Phylogenetic Studies in the Commelinaceae Subfamily Commelinoideae Inferred from Nuclear Ribosomal and Chloroplast DNA Sequences

The Commelinaceae are a pantropical family of monocotyledonous herbs. Previous phylogenies in Commelinaceae have emphasized sampling among genera. We extended this previous work by sampling multiple species within some of the largest genera of Commelinaceae (especially Commelina and Tradescantia , and also including Callisia , Cyanotis, Gibasis, and Murdannia ), and by sequencing noncoding regions both of the nuclear ribosomal DNA region, 5S NTS, and the chloroplast region, trnL-trnF . We generated a phylogenetic hypothesis for 68 Commelinaceae that partially tests previous morphological, taxonomic classifications. We found little evidence for conflict between nuclear and chloroplast regions for Tradescantia , Murdannia , and Callisia , and some evidence for conflict between the two regions for Commelina , though conflicting regions of the phylogeny were only weakly supported by bootstrap analyses. We found subtribe Tradescantieae to be paraphyletic, consistent with an rbcL study, though with a different topology than that produced by rbcL . In addition, subtribe Commelineae was monophyletic with strong support. We found Callisia to be polyphyletic, consistent with some previous molecular phylogenetic studies, and we found Tradescantia , Gibasis , Cyanotis , Commelina , and Murdannia , to be monophyletic. The molecular phylogenies presented here generally supported previous taxonomic classifications. Keywords—Callisia , Commelina , cpDNA , Murdannia , nrDNA , Tradescantia . 2011] BURNS ET AL.: COMMELINACEAE PHYLOGENY 269 et al. 2000 ) and molecular ( Bergamo 2003 ; Evans et al. 2003 ; Wade et al. 2006 ) data. Evans et al. (2000 ; 2003 ) sampled most of the currently recognized genera, Wade et al. (2006) focused on the tribe Tradescantieae, and Bergamo’s (2003) study was restricted to Callisia . Analyses with the chloroplast region rbcL provided strong support for a monophyletic Commelinaceae ( Evans et al. 2003 ). Subtribal relationships in the Tradescantieae were also partly supported in Evans et al. (2003) and Wade et al. (2006) , with the exceptions that subtribes Tradescantiinae Rohw., Thyrsantheminae Faden & D. R. Hunt, and Dichorisandrinae (Pichon) Faden & D. R. Hunt were not monophyletic in parsimony analyses of rbcL and ndhF ( Wade et al. 2006 ). Within the Tradescantieae, Gibasis was nested within Tradescantia , and Callisia was not monophyletic ( Evans et al. 2003 ), containing Tripogandra within it. Bergamo (2003) also found Callisia to be nonmonophyletic with respect to Gibasis and Tripoganda . Within the tribe Commelineae, Evans et al. (2003) found that Commelina and Pollia were part of a clade sister to Murdannia with strong support. However, the monophyly of most genera has not been tested, and little is known about phylogenetic relationships within most genera, including the largest, Commelina (~170 species) and Tradescantia (~70 species; Faden 1998 ). The family has been taxonomically difficult because of a high degree of morphological homoplasy ( Evans et al. 2000 , 2003 ). The goal of this study was to develop a molecular phylogeny of Commelinaceae subfamily Commelinoideae, with particular focus on within genus sampling, especially for the largest two genera, Commelina and Tradescantia , using two DNA regions, the chloroplast intergenic spacer trnL-trnF and the 5S nuclear ribosomal nontranscribed spacer region (NTS). We sampled extensively within some genera and present a model-based (maximum likelihood, Bayesian) analysis of Commelinaceae subfamily Commelinoideae. Materials and Methods Taxon Sampling— We sampled 68 species of Commelinaceae (out of 650 species; Faden 1998 ), all of which are members of subfamily Commelinoideae, including representatives of 15 of the 41 genera in Commelinaceae ( Faden 1998 ). We sampled six of these genera, Commelina , Tradescantia , Murdannia , Cyanotis , Callisia , and Gibasis , more intensively than other genera to aid in phylogenetic comparative analyses of these taxa ( Burns 2006 ; Burns, Faden and Steppan, unpublished data). The present study did not sample members of the Cartonematoideae, which has only two genera, so it will not be able to address hypotheses about subfamily relationships (but see Evans et al. 2003 ). Among the more densely sampled genera, we sequenced 22 Commelina out of ~170 species ( Faden 1998 ). We also sequenced 17 out of ~70 species of Tradescantia , including T. standleyi (sect. Cymbispatha ), and T. zebrina (sect. Zebrina ), the monotypic sects. Rhoeo ( T. spathacea ), Campelia (T. zanonia ), and Corinna ( T. soconuscana ), sect. Cymbispatha member T. standleyi ( Hunt 1980 ), sect. Setcreasea members T. brevifolia , T. buckleyi, and T. pallida , sect. Austrotradescantia members T. blossfeldiana and T. fluminensis , and sect. Tradescantia ser. Virginianae members T. bracteata , T. occidentalis , T. ohiensis , T. roseolens , and T. hirsutiflora. Further, we sequenced five out of ~20 species of Callisia, five out of 50 species of Murdannia , five out of 50 Cyanotis species, and three out of 11 Gibasis ( Faden 1998 ). We did not sequence additional species in Callisia to avoid duplicating the ongoing efforts of Bergamo (2003) . The choice of outgroup taxa was based on published accounts of higher order relationships among Commelinaceae and related families, including the Philydraceae, Pontederiaceae, and Haemodoraceae ( Clark et al. 1993 ; Smith et al. 1993 ; Givnish et al. 1999 ; Kress et al. 2001 ; Evans et al. 2003 ; see also Chase et al. 2000 ). Outgroup sequences were downloaded from GenBank for the trnL-trnF region, including two species of Philydraceae, a single species of Pontederiaceae, and 17 species of Haemodoraceae ( Givnish et al. 1999 ; Chase et al. 2000 ; Evans et al. 2003 ). DNA Region Selection— DNA regions for phylogenetic analysis were chosen based on appropriate rate of molecular evolution, primer availability, and origin (e.g. cp vs. nuclear DNA) ( Olmstead and Palmer 1994 ). The trnL-trnF intergenic spacer is in the large single-copy region of the chloroplast DNA, is 120–350 bp in length in Commelinaceae, and is noncoding. Because the size of trnL-trnF is fairly small for phylogenetic analysis, and because preliminary phylogenies indicated that trnL-trnF would provide insufficient resolution at the tips of the phylogeny, we also sequenced the nuclear ribosomal 5S NTS for 60 species of Commelinaceae to improve intrageneric resolution. The 5S NTS is a more quickly evolving noncoding region with sufficient variability to clarify intrageneric relationships in a variety of taxa (e.g. Clivia Lindl., Amaryllidaceae, Ran et al. 2001 ; Ilex L., Aquifoliaceae, Manen et al. 2002 ; Lampranthus N. E. Br., Aizoaceae, Klak et al. 2003 ), and has yielded suitable levels of variation in other taxa in Commelinaceae ( Hardy 2001 ). The 5S NTS region is 98–424 bp in length in Commelinaceae. DNA Extraction, Amplification, and Sequencing— Extractions were performed on fresh leaf material following Hsiao et al. (1994) up to the first digestion. Two rounds of phenol-chloroform extraction were performed using standard methods with modifications for phaselock tubes (2 ml Phase Lock Gel Light; Eppendorf 1999 ). Phenol-chloroform solutions were incubated in a water bath at 55°C for 5 min for each of two rounds of extraction. A third extraction was performed with chloroform and was also incubated at 55°C for 5 min. DNA was diluted in distilled water or TLE to a concentration of 25 ng/ml. Amplification of trnL-trnF was conducted with primers “e” and “f” ( Taberlet et al. 1991 ) or trnLF and trnLR ( Sang et al. 1997 ). Amplification of trnL-trnF was modified from Taberlet et al. (1991) , with dimethyl sulphoxide (DMSO) added to the PCR reactions to minimize the effects of secondary structure and enhance the sequence signal ( Winship 1989 ). Amplification in 25 μL reactions included a mixture of 15.65 μL distilled water, 2.5 μL 10 × buffer, 1.5 μL MgCl 2 (25 mM), 1.25 μL DMSO, 0.1 μL dNTPs (25 mM), 0.5 μL Taq (Amplitaq Gold, Applied Biosystems, Foster City, California) polymerase (5 U/μL). Each reaction used 1.25 μL of each primer for a concentration of 10 mM per 25 μL reaction. For taxa that were difficult to amplify, 2.5 μL of BSA (10 μM) was added to the reaction mix. The PCR profile for trnL-trnF was 12 min at 94°C, then 40 cycles of denaturation at 94°C for 30 sec, annealing at 53°C for 30 sec, and extension at 72°C for 1 min, followed by 6 min at 72°C. The trnL-trnF intergenic spacer was sequenced with the amplification primers, with the exception of Pollia secundiflora , which was sequenced in the forward direction with primer “e” and in the reverse direction with primer “trnLR”, and Callisia navicularis , which was sequenced with “e” and “f”. Initial chromatograms of Commelina nairobiensis contained multiple, overlapping peaks and could not be interpreted; this species was TA cloned using TA vector pCR2.1-TOPO (Invitrogen, Carlsbad, California) following the manufacturer’s protocol, and four clones were sequenced per accession. Redundant clones of Commelina nairobiensis were excluded from analysis. Universal primers (PI = forward, PII = reverse) were used to amplify 5S NTS ( Cox et al. 1992 ). Amplification in a 25 μL reaction included: 16.65 μL water, 2.5 μL 10 × buffer, 1.25 μL DMSO, 0.5 μL MgCl 2 , (25 mM), 0.1 μL dNTPs (25 mM), and 0.5 μL Taq (or Amplitaq Gold) (5 U/μL). 1.25 μL of each primer was used at a concentration of 10 mM. Bovine serum albumen was added to the reaction mix for taxa that were difficult to amplify. Amplification of 5S NTS used 27 cycles of denaturing at 94°C for 1 min, annealing at 55°C for 1 min, and extension at 72°C for 2 min, followed by 6 mins at 72°C (modified from Cox et al. 1992 ). The Florida State University sequencing facility conducted sequencing on an Applied Biosystems 3100 Genetic Analyzer. Because 5S NTS is a nuclear repeat, multiple copies (paralogues) exist in any individual. These multiple repeats could have divergent paralogues, in some cases potentially misleading phylogenetic analyses ( Buckler et al. 1997 ). To ensure that sequences of 5S NTS were orthologous, PCR products of ~400 bp were isolated and cloned using TA cloning by the Florida State University cloning laboratory (vector pCR2.1-TOPO, from Invitrogen following the manufacturer’s protocol). Clones of 5S NTS were sequenced using the cloning primers M13F and M13R. Multiple clones (two to four per accession) of five closely related Commelina species (based on preliminary trnL-trnF phylogenies) were sequenced to assess the monophyly of 5S NTS sequences within individuals. To determine whether 5S NTS copies yield a meaningful signal at the species level, maximum likelihood analyses were conducted with multiple clones per accession. The longest sequence per species was analyzed for species with multiple clones. Then a single arbitrary clone of 5S NTS was sequenced for the remaining species, for a total of 60 Commelinaceae species sequenced for 5S NTS (not all sequences were alignable or contained useful phylogenetic information, 270 SYSTEMATIC BOTANY [Volume 36 and phylogenies from only 48 sequences are shown). Voucher information, GenBank accession numbers, and authorities for sampled species are reported in Appendix 1. Data Analysis— Sequences of trnL-trnF were aligned in Clustal W (Version 1.82; Thompson et al. 1994 ; Chenna et al. 2003 ) using the default parameters: IUB pairwise mass matrix, 10.0 Gap opening penalty, 6.66 gap extension penalty, 0.50 transition weighing, 8 gap separation distance, 40% identity for delay, and the end gap separation penalty off. Sequences of 5S NTS were aligned using the default parameters (Clustal W Version 1.82; Thompson et al. 1994 ; Chenna et al. 2003 ) with a modified gap extension penalty of 2.5 following Hardy (2001) . Because 5S NTS was quickly evolving and alignments across distantly related taxa proved difficult, separate alignments were conducted for Murdannia , Tradescantia , Gibasis , Callisia , Cyanotis and two clades of Commelina (Commelina 1 and Commelina 2; also see Swain 2009 for a similar approach). The choice to align within the genera Murdannia , Tradescantia , Gibasis , Callisia , and Cyanotis was independently supported by taxonomy and trnL-trnF analyses ( Fig. 1 ). It was not possible to create a reliable alignment across all of the Commelina sampled for this data set, and we choose the subsets for Commelina using an iterative procedure. First, we attempted to align all Commelina and Pollia, chosen as an outgroup based on trnL-trnF , following the procedure described above (TreeBASE study number S10425). Then, we chose two blocks of taxa that the algorithm had aligned well. These initial alignments were then modified by eye and ambiguous regions were excluded for a Callisia alignment with 13.98% missing data (unalignable sequence which was excluded from analysis), a Commelina 1 alignment with 12.93% missing data, a Commelina 2 alignment with 19.33% missing data, a Murdannia alignment with 48.23% missing data, a Tradescantia alignment with 19.48% missing data, and a Cyanotis alignment with 14.40% missing data. Two criteria were used to determine whether to combine data partitions. First, an incongruence length difference test (ILD) was used, which compares the length of the most parsimonious trees from two data partitions ( Farris et al. 1995 ). The ILD tests were conducted in PAUP for each 5S NTS alignment ( Callisia , Commelina 1 and Commelina 2, Cyanotis , Murdannia , and Tradescantia ) with 1,000 replicates (HomPart command), and compared with a trnL-trnF phylogeny for that same subset of taxa. In addition, separate analyses were conducted for each DNA region, and the resulting phylogenies were examined for strongly supported (i.e. ≥ 80% bootstrap support) areas of conflict ( Wiens 1998 ; see, e.g. Taylor and Hellberg 2005 ; but also see Bull et al. 1993 ; Cunningham 1997 ). We then combined alignments for trnL-trnF and 5S NTS partitions into a single alignment. Each data partition and the combined data were analyzed with equalweighting maximum parsimony (MP), maximum likelihood (ML), and Bayesian inference. Modeltest (Version 3.7; Posada and Crandall 1998 ) was used to determine the optimal model for ML and Bayesian analyses using Fig. 1. One of the two maximum likelihood trees for the trnL-trnF data partition. Maximum likelihood bootstrap values are shown above the node and Bayesian posterior probabilities are shown below the node for values over 50%. Numbers refer to multiple genotypes sequenced per taxon. Section/ subgenera (for Tradescantia and Commelina , respectively), subfamilies, and families are shown on the right. 2011] BURNS ET AL.: COMMELINACEAE PHYLOGENY 271 Akaike’s Information Criterion (AIC; Burnham and Anderson 2002 ). To determine whether there are multiple, equally parsimonious “islands” of trees, a multiple islands search approach was used, with five independent searches conducted with different random seeds ( Maddison 1991 ). All MP and ML analyses used a heuristic search with five replicates, random addition sequences, TBR branch swapping, and a starting tree generated by stepwise addition. Bootstrap analyses were conducted using the same search parameters and 100 replicates. Each ML bootstrap replicate was stopped after ~50 hrs, about twice the amount of time necessary to reach the most likely tree in the full likelihood analysis. Maximum parsimony and ML phylogenies were generated using PAUP* version 4.0b10 for Unix ( Swofford 1998 ). Bayesian analyses were conducted in MrBayes (version 3.1.2; Huelsenbeck and Ronquist 2001 ; Ronquist and Huelsenbeck 2003 ) with models chosen using Modeltest as described above. The trnL-trnF data set was run for 10 million generations sampled every 100 generations and the first 50% of the run was discarded. Data sets of 5S NTS were analyzed using four million generations sampled every 100 generations, with a discarded burnin of 10%. The combined data set was run for four million generations sampled every 100 generations, parameters estimated separately for the two data partitions, with branch lengths unlinked, with a discarded burnin of 10%. The chains were stopped when the average standard deviation of split frequencies (between two parallel chains) was less than 0.01 ( Ronquist et al. 2005 ), and cumulative posterior probabilities of split frequencies had stabilized ( Wilgenbusch et al. 2004 ). The trnL-trnF and combined analyses were rooted using members of Haemodoraceae, Philydraceae, and Pontederiaceae as outgroups. The 5S NTS analyses were rooted using the trnL-trnF topology to inform the rooting.