous benefits of phylogenetic names relative to the taxonomy of core Malvales are: they are explicitly defined, independent of subjective rank assessment; and they allowed us to name all significant, well-supported clades. The phylogenetic names used here are in most cases identical in name and inferred content to the traditional subfamily, family, and ordinal names proposed by Bayer et al. (1999). However, the correspondence between our named clades and the named groups of Bayer et al. is not always exact because that study did not provide phylogenetic definitions and sometimes recognized paraphyletic groups. Additionally, some of the clades, discussed below, were not named by Bayer et al. To distinguish traditional taxa (e.g., Malvaceae) from unranked clade names, the latter are marked here by a preceeding slash mark (e.g., /Malvaceae).

RESULTS

The aligned sequences used for analysis are 2226 base pairs (bp) in length and correspond to positions 1-2109 of the ndhF gene of tobacco (Olmstead, Sweere, and Wolfe, 1993). Data are not available for 6-53 bases at either the 5′ or 3′ end of the analyzed sequence from 15 exemplars. Data are also missing for 44-338 internal bases (mean = 161) from five taxa: Colona, Eriotheca, Patinoa, Tilia, and Triumfetta. We ran all analyses with data from only the 5′ end of Nesogordonia because of a possible contamination of the 3′ end of the sequence, discovered late in the study.

The shortest, uncorrected pairwise distance within the ingroup is between Chiranthodendron and Fremontodendron (0.24%) and the largest is between Waltheria and Malope (6.91%). The largest distance overall is 11.29%, between Waltheria and Muntingia (an outgroup). The distribution of variable sites ranges from 3 (10%) to 21 (70%) per 30-bp segment. The contrast between the abundance of variable sites in the 5′ region (31% for bases 1-1350) and the 3′ region (47.2% for bases 1351-2226) is not as strong as that seen by Kim and Jansen (1995) for Asteraceae. Overall GC content is 32.6%, with a marked difference between the 5′ region (35.2%) and the 3′ region (28.2%). The 2-fold transition sites comprise 15.8% of the variable sites and are evenly distributed over the gene with no marked differences between the 5′ and 3′ regions (Table 2). However, 2-fold transitions at the second codon position were more than twice as common for the 3′ as for the 5′ end, whereas transitions at the third codon position were proportionately more common at the 5′ end. This pattern of reduced second position variation at the 5′ end and reduced third position variation at the 3′ end reoccurs for other types of sites (Table 2), suggesting a greater conservation of amino acids at the 5′ end.

A total of 32 indel events were indicated by these data, of which 19 are potentially synapomorphic. Seven of these 19 characters are perfectly consistent with the most parsimonious trees, whereas 12 are homoplastic (Fig. 1). The remaining 13 indel events were autapomorphic. The taxa in which these autapomorphies occurred, the nature and size of the indel, and position relative to the ndhF gene of Nicotiana (Olmstead, Sweere, and Wolfe, 1993) are as follows: Apeiba (+9, 23/24), Bernoullia (+15, 1335/1336), Byttneria (−6, 1546/1547), Carpodiptera (+6, 1605/1606), Cochlospermum (+6, 1584/1585), Colona (−6, 1548/1549), Heritiera (−6, 1779/1780), Hibiscus (+3, 1567/1568), Muntingia (+6, 1584/1585), Rhopalocarpus (+3, 1539/1540), Scleronema (+6, 92/93), Sterculia (+9, 1482/1483), and Trichospermum (+6, 1491/1492).

The equally weighted parsimony analysis of the full data set yielded 3663 trees of 1742 steps (including autapomorphies), distributed on one island that was found in all 100 RAS searches. The trees have a consistency index (CI) of 0.645 (including all characters) and a retention index (RI) of 0.679. The data are significantly skewed as judged by a g1 = −0.5878 (Hillis and Huelsenbeck, 1992). This preliminary 70-taxon analysis failed to resolve structure within a “core bombacoid” clade consisting of Adansonia, Bombax, Catostemma, Cavanillesia, Ceiba, Chorisia, Eriotheca, Pachira, Pseudobombax, Scleronema, and Spirotheca, all from Takhtajan's (1997) tribes Bombaceae and Ceibeae. This clade had a decay value of +2 steps (at which point the clade collapsed to include Bernoullia, Gyranthera, and Huberodendron). Although the support for this clade is not significant based on a Templeton test, branch lengths within the clade are so short that removing some taxa is unlikely to affect the overall tree structure.

Therefore, we removed six of these taxa and reran the equally weighted parsimony analysis with 64 taxa. This yielded 46 trees at 1666 steps on one island found in all 100 RAS searches (CI = 0.650, RI = 0.682, g1 = −0.5739, 431 informative characters). The strict consensus of the 46 trees is shown in Fig. 2. One of the 46 trees is shown in Fig. 1, onto which phylogenetically informative indel events were mapped by simple parsimony. Unequally weighted parsimony searches of the 64-taxon data set produced trees with a reduced resolution of core bombacoids and basal malvoids (see below) but with only one minor change in the composition of subclades, as noted below. Overall, branch lengths in the shortest trees found in these searches differed noticeably among subclades in a way that did not appear to be an artifact of our sampling. The shortest average branch lengths occurred in the core bombacoids (Fig. 1). This may be due to the fact that core bombacoids are generally large trees, given the general negative correlation between generation time and the rate of molecular evolution (Gaut et al., 1992; Baum, Small, and Wendel, 1998), or may relate to the high chromosome numbers found in this group (cf. Baum and Oginuma, 1994).

Relative to the three outgroups used here, traditional core Malvales (here called /Malvaceae; Fig. 2) formed a clade with strong support in all analyses (di = 20; bs = 100). Within the core Malvales, the consensus trees from both the preliminary and final analyses exhibited phylogenetic structure in the form of two major clades, whose composition differs from any previously published hypotheses but which are also found by Bayer et al. (1999).

The first major clade, /Malvadendrina (Fig. 2), contains all exemplars of traditional Malvaceae and Bombacaceae, plus elements of traditional Sterculiaceae and Tiliaceae. The clade is strongly supported by bootstrap and decay values (di = 4; bs = 87) and is marked by a distinctive and unreversed 21-bp deletion (Fig. 1). Relationships within /Malvadendrina are poorly resolved, but eight subclades can be recognized (Fig. 2). We discuss the composition of these subclades and their support from morphological data, below. The second major clade, /Byttneriina (Fig. 2), contains exemplars from several tribes of Sterculiaceae and Tiliaceae. It is well supported by bootstrap and decay values (di = 4; bs = 85) and by a 6-bp insertion that is subsequently lost in three member taxa (Fig. 1). Two subclades within /Byttneriina are shown in Fig. 2 and considered further, below.

DISCUSSION

The ndhF sequence data provide considerable resolution of relationships within core Malvales. Many clades receive significant support, as judged by bootstraps and decay indices. Furthermore, even though indel events were not included in the analysis, they are largely concordant with the most parsimonious trees (Fig. 1). These analyses provide support for the monophyly of traditional Malvaceae, as predicted by Judd and Manchester (1997). In contrast, none of the other three families, as traditionally delimited, is monophyletic, a result that corroborates the findings of Bayer et al. (1999). By excluding Durioneae and Matisieae, the most parsimonious trees are compatible with, but offer no support of, the monophyly of the remaining, traditional Bombacaceae (see below). However, the presence of a unique 21-bp deletion as a synapomorphy of the /Malvadendrina clade (which contains all exemplars of traditional Malvaceae and Bombacaceae but only some Sterculiaceae and Tiliaceae) clearly indicates that the traditional limits of the latter two families are incompatible with either group being monophyletic.

/Malvadendrina (Fig. 2; unnamed by Bayer et al., 1999)

This clade is strongly supported by molecular data (Figs. 1 and 2; also Bayer et al., 1999). It is divided here into seven subclades: /Malvoideae, /Bombacoideae, /Sterculioideae, /Dombeyoideae, /Tilioideae, /Brownlowioideae, and /Helicteroideae. The relationships among these subclades are unresolved, except for the clade formed by /Malvoideae and /Bombacoideae (the /Malvatheca clade; Fig. 2; Baum, Alverson, and Nyffeler, 1998), which has moderate molecular support (di = 4; bs = 84).

Although the relationships among the /Malvatheca clade and the other five subclades of /Malvadendrina are not resolved by these data, there exist morphological characters that could serve to link some of these subclades. Members of /Brownlowioideae, /Helicteroideae, /Sterculioideae, and /Malvatheca have sepals fused into a tubular or campanulate, often lobed calyx. Burret (1926) recognized the potential value of this character and divided Tiliaceae into two major groups based on “;t1calyx symphyllus, campanulatus, superne lobatus” vs. “sepala libera,” with the former group containing his Brownlowioideae, all exemplars of which here fall into /Brownlowioideae (Fig. 2). In contrast, the sepals of most /Dombeyoideae and /Tilioideae are fused only slightly at the base or are essentially unfused. If one assumes that the plesiomorphic condition for sepals in core Malvales is of unfused sepals (as found for example in Bixaceae), then the fusion of sepals would tend to unite /Malvatheca with /Brownlowioideae, /Helicteroideae, and /Sterculioideae. However, homoplasy cannot be avoided under this scenario because of the essentially unfused sepals of Robinsonella (Malvaceae, probably /Malvoideae) and the occurrence of fused sepals in some members of the /Byttneriina clade (see below).

/Malvatheca, and most /Helicteroideae, /Dombeyoideae, and /Sterculioideae also have staminal tubes, i.e., filaments partially to completely fused into a ring. This provides another potential character for resolving relationships among the lineages of /Malvadendrina, perhaps suggesting a basal position for /Brownlowioideae and /Tilioideae.

/Malvoideae (Fig. 2; cf. the Malvoideae of Bayer et al., 1999)

At the core of /Malvoideae there is a weakly supported monophyletic group (di = 1; bs = 69) containing all exemplars traditionally placed in the Malvaceae. Hampea, which has been sometimes placed in the Bombacaceae (e.g., Hutchinson, 1967; Takhtajan, 1987), also falls in this core clade, as predicted by Fryxell (1968) based on chromosome number and other characters. Our study suggests further additions to the base of the /Malvoideae, particularly Camptostemon (di = 8; bs = 100; Templeton P = 0.079), which was described by Masters (1875) as a “genus curiously intermediate between Durioneae and Malveae.” This mangrove genus and its apparent close relative Papuodendron have been placed either in Durioneae (Bombacaceae) because of their indumentum of peltate scales and cup-like epicalices, or in Malvaceae (White, 1946; Kostermans, 1960). Features linking both genera to other /Malvoideae are staminal tubes, spiny pollen, and seeds with a medial ridge of long, bristly hairs, resembling those of some species of Hibiscus.

Matisia, Phragmotheca, and Quararibea, traditionally Bombacaceae, tribe Matisieae, appear monophyletic, as predicted from their shared, fleshy-fibrous, drupaceous fruits (Alverson, 1991). Baum, Alverson, and Nyffeler (1998) named this clade /Matisieae. The weakly supported inclusion of /Matisieae in /Malvoideae (di = 1; bs = 54) is quite plausible because of their seemingly “monothecate” anthers, highly fused staminal filaments (but with thecae sessile, in contrast to the stalked thecae of Camptostemon and traditional Malvaceae), tubular calyces, and simple, usually palmately veined leaves. However, none of these characteristics is a unique synapomorphy linking them with other /Malvoideae.

Chiranthodendron and Fremontodendron (Fremontieae, Sterculiaceae) here form a strongly supported clade (di = 4; bs = 98), as was expected (e.g., Bentham and Hooker, 1867, p. 982, as Cheirosternon and Fremontia). The uncertain relationship of these apetalous genera to other core Malvales was reviewed by Kelman (1991). Data from ndhF place them as a sister group to all other /Malvoideae, but the position is weakly supported (di = 1; bs = 52). In view of this, it is noteworthy that Fremontodendron and Quararibea were placed close to the /Malvoideae in some of the most parsimonious trees found by Judd and Manchester (1997). Alternatively, they have often been placed in the Sterculiaceae on the basis of their seemingly bithecate, tetrasporangiate anthers and apetalous flowers with colorful calyces (Table 1). However, the nature of these anthers and their potential utility as a synapomorphy are considered in the separate discussion of staminal characters, below. Apetaly and concomitant changes in calyces appear to have arisen in several lineages (e.g., some /Sterculioideae, Grewia, Heliocarpus, Triumfetta, and Lasiopetaleae), and thus do not provide compelling synapomorphies that would serve to unite Chiranthodendron and Fremontodendron with other traditional Sterculiaceae.

/Bombacoideae (Fig. 2; cf. the Bombacoideae of Bayer et al., 1999)

The monophyly of the group containing most traditional Bombacaceae, exclusive of Durioneae and Matisieae, is neither supported nor rejected by these data. In other words, our study was equivocal as to whether Ochroma and Patinoa comprise a single clade with the other core Bombacaceae. If they form a clade, then these two genera would seem to occupy the basal position. Ochroma and Patinoa have simple leaves, many seeds, and highly fused staminal filaments and, thus, resemble Chiranthodendron and Fremontodendron, which fall at the base of the /Malvoideae. Furthermore, the leaf characteristics of Ochroma (shape, indument, texture) and its petaloid-margined calyces are extremely similar to those of Chiranthodendron (and only slightly less so to those of Fremontodendron), suggesting either a close relationship between these genera or that these characters are plesiomorphic for the /Malvatheca. One of the parsimony analyses (with both stepmatrices, equal codon weights, and gapped characters deleted) moved Patinoa to a sister position with Bernoullia, accentuating the need for further investigation of relationships within the /Bombacoideae.

Exclusive of Ochroma and Patinoa, the core /Bombacoideae clade found in the analysis of the full data set of 70 taxa consists of 14 genera. Based on morphology, the unexamined genera Aguiaria, Neobuchia, Rhodognaphalon, and Rhodognaphalopsis almost certainly will fall here also. These core /Bombacoideae share the probable synapomorphy of palmately compound leaves (sometimes unifoliolate, e.g., all Huberodendron and Scleronema, most Catostemma, one Pseudobombax), with a single reversal to simple, palmately veined leaves in Cavanillesia.

/Sterculioideae (Fig. 2; the Sterculioideae of Bayer et al., 1999)

This strongly supported (di = 6; bs = 99), largely Old World clade consists of exemplars of three traditional tribes of Sterculiaceae: Mansonieae, Sterculieae, and Tarrietieae (Table 1). It exhibits several morphological novelties that may serve as local synapomorphies: unisexual or polygamous flowers, apetaly with petaloid sepals, and apocarpy. As noted by Jenny (1988), the apocarpy of Sterculia and relatives appears to involve a mechanism differing from that seen in Dombeya (traditionally Dombeyeae) or Keraudrenia (traditionally Lasiopetaleae). Kubitzki (1995) also reported apocarpy in Brownlowia and Christiana (Tiliaceae). Palmately compound leaves are plesiomorphic in the clade, or evolved at least once in this clade (e.g., Cola, Sterculia), possibly with reversals to cordate, palmately veined, simple leaves. The androgynophores and fused staminal filaments of /Sterculioideae might serve as synapomorphies of this clade, depending on a future assessment of the sister lineage, which was not identified by our analysis.

/Dombeyoideae (Fig. 2; the Dombeyoideae of Bayer et al., 1999)

Exemplars of four traditional tribes of Sterculiaceae—Dombeyeae, Eriolaeneae, Helmiopsidae, and Helictereae—form this well-supported clade (di = 3; bs = 97), which in the most parsimonious topology is a weakly supported sister clade to the /Tilioideae (di = 1; bs = 58). Within /Dombeyoideae, Dombeya and Eriolaena share a unique 6-bp insertion (Fig. 1). The /Dombeyoideae are native to the Old World. Staminal filaments are fused into short (e.g., some Dombeya) to long (e.g., Pterospermum) staminal tubes, except in Nesogordonia, the basal genus, which has free filaments (Hutchinson, 1967; Barnett, 1987a). Ovaries are mostly sessile but borne on a gynophore in Pterospermum. The topology here corroborates the work of Barnett (1987b), who proposed that tribes Dombeyeae, Eriolaeneae, and Helmiopsidae (except Nesogordonia) should be combined, based on the combination of hermaphroditic flowers with flat, unappendaged petals, ten or more monadelphous stamens, sessile, syncarpous gynoecia, bifid cotyledons, umbonate sarcotestal projections, and wood anatomical features. She also considered Nesogordonia to be anomalous but linked to other /Dombeyoideae by its wood anatomy.

If further lines of evidence corroborate the membership of the /Dombeyoideae seen here, the polyphyly of traditional Helictereae (with exemplars also in /Helicteroideae and /Byttneriina, below), may be explained by the extensive homoplasy of androgynophores at and below the tribal level in core Malvales. Notably, Zebe (1915) stated that Pterospermum was not closely related to other Helictereae, despite its stalked ovaries.

/Tilioideae (Fig. 2; the Tilioideae of Bayer et al., 1999)

The familiar genus Tilia appears to be sister to Craigia, a Chinese genus, usually placed in the Tiliaceae (e.g., Ren, 1989; Takhtajan, 1997). This clade is well supported (di = 4; bs = 96) and agrees with the morphological cladistic analysis of Judd and Manchester (1997), who found that these genera were related based on their oblate, pleurotreme pollen grains, flowers with five, elongate, antipetalous staminodia (in at least some Tilia), and embryos with folded cotyledons.

/Brownlowioideae (Fig. 2; the Brownlowioideae of Bayer et al., 1999)

Exemplars of Tiliaceae, subfamily Brownlowoideae, tribes Berryeae and Brownlowieae, form a strongly supported subclade (di = 6; bs = 96). Based on preliminary ndhF data, Diplodiscus and Jarandersonia (Diplodisceae) also fall into this clade (R. Nyffeler, unpublished data). The /Brownlowioideae are characterized by sepals fused into a campanulate tube (Burret, 1926), many stamens that are unfused or slightly fused into fascicles at their base, with or without staminodia (Ridley, 1922; Hutchinson, 1967), and ovaries sessile or borne on a short stalk (“torus”), i.e., a gynophore. Judd and Manchester (1997) noted that Brownlowia and Pentace, but not Berrya, have five elongate antipetalous staminodia, like some Tilia. However, it is not clear which of these traits, or others, may serve as a unique synapomorphy for the clade because of its uncertain relationship to other lineages of the /Malvadendrina.

/Helicteroideae (Fig. 2; the Helicteroideae of Bayer et al., 1999)

The /Helicteroideae subclade contains some exemplars of traditional Helictereae (Sterculiaceae) and all core Durioneae (Bombacaceae). This unanticipated but well-supported subclade (di = 4; bs = 100) was sister to all other subclades of /Malvadendrina. However, since the /Helicteroideae could be forced to be a sister to /Malvatheca with only one additional step (Fig. 2), their position is perhaps best regarded as unresolved. We evaluated the traditional hypothesis that members of tribe Durioneae (here represented by Durio and Neesia) are part of Bombacaceae by using monophyly-constraint searches to force Durio and Neesia into the /Malvatheca clade. This requires an additional six steps.

The /Helicteroideae have petals, fused sepals, five to many stamens with mostly unfused (e.g., fascicles in Durio and Neesia) to mostly fused filaments (Helicteres and Reevesia). Some Helicteres and Reevesia have staminodes and staminal thecae that become confluent (Robyns and Cuatrecasas, 1964; Solheim, 1991; see also Soepadmo and Eow, 1977, on Durio). The flattened cotyledons of Durio and Neesia are similar to those found in other Sterculiaceae, as noted by Hutchinson (1967), Baum and Oginuma (1994), and Judd and Manchester (1997) and may provide a synapomorphy at some level within /Malvadendrina upon further study.

/Byttneriina (Fig. 2; unnamed by Bayer et al., 1999)

This major clade consists of exemplars of Sterculiaceae and Tiliaceae. The /Byttneriina clade is well supported (di = 4; bs = 85) but, like its sister clade, /Malvadendrina, does not exhibit obvious morphological synapomorphies. Many members of /Byttneriina have leaves with coarsely dentate or doubly serrate margins, but this is likely a plesiomorphic state for core Malvales. This leaf characteristic is also seen in closely related outgroups (e.g., Cochlospermum), as well as in /Dombeyoideae and /Tilioideae, and may be lost under certain ecological conditions (e.g., in some rainforest /Byttnerioideae). The /Byttneriina clade includes at least two subclades supported by this analysis: /Grewioideae and/Byttnerioideae.

/Grewioideae (Fig. 2; the Grewioideae of Bayer et al., 1999)

This strongly supported clade (di = 13; bs = 100; Templeton Ts = −170, p = 0.048) contains exemplars of Tiliaceae, tribes Apeibeae, Coloneae, Corchoreae, Enteleeae, Grewieae, Lueheeae, Pseudocorchoreae, Sparmannieae, and Triumfetteae. The clade gains further support from an unreversed 6-bp insertion (Fig. 1), and a homoplasious 6-bp insertion marks the sublineage to which the exemplars of Corchoreae, Pseudocorchoreae, and Triumfetteae belong.

A potential synapomorphy for /Grewioideae is an epidermal modification of the adaxial petal surfaces. This supposes, however, that the pitted-glandular petals of Colona, Luehea, and Trichospermum are homologous with the carinate surfaces of Corchoreae and Sparmannieae (Hutchinson, 1967). Loss of sepal fusion also may be a synapomorphy for the group if the basal condition in core Malvales is fused sepals. Assuming that the basal condition for core Malvales is many, free, fertile stamens (cf. potential sister clades such as Cochlospermaceae; see also Judd and Manchester, 1997), then another potential morphological synapomorphy is the sterilization of the outer stamens, which is seen in some members of /Grewioideae. Stamens with appendages of connective tissue also occur in some members (e.g., Apeiba and Glyphaea) and may help define subclades within /Grewioideae upon further study.

/Byttnerioideae (Fig. 2; the Byttnerioideae of Bayer et al., 1999)

Exemplars of Sterculiaceae tribes Byttnerieae, Helictereae, Hermannieae, Lasiopetaleae, and Theobromeae form this weakly supported clade (di = 1; bs = 52). Two subclades of /Byttnerioideae are suggested by this study. These “byttnerioid” and “lasiopetaloid” subclades were not given formal rankless names by Baum, Alverson, and Nyffeler (1998). A separate study by Whitlock et al. (unpublished data) will provide a more thorough sampling and greater phylogenetic detail.

Members of the /Byttnerioideae clade show a general reduction of stamen number compared to /Grewioideae and putative outgroups. Glossostemon, which also falls in the /Byttnerioideae (B. Whitlock, unpublished data), is an exception with 30 stamens (Freytag, 1951). Another potential synapomorphy of /Byttnerioideae is that the fertile stamens are gathered into antipetalous fascicles and that staminodia, when present, are antisepalous. A preliminary assessment of Rulingia and Commersonia, which fall near the base of the lasiopetaloid clade (B. Whitlock, unpublished data), suggests that another possible synapomorphy for /Byttnerioideae are petals that are broad or concave at their base. Finally, Bayer and Kubitzki (1996) note that the inflorescence morphology of Keraudrenia, a putatively unspecialized member of the Lasiopetaleae, shows homology to members of Byttnerieae, indicating another potential synapomorphy for /Byttnerioideae.

Traditional Byttnerieae and Theobromeae constitute the moderately supported byttnerioid subclade seen here (di = 3; bs = 58); Kleinhovia (traditionally Helictereae) also falls into this subclade. The separation of Byttnerieae and Theobromeae (subfamily Byttnerioideae) from the rest of traditional Sterculiaceae, predicted by Edlin (1935), is supported also by rbcL sequence data (Whitlock, Alverson, and Baum, 1996; Alverson et al., 1998; Bayer et al., 1999). Probable synapomorphies for this subclade include the fusion of filaments into staminal tubes bearing antipetalous fertile stamens at their apices (generally in threes), alternating with antisepalous staminodia, and cucullate-elaborate petal morphology. Kleinhovia shares these morphological traits and is strongly supported as a member of the byttnerioid subclade, an association suggested on morphological grounds by Zebe (1915).

The “lasiopetaloids,” a second subclade of /Byttnerioideae suggested by this study (di = 5; bs = 88), is composed of Waltheria and Hannafordia, exemplars from tribes Hermannieae and Lasiopetaleae, respectively (both Sterculiaceae). Within lasiopetaloids, staminodia may have been lost in some Hermannieae and, presumably independently, in some Lasiopetaleae. If broadly based petals are a symplesiomorphic condition in the lasiopetaloid subclade, as we suspect, then they have been lost in most members of Lasiopetaleae.

The correspondence of morphology with the ndhF topology

These molecular data suggest that core Malvales have experienced multiple gains (and/or losses) of androgynophores, staminal tubes, palmately compound leaves, palmate- (vs. pinnate-) veined leaves, apetaly, spiny pollen, apocarpy, winged seeds or fruits, the herbaceous habit, and “monothecate“ anthers. Thus, none of these character states can serve as an unambigous guide to monophyly at higher levels within /Malvaceae. Nonetheless, this does not preclude the utility of these characters at lower levels. For instance, the presence of a gynophore does not distinguish Sterculiaceae from Tiliaceae in the traditional sense, nor does it appear to serve as a synapomorphy for traditional Helictereae, yet the gain or loss of gynophores may help diagnose lineages within some of the subclades identified by our study. Similarly, the current tree topology suggests that palmately compound leaves were derived at least four times in the core Malvales: in the /Sterculioideae, /Bombacoideae, /Byttnerioideae (Herrania), and /Malvoideae (Cienfuegosia and Gossypium). Despite this homoplasy, the gain of compound leaves may serve as a legitimate synapomorphy for, or within, each of these subclades.

The nature and significance of staminal characteristics in core Malvales have been particularly intriguing. It has long been assumed that monothecate anthers (“half-anthers” or “monoloculate” anthers) offered some clear insight into the structure of the core Malvales, namely, that the presence of this character state was a derived character uniting Malvaceae and Bombacaceae (van Heel, 1966; Cronquist, 1988; Judd, Sanders, and Donoghue, 1994; Judd and Manchester, 1997).

In contrast, the topologies obtained here suggest that the “monothecate” anthers of some Durio, which have been the basis for traditional placement of this and several other, similar genera (Coelostegia, Cullenia, Kostermansia, and Neesia) in the traditional family Bombacaceae, may not be homologous with the “monothecate” anthers seen in /Malvatheca (cf. van Borssum Waalkes, 1966). We postulate that the “monothecate” anthers of some Durio and allies were derived from a dithecate ancestor, perhaps by a direct fusion of the adjacent thecae of a single anther (Soepadmo and Eow, 1977; the secondary disporangy of Endress and Stumpf, 1990), or by simple splitting of each anther into two separate thecae (e.g., Kostermans, 1960). In contrast, a superficially similar condition in /Malvatheca may have been derived by an independent, more complicated route, as follows.

The basal branches within the /Malvatheca (including Fremontodendron, Gyranthera, Huberodendron, Ochroma, and Patinoa) all show elongate anthers with transverse septa, suggesting that anthers are “polysporangiate by transverse septation” (Endress and Stumpf, 1990), rather than “monothecate.” Transversely septate anthers are also known from Septotheca (Bombacaceae), which was not included in this study, but which will almost certainly fall near the base of /Malvatheca. All of these genera have their staminal filaments highly fused into a tube, also suggesting that this may be a basal condition in /Malvatheca. In light of the position of these taxa in the phylogeny, it becomes more parsimonious to place polysporangiate anthers as the basal condition, rather than monothecate anthers.

We postulate that one or more reductions in the degree of filament fusion may have occurred subsequently within /Bombacoideae to partially fused (e.g., Ceiba and Pachira) or almost unfused (e.g., Bombax and Catostemma) filaments. In this scenario, as the androecia evolutionarily “unzipped,” the free portions of the filaments retained only two of the previously many sporangia from a single anther (with exceptions such as Spirotheca and some Ceiba s.l. which retained more), thus yielding pollen sacs that appear to be monothecate and bisporangiate.

A parallel process may have occurred in /Malvoideae. Each anther of Fremontodendron, the putative basal lineage of /Malvoideae, exhibits elongate, parallel pollen sacs that are transversely divided by septa into 20–30 contiguous sporangia (Endress and Stumpf, 1990). These have been interpreted as dithecate, tetrasporangiate anthers in the past, and hence this genus and the closely related Chiranthodendron have commonly been seen as members of Sterculiaceae (e.g., Kelman, 1991). We suggest that these anthers may be interpreted alternatively as dithecate and polysporangiate. In /Matiseae, another basal branch of /Malvoideae, Phragmotheca exhibits elongate, transversely septate thecae of irregular lengths with varying numbers of component sporangia (Alverson, 1991). This genus may represent a developmental state in which some but not all of the thecae have disaggregated to become bi- (vs. poly-) sporangiate, while the filaments themselves remain almost completely fused. Finally, in core /Malvoideae, staminal filaments are completely unfused apically and bear single (“monothecate”), bisporangiate pollen sacs, sometimes in pairs on bifurcating filaments. In sum, under both the traditional and this alternative hypothesis of stamen evolution, it is clear that the androecium of core Malvales shows a level of developmental plasticity that is unusual among eudicots and warrants further study.

CONCLUSIONS

This study demonstrates that using molecular tools, we can identify robust phylogenetic structure within the core malvalean clade. However, the relationships among many subclades noted here remain unresolved. To extend these results will require further sampling of taxa and corroboration from additional gene sequences. Additional exemplars of Sterculiaceae and Tiliaceae are needed, as well as odd taxa whose tribal affiliations have always been obscure. To this end, we are currently sequencing remaining taxa of potential /Bombacoideae and basal /Malvoideae to better determine the relationship between these two groups. In addition, the results reported here must be compared and combined with data from other genes (especially nuclear), as well as with morphological data. The significance of transversely septate anthers seen in Nesogordonia (/Dombeyoideae) and fossil Florissantia (Manchester, 1992) should be evaluated with regard to the ancestral condition of stamens of /Malvatheca. Careful assessments of the morphology and development of other, seemingly homoplasious structures, such as epicalices (cf. Bayer, 1999) and gynophores (cf. Jenny, 1988), may also help elucidate the evolutionary diversification of the core Malvales.

Table 1. Vouchers for ndhF sequences used in this study and the classification of each genus according to three modern treatments

Table 1. Continued

Table 1. Continued

Table 2. Distribution of variable sites in the ndhF molecule of core Malvales. The proportions given in the first row of each base substitution category represent the percentage of sites with the given type of base substitution over the entire ndhF sequence used in the analyses (2226 bp), or within the 5′ region (1350 bp) or the 3′ region (876 bp) of the sequence only. The proportions in rows 2 through 4 of each category represent the proportion of that type of base substitution, within the specified region, that occur at the first, second, and third codon positions

Figure 1

Phylogram of one of the 46 shortest trees found in the 64-taxon, equally weighted parsimony analysis. Twelve phylogenetically informative indel events within core Malvales are mapped onto this tree by simple parsimony. Insertions are indicated by solid squares and deletions by hollow squares. Letters above the bars identify particular indels and numbers below the bars indicate the number of bases involved in the indel event. Letters to the right of the bars show associated reversals (e.g., b reversing to b′). Forced monophyly of taxa with indels b, i, j, and k, using ten TBR/MULPARS RAS per analysis, resulted in trees with two, 41, 26, and six additional steps, respectively. These phylogenetically informative indel events map to the following positions on the ndhF gene of Nicotiana (Olmstead et al., 1993): a, 1704/1705; b, 1482/1483; c, 1530/1531; d, 1561/1562; e, 1545/1546; f, 1491/1492; g, 1443/1444; h, 1482/1483; i, 1593/1594; j, 1524/1525; k and k′, 651/652. Two additional indel events represent homoplasy and are not shown here: insertion e also occurred in Spirotheca of the core bombacoids, which was included in the 70-taxon data set; and a 9-bp insertion at location 1491/1492 (relative to Nicotiana) was seen in all taxa except Rhopalocarpus.

Figure 2

Tree representing the consensus of 46 trees at 1666 steps. CI = 0.650 (including all characters) and RI = 0.682. Numbers above the branches represent decay values. Numbers below these branches are bootstrap values. Rankless names (after Baum, Alverson, and Nyffeler, 1998) are given for the 12 major clades and subclades of core Malvales (i.e., /Malvaceae). Ranked names proposed by Bayer et al. (1999) for most clades are identical to the rankless names shown here, except that the rankless names are preceded by a forward slash to indicate that they are phylogenetically defined clade names. Ranked names have not been proposed for the clades shown in boldface, i.e., /Malvatheca, /Malvadendrina, and /Byttneriina.