Molecular phylogenetics of Melastomataceae and Memecylaceae: implications for character evolution^†

Taxon sampling, DNA isolation and amplification, and sequence alignment

Table 1 lists all species newly sequenced for this study, with sources and GenBank accession numbers. The species represent 42 of ∼150 currently recognized genera. Trees were rooted with species of Crypteroniaceae, Alzateaceae, Rhynchocalycaceae, Oliniaceae, Penaeaceae, and Myrtaceae, with Ludwigia (Onagraceae) added to represent more distant Myrtales (Conti, Litt, and Systma, 1996).

Total DNA was isolated from silica gel-dried, herbarium, or fresh leaves using a modified CTAB procedure (Smith et al., 1991), DNeasy plant mini kits (QIAGEN Inc., Valencia, California, USA), or NucleoSpin plant DNA extraction kits (Macherey-Nagel GmbH & CoKG, Dören, Germany) according to manufacturers' instructions. Standard polymerase chain reaction (PCR) protocols were used, but since Melastomataceae DNA generally works poorly, amplifications often had to be repeated several times to obtain enough product.

The rbcL gene was amplified using primers developed by Fay, Swensen, and Chase (1997) and the ndhF gene with primers developed by Olmstead and Sweere (1994). We amplified the exon between positions 972 (i.e., codon 305 of solanaceous sequences; Olmstead and Sweere, 1994) and 1955, using forward primer ndhF-972F, reverse primer ndhF-1955R, and one or two pairs of internal primers (ndhF-1318F, ndhF-1318R, ndhF-1603F, and ndhF- 1603R). The large intron that interrupts the rpl16 gene was amplified using primers 1067F and 18R (Asmussen, 1999). PCR products were purified either by running the entire product on a low-melting point agarose gel and then recovering the amplified DNA with the help of QIAquick gel extraction kits (QIAGEN) or by using QIAquick PCR purification columns directly, without a prior gel purification step. Cycle sequencing of the amplified double-stranded products was conducted with the ABI Prism Dye Terminator Cycle Sequencing Ready Reaction kit (Perkin Elmer, Norwalk, Connecticut, USA), using 2.5 ng of primer in a 5-μL reaction volume. Sequencing reactions were purified by ethanol precipitation and run on ABI 373 or ABI 377 automated sequencers at the universities of Mainz (ndhF, rbcL p.p.) or Missouri-St. Louis (rpl16, rbcL p.p.). Usually, both strands of DNA were sequenced and used to generate a consensus sequence using Sequencher software (version 3.1; GeneCodes Corp., Ann Arbor, Michigan, USA). Alignment was done manually.

For the combined 3-genome region-53-taxon analysis, sequences from the same species and usually from the same total DNA extract were spliced together, with the following exceptions (compare Table 1): Gravesia viscosa rbcL and rpl16 were combined with Gravesia guttata ndhF; Melastoma malabathricum rbcL was combined with M. sanguineum ndhF and rpl16. Memecylon bakerianum rbcL and ndhF were combined with M. edule rpl16; Tibouchina urvilleana rbcL was combined with T. longifolia rpl16 and ndhF; and Triolena obliqua ndhF and rbcL were combined with T. pustulata rpl16. In one case, sequences to be spliced came from different genera; rbcL and rpl16 of Tococa were supplemented by ndhF from Maieta. In a few other cases where rbcL or ndhF could not be obtained for a species (Table 1), missing data symbols (“nnnn”) were entered for that region.

Phylogenetic analyses

Phylogenetic analyses of the aligned sequences were conducted with test version 4.0b.2 of PAUP* (Swofford, 1998). Parsimony analyses were performed using heuristic searches, ten random-taxon- addition replicates and tree bisection-reconnection (TBR) swapping. All minimal trees were saved. The COLLAPSE, but not the STEEPEST DESCENT, options of PAUP were in effect during all searches, and character changes were interpreted under ACCTRAN optimization. Characters were unweighted and unordered, and gaps were treated as missing data. Under parsimony, nonparametric bootstrap support (Felsenstein, 1985) for each clade was estimated based on 1000 replications, using closest taxon addition and TBR swapping. Most-parsimonious trees were generated independently for the three data sets, followed by bootstrap analyses, to assess whether there was statistically supported conflict (i.e., with >50% bootstrap support) among data sets. In the absence of such conflict, the data were combined in a global analysis.

Maximum likelihood (ML) analyses were performed using the general time-reversible model (GTR; Yang, 1994), which estimates independent probabilities for all possible substitutions types in addition to accounting for unequal base frequencies. Rate heterogeneity among sites affects the performance of different tree reconstruction methods, and its estimation has received considerable recent attention (Yang, 1996; Sullivan, Swofford, and Naylor, 1999; Takezaki and Gojobori, 1999). A method for explicitly dealing with this kind of rate variation is the combination of an invariable-sites model, in which some proportion of sites (P_inv) is assumed to be completely resistant to change, with a gamma (Γ)-distributed-rates model in which the distribution of relative rates over sites is assumed to follow a Γ distribution whose shape parameter (α) determines rate heterogeneity. The dependence of P_inv and α on tree topology is minor as long as strongly supported groups are maintained (Yang and Kumar, 1996; Sullivan, Swofford, and Naylor, 1999). Both parameters can therefore be estimated for distance trees from the same data without complete branch swapping, which greatly reduces the computational demands of maximum likelihood searches. We estimated P_inv and α simultaneously, using the discrete gamma approximation of Yang (1994; implemented in PAUP*) with four rate categories to approximate the continuous gamma distribution. Base frequencies were the empirically observed ones.

Starting trees for ML searches were minimum-evolution trees, using LogDet distances, and the swapping strategy was nearest-neighbor-interchange swapping. We used quartet puzzling (Strimmer and von Haeseler, 1996; implemented in PAUP*), a fast tree search algorithm that allows analysis of large data sets, to obtain estimations of support for internal branches in the ML trees. These values are thought to have the same practical meaning as bootstrap values (Strimmer and von Haeseler, 1996).

We also calculated the likelihood score for a tree obtained under the Hasegawa-Kishino-Yano + P_inv + Γ model (HKY; Hasegawa, Kishino, and Yano, 1985) to assess whether the more parameter-rich GTR model fit the data significantly better, as judged by a likelihood ratio test, using four degrees of freedom (cf. Sullivan, Swofford, and Naylor, 1999). Both models yielded a single best trees that differed only in the placements of Heterocentron, Monochaetum, Pterolepis, and Tibouchina. However, the placement of these genera relative to each other was not well supported in any of the reconstructions. The likelihood ratio test rejected the HKY model in favor of the GTR model (χ² = 2(16427.95 − 16363.10) = 129.7; P < 0.001; 4 df). We therefore used the GTR + P_inv + Γ model as the most appropriate for our data.

Sequence data

Each of the sequenced regions is characterized in Table 2. In the case of rbcL, a total of 1398 nucleotides, from positions 30 to 1428 of the rbcL exon were used in the analyses. For ndhF, the aligned sequences, with length variations that introduced gaps, had a length of 1021 nucleotides. The completed alignment of the rpl16 sequences, with gaps, comprised 1045 nucleotides. We excluded base pairs 217–387 because of alignment ambiguity between the ingroup and the outgroup, mainly due to huge inserts in Crypteronia and Rhynchocalyx. The concatenated sequences thus comprised 3464 nucleotides, of which 170 were eliminated. This matrix contained 10% autapomorphic variable sites and 23% parsimony-informative sites when all 53 genera were included. Six Melastomataceae lacking rbcL sequences (Aciotis, Blastus, Centradenia, Nepsera, Phyllagathis) were excluded from most ML searches.

Of 35 sequence-length mutations, most occurred in the rpl16 intron (Table 2), and several were diagnostic of the ingroup. Nucleotide compositions of the two genes and the intron differ barely (Table 2). Under the HKY model, the average transition-to-transversion ratio across all sequences was 0.77. It was 0.95 and 0.88 for the two genes, and 0.78 for the intron (Table 2).

Rate heterogeneity among sites is measured by α, which is inversely related to the extent of rate variation. Table 2 shows the values for α, estimated under the GTR model. For rpl16, α is almost 1 (1.01), indicating a random distribution of the rates at which sites are changing, while for rbcL, α is 0.74, indicating that most sites have very low rates while some change at very high rates. For ndhF, α is >1, indicating that most sites have intermediate substitution rates, while a few have very high or very low rates.

Phylogenetic analyses

No hard incongruencies were found among strict consensus trees obtained from the individual data sets (not shown), and parsimony analysis of the concatenated sequences showed the same clades as seen in the individual analyses, only with higher bootstrap values. Figure 1 shows the strict consensus of the three equally parsimonious trees (Length = 2443, consistency index [CI] = 0.62, retention index [RI] = 0.80), all in a single tree island. A long branch separates Memecylaceae + Melastomataceae (with 100% bootstrap support) from their closest relatives, a clade of Crypteronia (Crypteroniaceae) + Alzatea (Alzateaceae) + Rhynchocalyx (Rhynchocalycaceae) + Olinia (Oliniaceae) + Penaea (Penaeaceae), Myrtaceae, and Onagraceae (see the midpoint-rooted trees in Fig. 1). We will subsequently refer to the former group of families as the CAROP clade. If trees are rooted with Ludwigia (Onagraceae), a sister-group relationship between the CAROP clade and Memecylaceae/Melastomataceae has 100% bootstrap support. Pternandra appears to be the first-branching Melastomataceae, but support for this placement of Pternandra is low (72%). The next-basal branch is Astronia, with a bootstrap support of 100%.

The single best tree resulting from the ML analysis under the GTR + P_inv + Γ model (Fig. 1) shows the same topology as the parsimony tree. The monophyly of Melastomataceae + Memecylaceae again is well supported (98%) and the placement of Pternandra as basal in Melastomataceae poorly (65%).

Within core Melastomataceae, several major groups can be discerned (Fig. 1; tribe names for these groups are shown in Fig. 2). They are: (1) a clade comprising the two species of Pternandra included in the combined analysis (five species of this genus of 15 species were sequenced [Table 1], but not for all genome regions); (2) a clade comprising the two species of Astronia (Astronieae), (3) a clade consisting of Merianieae (Adelobotrys, Graffenrieda, Meriania) and (4) their sister group Miconieae (Clidemia, Leandra, Maieta, Tetrazygia, Tococa) plus Macrocentrum, a genus traditionally placed in Bertolonieae; (5) a clade comprising Bertolonia, Monolena, and Triolena (Bertolonieae); (6) a clade comprising Dissochaeteae (Diplectria, Medinilla) and, nested within them, Sonerileae/ Oxysporeae (Amphiblemma, Blastus, Calvoa, Driessenia, Gravesia, Phyllagathis) plus Blakea, the sole representative of Blakeeae; (7) a clade comprising Melastomeae/Rhexieae; and (8) a Microlicieae clade. The second genus of Blakeeae, Topobea, was sequenced for ndhF and is sister to Blakea in terms of that gene (tree not shown).

The degree of genetic differentiation among Memecylaceae, Melastomataceae, and Pternandra becomes apparent when the data are visualized as a phylogram (Fig. 2). The branches leading to these taxa are among the longest in the ingroup.

Interfamilial relationships of Melastomataceae

The results of this study support Conti, Litt, and Sytsma's finding (1996) that Melastomataceae and Memecylaceae are sister to a small Southeast Asian/Neotropical/South African clade. This clade consists of Crypteroniaceae (ten spp. in Southeast Asia; long considered a close relative of Melastomataceae on the basis of morphology [cf. Renner, 1993, and references therein]), Alzateaceae (one sp. in South and Central America; Silverstone-Sopkin and Graham, 1986), Rhynchocalycaceae (one sp. in Natal), Oliniaceae (8–10 spp. in South Africa), and Penaeaceae (20 spp. in South Africa). These families share leaf stipules, haplostemonous flowers (Johnson and Briggs, 1984), and ephemeral endothecia (Tobe and Raven, 1983, 1984a, b, 1987a, b). Ephemeral endothecia degenerate early on, and anthers therefore dehisce not via differential shrinking of endothecium cells, but via rupture of walls along their thinnest sections caused by the shrinking of connective cells (H. Tobe, Kyoto University, personal communication). The sister-group relationship between Melastomataceae/Memecylaceae and the CAROP clade is supported by two morphological characters, viz. opposite leaves and stamen connectives that are dorsally enlarged and often massive (Fig. 3). The latter trait may relate to the connectives' role in anther dehiscence or may be correlated with the incurved position of the stamens in bud found in the CAROP families, Melastomataceae, and Memecylaceae. An inflexed bud position may create a tendency for abnormal growth at the points of greatest curvature (Ziegler, 1925; Leinfellner, 1958; Jacques-Félix, 1994).

Monophyly of Melastomataceae

Arguments about the circumscription of Melastomataceae, whether narrowly to exclude Memecylaceae or widely to include that family, have always hinged on the placement of Pternandra. Pternandra is a genus of 15 species of trees that is most species rich in Borneo, but extends into peninsular Malaysia (Maxwell [1981] included in Pternandra the genus Kibessia and two others that traditionally made up Kibessieae [Krasser, 1893]). Pternandra is characterized by fleshy capsules with dorsal-median placentas (Maxwell, 1981; Clausing, Meyer, and Renner, 2000) and wood with interxylary phloem islands. The latter trait is also found in Memecylaceae, causing wood anatomists to argue that Pternandra was closer to Memecylaceae than to Melastomataceae (van Tieghem, 1891a, b; Janssonius, 1950; van Vliet, 1981; van Vliet, Koek-Noorman, and ter Welle, 1981). This provided an argument for circumscribing the family widely, with Pternandra as the “link” between two phenetic groups. Similar interxylary phloem, however, is found in at least one species of Melastomataceae, Dissotis leonensis (D. Normand in Jacques-Félix, 1994, p. 250; Dissotis is nested in Melastomeae; Figs. 1 and 2) and is common in alliances with intraxylary phloem, such as Myrtales. It may be present or absent within single genera or individuals, for example, in the roots, but not stem, of Lythrum salicaria, also a myrtalean taxon (van Tieghem, 1891a, b; Metcalfe and Chalk, 1983). Indeed, van Vliet (1981) concluded that “Pternandra is […] nearest to the Melastomatoideae [= Melastomataceae], being similar in the ray type and the coarse vessel-ray and vessel- parenchyma pits and the scanty paratracheal parenchyma.” Vessel-ray pits in Pternandra are simple as in Melastomataceae. In contrast, Memecylaceae have half-bordered vessel-ray pits. These and other anatomical similarities of Pternandra to Melastomataceae—for example, Pternandra has radially included phloem in addition to its axially included phloem, a trait otherwise only found in the higher Melastomataceae Medinilla (van Vliet, 1981)—argue against the possibility that interxylary phloem is plesiomorphic in Memecylaceae and Melastomataceae, and lost in higher Melastomataceae.

A second character possibly linking Pternandra and Memecylaceae, discovered during the course of this investigation, is the presence of a fibrous endothecium in both lineages (Fig. 4). The presence or absence of an endothecium appears to be an important phylogenetic marker in the CAROP/Melastomataceae/Memecylaceae alliance, as well as within Melastomataceae, and the character is discussed in detail below (see Relationships within Melastomataceae).

Vliet, Koek-Noorman, and ter Welle's (1981) placement of Pternandra in Memecylaceae on the basis of the axially included phloem is contradicted by leaf venation. Pternandra and Melastomataceae both have acrodromal venation, while Memecylaceae have pinnate or brochidodromal venation (Morley, 1953; Dahlgren and Thorne, 1984; Johnson and Briggs, 1984; G. Clausing and S. S. Renner, personal observations, but see below). Brochidodromal venation is a subtype of pinnate venation in which the secondary veins anastomose close to the leaf margin, which can result in venation patterns that resemble acrodromal venation. Also, leaf clearings by Jacques-Félix, Mouton, and Chalopin (1978) and Klucking (1989) of species from five of the six genera of Memecylaceae—Memecylon, Mouriri, Lijndenia, Spathandra, and Warneckea (Votomita was not studied)—show that Memecylaceae venation can occasionally be truly acrodromal. The thick, opaque leaves of Memecylaceae make observation difficult, and the deeply scalloped courses of the lateral pair of primaries in both brochidodromally and acrodromally veined Memecylaceae obscure the venation's true nature (Klucking, 1989). A detailed phylogeny of Memecylaceae is needed to evaluate whether acrodromal venation is ancestral in this family, and pinnate and brochidodromous venations are secondarily derived as argued by Jacques-Félix, Mouton, and Chalopin (1978; see also Jacques-Félix, 1978, 1994), or whether Memecylaceae are ancestrally pinnate/brochidodromous (Johnson and Briggs, 1984; Renner, 1993). Scalloped primary veins are not seen in Melastomataceae (including Pternandra), indicating that there may be family-specific differences between Melastomataceae and Memecylaceae in the timing of lateral leaf expansion relative to the time when secondary veins join the lateral primaries (Klucking, 1989).

With Pternandra being the first-branching Melastomataceae, the question whether Memecylaceae should be included in Melastomataceae or ranked as a family, reduces to a matter of ranking and pragmatics of family identification. Among the morphological synapomorphies of Memecylaceae are dorsal glands on the stamen connectives (Fig. 3d), terminal leaf sclereids, paracytic stomates, axially included phloem islands in the secondary wood, fixed epigyny, and one or few large seeds with storage cotyledons (additional differences are listed in Table 3 in Renner, 1993). These traits are not found in the CAROP clade or Melastomataceae (except that Pternandra has the phloem islands) and can serve to distinguish Memecylaceae from Melastomataceae in cases where a look at the leaf venation does not suffice.

In the current DNA data, Memecylaceae are represented by two species of Memecylon, one from Madagascar and one from Southeast Asia, and two species of Mouriri, one from the Amazon basin and the other from Puerto Rico. Two Memecylaceae from Africa (Warneckea membranifolia, Memecylon cogniauxii) were sequenced for ndhF, and in an ndhF tree they group with the other species of Memecylon. We are sequencing additional species of Memecylaceae for a low- copy nuclear gene to further test the position of Pternandra.

Relationships and major morphological transitions within Melastomataceae

Melastomataceae form a monophyletic clade that is supported morphologically by the fixation of acrodromal venation (Fig. 2). The family appears to be the largest clade of flowering plants characterized by this type of venation; only a few isolated taxa, for example, Heterocentron, Sonerila, Loreya nigricans, and Macairea rufescens, have pinnate venation (Renner, 1989a, 1993).

The next deepest split in the family is that between Pternandra and all other Melastomataceae (compare the phylogram, Fig. 2). Melastomataceae above Pternandra are characterized by lack of an endothecium in mature anthers. The absence of endothecia in Melastomataceae has often been noted (Ziegler, 1925; Matthews and Maclachlan, 1929; Subrananyam, 1948; Favarger, 1952; Eyde and Teeri, 1967; G. Clausing and S. S. Renner, personal observations), but Pternandra had not been investigated prior to this study. It has a fibrous endothecium resembling that of Memecylaceae (compare illustrations in Venkatesh, 1955), except that in Pternandra the endothecium surrounds the ventral half of each locule, whereas in Memecylon it encloses the entire locule (Fig. 4). Memecylaceae anthers open by slits in Memecylon (Fig. 3d) and by short, drop-shaped slits that function as pores in Mouriri. By contrast, most Melastomataceae have anthers that open by pores. Melastomataceae pores develop in a patch at the tip of the anthers, where the epidermis is reduced and exposed mesophyll dries out and shrivels up. Poricidal dehiscence is an adaptation to pollinators capable of collecting pollen by high- frequency vibration of stamens (Harris, 1905; Buchmann and Buchmann, 1981; Renner, 1989b, 1990a; Gross, 1993; Larson and Barrett, 1999). Poricidally dehiscent Memecylaceae and Melastomataceae are both pollinated by pollen-collecting bees, but they may have acquired this mode of pollination independently. Unfortunately, nothing is known about the mode of pollen collection in Pternandra and Astronia.

A morphology-based cladistic analysis showed the Southeast Asian Astronieae—Astronia, Astronidium, Astrocalyx, and Beccarianthus (together 150 spp.)—as sister to all Melastomataceae except Pternandra, which had been designated as the functional outgroup (Renner, 1993). This placement of Astronieae was supported by the fixation of poricidal anther dehiscence and axillary placentation in the sister clade to Astronieae (Fig. 2). Astronieae anthers open by longitudinal slits (Fig. 3b), albeit without the help of an endothecium. Their capsules have basal to basal-axile placentas (Maxwell and Veldkamp, 1990a, b). This morphological topology is strongly supported by the molecular data (Fig. 1).

Of the major clades found within the higher Melastomataceae, two had been proposed based on morphology, viz. Melastomeae sensu lato (uniting the neotropical Tibouchineae with the paleotropical Osbeckieae) and a Microlicieae + Melastomeae sister-group relationship (Renner, 1993), but others, such as the Blakeeae + Dissochaeteae (including Sonerileae) clade, contradict morphological hypotheses. Also, although current sampling of Miconieae and Sonerileae, tribes that each comprise 27–30 genera, is sparse, our data refute Renner's (1993) merging of Miconieae with Dissochaeteae and of Sonerileae with Bertolonieae. The former two tribes were thought to uniquely share fleshy berries. However, an anatomical comparison of fruits of Miconieae and Dissochaeteae (Clausing, Meyer, and Renner, 2000) has shown that they are heterogeneous, in agreement with an independent evolution of berries from capsules in the paleotropical Dissochaeteae and neotropical Miconieae.

The traditionally recognized Bertolonieae (Bertolonia, Diplarpea, Macrocentrum, Monolena, Salpinga, and Triolena and the phenetically more isolated Maguireanthus, Opisthocentra, Tateanthus, and Boyania; Wurdack, 1964) are predominantly herbaceous and share (usually) triquetrous capsules, ovaries with apical scales surrounding the style, and (often) scorpioid inflorescences. These characters were thought to unite them with the paleotropical Sonerileae. That this needs to be reevaluated is suggested by the widely separate placements of Bertolonia/Triolena and Monolena from Macrocentrum, and of Bertolonieae from Sonerileae.

Instead of grouping with Bertolonieae, Sonerileae sensu stricto (Amphiblemma, Calvoa, Gravesia, Phyllagathis) and Oxysporeae (Blastus, Driessenia) were found to be nested within Dissochaeteae (Diplectria, Medinilla; Figs. 1 and 2). That Oxysporeae and Sonerileae form a close alliance and should be merged has been pointed out repeatedly (van Vliet, 1981; Renner, 1993). The separation of these tribes was based on whether capsules were more or less round and had a conical apex (Oxysporeae) or strongly 3–5-angled with a concave apex (Sonerileae). Several genera, e.g., Bredia and Driessenia, have been moved back and forth by different workers, indicating the tribes' problematic distinction (compare Cogniaux, 1891; Diels, 1932; Hansen, 1985). On the other hand, the evolution of the capsular-fruited and partly herbaceous Sonerileae/ Oxysporeae from the berry-fruited and often climbing Dissochaeteae implied by molecular topologies is surprising. However, the same nesting is found in a larger ndhF analysis that includes species from ten genera of Dissochaeteae and nine of Sonerileae/Oxysporeae (Clausing, 1999; Clausing and Renner, 2001). Also, fruit characters are highly labile in the family and conserve little phylogenetic signal (Clausing, Meyer, and Renner, 2000).

The three sampled genera of Merianieae, on the other hand, form a robust clade, sister to Miconieae. Merianieae have large dorsal connective spurs and elongate cuneate seeds, and these merianoid characters should now be compared to stamens and seeds of Miconieae for clues of derivation from shared ancestral morphologies. The sister-group relationship of the predominantly Andean Merianieae and Miconieae is also seen in a nuclear internal transcribed spacer phylogeny (Clausing, 1999).

Among the groups that agree with earlier morphological hypotheses are the robust clade formed by Microlicieae and Melastomeae (including Rhexia). This sister-group relationship is supported by a stamen character, namely basally prolonged connectives (Fig. 3c) that serve as a hinge between pollen sacs and filament and that appear uniquely shared by these two groups (Renner, 1993). Connectives in Melastomataceae, however, are in need of reinvestigation. For example, it is unclear whether there are anatomical or ontogenetic differences between the independently derived basal-ventrally prolonged connectives in Dissochaeteae (Macrolenes, Dissochaeta) and the ones of Microlicieae and Melastomeae. These hinges between pollen sacs and filaments, termed pedoconnectives by Jacques-Félix (1953, 1981, 1994), increase the flexibility of anther positioning during anthesis, facilitate the bees' hold on the androecium during vibration, and standardize bee position to ensure stigma contact. Pedoconnectives and their often differentially colored appendages also function to enhance the flowers' visual display (Renner, 1989b; Larson and Barrett, 1999).

Microlicieae sequenced so far share a 66-bp deletion in their ndhF sequences. Morphologically, they comprise a cohesive assemblage of genera centered in south-central Brazilian savannas. Potentially synapomorphic are their straight or slightly winged seeds with a foveolate surface (SEMs: Whiffin and Tomb, 1972; Renner, 1990b). Their sister group, Melastomeae sensu lato (i.e., including Tibouchineae, Osbeckieae, and now also Rhexia), has two morphological synapomorphies, cochleate seeds and ovaries crowned by persistent trichomes. Cochleate seeds contain curved (campylotropous) embryos, the likely adaptive advantage being that campylotropous seeds contain embryos twice as long as the seed itself, giving better opportunities for early seedling establishment (Bouman and Boesewinkel, 1991). The significance of ovary apex hairs or scales has not been studied, but such emergences may afford protection against insects that oviposit into developing ovaries. These characters are lacking only in a few odd species in the 45–47 genera. Thus, Rhexia has glabrous ovaries and lacks ventral connective appendages and so do a few other Melastomeae, which, however, all have the typical cochleate seeds. Because of these and other unusual traits, such as occasionally atropous ovules (Etheridge and Herr, 1968) and mature unilocular anthers, Rhexia had been assigned tribal rank, either together with Monochaetum and Pachyloma (Cogniaux, 1891) or by itself (Renner, 1993). Molecular data now solidly place Rhexia and Monochaetum in Melastomeae (Pachyloma has not been sampled) and indicate that Rhexia is sister to the Central American Arthrostemma. Arthrostemma and Rhexia share a strongly costate-tuberculate seed testa (Whiffin and Tomb, 1972), four-merous flowers (also found elsewhere in Melastomeae), and hypanthia with sparse glandular pubescence. Arthrostemma comprises seven species in Central America, while Rhexia consists of 11 species in North America and is the only genus of Melastomataceae endemic in the northern hemisphere.

Another genus placed in Melastomeae by the molecular data, but with ovoid rather than cochleate seeds, is Centradenia. Centradenia was treated in Microlicieae by Cogniaux (1891) and in Bertolonieae [sub Sonerileae sensu lato] by Almeda (1977, 1997a) and following him Renner (1993), but it now appears that its seed morphology represents a secondary modification.

Unexpectedly, the African and Asian Melastomeae (Dichaetanthera, Dissotis, Melastoma, and Osbeckia; Fig. 2) form a clade that is robustly nested within neotropical Melastomeae. A relatively recent derivation of Old World Melastomeae from New World Melastomeae, ∼15–12 million years ago judging from molecular clock-based estimates based on a dense sample of ndhF sequences from New World and Old World Melastomaeae (Renner and Meyer, in press), agrees with Almeda's (1997b) suggestion that paleotropical Melastomeae retain the same base chromosome number found in many neotropical Melastomeae. However, there is much intrageneric polyploidy and dysploidy. The African-Asian clade also has not yet acquired obvious morphological synapomorphies.

Within Melastomeae, the deepest splits are between Arthrostemma + Rhexia, Nepsera + Aciotis, and the remaining genera (Figs. 1 and 2). Nepsera and Aciotis both prefer wet habitats, have four-merous flowers, acute white petals, and much-branched fragile inflorescences, traits in which they differ from most Melastomeae, which usually have five-merous flowers, purple petals, and more robust inflorescences than those of Nepsera and Aciotis. However, denser sampling of neotropical Melastomeae is needed to break up the long branch currently leading to these two genera.

Perspectives

This first molecular phylogenetic assessment of Melastomataceae and Memecylaceae shows that leaf venation, stamen anatomy and morphology, and seed shape and size underwent major transformations early during the families' history, while fruit fleshiness and mode of dehiscence were modified frequently and more recently. The ancestor of Melastomataceae likely had capsular fruits with numerous small seeds, this being the condition in families most closely related to Melastomataceae + Memecylaceae, and in basalmost Melastomataceae (Pternandra, Astronia). Within Melastomataceae, berries appear to have evolved from capsules minimally four times, namely in Miconieae, Blakeeae, within Dissochaeteae, and within Melastoma (Meyer, 2000; Clausing, Meyer, and Renner, in press). The molecular trees also imply that dorsally massive connectives are ancestral in Memecylaceae and Melastomataceae, characterizing these families and their closest relatives. Future work will have to test the explicit hypotheses that (1) melastome stamen evolution went from dorsally enlarged connectives to basal-ventrally enlarged ones; (2) loss of an endothecium preceded poricidal dehiscence (perhaps as a preadaptation); (3) herbaceous, capsular-fruited Sonerileae/Oxysporeae evolved from woody Dissochaeta-like plants with fleshy fruits; and (4) African and Asian Melastomeae are derived from neotropical Melastomeae. Biogeogeographic implications of a larger ndhF phylogeny for the family, especially with regard to the apparently recent diversification of African and Madagascan melastomes as judged by fossil-calibrated genetic distances, are considered elsewhere (Renner, Clausing, and Meyer, in press).