Taxa comprising subtribes Tradescantiinae and Thyrsantheminae of tribe Tradescantieae in the monocot family Commelinaceae (dayflower family, Faden and Hunt 1991) are known by a variety of amusing common names, including bridal veil, widow's tears, spiderwort, snotweed, grass violet, wandering Jew, and Moses-in-a-basket. Henceforth referred to as the Tradescantia alliance, these eleven New World genera (Tradescantia, Gibasis, Callisia, Tripogandra, Elasis, Tinantia, Thyrsanthemum, Weldenia, Gibasoides, Matudanthus, Sauvallea) maintain variable levels of genome change, including polyploidy, aneuploidy, hybridization, and genomic rearrangements (Jones and Jopling 1972, Jones and Kenton 1984), and exhibit considerable ecological diversity (Faden 1998). Of the 650 species in Commelinaceae, the Tradescantia alliance contains ca. 144; genera range from large (Tradescantia, ca. 70 species) to monotypic (Elasis, Weldenia, Gibasoides, Matudanthus, Sauvallea; Faden 1998).

Several factors contribute to a complicated history of classification in Commelinaceae, some of which are particularly problematic in the Tradescantia alliance. First, petals in Commelinaceae are short-lived and deliquescent; herbarium specimens rarely preserve floral characteristics relevant to some classification attempts (Fig. 1; Woodson 1942). Second, morphological characters in Commelinaceae are homoplasious (Evans et al. 2000a); both circumscription of the Tradescantia alliance and relationships among genera using a cladistic analysis of morphology were incongruent with previous classification schemes (Evans et al. 2000b). Collection of anatomical data confirmed the presence of convergent evolution for some diagnostic traits (Tomlinson 1966). Third, interspecific hybridization may have played a role in the evolution of the group. Historical hybridization in some Tradescantia species may have led to speciation by chromosomal differentiation (Jones 1990), and ongoing gene flow continues between closely related species, such as the erect Tradescantia (Sect. Tradescantia, series Virginianae, Anderson 1936). While much progress has been made in circumscribing subtribes and genera, clarifying relationships between members of the Tradescantia alliance would allow for further explorations into the evolution of ecological and genomic traits.

The most recent Commelinaceae classification (Faden 1998) effectively resolves the shuffling of taxa between groups from several previous classification schemes (Faden and Hunt 1991). In the most recent treatment (Faden 1998), tribe Tradescantieae comprises 25 genera, 285 species, and is divided into seven subtribes: three from the Old World and four from the New World. This system places Gibasis, Tradescantia, Callisia and Tripogandra in subtribe Tradescantiinae; Thyrsanthemum, Gibasoides, Tinantia, Elasis, Matudanthus, and Weldenia into subtribe Thyrsantheminae. Sauvallea is an enigmatic monotypic genus from Cuba with uncertain placement in either of the two subfamilies (Faden and Hunt 1991).

Current generic circumscriptions for subtribes Tradescantiinae and Thyrsantheminae are the result of gradual dismemberment and then restructuring of groups. In his description of Mexican Commelinaceae, Hunt (1993) favored the inclusion of several minor genera into larger, broader genera of Tradescantiinae: Tradescantia (including Campelia, Cymbispatha, Rhoeo, Separotheca, Setcreasea, Zebrina), Gibasis (including Aneilema sensu Matuda, in part), Callisia (including Aploleia, Cuthbertia, Hadrodemas, Leptorrhoeo, Phyodina, Spironema) and Tripogandra (including Neodonellia). Hunt's (1980) treatment of Tradescantia initially identified eight sections, with four more subsequently added (Appendix 1, Hunt 1986a). Gibasis is classified into two sections (Heterobasis and Gibasis) using a suite of characters including chromosome morphology (Hunt 1985). Hunt (1986b) favored dividing Callisia into five sections rather than splitting into many genera of few species each.

Fig.1.

Floral morphological diversity in the Tradescantia alliance. Selected exemplars represent characteristic features of each genus. Floral morphology: A. Gibasis. B. Tripogandra. C. Tinantia. D. Tradescantia. Inflorescence morphology: E. Gibasis. F. Tradescantia. Plates A–D taken by Travis Columbus.

Fig.2.

Previous hypothesis for phylogenetic relationships in tribe Tradescantieae. Modified from Wade et al. (2006), inferred from one taxon per genus from morphological and molecular data. Numbers by nodes represent bootstrap support.

The first molecular phylogeny of the family suggested that tribe Tradescantieae is monophyletic with the exception of Palisota (Evans et al. 2003). As sampling was limited to one to few species per genus, however, further exploration of the relationships among genera is needed. A more recent phylogeny including comprehensive sampling of genera in tribe Tradescantieae used morphological and molecular data, and is the basis for sampling in the present study (Wade et al. 2006). It revealed a more derived New World clade composed of Tradescantia, Gibasis, Callisia, Tripogandra, Elasis, Tinantia, Thyrsanthemum, and Weldenia (Fig. 2). A combined analysis of a cpDNA locus (trnL-trnF) and a multiple copy nuclear locus (5S NTS) presents Tradescantia and Gibasis as both monophyletic, with Callisia as paraphyletic (Burns et al. 2011). Another phylogenetic study focused sampling on Callisia; two pastid loci nested Tripogandra inside Callisia with Tradescantia sister to that clade (Bergamo 2003).

Inflorescence structure is one of the most important characters for taxonomic classification in the Tradescantia alliance, especially to distinguish subtribes and genera (Table 1), although botanists historically disagreed on interpretation of relevant structures (Brenan 1966). The basic unit of inflorescence in Commelinaceae is a scorpioid cyme, or cincinnus. Inflorescences of taxa belonging to subtribe Tradescantiinae are characterized by a pair of such cymes fused back-to-back. The exception to this pattern is Gibasis, in which cymes are not fused but may be grouped in pairs or umbelliform clusters (Hunt 1985). Cymes from inflorescences representing subtribe Thyrsantheminae are never fused and appear thyrsiform or as a single cincinni (Faden and Hunt 1991). The hypothesized trend in inflorescence evolution is towards reduced parts and axes (Brenan 1966). For example, Gibasis cyme bracts are reduced to the point of appearing absent while Tradescantia possesses large, spathaceous bracts (Faden 1998).

Variation in inflorescence structure in the Tradescantia alliance is matched by switches in breeding system between self compatible (SC) and self incompatible (SI). Owens (1981) comprehensively surveyed breeding system in Commelinaceae and noted that genera belonging to tribe Tradescantieae are predominantly SI, presumably to control outbreeding. The exception is Tripogandra, in which only one of six sampled species was SI. Four additional genera (Callisia, Dichorisandra, Gibasis, Tradescantia) contained both SI and SC species. A partial breakdown in SI appeared to be occurring in several species of Tradescantia, and additional intraspecific variation in breeding system in the alliance was attributed to different cytotypes (Owens 1981). Variation in both inflorescence structure and self incompatibility allows for examination of the relevance of these life history traits in the evolution of the Tradescantia alliance.

While it is clear which genera belong in the Tradescantia alliance, relationships among these genera remain confusing. The questions addressed by this research are twofold. First, are subtribes and genera monophyletic? The current classifications of family Commelinaceae (Faden 1998; Faden and Hunt 1991) and treatments for Tradescantia (Hunt 1980), Gibasis (Hunt 1985), Tripogandra (Handlos 1975) and Callisia (Hunt 1986b) serve as hypotheses of generic composition. Second, how does a molecular phylogeny inform evolution of inflorescence structures relevant to taxonomy in the Tradescantia alliance? We expect that selfing species will possess condensed inflorescences (Goodwillie et al. 2010). In this study, we examine these issues by inferring a molecular phylogeny from two plastid loci, trnL-trnF and rpL16, for 85 taxa and investigating trait evolution. Given the complicated nature of evolution and hypothesized hybridization in the Tradescantia alliance, a phylogeny estimated from plastid loci can provide a simplified version of just matrilineal relationships.

Table 1.

Morphological characteristics of taxonomic groups in the Tradescantia alliance following Faden (1998). An asterisk (^*) indicates genera not included in the current study, and a carat () indicates monotypic genera.

Materials and Methods

Taxon Selection—Sampling in our study includes 85 taxa obtained from field collections, botanical gardens, commercial sources, and research collections, as well as sequences previously published in GenBank (Appendix 1). When possible, living specimens were maintained in greenhouses at the University of Missouri for DNA extraction. Herbarium specimens have been deposited in the University of Missouri Dunn-Palmer Herbarium (UMO). The ingroup includes 70 taxa from eight genera, including 30 Tradescantia (ca. 70 species total in genus), nine Gibasis (11 species), 15 Callisia (ca. 20 spp.), five Tripogandra (ca. 22 spp.), one Thyrsanthemum (3 spp.), six Tinantia (14 spp.) and monotypic Elasis and Weldenia. Obtaining monotypic genera Sauvallea, Gibasoides, and Matudanthus was not possible for this study, and we also lack sampling for a handful of sections in Tradescantia and Callisia. The outgroup is represented by 17 species from other subtribes in tribe Tradescantieae as well as tribe Commelineae (Faden and Hunt 1991).

Molecular Methods—DNA extraction necessitated a 3 × -6 × CTAB method (Smith et al. 1991) from fresh or frozen leaf tissue. We amplified two plastid loci generally following PCR parameters in Shaw et al. (2005) with minor alterations in MgCl₂ concentrations for recalcitrant taxa. Conserved primers (F71, R1516, Shaw et al. 2005) amplified the rpL16 intron and two additional internal primers assisted in sequencing (rpL16F692 ATGGAGAAGCTGTGGGAACGA, rpL16R690 CGTTCCCA CAGCTTCTCCATTA). Conserved primers TabC and TabF amplified the trnL intron/trnL-trnF intergenic spacer with additional sequencing via internal primers TabD and TabE (Taberlet et al. 1991). The University of Missouri's DNA Core directly sequenced purified products.

Sequence Alignment and Phylogenetic Analysis—We edited resulting sequences using the Lasergene Core Suite (DNASTAR, Madison, Wisconsin) with manual curation and aligned each locus using MUSCLE (Edgar 2004a; Edgar 2004b). We constructed all phylogenetic inferences using RAxML v7.2.8 (Stamatakis 2006) implemented on-line in RAxML BlackBox (Stamatakis et al. 2008). We partitioned the analysis into two loci (rpL16 and trnL-trnF) and implemented a GTR + GAMMA model of molecular evolution for each partition. We assigned members of tribe Commelineae (Commelina, Pollia, Aneilema, Murdannia) to the outgroup following Faden and Hunt's (1991) classification system. We used several methods to evaluate confidence intervals and explore alternative hypotheses in our resulting phylogeny. First, we obtained 100 bootstrap replicates in RAxML. Second, we conducted constraint tests to evaluate support for monophyly of subtribes (Tradescantiinae: Tradescantia, Gibasis, Callisia, Tripogandra; Thyrsantheminae: Elasis, Thyrsanthemum, Tinantia) and individual genera (Tradescantia, Gibasis, Callisia, Tripogandra, Tinantia). Constraint trees were inferred using the same parameters as the unconstrained trees. We compared constraint trees using several topology-based tests implemented in CONSEL (Shimodaira and Hasegawa 2001). Sequences were deposited in GenBank (accession numbers in Appendix 1), and alignments and trees were submitted to TreeBASE (study number 12595).

Trait Evolution—We assembled a dataset of two traits from literature and greenhouse observations (Appendix 1). Although sampling in the outgroup was sparse, exclusion of these taxa did not substantially alter results (data not shown). Assignment of species as self incompatible (SI) or compatible (SC) largely followed Owens (1981) and Burns Moriuchi (2006). When SI and SC were reported for the same taxon, we scored the species according to the most common occurrence, or in a few cases, as ambiguous/missing data. Assignation of inflorescence structure state was complicated by three issues. First, interpretation of inflorescence structures in the Tradescantia alliance varies widely among taxonomists (Brenan 1966). Second, terminology for inflorescence structure is inconsistent in historical literature (Endress 2010). Finally, we lacked developmental data for all taxa. As a result, we could only assign a general qualitative description for inflorescence condensation. Condensed inflorescences include single cymes, sessile inflorescences, and those subtended by spathaceous bracts (which in the family are characteristically compressed). Uncondensed inflorescences included pairs or groups of expanded cymes (or cyme pairs) attached to a common rachis, which are thyrses, although sometimes called panicles in the literature.

We conducted trait analyses on inflorescence condensation and breeding system using Mesquite version 2.75 ( www.mesquiteproject.org, Maddison and Maddison 2011). We calculated ancestral states for both traits using an Mk1 model (Xiang and Thomas 2008), and evaluated the relationship between inflorescence condensation and breeding system using Pagel's (1994) correlation test implemented in Mesquite's correl package (Midford and Maddison 2006). Because we were testing for relationships between two binary traits for the Tradescantia alliance, outgroup species and taxa for which compatibility data were missing or ambiguous (SI/SC, Appendix 1) were removed from the dataset prior to correlation analyses (59 total taxa included for correlation analysis). We ran the analysis for 10 iterations and 1,000 simulations.

Results

Phylogenetic Inference—A description of each data partition and the combined two locus dataset is available in Table 2. The best-scoring ML tree is well supported along the backbone (Fig. 3); specific taxonomic groups are discussed below.

Subtribe Tradescantiinae—The phylogeny constraining subtribe Tradescantiinae as monophyletic possessed a higher, albeit not statistically significant, likelihood than the unconstrained tree (Table 3). Additionally, there was strong bootstrap support for the inclusion of Elasis in the Tradescantiinae (BS = 99), which suggests polyphyly of the subtribe as currently circumscribed (Fig. 3). Topology tests do not support Tradescantia as monophyletic (Table 3). Tradescantia species comprise a strongly supported clade with the inclusion of Gibasis geniculata and G. linearis (BS = 99), as well as the sister taxon G. oaxacana (BS = 98, Fig. 3). There is little reinforcement for taxonomic classification within Tradescantia, as only weak bootstrap support exists for most internal nodes in the clade. No currently named sections emerge as monophyletic (Fig. 3).

As two species of Gibasis are nested within Tradescantia, and a third species is sister to Tradescantia, there is no support for this genus as monophyletic (Fig. 3). Topology tests reinforce this interpretation, as the constraint tree with Gibasis monophyletic is significantly less likely than the unconstrained tree (Table 3). The exception is the SH test (p = 0.79), but this test is known to have a relatively high error rate in some cases (Goldman et al. 2000). With the exception of the three species mentioned in association with Tradescantia, Gibasis forms a strongly supported monophyletic clade (BS = 97) and is sister to the monotypic genus Elasis (BS = 91). This Gibasis + Elasis clade is sister to the Tradescantia clade. The Gibasis taxa grouping together are all from sect. Gibasis; the only member of this section not in the clade is G. linearis. The other two Gibasis species, G. geniculata and G. oaxacana, comprise sect. Heterobasis.

Tripogandra is a strongly supported clade with the inclusion of Callisia gracilis (BS = 99), although topology tests do not reject the monophyly of Tripogandra (Table 3). This clade is nested within a strongly supported Tripogandra + Callisia clade (BS = 98) that is sister to Gibasis + Tradescantia (Fig. 3). Callisia is not supported as monophyletic by topology tests (Table 3). There is substantial substructure within Callisia, including support for several taxonomic sections. Section Cuthbertia (BS = 100) and sect. Brachyphylla (BS = 100, including previously unplaced C. hintoniorum) are sister to each other (BS = 100) as the earliest diverging Callisia lineage. Three taxa of sect. Leptocallisia are monophyletic (BS = 99) and next to diverge (BS = 98); the other sampled member of the section is the previously mentioned C. gracilis. Remaining Callisia species are represented by two strongly supported clades: first, the aforementioned Tripogandra + C. gracilis, and second, C. warszewicziana (sect. Hadrodemas), sister to sect. Callisia (BS = 100). Hunt (1986b) described three informal but well-marked “groups” within section Callisia which our analysis supports collectively as monophyletic (BS = 91).

Table 2.

Characteristics of the two locus plastid dataset.

Fig.3.

cpDNA ML phylogram of the Tradescantia alliance from trnL-trn-F and rpL16. Numbers by nodes represent bootstrap support (BS, 100 replicates). An asterisk (^*) indicates BS = 100. Relevant taxonomic groups are labeled. Taxa labeled with a caret () are displaced from their current taxonomically assigned clade. Tinantia alone is confirmed as monophyletic; Callisia, Gibasis and Tradescantia are polyphyletic. Tripogandra is nested within Callisia.

Table 3.

Constraint tests for monophyly of taxonomic groups. Asterisks (^*) indicate constrained trees that were significantly different from the best (unconstrained) tree. Tree likelihoods and significance scores reported from CONSEL (Shimodaira and Hasegawa 2001). p values are indicated for each of the following topological hypothesis tests: AU = Approximately Unbiased (Shimodaira 2002), KH = Kishino-Hasegawa (Kishino and Hasegawa 1989), SH = Shimodaira-Hasegawa (Shimodaira and Hasegawa 1999), WKH = weighted KH, WSH = weighted SH.

Subtribe Thyrsantheminae—Relationships among genera of subtribe Thyrsantheminae have moderate support along the tree's backbone (Fig. 3). With Elasis sister to Gibasis sect. Gibasis as previously mentioned, Weldenia + Thyrsanthemum are sister to subtribe Tradescantiinae. Constraint tests for the subtribe are consistent with paraphyly (Table 3). The largest genus in subtribe Thyrsantheminae, Tinantia, is supported as monophyletic (Fig. 3, BS = 85), and a constrained tree was not significantly different from the unconstrained tree (Table 3). Tinantia is strongly placed as the earliest diverging lineage of the Tradescantia alliance.

Trait Analysis—The trait matrix contained no missing data for inflorescence condensation; sampled taxa were about evenly split between condensed and incondensed states (Appendix 1, Dryad  http://dx.doi.org/10.5061/dryad.s2878). For breeding system, 45% of sampled taxa were SI and 22% were SC. Remaining taxa were ambiguous/multistate (7%) or unknown (26%). ML ancestral state reconstructions indicate that both the Tradescantia alliance and subtribe Tradescantiinae evolved from ancestors possessing uncondensed inflorescences (Fig. 4, proportional likelihood 0.9906 and 0.9974, respectively). Tinantia, Callisia + Tripogandra, and Gibasis sect. Gibasis are also derived from ancestors with uncondensed inflorescences (proportional likelihoods 0.9984, 0.9887, 0.9995) while Tradescantia is the sole genus derived from an ancestor with condensed inflorescences (proportional likelihood 0.9904). There are more shifts between breeding systems than for inflorescence condensation and greater ambiguity of ancestral state assignation (Fig. 4). We found no correlations between inflorescence structure and breeding system (p = 0.49); various methods of assigning a binary trait to taxa with missing and/or ambiguous compatibility data did not alter our results (data not shown).

Discussion

A molecular phylogeny of the Tradescantia alliance from two plastid loci resolves relationships between notoriously difficult genera. Resulting implications for circumscription of genera provide insight into interpretation of morphological characters and their lability over evolutionary time.

Limitations of Data —We would be remiss if we did not mention inherent caveats in the methods we employ here to narrate the evolutionary history of this complex group. Both loci sampled for this study are from the plant plastomes; their relatively high rates of evolution often result in complex insertion/deletion polymorphisms (indels) that cause alignment difficulties (Golubchik et al. 2007). Despite the rapidly evolving nature of the two plastid loci utilized in this study, virtually no variation was found to differentiate the erect Tradescantia. As several members of the Tradescantia alliance are hypothesized to have arisen via hybridization (Anderson 1936), nuclear data will illuminate these issues. Increased taxon collection and data sampling from the nuclear genome may resolve some of the more difficult questions in the group, including the placement of additional uncertain and as yet unsampled taxa. Finally, coding of the traits analyzed here as binary characters simplifies the variation within both traits and species, so resulting conclusions should be interpreted accordingly.

Phylogenetic Classification —The phylogenetic reconstruction from two plastid loci recapitulates the evolutionary relationships between genera posited by previous studies with more limited taxon sampling (Fig. 2). Topological constraint tests provide information about the monophyly of genera and subtribes, which as a result inform understanding of morphological characters used to define taxonomic groups (see below, Evolution of inflorescence structure). The ingroup of the Tradescantia alliance is comprised of two closely related subtribes, Tradescantiinae and Thyrsantheminae, which while strongly supported as a single clade are both paraphyletic according to current classification. The polyphyly of subtribe Thyrsantheminae confirms previous findings from phylogenies constructed from both morphological and molecular loci (Faden and Hunt 1991; Evans et al. 2000b; Evans et al. 2003).

Previous phylogenetic research indicated substantial issues with poly- and paraphyly for several groups in the Tradescantia alliance (Bergamo 2003; Burns Moriuchi 2006); additional intrageneric sampling presented here increases these concerns. None of the currently circumscribed genera in subtribe Tradescantiinae are monophyletic; we identified two clades comprising Callisia + Tripogandra and Tradescantia + Gibasis (and Elasis). Classification of taxa in Callisia has been historically difficult, resulting in dissatisfying conclusions for both systematics (Hunt 1986b) and molecular phylogeneticists (Bergamo 2003). We confirmed monophyly of most sections in Callisia and resolved relationships among them. The exception is the placement of Callisia gracilis with Tripogandra instead of sect. Leptocallisia, which Bergamo (2003) also noted. Tripogandra is a relatively clearly marked genus characterized by slightly zygomorphic flowers and dimorphic stamens (Handlos 1975). While it is still nested within Callisia, the lack of resolution within the Tripogandra clade cannot preclude the genus as monophyletic. What accounts for the difficulty in circumscribing Callisia and Tripogandra? Our findings reaffirm the conclusions of Bergamo (2003); systematic problems in Callisia appear to be the result of rapid evolution, as shown by very short branch lengths throughout the clade but prolific insertion-deletion polymorphism (data not shown, but see TreeBASE study number 12595). Unlike many species in Tradescantia (Anderson 1936) and Gibasis (Kenton 1984), Tripogandra species lack the ability to hybridize (Handlos 1975), and there is little to no evidence of hybridization in Callisia (Bergamo 2003). Rapidly changing morphological characters coupled with emergence and reinforcement of pre-zygotic isolation mechanisms may account for the lack of consistency in these genera.

Fig.4.

Ancestral state reconstruction of inflorescence condensation and breeding system in the Tradescantia alliance. Phylogeny is the two-locus cpDNA ML analysis shown in Fig. 3. Traits were assigned according to Appendix 1. For inflorescence condensation, white and black circles indicate expanded (uncondensed) and condensed inflorescences, respectively. White and black circles represent self incompatible and self compatible breeding systems, respectively.

Burns Moriuchi (2006) found Gibasis to be strongly monophyletic; however, all three species included in that analysis were from section Gibasis (species from section Heterobasis were not included). Our results from molecular data suggest Tradescantia and Gibasis intergrade substantially with each other, and additional lines of evidence support this possibility (also see below, Evolution of inflorescence structure). First, chemotaxonomic studies in Commelinaceae indicate possible relationships between Gibasis sect. Heterobasis and some Tradescantia species (Del Pero Martinez and Swain, 1985). Gibasis oaxacana and G. geniculata share the presence of phenolic and sulfate derivatives, a trait also found in some Tradescantia and Tripogandra. Second, silica cells, which superficially seem to be a distinctive and uniting taxonomic feature, occur in at least two different taxonomic groups in Commelinaceae (Tomlinson 1966). Finally, G. linearis and G. geniculata both possess Tradescantia-type pollen (tectum insulate, insulae forming cerebroid pattern; Poole and Hunt 1980), consistent with their placement in the Tradescantia clade. Gibasis oaxacana, which we place sister to Tradescantia, has an intermediate pollen type which suggests this species may serve as a link between Tradescantia and other genera (Poole and Hunt 1980). It is unlikely that these traits arose independently in the same family multiple times, but historical hybridization cannot be ruled out as a mechanism for traits to appear in seemingly disparate clades.

This is the first study to include substantial sampling from Tinantia. Horal zygomorphy and corresponding staminal characteristics make this a robustly delineated genus morphologically. Tinantia anomala was described as a monotypic genus, Commelinantia, because of morphological characters reminiscent of Commelina (Tharp 1922, 1956). Subsequent researchers, however, rejected this analysis and instead grouped it with Tinantia (e.g. Brenan 1966); our results confirm strong support for its inclusion in the genus. The remaining genera in subtribe Thrysantheminae are monotypic or only represented by one species. Of particular systematic interest are the still unsampled monotypic genera Gibasoides, Matudanthus, and Sauvallea; their inclusion in a molecular phylogeny could potentially solidify placement of the other genera and circumscription of subtribes. However, they possess distinct inflorescence variation, which could potentially complicate interpretation of evolution of such structural traits.

Evolution of Inflorescence Structure—Variation in inflorescence structure is an important driver of angiosperm evolution because of relationships with plant reproduction. Predictably, selfing species exhibit reduced allocation to flowers, since the need to attract pollinators is reduced (Goodwillie et al. 2010). Long-standing hypotheses indicate that angiosperm inflorescences evolved from highly branched displays, and suppression of various inflorescence structures results in condensation, like heads (Parkin 1914; Stebbins 1974; Wyatt 1982; Harris 1999). Pollination biology is traditionally floricentric, in which individual flowers are the focus of research (Harder et al. 2004). However, variation from pollinator movement and interactions with the entire inflorescence contribute to evolution of diverse structures (Harder et al. 2004). Moreover, inflorescence architecture can have widespread effects on pollinator interactions (Wyatt 1982). Empirical evidence suggests pollinators have little effect on inflorescence structure in Cornus (Feng et al. 2011), but inflorescence architecture is correlated with pollinator types in Arecaceae (Henderson 2002).

We did not find a correlation between breeding system and inflorescence structure in the Tradescantia alliance, but ancestral state reconstructions inform our knowledge of inflorescence evolution. The main distinction between subtribes Tradescantiinae and Thyrsantheminae is the inflorescence structure. Our results indicate that this morphological feature is labile throughout the phylogeny. The inclusion of Elasis in subtribe Tradescantiinae is strongly supported in this analysis by at least two robust nodes in the backbone of the phylogeny. As a result, the single cyme of Elasis represents a reduced form of bifacially fused cyme pairs characteristic of subtribe Tradescantiinae, confirming the hypothesis of Evans et al. (2003). Further evidence of the lability of inflorescence characters over evolutionary time comes from the Tradescantia clade. Despite condensed inflorescences being a uniting character for most Tradescantia species, T. standleyi possesses diffuse, pedunculate umbels (Standley and Steyermark 1944). The strong support for two Gibasis species in two different Tradescantia clades also indicates that the diagnostic character of a paired, condensed inflorescence structure is perhaps reversible. We support the assertion of Wade et al. (2006) that developmental evidence is required to determine the mode of inflorescence evolution in this problematic clade.

The lability in inflorescence morphology in the Tradescantia alliance may be related to variation in flowering phenology and pollination ecology among species. Evolution of condensed inflorescence structures may be an adaptation to early flowering or scarce pollinators (Feng et al. 2011); this ecological scenario is possible in the Tradescantia alliance given the deliquescent nature of most species' flowers. The only evidence for specialized pollinators in Commelinaceae comes from outgroup genera, and Callisia repens is the only known wind pollinated species in the Tradescantia alliance (Faden 1992). Some species, including erect Tradescantia (section Tradescantia, series Virginianae, Sinclair 1968) and Tripogandra serrulata (Schuster and Schuster 1971) utilize a wide variety of pollinating insects. However, despite having an expanded inflorescence and striking floral specializations, Tinantia anomala exhibits a paucity of insect pollinators and is self compatible (Simpson et al. 1986).

The failure of inflorescence condensation and pollinator syndromes to explain patterns of self compatibility in the Tradescantia alliance suggests that other floral characters may be contributing to this variation. Floral organogenesis in the Tradescantia alliance represents developmental variation found across the entire family, which may be coincident with floral morphological diversity (Hardy and Stevenson 2000). A dependence on insect pollinators and outcrossing may have influenced the arrangement of stamens and staminodes in Tripogandra (Moore 1960). Studies in Commelina reveal vertical orientation of bilateral flowers (Ushimaru and Hyodo 2005), and colored floral organs affect the frequency of pollinator visitation (Ushimaru et al. 2007). Such variation exists in the Tradescantia alliance as well. Faden (1992) notes several visual floral pollinator attractants for members of the alliance, including colored floral structures (axes, pedicels, calyces), anthers (and anther connectives), filament hairs and bearding on the androecium. Additional rewards like pollen and occasionally scent may also serve as attractants (Faden 1992).

A growing body of evidence indicates that a combination of factors, ranging from floral structures to spatial arrangement of entire inflorescences, best explains floral evolution in relation to pollinators. Floral display, a characteristic incorporating number and size of flowers, is a more accurate metric for measuring the attractant power of flowers compared to measures of individual flowers (Goodwillie et al. 2010). Even small changes in pedicel length can alter the three-dimensional arrangement of flowers, which alters pollinator behavior (Jordan and Harder 2006). Pedicel length in particular is a trait that varies among species in the erect Tradescantia group, where little to no molecular divergence exists. Seemingly uniform inflorescences can also vary in spatial and temporal arrangements of flowers; modular construction of the plant and flowering sequence are more important to the mating system than inflorescence architecture (Reuther and Claßen-Bockhoff 2010). Modularity is particularly important to the alliance, in which many species can be propagated by cuttings and rely heavily on vegetative growth. A comparison of umbels, panicles, and racemes indicates differences in bee visitation and frequency of self pollination (Jordan and Harder 2006). The combined effects of many factors, including spatial arrangement of flowers, display size and plant density, can alter pollinator behavior (Ishii et al. 2008). The Tradescantia alliance would be an interesting group in which to model the combined effects of floral/inflorescence characteristics on breeding system given the variation among species in those traits, including clonal reproduction through vegetative growth.

Our two-locus plastid phylogeny of the Tradescantia alliance indicates a complex evolutionary history for this notoriously difficult group of plants. While inflorescence condensation is not correlated with the breeding system, our ancestral state reconstructions of inflorescence structure indicate lability in the character and the possible signature of historical hybridization. Further research would benefit from incorporating karyotype and other genomic data into analyses of life history traits, as chromosomal restructuring is hypothesized to reduce the importance of reproduction in speciation (Jones and Jopling 1972).