Subtribal Relationships in Tribe Tradescantieae (Commelinaceae)Based on Molecular and Morphological Data

Tribe Tradescantieae (Commelinaceae) consists of seven subtribes and 25 genera. Previous attempts to evaluate phylogenetic relationships within the group using morphology or the chloroplast-encoded rbcL have either been highly homoplasious (morphology) or provided only weak support for subtribal relationships due to insufficient variability (rbcL). In this study, phylogenetic analysis of nucleotide sequence data from the chloroplast-encoded ndhF and rbcL genes, as well as 47 morphological and anatomical characters, were used to evaluate relationships within and among the subtribes of Tradescantieae. The addition of ndhF resulted in a more highly resolved phylogeny and greater bootstrap and decay values than were obtained by rbcL alone or rbcL and morphology. The analyses suggest the following: (I) subtribes Coleotrypinae, Cyanotinae, and Tradescantiinae (with the addition of Elasis) are monophyletic; (2) subtribe Thyrsantheminae is polyphyletic; and (3) subtribe Dichorisandrinae is polyphyletic. Members of Dichorisandrinae are united into two clades (Dichorisandra and Siderasis; Cochliostema, Geogenanthus, and Plowmanianthus) whose relationships are more clearly resolved. The position of Old World subtribes Cyanotinae and Coleotrypinae, nested within New World taxa suggested by rbcL studies, are supported by the addition of ndhF data.


INTRODUCTION
Tribe Tradescantieae (Meisn.) Faden & D. R. Hunt is the most diverse group within subfamily Commelinoideae (Faden and Hunt 1991). Meisner (1842) defined the tribe based on the presence of six fertile stamens. Clarke ( 1881) used both stamina! characters and fruit type to separate Tradescantieae from tribes Commelineae and Pollieae. Woodson (1942) and Rohweder (1956) each emphasized inflorescence characters. Brenan ( 1966) used several characters to divide the whole family into 15 informal groups and this classification was followed until Faden and Hunt (1991). Faden and Hunt (1991) and Faden (1998) employed a broad array of morphological and anatomical characters to divide the tribe into seven subtribes, containing 26 genera and approximately 285 species.
Circumscription of tribe Tradescantieae has varied greatly due to high amounts of homoplasy in morphological characters (Evans et al. 2000b ). The tribe is naturally split into Old World and New World components, with Cyanotinae, Coleotrypinae, Palisotinae, and Streptoliriinae restricted to the Old World, and Dichorisandrinae, Thyrsantheminae, and Tradescantiinae to the New World (Faden and Hunt 1991;Evans et al. 2000aEvans et al. , 2003. Evans (1995) and Evans et al. (2000a) conducted a cladistic analysis of morphological characters in Commelinaceae, and the results were largely incongruent with Faden and Hunt's classification, presumably due to a high degree of homoplasy in the data. Evans et al. (2003)  phylogenetic analysis using the chloroplast-encoded gene rbcL as well as a combined molecular/morphological data set. Both molecular and combined analyses produced phylogenies that were largely congruent with Faden and Hunt's classification and incongruent with the morphological phylogeny. The phylogenies are in disagreement with Faden and Hunt's classification in that: (1) Palisota Rchb. ex Endl. is basal to both Tradescantieae and Commelineae (making Tradescantieae paraphyletic); (2) Thyrsantheminae are polyphyletic; and (3) the monophyly of Dichorisandrinae is in question, as it is weakly supported by the combined rbcL/morphology analysis, but not supported by the rbcL data alone. Hardy (2001), with a more detailed study of morphological and molecular characters in Dichorisandrinae provided support for a monophyletic subtribe. Finally, the DNA data exhibited less homoplasy than the morphological data.
Relationships among subtribes of Tradescantieae were only weakly supported in the molecular analysis of Evans et al. (2003) as evidenced by low bootstrap and decay values. Thus, there was a need to perform an analysis using another gene to aid in providing a well-supported phylogeny for members of Tradescantieae. Givnish and Sytsma (1997) demonstrated that including a higher number of variable or informative characters in an analysis increased the chances of obtaining the correct phylogeny. The chloroplast-encoded gene ndhF was chosen for this study because: (1) ndhF is 1.5 times longer than rbcL (Olmstead and Palmer 1994;Kim and Jansen 1995); (2) ndhF has a relatively high substitution rate (approximately twice that of rbcL) (Olmstead and Palmer 1994;Kim and Jansen 1995); and (3) ndhF has been known to provide informative characters in several families Horticulture-Missouri Botanical Gardens. n.
pies used in this study were from the same DNA samples used in Evans et al. (2003). The objectives of this study were to determine phylogenetic relationships among members of tribe Tradescantieae using ndhF and rbcL sequence data and to use the resulting phylogenies to evaluate systematic and biogeographical trends within tribe Tradescantieae.

MATERIALS AND METHODS
The gene ndhF was sequenced from a single plant of 19 species, representing 19 genera from Tradescantieae (Table  1). Additionally, a single plant from one species each of Cartonema and Aneilema were included for outgroup comparison, based upon results of Evans et al. (2003). All sam-Total DNA for all species was extracted from frozen leaf tissue following the CTAB procedure of Doyle and Doyle (1987) as modified by Smith et al. (1991). The ndhF gene was amplified in two fragments on a Hybaid thermocycler (Thermo Electron Corporation, Marietta, Ohio, USA), using deoxynucleotides from United States Biochemical (Cleveland, Ohio, USA), and Taq polymerase from Promega (Madison, Wisconsin, USA). Primers for the 5'-region annealed near positions 32 (forward) and 1318 (reverse) of ndhF (Terry et al. 1997). For amplification of the 3' -region, primers that annealed near position 972 (forward) and 2110 (reverse) were used (Olmstead and Sweere 1994 ). Sequencing reactions were performed using BigDye® Terminator Reaction Mix (Applied Biosystems, Inc., Foster City, California, USA) or the DYEnamic ET terminator cycle sequencing kit (Amersham Biosciences Corporation, Piscatawy, New Jersey, USA). Cycle sequencing fragments were purified using Centri-Sep columns (Princeton Separations, Inc., Adelphia, New Jersey, USA) and sequenced on an ABI 310 automated sequencer before being assembled using Autoassembler vers. 2.0 (ABI Prism®). All sequences were manually aligned before phylogenetic analysis. The resulting ndhF data set was combined with rbcL data from Evans et al. (2003) and 47 morphological characters from Evans et al. (2000a).

Phylogenetic Analyses
All phylogenetic analyses were performed using PAUP* vers. 4.0b4a (Swofford 2003). A multiple-islands approach was used to find the most parsimonious trees (Maddison 1991). A heuristic search was conducted using a random addition sequence with 1000 replicates, tree-bisection-reconnection (TBR) branch swapping, steepest descent on, and 100 trees saved for each replicate. Bootstrap and decay values were determined to evaluate support for each node. For the bootstrap analysis, one hundred replicate searches were performed using TBR with random addition of 100 replicates and 100 trees saved from each replicate. Decay values were determined using AutoDecay vers. 2.9.9 (Eriksson 1997) to produce a constraint command file. This file was executed in PAUP* using a heuristic search, TBR branch swapping, and 10 replications of the random addition sequence. The "Converse Enforce" command in PAUP* was employed to save only those trees lacking the clade being examined.

Character State Mapping
To examine biogeographical trends within the tribe, geographic distributions were overlaid onto the total data phylogeny using MacClade vers. 4.0 (Maddison and Maddison 2000) assuming accelerated transformation (ACCTRANS).

RESULTS
One most parsimonious tree of 1392 steps was produced from the combined ndhF/rbcL data set; consistency index (CI) = 0.69, retention index (RI) = 0.61 without autapomorphies (Fig. 1 ). The phylogeny was largely congruent with the rbcL phylogeny (see Evans et al. 2003), though the support for the deeper clades was notably higher in the combined analysis (Fig. 1). The shallow branches were well supported, with the exception of the clade containing Callisia, Tripogandra, and Elasis (62% bootstrap). The deeper branches were also relatively well supported, though two of the deeper branches were supported by bootstrap values of less than 70% (Fig. 1).
When morphological data were added to the rbcL!ndhF data, two most-parsimonious trees of 1540 steps were found; CI = 0.66, RI = 0.58 without autapomorphies (Fig. 2). One tree was identical to the rbcL!ndhF phylogeny and the other differed in the position of Elasis.
Of the seven subtribes within Tradescantieae, Coleotrypinae, and Cyanotinae were monophyletic, Tradescantiinae were paraphyletic (due to the inclusion of Elasis, a member of Thyrsantheminae, in the clade), and Thyrsantheminae and Dichorisandrinae were polyphyletic. Members of Dichorisandrinae were placed into two clades: a Dichorisandra and a Siderasis clade, and a Cochliostema/Plowmanianthus/Geogenanthus clade. Subtribe Palisotinae is comprised of a single genus, Palisota, and subtribe Streptoliriinae (three genera) was represented by a single genus, Spatholirion.

DISCUSSION
The rbcL and combined rbcL/morphology data sets placed Palisota as sister to all genera of Commelinaceae except Cartonema, making tribe Tradescantieae paraphyletic (Evans et al. 2003). Deep branches in the rbcL and rbcL!morphology phylogenies were only weakly supported, however, as determined by bootstrap and decay values, and basal relationships within the tribe could not be inferred with confidence.
Addition of ndhF produced a monophyletic Tradescantieae (both the ndhF/rbcL and ndhF!rbcL!morphology data sets), with Palisota sister to the rest of the tribe (Fig. 1, 2). While support for most clades in the total data phylogeny was high, the branch uniting Palisota with the remainder of Tradescantieae is supported by a decay value of only 1 (or 2 when morphology is included), and a bootstrap value of 56% (less than 50% when morphology is included) (Fig. 1,  2). Additionally, only a single representative of tribe Commelineae, Aneilema, was included in this study. Until additional representatives of Commelineae are examined, as well as sequences from additional rapidly evolving regions, the exact placement of Palisota, and thus the monophyly of tribe Tradescantieae, will remain unclear.
Members of subtribe Dichorisandrinae are found in two separate clades (Fig. 1, 2). Analysis of morphological data produced a highly polyphyletic Dichorisandrinae, but a high degree of homoplasy among specific morphological characters makes those relationships suspect (Evans et al. 2000a). The combined rbcL!morphology data yielded a monophyletic Dichorisandrinae, albeit with low bootstrap and decay support (Evans et al. 2003). Hardy (2001) examined morphological and molecular data to evaluate relationships within Dichorisandrinae and found support for a monophyletic subtribe. That study, while providing a thorough sampling within Dichorisandrinae, did not include many representatives from other subtribes of Tradescantieae. Additionally, the jackknife value (Farris 1997) was relatively low for the branch supporting the monophyly of the subtribe.
Nearly every analysis to date (except for morphology alone) places the five genera of Dichorisandrinae into two well-supported clades with Dichorisandra and Siderasis in one and Cochliostema, Plowmanianthus (represented as "undescribed genus" in Evans et al. [2000aEvans et al. [ , b, 2003), and Geogenanthus in the other. While this analysis places these clades separate from each other, a tree of only one additional step is required to obtain a monophyletic subtribe. All members of the subtribe share a similar karyotype of 19 large chromosomes (Jones and Jopling 1972;Faden and Hunt 1991;Faden 1998), but no unique morphological characters are known that unambiguously unite these five genera. Inclusion of sequences from additional rapidly evolving regions of the genome, as well as more thorough sampling of taxa within Dichorisandrinae will likely be needed to confidently determine the monophyly of the subtribe. Subtribe Thyrsantheminae is polyphyletic, with representatives appearing in three different clades (Fig. 1, 2). The rbcL!morphology data united Weldenia and Thyrsanthemum, placed Elasis in a clade with members of subtribe Tradescantiinae, but failed to resolve the position of Tinantia (Evans et al. 2003). Addition of ndhF has yielded the same set of relationships, but with stronger support for each clade. Additionally, the placement of Tinantia has been resolved as sister to a clade containing the remainder of Thyrsantheminae and tribe Tradescantieae, again with strong support (Fig.  1, 2).
There is clearly a relationship between Elasis (subtribe Thyrsantheminae) and members of subtribe Tradescantiinae. Molecular data alone (rbcL!ndhF) place Elasis well within the Tradescantiinae clade (Fig. 1). With the addition of mor-phological data, the position of Elasis with respect to Tradescantiinae becomes unresolved, with Elasis being placed either within or sister to the subtribe (Fig. 2). All members of subtribe Tradescantiinae share a common inflorescence type, in which two cincinni are fused back-to-back or in which two-to-several stipitate cincinni form a pseudoumbel (Faden and Hunt 1991). Evans et al. (2003) hypothesized that Elasis, which lacks this inflorescence type, may represent a reduced form in which one of the two cincinni has been lost. Alternatively, if Elasis is resolved as sister to Tradescantiinae, then fused cincinni within the subtribe may represent the derived condition with respect to Elasis. As there are currently no morphological characters known that clearly unite Elasis with members of Tradescantiinae, examination of inflorescence structure and development might shed light upon this unresolved node of the phylogeny. branch indicate the number of additional steps required before that branch collapses (decay value). Gray line represents branch that collapses in strict consensus of the two most-parsimonious trees. Arrow indicates the position of branches in the second most-parsimonious tree. Subtribal and tribal affinities are indicated with the bars to the right of the cladogram.
are each monophyletic and together form a monophyletic Old World clade (Fig. 1, 2). The monophyly of each of these two subtribes is strongly supported by both molecular and morphological data. The inflorescence of members of Coleotrypinae consists of axillary, highly congested cincinni and perforates the leaf sheath. The Cyanotinae are united by the seeds with a terminal embryotega. As noted in Evans et al. (2003), biogeography provides some evidence of relationship between these two subtribes, but no morphological characters are known that unambiguously unite them. The addition of ndhF data has helped to clarify the biogeographical relationships of Cyanotinae and Coleotrypinae to other Tradescantieae subtribes. Of the seven subtribes within Tradescantieae, three (Dichorisandrinae, Thyrsantheminae, and Tradescantiinae) are found exclusively in the New World and four (Coleotrypinae, Cyanotinae, Palisotinae, and Streptoliriinae) are found exclusively in the Old World (Faden and Hunt 1991;Hunt 1993Hunt , 1994Faden 1998) (Fig. 3). The placement of Coleotrypinae and Cyanotinae within, but not sister to, the New World clade was noted by Evans et al. (2003) in their rbcL!morphology analysis. Three possible scenarios were proposed to explain the distribution: (1) a single shift from the Old World to the New World, either through vicariance or dispersal, followed by a single dispersal back to the Old World (ACCTRANS optimization); (2) two independent introductions to the New World (DELTRANS optimization); or (3) the Boreotropical Flora Hypothesis (Wolfe 1975), in which the current distribution reflects a relictual distribution of a formerly widespread northern temperate group. The third scenario was determined to be unlikely due to the relatively early divergence of Dichorisandrinae and the relatively derived position of Tradescantiinae. The first two scenarios, however, were equally likely as a result of the ambiguous optimization of biogeography onto the rbcL!morphology phylogeny (Fig. 6 of Evans et al. 2003). The addition of ndhF to the analysis clarifies the issue by placing the Old World subtribes Coleotrypinae and Cyanotinae well within the New World clade, thus removing ambiguity to the optimization of biogeography and favoring the first hypothesis (one shift from the Old World to the New World followed by dispersal back to the Old World; Fig 3).