Blackwell Science, LtdOxford, UKBIJBiological Journal of the Linnean Society0024-4066The Linnean Society of London, 2003? 2003 79? 565576 Original Article WOLFFIELLA PHYLOGENETICS and BIOGEOGRAPHY R. T. KIMBALL Biological Journal of the Linnean Society, 2003, 79, 565–576. With 3 figures Out of Africa: molecular phylogenetics and biogeography of Wolffiella (Lemnaceae) REBECCA T. KIMBALL1*, DANIEL J. CRAWFORD2, DONALD H. LES3 and ELIAS LANDOLT4 1 Department of Zoology, University of Florida, PO Box 118525, Gainesville, Florida 32611, USA Department of Ecology and Evolutionary Biology & The Museum of Natural History and Biodiversity Center, University of Kansas, Lawrence, Kansas 66045, USA 3 Department of Ecology and Evolutionary Biology, University of Connecticut, Storrs, Connecticut 06269– 3043, USA 4 Geobotanisches Institut ETH, Zürichbergstrasse 38, CH-8044, Zürich, Switzerland 2 Received 5 August 2002; accepted for publication 29 January 2003 The monophyletic genus Wolffiella (Lemnaceae) comprises 10 species divided taxonomically into three sections. Relative to other genera of Lemnaceae, Wolffiella has a restricted range, with species distributed in warm temperate to tropical areas of Africa and the Americas, with only one species occurring in both areas. Sequence data from coding (rbcL and matK) and non-coding (trnK and rpl16 introns) regions of cpDNA were analyzed phylogenetically to resolve relationships within Wolffiella, and these results were compared to earlier allozyme and morphological studies. Allozymes, cpDNA and morphology all supported the recognition of three sections. Relationships among species were similar in most respects between the allozyme and cpDNA trees, as well as among the different plastid partitions. In Wolffiella, both non-synonymous and synonymous substitutions were greater in matK than in rbcL, as observed in other taxa. The synonymous substitution rate in matK was similar to the substitution rate of the noncoding regions. All partitions, including coding regions, exhibited some homoplasy. Biogeographical reconstructions from a combination of cpDNA partitions indicated that Wolffiella originated in Africa with early movement to and radiation in the Americas. The one species found in both Africa and the Americas, W. welwitschii, likely originated in the Americas and subsequently dispersed to Africa. Using the SOWH test, the cpDNA data could reject two alternative biogeographical hypotheses suggested from analyses of morphological and allozyme data. The present distribution of Wolffiella can be explained by two major dispersal events and this contrasts with the more complex species distributions in other Lemnaceae genera. Limited dispersal in Wolffiella relative to other Lemnaceae genera may be due to more recent origins of species, lower dispersibility and poorer colonizing ability. © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576. ADDITIONAL KEYWORDS: allozymes – duckweeds – matK – rbcL – rpl16. INTRODUCTION The duckweed family (Lemnaceae) includes the smallest of all flowering plants. These cosmopolitan aquatic monocots are reduced to tiny thalloid fronds that float on or below the surface of the water (Landolt, 1986). Although all species produce flowers, the frequency of flowering differs among the species (Landolt, 1986), *Corresponding author. E-mail: rkimball@zoo.ufl.edu and reproduction in the family is thought to be primarily vegetative. Wolffiella is unique biogeographically among genera of Lemnaceae in being restricted to warm temperate, subtropical and tropical areas in the Americas and Africa, with a relatively recent introduction of one species (W. hyalina) into India (Landolt, 1986). In contrast, other Lemnaceae genera have one or more species distributed widely in temperate zones (Landolt, 1986). The biogeographical origin of Wolffiella is uncertain. Allozyme analyses indicated an African origin (Crawford et al., 1997), although © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576 565 566 R. T. KIMBALL ET AL. Landolt (1986) concluded that the origins were likely in South America. In addition, it is not known whether there have been one or multiple dispersal events between the Old and New Worlds. The family as currently recognized consists of 37 species in five genera: Landoltia (one species), Lemna (13 species), Spirodela (two species), Wolffia (11 species) and Wolffiella (ten species) (Table 1; Les & Crawford, 1999; Landolt, 2000; Les et al., 2002). The taxonomic history of Wolffiella, as reviewed by Landolt (1986), has been intertwined with that of Wolffia, the other genus of the subfamily Wolffioideae. Landolt (1986) initially circumscribed Wolffiella as comprising nine species in three sections, with an additional species being described later (Landolt, 1992). Phylogenetic analyses of morphological and anatomical characters (Les, Landolt & Crawford, 1997b) indicated monophyly of the sections in Wolffiella but paraphyly of the genus. However, a tree five steps longer could resolve Wolffiella and the other genera as being monophyletic. A recent molecular phylogenetic study (Les et al., 2002) provided strong support for both the monophyly of the sections in Wolffiella and the monophyly of the genus (amended to include W. caudata as recognized by Landolt, 1992). In this study, we examine phylogenetic relation- Table 1. Accessions used in analyses Species rbcL matK, trnK rpl16 Wolffiella Section Wolffiella W. caudata Landolt 9158 (Bolivia) 9173 (Bolivia) W. denticulata (Hegelm.) Hegelm. W. gladiata (Hegelm.) Hegelm. 8221 (S. Africa) 8261 (USA) 8221 8261 W. lingulata (Hegelm.) Hegelm. 7289 (Brazil) 7289 W. neotropica Landolt 8848 (Brazil) 8848 W. oblonga (Phil.) Hegelm. 8984 (Columbia) 8984 W. welwitschii (Hegelm.) Monod 7468 (Columbia) 7468 9158* 9214 (Bolivia) 8221 7173 (USA) 8768 (USA)* 7289* 7655 (Mexico) 7290 (Brazil) 8848* 7997 (Brazil)* 8072 (USA) 8393 (USA) 8984 7468* 9096 (Zimbabwe) 8640 (Tanzania) 8640 9122 (Zimbabwe) 9122 Section Rotundae W. rotunda Landolt 9121 (Zimbabwe) 9121 9072 (Zimbabwe)* 9121 Wolffia W. australiana (Benth.) Hartog & Plas W. borealis (Englem.) Landolt W. brasiliensis Wedd. W. microscopica (Griff.) Kurz 7733 9123 8743 8359 7733 9123 8743 8359 7631 (Australia) 9123 9134 (Brazil) 8359 Section Stipitatae W. hyalina (Del.) Monod W. repanda (Hegelm.) Monod (Australia) (USA) (Argentina) (India) 7376 (Egypt) 8640* 9054 (Zimbabwe) 9062 (Zimbabwe)* 9104 (Botswana) 9107 (Botswana) 9122 Numbers are those of E.L. with vouchers in ZT. Geographic locality is indicated in parentheses for the first listing for each accession.*Accession used in combined analyses (and in Les et al., 2002). © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576 WOLFFIELIA PHYLOGENETICS AND BIOGEOGRAPHY ships in Wolffiella using cpDNA sequence data from all species. For most species, multiple, geographically divergent accessions were included. The phylogenetic trees of Wolffiella obtained from different cpDNA partitions were evaluated for congruence. Aspects of molecular evolution were evaluated for each data partition in order to better understand and to interpret instances of incongruence. A well-supported phylogeny for the genus is presented and is used to reconstruct the biogeographical history of Wolffiella. Alternative biogeographical hypotheses from earlier allozyme and morphological studies are then examined to better clarify the evolution of this unusual Lemnaceae genus. MATERIAL AND METHODS DNA AMPLIFICATION AND SEQUENCING Four regions of the chloroplast genome were used, including two protein coding loci (rbcL, matK) and two intron regions (rpl16, trnK – combining both the 5¢- and the 3¢ regions). Previously published data were supplemented by sequencing additional accessions for the rpl16 intron. PCR amplification of the rpl16 intron was performed using primers F71 (Jordan, Courtney & Neigel, 1996) and R622 (Les et al., 2002). PCR reactions were carried out using standard protocols, and products were cleaned using either QIAquick PCR purification columns (Qiagen, Inc., Valencia, CA) or by precipitation using an equal volume of PEG/NaCl (20%/2.5 M). Sequencing of the rpl16 intron was conducted using the amplification primers. Cycle sequencing reactions ( 1/4 or 1/2 volumes) were performed using the BigDye Terminator kit (PE Applied Biosystems, Foster City, CA) and by following the standard protocol provided for the ABI Prism 310 automated sequencer (PE Applied Biosystems, Foster City, CA). Sequence chromatographs were edited manually and assembled into double-stranded contigs. Sequences were aligned initially in Clustal W (Thompson, Higgins & Gibson, 1994), then manually optimized by visual inspection. Alignments were trimmed to exclude highly variable regions (e.g. near exon 1 of rpl16) where positional homology was difficult to establish. GenBank accession numbers for all molecular data, including our previously published (Les et al., 2002) and new sequences are: AY034200AY034209 (rbcL), AY034316-AY034325 (trnK 3¢ intron), AY034355-AY034364 (trnK 5¢ intron), AY034200-AY034209 (matK), AY034277-AY034286 and AY131184-AY131197 (rpl16 intron). We used DNA sequence data from each species of Wolffiella; four species in the genus Wolffia (W. australiana, W. borealis, W. brasiliensis and W. microscopica; 567 Table 1) were included as outgroups. For the rpl16 intron, two or more independent accessions were sequenced for each species except Wolffiella denticulata; when possible, additional accessions were sampled from geographically distinct regions (Table 1). PHYLOGENETIC ANALYSES Phylogenetic analyses were performed using default values in PAUP* 4.0b8 (Swofford, 1999) unless noted otherwise. To obtain the most parsimonious (MP) tree using equally weighted parsimony, a heuristic search was performed with 100 random sequence additions and tree bisection–reconnection (TBR) branch swapping. The reliability of specific taxon groupings under parsimony was examined using 1000 bootstrap replicates with ten random sequence additions per replicate. For parsimony analyses, data were analyzed both by treating indels as missing data and by including indels as characters following the ‘simple indel coding’ method of Simmons & Ochoterena (2000). To determine the appropriate evolutionary models for maximum likelihood (ML) analyses we used the hierarchical likelihood ratio test as implemented in MODELTEST 3.04 (Posada & Crandall, 1998). Parameters used in ML analyses were those recommended by MODELTEST. To compare parameters across data partitions, the transition/transversion ratio and the shape parameter (a) of a gamma distribution were also estimated for all four partitions using HKY85 + G (Hasegawa–Kishino–Yano model with gamma distributed rates) and the MP topology obtained from the combined dataset. To determine whether the different plastid partitions were concordant, we performed the partition homogeneity test (incongruence length difference test (ILD); Farris et al., 1995). Tests were performed using only the informative sites, with 1000 replicates and 10 random sequence additions per replicate. To estimate the degree of sequence divergence among the four partitions (rpl16, trnK, matK and rbcL), the sums of the branch lengths from the ML trees for each partition were obtained. In addition, pdistances were estimated for the two non-coding partitions. Non-synonymous and synonymous p-distances were estimated using the method of Nei & Gojobori (1986), as implemented in MEGA 1.02 (Kumar, Tamura & Nei, 1993). Corrected non-synonymous and synonymous distances were estimated using the method of Yang & Nielsen (2000), as implemented in PAML 3.12 (Yang, 2002). TESTS OF BIOGEOGRAPHICAL HYPOTHESES We used the SOWH test (Swofford et al., 1996; Goldman, Anderson & Rodrigio, 2000) to examine © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576 568 R. T. KIMBALL ET AL. specific biogeographical hypotheses for Wolffiella. This approach compared a test statistic, 2d (two times the difference in likelihood values), for the ML tree estimated from cpDNA data with that of an alternative topology. The two alternative topologies (representing alternative biogeographical hypotheses) were those indicated by analyses of either morphology (Landolt, 1986, 1992) or allozymes (Crawford et al., 1997). To determine whether an alternative topology was less likely statistically than was the cpDNA ML topology, a null distribution of the test statistic was generated using 500 simulated data sets. For each topology tested, we simulated data sets based upon the complete cpDNA data set using Seq-Gen 1.1 (Rambaut & Grassly, 1997). Parameters and branch length information were estimated using the alternative ML topology being tested in that specific SOWH test. For each of the 500 simulated data sets, a heuristic search was used to find the ML tree (as performed on the raw data above). Parameter estimates for each simulated data set were estimated using ML and the topology of the alternative tree being considered, as recommended by Goldman et al. (2000). The 2d test statistic generated from the simulated data sets was used to establish the null distribution for this statistic. We rejected the null hypothesis (that the ML and the alternative topology did not differ significantly) if fewer than 5% of the simulated data sets had 2d values greater than the observed 2d value. RESULTS MOLECULAR EVOLUTION The variability of the partitions differed, with the two non-coding partitions having similar levels of variability and being only slightly more variable than matK (Table 2). By summing the ML branch lengths across the most likely tree for each partition (Table 2), which gives a divergence rate corrected for homoplasy, the trnK and rpl16 introns appeared to be diverging most rapidly, with rbcL diverging much more slowly. To compare the rate of divergence at sites that should not be under selection, we compared pdistances among the synonymous sites in matK and rbcL with the non-coding sequences of rpl16 and trnK. Synonymous sites in matK have evolved the most rapidly, accumulating substitutions at 2.0, 1.6 and 1.5 times the rate of rbcL, trnK and rpl16, respectively. Non-synonymous substitutions have also accumulated much more rapidly in matK than in rbcL, occurring about 4.8 times faster than the rate in rbcL. The partitions also differed in the ratio of transitions to transversions and the shape parameter of a gamma distribution (Table 2), indicating different patterns of molecular evolution. In general, matK, trnK and rpl16 were the most similar, while rbcL was generally quite different. PHYLOGENETIC RELATIONSHIPS Analyses including indels as characters provided similar topologies for each partition (data not shown), with little or no improvement in resolution and consistency compared with analyses treating gaps as missing data (Table 3). Therefore, we elected to show our results with gaps treated as missing data. The ILD test indicated that the four cpDNA partitions were not significantly incongruent (P = 0.11). Some of the observed incongruence was distributed among the outgroup taxa, and their removal increased congruence of the different plastid partitions substantially (P = 0.40). Because significant incongruence was not observed, we analyzed the data partitions separately and in combination. The phylogenetic tree estimated from all four parti- Table 2. Parameters of different molecular data partitions Parameter rbcL matK trnK rpl16 No. of sites % variable sites % informative sites Sum ML branches Best model* Shape parameter (a)† ti/tv† No. of gap characters 1348 5.8 2.9 0.08707 HKY85 + G + I 0.01 0.94 – 1548 15.9 6.7 0.21412 F81 + G 0.34 0.59 4 1017 17.1 8.4 0.26820 F81 + G 0.23 0.56 49 460 17.6 8.7 0.29567 F81 + G 0.21 0.45 20 *Determined using MODELTEST 3.04. G = incorporates site-to-site rate heterogeneity using a gamma distribution, I = incorporates invariant sites. †Estimated using HKY85 + G and the topology of the most parsimonious tree from the combined analysis.ti/tv = transition/ transversion ratio. © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576 WOLFFIELIA PHYLOGENETICS AND BIOGEOGRAPHY 569 Table 3. Results of analysis on different molecular data partitions Parameter rbcL matK trnK rpl16 Gaps as missing data No. of MP trees CI, excluding uninformative 5 0.65 3 0.72 3 0.75 1 0.77 With gap matrix No. of MP trees CI, excluding uninformative – – 3 0.72 1 0.70 4 0.76 With rapidly evolving sites removed No. of MP trees CI, excluding uninformative 1 0.75 3 0.74 6 0.80 1 0.78 MP = most parsimonious. 100 85 98 99 75 100 94 100 99 94 100 --100 100 90 83 ----83 74 95 --66 --95 82 75 --58 --- 98 34 97 78 XX Distribution W. lingulata NA, SA W. oblonga NA, SA W. gladiata NA W. caudata SA W. neotropica SA Section Wolffiella W. welwitschii AF, SA 100 95 100 100 100 100 77 100 99 70 W. denticulata AF W. hyalina AF W. repanda AF W. rotunda AF Section Rotundae Section Stipitatae Figure 1. Most parsimonious tree using the combined sequence partitions, rooted with Wolffia (outgroup taxa not shown). Values at nodes represent percent of 1000 parsimony bootstrap replicates. Top value is from analysis of the combined sequence partitions, followed by analysis of rbcL, matK, trnK intron and rpl16 intron. Dashes indicate that the node was not found, but that alternative placements of those taxa were supported by fewer than 50% of bootstrap replicates; XX indicates that the node was not found, and that alternative placements of the taxa were supported by more than 50% of bootstrap replicates. Distribution: AF = Africa, NA = North America, SA = South America. tions combined gave a single well-resolved MP tree (CI excluding uninformative sites = 0.72) with greater than 70% bootstrap support at all nodes (Fig. 1). There was strong support for the monophyly of the three sections recognized by Landolt (1986) based on morphology and anatomy, and also by Crawford et al. (1997) based on allozyme analyses (Fig. 2). However, cpDNA sequence data indicated slightly different relationships within Section Wolffiella than those suggested by previous studies (cf. Figs 1, 2; see also Les et al., 2002). When the different cpDNA sequence partitions were analysed independently, slightly different topologies were obtained. These differences were primarily due to a lack of resolution among particular nodes (matK, trnK and one relationship in rpl16). In rbcL, W. caudata was placed basal to W. denticulata, though with less than 50% bootstrap support. Only in rpl16 was a topological difference supported, and this involved 60% bootstrap support for a clade containing W. caudata and W. oblonga. However, within each partition, there was no conflict between the strict consensus of all MP trees and the ML tree. Therefore, only the parsimony results are shown. Trees constructed © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576 570 R. T. KIMBALL ET AL. Distribution (a) W. gladiata NA W. oblonga NA, SA W. lingulata NA, SA W. caudata SA W. denticulata AF W. welwitschii AF, SA W. neotropica SA W. hyalina AF W. repanda AF W. rotunda AF W. lingulata NA, SA W. oblonga NA, SA W. gladiata NA W. welwitschii (b) 0.00 0.25 0.50 0.75 AF, SA W. caudata SA W. neotropica SA W. denticulata AF W. hyalina AF W. repanda AF W. rotunda AF Section Wolffiella Section Stipitatae Section Rotundae Section Wolffiella Section Stipitatae Section Rotundae 1.00 Figure 2. Previous hypotheses of relationships in Wolffiella. (a) Redrawn from Landolt (1986) with the placement of W. caudata based on comments of Landolt (1992); (b) Phenogram (UPGMA) based on genetic identity at allozyme loci. Scale across bottom indicates genetic identity. Redrawn from Crawford et al. (1997). Designations of geographical distributions are the same as in Fig. 1. from individual partitions had few conflicts with the tree estimated from combined data (Fig. 1). The phylogeny estimated from rbcL differed most from the combined data tree (Fig. 1) in that W. caudata rather than W. denticulata was basal within Section Wolffiella. One other difference found in analyses of rpl16 data was that one accession of W. oblonga formed a clade with W. caudata rather than with W. lingulata (Fig. 1). However, neither of these conflicting nodes was supported by a bootstrap value greater than 50% (Fig. 1). Interestingly, the CIs of the non-coding regions were higher than those for either coding region (Table 3). When comparing the number of parsimony steps across the best tree (obtained using all four partitions combined), a small proportion of sites in each partition appeared to be evolving rapidly, accumulating three to four steps across the tree. Although the proportion of rapidly evolving sites was small, there was variation among partitions (1.08% rpl16, 1.28% trnK, 0.32% matK, 0.52% rbcL). These sites were not confined to third positions in the two coding partitions, as might have been expected. In particular, the seven rapidly evolving sites for rbcL included both a first and a second position, resulting in non-synonymous substitutions in each case. Since rapidly evolving sites may exhibit high levels of homoplasy, we reanalyzed the data after removal of these sites. With the rapidly evolving sites removed, the CI of all partitions increased (Table 3), suggesting such sites did exhibit © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576 WOLFFIELIA PHYLOGENETICS AND BIOGEOGRAPHY homoplasy. For rbcL, this altered the topology of the MP tree so that it was congruent with the tree obtained from the combined data set (Fig. 1). However, analysis of the more slowly evolving first, second or first and second positions of rbcL combined did not result in a topology congruent with Figure 1. For the other partitions, removal of rapidly evolving sites did not result in topological changes. The larger rpl16 alignment, which included multiple accessions for most species, was very similar in base composition and parameters to the smaller alignment (Tables 2, 3). The alignment contained 461 sites, 99 95 60 58 60 571 of which 18% were variable and 8% were parsimony informative. As with the smaller rpl16 alignment, the best model was F81 + G. To compare this alignment with the other data partitions, we used HKY85 + G. The shape parameter was estimated to be 0.22 and the transition/transversion ratio was 0.43. Analysis of this larger data set produced three MP trees with a CI (excluding uninformative sites) of 0.78. As may occur when analyzing alignments of relatively short sequences (Fehrer, 1996), bootstrap values were low at many nodes (Fig. 3), even when sequence identity was high. For example, the two accessions of W. neotropica W. lingulata 7289 * Brazil W. lingulata 7655 Mexico W. oblonga 8393 USA W. oblonga 8984 Columbi a W. oblonga 8072 USA W. oblonga 7997 * Brazil W. caudata 9158 * Bolivia W. caudata 9214 Bolivia W. gladiata 8768 * USA W. gladiata 7173 USA W. neotropica 8848 * Brazil 55 93 W. neotropica 7290 Brazil W. welwitschii 9096 Zimbabwe 93 W. welwitschii 7468 * Columbi a W. denticulata 8221 * S. Africa 73 70 63 100 98 W. hyalina 8640 * Tanzania W. hyalina 7376 Egypt W. repanda 9062 * Zimbabwe W. repanda 9054 Zimbabwe W. repanda 9122 Zimbabwe W. repanda 9104 Botswana W. repanda 9107 Botswana W. rotunda 9072 * Zimbabwe W. rotunda 9121 Zimbabwe Figure 3. Phylogeny estimated using the complete rpl16 alignment with multiple accessions for most species, rooted with Wolffia (outgroup taxa not shown). Asterisks identify the accession included in the combined analysis above. Geographic locality of each accession is given. Values at nodes represent per cent of 1000 parsimony bootstrap replicates. Nodes with less than 50% bootstrap support were collapsed. © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576 572 R. T. KIMBALL ET AL. Table 4. Number of nucleotide differences (excluding indels) in rpl16 sequences between two groups of closely related species W. caudata 9158 W. caudata 9214 W. gladiata 7173 W. gladiata 8768 W. lingulata 7289 W. lingulata 7655 W. oblonga 7997 W. oblonga 8072 W. oblonga 8393 W. oblonga 8984 W. hyalina 7376 W. hyalina 8640 W. repanda 9054 W. repanda 9062 W. repanda 9104 W. repanda 9107 W. repanda 9122 W. rotunda 9072 W. rotunda 9121 W. caudata W. gladiata W. lingulata W. oblonga 9158 9214 7173 8768 7289 7655 7997 8072 8393 8984 – 2 2 2 6 6 2 4 6 6 – 2 2 6 6 2 4 6 6 – 0 4 4 2 2 4 4 – 4 4 2 2 4 4 – 0 6 6 0 0 – 6 6 0 0 – 4 6 6 – 6 6 – 0 – W. hyalina W. repanda 7376 8640 9054 9062 – 0 1 2 6 6 1 5 5 – 1 2 6 6 1 5 5 – 1 7 7 0 6 6 – 8 8 1 7 7 were identical, yet W. neotropica was supported in only 55% of bootstrap replicates (Fig. 3) and the two identical accessions of W. gladiata did not even form a clade (Fig. 3). In addition to instances of similar rpl16 sequences that did not cluster together, there were species with some divergent accessions that showed a high degree of similarity to other species (Fig. 3; Table 4). For example, two accessions of W. oblonga (8393 and 8984) were identical to each other and also to both accessions of W. lingulata. Yet these two accessions differed from putative conspecifics (7997 and 8072) by six nucleotide substitutions. Wolffiella oblonga (7997) showed greater similarity to two accessions of W. caudata (Table 4), whereas W. oblonga (8072) was not closely related to any other accession. BIOGEOGRAPHY Different biogeographical hypotheses were indicated when relationships within Wolffiella were estimated using different types of data (morphology, allozymes, W. rotunda 9104 9107 9122 9072 9121 – 6 6 – 0 – – 0 7 11 10 – 7 11 10 cpDNA). Morphological data (Fig. 2a) indicated that Wolffiella originated either in Africa (distribution of Sections Stipitatae and Rotundae) or South America (distribution of W. neotropica). Whether an African or South American origin is assumed, dispersal between these two continents would have occurred at least three times if the morphological tree were correct. In contrast, both the allozyme and the cpDNA data (Figs 1, 2b) indicated an African origin, followed by two dispersal events between Africa and South America. The allozyme and cpDNA hypotheses differ in that the allozyme data suggest that W. welwitschii originated in South America, and then dispersed to Africa while the combined cpDNA data are equivocal on this point (Fig. 1; but see Fig. 3). Regardless of the topology, there have been at least three dispersal events between North and South America, assuming both W. lingulata and W. oblonga are each monophyletic. Results of the SOWH test indicated that the phylogeny estimated from the cpDNA data (Fig. 1), and the biogeographical hypothesis it supported, were significantly more likely than those supported by either morphological data (2d = 249.27, 2d-critical = 11.84, © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576 WOLFFIELIA PHYLOGENETICS AND BIOGEOGRAPHY P < 0.002) or allozyme data critical = 5.21, P < < 0.002). (2d = 57.817, 2d- DISCUSSION MOLECULAR EVOLUTION This study is similar to other published results (e.g. Steele & Vilgalys, 1994; Johnson & Soltis, 1995; Manos & Steele, 1997; Xiang, Soltis & Soltis, 1998) in showing that matK is more variable and diverges more rapidly than rbcL for both synonymous and nonsynonymous substitutions. Although rbcL appeared to be under greater constraint at non-synonymous sites, it was surprising to find a rapidly evolving first and second position that in both cases resulted in nonsynonymous substitutions in the rbcL data set. The different amino acids encoded at these sites are those typical of rbcL sequences (e.g. Kellogg & Juliano, 1997), and it may be that there is selection for one of several amino acids at these sites, but little selection for a specific one of those amino acids among the taxa examined here. Analysis of the rbcL data set that included rapidly evolving first and second position sites resulted in a topology that was inconsistent with all other analyses, suggesting that even nonsynonymous sites in rbcL may exhibit problematic levels of homoplasy (see also Manos & Steele, 1997). Although the degree of divergence among noncoding chloroplast regions varies greatly (e.g. Small et al., 1998), the two regions examined here both appeared to be diverging at relatively high rates, and this, combined with a lesser degree of constraint, might be expected to lead to greater homoplasy in these regions. However, the CIs of both non-coding regions were higher than for the coding regions, and removal of rapidly evolving (and potentially more homoplasious) sites had little effect on the resulting topology, indicating that the non-coding regions were not excessively homoplasious at the taxonomic level examined in this study. PHYLOGENETIC RELATIONSHIPS IN WOLFFIELLA Phylogenetic relationships in Wolffiella inferred from the present study may be compared to those hypothesized previously for the genus. In an attempt to resolve relationships within Wolffiella, Landolt (1986) used primitive and derived states of 26 anatomorphological characters and ranked species phylogenetically according to an ‘index of primitivity’ for those characters (Fig. 2a). We placed the subsequently named W. caudata as the sister to W. lingulata in Fig. 2a because Landolt (1992) viewed it as most closely related to that species. Landolt (1986) recognized a basal split within the genus, with one lineage consisting of Sections Rotundae and Stipitatae and the other 573 comprising Section Wolffiella. Les et al. (1997a,b) used 41 anatomorphological characters (including many of the same ones employed by Landolt, 1986) to conduct a phylogenetic analysis using maximum parsimony. That analysis showed the same basal split as Landolt (1986) hypothesized with monophyly of the three sections. However, the genus was not shown to be monophyletic. The uniformity of the anatomorphological traits used by Les et al. (1997a,b) precluded resolution of relationships among species. Crawford et al. (1997) inferred relationships within Wolffiella using allozyme data (Fig. 2b) and recovered the same basal split in the genus as well as clustering of the sections (Fig. 2a,b). Our combined cpDNA sequence phylogeny also supports the basal split as well as the sectional relationships depicted by Landolt (1986) and supported by later studies (Crawford et al., 1997; Les et al., 1997a,b). Of interest may be the tree based on the larger rpl16 alignment, in which Section Stipitatae is not monophyletic (Fig. 3; see also Table 4). Whether this result is real or due to an artefact remains to be determined. Species in the two lineages resulting from the basal split in Wolffiella differ in several ecological features (Landolt, 1986). Members of Sections Rotundae and Stipitatae live on the surface of seasonal waters and survive dry periods by producing seeds. The seeds germinate quickly, and rapid vegetative reproduction produces large populations that cover the surface of the water with the onset of the wet period (Landolt, 1994). In contrast, members of Section Wolffiella, both in the Americas and in Africa, live submersed in permanent waters. When overgrown by other plants, they survive by using organic substances from the water. They are also capable of sinking to the bottom of the water and using nutrients released from the soil. With the exception of W. welwitschii, members of Section Wolffiella live in permanent waters and do not rely on seeds for reproduction (Landolt, 1986). Thus, the initial phylogenetic split in Wolffiella reflects two lineages differing by several fundamental life history and ecological attributes. Relationships among species in Section Wolffiella have been difficult to infer. Les et al. (1997a,b) were unable to clarify these relationships using anatomorphological data. Greater resolution was achieved in our combined cpDNA phylogeny and in the allozyme dendrogram which agreed topologically in most respects, differing only in the placement of W. welwitschii (cf. Figs 1, 2b). Landolt (1986) portrayed close phylogenetic relationships among W. gladiata, W. lingulata and W. oblonga. He also viewed the newly described W caudata as closely related to these three species, and especially to W. lingulata (Landolt, 1992). A close relationship hypothesized between W. caudata and the other three © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576 574 R. T. KIMBALL ET AL. species is concordant with our cpDNA phylogeny. However, W. caudata is sister to the other three species in the cpDNA phylogeny whereas Landolt considered it to be most closely related to W. lingulata. The phylogeny of Landolt (1986) also differed in the placement of W. denticulata and W. neotropica (cf. Figs 1, 2a). The rpl16 sequences failed to group all accessions of Wolffiella oblonga (Fig. 3), with accessions occurring in three different regions of the tree (see also Table 4). The clustering of two accessions of W. oblonga(8393 and 8394) with W. lingulata is concordant with allozyme data; Crawford et al. (1997) found very high allozyme identities between these taxa (the highest yet found between any two Lemnaceae species) with no alleles unique to either species. This result is also supported by morphology, as Landolt (1986) observed that W. gladiata, W. oblonga and W. lingulata were ‘very difficult to recognize’ and that distinctions between W. oblonga and W. lingulata were ‘especially unclear’ due to extensive variability in key, defining characteristics. He further stated that it was difficult to determine whether certain collections contained both species, or a single species displaying different developmental forms (Landolt, 1986). Given this, it is possible that W. oblongaand W. lingulata may not form distinct species or may have diverged very recently. Also of interest is the accession of W. oblonga (7997) that grouped with W. caudata in the rpl16 analysis (Fig. 3, Table 4), though the reasons for this anomalous placement are unknown. Another accession of W. oblonga (8072) was distinct in the rpl16 analysis (Fig. 3) and did not show a high degree of similarity to any other accession. Too few data exist to determine whether this accession represents a novel lineage or whether its position is due to incomplete coalescence. Whatever the case, it is clear that a more detailed study of W. oblonga and other closely related species could be fruitful for elucidating their relationships. MOLECULAR BIOGEOGRAPHY OF WOLFFIELLA Determining the biogeographical history of Lemnaceae is challenging because they are readily dispersed by virtue of their minute size. Furthermore, there are several documented examples of species recently introduced by humans (Landolt, 1986). These factors have led to widespread distributions of Lemna, Spirodela and Wolffia. In contrast, Wolffiella is restricted to North and South America and Africa. Landolt (1986) suggested that Wolffiella originated in the warmer regions of South America because the ‘most primitive and probably most ancient’ species (W. neotropica) occurred there, although his tree was equivocal with regard to the origins of Wolffiella (Fig. 2a). Relationships inferred from divergence at allozyme loci and cpDNA data indicated an African origin for Wolffiella, with dispersal to America, followed by a later dispersal of W. welwitschii back into Africa (Figs 2b, 3). An African origin is more parsimonious than is a South American origin, as it requires only two dispersal events between Africa and the Americas, rather than the three required if Wolffiella originated in South America. We assumed a synonymous substitution rate of 0.12% per Myr for rbcL, which is similar to rates calculated for a variety of flowering plants (e.g. Xiang et al., 2000) to estimate the timing of the dispersal from Africa to the Americas. The divergence between W. denticulata and W. welwitschii was 2.35% at synonymous sites, which sets the estimated time for the divergence between American and African species at approximately 9.8 million years. Although it appears most likely that Wolffiella originated in Africa, it is less clear where W. welwitschii (which is distributed in both South America and Africa) evolved. Allozyme data (and morphology) suggest that W. welwitschii originated in South America, and then dispersed to Africa, while the combined cpDNA data are equivocal (Figs 1, 2). The accession of W. welwitschii that we sequenced for the combined data set was from South America. If the origin of the species were in Africa with dispersal to America, then it should be at least as divergent from the American species as it is from W. denticulata. In contrast, if the species originated in America with dispersal back to Africa the American accession should be less divergent from exclusively American species than from the African W. denticulata (Fig. 1). The accession of W. welwitschii from South America was more divergent from W. denticulata than from any American species of Wolffiella (about twice as divergent in matK and rbcL), supporting an American origin for W. welwitschii with dispersal back to Africa. Additional support for an American origin occurred in the rpl16 data, for which both an African and American accession were sequenced (Table 1; Fig. 3). The two accessions of W. welwitschii differed by only two substitutions (0.43%) whereas the mean divergence between W. welwitschii and the South American W. neotropica was 1.1% and divergence between W. welwitschii (whether from America or Africa) and its closest African species, W. denticulata, was over 2.5%. In Wolffiella, the rate of rpl16 sequence divergence was 1.33 times faster than at rbcL synonymous sites (see Results), giving an estimated time of 1.35 Myr for dispersal of W. welwitschii back to Africa. Among the species of Wolffiella, W. welwitschii is best adapted for dispersal because it flowers and sets seed rather frequently, and Lemnaceae seeds are able to survive out of water much longer than can their fronds (Landolt, 1997). © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576 WOLFFIELIA PHYLOGENETICS AND BIOGEOGRAPHY The distribution of Wolffiella differs markedly from the distributions of other Lemnaceae genera because it can be explained by only two major dispersal events. In contrast, attempts to reconstruct distributions of other genera using the methods employed in the present study have given ambiguous results (D.H. Les, unpubl. data) due to widespread distributions of several species in other genera. There are several historical and ecological factors that could account for the fewer dispersal events in Wolffiella than in the other genera. Molecular data (Crawford & Landolt, 1995; Crawford et al., 1997; Les et al., 2002) showed that species of Wolffiella are less divergent on average than are species of other genera of Lemnaceae, suggesting that species of Wolffiella are ‘younger’ than those of other genera. Thus, the lack of dispersal could be a reflection of time since origin. As indicated earlier, Wolffiella is restricted to tropical and subtropical regions where distances between continents are greater than in the boreal and nemoral regions of the northern hemisphere, and low dispersal may reflect the greater distances between continents. The fronds of Wolffiella are very thin, and cuticles function less efficiently against dehydration than in other Lemnaceae, making it difficult for the fronds to be transported long distances even when covered by feathers (Landolt, 1986). Seeds, which allow for more efficient transport than fronds, are rare in all species except members of Sections Rotundae and Stipitatae and in W. welwitschii (Landolt, 1986). The specialized adaptations of Sections Rotundae and Stipitatae to seasonally dry, local pools in Africa make them poor colonizers, so dispersal events would likely not result in effective colonization (Landolt, 1994). ACKNOWLEDGEMENTS We would like to thank E. L. Braun for assisting us with the SOWH test and allowing us to use his computer facilities. We thank two anonymous reviewers for critically reading an earlier version of the manuscript. Research was funded in part by NSF grant DEB-9806537 to D.H.L. and D.J.C. REFERENCES Crawford DJ, Landolt E. 1995. Allozyme divergence among species of Wolffia (Lemnaceae). Plant Systematics and Evolution 197: 59–70. Crawford DJ, Landolt E, Les DH, Tepe E. 1997. Allozyme variation and the taxonomy of Wolffiella (Lemnaceae). Aquatic Botany 58: 43–54. Farris JS, Källersjo M, Kluge AG, Bult C. 1995. Testing significance of incongruence. Cladistics 10: 315–319. Fehrer J. 1996. Conflicting character distribution within dif- 575 ferent data sets of cardueline finches: Artifact or history? Molecular Biology and Evolution 13: 7–20. Goldman N, Anderson JP, Rodrigo AG. 2000. Likelihoodbased tests of topologies in phylogenetics. Systematic Biology 49: 652–670. Johnson LA, Soltis DE. 1995. Phylogenetic inference in Saxifragaceae sensu stricto and Gilia (Polemoniaceae) using matK sequences. Annals of the Missouri Botanical Garden 82: 149–175. Jordan WC, Courtney MW, Neigel JE. 1996. Low levels of intraspecific genetic variation at a rapidly evolving chloroplast DNA locus in North American duckweeds (Lemnaceae). American Journal of Botany 83: 430–439. Kellogg EA, Juliano ND. 1997. The structure and function of rubisco and their implications for systematic studies. American Journal of Botany 84: 413–428. Kumar S, Tamura K, Nei M. 1993. MEGA: molecular evolution genetic analysis, Vers. 1.0. Pennsylvania: Pennsylvania State University. Landolt E. 1986. Biosystematic investigations in the family of duckweeds (Lemnaceae) volume 2. The family of Lemnaceae – a monographic study, volume 1. Veroffentlichungen des Geobotanischen Institutes der ETH, Stiftung Rubel Zurich 71. Landolt E. 1992. Wolffiella caudata, a new Lemnaceae species from the Bolivian Amazon region. Berichte des Geobotanischen Institutes der ETH, Stiftung Rubel, in Zurich 58: 121– 123. Landolt E. 1994. The Lemnaceae of Zimbabwe and Botswana. Berichte des Geobotanischen Institutes der ETH, Stiftung Rubel, in Zurich 60: 110–136. Landolt E. 1997. How do Lemnaceae (duckweeds) survive dry conditions? Bulletin of the Geobotanical Institute ETH. 63: 25–31. Landolt E. 2000. Contribution on the Lemnaceae of Ecuador. Fragmenta Floristica et Geobotanica 45: 221–237. Les DH, Crawford DJ. 1999. Landoltia (Lemnaceae), a new genus of duckweeds. Novon 9: 530–533. Les DH, Crawford DJ, Landolt E, Aakjar R, Tepe E. 1997a. Systematics of Lemnaceae revisited. American Journal of Botany 84: 211. Les DH, Crawford DJ, Landolt E, Gabel JD, Kimball RT. 2002. Phylogeny and systematics of Lemnaceae, the duckweed family. Systematic Botany 27: 221–240. Les DH, Landolt E, Crawford DJ. 1997b. Systematics of Lemnaceae: inferences from micromolecular and morphological data. Plant Systematics and Evolution 204: 161–177. Manos PS, Steele KP. 1997. Phylogenetic analyses of ‘higher’ Hamamelididae based on plastid sequence data. American Journal of Botany 84: 1407–1419. Nei M, Gojobori T. 1986. Simple methods for estimating the numbers of synonymous and nonsynonymous nucleotide substitutions. Molecular Biology and Evolution 3: 418– 426. Posada D, Crandall KA. 1998. MODELTEST: testing the model of DNA substitution. Bioinformatics 14: 817–818. Rambaut A, Grassly NC. 1997. Seq-Gen: An application for the Monte Carlo simulation of DNA sequence evolution © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576 576 R. T. KIMBALL ET AL. along phylogenetic trees. Computer Applications in Biosciences 13: 235–238. Simmons MP, Ochoterena H. 2000. Gaps as characters in sequence-based phylogenetic analyses. Systematic Biology. 49: 369–381. Small RL, Ryburn JA, Cronn RC, Seelanan T, Wendel JF. 1998. The tortoise and the hare: choosing between noncoding plastome and nuclear ADH sequences for phylogeny reconstruction in a recently diverged plant group. American Journal of Botany 85: 1301–1315. Steele KP, Vilgalys R. 1994. Phylogenetic analyses of Polemoniaceae using nucleotide sequences of the plastid gene matK. Systematic Botany 19: 126–142. Swofford DL. 1999. PAUP*: Phylogenetic analysis using parsimony (*and other methods), Version 4.0. Sunderland, MA: Sinauer Associates. Swofford DL, Olsen GJ, Waddell PJ, Hillis DM. 1996. Phylogenetic inference. In: Hillis DM, Moritz C, Marble BK, eds. Molecular systematics, 2nd edn. Sunderland, MA: Sinauer Associates, 404–514. Thompson JD, Higgins DG, Gibson TJ. 1994. Clustal W: Improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties and weight matrix choice. Nucleic Acids Research 22: 4673–4680. Xiang Q-Y, Soltis DE, Soltis PS. 1998. Phylogenetic relationships of Cornaceae and close relatives inferred from matK and rbcL sequences. American Journal of Botany 85: 285– 297. Xiang Q-Y, Soltis DE, Soltis PS, Manchester SR, Crawford DJ. 2000. Timing the eastern Asian–eastern North American floristic disjunction: molecular clock corroborates paleontological estimates. Molecular Phylogenetics and Evolution 15: 462–472. Yang Z. 2002. Phylogenetic analysis by maximum likelihood (PAML). Vers. 3.12. London: University College. Yang Z, Nielsen R. 2000. Estimating synonymous and nonsynonymous substitution rates under realistic evolutionary models. Molecular Biology and Evolution 17: 32– 43. © 2003 The Linnean Society of London, Biological Journal of the Linnean Society, 2003, 79, 565–576